Not sure how CMPPS and CMPPD had already ever worked, I guess it didn't.

However with this fix it does now.

Basically the operand order for the x86 target specific node
is not the same as the instruction, but since the intrinsic need that
specific order at the instruction definition, just change the order
during legalization. Also, there were some wrong invertions of condition
codes, such as GE => LE, GT => LT, fix that too. Fix PR10907.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139528 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 55 insertions, 9 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index a3fb00af1e..f211c7260f 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8445,16 +8445,25 @@ SDValue X86TargetLowering::LowerVSETCC(SDValue Op, SelectionDAG &DAG) const {
     unsigned Opc = EltVT == MVT::f32 ? X86ISD::CMPPS : X86ISD::CMPPD;
     bool Swap = false;
 
+    // SSE Condition code mapping:
+    //  0 - EQ
+    //  1 - LT
+    //  2 - LE
+    //  3 - UNORD
+    //  4 - NEQ
+    //  5 - NLT
+    //  6 - NLE
+    //  7 - ORD
     switch (SetCCOpcode) {
     default: break;
     case ISD::SETOEQ:
     case ISD::SETEQ:  SSECC = 0; break;
-    case ISD::SETOGT:
-    case ISD::SETGT: Swap = true; // Fallthrough
-    case ISD::SETLT:
-    case ISD::SETOLT: SSECC = 1; break;
     case ISD::SETOGE:
     case ISD::SETGE: Swap = true; // Fallthrough
+    case ISD::SETLT:
+    case ISD::SETOLT: SSECC = 1; break;
+    case ISD::SETOGT:
+    case ISD::SETGT: Swap = true; // Fallthrough
     case ISD::SETLE:
     case ISD::SETOLE: SSECC = 2; break;
     case ISD::SETUO:  SSECC = 3; break;
@@ -8473,20 +8482,20 @@ SDValue X86TargetLowering::LowerVSETCC(SDValue Op, SelectionDAG &DAG) const {
     if (SSECC == 8) {
       if (SetCCOpcode == ISD::SETUEQ) {
         SDValue UNORD, EQ;
-        UNORD = DAG.getNode(Opc, dl, VT, Op0, Op1, DAG.getConstant(3, MVT::i8));
-        EQ = DAG.getNode(Opc, dl, VT, Op0, Op1, DAG.getConstant(0, MVT::i8));
+        UNORD = DAG.getNode(Opc, dl, VT, Op1, Op0, DAG.getConstant(3, MVT::i8));
+        EQ = DAG.getNode(Opc, dl, VT, Op1, Op0, DAG.getConstant(0, MVT::i8));
         return DAG.getNode(ISD::OR, dl, VT, UNORD, EQ);
       }
       else if (SetCCOpcode == ISD::SETONE) {
         SDValue ORD, NEQ;
-        ORD = DAG.getNode(Opc, dl, VT, Op0, Op1, DAG.getConstant(7, MVT::i8));
-        NEQ = DAG.getNode(Opc, dl, VT, Op0, Op1, DAG.getConstant(4, MVT::i8));
+        ORD = DAG.getNode(Opc, dl, VT, Op1, Op0, DAG.getConstant(7, MVT::i8));
+        NEQ = DAG.getNode(Opc, dl, VT, Op1, Op0, DAG.getConstant(4, MVT::i8));
         return DAG.getNode(ISD::AND, dl, VT, ORD, NEQ);
       }
       llvm_unreachable("Illegal FP comparison");
     }
     // Handle all other FP comparisons here.
-    return DAG.getNode(Opc, dl, VT, Op0, Op1, DAG.getConstant(SSECC, MVT::i8));
+    return DAG.getNode(Opc, dl, VT, Op1, Op0, DAG.getConstant(SSECC, MVT::i8));
   }
 
   // Break 256-bit integer vector compare into smaller ones.
diff --git a/test/CodeGen/X86/avx-blend.ll b/test/CodeGen/X86/avx-blend.ll
index e025e26bde..e3008d9e16 100644
--- a/test/CodeGen/X86/avx-blend.ll
+++ b/test/CodeGen/X86/avx-blend.ll
@@ -82,4 +82,23 @@ define <8 x i64> @vsel_i648(<8 x i64> %v1, <8 x i64> %v2) {
   ret <8 x i64> %vsel
 }
 
+;; TEST blend + compares
+; CHECK: A
+define <2 x double> @A(<2 x double> %x, <2 x double> %y) {
+  ; CHECK: vcmpltpd
+  ; CHECK: vblendvpd
+  %max_is_x = fcmp oge <2 x double> %x, %y
+  %max = select <2 x i1> %max_is_x, <2 x double> %x, <2 x double> %y
+  ret <2 x double> %max
+}
+
+; CHECK: B
+define <2 x double> @B(<2 x double> %x, <2 x double> %y) {
+  ; CHECK: vcmplepd
+  ; CHECK: vblendvpd
+  %max_is_x = fcmp ogt <2 x double> %x, %y
+  %max = select <2 x i1> %max_is_x, <2 x double> %x, <2 x double> %y
+  ret <2 x double> %max
+}
+
 
diff --git a/test/CodeGen/X86/sse41-blend.ll b/test/CodeGen/X86/sse41-blend.ll
index 3c854ac606..aba2c129e3 100644
--- a/test/CodeGen/X86/sse41-blend.ll
+++ b/test/CodeGen/X86/sse41-blend.ll
@@ -44,4 +44,22 @@ define <16 x i8> @vsel_i8(<16 x i8> %v1, <16 x i8> %v2) {
   ret <16 x i8> %vsel
 }
 
+;; TEST blend + compares
+; CHECK: A
+define <2 x double> @A(<2 x double> %x, <2 x double> %y) {
+  ; CHECK: cmpltpd
+  ; CHECK: blendvpd
+  %max_is_x = fcmp oge <2 x double> %x, %y
+  %max = select <2 x i1> %max_is_x, <2 x double> %x, <2 x double> %y
+  ret <2 x double> %max
+}
+
+; CHECK: B
+define <2 x double> @B(<2 x double> %x, <2 x double> %y) {
+  ; CHECK: cmplepd
+  ; CHECK: blendvpd
+  %max_is_x = fcmp ogt <2 x double> %x, %y
+  %max = select <2 x i1> %max_is_x, <2 x double> %x, <2 x double> %y
+  ret <2 x double> %max
+}