Fix detection of FP16 support: test __ARM_FP16_FORMAT_IEEE instead of __ARM_NEON_FP16_INTRINSICS.

16 files changed, 126 insertions, 128 deletions

diff --git a/compute_ref_data.c b/compute_ref_data.c
index 559be6d..ef7a6e0 100644
--- a/compute_ref_data.c
+++ b/compute_ref_data.c
@@ -143,7 +143,7 @@ VECT_VAR_DECL_INIT(buffer, uint, 64, 1);
 PAD(buffer_pad, uint, 64, 1);
 VECT_VAR_DECL_INIT(buffer, float, 32, 2);
 PAD(buffer_pad, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 /* We need a different initialization for ARMCC, because the compiler
    performs the conversion to half-precision internal
    representation.  */
@@ -177,7 +177,7 @@ VECT_VAR_DECL_INIT(buffer, poly, 16, 8);
 PAD(buffer_pad, poly, 16, 8);
 VECT_VAR_DECL_INIT(buffer, float, 32, 4);
 PAD(buffer_pad, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_float16x8[8] = {-16, -15, -14, -13, -12, -11, -10, -9};
 #else
@@ -214,7 +214,7 @@ VECT_VAR_DECL_INIT(buffer_dup, poly, 16, 4);
 VECT_VAR_DECL(buffer_dup_pad, poly, 16, 4);
 VECT_VAR_DECL_INIT4(buffer_dup, float, 32, 2);
 VECT_VAR_DECL(buffer_dup_pad, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_dup_float16x4[4] = {-16, -15, -14, -13};
 #else
@@ -245,7 +245,7 @@ VECT_VAR_DECL_INIT(buffer_dup, poly, 16, 8);
 VECT_VAR_DECL(buffer_dup_pad, poly, 16, 8);
 VECT_VAR_DECL_INIT(buffer_dup, float, 32, 4);
 VECT_VAR_DECL(buffer_dup_pad, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_dup_float16x8[8] = {-16, -15, -14, -13, -12, -11, -10, -9};
 #else
@@ -280,7 +280,7 @@ VECT_ARRAY_INIT2(buffer_vld2, poly, 16, 4);
 PAD(buffer_vld2_pad, poly, 16, 4);
 VECT_ARRAY_INIT2(buffer_vld2, float, 32, 2);
 PAD(buffer_vld2_pad, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld2_float16x4x2[4*2] = {-16, -15, -14, -13, -12, -11, -10, -9};
 #else
@@ -313,7 +313,7 @@ VECT_ARRAY_INIT2(buffer_vld2, poly, 16, 8);
 PAD(buffer_vld2_pad, poly, 16, 8);
 VECT_ARRAY_INIT2(buffer_vld2, float, 32, 4);
 PAD(buffer_vld2_pad, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld2_float16x8x2[8*2] = {-16, -15, -14, -13, -12, -11, -10, -9,
 				       -8, -7, -6, -5, -4, -3, -2, -1};
@@ -353,7 +353,7 @@ VECT_ARRAY_INIT3(buffer_vld3, poly, 16, 4);
 PAD(buffer_vld3_pad, poly, 16, 4);
 VECT_ARRAY_INIT3(buffer_vld3, float, 32, 2);
 PAD(buffer_vld3_pad, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld3_float16x4x3[4*3] = {-16, -15, -14, -13, -12, -11, -10, -9,
 				       -8, -7, -6, -5};
@@ -389,7 +389,7 @@ VECT_ARRAY_INIT3(buffer_vld3, poly, 16, 8);
 PAD(buffer_vld3_pad, poly, 16, 8);
 VECT_ARRAY_INIT3(buffer_vld3, float, 32, 4);
 PAD(buffer_vld3_pad, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld3_float16x8x3[8*3] = {-16, -15, -14, -13, -12, -11, -10, -9,
 				       -8, -7, -6, -5, -4, -3, -2, -1,
@@ -434,7 +434,7 @@ VECT_ARRAY_INIT4(buffer_vld4, poly, 16, 4);
 PAD(buffer_vld4_pad, poly, 16, 4);
 VECT_ARRAY_INIT4(buffer_vld4, float, 32, 2);
 PAD(buffer_vld4_pad, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld4_float16x4x4[4*4] = {-16, -15, -14, -13, -12, -11, -10, -9,
 				       -8, -7, -6, -5, -4, -3, -2, -1};
@@ -472,7 +472,7 @@ VECT_ARRAY_INIT4(buffer_vld4, poly, 16, 8);
 PAD(buffer_vld4_pad, poly, 16, 8);
 VECT_ARRAY_INIT4(buffer_vld4, float, 32, 4);
 PAD(buffer_vld4_pad, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld4_float16x8x4[8*4] = {-16, -15, -14, -13, -12, -11, -10, -9,
 				       -8, -7, -6, -5, -4, -3, -2, -1,
@@ -511,7 +511,7 @@ VECT_VAR_DECL_INIT(buffer_vld2_lane, uint, 64, 2);
 VECT_VAR_DECL_INIT(buffer_vld2_lane, poly, 8, 2);
 VECT_VAR_DECL_INIT(buffer_vld2_lane, poly, 16, 2);
 VECT_VAR_DECL_INIT(buffer_vld2_lane, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld2_lane_float16x2[2] = {-16, -15};
 #else
@@ -532,7 +532,7 @@ VECT_VAR_DECL_INIT(buffer_vld3_lane, uint, 64, 3);
 VECT_VAR_DECL_INIT(buffer_vld3_lane, poly, 8, 3);
 VECT_VAR_DECL_INIT(buffer_vld3_lane, poly, 16, 3);
 VECT_VAR_DECL_INIT(buffer_vld3_lane, float, 32, 3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld3_lane_float16x3[3] = {-16, -15, -14};
 #else
@@ -554,7 +554,7 @@ VECT_VAR_DECL_INIT(buffer_vld4_lane, uint, 64, 4);
 VECT_VAR_DECL_INIT(buffer_vld4_lane, poly, 8, 4);
 VECT_VAR_DECL_INIT(buffer_vld4_lane, poly, 16, 4);
 VECT_VAR_DECL_INIT(buffer_vld4_lane, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #ifdef __ARMCC_VERSION
 __fp16 buffer_vld4_lane_float16x4[4] = {-16, -15, -14, -13};
 #else
diff --git a/ref_vcombine.c b/ref_vcombine.c
index 1d9f61b..570c2c5 100644
--- a/ref_vcombine.c
+++ b/ref_vcombine.c
@@ -46,7 +46,7 @@ void exec_vcombine (void)
   DECL_VARIABLE_64BITS_VARIANTS(vector64_a);
   DECL_VARIABLE_64BITS_VARIANTS(vector64_b);
   DECL_VARIABLE_128BITS_VARIANTS(vector128);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector64_a, float, 16, 4);
   DECL_VARIABLE(vector64_b, float, 16, 4);
   DECL_VARIABLE(vector64_b_init, uint, 16, 4);
@@ -55,7 +55,7 @@ void exec_vcombine (void)
 
