From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sun, 20 Jan 2019 16:40:33 +0000 (+0000)
Subject: [X86] Replace VPCOM/VPCOMU with generic integer comparisons (clang)
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=70faccd01059ccabbad982410d9f8c455ff90bec;p=clang

[X86] Replace VPCOM/VPCOMU with generic integer comparisons (clang)

These intrinsics can always be replaced with generic integer comparisons without any regression in codegen, even for -O0/-fast-isel cases.

Noticed while cleaning up vector integer comparison costs for PR40376.

A future commit will remove/autoupgrade the existing VPCOM/VPCOMU llvm intrinsics.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@351687 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 6732915413..390d28fd08 100644
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -9198,6 +9198,46 @@ static Value *EmitX86FunnelShift(CodeGenFunction &CGF, Value *Op0, Value *Op1,
   return CGF.Builder.CreateCall(F, {Op0, Op1, Amt});
 }
 
+static Value *EmitX86vpcom(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
+                           bool IsSigned) {
+  Value *Op0 = Ops[0];
+  Value *Op1 = Ops[1];
+  llvm::Type *Ty = Op0->getType();
+  uint64_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
+
+  CmpInst::Predicate Pred;
+  switch (Imm) {
+  case 0x0:
+    Pred = IsSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
+    break;
+  case 0x1:
+    Pred = IsSigned ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE;
+    break;
+  case 0x2:
+    Pred = IsSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
+    break;
+  case 0x3:
+    Pred = IsSigned ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE;
+    break;
+  case 0x4:
+    Pred = ICmpInst::ICMP_EQ;
+    break;
+  case 0x5:
+    Pred = ICmpInst::ICMP_NE;
+    break;
+  case 0x6:
+    return llvm::Constant::getNullValue(Ty); // FALSE
+  case 0x7:
+    return llvm::Constant::getAllOnesValue(Ty); // TRUE
+  default:
+    llvm_unreachable("Unexpected XOP vpcom/vpcomu predicate");
+  }
+
+  Value *Cmp = CGF.Builder.CreateICmp(Pred, Op0, Op1);
+  Value *Res = CGF.Builder.CreateSExt(Cmp, Ty);
+  return Res;
+}
+
 static Value *EmitX86Select(CodeGenFunction &CGF,
                             Value *Mask, Value *Op0, Value *Op1) {
 
@@ -10928,6 +10968,16 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
     return EmitX86MaskedCompare(*this, CC, false, Ops);
   }
+  case X86::BI__builtin_ia32_vpcomb:
+  case X86::BI__builtin_ia32_vpcomw:
+  case X86::BI__builtin_ia32_vpcomd:
+  case X86::BI__builtin_ia32_vpcomq:
+    return EmitX86vpcom(*this, Ops, true);
+  case X86::BI__builtin_ia32_vpcomub:
+  case X86::BI__builtin_ia32_vpcomuw:
+  case X86::BI__builtin_ia32_vpcomud:
+  case X86::BI__builtin_ia32_vpcomuq:
+    return EmitX86vpcom(*this, Ops, false);
 
   case X86::BI__builtin_ia32_kortestcqi:
   case X86::BI__builtin_ia32_kortestchi:
diff --git a/test/CodeGen/xop-builtins-cmp.c b/test/CodeGen/xop-builtins-cmp.c
index a805352ad3..4fbe6d0e70 100644
--- a/test/CodeGen/xop-builtins-cmp.c
+++ b/test/CodeGen/xop-builtins-cmp.c
@@ -8,49 +8,57 @@
 
 __m128i test_mm_comlt_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comlt_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 0)
+  // CHECK: icmp ult <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comlt_epu8(a, b);
 }
 
 __m128i test_mm_comlt_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comlt_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 0)
+  // CHECK: icmp ult <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comlt_epu16(a, b);
 }
 
 __m128i test_mm_comlt_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comlt_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 0)
+  // CHECK: icmp ult <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comlt_epu32(a, b);
 }
 
 __m128i test_mm_comlt_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comlt_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 0)
+  // CHECK: icmp ult <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comlt_epu64(a, b);
 }
 
 __m128i test_mm_comlt_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comlt_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 0)
+  // CHECK: icmp slt <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comlt_epi8(a, b);
 }
 
