[AVX-512] Remove 128/256-bit masked vpermilvar builtins and replace with select and the avx unmasked builtins.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@289338 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 48 insertions, 62 deletions

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 9b3d942bf9..bce9a09c49 100644
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -1495,10 +1495,6 @@ TARGET_BUILTIN(__builtin_ia32_vpermt2vard512_maskz, "V16iV16iV16iV16iUs","","avx
 TARGET_BUILTIN(__builtin_ia32_vpermt2varpd512_maskz, "V8dV8LLiV8dV8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_vpermt2varps512_maskz, "V16fV16iV16fV16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_vpermt2varq512_maskz, "V8LLiV8LLiV8LLiV8LLiUc","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_vpermilvarpd_mask, "V2dV2dV2LLiV2dUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_vpermilvarpd256_mask, "V4dV4dV4LLiV4dUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_vpermilvarps_mask, "V4fV4fV4iV4fUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_vpermilvarps256_mask, "V8fV8fV8iV8fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_ptestmb512, "ULLiV64cV64cULLi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_ptestmw512, "UiV32sV32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_ptestnmb512, "ULLiV64cV64cULLi","","avx512bw")
diff --git a/lib/Headers/avx512vlintrin.h b/lib/Headers/avx512vlintrin.h
index 0974fc89ad..f3744da6ab 100644
--- a/lib/Headers/avx512vlintrin.h
+++ b/lib/Headers/avx512vlintrin.h
@@ -6424,85 +6424,67 @@ _mm256_maskz_rcp14_ps (__mmask8 __U, __m256 __A)
                                       (__v8sf)_mm256_setzero_ps()); })
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
-_mm_mask_permutevar_pd (__m128d __W, __mmask8 __U, __m128d __A,
-      __m128i __C)
+_mm_mask_permutevar_pd(__m128d __W, __mmask8 __U, __m128d __A, __m128i __C)
 {
-  return (__m128d) __builtin_ia32_vpermilvarpd_mask ((__v2df) __A,
-                 (__v2di) __C,
-                 (__v2df) __W,
-                 (__mmask8) __U);
+  return (__m128d)__builtin_ia32_selectpd_128((__mmask8)__U,
+                                            (__v2df)_mm_permutevar_pd(__A, __C),
+                                            (__v2df)__W);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
-_mm_maskz_permutevar_pd (__mmask8 __U, __m128d __A, __m128i __C)
+_mm_maskz_permutevar_pd(__mmask8 __U, __m128d __A, __m128i __C)
 {
-  return (__m128d) __builtin_ia32_vpermilvarpd_mask ((__v2df) __A,
-                 (__v2di) __C,
-                 (__v2df)
-                 _mm_setzero_pd (),
-                 (__mmask8) __U);
+  return (__m128d)__builtin_ia32_selectpd_128((__mmask8)__U,
+                                            (__v2df)_mm_permutevar_pd(__A, __C),
+                                            (__v2df)_mm_setzero_pd());
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
-_mm256_mask_permutevar_pd (__m256d __W, __mmask8 __U, __m256d __A,
-         __m256i __C)
+_mm256_mask_permutevar_pd(__m256d __W, __mmask8 __U, __m256d __A, __m256i __C)
 {
-  return (__m256d) __builtin_ia32_vpermilvarpd256_mask ((__v4df) __A,
-              (__v4di) __C,
-              (__v4df) __W,
-              (__mmask8)
-              __U);
+  return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__U,
+                                         (__v4df)_mm256_permutevar_pd(__A, __C),
+                                         (__v4df)__W);
 }
 