   TEST_MACRO_64BITS_VARIANTS_2_5(TEST_VLOAD, vector64_a, buffer);
   TEST_VLOAD(vector64_a, buffer, , float, f, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VLOAD(vector64_a, buffer, , float, f, 16, 4);
 #endif
 
@@ -71,7 +71,7 @@ void exec_vcombine (void)
   TEST_VDUP(vector64_b, , poly, p, 16, 4, 0x66);
   TEST_VDUP(vector64_b, , float, f, 32, 2, 3.3f);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   /* There is no vdup_n_f16, so we need another initialization
      method.  */
   TEST_VDUP(vector64_b_init, , uint, u, 16, 4, 0x4b80 /* 15 */);
@@ -92,7 +92,7 @@ void exec_vcombine (void)
   TEST_VCOMBINE(poly, p, 8, 8, 16);
   TEST_VCOMBINE(poly, p, 16, 4, 8);
   TEST_VCOMBINE(float, f, 32, 2, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VCOMBINE(float, f, 16, 4, 8);
 #endif
 
diff --git a/ref_vcreate.c b/ref_vcreate.c
index 6192879..e11cceb 100644
--- a/ref_vcreate.c
+++ b/ref_vcreate.c
@@ -63,7 +63,7 @@ FNNAME (INSN_NAME)
   DECL_VAL(val, uint, 64, 1);
   DECL_VAL(val, poly, 8, 8);
   DECL_VAL(val, poly, 16, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VAL(val, float, 16, 4);
 #endif
 
@@ -72,7 +72,7 @@ FNNAME (INSN_NAME)
   DECL_VARIABLE(vector_res, int, 32, 2);
   DECL_VARIABLE(vector_res, int, 64, 1);
   DECL_VARIABLE(vector_res, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector_res, float, 16, 4);
 #endif
   DECL_VARIABLE(vector_res, uint, 8, 8);
@@ -90,7 +90,7 @@ FNNAME (INSN_NAME)
   VECT_VAR(val, int, 32, 2) = 0x123456789abcdef0LL;
   VECT_VAR(val, int, 64, 1) = 0x123456789abcdef0LL;
   VECT_VAR(val, float, 32, 2) = 0x123456789abcdef0LL;
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   VECT_VAR(val, float, 16, 4) = 0x123456789abcdef0LL;
 #endif
   VECT_VAR(val, uint, 8, 8) = 0x123456789abcdef0ULL;
@@ -104,7 +104,7 @@ FNNAME (INSN_NAME)
   TEST_VCREATE(int, s, 16, 4);
   TEST_VCREATE(int, s, 32, 2);
   TEST_VCREATE(float, f, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VCREATE(float, f, 16, 4);
 #endif
   TEST_VCREATE(int, s, 64, 1);
diff --git a/ref_vcvt.c b/ref_vcvt.c
index dcabe99..94a570f 100644
--- a/ref_vcvt.c
+++ b/ref_vcvt.c
@@ -52,7 +52,7 @@ void exec_vcvt (void)
 		    VECT_VAR(vector_res, T1, W, N));		\
   DUMP_FP(TEST_MSG, T1, W, N, PRIx##W);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_VCVT_FP16(T1, T2, W1, W2, N)			\
   VECT_VAR(vector_res, T1, W1, N) =				\
     vcvt_##T2##W1##_##T2##W2(VECT_VAR(vector, T1, W2, N));	\
@@ -86,7 +86,7 @@ void exec_vcvt (void)
      statement */
   DECL_VARIABLE_ALL_VARIANTS(vector);
   DECL_VARIABLE_ALL_VARIANTS(vector_res);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector_init, uint, 16, 4);
   DECL_VARIABLE(vector_init, uint, 16, 8);
   DECL_VARIABLE(vector, float, 16, 4);
@@ -102,7 +102,7 @@ void exec_vcvt (void)
   TEST_MACRO_ALL_VARIANTS_2_5(TEST_VLOAD, vector, buffer);
   TEST_VLOAD(vector, buffer, , float, f, 32, 2);
   TEST_VLOAD(vector, buffer, q, float, f, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VLOAD(vector, buffer, , float, f, 16, 4);
   TEST_VLOAD(vector, buffer, q, float, f, 16, 8);
 #endif
@@ -114,7 +114,7 @@ void exec_vcvt (void)
   TEST_VSET_LANE(vector, q, float, f, 32, 4, 2, -15.3f);
   TEST_VSET_LANE(vector, q, float, f, 32, 4, 3, 5.3f);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   /* FP16 tests.  */
   /* There is no vdup_n_f16, so we need another initialization
      method.  */
@@ -136,7 +136,7 @@ void exec_vcvt (void)
   /* vcvt_f32_xx */
   TEST_VCVT_FP(, float, f, 32, 2, int, s);
   TEST_VCVT_FP(, float, f, 32, 2, uint, u);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VCVT_FP16(float, f, 32, 16, 4);
 #endif
 
