[X86] Implement _builtin_ia32_vfmaddss and _builtin_ia32_vfmaddsd with native IR...

This generates some extra zeroing currently, but we should be able to quickly address that with some isel patterns.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@336417 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 56cf6a9952491b689954b8a96e0e326a49d6ee21..f2efca8cc9c79daa9392593d8f151bb4c8675a0e 100644 (file)
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -9134,6 +9134,16 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     Value *Res = Builder.CreateCall(FMA, {A, B, C} );
     return Builder.CreateInsertElement(Ops[0], Res, (uint64_t)0);
   }
+  case X86::BI__builtin_ia32_vfmaddss:
+  case X86::BI__builtin_ia32_vfmaddsd: {
+    Value *A = Builder.CreateExtractElement(Ops[0], (uint64_t)0);
+    Value *B = Builder.CreateExtractElement(Ops[1], (uint64_t)0);
+    Value *C = Builder.CreateExtractElement(Ops[2], (uint64_t)0);
+    Function *FMA = CGM.getIntrinsic(Intrinsic::fma, A->getType());
+    Value *Res = Builder.CreateCall(FMA, {A, B, C} );
+    Value *Zero = Constant::getNullValue(Ops[0]->getType());
+    return Builder.CreateInsertElement(Zero, Res, (uint64_t)0);
+  }
   case X86::BI__builtin_ia32_vfmaddps:
   case X86::BI__builtin_ia32_vfmaddpd:
   case X86::BI__builtin_ia32_vfmaddps256:

diff --git a/test/CodeGen/fma4-builtins.c b/test/CodeGen/fma4-builtins.c
index 8d88a51519302f468a658fb97f6b2c2bb55a2a85..8078a6ed69b2899199c114f79ccd94708a840107 100644 (file)
--- a/test/CodeGen/fma4-builtins.c
+++ b/test/CodeGen/fma4-builtins.c
@@ -17,13 +17,21 @@ __m128d test_mm_macc_pd(__m128d a, __m128d b, __m128d c) {
 
 __m128 test_mm_macc_ss(__m128 a, __m128 b, __m128 c) {
   // CHECK-LABEL: test_mm_macc_ss
-  // CHECK: @llvm.x86.fma4.vfmadd.ss
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK: call float @llvm.fma.f32(float %{{.*}}, float %{{.*}}, float %{{.*}})
+  // CHECK: insertelement <4 x float> zeroinitializer, float %{{.*}}, i64 0
   return _mm_macc_ss(a, b, c);
 }
 
 __m128d test_mm_macc_sd(__m128d a, __m128d b, __m128d c) {
   // CHECK-LABEL: test_mm_macc_sd
-  // CHECK: @llvm.x86.fma4.vfmadd.sd
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK: call double @llvm.fma.f64(double %{{.*}}, double %{{.*}}, double %{{.*}})
+  // CHECK: insertelement <2 x double> zeroinitializer, double %{{.*}}, i64 0
   return _mm_macc_sd(a, b, c);
 }
 
@@ -44,14 +52,22 @@ __m128d test_mm_msub_pd(__m128d a, __m128d b, __m128d c) {
 __m128 test_mm_msub_ss(__m128 a, __m128 b, __m128 c) {
   // CHECK-LABEL: test_mm_msub_ss
   // CHECK: [[NEG:%.+]] = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{.+}}
-  // CHECK: @llvm.x86.fma4.vfmadd.ss(<4 x float> %{{.+}}, <4 x float> %{{.+}}, <4 x float> [[NEG]])
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK: [[C:%.+]] = extractelement <4 x float> [[NEG]], i64 0
+  // CHECK: call float @llvm.fma.f32(float %{{.*}}, float %{{.*}}, float [[C]])
+  // CHECK: insertelement <4 x float> zeroinitializer, float %{{.*}}, i64 0
   return _mm_msub_ss(a, b, c);
 }
 
