[AVX-512] Use scalar vfmsub/vfnmsub mask3 intrinsics instead of inverting the mask...

Summary: Inverting the mask argument does not reflect the intended semantics of the intrinsic.

Reviewers: igorb, delena

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D26019

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@286733 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 07b20b3cf5d140c320d4e370aa90ffb28c44b947..e65fb1145d4e4d5d1423216ea573467692d95a95 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -1743,6 +1743,10 @@ TARGET_BUILTIN(__builtin_ia32_vfmaddss3_mask3, "V4fV4fV4fV4fUcIi", "", "avx512f"
 TARGET_BUILTIN(__builtin_ia32_vfmaddsd3_mask,  "V2dV2dV2dV2dUcIi", "", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_vfmaddsd3_maskz, "V2dV2dV2dV2dUcIi", "", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_vfmaddsd3_mask3, "V2dV2dV2dV2dUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfmsubsd3_mask3, "V2dV2dV2dV2dUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfmsubss3_mask3, "V4fV4fV4fV4fUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfnmsubsd3_mask3, "V2dV2dV2dV2dUcIi", "", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_vfnmsubss3_mask3, "V4fV4fV4fV4fUcIi", "", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_permvarhi512_mask, "V32sV32sV32sV32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_permvardf512_mask, "V8dV8dV8LLiV8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_permvardi512_mask, "V8LLiV8LLiV8LLiV8LLiUc","","avx512f")

diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h
index d7f8cf42bc1622c305edddd1bba17cc84d3c6ef7..1992bd84797628dc4a2c9233912fd09f7c1463e4 100644 (file)
--- a/lib/Headers/avx512fintrin.h
+++ b/lib/Headers/avx512fintrin.h
@@ -8470,17 +8470,17 @@ _mm_maskz_fmsub_ss (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C)
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_mask3_fmsub_ss (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U)
 {
- return (__m128) __builtin_ia32_vfmaddss3_mask3 ((__v4sf) __W,
+ return (__m128) __builtin_ia32_vfmsubss3_mask3 ((__v4sf) __W,
           (__v4sf) __X,
-          -(__v4sf) __Y,
+          (__v4sf) __Y,
           (__mmask8) __U,
           _MM_FROUND_CUR_DIRECTION);
 }
 
 #define _mm_mask3_fmsub_round_ss(W, X, Y, U, R) __extension__ ({\
-  (__m128)__builtin_ia32_vfmaddss3_mask3((__v4sf)(__m128)(W), \
+  (__m128)__builtin_ia32_vfmsubss3_mask3((__v4sf)(__m128)(W), \
                                          (__v4sf)(__m128)(X), \
-                                         -(__v4sf)(__m128)(Y), (__mmask8)(U), \
+                                         (__v4sf)(__m128)(Y), (__mmask8)(U), \
                                          (int)(R)); })
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS
@@ -8566,17 +8566,17 @@ _mm_maskz_fnmsub_ss (__mmask8 __U, __m128 __A, __m128 __B, __m128 __C)
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_mask3_fnmsub_ss (__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U)
 {
- return (__m128) __builtin_ia32_vfmaddss3_mask3 (-(__v4sf) __W,
+ return (__m128) __builtin_ia32_vfnmsubss3_mask3 ((__v4sf) __W,
           (__v4sf) __X,
-          -(__v4sf) __Y,
+          (__v4sf) __Y,
           (__mmask8) __U,
           _MM_FROUND_CUR_DIRECTION);
 }
 
 #define _mm_mask3_fnmsub_round_ss(W, X, Y, U, R) __extension__({\
-  (__m128)__builtin_ia32_vfmaddss3_mask3(-(__v4sf)(__m128)(W), \
+  (__m128)__builtin_ia32_vfnmsubss3_mask3((__v4sf)(__m128)(W), \
                                          (__v4sf)(__m128)(X), \
-                                         -(__v4sf)(__m128)(Y), (__mmask8)(U), \
+                                         (__v4sf)(__m128)(Y), (__mmask8)(U), \
                                          (int)(R)); })
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
@@ -8662,17 +8662,17 @@ _mm_maskz_fmsub_sd (__mmask8 __U, __m128d __A, __m128d __B, __m128d __C)
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_mask3_fmsub_sd (__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U)
 {
- return (__m128d) __builtin_ia32_vfmaddsd3_mask3 ((__v2df) __W,
+ return (__m128d) __builtin_ia32_vfmsubsd3_mask3 ((__v2df) __W,
           (__v2df) __X,
-          -(__v2df) __Y,
+          (__v2df) __Y,
           (__mmask8) __U,
           _MM_FROUND_CUR_DIRECTION);
 }
 