@@ -152,7 +152,7 @@ void exec_vcvt (void)
   TEST_VSET_LANE(vector, q, float, f, 32, 4, 1, -0.0f);
   TEST_VSET_LANE(vector, q, float, f, 32, 4, 2, 15.12f);
   TEST_VSET_LANE(vector, q, float, f, 32, 4, 3, -15.12f);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VCVT_2FP16(float, f, 16, 32, 4);
 #endif
 
@@ -214,7 +214,7 @@ void exec_vcvt (void)
   /* vcvtq_n_xx_f32 */
   TEST_VCVT_N(q, int, s, 32, 4, float, f, 31);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #undef TEST_MSG
 #define TEST_MSG "VCVT FP16"
   fprintf(ref_file, "\n%s output:\n", TEST_MSG " (check fp16-fp32 inf/nan/denormal)");
diff --git a/ref_vget_high.c b/ref_vget_high.c
index b15b962..e067f7b 100644
--- a/ref_vget_high.c
+++ b/ref_vget_high.c
@@ -44,14 +44,14 @@ void exec_vget_high (void)
      statement */
   DECL_VARIABLE_64BITS_VARIANTS(vector64);
   DECL_VARIABLE_128BITS_VARIANTS(vector128);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector64, float, 16, 4);
   DECL_VARIABLE(vector128, float, 16, 8);
 #endif
 
   TEST_MACRO_128BITS_VARIANTS_2_5(TEST_VLOAD, vector128, buffer);
   TEST_VLOAD(vector128, buffer, q, float, f, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VLOAD(vector128, buffer, q, float, f, 16, 8);
 #endif
 
@@ -68,7 +68,7 @@ void exec_vget_high (void)
   TEST_VGET_HIGH(poly, p, 8, 8, 16);
   TEST_VGET_HIGH(poly, p, 16, 4, 8);
   TEST_VGET_HIGH(float, f, 32, 2, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VGET_HIGH(float, f, 16, 4, 8);
 #endif
 
diff --git a/ref_vget_low.c b/ref_vget_low.c
index b42548d..07ee684 100644
--- a/ref_vget_low.c
+++ b/ref_vget_low.c
@@ -44,14 +44,14 @@ void exec_vget_low (void)
      statement */
   DECL_VARIABLE_64BITS_VARIANTS(vector64);
   DECL_VARIABLE_128BITS_VARIANTS(vector128);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector64, float, 16, 4);
   DECL_VARIABLE(vector128, float, 16, 8);
 #endif
 
   TEST_MACRO_128BITS_VARIANTS_2_5(TEST_VLOAD, vector128, buffer);
   TEST_VLOAD(vector128, buffer, q, float, f, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VLOAD(vector128, buffer, q, float, f, 16, 8);
 #endif
 
@@ -68,7 +68,7 @@ void exec_vget_low (void)
   TEST_VGET_LOW(poly, p, 8, 8, 16);
   TEST_VGET_LOW(poly, p, 16, 4, 8);
   TEST_VGET_LOW(float, f, 32, 2, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VGET_LOW(float, f, 16, 4, 8);
 #endif
 
diff --git a/ref_vld1.c b/ref_vld1.c
index d62e535..0e7eebd 100644
--- a/ref_vld1.c
+++ b/ref_vld1.c
@@ -43,7 +43,7 @@ void exec_vld1 (void)
   /* With ARM RVCT, we need to declare variables before any executable
      statement */
   DECL_VARIABLE_ALL_VARIANTS(vector);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector, float, 16, 4);
   DECL_VARIABLE(vector, float, 16, 8);
 #endif
@@ -55,7 +55,7 @@ void exec_vld1 (void)
   TEST_VLD1(vector, buffer, , float, f, 32, 2);
   TEST_VLD1(vector, buffer, q, float, f, 32, 4);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VLD1(vector, buffer, , float, f, 16, 4);
   TEST_VLD1(vector, buffer, q, float, f, 16, 8);
 #endif
diff --git a/ref_vld1_dup.c b/ref_vld1_dup.c
index 45f02a6..66fe6a2 100644
--- a/ref_vld1_dup.c
+++ b/ref_vld1_dup.c
@@ -45,7 +45,7 @@ void exec_vld1_dup (void)
   /* With ARM RVCT, we need to declare variables before any executable
      statement */
   DECL_VARIABLE_ALL_VARIANTS(vector);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector, float, 16, 4);
   DECL_VARIABLE(vector, float, 16, 8);
 #endif
@@ -59,7 +59,7 @@ void exec_vld1_dup (void)
     TEST_VLD1_DUP(vector, buffer_dup, , float, f, 32, 2);
     TEST_VLD1_DUP(vector, buffer_dup, q, float, f, 32, 4);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
     TEST_VLD1_DUP(vector, buffer_dup, , float, f, 16, 4);
     TEST_VLD1_DUP(vector, buffer_dup, q, float, f, 16, 8);
 #endif
diff --git a/ref_vld1_lane.c b/ref_vld1_lane.c
index 742c357..f4a9af6 100644
--- a/ref_vld1_lane.c
+++ b/ref_vld1_lane.c
@@ -47,12 +47,12 @@ void exec_vld1_lane (void)
   /* With ARM RVCT, we need to declare variables before any executable
      statement */
   DECL_VARIABLE_ALL_VARIANTS(vector);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector, float, 16, 4);
   DECL_VARIABLE(vector, float, 16, 8);
 #endif
   DECL_VARIABLE_ALL_VARIANTS(vector_src);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector_src, float, 16, 4);
   DECL_VARIABLE(vector_src, float, 16, 8);
 #endif
@@ -68,7 +68,7 @@ void exec_vld1_lane (void)
   ARRAY(buffer_src, poly, 8, 8);
   ARRAY(buffer_src, poly, 16, 4);
   ARRAY(buffer_src, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   ARRAY(buffer_src, float, 16, 4);
 #endif
 
