Dump fp16 values in expected_input4gcc.txt too.

Add expected_input4gcc-nofp16.txt.

4 files changed, 7619 insertions, 11 deletions

diff --git a/compute_ref.axf b/compute_ref.axf
index 20474d1..4453bb2 100644
--- a/compute_ref.axf
+++ b/compute_ref.axf
diff --git a/expected_input4gcc-nofp16.txt b/expected_input4gcc-nofp16.txt
new file mode 100644
index 0000000..5cefedb
--- /dev/null
+++ b/expected_input4gcc-nofp16.txt
@@ -0,0 +1,7089 @@
+
+VLD1/VLD1Q output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+
+VADD/VADDQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffec, 0xffed, 0xffee, 0xffef };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff3, 0xfffffff4 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x54 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xe, 0xf, 0x10, 0x11 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x18, 0x19 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffdc, 0xffdd, 0xffde, 0xffdf, 0xffe0, 0xffe1, 0xffe2, 0xffe3 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffd2, 0xffffffd3, 0xffffffd4, 0xffffffd5 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x8, 0x9 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xfc, 0xfd, 0xfe, 0xff, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7, 0xfff8, 0xfff9, 0xfffa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x27, 0x28, 0x29, 0x2a };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff3, 0xfffffffffffffff4 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x40d9999a, 0x40d9999a };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x41100000, 0x41100000, 0x41100000, 0x41100000 };
+
+VLD1_LANE/VLD1_LANEQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xaa };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xaaaaaaaa, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xaaaaaaaa, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xc1800000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xfffffff0, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xaaaaaaaaaaaaaaaa, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xfffffff0, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xaaaaaaaaaaaaaaaa };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xc1800000, 0xaaaaaaaa };
+
+VLD1_DUP/VLD1_DUPQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1800000, 0xc1800000, 0xc1800000 };
+
+VLD1_DUP/VLD1_DUPQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1700000, 0xc1700000, 0xc1700000 };
+
+VLD1_DUP/VLD1_DUPQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xc1600000, 0xc1600000, 0xc1600000 };
+
+VDUP/VDUPQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1800000, 0xc1800000, 0xc1800000 };
+
+VDUP/VDUPQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1700000, 0xc1700000, 0xc1700000 };
+
+VDUP/VDUPQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xc1600000, 0xc1600000, 0xc1600000 };
+
+VMOV/VMOVQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1800000, 0xc1800000, 0xc1800000 };
+
+VMOV/VMOVQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1700000, 0xc1700000, 0xc1700000 };
+
+VMOV/VMOVQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xc1600000, 0xc1600000, 0xc1600000 };
+
+VGET_HIGH output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VGET_LOW output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQDMLAL_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7c1e, 0x7c1f, 0x7c20, 0x7c21 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7c1e, 0x7c1f };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQDMLAL_LANE (mul with input=0) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQDMLAL_LANE (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffef, 0x7ffffff0, 0x7ffffff1, 0x7ffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffef, 0x7ffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQDMLSL_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffff83c2, 0xffff83c3, 0xffff83c4, 0xffff83c5 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffff83c2, 0xffffffffffff83c3 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQDMLSL_LANE (mul with input=0) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQDMLSL_LANE (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQDMLAL_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x1684, 0x1685, 0x1686, 0x1687 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x21ce, 0x21cf };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQDMLAL_N (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffef, 0x7ffffff0, 0x7ffffff1, 0x7ffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffef, 0x7ffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQDMLSL_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffe95c, 0xffffe95d, 0xffffe95e, 0xffffe95f };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffde12, 0xffffffffffffde13 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQDMLSL_N (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VEXT/VEXTQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf7, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff3, 0x22, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0x33 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf6, 0xf7, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff2, 0xfff3, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0x77 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf6, 0xf7, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff3, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0x42066666 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xfe, 0xff, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff7, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff3, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff1, 0x44 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xfc, 0xfd, 0xfe, 0xff, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff6, 0xfff7, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff3, 0x77, 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff1, 0x88 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xfc, 0xfd, 0xfe, 0xff, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff6, 0xfff7, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1500000, 0x4204cccd, 0x4204cccd, 0x4204cccd };
+
+VSHR_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffc };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x3c, 0x3c, 0x3c, 0x3c, 0x3d, 0x3d, 0x3d, 0x3d };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1ffe, 0x1ffe, 0x1ffe, 0x1ffe };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x7ffffff, 0x7ffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x7fffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc, 0xfc, 0xfd, 0xfd, 0xfe, 0xfe, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffc, 0xfffffffc, 0xfffffffc, 0xfffffffc };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x3c, 0x3c, 0x3c, 0x3c, 0x3d, 0x3d, 0x3d, 0x3d, 0x3e, 0x3e, 0x3e, 0x3e, 0x3f, 0x3f, 0x3f, 0x3f };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1ffe, 0x1ffe, 0x1ffe, 0x1ffe, 0x1ffe, 0x1ffe, 0x1ffe, 0x1ffe };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x7ffffff, 0x7ffffff, 0x7ffffff, 0x7ffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSHRN_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff8, 0xfff8, 0xfff9, 0xfff9 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffc };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xfc, 0xfc, 0xfc, 0xfc, 0xfd, 0xfd, 0xfd, 0xfd };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfffe, 0xfffe, 0xfffe, 0xfffe };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHRN_N (with input = 0) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHRN_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff8, 0xfff9, 0xfff9, 0xfffa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffc };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xfc, 0xfc, 0xfd, 0xfd, 0xfd, 0xfd, 0xfe, 0xfe };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfffe, 0xfffe, 0xfffe, 0xfffe };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHRN_N (with large shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQRSHRN_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff8, 0xfff9, 0xfff9, 0xfffa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffc };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQRSHRN_N (check saturation: shift by 3) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQRSHRN_N (check saturation: shift by max) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSET_LANE/VSET_LANEQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0x11 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0x22 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0x33 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x44 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0x55, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0x66, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0x77 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x88 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0x55, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0x66, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0x4204cccd };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0x99 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xaa, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xbb };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xcc };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xdd, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xee, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xff, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0x11 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xdd, 0xff };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xee, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0x41333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQSUB/VQSUBQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xdf, 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffce, 0xffcf, 0xffd0, 0xffd1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffffbd, 0xffffffbe };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffac };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x9b, 0x9c, 0x9d, 0x9e, 0x9f, 0xa0, 0xa1, 0xa2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xff8a, 0xff8b, 0xff8c, 0xff8d };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffff79, 0xffffff7a };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffff68 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xdf, 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffce, 0xffcf, 0xffd0, 0xffd1, 0xffd2, 0xffd3, 0xffd4, 0xffd5 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffbd, 0xffffffbe, 0xffffffbf, 0xffffffc0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffac, 0xffffffffffffffad };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x9b, 0x9c, 0x9d, 0x9e, 0x9f, 0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xff8a, 0xff8b, 0xff8c, 0xff8d, 0xff8e, 0xff8f, 0xff90, 0xff91 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffff79, 0xffffff7a, 0xffffff7b, 0xffffff7c };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffff68, 0xffffffffffffff69 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffac };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffff68 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffac, 0xffffffffffffffad };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffff68, 0xffffffffffffff69 };
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQDMULH output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+
+VQDMULH output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQDMULH_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+
+VQDMULH_LANE (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQDMULH_N output:
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x19, 0x19, 0x19, 0x19 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x4, 0x4 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xa, 0xa, 0xa, 0xa };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+
+VQDMULH_N (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+
+VQDMULL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x200, 0x1c2, 0x188, 0x152 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x200, 0x1c2 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+
+VQDMULL (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQDMLAL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7c1e, 0x7c1f, 0x7c20, 0x7c21 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7c1e, 0x7c1f };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQDMLAL (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffef, 0x7ffffff0, 0x7ffffff1, 0x7ffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffef, 0x7ffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQDMLSL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffff83c2, 0xffff83c3, 0xffff83c4, 0xffff83c5 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffff83c2, 0xffffffffffff83c3 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQDMLSL (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VCEQ/VCEQQ output:
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0xff, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0xff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0xff, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0xff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+
+VCGE/VCGEQ output:
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+
+VCLE/VCLEQ output:
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+
+VCGT/VCGTQ output:
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+
+VCLT/VCLTQ output:
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+
+VBSL/VBSLQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf6, 0xf6, 0xf6, 0xf6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffffd };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800004, 0xc1700004 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf6, 0xf6, 0xf6, 0xf6, 0xf2, 0xf2, 0xf2, 0xf2, 0xf6, 0xf6, 0xf6, 0xf6 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2, 0xfff4, 0xfff4, 0xfff6, 0xfff6 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffffd, 0xfffffffffffffffd };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7, 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2, 0xfff4, 0xfff4, 0xfff6, 0xfff6 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7, 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2, 0xfff4, 0xfff4, 0xfff6, 0xfff6 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800001, 0xc1700001, 0xc1600001, 0xc1500001 };
+
+VSHL/VSHLQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xff80, 0xff88, 0xff90, 0xff98 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffff000, 0xfffff100 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffff80 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xff80, 0xff88, 0xff90, 0xff98 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffff000, 0xfffff100 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffff80 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0, 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x1000, 0x2000, 0x3000, 0x4000, 0x5000, 0x6000, 0x7000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x40000000, 0x80000000, 0xc0000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0, 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x1000, 0x2000, 0x3000, 0x4000, 0x5000, 0x6000, 0x7000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x40000000, 0x80000000, 0xc0000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSHL/VSHLQ (large shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSHL/VSHLQ (negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff8, 0xfff8, 0xfff9, 0xfff9 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffc };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x78, 0x78, 0x79, 0x79, 0x7a, 0x7a, 0x7b, 0x7b };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x7ff8, 0x7ff8, 0x7ff9, 0x7ff9 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x3ffffffc, 0x3ffffffc };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xfc, 0xfc, 0xfc, 0xfc, 0xfd, 0xfd, 0xfd, 0xfd, 0xfe, 0xfe, 0xfe, 0xfe, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffe, 0xfffffffe, 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x3c, 0x3c, 0x3c, 0x3c, 0x3d, 0x3d, 0x3d, 0x3d, 0x3e, 0x3e, 0x3e, 0x3e, 0x3f, 0x3f, 0x3f, 0x3f };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x7ff, 0x7ff, 0x7ff, 0x7ff, 0x7ff, 0x7ff, 0x7ff, 0x7ff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1ffffffe, 0x1ffffffe, 0x1ffffffe, 0x1ffffffe };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7ffffffffffffff, 0x7ffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSHL_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffe0, 0xffe2, 0xffe4, 0xffe6 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffff80, 0xffffff88 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffc0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xc0, 0xc4, 0xc8, 0xcc, 0xd0, 0xd4, 0xd8, 0xdc };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xff00, 0xff10, 0xff20, 0xff30 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffff80, 0xffffff88 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffe0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0, 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe0, 0xffe2, 0xffe4, 0xffe6, 0xffe8, 0xffea, 0xffec, 0xffee };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffc0, 0xffffffc4, 0xffffffc8, 0xffffffcc };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffc0, 0xffffffffffffffc4 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xc0, 0xc4, 0xc8, 0xcc, 0xd0, 0xd4, 0xd8, 0xdc, 0xe0, 0xe4, 0xe8, 0xec, 0xf0, 0xf4, 0xf8, 0xfc };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xff80, 0xff88, 0xff90, 0xff98, 0xffa0, 0xffa8, 0xffb0, 0xffb8 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffc0, 0xffffffc4, 0xffffffc8, 0xffffffcc };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffe0, 0xffffffffffffffe2 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQSHL/VQSHLQ (with input = 0) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQSHL/VQSHLQ (input 0 and negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQSHL/VQSHLQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xff80, 0xff88, 0xff90, 0xff98 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffff000, 0xfffff100 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffffe };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1ffffffffffffffe };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQSHL/VQSHLQ (negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfffc, 0xfffc, 0xfffc, 0xfffc };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x78, 0x78, 0x79, 0x79, 0x7a, 0x7a, 0x7b, 0x7b };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3ffc, 0x3ffc, 0x3ffc, 0x3ffc };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1ffffffe, 0x1ffffffe };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x7ffff, 0x7ffff, 0x7ffff, 0x7ffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffff, 0xfffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQSHL/VQSHLQ (large shift amount, negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQSHL/VQSHLQ (check cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3fff, 0x3fff, 0x3fff, 0x3fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x3fffffff, 0x3fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3fff, 0x3fff, 0x3fff, 0x3fff, 0x3fff, 0x3fff, 0x3fff, 0x3fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x3fffffff, 0x3fffffff, 0x3fffffff, 0x3fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3fffffffffffffff, 0x3fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQSHL/VQSHLQ (large shift amount, positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQSHL/VQSHLQ (check saturation on 64 bits) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQSHL_N/VQSHLQ_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xc0, 0xc4, 0xc8, 0xcc, 0xd0, 0xd4, 0xd8, 0xdc };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffe0, 0xffe2, 0xffe4, 0xffe6 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffffe0, 0xffffffe2 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffc0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xc0, 0xc4, 0xc8, 0xcc, 0xd0, 0xd4, 0xd8, 0xdc, 0xe0, 0xe4, 0xe8, 0xec, 0xf0, 0xf4, 0xf8, 0xfc };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe0, 0xffe2, 0xffe4, 0xffe6, 0xffe8, 0xffea, 0xffec, 0xffee };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe0, 0xffffffe2, 0xffffffe4, 0xffffffe6 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffc0, 0xffffffffffffffc4 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQSHL_N/VQSHLQ_N (check saturation with large positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ (with input = 0) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ (input 0 and negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xff80, 0xff88, 0xff90, 0xff98 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffff000, 0xfffff100 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffffe };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xff80, 0xff88, 0xff90, 0xff98 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffff000, 0xfffff100 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1ffffffffffffffe };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x1000, 0x2000, 0x3000, 0x4000, 0x5000, 0x6000, 0x7000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x1000, 0x2000, 0x3000, 0x4000, 0x5000, 0x6000, 0x7000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ (negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfffc, 0xfffc, 0xfffd, 0xfffd };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x78, 0x79, 0x79, 0x7a, 0x7a, 0x7b, 0x7b, 0x7c };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3ffc, 0x3ffc, 0x3ffd, 0x3ffd };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1ffffffe, 0x1ffffffe };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x80000, 0x80000, 0x80000, 0x80000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x100000000000, 0x100000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ (checking round_const overflow: shift by -1) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x4000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x4000000000000000, 0x4000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ (checking round_const overflow: shift by -3) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x1000, 0x1000, 0x1000, 0x1000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x10000000, 0x10000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x1000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10000000, 0x10000000, 0x10000000, 0x10000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x1000000000000000, 0x1000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x20000000, 0x20000000, 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ (checking negative shift amount as large as input vector width) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ (large shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHL/VRSHLQ (large negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD2/VLD2Q chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+
+VLD2/VLD2Q chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1400000, 0xc1300000, 0xc1200000, 0xc1100000 };
+
+VLD3/VLD3Q chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+
+VLD3/VLD3Q chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1400000, 0xc1300000, 0xc1200000, 0xc1100000 };
+
+VLD3/VLD3Q chunk 2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff4, 0xfffffff5 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff4, 0xfffffff5 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1400000, 0xc1300000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff8, 0xfffffff9, 0xfffffffa, 0xfffffffb };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff8, 0xfffffff9, 0xfffffffa, 0xfffffffb };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1000000, 0xc0e00000, 0xc0c00000, 0xc0a00000 };
+
+VLD4/VLD4Q chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+
+VLD4/VLD4Q chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1400000, 0xc1300000, 0xc1200000, 0xc1100000 };
+
+VLD4/VLD4Q chunk 2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff4, 0xfffffff5 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff4, 0xfffffff5 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1400000, 0xc1300000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff8, 0xfffffff9, 0xfffffffa, 0xfffffffb };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff8, 0xfffffff9, 0xfffffffa, 0xfffffffb };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1000000, 0xc0e00000, 0xc0c00000, 0xc0a00000 };
+
+VLD4/VLD4Q chunk 3 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff3 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff3 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1200000, 0xc1100000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffc, 0xfffffffd, 0xfffffffe, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffffc, 0xfffffffd, 0xfffffffe, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc0800000, 0xc0400000, 0xc0000000, 0xbf800000 };
+
+VDUP_LANE/VDUP_LANEQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1700000, 0xc1700000, 0xc1700000 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+
+VQDMULL_LANE output:
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x4000, 0x4000 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+
+VQDMULL_N output:
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x44000, 0x44000, 0x44000, 0x44000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xaa000, 0xaa000 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+
+VST1_LANE/VST1_LANEQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf7, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff3, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf6, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff2, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf6, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xff, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff5, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff1, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xfa, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff4, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff3, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xfa, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff4, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0x33333333, 0x33333333, 0x33333333 };
+
