[X86] Revert part of r332266: Use __builtin_convertvector to replace some of the avx512 truncate builtins.

The masking doesn't work right in the backend for the ones that produce byte or word elements without avx512bw.

llvm-svn: 332322

5 files changed, 32 insertions, 40 deletions

diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index 9de8e86d003..f7923919b3d 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -1717,12 +1717,15 @@ TARGET_BUILTIN(__builtin_ia32_pmovusqw128_mask, "V8sV2LLiV8sUc", "nc", "avx512vl
 TARGET_BUILTIN(__builtin_ia32_pmovusqw128mem_mask, "vV8s*V2LLiUc", "n", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqw256_mask, "V8sV4LLiV8sUc", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovusqw256mem_mask, "vV8s*V4LLiUc", "n", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmovdb512_mask, "V16cV16iV16cUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovdb512mem_mask, "vV16c*V16iUs", "n", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovwb512mem_mask, "vV32c*V32sUi", "n", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pmovdw512_mask, "V16sV16iV16sUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovdw512mem_mask, "vV16s*V16iUs", "n", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqb512_mask, "V16cV8LLiV16cUc", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqb512mem_mask, "vV16c*V8LLiUc", "n", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqd512mem_mask, "vV8i*V8LLiUc", "n", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_pmovqw512_mask, "V8sV8LLiV8sUc", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovqw512mem_mask, "vV8s*V8LLiUc", "n", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_pmovdb128_mask, "V16cV4iV16cUc", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovwb128mem_mask, "vV16c*V8sUc", "n", "avx512vl,avx512bw")
@@ -1732,6 +1735,7 @@ TARGET_BUILTIN(__builtin_ia32_pmovdb256mem_mask, "vV16c*V8iUc", "n", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovwb256mem_mask, "vV16c*V16sUs", "n", "avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_pmovdw128_mask, "V8sV4iV8sUc", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovdw128mem_mask, "vV8s*V4iUc", "n", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pmovdw256_mask, "V8sV8iV8sUc", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovdw256mem_mask, "vV8s*V8iUc", "n", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqb128_mask, "V16cV2LLiV16cUc", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqb128mem_mask, "vV16c*V2LLiUc", "n", "avx512vl")
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 24a4e9fef24..ee5fb5ae0be 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -7607,17 +7607,16 @@ _mm512_cvtepi32_epi8 (__m512i __A)
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi32_epi8 (__m128i __O, __mmask16 __M, __m512i __A)
 {
-  return (__m128i)__builtin_ia32_selectb_128((__mmask8)__M,
-                                             (__v16qi)_mm512_cvtepi32_epi8(__A),
-                                             (__v16qi)__O);
+  return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A,
+              (__v16qi) __O, __M);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi32_epi8 (__mmask16 __M, __m512i __A)
 {
-  return (__m128i)__builtin_ia32_selectb_128((__mmask8)__M,
-                                             (__v16qi)_mm512_cvtepi32_epi8(__A),
-                                             (__v16qi)_mm_setzero_si128());
+  return (__m128i) __builtin_ia32_pmovdb512_mask ((__v16si) __A,
+              (__v16qi) _mm_setzero_si128 (),
+              __M);
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
@@ -7635,17 +7634,16 @@ _mm512_cvtepi32_epi16 (__m512i __A)
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi32_epi16 (__m256i __O, __mmask16 __M, __m512i __A)
 {
-  return (__m256i)__builtin_ia32_selectw_256((__mmask16)__M,
-                                             (__v16hi)_mm512_cvtepi32_epi16(__A),
-                                             (__v16hi)__O);
+  return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A,
+              (__v16hi) __O, __M);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi32_epi16 (__mmask16 __M, __m512i __A)
 {
-  return (__m256i)__builtin_ia32_selectw_256((__mmask16)__M,
-                                             (__v16hi)_mm512_cvtepi32_epi16(__A),
-                                             (__v16hi)_mm256_setzero_si256());
+  return (__m256i) __builtin_ia32_pmovdw512_mask ((__v16si) __A,
+              (__v16hi) _mm256_setzero_si256 (),
+              __M);
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
@@ -7720,17 +7718,16 @@ _mm512_cvtepi64_epi16 (__m512i __A)
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_mask_cvtepi64_epi16 (__m128i __O, __mmask8 __M, __m512i __A)
 {
-  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__M,
-                                             (__v8hi)_mm512_cvtepi64_epi16(__A),
-                                             (__v8hi)__O);
+  return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A,
+              (__v8hi) __O, __M);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm512_maskz_cvtepi64_epi16 (__mmask8 __M, __m512i __A)
 {
-  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__M,
-                                             (__v8hi)_mm512_cvtepi64_epi16(__A),
-                                             (__v8hi)_mm_setzero_si128());
+  return (__m128i) __builtin_ia32_pmovqw512_mask ((__v8di) __A,
+              (__v8hi) _mm_setzero_si128 (),
+              __M);
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
index 07429953f91..de89cfdfacb 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -7633,17 +7633,16 @@ _mm256_cvtepi32_epi16 (__m256i __A)
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_mask_cvtepi32_epi16 (__m128i __O, __mmask8 __M, __m256i __A)
 {
-  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__M,
-                                             (__v8hi)_mm256_cvtepi32_epi16(__A),
-                                             (__v8hi)__O);
+  return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A,
+              (__v8hi) __O, __M);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm256_maskz_cvtepi32_epi16 (__mmask8 __M, __m256i __A)
 {
-  return (__m128i)__builtin_ia32_selectw_128((__mmask8)__M,
-                                             (__v8hi)_mm256_cvtepi32_epi16(__A),
-                                             (__v8hi)_mm_setzero_si128());
+  return (__m128i) __builtin_ia32_pmovdw256_mask ((__v8si) __A,
+              (__v8hi) _mm_setzero_si128 (),
+              __M);
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS
diff --git a/clang/test/CodeGen/avx512f-builtins.c b/clang/test/CodeGen/avx512f-builtins.c
index 3b053f87621..0bff5c8dbd4 100644
--- a/clang/test/CodeGen/avx512f-builtins.c
+++ b/clang/test/CodeGen/avx512f-builtins.c
@@ -5108,15 +5108,13 @@ __m128i test_mm512_cvtepi32_epi8(__m512i __A) {
 