@@ -83,7 +83,7 @@ void exec_vld1_lane (void)
   ARRAY(buffer_src, poly, 8, 16);
   ARRAY(buffer_src, poly, 16, 8);
   ARRAY(buffer_src, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   ARRAY(buffer_src, float, 16, 8);
 #endif
 
@@ -101,7 +101,7 @@ void exec_vld1_lane (void)
   TEST_VLD1_LANE(, poly, p, 8, 8, 7);
   TEST_VLD1_LANE(, poly, p, 16, 4, 3);
   TEST_VLD1_LANE(, float, f, 32, 2, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VLD1_LANE(, float, f, 16, 4, 2);
 #endif
 
@@ -116,7 +116,7 @@ void exec_vld1_lane (void)
   TEST_VLD1_LANE(q, poly, p, 8, 16, 12);
   TEST_VLD1_LANE(q, poly, p, 16, 8, 6);
   TEST_VLD1_LANE(q, float, f, 32, 4, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VLD1_LANE(q, float, f, 16, 8, 5);
 #endif
 
diff --git a/ref_vldX.c b/ref_vldX.c
index c3e8e4f..7be336d 100644
--- a/ref_vldX.c
+++ b/ref_vldX.c
@@ -85,7 +85,7 @@ void exec_vldX (void)
   DECL_VLDX(poly, 16, 8, X);			\
   DECL_VLDX(float, 32, 4, X)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define DECL_ALL_VLDX_FP16(X)			\
   DECL_VLDX(float, 16, 4, X);			\
   DECL_VLDX(float, 16, 8, X)
@@ -113,7 +113,7 @@ void exec_vldX (void)
   TEST_VLDX(q, poly, p, 16, 8, X);		\
   TEST_VLDX(q, float, f, 32, 4, X)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_ALL_VLDX_FP16(X)			\
   TEST_VLDX(, float, f, 16, 4, X);		\
   TEST_VLDX(q, float, f, 16, 8, X)
@@ -141,7 +141,7 @@ void exec_vldX (void)
   TEST_EXTRA_CHUNK(poly, 16, 8, X, Y);		\
   TEST_EXTRA_CHUNK(float, 32, 4, X, Y)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_ALL_EXTRA_CHUNKS_FP16(X, Y)	\
   TEST_EXTRA_CHUNK(float, 16, 4, X, Y);		\
   TEST_EXTRA_CHUNK(float, 16, 8, X, Y)
@@ -151,7 +151,7 @@ void exec_vldX (void)
   DECL_ALL_VLDX(3);
   DECL_ALL_VLDX(4);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_ALL_VLDX_FP16(2);
   DECL_ALL_VLDX_FP16(3);
   DECL_ALL_VLDX_FP16(4);
@@ -161,13 +161,13 @@ void exec_vldX (void)
   clean_results ();
 #define TEST_MSG "VLD2/VLD2Q"
   TEST_ALL_VLDX(2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_FP16(2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
 
   TEST_ALL_EXTRA_CHUNKS(2, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(2, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
@@ -177,18 +177,18 @@ void exec_vldX (void)
 #undef TEST_MSG
 #define TEST_MSG "VLD3/VLD3Q"
   TEST_ALL_VLDX(3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_FP16(3);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
 
   TEST_ALL_EXTRA_CHUNKS(3, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(3, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
   TEST_ALL_EXTRA_CHUNKS(3, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(3, 2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 2");
@@ -198,23 +198,23 @@ void exec_vldX (void)
 #undef TEST_MSG
 #define TEST_MSG "VLD4/VLD4Q"
   TEST_ALL_VLDX(4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_FP16(4);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
 