+VSUB/VSUBQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffffed, 0xffffffee };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffff8c };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xdc, 0xdd, 0xde, 0xdf, 0xe0, 0xe1, 0xe2, 0xe3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffd2, 0xffd3, 0xffd4, 0xffd5 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffc8, 0xffffffc9 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffee };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xe, 0xf, 0x10, 0x11 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffd8, 0xffffffffffffffd9 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffed, 0xffee, 0xffef, 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffb9, 0xffffffba, 0xffffffbb, 0xffffffbc };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffed, 0xffffffffffffffee };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc00ccccd, 0xc00ccccd };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc00ccccc, 0xc00ccccc, 0xc00ccccc, 0xc00ccccc };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQADD/VQADDQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x12, 0x13, 0x14, 0x15 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x23, 0x24 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x34 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf, 0x10 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x23, 0x24, 0x25, 0x26 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x34, 0x35 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x34 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x34, 0x35 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+
+VABS/VABSQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x10, 0xf, 0xe, 0xd };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x10, 0xf };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0xf, 0xe, 0xd };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x40133333, 0x40133333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x4059999a, 0x4059999a, 0x4059999a, 0x4059999a };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQABS/VQABSQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x10, 0xf, 0xe, 0xd };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x10, 0xf };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0xf, 0xe, 0xd };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+
+VQABS/VQABSQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VCOMBINE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x22, 0x22, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0x44 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x66, 0x66, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0x88 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x66, 0x66, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0x40533333, 0x40533333 };
+
+VMAX/VMAXQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1780000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf9, 0xf9, 0xf9, 0xf9, 0xf9, 0xf9, 0xf9, 0xf9, 0xf9, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1680000, 0xc1680000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VMIN/VMINQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf3, 0xf3, 0xf3, 0xf3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf3, 0xf3, 0xf3, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff1, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1780000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xf9, 0xf9, 0xf9, 0xf9, 0xf9, 0xf9 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1680000, 0xc1680000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xff800000, 0xff800000, 0xff800000, 0xff800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+
+VNEG/VNEGQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x10, 0xf, 0xe, 0xd };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x10, 0xf };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0xf, 0xe, 0xd };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc0133333, 0xc0133333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc059999a, 0xc059999a, 0xc059999a, 0xc059999a };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQNEG/VQNEGQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x10, 0xf, 0xe, 0xd };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x10, 0xf };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0xf, 0xe, 0xd };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+
+VQNEG/VQNEGQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMLAL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xe907, 0xe908, 0xe909, 0xe90a, 0xe90b, 0xe90c, 0xe90d, 0xe90e };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3e07, 0x3e08 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a, 0x3e0b, 0x3e0c, 0x3e0d, 0x3e0e };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3e07, 0x3e08 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMLSL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x16d9, 0x16da, 0x16db, 0x16dc, 0x16dd, 0x16de, 0x16df, 0x16e0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffc1d9, 0xffffffffffffc1da };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xc1d9, 0xc1da, 0xc1db, 0xc1dc, 0xc1dd, 0xc1de, 0xc1df, 0xc1e0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffc1d9, 0xffffffffffffc1da };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMLAL_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3e07, 0x3e08 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3e07, 0x3e08 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMLSL_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffc1d9, 0xffffffffffffc1da };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffc1d9, 0xffffffffffffc1da };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMLAL_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x595, 0x596, 0x597, 0x598 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xb3a, 0xb3b };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x10df, 0x10e0, 0x10e1, 0x10e2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x10df, 0x10e0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMLSL_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffa4b, 0xfffffa4c, 0xfffffa4d, 0xfffffa4e };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffff4a6, 0xfffffffffffff4a7 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffef01, 0xffffef02, 0xffffef03, 0xffffef04 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffef01, 0xffffffffffffef02 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMOVL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMOVN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMULL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x100, 0xe1, 0xc4, 0xa9, 0x90, 0x79, 0x64, 0x51 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x100, 0xe1, 0xc4, 0xa9 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x100, 0xe1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xe100, 0xe2e1, 0xe4c4, 0xe6a9, 0xe890, 0xea79, 0xec64, 0xee51 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffe00100, 0xffe200e1, 0xffe400c4, 0xffe600a9 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe000000100, 0xffffffe2000000e1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x5500, 0x5501, 0x5504, 0x5505, 0x5510, 0x5511, 0x5514, 0x5515 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMULL_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VREV16 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6, 0xf9, 0xf8, 0xfb, 0xfa, 0xfd, 0xfc, 0xff, 0xfe };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6, 0xf9, 0xf8, 0xfb, 0xfa, 0xfd, 0xfc, 0xff, 0xfe };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6, 0xf9, 0xf8, 0xfb, 0xfa, 0xfd, 0xfc, 0xff, 0xfe };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VREV32 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4, 0xfb, 0xfa, 0xf9, 0xf8, 0xff, 0xfe, 0xfd, 0xfc };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2, 0xfff5, 0xfff4, 0xfff7, 0xfff6 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4, 0xfb, 0xfa, 0xf9, 0xf8, 0xff, 0xfe, 0xfd, 0xfc };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2, 0xfff5, 0xfff4, 0xfff7, 0xfff6 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4, 0xfb, 0xfa, 0xf9, 0xf8, 0xff, 0xfe, 0xfd, 0xfc };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2, 0xfff5, 0xfff4, 0xfff7, 0xfff6 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VREV64 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1800000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0, 0xff, 0xfe, 0xfd, 0xfc, 0xfb, 0xfa, 0xf9, 0xf8 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0, 0xfff7, 0xfff6, 0xfff5, 0xfff4 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff0, 0xfffffff3, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0, 0xff, 0xfe, 0xfd, 0xfc, 0xfb, 0xfa, 0xf9, 0xf8 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0, 0xfff7, 0xfff6, 0xfff5, 0xfff4 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff1, 0xfffffff0, 0xfffffff3, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0, 0xff, 0xfe, 0xfd, 0xfc, 0xfb, 0xfa, 0xf9, 0xf8 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0, 0xfff7, 0xfff6, 0xfff5, 0xfff4 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1800000, 0xc1500000, 0xc1600000 };
+
+VSRA_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffd };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff3, 0xfffffff4 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffc, 0xfffffffd, 0xfffffffe, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf, 0x10, 0x11, 0x12, 0x13, 0x14 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff, 0x0, 0x1, 0x2, 0x3 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff3, 0xfffffff4, 0xfffffff5, 0xfffffff6 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTRN/VTRNQ chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0x11, 0x11, 0xf2, 0xf3, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0x55, 0x55, 0xf2, 0xf3, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0x55, 0x55, 0xf2, 0xf3, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0x11, 0x11, 0xf2, 0xf3, 0x11, 0x11, 0xf4, 0xf5, 0x11, 0x11, 0xf6, 0xf7, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0x22, 0x22, 0xfff2, 0xfff3, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0x55, 0x55, 0xf2, 0xf3, 0x55, 0x55, 0xf4, 0xf5, 0x55, 0x55, 0xf6, 0xf7, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0x66, 0x66, 0xfff2, 0xfff3, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0x55, 0x55, 0xf2, 0xf3, 0x55, 0x55, 0xf4, 0xf5, 0x55, 0x55, 0xf6, 0xf7, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0x66, 0x66, 0xfff2, 0xfff3, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0x42073333, 0x42073333 };
+
+VTRN/VTRNQ chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf4, 0xf5, 0x11, 0x11, 0xf6, 0xf7, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff2, 0xfff3, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf4, 0xf5, 0x55, 0x55, 0xf6, 0xf7, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff2, 0xfff3, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf4, 0xf5, 0x55, 0x55, 0xf6, 0xf7, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff3, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x42066666, 0x42066666 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf8, 0xf9, 0x11, 0x11, 0xfa, 0xfb, 0x11, 0x11, 0xfc, 0xfd, 0x11, 0x11, 0xfe, 0xff, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff4, 0xfff5, 0x22, 0x22, 0xfff6, 0xfff7, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff3, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf8, 0xf9, 0x55, 0x55, 0xfa, 0xfb, 0x55, 0x55, 0xfc, 0xfd, 0x55, 0x55, 0xfe, 0xff, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff4, 0xfff5, 0x66, 0x66, 0xfff6, 0xfff7, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff2, 0xfffffff3, 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf8, 0xf9, 0x55, 0x55, 0xfa, 0xfb, 0x55, 0x55, 0xfc, 0xfd, 0x55, 0x55, 0xfe, 0xff, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff4, 0xfff5, 0x66, 0x66, 0xfff6, 0xfff7, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xc1500000, 0x42073333, 0x42073333 };
+
+VUZP/VUZPQ chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+
+VUZP/VUZPQ chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x22, 0x22, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x66, 0x66, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x66, 0x66, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x42066666, 0x42066666 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x77, 0x77, 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x42073333, 0x42073333, 0x42073333, 0x42073333 };
+
+VZIP/VZIPQ chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf4, 0x11, 0x11, 0xf1, 0xf5, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff2, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf4, 0x55, 0x55, 0xf1, 0xf5, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff2, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf4, 0x55, 0x55, 0xf1, 0xf5, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff2, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf8, 0x11, 0x11, 0xf1, 0xf9, 0x11, 0x11, 0xf2, 0xfa, 0x11, 0x11, 0xf3, 0xfb, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff4, 0x22, 0x22, 0xfff1, 0xfff5, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff2, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf8, 0x55, 0x55, 0xf1, 0xf9, 0x55, 0x55, 0xf2, 0xfa, 0x55, 0x55, 0xf3, 0xfb, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff4, 0x66, 0x66, 0xfff1, 0xfff5, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff2, 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf8, 0x55, 0x55, 0xf1, 0xf9, 0x55, 0x55, 0xf2, 0xfa, 0x55, 0x55, 0xf3, 0xfb, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff4, 0x66, 0x66, 0xfff1, 0xfff5, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1600000, 0x42073333, 0x42073333 };
+
+VZIP/VZIPQ chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf6, 0x11, 0x11, 0xf3, 0xf7, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff3, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf2, 0xf6, 0x55, 0x55, 0xf3, 0xf7, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff3, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf6, 0x55, 0x55, 0xf3, 0xf7, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff3, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x42066666, 0x42066666 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf4, 0xfc, 0x11, 0x11, 0xf5, 0xfd, 0x11, 0x11, 0xf6, 0xfe, 0x11, 0x11, 0xf7, 0xff, 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff6, 0x22, 0x22, 0xfff3, 0xfff7, 0x22, 0x22 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff3, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf4, 0xfc, 0x55, 0x55, 0xf5, 0xfd, 0x55, 0x55, 0xf6, 0xfe, 0x55, 0x55, 0xf7, 0xff, 0x55, 0x55 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff2, 0xfff6, 0x66, 0x66, 0xfff3, 0xfff7, 0x66, 0x66 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff1, 0xfffffff3, 0x77, 0x77 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf4, 0xfc, 0x55, 0x55, 0xf5, 0xfd, 0x55, 0x55, 0xf6, 0xfe, 0x55, 0x55, 0xf7, 0xff, 0x55, 0x55 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff2, 0xfff6, 0x66, 0x66, 0xfff3, 0xfff7, 0x66, 0x66 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1500000, 0x42073333, 0x42073333 };
+
+VREINTERPRET/VREINTERPRETQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xffffffff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xffffffff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xf7f6f5f4f3f2f1f0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfff3fff2fff1fff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffff1fffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xf7f6f5f4f3f2f1f0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfff3fff2fff1fff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffff1fffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xf7f6f5f4f3f2f1f0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfff3fff2fff1fff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xf7f6f5f4f3f2f1f0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfff3fff2fff1fff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffff1fffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xf7f6f5f4f3f2f1f0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfff3fff2fff1fff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffff1fffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xf7f6f5f4f3f2f1f0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfff3fff2fff1fff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff, 0xf4, 0xff, 0xf5, 0xff, 0xf6, 0xff, 0xf7, 0xff };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff, 0xf2, 0xff, 0xff, 0xff, 0xf3, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff, 0xf4, 0xff, 0xf5, 0xff, 0xf6, 0xff, 0xf7, 0xff };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff, 0xf2, 0xff, 0xff, 0xff, 0xf3, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff, 0xf4, 0xff, 0xf5, 0xff, 0xf6, 0xff, 0xf7, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff, 0xfff2, 0xffff, 0xfff3, 0xffff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xffff, 0xffff, 0xffff, 0xfff1, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff, 0xfff2, 0xffff, 0xfff3, 0xffff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xffff, 0xffff, 0xffff, 0xfff1, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xffffffff, 0xfffffff1, 0xffffffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xffffffff, 0xfffffff1, 0xffffffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xf7f6f5f4f3f2f1f0, 0xfffefdfcfbfaf9f8 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfff3fff2fff1fff0, 0xfff7fff6fff5fff4 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffff1fffffff0, 0xfffffff3fffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xf7f6f5f4f3f2f1f0, 0xfffefdfcfbfaf9f8 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfff3fff2fff1fff0, 0xfff7fff6fff5fff4 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffff1fffffff0, 0xfffffff3fffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xf7f6f5f4f3f2f1f0, 0xfffefdfcfbfaf9f8 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfff3fff2fff1fff0, 0xfff7fff6fff5fff4 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff, 0xfff2, 0xffff, 0xfff3, 0xffff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xffff, 0xffff, 0xffff, 0xfff1, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff, 0xfff2, 0xffff, 0xfff3, 0xffff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xffff, 0xffff, 0xffff, 0xfff1, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xffffffff, 0xfffffff1, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xffffffff, 0xfffffff1, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xf7f6f5f4f3f2f1f0, 0xfffefdfcfbfaf9f8 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfff3fff2fff1fff0, 0xfff7fff6fff5fff4 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffff1fffffff0, 0xfffffff3fffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xf7f6f5f4f3f2f1f0, 0xfffefdfcfbfaf9f8 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfff3fff2fff1fff0, 0xfff7fff6fff5fff4 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffff1fffffff0, 0xfffffff3fffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xf7f6f5f4f3f2f1f0, 0xfffefdfcfbfaf9f8 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfff3fff2fff1fff0, 0xfff7fff6fff5fff4 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff, 0xf4, 0xff, 0xf5, 0xff, 0xf6, 0xff, 0xf7, 0xff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff, 0xf2, 0xff, 0xff, 0xff, 0xf3, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff, 0xf4, 0xff, 0xf5, 0xff, 0xf6, 0xff, 0xf7, 0xff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff, 0xf2, 0xff, 0xff, 0xff, 0xf3, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xf1, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff, 0xf4, 0xff, 0xf5, 0xff, 0xf6, 0xff, 0xf7, 0xff };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xfffffff0, 0xffffffff };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xfffffff0, 0xffffffff };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xf3f2f1f0, 0xf7f6f5f4 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xfff1fff0, 0xfff3fff2 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xfffffff0, 0xffffffff, 0xfffffff1, 0xffffffff };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xfffffff0, 0xffffffff, 0xfffffff1, 0xffffffff };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xf3f2f1f0, 0xf7f6f5f4, 0xfbfaf9f8, 0xfffefdfc };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xfff1fff0, 0xfff3fff2, 0xfff5fff4, 0xfff7fff6 };
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x80, 0xc1, 0x0, 0x0, 0x70, 0xc1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0xc180, 0x0, 0xc170 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xc1700000c1800000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x80, 0xc1, 0x0, 0x0, 0x70, 0xc1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0xc180, 0x0, 0xc170 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xc1700000c1800000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x80, 0xc1, 0x0, 0x0, 0x70, 0xc1 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0xc180, 0x0, 0xc170 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x80, 0xc1, 0x0, 0x0, 0x70, 0xc1, 0x0, 0x0, 0x60, 0xc1, 0x0, 0x0, 0x50, 0xc1 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0xc180, 0x0, 0xc170, 0x0, 0xc160, 0x0, 0xc150 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xc1700000c1800000, 0xc1500000c1600000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x80, 0xc1, 0x0, 0x0, 0x70, 0xc1, 0x0, 0x0, 0x60, 0xc1, 0x0, 0x0, 0x50, 0xc1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0xc180, 0x0, 0xc170, 0x0, 0xc160, 0x0, 0xc150 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xc1700000c1800000, 0xc1500000c1600000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x0, 0x0, 0x80, 0xc1, 0x0, 0x0, 0x70, 0xc1, 0x0, 0x0, 0x60, 0xc1, 0x0, 0x0, 0x50, 0xc1 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0xc180, 0x0, 0xc170, 0x0, 0xc160, 0x0, 0xc150 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQRDMULH output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff5, 0xfff6, 0xfff7, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+
+VQRDMULH (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQRDMULH (check rounding cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQRDMULH_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+
+VQRDMULH_LANE (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQRDMULH_LANE (check rounding cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQRDMULH_N output:
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfffc, 0xfffc, 0xfffc, 0xfffd };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x6, 0x6, 0x6, 0x5, 0x5, 0x4, 0x4, 0x4 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffe, 0xfffffffe, 0xfffffffe, 0xfffffffe };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+
+VQRDMULH_N (check mul cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+
+VQRDMULH_N (check rounding cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQRSHL/VQRSHLQ (with input = 0) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQRSHL/VQRSHLQ (input 0 and negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQRSHL/VQRSHLQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xff80, 0xff88, 0xff90, 0xff98 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffff000, 0xfffff100 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffff80 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQRSHL/VQRSHLQ (negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xfc, 0xfc, 0xfd, 0xfd, 0xfd, 0xfd, 0xfe, 0xfe };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfffc, 0xfffc, 0xfffd, 0xfffd };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x3c, 0x3c, 0x3d, 0x3d, 0x3d, 0x3d, 0x3e, 0x3e };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3ffc, 0x3ffc, 0x3ffd, 0x3ffd };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1ffffffe, 0x1ffffffe };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x80000, 0x80000, 0x80000, 0x80000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x100000000000, 0x100000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQRSHL/VQRSHLQ (checking cumulative saturation: shift by -1) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x4000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x4000000000000000, 0x4000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQRSHL/VQRSHLQ (checking cumulative saturation: shift by -3) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x1000, 0x1000, 0x1000, 0x1000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x10000000, 0x10000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x1000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10000000, 0x10000000, 0x10000000, 0x10000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x1000000000000000, 0x1000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x20000000, 0x20000000, 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQRSHL/VQRSHLQ (checking cumulative saturation: large shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQRSHL/VQRSHLQ (checking cumulative saturation: large shift amount with negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQRSHL/VQRSHLQ (checking cumulative saturation: large negative shift amount) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQRSHL/VQRSHLQ (checking cumulative saturation: large shift amount with 0 input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VABA/VABAQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x16, 0x17, 0x18, 0x19 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x20, 0x21 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5a };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x907, 0x908, 0x909, 0x90a };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffe7, 0xffffffe8 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x5e, 0x5f, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6a, 0x6b, 0x6c, 0x6d };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xb9c, 0xb9d, 0xb9e, 0xb9f, 0xba0, 0xba1, 0xba2, 0xba3 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x26e0, 0x26e1, 0x26e2, 0x26e3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VABAL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff6, 0xfff7, 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x16, 0x17, 0x18, 0x19 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x20, 0x21 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5a };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x907, 0x908, 0x909, 0x90a };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe7, 0xffffffe8 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VABAL test intermediate overflow output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffef, 0xfff0, 0xfff1, 0xfff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffef, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffe2, 0xffe3, 0xffe4, 0xffe5 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe7, 0xffffffe8 