 __m128d test_mm_msub_sd(__m128d a, __m128d b, __m128d c) {
   // CHECK-LABEL: test_mm_msub_sd
   // CHECK: [[NEG:%.+]] = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{.+}}
-  // CHECK: @llvm.x86.fma4.vfmadd.sd(<2 x double> %{{.+}}, <2 x double> %{{.+}}, <2 x double> [[NEG]])
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK: [[C:%.+]] = extractelement <2 x double> [[NEG]], i64 0
+  // CHECK: call double @llvm.fma.f64(double %{{.*}}, double %{{.*}}, double [[C]])
+  // CHECK: insertelement <2 x double> zeroinitializer, double %{{.*}}, i64 0
   return _mm_msub_sd(a, b, c);
 }
 
@@ -72,14 +88,22 @@ __m128d test_mm_nmacc_pd(__m128d a, __m128d b, __m128d c) {
 __m128 test_mm_nmacc_ss(__m128 a, __m128 b, __m128 c) {
   // CHECK-LABEL: test_mm_nmacc_ss
   // CHECK: [[NEG:%.+]] = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{.+}}
-  // CHECK: @llvm.x86.fma4.vfmadd.ss(<4 x float> [[NEG]], <4 x float> %{{.+}}, <4 x float> %{{.+}})
+  // CHECK: [[A:%.+]] = extractelement <4 x float> [[NEG]], i64 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK: call float @llvm.fma.f32(float [[A]], float %{{.*}}, float %{{.*}})
+  // CHECK: insertelement <4 x float> zeroinitializer, float %{{.*}}, i64 0
   return _mm_nmacc_ss(a, b, c);
 }
 
 __m128d test_mm_nmacc_sd(__m128d a, __m128d b, __m128d c) {
   // CHECK-LABEL: test_mm_nmacc_sd
   // CHECK: [[NEG:%.+]] = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{.+}}
-  // CHECK: @llvm.x86.fma4.vfmadd.sd(<2 x double> [[NEG]], <2 x double> %{{.+}}, <2 x double> %{{.+}})
+  // CHECK: [[A:%.+]] = extractelement <2 x double> [[NEG]], i64 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK: call double @llvm.fma.f64(double [[A]], double %{{.*}}, double %{{.*}})
+  // CHECK: insertelement <2 x double> zeroinitializer, double %{{.*}}, i64 0
   return _mm_nmacc_sd(a, b, c);
 }
 
@@ -103,7 +127,11 @@ __m128 test_mm_nmsub_ss(__m128 a, __m128 b, __m128 c) {
   // CHECK-LABEL: test_mm_nmsub_ss
   // CHECK: [[NEG:%.+]] = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{.+}}
   // CHECK: [[NEG2:%.+]] = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{.+}}
-  // CHECK: @llvm.x86.fma4.vfmadd.ss(<4 x float> [[NEG]], <4 x float> %{{.+}}, <4 x float> [[NEG2]])
+  // CHECK: [[A:%.+]] = extractelement <4 x float> [[NEG]], i64 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK: [[C:%.+]] = extractelement <4 x float> [[NEG2]], i64 0
+  // CHECK: call float @llvm.fma.f32(float [[A]], float %{{.*}}, float [[C]])
+  // CHECK: insertelement <4 x float> zeroinitializer, float %{{.*}}, i64 0
   return _mm_nmsub_ss(a, b, c);
 }
 
@@ -111,7 +139,11 @@ __m128d test_mm_nmsub_sd(__m128d a, __m128d b, __m128d c) {
   // CHECK-LABEL: test_mm_nmsub_sd
   // CHECK: [[NEG:%.+]] = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{.+}}
   // CHECK: [[NEG2:%.+]] = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{.+}}
-  // CHECK: @llvm.x86.fma4.vfmadd.sd(<2 x double> [[NEG]], <2 x double> %{{.+}}, <2 x double> [[NEG2]])
+  // CHECK: [[A:%.+]] = extractelement <2 x double> [[NEG]], i64 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK: [[C:%.+]] = extractelement <2 x double> [[NEG2]], i64 0
+  // CHECK: call double @llvm.fma.f64(double [[A]], double %{{.*}}, double [[C]])
+  // CHECK: insertelement <2 x double> zeroinitializer, double %{{.*}}, i64 0
   return _mm_nmsub_sd(a, b, c);
 }