[X86] Use separate builtins for fma4 scalar intrinsics. Use negations to remove some of the scalar fma3 builtins.

fma4 instructions zero the upper bits of the xmm register. fma3 instructions leave the bits unmodified. This requires separate builtins for the different semantics.

While we're cleaning up the scalar builtins this also removes the fma3 fmsub/fnmadd/fnmsub builtins by using negates in the header file.

llvm-svn: 318985

2 files changed, 30 insertions, 14 deletions

diff --git a/clang/test/CodeGen/fma-builtins.c b/clang/test/CodeGen/fma-builtins.c
index d5905d33a97..6f792a78d84 100644
--- a/clang/test/CodeGen/fma-builtins.c
+++ b/clang/test/CodeGen/fma-builtins.c
@@ -43,13 +43,15 @@ __m128d test_mm_fmsub_pd(__m128d a, __m128d b, __m128d c) {
 
 __m128 test_mm_fmsub_ss(__m128 a, __m128 b, __m128 c) {
   // CHECK-LABEL: test_mm_fmsub_ss
-  // CHECK: @llvm.x86.fma.vfmsub.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.fma.vfmadd.ss(<4 x float> %{{.+}}, <4 x float> %{{.+}}, <4 x float> [[NEG]])
   return _mm_fmsub_ss(a, b, c);
 }
 
 __m128d test_mm_fmsub_sd(__m128d a, __m128d b, __m128d c) {
   // CHECK-LABEL: test_mm_fmsub_sd
-  // CHECK: @llvm.x86.fma.vfmsub.sd
+  // CHECK: [[NEG:%.+]] = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{.+}}
+  // CHECK: @llvm.x86.fma.vfmadd.sd(<2 x double> %{{.+}}, <2 x double> %{{.+}}, <2 x double> [[NEG]])
   return _mm_fmsub_sd(a, b, c);
 }
 
@@ -69,13 +71,15 @@ __m128d test_mm_fnmadd_pd(__m128d a, __m128d b, __m128d c) {
 
 __m128 test_mm_fnmadd_ss(__m128 a, __m128 b, __m128 c) {
   // CHECK-LABEL: test_mm_fnmadd_ss
-  // CHECK: @llvm.x86.fma.vfnmadd.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.fma.vfmadd.ss(<4 x float> %{{.+}}, <4 x float> [[NEG]], <4 x float> %{{.+}})
   return _mm_fnmadd_ss(a, b, c);
 }
 
 __m128d test_mm_fnmadd_sd(__m128d a, __m128d b, __m128d c) {
   // CHECK-LABEL: test_mm_fnmadd_sd
-  // CHECK: @llvm.x86.fma.vfnmadd.sd
+  // CHECK: [[NEG:%.+]] = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{.+}}
+  // CHECK: @llvm.x86.fma.vfmadd.sd(<2 x double> %{{.+}}, <2 x double> [[NEG]], <2 x double> %{{.+}})
   return _mm_fnmadd_sd(a, b, c);
 }
 
@@ -97,13 +101,17 @@ __m128d test_mm_fnmsub_pd(__m128d a, __m128d b, __m128d c) {
 
 __m128 test_mm_fnmsub_ss(__m128 a, __m128 b, __m128 c) {
   // CHECK-LABEL: test_mm_fnmsub_ss
-  // CHECK: @llvm.x86.fma.vfnmsub.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.fma.vfmadd.ss(<4 x float> %{{.+}}, <4 x float> [[NEG]], <4 x float> [[NEG2]])
   return _mm_fnmsub_ss(a, b, c);
 }
 
 __m128d test_mm_fnmsub_sd(__m128d a, __m128d b, __m128d c) {
   // CHECK-LABEL: test_mm_fnmsub_sd
-  // CHECK: @llvm.x86.fma.vfnmsub.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.fma.vfmadd.sd(<2 x double> %{{.+}}, <2 x double> [[NEG]], <2 x double> [[NEG2]])
   return _mm_fnmsub_sd(a, b, c);
 }
 
diff --git a/clang/test/CodeGen/fma4-builtins.c b/clang/test/CodeGen/fma4-builtins.c
index 68a2ec6e3a9..c848d4a751d 100644
--- a/clang/test/CodeGen/fma4-builtins.c
+++ b/clang/test/CodeGen/fma4-builtins.c
@@ -17,13 +17,13 @@ __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.fma.vfmadd.ss
+  // CHECK: @llvm.x86.fma4.vfmadd.ss
   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.fma.vfmadd.sd
+  // CHECK: @llvm.x86.fma4.vfmadd.sd
   return _mm_macc_sd(a, b, c);
 }
 
@@ -43,13 +43,15 @@ __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: @llvm.x86.fma.vfmsub.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]])
   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: @llvm.x86.fma.vfmsub.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]])
   return _mm_msub_sd(a, b, c);
 }
 
@@ -69,13 +71,15 @@ __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: @llvm.x86.fma.vfnmadd.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> %{{.+}})
   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: @llvm.x86.fma.vfnmadd.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> %{{.+}})
   return _mm_nmacc_sd(a, b, c);
 }
 
@@ -97,13 +101,17 @@ __m128d test_mm_nmsub_pd(__m128d a, __m128d b, __m128d c) {
 
 __m128 test_mm_nmsub_ss(__m128 a, __m128 b, __m128 c) {
   // CHECK-LABEL: test_mm_nmsub_ss
-  // CHECK: @llvm.x86.fma.vfnmsub.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]])
   return _mm_nmsub_ss(a, b, c);
 }
 
 __m128d test_mm_nmsub_sd(__m128d a, __m128d b, __m128d c) {
   // CHECK-LABEL: test_mm_nmsub_sd
-  // CHECK: @llvm.x86.fma.vfnmsub.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]])
   return _mm_nmsub_sd(a, b, c);
 }