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VABD/VABDQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x11, 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3, 0x2, 0x1, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x18, 0x17 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffe3, 0xffe4, 0xffe5, 0xffe6 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffe8, 0xffffffe9 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x41c26666, 0x41ba6666 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x1a, 0x19, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0xf, 0xe, 0xd, 0xc, 0xb };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x4, 0x3, 0x2, 0x1, 0x0, 0x1, 0x2, 0x3 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x30, 0x2f, 0x2e, 0x2d };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffe4, 0xffe5, 0xffe6, 0xffe7, 0xffe8, 0xffe9, 0xffea, 0xffeb };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffd0, 0xffffffd1, 0xffffffd2, 0xffffffd3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x42407ae1, 0x423c7ae1, 0x42387ae1, 0x42347ae1 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VABDL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x11, 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x3, 0x2, 0x1, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x18, 0x17 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffe3, 0xffe4, 0xffe5, 0xffe6 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe8, 0xffffffe9 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VAND/VANDQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x2, 0x2, 0x0, 0x0, 0x2, 0x2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x60 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x14, 0x14, 0x14, 0x14 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x10, 0x10, 0x12, 0x12 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf2, 0xf2, 0xf4, 0xf4, 0xf6, 0xf6, 0xf0, 0xf0, 0xf2, 0xf2, 0xf4, 0xf4, 0xf6, 0xf6 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe0, 0xffe0, 0xffe0, 0xffe0, 0xffe4, 0xffe4, 0xffe4, 0xffe4 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe0, 0xffffffe0, 0xffffffe2, 0xffffffe2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x10, 0x10 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x4, 0x4, 0x4, 0x4, 0x8, 0x8, 0x8, 0x8, 0xc, 0xc, 0xc, 0xc };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x0, 0x1, 0x2, 0x3 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x30, 0x31, 0x32, 0x33 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VORR/VORRQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf3, 0xf2, 0xf3, 0xf6, 0xf7, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff3, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff4 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf4, 0xf5, 0xf6, 0xf7, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfffe, 0xffff, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff8, 0xfffffff9 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf6, 0xf7, 0xf6, 0xf7, 0xf6, 0xf7, 0xf6, 0xf7, 0xfe, 0xff, 0xfe, 0xff, 0xfe, 0xff, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff3, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff8, 0xfffffffffffffff9 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xfc, 0xfd, 0xfe, 0xff, 0xfc, 0xfd, 0xfe, 0xff, 0xfc, 0xfd, 0xfe, 0xff, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff7, 0xfff7, 0xfff7, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff7, 0xfffffff7, 0xfffffff7, 0xfffffff7 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff3, 0xfffffffffffffff3 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VORN/VORNQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xfd, 0xfd, 0xff, 0xff, 0xfd, 0xfd, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffd };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffffb };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xfb, 0xfb, 0xfb, 0xfb, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff1, 0xfff3, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff7, 0xfffffff7 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffffd };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf9, 0xf9, 0xfb, 0xfb, 0xfd, 0xfd, 0xff, 0xff, 0xf9, 0xf9, 0xfb, 0xfb, 0xfd, 0xfd, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff7, 0xfff7, 0xfff7, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffd, 0xfffffffd, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff7, 0xfffffffffffffff7 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7, 0xfb, 0xfb, 0xfb, 0xfb, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff, 0xfffc, 0xfffd, 0xfffe, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff8, 0xfffffff9, 0xfffffffa, 0xfffffffb };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffffc, 0xfffffffffffffffd };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VEOR/VEORQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf3, 0xf0, 0xf1, 0xf6, 0xf7, 0xf4, 0xf5 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xc, 0xd, 0xe, 0xf };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff3, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffff94 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xe4, 0xe5, 0xe6, 0xe7, 0xe0, 0xe1, 0xe2, 0xe3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffee, 0xffef, 0xffec, 0xffed };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffd8, 0xffffffd9 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x6, 0x7, 0x4, 0x5, 0x2, 0x3, 0x0, 0x1, 0xe, 0xf, 0xc, 0xd, 0xa, 0xb, 0x8, 0x9 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x1c, 0x1d, 0x1e, 0x1f, 0x18, 0x19, 0x1a, 0x1b };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x12, 0x13, 0x10, 0x11 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffe8, 0xffffffffffffffe9 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xfc, 0xfd, 0xfe, 0xff, 0xf8, 0xf9, 0xfa, 0xfb, 0xf4, 0xf5, 0xf6, 0xf7, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0, 0xfff7, 0xfff6, 0xfff5, 0xfff4 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffc7, 0xffffffc6, 0xffffffc5, 0xffffffc4 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff3, 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VBIC/VBICQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf4, 0xf5, 0xf4, 0xf5 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x1, 0x2, 0x3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffff90 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xe0, 0xe1, 0xe2, 0xe3, 0xe0, 0xe1, 0xe2, 0xe3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffe0, 0xffe1, 0xffe0, 0xffe1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffd0, 0xffffffd1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x1, 0x0, 0x1, 0x0, 0x1, 0x0, 0x1, 0x8, 0x9, 0x8, 0x9, 0x8, 0x9, 0x8, 0x9 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0x11, 0x12, 0x13, 0x10, 0x11, 0x12, 0x13 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0x11, 0x10, 0x11 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffe0, 0xffffffffffffffe1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff4, 0xfff4, 0xfff4, 0xfff4 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffc0, 0xffffffc0, 0xffffffc0, 0xffffffc0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VCREATE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xde, 0xbc, 0x9a, 0x78, 0x56, 0x34, 0x12 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xdef0, 0x9abc, 0x5678, 0x1234 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x9abcdef0, 0x12345678 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x123456789abcdef0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xde, 0xbc, 0x9a, 0x78, 0x56, 0x34, 0x12 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xdef0, 0x9abc, 0x5678, 0x1234 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x9abcdef0, 0x12345678 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x123456789abcdef0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xde, 0xbc, 0x9a, 0x78, 0x56, 0x34, 0x12 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xdef0, 0x9abc, 0x5678, 0x1234 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x9abcdef0, 0x12345678 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD2_LANE/VLD2Q_LANE chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+
+VLD2_LANE/VLD2Q_LANE chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xaaaaaaaa, 0xaaaaaaaa };
+
+VLD3_LANE/VLD3Q_LANE chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+
+VLD3_LANE/VLD3Q_LANE chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1, 0xf2, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xaaaaaaaa, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1, 0xf2, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xc1800000, 0xc1700000 };
+
+VLD3_LANE/VLD3Q_LANE chunk 2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1, 0xf2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff2, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xfff0, 0xfff1, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xfff0, 0xfff1, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xfff2, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff1, 0xfff2, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff2, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+
+VLD4_LANE/VLD4Q_LANE chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+
+VLD4_LANE/VLD4Q_LANE chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+
+VLD4_LANE/VLD4Q_LANE chunk 2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+
+VLD4_LANE/VLD4Q_LANE chunk 3 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+
+VLD2_DUP/VLD2Q_DUP chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD2_DUP/VLD2Q_DUP chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD3_DUP/VLD3Q_DUP chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD3_DUP/VLD3Q_DUP chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff2, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff2, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff2, 0xfff0, 0xfff1 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1800000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD3_DUP/VLD3Q_DUP chunk 2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff2, 0xfff0, 0xfff1, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff2, 0xfff0, 0xfff1, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff0, 0xfff1, 0xfff2 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1600000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD4_DUP/VLD4Q_DUP chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD4_DUP/VLD4Q_DUP chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD4_DUP/VLD4Q_DUP chunk 2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VLD4_DUP/VLD4Q_DUP chunk 3 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff3 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff3 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMLA output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xdf, 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x1f8c, 0x1f8d, 0x1f8e, 0x1f8f };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x2bf7, 0x2bf8 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x43ac, 0x43ad };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x43a14e76, 0x43a1ce76 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x4830, 0x4831, 0x4832, 0x4833, 0x4834, 0x4835, 0x4836, 0x4837 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x470f, 0x4710, 0x4711, 0x4712 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xac, 0xad, 0xae, 0xaf, 0xb0, 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xbb };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a, 0x3e0b, 0x3e0c, 0x3e0d, 0x3e0e };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x3620, 0x3621, 0x3622, 0x3623 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x45f0ae15, 0x45f0b615, 0x45f0be15, 0x45f0c615 };
+
+VMLS output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xe054, 0xe055, 0xe056, 0xe057 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffd3e9, 0xffffd3ea };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xc1d9, 0xc1da, 0xc1db, 0xc1dc };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffbc34, 0xffffbc35 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc3b14e76, 0xc3b0ce76 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf, 0xe0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xb7b0, 0xb7b1, 0xb7b2, 0xb7b3, 0xb7b4, 0xb7b5, 0xb7b6, 0xb7b7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffb8d1, 0xffffb8d2, 0xffffb8d3, 0xffffb8d4 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f, 0x40, 0x41, 0x42, 0x43 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xc1d9, 0xc1da, 0xc1db, 0xc1dc, 0xc1dd, 0xc1de, 0xc1df, 0xc1e0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffc9c0, 0xffffc9c1, 0xffffc9c2, 0xffffc9c3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc5f1ae15, 0xc5f1a615, 0xc5f19e15, 0xc5f19615 };
+
+VMUL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0x1, 0x12, 0x23, 0x34, 0x45, 0x56, 0x67 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfde0, 0xfe02, 0xfe24, 0xfe46 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffcd0, 0xfffffd03 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xc0, 0x4, 0x48, 0x8c, 0xd0, 0x14, 0x58, 0x9c };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfab0, 0xfb05, 0xfb5a, 0xfbaf };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffff9a0, 0xfffffa06 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xc0, 0x84, 0x48, 0xc, 0xd0, 0x94, 0x58, 0x1c };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc4053333, 0xc3f9c000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x90, 0x7, 0x7e, 0xf5, 0x6c, 0xe3, 0x5a, 0xd1, 0x48, 0xbf, 0x36, 0xad, 0x24, 0x9b, 0x12, 0x89 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xf780, 0xf808, 0xf890, 0xf918, 0xf9a0, 0xfa28, 0xfab0, 0xfb38 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffff670, 0xfffff709, 0xfffff7a2, 0xfffff83b };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x60, 0xa, 0xb4, 0x5e, 0x8, 0xb2, 0x5c, 0x6, 0xb0, 0x5a, 0x4, 0xae, 0x58, 0x2, 0xac, 0x56 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xf450, 0xf50b, 0xf5c6, 0xf681, 0xf73c, 0xf7f7, 0xf8b2, 0xf96d };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffff340, 0xfffff40c, 0xfffff4d8, 0xfffff5a4 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x60, 0xca, 0x34, 0x9e, 0xc8, 0x62, 0x9c, 0x36, 0x30, 0x9a, 0x64, 0xce, 0x98, 0x32, 0xcc, 0x66 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc4c73333, 0xc4bac000, 0xc4ae4ccd, 0xc4a1d999 };
+
+VMUL_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffc0, 0xffc4, 0xffc8, 0xffcc };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffde0, 0xfffffe02 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xbbc0, 0xc004, 0xc448, 0xc88c };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffface0, 0xffffb212 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc3b66666, 0xc3ab0000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffc0, 0xffc4, 0xffc8, 0xffcc, 0xffd0, 0xffd4, 0xffd8, 0xffdc };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffde0, 0xfffffe02, 0xfffffe24, 0xfffffe46 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xbbc0, 0xc004, 0xc448, 0xc88c, 0xccd0, 0xd114, 0xd558, 0xd99c };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffface0, 0xffffb212, 0xffffb744, 0xffffbc76 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc3b66666, 0xc3ab0000, 0xc39f9999, 0xc3943333 };
+
+VMUL_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfef0, 0xff01, 0xff12, 0xff23 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffde0, 0xfffffe02 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfcd0, 0xfd03, 0xfd36, 0xfd69 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffbc0, 0xfffffc04 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc3b26666, 0xc3a74000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfab0, 0xfb05, 0xfb5a, 0xfbaf, 0xfc04, 0xfc59, 0xfcae, 0xfd03 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffff9a0, 0xfffffa06, 0xfffffa6c, 0xfffffad2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xf890, 0xf907, 0xf97e, 0xf9f5, 0xfa6c, 0xfae3, 0xfb5a, 0xfbd1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffff780, 0xfffff808, 0xfffff890, 0xfffff918 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc4b1cccd, 0xc4a6b000, 0xc49b9333, 0xc4907667 };
+
+VMULL_N output:
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x11000, 0x11000, 0x11000, 0x11000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x22000, 0x22000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33000, 0x33000, 0x33000, 0x33000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x44000, 0x44000 };
+
+VMLA_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x3e07, 0x3e08 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x3e07, 0x3e08 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x4418c687, 0x44190687 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a, 0x3e0b, 0x3e0c, 0x3e0d, 0x3e0e };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a, 0x3e0b, 0x3e0c, 0x3e0d, 0x3e0e };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x441a3168, 0x441a7168, 0x441ab168, 0x441af168 };
+
+VMLS_LANE output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xc1d9, 0xc1da, 0xc1db, 0xc1dc };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffc1d9, 0xffffc1da };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xc1d9, 0xc1da, 0xc1db, 0xc1dc };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffc1d9, 0xffffc1da };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc420c687, 0xc4208687 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xc1d9, 0xc1da, 0xc1db, 0xc1dc, 0xc1dd, 0xc1de, 0xc1df, 0xc1e0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xc1d9, 0xc1da, 0xc1db, 0xc1dc, 0xc1dd, 0xc1de, 0xc1df, 0xc1e0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc4223168, 0xc421f168, 0xc421b168, 0xc4217168 };
+
+VMLA_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x595, 0x596, 0x597, 0x598 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xb3a, 0xb3b };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x10df, 0x10e0, 0x10e1, 0x10e2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1684, 0x1685 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x4497deb8, 0x4497feb8 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x1c29, 0x1c2a, 0x1c2b, 0x1c2c, 0x1c2d, 0x1c2e, 0x1c2f, 0x1c30 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x21ce, 0x21cf, 0x21d0, 0x21d1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x2773, 0x2774, 0x2775, 0x2776, 0x2777, 0x2778, 0x2779, 0x277a };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x2d18, 0x2d19, 0x2d1a, 0x2d1b };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x4568087b, 0x4568187b, 0x4568287b, 0x4568387b };
+
+VMLS_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfa4b, 0xfa4c, 0xfa4d, 0xfa4e };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffff4a6, 0xfffff4a7 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xef01, 0xef02, 0xef03, 0xef04 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffe95c, 0xffffe95d };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc49bdeb8, 0xc49bbeb8 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xe3b7, 0xe3b8, 0xe3b9, 0xe3ba, 0xe3bb, 0xe3bc, 0xe3bd, 0xe3be };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffde12, 0xffffde13, 0xffffde14, 0xffffde15 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xd86d, 0xd86e, 0xd86f, 0xd870, 0xd871, 0xd872, 0xd873, 0xd874 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffd2c8, 0xffffd2c9, 0xffffd2ca, 0xffffd2cb };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc56a087b, 0xc569f87b, 0xc569e87b, 0xc569d87b };
+
+VSLI_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffe0, 0xffe1, 0xffe2, 0xffe3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x6, 0x7 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x64fffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x50, 0x51, 0x52, 0x53, 0x50, 0x51, 0x52, 0x53 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x7bf0, 0x7bf1, 0x7bf2, 0x7bf3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x3ffffff0, 0x3ffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x10 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x50, 0x51, 0x52, 0x53, 0x50, 0x51, 0x52, 0x53 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x7bf0, 0x7bf1, 0x7bf2, 0x7bf3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xff60, 0xff61, 0xff62, 0xff63, 0xff64, 0xff65, 0xff66, 0xff67 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfe2ffff0, 0xfe2ffff1, 0xfe2ffff2, 0xfe2ffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x18fff0, 0x18fff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3ff0, 0x3ff1, 0x3ff2, 0x3ff3, 0x3ff4, 0x3ff5, 0x3ff6, 0x3ff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1bfffff0, 0x1bfffff1, 0x1bfffff2, 0x1bfffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7ffffffffffff0, 0x7ffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3ff0, 0x3ff1, 0x3ff2, 0x3ff3, 0x3ff4, 0x3ff5, 0x3ff6, 0x3ff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSLI_Nmax shift amount output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7ff0, 0x7ff1, 0x7ff2, 0x7ff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x7ffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x7ff0, 0x7ff1, 0x7ff2, 0x7ff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x7ffffff0, 0x7ffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x7ffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x7ff0, 0x7ff1, 0x7ff2, 0x7ff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x7ff0, 0x7ff1, 0x7ff2, 0x7ff3, 0x7ff4, 0x7ff5, 0x7ff6, 0x7ff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7ffffff0, 0x7ffffff1, 0x7ffffff2, 0x7ffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x7ffffffffffffff0, 0x7ffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSRI_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x80000001, 0x80000001 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffff00000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffc0, 0xffc0, 0xffc0, 0xffc0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xe000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xffc0, 0xffc0, 0xffc0, 0xffc0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffd, 0xfffd, 0xfffd, 0xfffd, 0xfffd, 0xfffd, 0xfffd, 0xfffd };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffff000000000000, 0xffff000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffe00, 0xfffffe00, 0xfffffe00, 0xfffffe00 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffff800, 0xfffffffffffff800 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSRI_N max shift amount output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTST/VTSTQ (signed input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0xffff, 0x0, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0xffff, 0x0, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0xffffffff, 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTST/VTSTQ (unsigned input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0xffff, 0x0, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0xffff, 0x0, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0xffffffff, 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VADDHN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x32, 0x32, 0x32, 0x32, 0x32, 0x32, 0x32, 0x32 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x32, 0x32, 0x32, 0x32 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x18, 0x18 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x37, 0x37, 0x37, 0x37 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x3, 0x3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRADDHN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x19, 0x19 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x38, 0x38, 0x38, 0x38 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x4, 0x4 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VADDL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe3, 0xffe4, 0xffe5, 0xffe6, 0xffe7, 0xffe8, 0xffe9, 0xffea };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe2, 0xffffffe3, 0xffffffe4, 0xffffffe5 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffe0, 0xffffffffffffffe1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1e3, 0x1e4, 0x1e5, 0x1e6, 0x1e7, 0x1e8, 0x1e9, 0x1ea };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1ffe1, 0x1ffe2, 0x1ffe3, 0x1ffe4 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1ffffffe0, 0x1ffffffe1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VADDW output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe3, 0xffe4, 0xffe5, 0xffe6, 0xffe7, 0xffe8, 0xffe9, 0xffea };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe2, 0xffffffe3, 0xffffffe4, 0xffffffe5 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffe0, 0xffffffffffffffe1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffe1, 0xffe2, 0xffe3, 0xffe4 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe0, 0xffffffe1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VHADD/VHADDQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf2, 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff1, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf1, 0xf2, 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff1, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5, 0xf5, 0xf6, 0xf6, 0xf7, 0xf7, 0xf8, 0xf8, 0xf9, 0xf9 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff2, 0xfff2, 0xfff3, 0xfff3, 0xfff4, 0xfff4, 0xfff5 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff1, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf4, 0xf5, 0xf5, 0xf6, 0xf6, 0xf7, 0xf7, 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff1, 0xfff1, 0xfff2, 0xfff2, 0xfff3, 0xfff3, 0xfff4, 0xfff4 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff1, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRHADD/VRHADDQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff2, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf2, 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff1, 0xfff2, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5, 0xf5, 0xf6, 0xf6, 0xf7, 0xf7, 0xf8, 0xf8, 0xf9, 0xf9, 0xfa };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff2, 0xfff3, 0xfff3, 0xfff4, 0xfff4, 0xfff5, 0xfff5 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf5, 0xf5, 0xf6, 0xf6, 0xf7, 0xf7, 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc, 0xfc };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff1, 0xfff2, 0xfff2, 0xfff3, 0xfff3, 0xfff4, 0xfff4, 0xfff5 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff2, 0xfffffff2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VHSUB/VHSUBQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xfe, 0xff, 0xff, 0x0, 0x0, 0x1, 0x1, 0x2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffff, 0xffff, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xfe, 0xff, 0xff, 0x0, 0x0, 0x1, 0x1, 0x2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0x0, 0x0, 0x1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xfe, 0xfe, 0xff, 0xff, 0x0, 0x0, 0x1, 0x1, 0x2, 0x2, 0x3, 0x3, 0x4, 0x4, 0x5, 0x5 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffe, 0xffff, 0xffff, 0x0, 0x0, 0x1, 0x1, 0x2 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0x0, 0x0, 0x1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xfb, 0xfc, 0xfc, 0xfd, 0xfd, 0xfe, 0xfe, 0xff, 0xff, 0x0, 0x0, 0x1, 0x1, 0x2, 0x2, 0x3 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0x0, 0x0, 0x1, 0x1, 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0x0, 0x0, 0x1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSUBL output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffd, 0xfffe, 0xffff, 0x0, 0x1, 0x2, 0x3, 0x4 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffe, 0xffffffff, 0x0, 0x1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfffd, 0xfffe, 0xffff, 0x0, 0x1, 0x2, 0x3, 0x4 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0x0, 0x1, 0x2 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSUBW output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffd, 0xfffe, 0xffff, 0x0, 0x1, 0x2, 0x3, 0x4 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffe, 0xffffffff, 0x0, 0x1 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfefd, 0xfefe, 0xfeff, 0xff00, 0xff01, 0xff02, 0xff03, 0xff04 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffeffff, 0xffff0000, 0xffff0001, 0xffff0002 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffff00000000, 0xffffffff00000001 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSUBHN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x31, 0x31, 0x31, 0x31, 0x31, 0x31, 0x31, 0x31 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x31, 0x31, 0x31, 0x31 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x17, 0x17 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x36, 0x36, 0x36, 0x36 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSUBHN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x31, 0x31, 0x31, 0x31, 0x31, 0x31, 0x31, 0x31 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x31, 0x31, 0x31, 0x31 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x17, 0x17 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x36, 0x36, 0x36, 0x36 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VMVN/VMVNQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xf, 0xe, 0xd, 0xc };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xf, 0xe };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xf, 0xe, 0xd, 0xc };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xf, 0xe };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xf, 0xe, 0xd, 0xc };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xf, 0xe, 0xd, 0xc };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1, 0x0 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+