 __m128i test_mm_comlt_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comlt_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 0)
+  // CHECK: icmp slt <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comlt_epi16(a, b);
 }
 
 __m128i test_mm_comlt_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comlt_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 0)
+  // CHECK: icmp slt <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comlt_epi32(a, b);
 }
 
 __m128i test_mm_comlt_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comlt_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 0)
+  // CHECK: icmp slt <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comlt_epi64(a, b);
 }
 
@@ -58,49 +66,57 @@ __m128i test_mm_comlt_epi64(__m128i a, __m128i b) {
 
 __m128i test_mm_comle_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comle_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 1)
+  // CHECK: icmp ule <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comle_epu8(a, b);
 }
 
 __m128i test_mm_comle_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comle_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 1)
+  // CHECK: icmp ule <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comle_epu16(a, b);
 }
 
 __m128i test_mm_comle_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comle_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 1)
+  // CHECK: icmp ule <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comle_epu32(a, b);
 }
 
 __m128i test_mm_comle_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comle_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 1)
+  // CHECK: icmp ule <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comle_epu64(a, b);
 }
 
 __m128i test_mm_comle_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comle_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 1)
+  // CHECK: icmp sle <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comle_epi8(a, b);
 }
 
 __m128i test_mm_comle_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comle_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 1)
+  // CHECK: icmp sle <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comle_epi16(a, b);
 }
 
 __m128i test_mm_comle_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comle_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 1)
+  // CHECK: icmp sle <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comle_epi32(a, b);
 }
 
 __m128i test_mm_comle_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comle_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 1)
+  // CHECK: icmp sle <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comle_epi64(a, b);
 }
 
@@ -108,49 +124,57 @@ __m128i test_mm_comle_epi64(__m128i a, __m128i b) {
 
 __m128i test_mm_comgt_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comgt_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 2)
+  // CHECK: icmp ugt <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comgt_epu8(a, b);
 }
 
 __m128i test_mm_comgt_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comgt_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 2)
+  // CHECK: icmp ugt <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comgt_epu16(a, b);
 }
 
 __m128i test_mm_comgt_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comgt_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 2)
+  // CHECK: icmp ugt <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comgt_epu32(a, b);
 }
 
 __m128i test_mm_comgt_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comgt_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 2)
+  // CHECK: icmp ugt <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comgt_epu64(a, b);
 }
 
 __m128i test_mm_comgt_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comgt_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 2)
+  // CHECK: icmp sgt <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comgt_epi8(a, b);
 }
 
 __m128i test_mm_comgt_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comgt_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 2)
+  // CHECK: icmp sgt <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comgt_epi16(a, b);
 }
 
 __m128i test_mm_comgt_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comgt_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 2)
+  // CHECK: icmp sgt <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comgt_epi32(a, b);
 }
 
 __m128i test_mm_comgt_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comgt_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 2)
+  // CHECK: icmp sgt <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comgt_epi64(a, b);
 }
 
@@ -158,49 +182,57 @@ __m128i test_mm_comgt_epi64(__m128i a, __m128i b) {
 
 __m128i test_mm_comge_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comge_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 3)
+  // CHECK: icmp uge <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comge_epu8(a, b);
 }
 
 __m128i test_mm_comge_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comge_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 3)
+  // CHECK: icmp uge <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comge_epu16(a, b);
 }
 
 __m128i test_mm_comge_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comge_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 3)
+  // CHECK: icmp uge <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comge_epu32(a, b);
 }
 
 __m128i test_mm_comge_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comge_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 3)
+  // CHECK: icmp uge <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comge_epu64(a, b);
 }
 
 __m128i test_mm_comge_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comge_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 3)
+  // CHECK: icmp sge <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comge_epi8(a, b);
 }
 
 __m128i test_mm_comge_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comge_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 3)
+  // CHECK: icmp sge <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comge_epi16(a, b);
 }
 
 __m128i test_mm_comge_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comge_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 3)
+  // CHECK: icmp sge <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comge_epi32(a, b);
 }
 
 __m128i test_mm_comge_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comge_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 3)
+  // CHECK: icmp sge <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comge_epi64(a, b);
 }
 
@@ -208,49 +240,57 @@ __m128i test_mm_comge_epi64(__m128i a, __m128i b) {
 
 __m128i test_mm_comeq_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comeq_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 4)
+  // CHECK: icmp eq <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comeq_epu8(a, b);
 }
 