   TEST_ALL_EXTRA_CHUNKS(4, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
   TEST_ALL_EXTRA_CHUNKS(4, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 2");
   TEST_ALL_EXTRA_CHUNKS(4, 3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 3);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 3");
diff --git a/ref_vldX_dup.c b/ref_vldX_dup.c
index 08d1a3b..a0a417d 100644
--- a/ref_vldX_dup.c
+++ b/ref_vldX_dup.c
@@ -77,7 +77,7 @@ void exec_vldX_dup (void)
   DECL_VLDX_DUP(poly, 16, 4, X);		\
   DECL_VLDX_DUP(float, 32, 2, X)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define DECL_ALL_VLDX_DUP_FP16(X)		\
   DECL_VLDX_DUP(float, 16, 4, X)
 #endif
@@ -95,7 +95,7 @@ void exec_vldX_dup (void)
   TEST_VLDX_DUP(, poly, p, 16, 4, X);		\
   TEST_VLDX_DUP(, float, f, 32, 2, X)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_ALL_VLDX_DUP_FP16(X)		\
   TEST_VLDX_DUP(, float, f, 16, 4, X)
 #endif
@@ -113,7 +113,7 @@ void exec_vldX_dup (void)
   TEST_EXTRA_CHUNK(poly, 16, 4, X, Y);		\
   TEST_EXTRA_CHUNK(float, 32, 2, X, Y)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_ALL_EXTRA_CHUNKS_FP16(X, Y)	\
   TEST_EXTRA_CHUNK(float, 16, 4, X, Y)
 #endif
@@ -122,7 +122,7 @@ void exec_vldX_dup (void)
   DECL_ALL_VLDX_DUP(2);
   DECL_ALL_VLDX_DUP(3);
   DECL_ALL_VLDX_DUP(4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_ALL_VLDX_DUP_FP16(2);
   DECL_ALL_VLDX_DUP_FP16(3);
   DECL_ALL_VLDX_DUP_FP16(4);
@@ -132,12 +132,12 @@ void exec_vldX_dup (void)
   clean_results ();
 #define TEST_MSG "VLD2_DUP/VLD2Q_DUP"
   TEST_ALL_VLDX_DUP(2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_DUP_FP16(2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
   TEST_ALL_EXTRA_CHUNKS(2, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(2, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
@@ -147,17 +147,17 @@ void exec_vldX_dup (void)
 #undef TEST_MSG
 #define TEST_MSG "VLD3_DUP/VLD3Q_DUP"
   TEST_ALL_VLDX_DUP(3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_DUP_FP16(3);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
   TEST_ALL_EXTRA_CHUNKS(3, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(3, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
   TEST_ALL_EXTRA_CHUNKS(3, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(3, 2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 2");
@@ -167,22 +167,22 @@ void exec_vldX_dup (void)
 #undef TEST_MSG
 #define TEST_MSG "VLD4_DUP/VLD4Q_DUP"
   TEST_ALL_VLDX_DUP(4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_DUP_FP16(4);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
   TEST_ALL_EXTRA_CHUNKS(4, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
   TEST_ALL_EXTRA_CHUNKS(4, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 2");
   TEST_ALL_EXTRA_CHUNKS(4, 3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 3);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 3");
diff --git a/ref_vldX_lane.c b/ref_vldX_lane.c
index 5901e83..c0d5a34 100644
--- a/ref_vldX_lane.c
+++ b/ref_vldX_lane.c
@@ -89,7 +89,7 @@ void exec_vldX_lane (void)
   DECL_VLDX_LANE(float, 32, 2, X);		\
   DECL_VLDX_LANE(float, 32, 4, X)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define DECL_ALL_VLDX_LANE_FP16(X)		\
   DECL_VLDX_LANE(float, 16, 4, X);		\
   DECL_VLDX_LANE(float, 16, 8, X)
@@ -122,7 +122,7 @@ void exec_vldX_lane (void)
   TEST_VLDX_LANE(, float, f, 32, 2, X, 0);	\
   TEST_VLDX_LANE(q, float, f, 32, 4, X, 2)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_ALL_VLDX_LANE_FP16(X)		\
   TEST_VLDX_LANE(, float, f, 16, 4, X, 0);	\
   TEST_VLDX_LANE(q, float, f, 16, 8, X, 2)
@@ -145,7 +145,7 @@ void exec_vldX_lane (void)
   TEST_EXTRA_CHUNK(float, 32, 2, X, Y);		\
   TEST_EXTRA_CHUNK(float, 32, 4, X, Y)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_ALL_EXTRA_CHUNKS_FP16(X, Y)	\
   TEST_EXTRA_CHUNK(float, 16, 4, X, Y);		\
   TEST_EXTRA_CHUNK(float, 16, 8, X, Y)
@@ -155,7 +155,7 @@ void exec_vldX_lane (void)
   DECL_ALL_VLDX_LANE(2);
   DECL_ALL_VLDX_LANE(3);
   DECL_ALL_VLDX_LANE(4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_ALL_VLDX_LANE_FP16(2);
   DECL_ALL_VLDX_LANE_FP16(3);
   DECL_ALL_VLDX_LANE_FP16(4);
@@ -177,7 +177,7 @@ void exec_vldX_lane (void)
   DUMMY_ARRAY(buffer_src, poly, 16, 8, 4);
   DUMMY_ARRAY(buffer_src, float, 32, 2, 4);
   DUMMY_ARRAY(buffer_src, float, 32, 4, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DUMMY_ARRAY(buffer_src, float, 16, 4, 4);
   DUMMY_ARRAY(buffer_src, float, 16, 8, 4);
 #endif
@@ -186,12 +186,12 @@ void exec_vldX_lane (void)
   clean_results ();
 #define TEST_MSG "VLD2_LANE/VLD2Q_LANE"
   TEST_ALL_VLDX_LANE(2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_LANE_FP16(2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
   TEST_ALL_EXTRA_CHUNKS(2, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(2, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
@@ -201,17 +201,17 @@ void exec_vldX_lane (void)
 #undef TEST_MSG
 #define TEST_MSG "VLD3_LANE/VLD3Q_LANE"
   TEST_ALL_VLDX_LANE(3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_LANE_FP16(3);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
   TEST_ALL_EXTRA_CHUNKS(3, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(3, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
   TEST_ALL_EXTRA_CHUNKS(3, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(3, 2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 2");
@@ -221,22 +221,22 @@ void exec_vldX_lane (void)
 #undef TEST_MSG
 #define TEST_MSG "VLD4_LANE/VLD4Q_LANE"
   TEST_ALL_VLDX_LANE(4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VLDX_LANE_FP16(4);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
   TEST_ALL_EXTRA_CHUNKS(4, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
   TEST_ALL_EXTRA_CHUNKS(4, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 2");
   TEST_ALL_EXTRA_CHUNKS(4, 3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 3);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 3");
diff --git a/ref_vreinterpret.c b/ref_vreinterpret.c
index e4da704..894431d 100644
--- a/ref_vreinterpret.c
+++ b/ref_vreinterpret.c
@@ -59,7 +59,7 @@ void exec_vreinterpret (void)
 		    VECT_VAR(vector_res, T1, W, N));			\
   DUMP_FP(TEST_MSG, T1, W, N, PRIx##W);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_VREINTERPRET_FP16(Q, T1, T2, W, N, TS1, TS2, WS, NS)	\
   VECT_VAR(vector_res, T1, W, N) =					\
     vreinterpret##Q##_##T2##W##_##TS2##WS(VECT_VAR(vector, TS1, WS, NS)); \
@@ -72,7 +72,7 @@ void exec_vreinterpret (void)
      statement */
   DECL_VARIABLE_ALL_VARIANTS(vector);
   DECL_VARIABLE_ALL_VARIANTS(vector_res);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector, float, 16, 4);
   DECL_VARIABLE(vector_res, float, 16, 4);
   DECL_VARIABLE(vector, float, 16, 8);
@@ -86,7 +86,7 @@ void exec_vreinterpret (void)
   TEST_MACRO_ALL_VARIANTS_2_5(TEST_VLOAD, vector, buffer);
   TEST_VLOAD(vector, buffer, , float, f, 32, 2);
   TEST_VLOAD(vector, buffer, q, float, f, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VLOAD(vector, buffer, , float, f, 16, 4);
   TEST_VLOAD(vector, buffer, q, float, f, 16, 8);
 #endif
@@ -342,7 +342,7 @@ void exec_vreinterpret (void)
   TEST_VREINTERPRET_POLY(q, poly, p, 8, 16, float, f, 32, 4);
   TEST_VREINTERPRET_POLY(q, poly, p, 16, 8, float, f, 32, 4);
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   /* vreinterpret_f16_xx */
   TEST_VREINTERPRET_FP16(, float, f, 16, 4, int, s, 8, 8);
   TEST_VREINTERPRET_FP16(, float, f, 16, 4, int, s, 16, 4);
diff --git a/ref_vst1_lane.c b/ref_vst1_lane.c
index 341b0a8..c7e533f 100644
--- a/ref_vst1_lane.c
+++ b/ref_vst1_lane.c
@@ -43,7 +43,7 @@ void exec_vst1_lane (void)
   /* With ARM RVCT, we need to declare variables before any executable
      statement */
   DECL_VARIABLE_ALL_VARIANTS(vector);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_VARIABLE(vector, float, 16, 4);
   DECL_VARIABLE(vector, float, 16, 8);
 #endif
@@ -62,7 +62,7 @@ void exec_vst1_lane (void)
   TEST_VST1_LANE(, poly, p, 8, 8, 6);
   TEST_VST1_LANE(, poly, p, 16, 4, 2);
   TEST_VST1_LANE(, float, f, 32, 2, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VST1_LANE(, float, f, 16, 4, 2);
 #endif
 