+VQMOVN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x12, 0x12, 0x12, 0x12, 0x12, 0x12, 0x12, 0x12 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x1278, 0x1278, 0x1278, 0x1278 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x12345678, 0x12345678 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, 0x82, 0x82 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x8765, 0x8765, 0x8765, 0x8765 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x87654321, 0x87654321 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+
+VQMOVN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
+
+VQMOVUN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x34, 0x34, 0x34, 0x34, 0x34, 0x34, 0x34, 0x34 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x5678, 0x5678, 0x5678, 0x5678 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x12345678, 0x12345678 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
+
+VQMOVUN (negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHR_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffc };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x3c, 0x3c, 0x3d, 0x3d, 0x3d, 0x3d, 0x3e, 0x3e };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1ffe, 0x1ffe, 0x1ffe, 0x1ffe };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x8000000, 0x8000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x80000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc, 0xfc, 0xfd, 0xfd, 0xfe, 0xfe, 0xff, 0xff, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffc, 0xfffffffc, 0xfffffffd, 0xfffffffd };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x3c, 0x3c, 0x3d, 0x3d, 0x3d, 0x3d, 0x3e, 0x3e, 0x3e, 0x3e, 0x3f, 0x3f, 0x3f, 0x3f, 0x40, 0x40 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1ffe, 0x1ffe, 0x1ffe, 0x1ffe, 0x1fff, 0x1fff, 0x1fff, 0x1fff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x8000000, 0x8000000, 0x8000000, 0x8000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHR_N (overflow test: max shift amount, positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHR_N (overflow test: shift by 1, with negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x4000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x4000000000000000, 0x4000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHR_N (overflow test: shift by 3, positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x1000, 0x1000, 0x1000, 0x1000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x10000000, 0x10000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x1000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10000000, 0x10000000, 0x10000000, 0x10000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x1000000000000000, 0x1000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x20000000, 0x20000000, 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHR_N (overflow test: shift by 1, with negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xc000, 0xc000, 0xc000, 0xc000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xc0000000, 0xc0000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xc000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xc0000000, 0xc0000000, 0xc0000000, 0xc0000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xc000000000000000, 0xc000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSHR_N (overflow test: shift by 3, with negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xf000, 0xf000, 0xf000, 0xf000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xf0000000, 0xf0000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xf000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xf0000000, 0xf0000000, 0xf0000000, 0xf0000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xf000000000000000, 0xf000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x20000000, 0x20000000, 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSRA_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffd, 0xfffffffe };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfffd, 0xfffe, 0xffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff4, 0xfffffff5 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffd, 0xfffffffe, 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf, 0x10, 0x11, 0x12, 0x13, 0x14 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfffd, 0xfffe, 0xffff, 0x0, 0x1, 0x2, 0x3, 0x4 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSRA_N (checking overflow: shift by 1, positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x4000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x4000000000000000, 0x4000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSRA_N (checking overflow: shift by 3, positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x1000, 0x1000, 0x1000, 0x1000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x10000000, 0x10000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x1000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x10000000, 0x10000000, 0x10000000, 0x10000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x1000000000000000, 0x1000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000, 0x2000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x20000000, 0x20000000, 0x20000000, 0x20000000 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSRA_N (checking overflow: shift by max, positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSRA_N (checking overflow: shift by 1, negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xc000, 0xc000, 0xc000, 0xc000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xc0000000, 0xc0000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xc000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xc0000000, 0xc0000000, 0xc0000000, 0xc0000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xc000000000000000, 0xc000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSRA_N (checking overflow: shift by max, negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xf000, 0xf000, 0xf000, 0xf000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xf0000000, 0xf0000000 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xf000000000000000 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xf0000000, 0xf0000000, 0xf0000000, 0xf0000000 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xf000000000000000, 0xf000000000000000 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VRSRA_N (checking overflow: shift by max, negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VSHLL_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe0, 0xffe2, 0xffe4, 0xffe6, 0xffe8, 0xffea, 0xffec, 0xffee };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe0, 0xffffffe2, 0xffffffe4, 0xffffffe6 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffff80, 0xffffffffffffff88 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3c0, 0x3c4, 0x3c8, 0x3cc, 0x3d0, 0x3d4, 0x3d8, 0x3dc };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfff00, 0xfff10, 0xfff20, 0xfff30 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7ffffff80, 0x7ffffff88 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VPADDL/VPADDLQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffe1, 0xffe5, 0xffe9, 0xffed };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffffe1, 0xffffffe5 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffe1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1e1, 0x1e5, 0x1e9, 0x1ed };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1ffe1, 0x1ffe5 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1ffffffe1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe1, 0xffe5, 0xffe9, 0xffed, 0xfff1, 0xfff5, 0xfff9, 0xfffd };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe1, 0xffffffe5, 0xffffffe9, 0xffffffed };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffe1, 0xffffffffffffffe5 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1e1, 0x1e5, 0x1e9, 0x1ed, 0x1f1, 0x1f5, 0x1f9, 0x1fd };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1ffe1, 0x1ffe5, 0x1ffe9, 0x1ffed };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1ffffffe1, 0x1ffffffe5 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VPADD output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xe1, 0xe5, 0xe9, 0xed, 0xe1, 0xe5, 0xe9, 0xed };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffe1, 0xffe5, 0xffe1, 0xffe5 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffffe1, 0xffffffe1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xe1, 0xe5, 0xe9, 0xed, 0xe1, 0xe5, 0xe9, 0xed };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffe1, 0xffe5, 0xffe1, 0xffe5 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffe1, 0xffffffe1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1f80000, 0xc1f80000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VPADAL/VPADALQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xffd1, 0xffd6, 0xffdb, 0xffe0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xffffffd1, 0xffffffd6 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0xffffffffffffffd1 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x1d1, 0x1d6, 0x1db, 0x1e0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x1ffd1, 0x1ffd6 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1ffffffd1 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xffd1, 0xffd6, 0xffdb, 0xffe0, 0xffe5, 0xffea, 0xffef, 0xfff4 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffd1, 0xffffffd6, 0xffffffdb, 0xffffffe0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0xffffffffffffffd1, 0xffffffffffffffd6 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x1d1, 0x1d6, 0x1db, 0x1e0, 0x1e5, 0x1ea, 0x1ef, 0x1f4 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1ffd1, 0x1ffd6, 0x1ffdb, 0x1ffe0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1ffffffd1, 0x1ffffffd6 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQSHLU_N/VQSHLUQ_N (negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQSHLU_N/VQSHLUQ_N (check cumulative saturation: shift by 1) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfffe, 0xfffe, 0xfffe, 0xfffe };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffffe };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfffe, 0xfffe, 0xfffe, 0xfffe, 0xfffe, 0xfffe, 0xfffe, 0xfffe };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffffe, 0xfffffffe, 0xfffffffe, 0xfffffffe };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffffe, 0xfffffffffffffffe };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 1;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 1;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQSHLU_N/VQSHLUQ_N (check cumulative saturation: shift by 2) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
+int VECT_VAR(expected_cumulative_sat,int,8,16) = 0;
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQSHLU_N/VQSHLUQ_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x8, 0x8, 0x8, 0x8 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x18, 0x18 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x40 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0, 0xa0 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x180, 0x180, 0x180, 0x180, 0x180, 0x180, 0x180, 0x180 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x380, 0x380, 0x380, 0x380 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x800, 0x800 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VCLZ/VCLZQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3, 0x3, 0x3, 0x3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x11, 0x11 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x5, 0x5 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x3, 0x3, 0x3, 0x3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xd, 0xd, 0xd, 0xd, 0xd, 0xd, 0xd, 0xd };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x1f, 0x1f, 0x1f, 0x1f };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VCLZ/VCLZQ (input=0) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x20, 0x20 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x20, 0x20, 0x20, 0x20 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VCLS/VCLSQ (positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x19, 0x19 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x14, 0x14, 0x14, 0x14 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VCLS/VCLSQ (negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x1, 0x1, 0x1, 0x1 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x1, 0x1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VCNT/VCNTQ output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQSHRN_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff8, 0xfff8, 0xfff9, 0xfff9 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffffc, 0xfffffffc };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 1;
+
+VQSHRN_N (check saturation: shift by 3) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,uint,64,2) = 0;
+
+VQSHRN_N (check saturation: shift by max) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VPMAX output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf3, 0xf5, 0xf7, 0xf1, 0xf3, 0xf5, 0xf7 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff1, 0xfff3, 0xfff1, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf1, 0xf3, 0xf5, 0xf7, 0xf1, 0xf3, 0xf5, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff1, 0xfff3, 0xfff1, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff1, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VPMIN output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf2, 0xf4, 0xf6, 0xf0, 0xf2, 0xf4, 0xf6 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff2, 0xfff0, 0xfff2 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf2, 0xf4, 0xf6, 0xf0, 0xf2, 0xf4, 0xf6 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff2, 0xfff0, 0xfff2 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQSHRUN_N (negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQSHRUN_N (check cumulative saturation) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQSHRUN_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x48, 0x48, 0x48, 0x48, 0x48, 0x48, 0x48, 0x48 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xdeadbe, 0xdeadbe };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQRSHRUN_N (negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQRSHRUN_N (check cumulative saturation: shift by 1) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xffff, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQRSHRUN_N (check cumulative saturation: shift by max, positive input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x8000, 0x8000, 0x8000, 0x8000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
+
+VQRSHRUN_N (check cumulative saturation: shift by max, negative input) output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
+int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
+int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
+
+VQRSHRUN_N output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x49, 0x49, 0x49, 0x49, 0x49, 0x49, 0x49, 0x49 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xdeadbf, 0xdeadbf };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VST2_LANE/VST2Q_LANE chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0x0, 0x0 };
+
+VST2_LANE/VST2Q_LANE chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VST3_LANE/VST3Q_LANE chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0x0 };
+
+VST3_LANE/VST3Q_LANE chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0x0 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VST3_LANE/VST3Q_LANE chunk 2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VST4_LANE/VST4Q_LANE chunk 0 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff0, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+
+VST4_LANE/VST4Q_LANE chunk 1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xfffffff2, 0xfffffff3 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VST4_LANE/VST4Q_LANE chunk 2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VST4_LANE/VST4Q_LANE chunk 3 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VTBL1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0xf2, 0xf2, 0xf2, 0x0, 0x0, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0xf3, 0xf3, 0xf3, 0x0, 0x0, 0xf3, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0xf3, 0xf3, 0xf3, 0x0, 0x0, 0xf3, 0xf3 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTBL2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf6, 0xf3, 0xf3, 0xf3, 0x0, 0x0, 0xf3, 0xf3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf6, 0xf5, 0xf5, 0xf5, 0x0, 0x0, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf6, 0xf5, 0xf5, 0xf5, 0x0, 0x0, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTBL3 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf4, 0xf4, 0xf4, 0xff, 0x0, 0xf4, 0xf4 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf8, 0xf7, 0xf7, 0xf7, 0xff, 0x0, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf7, 0xf7, 0xf7, 0xff, 0x0, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTBL4 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xfa, 0xf5, 0xf5, 0xf5, 0x3, 0x0, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xfa, 0xf9, 0xf9, 0xf9, 0x3, 0x0, 0xf9, 0xf9 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xfa, 0xf9, 0xf9, 0xf9, 0x3, 0x0, 0xf9, 0xf9 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTBX1 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0xf2, 0xf2, 0xf2, 0x33, 0x33, 0xf2, 0xf2 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xcc, 0xf3, 0xf3, 0xf3, 0xcc, 0xcc, 0xf3, 0xf3 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xcc, 0xf3, 0xf3, 0xf3, 0xcc, 0xcc, 0xf3, 0xf3 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTBX2 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf6, 0xf3, 0xf3, 0xf3, 0x33, 0x33, 0xf3, 0xf3 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf6, 0xf5, 0xf5, 0xf5, 0xcc, 0xcc, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf6, 0xf5, 0xf5, 0xf5, 0xcc, 0xcc, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTBX3 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf4, 0xf4, 0xf4, 0xff, 0x33, 0xf4, 0xf4 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xf8, 0xf7, 0xf7, 0xf7, 0xff, 0xcc, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf7, 0xf7, 0xf7, 0xff, 0xcc, 0xf7, 0xf7 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+
+VTBX4 output:
+VECT_VAR_DECL(expected,int,8,8) [] = { 0xfa, 0xf5, 0xf5, 0xf5, 0x3, 0x33, 0xf5, 0xf5 };
+VECT_VAR_DECL(expected,int,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,8) [] = { 0xfa, 0xf9, 0xf9, 0xf9, 0x3, 0xcc, 0xf9, 0xf9 };
+VECT_VAR_DECL(expected,uint,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,8) [] = { 0xfa, 0xf9, 0xf9, 0xf9, 0x3, 0xcc, 0xf9, 0xf9 };
+VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,int,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,uint,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,uint,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333 };
+VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
+VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x3f068000, 0x3f068000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x3c030000, 0x3c030000, 0x3c030000, 0x3c030000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xee800000, 0xee800000, 0xee800000, 0xee800000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xbdcc8000, 0xbdcc8000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xbc030000, 0xbc030000, 0xbc030000, 0xbc030000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x7f800000, 0x7f800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xff800000, 0xff800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x80000000, 0x80000000 };
+
+VRSQRTE/VRSQRTEQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x9c800000, 0x9c800000, 0x9c800000, 0x9c800000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x3e498000, 0x3e498000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x3e700000, 0x3e700000, 0x3e700000, 0x3e700000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x80000000, 0x80000000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xae800000, 0xae800000, 0xae800000, 0xae800000 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xb4800000, 0xb4800000 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xed000000, 0xed000000, 0xed000000, 0xed000000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xff800000, 0xff800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
+
+VCAGE/VCAGEQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0x0 };
+
+VCAGE/VCAGEQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+
+VCALE/VCALEQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xffffffff, 0xffffffff };
+
+VCALE/VCALEQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VCAGT/VCAGTQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0x0, 0x0 };
+
+VCAGT/VCAGTQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
+
+VCALT/VCALTQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0xffffffff };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0xffffffff };
+
+VCALT/VCALTQ output:
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+
+VCVT/VCVTQ output:
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x4f800000, 0x4f800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x41720000, 0xc1720000, 0x0, 0x80000000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0x5 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x5 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0xf, 0xfffffff1 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xf, 0x0 };
+
+VCVT_N/VCVTQ_N output:
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc0800000, 0xc0700000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x4c000000, 0x4c000000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xb2800000, 0xb2700000, 0xb2600000, 0xb2500000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x49800000, 0x49800000, 0x49800000, 0x49800000 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xff0b3333, 0x54cccd };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x15 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x1e3d7, 0xfffe1c29 };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0x1e, 0x0 };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xa, 0xa };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xa, 0xa };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7d, 0x7d, 0x7d, 0x7d };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0x7d, 0x7d, 0x7d, 0x7d };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0xa66666, 0xa66666 };
+VECT_VAR_DECL(expected,uint,32,2) [] = { 0xa66666, 0xa66666 };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0xfbccc, 0xfbccc, 0xfbccc, 0xfbccc };
+VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfbccc, 0xfbccc, 0xfbccc, 0xfbccc };
+VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x38640000, 0x7f800000, 0x7fc00000, 0xff800000 };
+
+VRECPS/VRECPSQ output:
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc2e19eb7, 0xc2e19eb7 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1db851f, 0xc1db851f, 0xc1db851f, 0xc1db851f };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xff800000, 0xff800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x40000000, 0x40000000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
+
+VRSQRTS/VRSQRTSQ output:
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc2796b84, 0xc2796b84 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc0e4a3d8, 0xc0e4a3d8, 0xc0e4a3d8, 0xc0e4a3d8 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xff800000, 0xff800000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x3fc00000, 0x3fc00000 };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000 };
diff --git a/expected_input4gcc.txt b/expected_input4gcc.txt
index 0205069..df294b9 100644
--- a/expected_input4gcc.txt
+++ b/expected_input4gcc.txt
@@ -11,6 +11,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -22,6 +23,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80, 0xca00, 0xc980, 0xc900, 0xc880 };
 