 __m128i test_mm_comeq_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comeq_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 4)
+  // CHECK: icmp eq <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comeq_epu16(a, b);
 }
 
 __m128i test_mm_comeq_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comeq_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 4)
+  // CHECK: icmp eq <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comeq_epu32(a, b);
 }
 
 __m128i test_mm_comeq_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comeq_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 4)
+  // CHECK: icmp eq <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comeq_epu64(a, b);
 }
 
 __m128i test_mm_comeq_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comeq_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 4)
+  // CHECK: icmp eq <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comeq_epi8(a, b);
 }
 
 __m128i test_mm_comeq_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comeq_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 4)
+  // CHECK: icmp eq <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comeq_epi16(a, b);
 }
 
 __m128i test_mm_comeq_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comeq_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 4)
+  // CHECK: icmp eq <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comeq_epi32(a, b);
 }
 
 __m128i test_mm_comeq_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comeq_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 4)
+  // CHECK: icmp eq <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comeq_epi64(a, b);
 }
 
@@ -258,49 +298,57 @@ __m128i test_mm_comeq_epi64(__m128i a, __m128i b) {
 
 __m128i test_mm_comneq_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comneq_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 5)
+  // CHECK: icmp ne <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comneq_epu8(a, b);
 }
 
 __m128i test_mm_comneq_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comneq_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 5)
+  // CHECK: icmp ne <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comneq_epu16(a, b);
 }
 
 __m128i test_mm_comneq_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comneq_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 5)
+  // CHECK: icmp ne <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comneq_epu32(a, b);
 }
 
 __m128i test_mm_comneq_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comneq_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 5)
+  // CHECK: icmp ne <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comneq_epu64(a, b);
 }
 
 __m128i test_mm_comneq_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comneq_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 5)
+  // CHECK: icmp ne <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_comneq_epi8(a, b);
 }
 
 __m128i test_mm_comneq_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comneq_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 5)
+  // CHECK: icmp ne <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_comneq_epi16(a, b);
 }
 
 __m128i test_mm_comneq_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comneq_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 5)
+  // CHECK: icmp ne <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_comneq_epi32(a, b);
 }
 
 __m128i test_mm_comneq_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comneq_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 5)
+  // CHECK: icmp ne <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_comneq_epi64(a, b);
 }
 
@@ -308,49 +356,49 @@ __m128i test_mm_comneq_epi64(__m128i a, __m128i b) {
 
 __m128i test_mm_comfalse_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comfalse_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 6)
+  // CHECK: ret <2 x i64> zeroinitializer
   return _mm_comfalse_epu8(a, b);
 }
 
 __m128i test_mm_comfalse_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comfalse_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 6)
+  // CHECK: ret <2 x i64> zeroinitializer
   return _mm_comfalse_epu16(a, b);
 }
 
 __m128i test_mm_comfalse_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comfalse_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 6)
+  // CHECK: ret <2 x i64> zeroinitializer
   return _mm_comfalse_epu32(a, b);
 }
 
 __m128i test_mm_comfalse_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comfalse_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 6)
+  // CHECK: ret <2 x i64> zeroinitializer
   return _mm_comfalse_epu64(a, b);
 }
 
 __m128i test_mm_comfalse_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comfalse_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 6)
+  // CHECK: ret <2 x i64> zeroinitializer
   return _mm_comfalse_epi8(a, b);
 }
 
 __m128i test_mm_comfalse_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comfalse_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 6)
+  // CHECK: ret <2 x i64> zeroinitializer
   return _mm_comfalse_epi16(a, b);
 }
 
 __m128i test_mm_comfalse_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comfalse_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 6)
+  // CHECK: ret <2 x i64> zeroinitializer
   return _mm_comfalse_epi32(a, b);
 }
 
 __m128i test_mm_comfalse_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comfalse_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 6)
+  // CHECK: ret <2 x i64> zeroinitializer
   return _mm_comfalse_epi64(a, b);
 }
 
@@ -358,48 +406,48 @@ __m128i test_mm_comfalse_epi64(__m128i a, __m128i b) {
 
 __m128i test_mm_comtrue_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comtrue_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 7)
+  // CHECK: ret <2 x i64> <i64 -1, i64 -1>
   return _mm_comtrue_epu8(a, b);
 }
 