 __m128i test_mm512_mask_cvtepi32_epi8(__m128i __O, __mmask16 __M, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvtepi32_epi8
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i8>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.pmov.db.512
   return _mm512_mask_cvtepi32_epi8(__O, __M, __A); 
 }
 
 __m128i test_mm512_maskz_cvtepi32_epi8(__mmask16 __M, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvtepi32_epi8
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i8>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.pmov.db.512
   return _mm512_maskz_cvtepi32_epi8(__M, __A); 
 }
 
@@ -5134,15 +5132,13 @@ __m256i test_mm512_cvtepi32_epi16(__m512i __A) {
 
 __m256i test_mm512_mask_cvtepi32_epi16(__m256i __O, __mmask16 __M, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvtepi32_epi16
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.pmov.dw.512
   return _mm512_mask_cvtepi32_epi16(__O, __M, __A); 
 }
 
 __m256i test_mm512_maskz_cvtepi32_epi16(__mmask16 __M, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvtepi32_epi16
-  // CHECK: trunc <16 x i32> %{{.*}} to <16 x i16>
-  // CHECK: select <16 x i1> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.pmov.dw.512
   return _mm512_maskz_cvtepi32_epi16(__M, __A); 
 }
 
@@ -5210,15 +5206,13 @@ __m128i test_mm512_cvtepi64_epi16(__m512i __A) {
 
 __m128i test_mm512_mask_cvtepi64_epi16(__m128i __O, __mmask8 __M, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvtepi64_epi16
-  // CHECK: trunc <8 x i64> %{{.*}} to <8 x i16>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.pmov.qw.512
   return _mm512_mask_cvtepi64_epi16(__O, __M, __A); 
 }
 
 __m128i test_mm512_maskz_cvtepi64_epi16(__mmask8 __M, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvtepi64_epi16
-  // CHECK: trunc <8 x i64> %{{.*}} to <8 x i16>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.pmov.qw.512
   return _mm512_maskz_cvtepi64_epi16(__M, __A); 
 }
 
diff --git a/clang/test/CodeGen/avx512vl-builtins.c b/clang/test/CodeGen/avx512vl-builtins.c
index 344631705be..27c0993b9b3 100644
--- a/clang/test/CodeGen/avx512vl-builtins.c
+++ b/clang/test/CodeGen/avx512vl-builtins.c
@@ -6583,15 +6583,13 @@ __m128i test_mm256_cvtepi32_epi16(__m256i __A) {
 
 __m128i test_mm256_mask_cvtepi32_epi16(__m128i __O, __mmask8 __M, __m256i __A) {
   // CHECK-LABEL: @test_mm256_mask_cvtepi32_epi16
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.pmov.dw.256
   return _mm256_mask_cvtepi32_epi16(__O, __M, __A); 
 }
 
 __m128i test_mm256_maskz_cvtepi32_epi16(__mmask8 __M, __m256i __A) {
   // CHECK-LABEL: @test_mm256_maskz_cvtepi32_epi16
-  // CHECK: trunc <8 x i32> %{{.*}} to <8 x i16>
-  // CHECK: select <8 x i1> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}}
+  // CHECK: @llvm.x86.avx512.mask.pmov.dw.256
   return _mm256_maskz_cvtepi32_epi16(__M, __A); 
 }