 VADD/VADDQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9 };
@@ -35,6 +37,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffdc, 0xffdd, 0xffde, 0xffdf, 0xffe0, 0xffe1, 0xffe2, 0xffe3 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffd2, 0xffffffd3, 0xffffffd4, 0xffffffd5 };
@@ -46,6 +49,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff3, 0xfffffffffffffff4
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x40d9999a, 0x40d9999a };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x41100000, 0x41100000, 0x41100000, 0x41100000 };
 
@@ -61,6 +65,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xaaaa, 0xaaaa, 0xcc00, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xfffffff0, 0xaaaaaaaa };
@@ -72,6 +77,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xaaaaaaaaaaaaaaaa
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xc1800000, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xcc00, 0xaaaa, 0xaaaa };
 
 VLD1_DUP/VLD1_DUPQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
@@ -85,6 +91,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcc00, 0xcc00, 0xcc00 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
@@ -96,6 +103,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1800000, 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcc00, 0xcc00, 0xcc00, 0xcc00, 0xcc00, 0xcc00, 0xcc00 };
 
 VLD1_DUP/VLD1_DUPQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
@@ -109,6 +117,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcb80, 0xcb80, 0xcb80, 0xcb80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
@@ -120,6 +129,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1700000, 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcb80, 0xcb80, 0xcb80, 0xcb80, 0xcb80, 0xcb80, 0xcb80, 0xcb80 };
 
 VLD1_DUP/VLD1_DUPQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
@@ -133,6 +143,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcb00, 0xcb00, 0xcb00, 0xcb00 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
@@ -144,6 +155,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xc1600000, 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcb00, 0xcb00, 0xcb00, 0xcb00, 0xcb00, 0xcb00, 0xcb00, 0xcb00 };
 
 VDUP/VDUPQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
@@ -157,6 +169,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
@@ -168,6 +181,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1800000, 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VDUP/VDUPQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
@@ -181,6 +195,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
@@ -192,6 +207,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1700000, 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VDUP/VDUPQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
@@ -205,6 +221,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
@@ -216,6 +233,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xc1600000, 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMOV/VMOVQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
@@ -229,6 +247,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff0, 0xfffffff0 };
@@ -240,6 +259,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1800000, 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMOV/VMOVQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
@@ -253,6 +273,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
@@ -264,6 +285,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff1, 0xfffffffffffffff1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1700000, 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMOV/VMOVQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
@@ -277,6 +299,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff2, 0xfffffff2, 0xfffffff2 };
@@ -288,6 +311,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff2, 0xfffffffffffffff2
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xc1600000, 0xc1600000, 0xc1600000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VGET_HIGH output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
@@ -301,6 +325,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xca00, 0xc980, 0xc900, 0xc880 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -312,6 +337,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VGET_LOW output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
@@ -325,6 +351,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -336,6 +363,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
 
@@ -351,6 +379,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7c1e, 0x7c1f, 0x7c20, 0x7c21 };
@@ -362,6 +391,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
 
@@ -377,6 +407,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -388,6 +419,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
 
@@ -403,6 +435,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffef, 0x7ffffff0, 0x7ffffff1, 0x7ffffff2 };
@@ -414,6 +447,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
 
@@ -429,6 +463,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffff83c2, 0xffff83c3, 0xffff83c4, 0xffff83c5 };
@@ -440,6 +475,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
 
@@ -455,6 +491,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -466,6 +503,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
 
@@ -481,6 +519,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
@@ -492,6 +531,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
 
@@ -507,6 +547,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x1684, 0x1685, 0x1686, 0x1687 };
@@ -518,6 +559,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
 
@@ -533,6 +575,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffef, 0x7ffffff0, 0x7ffffff1, 0x7ffffff2 };
@@ -544,6 +587,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
 
@@ -559,6 +603,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffe95c, 0xffffe95d, 0xffffe95e, 0xffffe95f };
@@ -570,6 +615,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
 
@@ -585,6 +631,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
@@ -596,6 +643,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VEXT/VEXTQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf7, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
@@ -609,6 +657,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf6, 0xf7, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff3, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0x42066666 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xfe, 0xff, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff7, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff3, 0x33, 0x33, 0x33 };
@@ -620,6 +669,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff1, 0x88 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xfc, 0xfd, 0xfe, 0xff, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff6, 0xfff7, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1500000, 0x4204cccd, 0x4204cccd, 0x4204cccd };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSHR_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb };
@@ -633,6 +683,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x7fffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc, 0xfc, 0xfd, 0xfd, 0xfe, 0xfe, 0xff, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffc, 0xfffffffc, 0xfffffffc, 0xfffffffc };
@@ -644,6 +695,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7fffffff, 0x7fffffff };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSHRN_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb };
@@ -657,6 +709,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -668,6 +721,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHRN_N (with input = 0) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -681,6 +735,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -692,6 +747,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHRN_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc };
@@ -705,6 +761,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -716,6 +773,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHRN_N (with large shift amount) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -729,6 +787,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -740,6 +799,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
@@ -759,6 +819,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -770,6 +831,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
@@ -789,6 +851,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -800,6 +863,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
@@ -819,6 +883,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -830,6 +895,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSET_LANE/VSET_LANEQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0x11 };
@@ -843,6 +909,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x88 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0x55, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0x66, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0x4204cccd };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0x99 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xaa, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xbb };
@@ -854,6 +921,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0x11 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xdd, 0xff };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xee, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0x41333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -883,6 +951,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffff68 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xdf, 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffce, 0xffcf, 0xffd0, 0xffd1, 0xffd2, 0xffd3, 0xffd4, 0xffd5 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffbd, 0xffffffbe, 0xffffffbf, 0xffffffc0 };
@@ -894,6 +963,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffff68, 0xffffffffffffff69
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
 int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
@@ -959,6 +1029,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
@@ -970,6 +1041,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
@@ -987,6 +1059,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -998,6 +1071,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
@@ -1015,6 +1089,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
@@ -1026,6 +1101,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
@@ -1043,6 +1119,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -1054,6 +1131,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
@@ -1081,6 +1159,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -1092,6 +1171,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 
@@ -1107,6 +1187,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x200, 0x1c2, 0x188, 0x152 };
@@ -1118,6 +1199,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
 
@@ -1133,6 +1215,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -1144,6 +1227,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
 
@@ -1159,6 +1243,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7c1e, 0x7c1f, 0x7c20, 0x7c21 };
@@ -1170,6 +1255,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
 
@@ -1185,6 +1271,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffef, 0x7ffffff0, 0x7ffffff1, 0x7ffffff2 };
@@ -1196,6 +1283,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
 
@@ -1211,6 +1299,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffff83c2, 0xffff83c3, 0xffff83c4, 0xffff83c5 };
@@ -1222,6 +1311,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
 
@@ -1237,6 +1327,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
@@ -1248,6 +1339,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VCEQ/VCEQQ output:
 VECT_VAR_DECL(expected,uint,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xff, 0x0 };
@@ -1393,6 +1485,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800004, 0xc1700004 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf6, 0xf6, 0xf6, 0xf6, 0xf2, 0xf2, 0xf2, 0xf2, 0xf6, 0xf6, 0xf6, 0xf6 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2, 0xfff4, 0xfff4, 0xfff6, 0xfff6 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff0, 0xfffffff2, 0xfffffff2 };
@@ -1404,6 +1497,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffff1, 0xfffffff1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7, 0xf3, 0xf3, 0xf3, 0xf3, 0xf7, 0xf7, 0xf7, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff2, 0xfff2, 0xfff4, 0xfff4, 0xfff6, 0xfff6 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800001, 0xc1700001, 0xc1600001, 0xc1500001 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSHL/VSHLQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
@@ -1417,6 +1511,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffff80 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0, 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x1000, 0x2000, 0x3000, 0x4000, 0x5000, 0x6000, 0x7000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x40000000, 0x80000000, 0xc0000000 };
@@ -1428,6 +1523,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x8000000000000000 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSHL/VSHLQ (large shift amount) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -1441,6 +1537,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -1452,6 +1549,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSHL/VSHLQ (negative shift amount) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb };
@@ -1465,6 +1563,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xfc, 0xfc, 0xfc, 0xfc, 0xfd, 0xfd, 0xfd, 0xfd, 0xfe, 0xfe, 0xfe, 0xfe, 0xff, 0xff, 0xff, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffe, 0xfffffffe, 0xfffffffe, 0xfffffffe };
@@ -1476,6 +1575,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7ffffffffffffff, 0x7ffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSHL_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
@@ -1489,6 +1589,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffe0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0, 0x0, 0x20, 0x40, 0x60, 0x80, 0xa0, 0xc0, 0xe0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe0, 0xffe2, 0xffe4, 0xffe6, 0xffe8, 0xffea, 0xffec, 0xffee };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffc0, 0xffffffc4, 0xffffffc8, 0xffffffcc };
@@ -1500,6 +1601,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffe0, 0xffffffffffffffe2
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -1529,6 +1631,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -1540,6 +1643,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -1569,6 +1673,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -1580,6 +1685,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -1609,6 +1715,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1ffffffffffffffe };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
@@ -1620,6 +1727,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -1649,6 +1757,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
@@ -1660,6 +1769,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffff, 0xfffffffffff };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -1689,6 +1799,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
@@ -1700,6 +1811,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -1729,6 +1841,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x7fffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f, 0x3f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3fff, 0x3fff, 0x3fff, 0x3fff, 0x3fff, 0x3fff, 0x3fff, 0x3fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x3fffffff, 0x3fffffff, 0x3fffffff, 0x3fffffff };
@@ -1740,6 +1853,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7fffffffffffffff, 0x7fffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -1769,6 +1883,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -1780,6 +1895,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -1809,6 +1925,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -1820,6 +1937,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -1849,6 +1967,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xc0, 0xc4, 0xc8, 0xcc, 0xd0, 0xd4, 0xd8, 0xdc, 0xe0, 0xe4, 0xe8, 0xec, 0xf0, 0xf4, 0xf8, 0xfc };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe0, 0xffe2, 0xffe4, 0xffe6, 0xffe8, 0xffea, 0xffec, 0xffee };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe0, 0xffffffe2, 0xffffffe4, 0xffffffe6 };
@@ -1860,6 +1979,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -1889,6 +2009,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -1900,6 +2021,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ (with input = 0) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -1913,6 +2035,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -1924,6 +2047,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ (input 0 and negative shift amount) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -1937,6 +2061,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -1948,6 +2073,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee };
@@ -1961,6 +2087,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1ffffffffffffffe };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x1000, 0x2000, 0x3000, 0x4000, 0x5000, 0x6000, 0x7000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -1972,6 +2099,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x8000000000000000 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ (negative shift amount) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc };
@@ -1985,6 +2113,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -1996,6 +2125,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x100000000000, 0x100000000000 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ (checking round_const overflow: shift by -1) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
@@ -2009,6 +2139,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
@@ -2020,6 +2151,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ (checking round_const overflow: shift by -3) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
@@ -2033,6 +2165,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10000000, 0x10000000, 0x10000000, 0x10000000 };
@@ -2044,6 +2177,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ (checking negative shift amount as large as input vector width) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -2057,6 +2191,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -2068,6 +2203,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ (large shift amount) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -2081,6 +2217,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -2092,6 +2229,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHL/VRSHLQ (large negative shift amount) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -2105,6 +2243,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -2116,6 +2255,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD2/VLD2Q chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
@@ -2129,6 +2269,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -2140,6 +2281,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80, 0xca00, 0xc980, 0xc900, 0xc880 };
 
