[X86] Remove masking from dbpsadbw builtins, use select builtin instead.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@334385 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 45 insertions, 51 deletions

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index fc580df083..6a3016bcd3 100644
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -1745,9 +1745,9 @@ TARGET_BUILTIN(__builtin_ia32_kunpckhi, "UsUsUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_kxnorhi, "UsUsUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_kxorhi, "UsUsUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_palignr512, "V64cV64cV64cIi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_dbpsadbw128_mask, "V8sV16cV16cIiV8sUc", "nc", "avx512bw,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_dbpsadbw256_mask, "V16sV32cV32cIiV16sUs", "nc", "avx512bw,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_dbpsadbw512_mask, "V32sV64cV64cIiV32sUi", "nc", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_dbpsadbw128, "V8sV16cV16cIi", "nc", "avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_dbpsadbw256, "V16sV32cV32cIi", "nc", "avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_dbpsadbw512, "V32sV64cV64cIi", "nc", "avx512bw")
 TARGET_BUILTIN(__builtin_ia32_psadbw512, "V8LLiV64cV64c", "nc", "avx512bw")
 TARGET_BUILTIN(__builtin_ia32_compressdf512_mask, "V8dV8dV8dUc", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_compressdi512_mask, "V8LLiV8LLiV8LLiUc", "nc", "avx512f")
diff --git a/lib/Headers/avx512bwintrin.h b/lib/Headers/avx512bwintrin.h
index 2590ec7154..471ec38191 100644
--- a/lib/Headers/avx512bwintrin.h
+++ b/lib/Headers/avx512bwintrin.h
@@ -1819,22 +1819,18 @@ _mm512_mask_permutexvar_epi16 (__m512i __W, __mmask32 __M, __m512i __A,
                               (__v64qi)(__m512i)_mm512_setzero_si512())
 
 #define _mm512_dbsad_epu8(A, B, imm) \
-  (__m512i)__builtin_ia32_dbpsadbw512_mask((__v64qi)(__m512i)(A), \
-                                           (__v64qi)(__m512i)(B), (int)(imm), \
-                                           (__v32hi)_mm512_undefined_epi32(), \
-                                           (__mmask32)-1)
+  (__m512i)__builtin_ia32_dbpsadbw512((__v64qi)(__m512i)(A), \
+                                      (__v64qi)(__m512i)(B), (int)(imm))
 
 #define _mm512_mask_dbsad_epu8(W, U, A, B, imm) \
-  (__m512i)__builtin_ia32_dbpsadbw512_mask((__v64qi)(__m512i)(A), \
-                                           (__v64qi)(__m512i)(B), (int)(imm), \
-                                           (__v32hi)(__m512i)(W), \
-                                           (__mmask32)(U))
+  (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
+                                  (__v32hi)_mm512_dbsad_epu8((A), (B), (imm)), \
+                                  (__v32hi)(__m512i)(W))
 
 #define _mm512_maskz_dbsad_epu8(U, A, B, imm) \
-  (__m512i)__builtin_ia32_dbpsadbw512_mask((__v64qi)(__m512i)(A), \
-                                           (__v64qi)(__m512i)(B), (int)(imm), \
-                                           (__v32hi)_mm512_setzero_si512(), \
-                                           (__mmask32)(U))
+  (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
+                                  (__v32hi)_mm512_dbsad_epu8((A), (B), (imm)), \
+                                  (__v32hi)_mm512_setzero_si512())
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_sad_epu8 (__m512i __A, __m512i __B)
diff --git a/lib/Headers/avx512vlbwintrin.h b/lib/Headers/avx512vlbwintrin.h
index a04ef6f17c..f2dfb567dd 100644
--- a/lib/Headers/avx512vlbwintrin.h
+++ b/lib/Headers/avx512vlbwintrin.h
@@ -2714,40 +2714,32 @@ _mm256_mask_permutexvar_epi16 (__m256i __W, __mmask16 __M, __m256i __A,
                               (__v32qi)_mm256_setzero_si256())
 