 __m128i test_mm_comtrue_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comtrue_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 7)
+  // CHECK: ret <2 x i64> <i64 -1, i64 -1>
   return _mm_comtrue_epu16(a, b);
 }
 
 __m128i test_mm_comtrue_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comtrue_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 7)
+  // CHECK: ret <2 x i64> <i64 -1, i64 -1>
   return _mm_comtrue_epu32(a, b);
 }
 
 __m128i test_mm_comtrue_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comtrue_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 7)
+  // CHECK: ret <2 x i64> <i64 -1, i64 -1>
   return _mm_comtrue_epu64(a, b);
 }
 
 __m128i test_mm_comtrue_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comtrue_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 7)
+  // CHECK: ret <2 x i64> <i64 -1, i64 -1>
   return _mm_comtrue_epi8(a, b);
 }
 
 __m128i test_mm_comtrue_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comtrue_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 7)
+  // CHECK: ret <2 x i64> <i64 -1, i64 -1>
   return _mm_comtrue_epi16(a, b);
 }
 
 __m128i test_mm_comtrue_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comtrue_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 7)
+  // CHECK: ret <2 x i64> <i64 -1, i64 -1>
   return _mm_comtrue_epi32(a, b);
 }
 
 __m128i test_mm_comtrue_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_comtrue_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 7)
+  // CHECK: ret <2 x i64> <i64 -1, i64 -1>
   return _mm_comtrue_epi64(a, b);
 }
diff --git a/test/CodeGen/xop-builtins.c b/test/CodeGen/xop-builtins.c
index e6a09007f7..01d77ce056 100644
--- a/test/CodeGen/xop-builtins.c
+++ b/test/CodeGen/xop-builtins.c
@@ -290,49 +290,57 @@ __m128i test_mm_sha_epi64(__m128i a, __m128i b) {
 
 __m128i test_mm_com_epu8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_com_epu8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomub(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 0)
+  // CHECK: icmp ult <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_com_epu8(a, b, 0);
 }
 
 __m128i test_mm_com_epu16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_com_epu16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomuw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 0)
+  // CHECK: icmp ult <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_com_epu16(a, b, 0);
 }
 
 __m128i test_mm_com_epu32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_com_epu32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomud(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 0)
+  // CHECK: icmp ult <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_com_epu32(a, b, 0);
 }
 
 __m128i test_mm_com_epu64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_com_epu64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomuq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 0)
+  // CHECK: icmp ult <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_com_epu64(a, b, 0);
 }
 
 __m128i test_mm_com_epi8(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_com_epi8
-  // CHECK: call <16 x i8> @llvm.x86.xop.vpcomb(<16 x i8> %{{.*}}, <16 x i8> %{{.*}}, i8 0)
+  // CHECK: icmp slt <16 x i8> %{{.*}}, %{{.*}}
+  // CHECK: sext <16 x i1> %{{.*}} to <16 x i8>
   return _mm_com_epi8(a, b, 0);
 }
 
 __m128i test_mm_com_epi16(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_com_epi16
-  // CHECK: call <8 x i16> @llvm.x86.xop.vpcomw(<8 x i16> %{{.*}}, <8 x i16> %{{.*}}, i8 0)
+  // CHECK: icmp slt <8 x i16> %{{.*}}, %{{.*}}
+  // CHECK: sext <8 x i1> %{{.*}} to <8 x i16>
   return _mm_com_epi16(a, b, 0);
 }
 
 __m128i test_mm_com_epi32(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_com_epi32
-  // CHECK: call <4 x i32> @llvm.x86.xop.vpcomd(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, i8 0)
+  // CHECK: icmp slt <4 x i32> %{{.*}}, %{{.*}}
+  // CHECK: sext <4 x i1> %{{.*}} to <4 x i32>
   return _mm_com_epi32(a, b, 0);
 }
 
 __m128i test_mm_com_epi64(__m128i a, __m128i b) {
   // CHECK-LABEL: test_mm_com_epi64
-  // CHECK: call <2 x i64> @llvm.x86.xop.vpcomq(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, i8 0)
+  // CHECK: icmp slt <2 x i64> %{{.*}}, %{{.*}}
+  // CHECK: sext <2 x i1> %{{.*}} to <2 x i64>
   return _mm_com_epi64(a, b, 0);
 }