 VLD2/VLD2Q chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
@@ -2153,6 +2295,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xca00, 0xc980, 0xc900, 0xc880 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
@@ -2164,6 +2307,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1400000, 0xc1300000, 0xc1200000, 0xc1100000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xc800, 0xc700, 0xc600, 0xc500, 0xc400, 0xc200, 0xc000, 0xbc00 };
 
 VLD3/VLD3Q chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
@@ -2177,6 +2321,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -2188,6 +2333,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80, 0xca00, 0xc980, 0xc900, 0xc880 };
 
 VLD3/VLD3Q chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
@@ -2201,6 +2347,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xca00, 0xc980, 0xc900, 0xc880 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
@@ -2212,6 +2359,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1400000, 0xc1300000, 0xc1200000, 0xc1100000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xc800, 0xc700, 0xc600, 0xc500, 0xc400, 0xc200, 0xc000, 0xbc00 };
 
 VLD3/VLD3Q chunk 2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
@@ -2225,6 +2373,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1400000, 0xc1300000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xc800, 0xc700, 0xc600, 0xc500 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff8, 0xfffffff9, 0xfffffffa, 0xfffffffb };
@@ -2236,6 +2385,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1000000, 0xc0e00000, 0xc0c00000, 0xc0a00000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x3c00, 0x4000, 0x4200, 0x4400, 0x4500, 0x4600, 0x4700 };
 
 VLD4/VLD4Q chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
@@ -2249,6 +2399,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -2260,6 +2411,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80, 0xca00, 0xc980, 0xc900, 0xc880 };
 
 VLD4/VLD4Q chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
@@ -2273,6 +2425,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xca00, 0xc980, 0xc900, 0xc880 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff4, 0xfffffff5, 0xfffffff6, 0xfffffff7 };
@@ -2284,6 +2437,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd, 0xfffe, 0xffff };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1400000, 0xc1300000, 0xc1200000, 0xc1100000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xc800, 0xc700, 0xc600, 0xc500, 0xc400, 0xc200, 0xc000, 0xbc00 };
 
 VLD4/VLD4Q chunk 2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
@@ -2297,6 +2451,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff8, 0xfff9, 0xfffa, 0xfffb };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1400000, 0xc1300000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xc800, 0xc700, 0xc600, 0xc500 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff8, 0xfffffff9, 0xfffffffa, 0xfffffffb };
@@ -2308,6 +2463,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1000000, 0xc0e00000, 0xc0c00000, 0xc0a00000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x3c00, 0x4000, 0x4200, 0x4400, 0x4500, 0x4600, 0x4700 };
 
 VLD4/VLD4Q chunk 3 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
@@ -2321,6 +2477,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff3 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1200000, 0xc1100000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xc400, 0xc200, 0xc000, 0xbc00 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffc, 0xfffffffd, 0xfffffffe, 0xffffffff };
@@ -2332,6 +2489,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc0800000, 0xc0400000, 0xc0000000, 0xbf800000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x4800, 0x4880, 0x4900, 0x4980, 0x4a00, 0x4a80, 0x4b00, 0x4b80 };
 
 VDUP_LANE/VDUP_LANEQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1, 0xf1 };
@@ -2345,6 +2503,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2, 0xf2 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
@@ -2356,6 +2515,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5, 0xf5 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1700000, 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 
@@ -2389,6 +2549,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf6, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcb00, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xff, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff5, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0x33333333, 0x33333333, 0x33333333 };
@@ -2400,6 +2561,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xfa, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff4, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xc980, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSUB/VSUBQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xee, 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5 };
@@ -2413,6 +2575,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffee };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xe, 0xf, 0x10, 0x11 };
@@ -2424,6 +2587,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffed, 0xffffffffffffffee
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc00ccccd, 0xc00ccccd };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc00ccccc, 0xc00ccccc, 0xc00ccccc, 0xc00ccccc };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
@@ -2455,6 +2619,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf, 0x10 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x23, 0x24, 0x25, 0x26 };
@@ -2466,6 +2631,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,64,1) = 0;
 int VECT_VAR(expected_cumulative_sat,uint,64,1) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
@@ -2527,6 +2693,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0xf, 0xe, 0xd };
@@ -2538,6 +2705,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x40133333, 0x40133333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x4059999a, 0x4059999a, 0x4059999a, 0x4059999a };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
@@ -2559,6 +2727,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0xf, 0xe, 0xd };
@@ -2570,6 +2739,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -2589,6 +2759,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -2600,6 +2771,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VCOMBINE output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2613,6 +2785,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x22, 0x22, 0x22, 0x22 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x33, 0x33 };
@@ -2624,6 +2797,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0x88 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x66, 0x66, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0x40533333, 0x40533333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80, 0x4b80, 0x4b80, 0x4b80, 0x4b80 };
 
 VMAX/VMAXQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf3, 0xf3, 0xf3, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
@@ -2637,6 +2811,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1780000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -2648,6 +2823,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1680000, 0xc1680000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 };
@@ -2667,6 +2843,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1780000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4, 0xf4 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff3 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff1, 0xfffffff1 };
@@ -2678,6 +2855,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1680000, 0xc1680000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x7fc00000, 0x7fc00000, 0x7fc00000, 0x7fc00000 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 };
@@ -2697,6 +2875,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0xf, 0xe, 0xd };
@@ -2708,6 +2887,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc0133333, 0xc0133333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc059999a, 0xc059999a, 0xc059999a, 0xc059999a };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
@@ -2729,6 +2909,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0xf, 0xe, 0xd };
@@ -2740,6 +2921,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -2759,6 +2941,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -2770,6 +2953,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLAL output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2783,6 +2967,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xe907, 0xe908, 0xe909, 0xe90a, 0xe90b, 0xe90c, 0xe90d, 0xe90e };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
@@ -2794,6 +2979,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3e07, 0x3e08 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLSL output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2807,6 +2993,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x16d9, 0x16da, 0x16db, 0x16dc, 0x16dd, 0x16de, 0x16df, 0x16e0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
@@ -2818,6 +3005,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffc1d9, 0xffffffffffffc1da
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLAL_LANE output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2831,6 +3019,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
@@ -2842,6 +3031,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3e07, 0x3e08 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLSL_LANE output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2855,6 +3045,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
@@ -2866,6 +3057,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffc1d9, 0xffffffffffffc1da
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLAL_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2879,6 +3071,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x595, 0x596, 0x597, 0x598 };
@@ -2890,6 +3083,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x10df, 0x10e0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLSL_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2903,6 +3097,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffa4b, 0xfffffa4c, 0xfffffa4d, 0xfffffa4e };
@@ -2914,6 +3109,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffef01, 0xffffffffffffef02
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMOVL output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2927,6 +3123,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -2938,6 +3135,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffff0, 0xfffffff1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMOVN output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
@@ -2951,6 +3149,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -2962,6 +3161,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMULL output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2975,6 +3175,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x100, 0xe1, 0xc4, 0xa9, 0x90, 0x79, 0x64, 0x51 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x100, 0xe1, 0xc4, 0xa9 };
@@ -2986,6 +3187,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe000000100, 0xffffffe2000000e1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x5500, 0x5501, 0x5504, 0x5505, 0x5510, 0x5511, 0x5514, 0x5515 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMULL_LANE output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -2999,6 +3201,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x4000, 0x4000, 0x4000, 0x4000 };
@@ -3010,6 +3213,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000, 0x2000 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VREV16 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6 };
@@ -3023,6 +3227,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6, 0xf9, 0xf8, 0xfb, 0xfa, 0xfd, 0xfc, 0xff, 0xfe };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -3034,6 +3239,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf1, 0xf0, 0xf3, 0xf2, 0xf5, 0xf4, 0xf7, 0xf6, 0xf9, 0xf8, 0xfb, 0xfa, 0xfd, 0xfc, 0xff, 0xfe };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VREV32 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4 };
@@ -3047,6 +3253,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4, 0xfb, 0xfa, 0xf9, 0xf8, 0xff, 0xfe, 0xfd, 0xfc };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2, 0xfff5, 0xfff4, 0xfff7, 0xfff6 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -3058,6 +3265,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf3, 0xf2, 0xf1, 0xf0, 0xf7, 0xf6, 0xf5, 0xf4, 0xfb, 0xfa, 0xf9, 0xf8, 0xff, 0xfe, 0xfd, 0xfc };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff0, 0xfff3, 0xfff2, 0xfff5, 0xfff4, 0xfff7, 0xfff6 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VREV64 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0 };
@@ -3071,6 +3279,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0, 0xff, 0xfe, 0xfd, 0xfc, 0xfb, 0xfa, 0xf9, 0xf8 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0, 0xfff7, 0xfff6, 0xfff5, 0xfff4 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff0, 0xfffffff3, 0xfffffff2 };
@@ -3082,6 +3291,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf7, 0xf6, 0xf5, 0xf4, 0xf3, 0xf2, 0xf1, 0xf0, 0xff, 0xfe, 0xfd, 0xfc, 0xfb, 0xfa, 0xf9, 0xf8 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff3, 0xfff2, 0xfff1, 0xfff0, 0xfff7, 0xfff6, 0xfff5, 0xfff4 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1800000, 0xc1500000, 0xc1600000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSRA_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
@@ -3095,6 +3305,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffc, 0xfffffffd, 0xfffffffe, 0xffffffff };
@@ -3106,6 +3317,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTRN/VTRNQ chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0x11, 0x11, 0xf2, 0xf3, 0x11, 0x11 };
@@ -3119,6 +3331,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0x55, 0x55, 0xf2, 0xf3, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0x11, 0x11, 0xf2, 0xf3, 0x11, 0x11, 0xf4, 0xf5, 0x11, 0x11, 0xf6, 0xf7, 0x11, 0x11 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0x22, 0x22, 0xfff2, 0xfff3, 0x22, 0x22 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x33, 0x33 };
@@ -3130,6 +3343,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0x55, 0x55, 0xf2, 0xf3, 0x55, 0x55, 0xf4, 0xf5, 0x55, 0x55, 0xf6, 0xf7, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0x66, 0x66, 0xfff2, 0xfff3, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0x42073333, 0x42073333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTRN/VTRNQ chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf4, 0xf5, 0x11, 0x11, 0xf6, 0xf7, 0x11, 0x11 };
@@ -3143,6 +3357,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf4, 0xf5, 0x55, 0x55, 0xf6, 0xf7, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff3, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x42066666, 0x42066666 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf8, 0xf9, 0x11, 0x11, 0xfa, 0xfb, 0x11, 0x11, 0xfc, 0xfd, 0x11, 0x11, 0xfe, 0xff, 0x11, 0x11 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff4, 0xfff5, 0x22, 0x22, 0xfff6, 0xfff7, 0x22, 0x22 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff3, 0x33, 0x33 };
@@ -3154,6 +3369,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf8, 0xf9, 0x55, 0x55, 0xfa, 0xfb, 0x55, 0x55, 0xfc, 0xfd, 0x55, 0x55, 0xfe, 0xff, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff4, 0xfff5, 0x66, 0x66, 0xfff6, 0xfff7, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xc1500000, 0x42073333, 0x42073333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VUZP/VUZPQ chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
@@ -3167,6 +3383,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -3178,6 +3395,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VUZP/VUZPQ chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
@@ -3191,6 +3409,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x66, 0x66, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x42066666, 0x42066666 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22, 0x22 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33, 0x33, 0x33, 0x33 };
@@ -3202,6 +3421,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x42073333, 0x42073333, 0x42073333, 0x42073333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VZIP/VZIPQ chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf4, 0x11, 0x11, 0xf1, 0xf5, 0x11, 0x11 };
@@ -3215,6 +3435,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf4, 0x55, 0x55, 0xf1, 0xf5, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff2, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf8, 0x11, 0x11, 0xf1, 0xf9, 0x11, 0x11, 0xf2, 0xfa, 0x11, 0x11, 0xf3, 0xfb, 0x11, 0x11 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff4, 0x22, 0x22, 0xfff1, 0xfff5, 0x22, 0x22 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff2, 0x33, 0x33 };
@@ -3226,6 +3447,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf8, 0x55, 0x55, 0xf1, 0xf9, 0x55, 0x55, 0xf2, 0xfa, 0x55, 0x55, 0xf3, 0xfb, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff4, 0x66, 0x66, 0xfff1, 0xfff5, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1600000, 0x42073333, 0x42073333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VZIP/VZIPQ chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf6, 0x11, 0x11, 0xf3, 0xf7, 0x11, 0x11 };
@@ -3239,6 +3461,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf6, 0x55, 0x55, 0xf3, 0xf7, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff3, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x42066666, 0x42066666 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf4, 0xfc, 0x11, 0x11, 0xf5, 0xfd, 0x11, 0x11, 0xf6, 0xfe, 0x11, 0x11, 0xf7, 0xff, 0x11, 0x11 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff6, 0x22, 0x22, 0xfff3, 0xfff7, 0x22, 0x22 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff3, 0x33, 0x33 };
@@ -3250,6 +3473,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf4, 0xfc, 0x55, 0x55, 0xf5, 0xfd, 0x55, 0x55, 0xf6, 0xfe, 0x55, 0x55, 0xf7, 0xff, 0x55, 0x55 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff2, 0xfff6, 0x66, 0x66, 0xfff3, 0xfff7, 0x66, 0x66 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1700000, 0xc1500000, 0x42073333, 0x42073333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VREINTERPRET/VREINTERPRETQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xff, 0xf1, 0xff, 0xf2, 0xff, 0xf3, 0xff };
@@ -3454,6 +3678,28 @@ VECT_VAR_DECL(expected,uint,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc
 VECT_VAR_DECL(expected,uint,64,2) [] = { 0xc1700000c1800000, 0xc1500000c1600000 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x0, 0x0, 0x80, 0xc1, 0x0, 0x0, 0x70, 0xc1, 0x0, 0x0, 0x60, 0xc1, 0x0, 0x0, 0x50, 0xc1 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0xc180, 0x0, 0xc170, 0x0, 0xc160, 0x0, 0xc150 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xfff0, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xfff0, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0xc180, 0x0, 0xc170 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff, 0xfff2, 0xffff, 0xfff3, 0xffff };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xfff0, 0xffff, 0xffff, 0xffff, 0xfff1, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xfff0, 0xffff, 0xfff1, 0xffff, 0xfff2, 0xffff, 0xfff3, 0xffff };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xfff0, 0xffff, 0xffff, 0xffff, 0xfff1, 0xffff, 0xffff, 0xffff };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xf1f0, 0xf3f2, 0xf5f4, 0xf7f6, 0xf9f8, 0xfbfa, 0xfdfc, 0xfffe };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0xc180, 0x0, 0xc170, 0x0, 0xc160, 0x0, 0xc150 };
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0xcc, 0x80, 0xcb, 0x0, 0xcb, 0x80, 0xca };
 VECT_VAR_DECL(expected,int,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,32,2) [] = { 0xcb80cc00, 0xca80cb00 };
@@ -3493,6 +3739,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -3504,6 +3751,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
@@ -3521,6 +3769,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -3532,6 +3781,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
@@ -3549,6 +3799,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -3560,6 +3811,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
@@ -3577,6 +3829,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -3588,6 +3841,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
@@ -3605,6 +3859,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -3616,6 +3871,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
@@ -3633,6 +3889,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -3644,6 +3901,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
@@ -3671,6 +3929,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -3682,6 +3941,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
@@ -3699,6 +3959,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -3710,6 +3971,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -3739,6 +4001,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -3750,6 +4013,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -3779,6 +4043,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -3790,6 +4055,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -3819,6 +4085,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
@@ -3830,6 +4097,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -3859,6 +4127,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -3870,6 +4139,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x100000000000, 0x100000000000 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -3899,6 +4169,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
@@ -3910,6 +4181,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -3939,6 +4211,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10000000, 0x10000000, 0x10000000, 0x10000000 };
@@ -3950,6 +4223,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -3979,6 +4253,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f, 0x7f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff, 0x7fff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
@@ -3990,6 +4265,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -4019,6 +4295,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
@@ -4030,6 +4307,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -4059,6 +4337,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -4070,6 +4349,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -4099,6 +4379,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -4110,6 +4391,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VABA/VABAQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd };
@@ -4123,6 +4405,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x5e, 0x5f, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6a, 0x6b, 0x6c, 0x6d };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xb9c, 0xb9d, 0xb9e, 0xb9f, 0xba0, 0xba1, 0xba2, 0xba3 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x26e0, 0x26e1, 0x26e2, 0x26e3 };
@@ -4134,6 +4417,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VABAL output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -4147,6 +4431,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff6, 0xfff7, 0xfff8, 0xfff9, 0xfffa, 0xfffb, 0xfffc, 0xfffd };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x16, 0x17, 0x18, 0x19 };
@@ -4158,6 +4443,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe7, 0xffffffe8 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VABAL test intermediate overflow output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -4171,6 +4457,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffef, 0xfff0, 0xfff1, 0xfff2 };
@@ -4182,6 +4469,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe7, 0xffffffe8 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VABD/VABDQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x11, 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa };
@@ -4195,6 +4483,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x41c26666, 0x41ba6666 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x1a, 0x19, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0xf, 0xe, 0xd, 0xc, 0xb };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x4, 0x3, 0x2, 0x1, 0x0, 0x1, 0x2, 0x3 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x30, 0x2f, 0x2e, 0x2d };
@@ -4206,6 +4495,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x42407ae1, 0x423c7ae1, 0x42387ae1, 0x42347ae1 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 
@@ -4221,6 +4511,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x11, 0x10, 0xf, 0xe, 0xd, 0xc, 0xb, 0xa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x3, 0x2, 0x1, 0x0 };
@@ -4232,6 +4523,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe8, 0xffffffe9 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VAND/VANDQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x2, 0x2, 0x0, 0x0, 0x2, 0x2 };
@@ -4245,6 +4537,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf2, 0xf2, 0xf4, 0xf4, 0xf6, 0xf6, 0xf0, 0xf0, 0xf2, 0xf2, 0xf4, 0xf4, 0xf6, 0xf6 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe0, 0xffe0, 0xffe0, 0xffe0, 0xffe4, 0xffe4, 0xffe4, 0xffe4 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe0, 0xffffffe0, 0xffffffe2, 0xffffffe2 };
@@ -4256,6 +4549,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VORR/VORRQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf3, 0xf2, 0xf3, 0xf6, 0xf7, 0xf6, 0xf7 };
@@ -4269,6 +4563,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf6, 0xf7, 0xf6, 0xf7, 0xf6, 0xf7, 0xf6, 0xf7, 0xfe, 0xff, 0xfe, 0xff, 0xfe, 0xff, 0xfe, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffc, 0xfffd, 0xfffe, 0xffff, 0xfffc, 0xfffd, 0xfffe, 0xffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xfffffff3, 0xfffffff2, 0xfffffff3 };
@@ -4280,6 +4575,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff3, 0xfffffffffffffff3
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VORN/VORNQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xfd, 0xfd, 0xff, 0xff, 0xfd, 0xfd, 0xff, 0xff };
@@ -4293,6 +4589,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffffd };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf9, 0xf9, 0xfb, 0xfb, 0xfd, 0xfd, 0xff, 0xff, 0xf9, 0xf9, 0xfb, 0xfb, 0xfd, 0xfd, 0xff, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff3, 0xfff3, 0xfff3, 0xfff3, 0xfff7, 0xfff7, 0xfff7, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffd, 0xfffffffd, 0xffffffff, 0xffffffff };
@@ -4304,6 +4601,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffffc, 0xfffffffffffffffd
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VEOR/VEORQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf3, 0xf0, 0xf1, 0xf6, 0xf7, 0xf4, 0xf5 };
@@ -4317,6 +4615,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x6, 0x7, 0x4, 0x5, 0x2, 0x3, 0x0, 0x1, 0xe, 0xf, 0xc, 0xd, 0xa, 0xb, 0x8, 0x9 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x1c, 0x1d, 0x1e, 0x1f, 0x18, 0x19, 0x1a, 0x1b };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x12, 0x13, 0x10, 0x11 };
@@ -4328,6 +4627,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff3, 0xfffffffffffffff2
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VBIC/VBICQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf4, 0xf5, 0xf4, 0xf5 };
@@ -4341,6 +4641,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x1, 0x0, 0x1, 0x0, 0x1, 0x0, 0x1, 0x8, 0x9, 0x8, 0x9, 0x8, 0x9, 0x8, 0x9 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0x11, 0x12, 0x13, 0x10, 0x11, 0x12, 0x13 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10, 0x11, 0x10, 0x11 };
@@ -4352,6 +4653,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff0
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VCREATE output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xde, 0xbc, 0x9a, 0x78, 0x56, 0x34, 0x12 };
@@ -4365,6 +4667,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x123456789abcdef0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xde, 0xbc, 0x9a, 0x78, 0x56, 0x34, 0x12 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xdef0, 0x9abc, 0x5678, 0x1234 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x9abcdef0, 0x12345678 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xdef0, 0x9abc, 0x5678, 0x1234 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4376,6 +4679,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD2_LANE/VLD2Q_LANE chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
@@ -4389,6 +4693,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
@@ -4400,6 +4705,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xcc00, 0xcb80, 0xaaaa, 0xaaaa };
 