 #define _mm_dbsad_epu8(A, B, imm) \
-  (__m128i)__builtin_ia32_dbpsadbw128_mask((__v16qi)(__m128i)(A), \
-                                           (__v16qi)(__m128i)(B), (int)(imm), \
-                                           (__v8hi)_mm_setzero_si128(), \
-                                           (__mmask8)-1)
+  (__m128i)__builtin_ia32_dbpsadbw128((__v16qi)(__m128i)(A), \
+                                      (__v16qi)(__m128i)(B), (int)(imm))
 
 #define _mm_mask_dbsad_epu8(W, U, A, B, imm) \
-  (__m128i)__builtin_ia32_dbpsadbw128_mask((__v16qi)(__m128i)(A), \
-                                           (__v16qi)(__m128i)(B), (int)(imm), \
-                                           (__v8hi)(__m128i)(W), \
-                                           (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectw_128((__mmask8)(U), \
+                                      (__v8hi)_mm_dbsad_epu8((A), (B), (imm)), \
+                                      (__v8hi)(__m128i)(W))
 
 #define _mm_maskz_dbsad_epu8(U, A, B, imm) \
-  (__m128i)__builtin_ia32_dbpsadbw128_mask((__v16qi)(__m128i)(A), \
-                                           (__v16qi)(__m128i)(B), (int)(imm), \
-                                           (__v8hi)_mm_setzero_si128(), \
-                                           (__mmask8)(U))
+  (__m128i)__builtin_ia32_selectw_128((__mmask8)(U), \
+                                      (__v8hi)_mm_dbsad_epu8((A), (B), (imm)), \
+                                      (__v8hi)_mm_setzero_si128())
 
 #define _mm256_dbsad_epu8(A, B, imm) \
-  (__m256i)__builtin_ia32_dbpsadbw256_mask((__v32qi)(__m256i)(A), \
-                                           (__v32qi)(__m256i)(B), (int)(imm), \
-                                           (__v16hi)_mm256_setzero_si256(), \
-                                           (__mmask16)-1)
+  (__m256i)__builtin_ia32_dbpsadbw256((__v32qi)(__m256i)(A), \
+                                      (__v32qi)(__m256i)(B), (int)(imm))
 
 #define _mm256_mask_dbsad_epu8(W, U, A, B, imm) \
-  (__m256i)__builtin_ia32_dbpsadbw256_mask((__v32qi)(__m256i)(A), \
-                                           (__v32qi)(__m256i)(B), (int)(imm), \
-                                           (__v16hi)(__m256i)(W), \
-                                           (__mmask16)(U))
+  (__m256i)__builtin_ia32_selectw_256((__mmask16)(U), \
+                                  (__v16hi)_mm256_dbsad_epu8((A), (B), (imm)), \
+                                  (__v16hi)(__m256i)(W))
 
 #define _mm256_maskz_dbsad_epu8(U, A, B, imm) \
-  (__m256i)__builtin_ia32_dbpsadbw256_mask((__v32qi)(__m256i)(A), \
-                                           (__v32qi)(__m256i)(B), (int)(imm), \
-                                           (__v16hi)_mm256_setzero_si256(), \
-                                           (__mmask16)(U))
+  (__m256i)__builtin_ia32_selectw_256((__mmask16)(U), \
+                                  (__v16hi)_mm256_dbsad_epu8((A), (B), (imm)), \
+                                  (__v16hi)_mm256_setzero_si256())
 
