[AVX-512] Remove masked 128/256-bit builtins for vpmaddwd and vpmaddubsw. Replace...

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@285516 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 1268592e65200cfed9e5b02e8ef9c8d71386004d..2cc62bcd05221f89bfd23f74dd23fc0ff92fe38a 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -1314,10 +1314,6 @@ TARGET_BUILTIN(__builtin_ia32_reduceps128_mask, "V4fV4fIiV4fUc", "", "avx512vl,a
 TARGET_BUILTIN(__builtin_ia32_reduceps256_mask, "V8fV8fIiV8fUc", "", "avx512vl,avx512dq")
 TARGET_BUILTIN(__builtin_ia32_reducesd_mask, "V2dV2dV2dV2dUcIiIi", "", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_reducess_mask, "V4fV4fV4fV4fUcIiIi", "", "avx512dq")
-TARGET_BUILTIN(__builtin_ia32_pmaddubsw128_mask, "V8sV16cV16cV8sUc", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmaddubsw256_mask, "V16sV32cV32cV16sUs", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmaddwd128_mask, "V4iV8sV8sV4iUc", "", "avx512vl,avx512bw")
-TARGET_BUILTIN(__builtin_ia32_pmaddwd256_mask, "V8iV16sV16sV8iUc", "", "avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovswb128_mask, "V16cV8sV16cUc", "", "avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovswb256_mask, "V16cV16sV16cUs", "", "avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovuswb128_mask, "V16cV8sV16cUc", "", "avx512vl,avx512bw")

diff --git a/lib/Headers/avx512vlbwintrin.h b/lib/Headers/avx512vlbwintrin.h
index 427dbeae6ba2ee5248ba016c3d6ec55090e7749c..ed92efa9baea3194b54be6881dbafc6cfe13cdb7 100644 (file)
--- a/lib/Headers/avx512vlbwintrin.h
+++ b/lib/Headers/avx512vlbwintrin.h
@@ -1664,69 +1664,60 @@ _mm256_maskz_permutex2var_epi16 (__mmask16 __U, __m256i __A,
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_maddubs_epi16 (__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
-  return (__m128i) __builtin_ia32_pmaddubsw128_mask ((__v16qi) __X,
-               (__v16qi) __Y,
-               (__v8hi) __W,
-               (__mmask8) __U);
+_mm_mask_maddubs_epi16(__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
+  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+                                            (__v8hi)_mm_maddubs_epi16(__X, __Y),
+                                            (__v8hi)__W);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_maddubs_epi16 (__mmask8 __U, __m128i __X, __m128i __Y) {
-  return (__m128i) __builtin_ia32_pmaddubsw128_mask ((__v16qi) __X,
-               (__v16qi) __Y,
-              (__v8hi) _mm_setzero_si128(),
-               (__mmask8) __U);
+_mm_maskz_maddubs_epi16(__mmask8 __U, __m128i __X, __m128i __Y) {
+  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__U,
+                                            (__v8hi)_mm_maddubs_epi16(__X, __Y),
+                                            (__v8hi)_mm_setzero_si128());
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_maddubs_epi16 (__m256i __W, __mmask16 __U, __m256i __X,
-         __m256i __Y) {
-  return (__m256i) __builtin_ia32_pmaddubsw256_mask ((__v32qi) __X,
-               (__v32qi) __Y,
-               (__v16hi) __W,
-               (__mmask16) __U);
+_mm256_mask_maddubs_epi16(__m256i __W, __mmask16 __U, __m256i __X,
+                          __m256i __Y) {
+  return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+                                        (__v16hi)_mm256_maddubs_epi16(__X, __Y),
+                                        (__v16hi)__W);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_maddubs_epi16 (__mmask16 __U, __m256i __X, __m256i __Y) {
-  return (__m256i) __builtin_ia32_pmaddubsw256_mask ((__v32qi) __X,
-               (__v32qi) __Y,
-               (__v16hi) _mm256_setzero_si256(),
-               (__mmask16) __U);
+_mm256_maskz_maddubs_epi16(__mmask16 __U, __m256i __X, __m256i __Y) {
+  return (__m256i)__builtin_ia32_selectw_256((__mmask16)__U,
+                                        (__v16hi)_mm256_maddubs_epi16(__X, __Y),
+                                        (__v16hi)_mm256_setzero_si256());
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_mask_madd_epi16 (__m128i __W, __mmask8 __U, __m128i __A,
-         __m128i __B) {
-  return (__m128i) __builtin_ia32_pmaddwd128_mask ((__v8hi) __A,
-               (__v8hi) __B,
-               (__v4si) __W,
-               (__mmask8) __U);
+_mm_mask_madd_epi16(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U,
+                                             (__v4si)_mm_madd_epi16(__A, __B),
+                                             (__v4si)__W);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_maskz_madd_epi16 (__mmask8 __U, __m128i __A, __m128i __B) {
-  return (__m128i) __builtin_ia32_pmaddwd128_mask ((__v8hi) __A,
-               (__v8hi) __B,
-               (__v4si) _mm_setzero_si128(),
-               (__mmask8) __U);
+_mm_maskz_madd_epi16(__mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U,
+                                             (__v4si)_mm_madd_epi16(__A, __B),
+                                             (__v4si)_mm_setzero_si128());
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_madd_epi16 (__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
-  return (__m256i) __builtin_ia32_pmaddwd256_mask ((__v16hi) __A,
-               (__v16hi) __B,
-               (__v8si) __W,
-               (__mmask8) __U);
+_mm256_mask_madd_epi16(__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U,
+                                            (__v8si)_mm256_madd_epi16(__A, __B),
+                                            (__v8si)__W);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_madd_epi16 (__mmask8 __U, __m256i __A, __m256i __B) {
-  return (__m256i) __builtin_ia32_pmaddwd256_mask ((__v16hi) __A,
-               (__v16hi) __B,
-               (__v8si) _mm256_setzero_si256(),
-               (__mmask8) __U);
+_mm256_maskz_madd_epi16(__mmask8 __U, __m256i __A, __m256i __B) {
+  return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U,
+                                            (__v8si)_mm256_madd_epi16(__A, __B),
+                                            (__v8si)_mm256_setzero_si256());
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS

diff --git a/test/CodeGen/avx512vlbw-builtins.c b/test/CodeGen/avx512vlbw-builtins.c
index 3a8b00ba7cebd2273921922ba84f1b0be9f160f4..795a1b48b19db18a6db69167453b7cc62dd832ef 100644 (file)
--- a/test/CodeGen/avx512vlbw-builtins.c
+++ b/test/CodeGen/avx512vlbw-builtins.c
@@ -1590,49 +1590,57 @@ __m256i test_mm256_maskz_permutex2var_epi16(__mmask16 __U, __m256i __A,
 }
 __m128i test_mm_mask_maddubs_epi16(__m128i __W, __mmask8 __U, __m128i __X, __m128i __Y) {
   // CHECK-LABEL: @test_mm_mask_maddubs_epi16
-  // CHECK: @llvm.x86.avx512.mask.pmaddubs.w.128
+  // CHECK: @llvm.x86.ssse3.pmadd.ub.sw
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_mask_maddubs_epi16(__W, __U, __X, __Y); 
 }
 
 __m128i test_mm_maskz_maddubs_epi16(__mmask8 __U, __m128i __X, __m128i __Y) {
   // CHECK-LABEL: @test_mm_maskz_maddubs_epi16
-  // CHECK: @llvm.x86.avx512.mask.pmaddubs.w.128
+  // CHECK: @llvm.x86.ssse3.pmadd.ub.sw
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
   return _mm_maskz_maddubs_epi16(__U, __X, __Y); 
 }
 
 __m256i test_mm256_mask_maddubs_epi16(__m256i __W, __mmask16 __U, __m256i __X, __m256i __Y) {
   // CHECK-LABEL: @test_mm256_mask_maddubs_epi16
-  // CHECK: @llvm.x86.avx512.mask.pmaddubs.w.256
+  // CHECK: @llvm.x86.avx2.pmadd.ub.sw
+  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_mask_maddubs_epi16(__W, __U, __X, __Y); 
 }
 
 __m256i test_mm256_maskz_maddubs_epi16(__mmask16 __U, __m256i __X, __m256i __Y) {
   // CHECK-LABEL: @test_mm256_maskz_maddubs_epi16
-  // CHECK: @llvm.x86.avx512.mask.pmaddubs.w.256
+  // CHECK: @llvm.x86.avx2.pmadd.ub.sw
+  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
   return _mm256_maskz_maddubs_epi16(__U, __X, __Y); 
 }
 
 __m128i test_mm_mask_madd_epi16(__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_mask_madd_epi16
-  // CHECK: @llvm.x86.avx512.mask.pmaddw.d.128
+  // CHECK: @llvm.x86.sse2.pmadd.wd
+  // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_madd_epi16(__W, __U, __A, __B); 
 }
 
 __m128i test_mm_maskz_madd_epi16(__mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: @test_mm_maskz_madd_epi16
-  // CHECK: @llvm.x86.avx512.mask.pmaddw.d.128
+  // CHECK: @llvm.x86.sse2.pmadd.wd
+  // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_madd_epi16(__U, __A, __B); 
 }
 
 __m256i test_mm256_mask_madd_epi16(__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_mask_madd_epi16
-  // CHECK: @llvm.x86.avx512.mask.pmaddw.d.256
+  // CHECK: @llvm.x86.avx2.pmadd.wd
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_madd_epi16(__W, __U, __A, __B); 
 }
 
 __m256i test_mm256_maskz_madd_epi16(__mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_madd_epi16
-  // CHECK: @llvm.x86.avx512.mask.pmaddw.d.256
+  // CHECK: @llvm.x86.avx2.pmadd.wd
+  // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_madd_epi16(__U, __A, __B); 
 }