@@ -77,7 +77,7 @@ void exec_vst1_lane (void)
   TEST_VST1_LANE(q, poly, p, 8, 16, 10);
   TEST_VST1_LANE(q, poly, p, 16, 8, 4);
   TEST_VST1_LANE(q, float, f, 32, 4, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_VST1_LANE(q, float, f, 16, 8, 5);
 #endif
 
diff --git a/ref_vstX_lane.c b/ref_vstX_lane.c
index 7a9533c..3c0f97a 100644
--- a/ref_vstX_lane.c
+++ b/ref_vstX_lane.c
@@ -92,7 +92,7 @@ void exec_vstX_lane (void)
   DECL_VSTX_LANE(poly, 16, 8, X);		\
   DECL_VSTX_LANE(float, 32, 4, X)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define DECL_ALL_VSTX_LANE_FP16(X)		\
   DECL_VSTX_LANE(float, 16, 4, X);		\
   DECL_VSTX_LANE(float, 16, 8, X)
@@ -119,7 +119,7 @@ void exec_vstX_lane (void)
   TEST_VSTX_LANE(q, poly, p, 16, 8, X, 5);	\
   TEST_VSTX_LANE(q, float, f, 32, 4, X, 2)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_ALL_VSTX_LANE_FP16(X)		\
   TEST_VSTX_LANE(, float, f, 16, 4, X, 3);	\
   TEST_VSTX_LANE(q, float, f, 16, 8, X, 6)
@@ -142,7 +142,7 @@ void exec_vstX_lane (void)
   TEST_EXTRA_CHUNK(poly, 16, 8, X, Y);		\
   TEST_EXTRA_CHUNK(float, 32, 4, X, Y)
 
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define TEST_ALL_EXTRA_CHUNKS_FP16(X, Y)	\
   TEST_EXTRA_CHUNK(float, 16, 4, X, Y);		\
   TEST_EXTRA_CHUNK(float, 16, 8, X, Y)
@@ -152,7 +152,7 @@ void exec_vstX_lane (void)
   DECL_ALL_VSTX_LANE(2);
   DECL_ALL_VSTX_LANE(3);
   DECL_ALL_VSTX_LANE(4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DECL_ALL_VSTX_LANE_FP16(2);
   DECL_ALL_VSTX_LANE_FP16(3);
   DECL_ALL_VSTX_LANE_FP16(4);
@@ -174,7 +174,7 @@ void exec_vstX_lane (void)
   DUMMY_ARRAY(buffer_src, uint, 32, 4, 4);
   DUMMY_ARRAY(buffer_src, poly, 16, 8, 4);
   DUMMY_ARRAY(buffer_src, float, 32, 4, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DUMMY_ARRAY(buffer_src, float, 16, 4, 4);
   DUMMY_ARRAY(buffer_src, float, 16, 8, 4);
 #endif
@@ -183,13 +183,13 @@ void exec_vstX_lane (void)
   clean_results ();
 #define TEST_MSG "VST2_LANE/VST2Q_LANE"
   TEST_ALL_VSTX_LANE(2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VSTX_LANE_FP16(2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
 