 #undef __DEFAULT_FN_ATTRS
 
diff --git a/lib/Sema/SemaChecking.cpp b/lib/Sema/SemaChecking.cpp
index bf1e3a8b4d..8730408c8d 100644
--- a/lib/Sema/SemaChecking.cpp
+++ b/lib/Sema/SemaChecking.cpp
@@ -2838,9 +2838,9 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_shufps:
   case X86::BI__builtin_ia32_shufps256:
   case X86::BI__builtin_ia32_shufps512:
-  case X86::BI__builtin_ia32_dbpsadbw128_mask:
-  case X86::BI__builtin_ia32_dbpsadbw256_mask:
-  case X86::BI__builtin_ia32_dbpsadbw512_mask:
+  case X86::BI__builtin_ia32_dbpsadbw128:
+  case X86::BI__builtin_ia32_dbpsadbw256:
+  case X86::BI__builtin_ia32_dbpsadbw512:
   case X86::BI__builtin_ia32_vpshldd128_mask:
   case X86::BI__builtin_ia32_vpshldd256_mask:
   case X86::BI__builtin_ia32_vpshldd512_mask:
diff --git a/test/CodeGen/avx512bw-builtins.c b/test/CodeGen/avx512bw-builtins.c
index 6803652a79..7d2e1fadf4 100644
--- a/test/CodeGen/avx512bw-builtins.c
+++ b/test/CodeGen/avx512bw-builtins.c
@@ -1934,19 +1934,21 @@ __m512i test_mm512_maskz_alignr_epi8(__mmask64 __U, __m512i __A,__m512i __B){
 
 __m512i test_mm512_mm_dbsad_epu8(__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mm_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.512
+  // CHECK: @llvm.x86.avx512.dbpsadbw.512
   return _mm512_dbsad_epu8(__A, __B, 170); 
 }
 
 __m512i test_mm512_mm_mask_dbsad_epu8(__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mm_mask_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.512
+  // CHECK: @llvm.x86.avx512.dbpsadbw.512
+  //CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
   return _mm512_mask_dbsad_epu8(__W, __U, __A, __B, 170); 
 }
 
 __m512i test_mm512_mm_maskz_dbsad_epu8(__mmask32 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mm_maskz_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.512
+  // CHECK: @llvm.x86.avx512.dbpsadbw.512
+  //CHECK: select <32 x i1> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}}
   return _mm512_maskz_dbsad_epu8(__U, __A, __B, 170); 
 }
 
diff --git a/test/CodeGen/avx512vlbw-builtins.c b/test/CodeGen/avx512vlbw-builtins.c
index 9cc58b6505..427037cafb 100644
--- a/test/CodeGen/avx512vlbw-builtins.c
+++ b/test/CodeGen/avx512vlbw-builtins.c
@@ -2964,37 +2964,41 @@ __m256i test_mm256_maskz_alignr_epi8(__mmask32 __U, __m256i __A, __m256i __B) {
 
 __m128i test_mm_dbsad_epu8(__m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.128
+  // CHECK: @llvm.x86.avx512.dbpsadbw.128
   return _mm_dbsad_epu8(__A, __B, 170); 
 }
 
 __m128i test_mm_mask_dbsad_epu8(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.128
+  // CHECK: @llvm.x86.avx512.dbpsadbw.128
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_mask_dbsad_epu8(__W, __U, __A, __B, 170); 
 }
 
 __m128i test_mm_maskz_dbsad_epu8(__mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.128
+  // CHECK: @llvm.x86.avx512.dbpsadbw.128
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_maskz_dbsad_epu8(__U, __A, __B, 170); 
 }
 
 __m256i test_mm256_dbsad_epu8(__m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.256
+  // CHECK: @llvm.x86.avx512.dbpsadbw.256
   return _mm256_dbsad_epu8(__A, __B, 170); 
 }
 
 __m256i test_mm256_mask_dbsad_epu8(__m256i __W, __mmask16 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.256
+  // CHECK: @llvm.x86.avx512.dbpsadbw.256
+  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_mask_dbsad_epu8(__W, __U, __A, __B, 170); 
 }
 
 __m256i test_mm256_maskz_dbsad_epu8(__mmask16 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_dbsad_epu8
-  // CHECK: @llvm.x86.avx512.mask.dbpsadbw.256
+  // CHECK: @llvm.x86.avx512.dbpsadbw.256
+  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_maskz_dbsad_epu8(__U, __A, __B, 170); 
 }
 __mmask8 test_mm_movepi16_mask(__m128i __A) {