 VLD2_LANE/VLD2Q_LANE chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1 };
@@ -4413,6 +4719,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xaaaaaaaa, 0xaaaaaaaa };
@@ -4424,6 +4731,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 
 VLD3_LANE/VLD3Q_LANE chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
@@ -4437,6 +4745,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
@@ -4448,6 +4757,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xcc00, 0xcb80 };
 
 VLD3_LANE/VLD3Q_LANE chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
@@ -4461,6 +4771,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1, 0xf2, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xfffffff0, 0xfffffff1 };
@@ -4472,6 +4783,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcb00, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 
 VLD3_LANE/VLD3Q_LANE chunk 2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1, 0xf2 };
@@ -4485,6 +4797,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xfff0, 0xfff1, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xfff2, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff2, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
@@ -4496,6 +4809,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff1, 0xfff2, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1600000, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 
 VLD4_LANE/VLD4Q_LANE chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
@@ -4509,6 +4823,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
@@ -4520,6 +4835,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 
 VLD4_LANE/VLD4Q_LANE chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
@@ -4533,6 +4849,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
@@ -4544,6 +4861,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 
 VLD4_LANE/VLD4Q_LANE chunk 2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
@@ -4557,6 +4875,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xaa, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -4568,6 +4887,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 
 VLD4_LANE/VLD4Q_LANE chunk 3 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xf0, 0xf1, 0xf2, 0xf3 };
@@ -4581,6 +4901,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa, 0xaa };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
@@ -4592,6 +4913,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa, 0xaaaa };
 
 VLD2_DUP/VLD2Q_DUP chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
@@ -4605,6 +4927,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff0, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcc00, 0xcb80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4616,6 +4939,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD2_DUP/VLD2Q_DUP chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
@@ -4629,6 +4953,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1, 0xf0, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff0, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcc00, 0xcb80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4640,6 +4965,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD3_DUP/VLD3Q_DUP chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1 };
@@ -4653,6 +4979,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xcc00 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4664,6 +4991,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD3_DUP/VLD3Q_DUP chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0 };
@@ -4677,6 +5005,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff1, 0xfff2, 0xfff0, 0xfff1 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcb80, 0xcb00, 0xcc00, 0xcb80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4688,6 +5017,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD3_DUP/VLD3Q_DUP chunk 2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2 };
@@ -4701,6 +5031,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf1, 0xf2, 0xf0, 0xf1, 0xf2, 0xf0, 0xf1, 0xf2 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff2, 0xfff0, 0xfff1, 0xfff2 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1600000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcb00, 0xcc00, 0xcb80, 0xcb00 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4712,6 +5043,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD4_DUP/VLD4Q_DUP chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
@@ -4725,6 +5057,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4736,6 +5069,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD4_DUP/VLD4Q_DUP chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
@@ -4749,6 +5083,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4760,6 +5095,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD4_DUP/VLD4Q_DUP chunk 2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
@@ -4773,6 +5109,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4784,6 +5121,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VLD4_DUP/VLD4Q_DUP chunk 3 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
@@ -4797,6 +5135,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff3 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf0, 0xf1, 0xf2, 0xf3 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -4808,6 +5147,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLA output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xdf, 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6 };
@@ -4821,6 +5161,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x43a14e76, 0x43a1ce76 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x4830, 0x4831, 0x4832, 0x4833, 0x4834, 0x4835, 0x4836, 0x4837 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x470f, 0x4710, 0x4711, 0x4712 };
@@ -4832,6 +5173,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x45f0ae15, 0x45f0b615, 0x45f0be15, 0x45f0c615 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLS output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8 };
@@ -4845,6 +5187,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc3b14e76, 0xc3b0ce76 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf, 0xe0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xb7b0, 0xb7b1, 0xb7b2, 0xb7b3, 0xb7b4, 0xb7b5, 0xb7b6, 0xb7b7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffb8d1, 0xffffb8d2, 0xffffb8d3, 0xffffb8d4 };
@@ -4856,6 +5199,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc5f1ae15, 0xc5f1a615, 0xc5f19e15, 0xc5f19615 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMUL output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0x1, 0x12, 0x23, 0x34, 0x45, 0x56, 0x67 };
@@ -4869,6 +5213,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xc0, 0x84, 0x48, 0xc, 0xd0, 0x94, 0x58, 0x1c };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc4053333, 0xc3f9c000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x90, 0x7, 0x7e, 0xf5, 0x6c, 0xe3, 0x5a, 0xd1, 0x48, 0xbf, 0x36, 0xad, 0x24, 0x9b, 0x12, 0x89 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xf780, 0xf808, 0xf890, 0xf918, 0xf9a0, 0xfa28, 0xfab0, 0xfb38 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffff670, 0xfffff709, 0xfffff7a2, 0xfffff83b };
@@ -4880,6 +5225,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x60, 0xca, 0x34, 0x9e, 0xc8, 0x62, 0x9c, 0x36, 0x30, 0x9a, 0x64, 0xce, 0x98, 0x32, 0xcc, 0x66 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc4c73333, 0xc4bac000, 0xc4ae4ccd, 0xc4a1d999 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMUL_LANE output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -4893,6 +5239,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc3b66666, 0xc3ab0000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffc0, 0xffc4, 0xffc8, 0xffcc, 0xffd0, 0xffd4, 0xffd8, 0xffdc };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffde0, 0xfffffe02, 0xfffffe24, 0xfffffe46 };
@@ -4904,6 +5251,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc3b66666, 0xc3ab0000, 0xc39f9999, 0xc3943333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMUL_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -4917,6 +5265,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc3b26666, 0xc3a74000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfab0, 0xfb05, 0xfb5a, 0xfbaf, 0xfc04, 0xfc59, 0xfcae, 0xfd03 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffff9a0, 0xfffffa06, 0xfffffa6c, 0xfffffad2 };
@@ -4928,6 +5277,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc4b1cccd, 0xc4a6b000, 0xc49b9333, 0xc4907667 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMULL_N output:
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x11000, 0x11000, 0x11000, 0x11000 };
@@ -4947,6 +5297,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x4418c687, 0x44190687 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a, 0x3e0b, 0x3e0c, 0x3e0d, 0x3e0e };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x3e07, 0x3e08, 0x3e09, 0x3e0a };
@@ -4958,6 +5309,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x441a3168, 0x441a7168, 0x441ab168, 0x441af168 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLS_LANE output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -4971,6 +5323,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc420c687, 0xc4208687 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xc1d9, 0xc1da, 0xc1db, 0xc1dc, 0xc1dd, 0xc1de, 0xc1df, 0xc1e0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffc1d9, 0xffffc1da, 0xffffc1db, 0xffffc1dc };
@@ -4982,6 +5335,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc4223168, 0xc421f168, 0xc421b168, 0xc4217168 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLA_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -4995,6 +5349,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x4497deb8, 0x4497feb8 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x1c29, 0x1c2a, 0x1c2b, 0x1c2c, 0x1c2d, 0x1c2e, 0x1c2f, 0x1c30 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x21ce, 0x21cf, 0x21d0, 0x21d1 };
@@ -5006,6 +5361,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x4568087b, 0x4568187b, 0x4568287b, 0x4568387b };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMLS_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5019,6 +5375,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc49bdeb8, 0xc49bbeb8 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xe3b7, 0xe3b8, 0xe3b9, 0xe3ba, 0xe3bb, 0xe3bc, 0xe3bd, 0xe3be };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffde12, 0xffffde13, 0xffffde14, 0xffffde15 };
@@ -5030,6 +5387,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc56a087b, 0xc569f87b, 0xc569e87b, 0xc569d87b };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSLI_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27 };
@@ -5043,6 +5401,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x10 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x50, 0x51, 0x52, 0x53, 0x50, 0x51, 0x52, 0x53 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x7bf0, 0x7bf1, 0x7bf2, 0x7bf3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xff60, 0xff61, 0xff62, 0xff63, 0xff64, 0xff65, 0xff66, 0xff67 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfe2ffff0, 0xfe2ffff1, 0xfe2ffff2, 0xfe2ffff3 };
@@ -5054,6 +5413,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7ffffffffffff0, 0x7ffffffffffff1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3ff0, 0x3ff1, 0x3ff2, 0x3ff3, 0x3ff4, 0x3ff5, 0x3ff6, 0x3ff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSLI_Nmax shift amount output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77 };
@@ -5067,6 +5427,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x7ffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x7ff0, 0x7ff1, 0x7ff2, 0x7ff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x7ff0, 0x7ff1, 0x7ff2, 0x7ff3, 0x7ff4, 0x7ff5, 0x7ff6, 0x7ff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7ffffff0, 0x7ffffff1, 0x7ffffff2, 0x7ffffff3 };
@@ -5078,6 +5439,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSRI_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
@@ -5091,6 +5453,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xe000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5, 0xc5 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xffc0, 0xffc0, 0xffc0, 0xffc0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xf7, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffd, 0xfffd, 0xfffd, 0xfffd, 0xfffd, 0xfffd, 0xfffd, 0xfffd };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff };
@@ -5102,6 +5465,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffff800, 0xfffffffffffff800
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1, 0xe1 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0, 0xfff0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSRI_N max shift amount output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
@@ -5115,6 +5479,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -5126,6 +5491,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTST/VTSTQ (signed input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5139,6 +5505,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5150,6 +5517,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTST/VTSTQ (unsigned input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5163,6 +5531,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5174,6 +5543,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VADDHN output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x32, 0x32, 0x32, 0x32, 0x32, 0x32, 0x32, 0x32 };
@@ -5187,6 +5557,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5198,6 +5569,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRADDHN output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5211,6 +5583,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5222,6 +5595,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VADDL output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5235,6 +5609,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe3, 0xffe4, 0xffe5, 0xffe6, 0xffe7, 0xffe8, 0xffe9, 0xffea };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe2, 0xffffffe3, 0xffffffe4, 0xffffffe5 };
@@ -5246,6 +5621,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1ffffffe0, 0x1ffffffe1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VADDW output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5259,6 +5635,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe3, 0xffe4, 0xffe5, 0xffe6, 0xffe7, 0xffe8, 0xffe9, 0xffea };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe2, 0xffffffe3, 0xffffffe4, 0xffffffe5 };
@@ -5270,6 +5647,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffe0, 0xffffffe1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VHADD/VHADDQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf2, 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5 };
@@ -5283,6 +5661,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5, 0xf5, 0xf6, 0xf6, 0xf7, 0xf7, 0xf8, 0xf8, 0xf9, 0xf9 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff1, 0xfff2, 0xfff2, 0xfff3, 0xfff3, 0xfff4, 0xfff4, 0xfff5 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff1, 0xfffffff2 };
@@ -5294,6 +5673,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRHADD/VRHADDQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf2, 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5, 0xf5 };
@@ -5307,6 +5687,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf2, 0xf3, 0xf3, 0xf4, 0xf4, 0xf5, 0xf5, 0xf6, 0xf6, 0xf7, 0xf7, 0xf8, 0xf8, 0xf9, 0xf9, 0xfa };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff2, 0xfff2, 0xfff3, 0xfff3, 0xfff4, 0xfff4, 0xfff5, 0xfff5 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff1, 0xfffffff1, 0xfffffff2, 0xfffffff2 };
@@ -5318,6 +5699,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VHSUB/VHSUBQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xfe, 0xff, 0xff, 0x0, 0x0, 0x1, 0x1, 0x2 };
@@ -5331,6 +5713,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xfe, 0xfe, 0xff, 0xff, 0x0, 0x0, 0x1, 0x1, 0x2, 0x2, 0x3, 0x3, 0x4, 0x4, 0x5, 0x5 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffe, 0xffff, 0xffff, 0x0, 0x0, 0x1, 0x1, 0x2 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffff, 0x0, 0x0, 0x1 };
@@ -5342,6 +5725,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSUBL output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5355,6 +5739,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffd, 0xfffe, 0xffff, 0x0, 0x1, 0x2, 0x3, 0x4 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffe, 0xffffffff, 0x0, 0x1 };
@@ -5366,6 +5751,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSUBW output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5379,6 +5765,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfffd, 0xfffe, 0xffff, 0x0, 0x1, 0x2, 0x3, 0x4 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffe, 0xffffffff, 0x0, 0x1 };
@@ -5390,6 +5777,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffff00000000, 0xffffffff00000001
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSUBHN output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x31, 0x31, 0x31, 0x31, 0x31, 0x31, 0x31, 0x31 };
@@ -5403,6 +5791,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5414,6 +5803,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSUBHN output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x31, 0x31, 0x31, 0x31, 0x31, 0x31, 0x31, 0x31 };
@@ -5427,6 +5817,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5438,6 +5829,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VMVN/VMVNQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8 };
@@ -5451,6 +5843,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xf, 0xe, 0xd, 0xc };
@@ -5462,6 +5855,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0xf, 0xe, 0xd, 0xc, 0xb, 0xa, 0x9, 0x8, 0x7, 0x6, 0x5, 0x4, 0x3, 0x2, 0x1, 0x0 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -5481,6 +5875,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5492,6 +5887,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -5511,6 +5907,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5522,6 +5919,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,uint,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,uint,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,uint,32,2) = 0;
@@ -5538,6 +5936,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5549,6 +5948,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,uint,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,uint,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,uint,32,2) = 1;
@@ -5565,6 +5965,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5576,6 +5977,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHR_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc };
@@ -5589,6 +5991,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x80000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf8, 0xf9, 0xf9, 0xfa, 0xfa, 0xfb, 0xfb, 0xfc, 0xfc, 0xfd, 0xfd, 0xfe, 0xfe, 0xff, 0xff, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffc, 0xfffffffc, 0xfffffffd, 0xfffffffd };
@@ -5600,6 +6003,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x80000000, 0x80000000 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHR_N (overflow test: max shift amount, positive input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -5613,6 +6017,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -5624,6 +6029,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHR_N (overflow test: shift by 1, with negative input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
@@ -5637,6 +6043,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
@@ -5648,6 +6055,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHR_N (overflow test: shift by 3, positive input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
@@ -5661,6 +6069,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10000000, 0x10000000, 0x10000000, 0x10000000 };
@@ -5672,6 +6081,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHR_N (overflow test: shift by 1, with negative input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0 };
@@ -5685,6 +6095,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xc0000000, 0xc0000000, 0xc0000000, 0xc0000000 };
@@ -5696,6 +6107,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSHR_N (overflow test: shift by 3, with negative input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
@@ -5709,6 +6121,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xf0000000, 0xf0000000, 0xf0000000, 0xf0000000 };
@@ -5720,6 +6133,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSRA_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0 };
@@ -5733,6 +6147,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0xfff4, 0xfff5, 0xfff6, 0xfff7 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffffd, 0xfffffffe, 0xffffffff, 0x0 };
@@ -5744,6 +6159,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffff0, 0xfffffffffffffff1
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSRA_N (checking overflow: shift by 1, positive input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
@@ -5757,6 +6173,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x8000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40, 0x40 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000, 0x4000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x40000000, 0x40000000, 0x40000000, 0x40000000 };
@@ -5768,6 +6185,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x8000000000000000, 0x8000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSRA_N (checking overflow: shift by 3, positive input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
@@ -5781,6 +6199,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x2000000000000000 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000, 0x1000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x10000000, 0x10000000, 0x10000000, 0x10000000 };
@@ -5792,6 +6211,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x2000000000000000, 0x2000000000000000
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSRA_N (checking overflow: shift by max, positive input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -5805,6 +6225,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -5816,6 +6237,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSRA_N (checking overflow: shift by 1, negative input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0 };
@@ -5829,6 +6251,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0, 0xc0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000, 0xc000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xc0000000, 0xc0000000, 0xc0000000, 0xc0000000 };
@@ -5840,6 +6263,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSRA_N (checking overflow: shift by max, negative input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
@@ -5853,6 +6277,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0, 0xf0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000, 0xf000 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xf0000000, 0xf0000000, 0xf0000000, 0xf0000000 };
@@ -5864,6 +6289,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VRSRA_N (checking overflow: shift by max, negative input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -5877,6 +6303,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -5888,6 +6315,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1, 0x1 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VSHLL_N output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5901,6 +6329,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe0, 0xffe2, 0xffe4, 0xffe6, 0xffe8, 0xffea, 0xffec, 0xffee };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe0, 0xffffffe2, 0xffffffe4, 0xffffffe6 };
@@ -5912,6 +6341,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x7ffffff80, 0x7ffffff88 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VPADDL/VPADDLQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5925,6 +6355,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1ffffffe1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffe1, 0xffe5, 0xffe9, 0xffed, 0xfff1, 0xfff5, 0xfff9, 0xfffd };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffe1, 0xffffffe5, 0xffffffe9, 0xffffffed };
@@ -5936,6 +6367,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1ffffffe1, 0x1ffffffe5 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VPADD output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xe1, 0xe5, 0xe9, 0xed, 0xe1, 0xe5, 0xe9, 0xed };
@@ -5949,6 +6381,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1f80000, 0xc1f80000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -5960,6 +6393,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VPADAL/VPADALQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
@@ -5973,6 +6407,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x1ffffffd1 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xffd1, 0xffd6, 0xffdb, 0xffe0, 0xffe5, 0xffea, 0xffef, 0xfff4 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xffffffd1, 0xffffffd6, 0xffffffdb, 0xffffffe0 };
@@ -5984,6 +6419,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x1ffffffd1, 0x1ffffffd6 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -6005,6 +6441,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x0 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6016,6 +6453,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -6037,6 +6475,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xfffffffffffffffe };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6048,6 +6487,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xfffffffffffffffe, 0xfffffffffffffffe
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 1;
@@ -6069,6 +6509,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0xffffffffffffffff };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6080,6 +6521,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0xffffffffffffffff, 0xffffffffffffffff
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,8,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,16,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,2) = 0;
@@ -6101,6 +6543,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x40 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6112,6 +6555,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x800, 0x800 };
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VCLZ/VCLZQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6125,6 +6569,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3, 0x3 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x3, 0x3, 0x3, 0x3 };
@@ -6136,6 +6581,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VCLZ/VCLZQ (input=0) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8 };
@@ -6149,6 +6595,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x20, 0x20, 0x20, 0x20 };
@@ -6160,6 +6607,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VCLS/VCLSQ (positive input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6 };
@@ -6173,6 +6621,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x14, 0x14, 0x14, 0x14 };
@@ -6184,6 +6633,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VCLS/VCLSQ (negative input) output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x7 };
@@ -6197,6 +6647,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -6208,6 +6659,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VCNT/VCNTQ output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8, 0x8 };
@@ -6221,6 +6673,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6232,6 +6685,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6, 0x6 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
@@ -6251,6 +6705,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6262,6 +6717,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
@@ -6281,6 +6737,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6292,6 +6749,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
@@ -6311,6 +6769,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6322,6 +6781,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VPMAX output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf1, 0xf3, 0xf5, 0xf7, 0xf1, 0xf3, 0xf5, 0xf7 };
@@ -6335,6 +6795,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1700000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6346,6 +6807,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VPMIN output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf2, 0xf4, 0xf6, 0xf0, 0xf2, 0xf4, 0xf6 };
@@ -6359,6 +6821,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1800000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6370,6 +6833,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
@@ -6386,6 +6850,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6397,6 +6862,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
@@ -6413,6 +6879,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6424,6 +6891,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
@@ -6440,6 +6908,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6451,6 +6920,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
@@ -6467,6 +6937,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6478,6 +6949,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
@@ -6494,6 +6966,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6505,6 +6978,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 0;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
@@ -6521,6 +6995,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6532,6 +7007,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 1;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 1;
@@ -6548,6 +7024,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6559,6 +7036,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 int VECT_VAR(expected_cumulative_sat,int,16,8) = 0;
 int VECT_VAR(expected_cumulative_sat,int,32,4) = 1;
 int VECT_VAR(expected_cumulative_sat,int,64,2) = 0;
@@ -6575,6 +7053,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6586,6 +7065,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VST2_LANE/VST2Q_LANE chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6599,6 +7079,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0x0, 0x0 };
@@ -6610,6 +7091,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VST2_LANE/VST2Q_LANE chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6623,6 +7105,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -6634,6 +7117,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VST3_LANE/VST3Q_LANE chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6647,6 +7131,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0x0 };
@@ -6658,6 +7143,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0xcb00, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VST3_LANE/VST3Q_LANE chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6671,6 +7157,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -6682,6 +7169,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VST3_LANE/VST3Q_LANE chunk 2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6695,6 +7183,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -6706,6 +7195,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VST4_LANE/VST4Q_LANE chunk 0 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0x0, 0x0, 0x0, 0x0 };
@@ -6719,6 +7209,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf0, 0xf1, 0xf2, 0xf3, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1800000, 0xc1700000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0xfffffff0, 0xfffffff1, 0xfffffff2, 0xfffffff3 };
@@ -6730,6 +7221,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0xfff0, 0xfff1, 0xfff2, 0xfff3, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0xcc00, 0xcb80, 0xcb00, 0xca80, 0x0, 0x0, 0x0, 0x0 };
 