   TEST_ALL_EXTRA_CHUNKS(2, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(2, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
@@ -199,18 +199,18 @@ void exec_vstX_lane (void)
 #undef TEST_MSG
 #define TEST_MSG "VST3_LANE/VST3Q_LANE"
   TEST_ALL_VSTX_LANE(3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VSTX_LANE_FP16(3);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
 
   TEST_ALL_EXTRA_CHUNKS(3, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(3, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
   TEST_ALL_EXTRA_CHUNKS(3, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(3, 2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 2");
@@ -220,23 +220,23 @@ void exec_vstX_lane (void)
 #undef TEST_MSG
 #define TEST_MSG "VST4_LANE/VST4Q_LANE"
   TEST_ALL_VSTX_LANE(4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_VSTX_LANE_FP16(4);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 0");
 
   TEST_ALL_EXTRA_CHUNKS(4, 1);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 1);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 1");
   TEST_ALL_EXTRA_CHUNKS(4, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 2);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 2");
   TEST_ALL_EXTRA_CHUNKS(4, 3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   TEST_ALL_EXTRA_CHUNKS_FP16(4, 3);
 #endif
   dump_results_hex2 (TEST_MSG, " chunk 3");
diff --git a/stm-arm-neon-ref.h b/stm-arm-neon-ref.h
index 1500905..b06b23f 100644
--- a/stm-arm-neon-ref.h
+++ b/stm-arm-neon-ref.h
@@ -150,11 +150,8 @@ static int result_idx = 0;
     fprintf(gcc_tests_file, " };\n");					\
   }
 
-/* ARMCC has internal knowledge of half-precision type. Define this
-   alias to avoid having to duplicate declarations.  */
-#ifdef __ARMCC_VERSION
+#if defined(__ARM_FP16_FORMAT_IEEE)
 #define float16_t __fp16
-#endif
 
 #define DUMP_FP16(MSG,T,W,N,FMT)					\
   fprintf(ref_file, "%s:%d:%s [] = { ", MSG, result_idx++,		\
@@ -182,6 +179,7 @@ static int result_idx = 0;
     fprintf(gcc_tests_file, "0x%" FMT, tmp);				\
     fprintf(gcc_tests_file, " };\n");					\
   }
+#endif
 
 #define CLEAN_PATTERN_8  0x33
 #define CLEAN_PATTERN_16 0x3333
@@ -228,7 +226,7 @@ extern ARRAY(buffer, uint, 64, 1);
 extern ARRAY(buffer, poly, 8, 8);
 extern ARRAY(buffer, poly, 16, 4);
 extern ARRAY(buffer, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern ARRAY(buffer, float, 16, 4);
 #endif
 extern ARRAY(buffer, int, 8, 16);
@@ -242,7 +240,7 @@ extern ARRAY(buffer, uint, 64, 2);
 extern ARRAY(buffer, poly, 8, 16);
 extern ARRAY(buffer, poly, 16, 8);
 extern ARRAY(buffer, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern ARRAY(buffer, float, 16, 8);
 #endif
 
@@ -260,7 +258,7 @@ extern ARRAY4(buffer_dup, uint, 64, 1);
 extern ARRAY(buffer_dup, poly, 8, 8);
 extern ARRAY(buffer_dup, poly, 16, 4);
 extern ARRAY4(buffer_dup, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern ARRAY4(buffer_dup, float, 16, 4);
 #endif
 extern ARRAY(buffer_dup, int, 8, 16);
@@ -274,7 +272,7 @@ extern ARRAY4(buffer_dup, uint, 64, 2);
 extern ARRAY(buffer_dup, poly, 8, 16);
 extern ARRAY(buffer_dup, poly, 16, 8);
 extern ARRAY(buffer_dup, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern ARRAY(buffer_dup, float, 16, 8);
 #endif
 
@@ -290,7 +288,7 @@ extern VECT_ARRAY(buffer_vld2, uint, 64, 1, 2);
 extern VECT_ARRAY(buffer_vld2, poly, 8, 8, 2);
 extern VECT_ARRAY(buffer_vld2, poly, 16, 4, 2);
 extern VECT_ARRAY(buffer_vld2, float, 32, 2, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_ARRAY(buffer_vld2, float, 16, 4, 2);
 #endif
 extern VECT_ARRAY(buffer_vld2, int, 8, 16, 2);
@@ -304,7 +302,7 @@ extern VECT_ARRAY(buffer_vld2, uint, 64, 2, 2);
 extern VECT_ARRAY(buffer_vld2, poly, 8, 16, 2);
 extern VECT_ARRAY(buffer_vld2, poly, 16, 8, 2);
 extern VECT_ARRAY(buffer_vld2, float, 32, 4, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_ARRAY(buffer_vld2, float, 16, 8, 2);
 #endif
 