 #define _mm_mask3_fmsub_round_sd(W, X, Y, U, R) __extension__ ({\
-  (__m128d)__builtin_ia32_vfmaddsd3_mask3((__v2df)(__m128d)(W), \
+  (__m128d)__builtin_ia32_vfmsubsd3_mask3((__v2df)(__m128d)(W), \
                                           (__v2df)(__m128d)(X), \
-                                          -(__v2df)(__m128d)(Y), \
+                                          (__v2df)(__m128d)(Y), \
                                           (__mmask8)(U), (int)(R)); })
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
@@ -8759,17 +8759,17 @@ _mm_maskz_fnmsub_sd (__mmask8 __U, __m128d __A, __m128d __B, __m128d __C)
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_mask3_fnmsub_sd (__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U)
 {
- return (__m128d) __builtin_ia32_vfmaddsd3_mask3 (-(__v2df) (__W),
+ return (__m128d) __builtin_ia32_vfnmsubsd3_mask3 ((__v2df) (__W),
           (__v2df) __X,
-          -(__v2df) (__Y),
+          (__v2df) (__Y),
           (__mmask8) __U,
           _MM_FROUND_CUR_DIRECTION);
 }
 
 #define _mm_mask3_fnmsub_round_sd(W, X, Y, U, R) __extension__({\
-  (__m128d)__builtin_ia32_vfmaddsd3_mask3(-(__v2df)(__m128d)(W), \
+  (__m128d)__builtin_ia32_vfnmsubsd3_mask3((__v2df)(__m128d)(W), \
                                           (__v2df)(__m128d)(X), \
-                                          -(__v2df)(__m128d)(Y), \
+                                          (__v2df)(__m128d)(Y), \
                                           (__mmask8)(U), (int)(R)); })
 
 #define _mm512_permutex_pd(X, C) __extension__ ({ \

diff --git a/test/CodeGen/avx512f-builtins.c b/test/CodeGen/avx512f-builtins.c
index 08011772067f9164bec2a11800dde794bcdc4819..1e3baf0d6a947502aa874e909a2bc5b6f4594184 100644 (file)
--- a/test/CodeGen/avx512f-builtins.c
+++ b/test/CodeGen/avx512f-builtins.c
@@ -5813,13 +5813,13 @@ __m128 test_mm_maskz_fmsub_round_ss(__mmask8 __U, __m128 __A, __m128 __B, __m128
 
 __m128 test_mm_mask3_fmsub_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
   // CHECK-LABEL: @test_mm_mask3_fmsub_ss
-  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmsub.ss
   return _mm_mask3_fmsub_ss(__W, __X, __Y, __U);
 }
 
 __m128 test_mm_mask3_fmsub_round_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
   // CHECK-LABEL: @test_mm_mask3_fmsub_round_ss
-  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  // CHECK: @llvm.x86.avx512.mask3.vfmsub.ss
   return _mm_mask3_fmsub_round_ss(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
 }
 
@@ -5885,13 +5885,13 @@ __m128 test_mm_maskz_fnmsub_round_ss(__mmask8 __U, __m128 __A, __m128 __B, __m12
 
 __m128 test_mm_mask3_fnmsub_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
   // CHECK-LABEL: @test_mm_mask3_fnmsub_ss
-  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  // CHECK: @llvm.x86.avx512.mask3.vfnmsub.ss
   return _mm_mask3_fnmsub_ss(__W, __X, __Y, __U);
 }
 
 __m128 test_mm_mask3_fnmsub_round_ss(__m128 __W, __m128 __X, __m128 __Y, __mmask8 __U){
   // CHECK-LABEL: @test_mm_mask3_fnmsub_round_ss
-  // CHECK: @llvm.x86.avx512.mask3.vfmadd.ss
+  // CHECK: @llvm.x86.avx512.mask3.vfnmsub.ss
   return _mm_mask3_fnmsub_round_ss(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
 }
 
@@ -5957,13 +5957,13 @@ __m128d test_mm_maskz_fmsub_round_sd(__mmask8 __U, __m128d __A, __m128d __B, __m
 
 __m128d test_mm_mask3_fmsub_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
   // CHECK-LABEL: @test_mm_mask3_fmsub_sd
-  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmsub.sd
   return _mm_mask3_fmsub_sd(__W, __X, __Y, __U);
 }
 
 __m128d test_mm_mask3_fmsub_round_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
   // CHECK-LABEL: @test_mm_mask3_fmsub_round_sd
-  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  // CHECK: @llvm.x86.avx512.mask3.vfmsub.sd
   return _mm_mask3_fmsub_round_sd(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
 }
 
@@ -6029,13 +6029,13 @@ __m128d test_mm_maskz_fnmsub_round_sd(__mmask8 __U, __m128d __A, __m128d __B, __
 
 __m128d test_mm_mask3_fnmsub_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
   // CHECK-LABEL: @test_mm_mask3_fnmsub_sd
-  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  // CHECK: @llvm.x86.avx512.mask3.vfnmsub.sd
   return _mm_mask3_fnmsub_sd(__W, __X, __Y, __U);
 }
 
 __m128d test_mm_mask3_fnmsub_round_sd(__m128d __W, __m128d __X, __m128d __Y, __mmask8 __U){
   // CHECK-LABEL: @test_mm_mask3_fnmsub_round_sd
-  // CHECK: @llvm.x86.avx512.mask3.vfmadd.sd
+  // CHECK: @llvm.x86.avx512.mask3.vfnmsub.sd
   return _mm_mask3_fnmsub_round_sd(__W, __X, __Y, __U, _MM_FROUND_CUR_DIRECTION);
 }