 VST4_LANE/VST4Q_LANE chunk 1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6743,6 +7235,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc1600000, 0xc1500000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -6754,6 +7247,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VST4_LANE/VST4Q_LANE chunk 2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6767,6 +7261,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -6778,6 +7273,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VST4_LANE/VST4Q_LANE chunk 3 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
@@ -6791,6 +7287,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
@@ -6802,6 +7299,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x0, 0x0, 0x0, 0x0 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTBL1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x0, 0xf2, 0xf2, 0xf2, 0x0, 0x0, 0xf2, 0xf2 };
@@ -6815,6 +7313,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0x0, 0xf3, 0xf3, 0xf3, 0x0, 0x0, 0xf3, 0xf3 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6826,6 +7325,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTBL2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf6, 0xf3, 0xf3, 0xf3, 0x0, 0x0, 0xf3, 0xf3 };
@@ -6839,6 +7339,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf6, 0xf5, 0xf5, 0xf5, 0x0, 0x0, 0xf5, 0xf5 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6850,6 +7351,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTBL3 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf4, 0xf4, 0xf4, 0xff, 0x0, 0xf4, 0xf4 };
@@ -6863,6 +7365,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf7, 0xf7, 0xf7, 0xff, 0x0, 0xf7, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6874,6 +7377,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTBL4 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xfa, 0xf5, 0xf5, 0xf5, 0x3, 0x0, 0xf5, 0xf5 };
@@ -6887,6 +7391,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xfa, 0xf9, 0xf9, 0xf9, 0x3, 0x0, 0xf9, 0xf9 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6898,6 +7403,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTBX1 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0x33, 0xf2, 0xf2, 0xf2, 0x33, 0x33, 0xf2, 0xf2 };
@@ -6911,6 +7417,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xcc, 0xf3, 0xf3, 0xf3, 0xcc, 0xcc, 0xf3, 0xf3 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6922,6 +7429,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTBX2 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf6, 0xf3, 0xf3, 0xf3, 0x33, 0x33, 0xf3, 0xf3 };
@@ -6935,6 +7443,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf6, 0xf5, 0xf5, 0xf5, 0xcc, 0xcc, 0xf5, 0xf5 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6946,6 +7455,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTBX3 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xf8, 0xf4, 0xf4, 0xf4, 0xff, 0x33, 0xf4, 0xf4 };
@@ -6959,6 +7469,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xf8, 0xf7, 0xf7, 0xf7, 0xff, 0xcc, 0xf7, 0xf7 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6970,6 +7481,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 
 VTBX4 output:
 VECT_VAR_DECL(expected,int,8,8) [] = { 0xfa, 0xf5, 0xf5, 0xf5, 0x3, 0x33, 0xf5, 0xf5 };
@@ -6983,6 +7495,7 @@ VECT_VAR_DECL(expected,uint,64,1) [] = { 0x3333333333333333 };
 VECT_VAR_DECL(expected,poly,8,8) [] = { 0xfa, 0xf9, 0xf9, 0xf9, 0x3, 0xcc, 0xf9, 0xf9 };
 VECT_VAR_DECL(expected,poly,16,4) [] = { 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,int,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,int,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
@@ -6994,6 +7507,7 @@ VECT_VAR_DECL(expected,uint,64,2) [] = { 0x3333333333333333, 0x3333333333333333
 VECT_VAR_DECL(expected,poly,8,16) [] = { 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33, 0x33 };
 VECT_VAR_DECL(expected,poly,16,8) [] = { 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333, 0x3333 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x33333333, 0x33333333, 0x33333333, 0x33333333 };
+VECT_VAR_DECL(expected,hfloat,16,8) [] = { 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 };
 VECT_VAR_DECL(expected,uint,32,2) [] = { 0xffffffff, 0xffffffff };
 VECT_VAR_DECL(expected,uint,32,4) [] = { 0xbf000000, 0xbf000000, 0xbf000000, 0xbf000000 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x3f068000, 0x3f068000 };
@@ -7066,6 +7580,7 @@ VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1800000, 0xc1700000, 0xc1600000, 0
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x4f800000, 0x4f800000, 0x4f800000, 0x4f800000 };
 VECT_VAR_DECL(expected,int,32,2) [] = { 0xfffffff1, 0x5 };
 VECT_VAR_DECL(expected,uint,32,2) [] = { 0x0, 0x5 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x0, 0x8000, 0x4b8f, 0xcb8f };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x0, 0x0, 0xf, 0xfffffff1 };
 VECT_VAR_DECL(expected,uint,32,4) [] = { 0x0, 0x0, 0xf, 0x0 };
 
@@ -7088,9 +7603,10 @@ VECT_VAR_DECL(expected,int,32,4) [] = { 0xfbccc, 0xfbccc, 0xfbccc, 0xfbccc };
 VECT_VAR_DECL(expected,uint,32,4) [] = { 0xfbccc, 0xfbccc, 0xfbccc, 0xfbccc };
 VECT_VAR_DECL(expected,int,32,2) [] = { 0x7fffffff, 0x7fffffff };
 VECT_VAR_DECL(expected,int,32,4) [] = { 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff };
+VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x38640000, 0x7f800000, 0x7fc00000, 0xff800000 };
+VECT_VAR_DECL(expected,hfloat,16,4) [] = { 0x7e00, 0x7c00, 0xfc00, 0x8000 };
 
 VRECPS/VRECPSQ output:
-VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x38640000, 0x7f800000, 0x7fc00000, 0xff800000 };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0xc2e19eb7, 0xc2e19eb7 };
 VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0xc1db851f, 0xc1db851f, 0xc1db851f, 0xc1db851f };
 VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x7fc00000, 0x7fc00000 };
diff --git a/stm-arm-neon-ref.h b/stm-arm-neon-ref.h
index b9b3fec..f0ff00d 100644
--- a/stm-arm-neon-ref.h
+++ b/stm-arm-neon-ref.h
@@ -150,22 +150,25 @@ static int result_idx = 0;
   for(i=0; i<N ; i++)							\
     {									\
       uint##W##_t tmp;							\
-      tmp = (uint##W##_t)VECT_VAR(result, T, W, N)[i];			\
-      fprintf(ref_file, "%" FMT ", ", tmp);				\
+	tmp = (uint##W##_t)VECT_VAR(result, T, W, N)[i];		\
+	fprintf(ref_file, "%" FMT ", ", tmp);				\
     }									\
-  fprintf(ref_file, " }\n");
+  fprintf(ref_file, " }\n");						\
+  DUMP4GCC_FP16(MSG,T,W,N,FMT);
 
 #define DUMP4GCC_FP16(MSG,T,W,N,FMT)					\
-  if (0) {								\
-    fprintf(gcc_tests_file, "%s:%d:%s [] = { ", MSG, result_idx++,	\
-	    STR(VECT_VAR(result, T, W, N)));				\
-    for(i=0; i<N ; i++)							\
+  {									\
+    uint##W##_t tmp;							\
+    fprintf(gcc_tests_file, "VECT_VAR_DECL(expected,%s,%d,%d) [] = { ", \
+	    "hfloat", W, N);						\
+    for(i=0; i<(N-1) ; i++)						\
       {									\
-	uint##W##_t tmp;						\
 	tmp = (uint##W##_t)VECT_VAR(result, T, W, N)[i];		\
-	fprintf(gcc_tests_file, "%" FMT ", ", tmp);			\
+	fprintf(gcc_tests_file, "0x%" FMT ", ", tmp);			\
       }									\
-    fprintf(gcc_tests_file, " }\n");					\
+    tmp = (uint##W##_t)VECT_VAR(result, T, W, N)[i];			\
+    fprintf(gcc_tests_file, "0x%" FMT, tmp);				\
+    fprintf(gcc_tests_file, " };\n");					\
   }
 
 #define CLEAN_PATTERN_8  0x33