@@ -320,7 +318,7 @@ extern VECT_ARRAY(buffer_vld3, uint, 64, 1, 3);
 extern VECT_ARRAY(buffer_vld3, poly, 8, 8, 3);
 extern VECT_ARRAY(buffer_vld3, poly, 16, 4, 3);
 extern VECT_ARRAY(buffer_vld3, float, 32, 2, 3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_ARRAY(buffer_vld3, float, 16, 4, 3);
 #endif
 extern VECT_ARRAY(buffer_vld3, int, 8, 16, 3);
@@ -334,7 +332,7 @@ extern VECT_ARRAY(buffer_vld3, uint, 64, 2, 3);
 extern VECT_ARRAY(buffer_vld3, poly, 8, 16, 3);
 extern VECT_ARRAY(buffer_vld3, poly, 16, 8, 3);
 extern VECT_ARRAY(buffer_vld3, float, 32, 4, 3);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_ARRAY(buffer_vld3, float, 16, 8, 3);
 #endif
 
@@ -350,7 +348,7 @@ extern VECT_ARRAY(buffer_vld4, uint, 64, 1, 4);
 extern VECT_ARRAY(buffer_vld4, poly, 8, 8, 4);
 extern VECT_ARRAY(buffer_vld4, poly, 16, 4, 4);
 extern VECT_ARRAY(buffer_vld4, float, 32, 2, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_ARRAY(buffer_vld4, float, 16, 4, 4);
 #endif
 extern VECT_ARRAY(buffer_vld4, int, 8, 16, 4);
@@ -364,7 +362,7 @@ extern VECT_ARRAY(buffer_vld4, uint, 64, 2, 4);
 extern VECT_ARRAY(buffer_vld4, poly, 8, 16, 4);
 extern VECT_ARRAY(buffer_vld4, poly, 16, 8, 4);
 extern VECT_ARRAY(buffer_vld4, float, 32, 4, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_ARRAY(buffer_vld4, float, 16, 8, 4);
 #endif
 
@@ -380,7 +378,7 @@ extern VECT_VAR_DECL(buffer_vld2_lane, uint, 64, 2)[2];
 extern VECT_VAR_DECL(buffer_vld2_lane, poly, 8, 2)[2];
 extern VECT_VAR_DECL(buffer_vld2_lane, poly, 16, 2)[2];
 extern VECT_VAR_DECL(buffer_vld2_lane, float, 32, 2)[2];
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_VAR_DECL(buffer_vld2_lane, float, 16, 2)[2];
 #endif
 
@@ -396,7 +394,7 @@ extern VECT_VAR_DECL(buffer_vld3_lane, uint, 64, 3)[3];
 extern VECT_VAR_DECL(buffer_vld3_lane, poly, 8, 3)[3];
 extern VECT_VAR_DECL(buffer_vld3_lane, poly, 16, 3)[3];
 extern VECT_VAR_DECL(buffer_vld3_lane, float, 32, 3)[3];
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_VAR_DECL(buffer_vld3_lane, float, 16, 3)[3];
 #endif
 
@@ -412,7 +410,7 @@ extern VECT_VAR_DECL(buffer_vld4_lane, uint, 64, 4)[4];
 extern VECT_VAR_DECL(buffer_vld4_lane, poly, 8, 4)[4];
 extern VECT_VAR_DECL(buffer_vld4_lane, poly, 16, 4)[4];
 extern VECT_VAR_DECL(buffer_vld4_lane, float, 32, 4)[4];
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 extern VECT_VAR_DECL(buffer_vld4_lane, float, 16, 4)[4];
 #endif
 
@@ -428,7 +426,7 @@ static ARRAY(result, uint, 64, 1);
 static ARRAY(result, poly, 8, 8);
 static ARRAY(result, poly, 16, 4);
 static ARRAY(result, float, 32, 2);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 static ARRAY(result, float, 16, 4);
 #endif
 static ARRAY(result, int, 8, 16);
@@ -442,7 +440,7 @@ static ARRAY(result, uint, 64, 2);
 static ARRAY(result, poly, 8, 16);
 static ARRAY(result, poly, 16, 8);
 static ARRAY(result, float, 32, 4);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
 static ARRAY(result, float, 16, 8);
 #endif
 
@@ -465,7 +463,7 @@ static void dump_results (char *test_name)
   DUMP_POLY(test_name, poly, 8, 8, PRIu8);
   DUMP_POLY(test_name, poly, 16, 4, PRIu16);
   DUMP_FP(test_name, float, 32, 2, PRIx32);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DUMP_FP16(test_name, float, 16, 4, PRIu16);
 #endif
 
@@ -480,7 +478,7 @@ static void dump_results (char *test_name)
   DUMP_POLY(test_name, poly, 8, 16, PRIu8);
   DUMP_POLY(test_name, poly, 16, 8, PRIu16);
   DUMP_FP(test_name, float, 32, 4, PRIx32);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DUMP_FP16(test_name, float, 16, 8, PRIu16);
 #endif
 }
@@ -504,7 +502,7 @@ static void dump_results_hex2 (const char *test_name, const char* comment)
   DUMP_POLY(test_name, poly, 8, 8, PRIx8);
   DUMP_POLY(test_name, poly, 16, 4, PRIx16);
   DUMP_FP(test_name, float, 32, 2, PRIx32);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DUMP_FP16(test_name, float, 16, 4, PRIx16);
 #endif
 
@@ -519,7 +517,7 @@ static void dump_results_hex2 (const char *test_name, const char* comment)
   DUMP_POLY(test_name, poly, 8, 16, PRIx8);
   DUMP_POLY(test_name, poly, 16, 8, PRIx16);
   DUMP_FP(test_name, float, 32, 4, PRIx32);
-#if __ARM_NEON_FP16_INTRINSICS
+#if defined(__ARM_FP16_FORMAT_IEEE)
   DUMP_FP16(test_name, float, 16, 8, PRIx16);
 #endif
 }