 static __inline__ __m256d __DEFAULT_FN_ATTRS
-_mm256_maskz_permutevar_pd (__mmask8 __U, __m256d __A, __m256i __C)
+_mm256_maskz_permutevar_pd(__mmask8 __U, __m256d __A, __m256i __C)
 {
-  return (__m256d) __builtin_ia32_vpermilvarpd256_mask ((__v4df) __A,
-              (__v4di) __C,
-              (__v4df)
-              _mm256_setzero_pd (),
-              (__mmask8)
-              __U);
+  return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__U,
+                                         (__v4df)_mm256_permutevar_pd(__A, __C),
+                                         (__v4df)_mm256_setzero_pd());
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS
-_mm_mask_permutevar_ps (__m128 __W, __mmask8 __U, __m128 __A,
-      __m128i __C)
+_mm_mask_permutevar_ps(__m128 __W, __mmask8 __U, __m128 __A, __m128i __C)
 {
-  return (__m128) __builtin_ia32_vpermilvarps_mask ((__v4sf) __A,
-                (__v4si) __C,
-                (__v4sf) __W,
-                (__mmask8) __U);
+  return (__m128)__builtin_ia32_selectps_128((__mmask8)__U,
+                                            (__v4sf)_mm_permutevar_ps(__A, __C),
+                                            (__v4sf)__W);
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS
-_mm_maskz_permutevar_ps (__mmask8 __U, __m128 __A, __m128i __C)
+_mm_maskz_permutevar_ps(__mmask8 __U, __m128 __A, __m128i __C)
 {
-  return (__m128) __builtin_ia32_vpermilvarps_mask ((__v4sf) __A,
-                (__v4si) __C,
-                (__v4sf)
-                _mm_setzero_ps (),
-                (__mmask8) __U);
+  return (__m128)__builtin_ia32_selectps_128((__mmask8)__U,
+                                            (__v4sf)_mm_permutevar_ps(__A, __C),
+                                            (__v4sf)_mm_setzero_ps());
 }
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_mask_permutevar_ps (__m256 __W, __mmask8 __U, __m256 __A,
-         __m256i __C)
+_mm256_mask_permutevar_ps(__m256 __W, __mmask8 __U, __m256 __A, __m256i __C)
 {
-  return (__m256) __builtin_ia32_vpermilvarps256_mask ((__v8sf) __A,
-                   (__v8si) __C,
-                   (__v8sf) __W,
-                   (__mmask8) __U);
+  return (__m256)__builtin_ia32_selectps_256((__mmask8)__U,
+                                          (__v8sf)_mm256_permutevar_ps(__A, __C),
+                                          (__v8sf)__W);
 }
 
 static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_maskz_permutevar_ps (__mmask8 __U, __m256 __A, __m256i __C)
+_mm256_maskz_permutevar_ps(__mmask8 __U, __m256 __A, __m256i __C)
 {
-  return (__m256) __builtin_ia32_vpermilvarps256_mask ((__v8sf) __A,
-                   (__v8si) __C,
-                   (__v8sf)
-                   _mm256_setzero_ps (),
-                   (__mmask8) __U);
+  return (__m256)__builtin_ia32_selectps_256((__mmask8)__U,
+                                          (__v8sf)_mm256_permutevar_ps(__A, __C),
+                                          (__v8sf)_mm256_setzero_ps());
 }
 
 static __inline__ __mmask8 __DEFAULT_FN_ATTRS
diff --git a/test/CodeGen/avx512vl-builtins.c b/test/CodeGen/avx512vl-builtins.c
index c4ae3a023b..fe4ebe12dd 100644
--- a/test/CodeGen/avx512vl-builtins.c
+++ b/test/CodeGen/avx512vl-builtins.c
@@ -5064,49 +5064,57 @@ __m256 test_mm256_maskz_permute_ps(__mmask8 __U, __m256 __X) {
 
 __m128d test_mm_mask_permutevar_pd(__m128d __W, __mmask8 __U, __m128d __A, __m128i __C) {
   // CHECK-LABEL: @test_mm_mask_permutevar_pd
-  // CHECK: @llvm.x86.avx512.mask.vpermilvar.pd
+  // CHECK: @llvm.x86.avx.vpermilvar.pd
+  // CHECK: select <2 x i1> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}
   return _mm_mask_permutevar_pd(__W, __U, __A, __C); 
 }
 
 __m128d test_mm_maskz_permutevar_pd(__mmask8 __U, __m128d __A, __m128i __C) {
   // CHECK-LABEL: @test_mm_maskz_permutevar_pd
-  // CHECK: @llvm.x86.avx512.mask.vpermilvar.pd
+  // CHECK: @llvm.x86.avx.vpermilvar.pd
+  // CHECK: select <2 x i1> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}
   return _mm_maskz_permutevar_pd(__U, __A, __C); 
 }
 
 __m256d test_mm256_mask_permutevar_pd(__m256d __W, __mmask8 __U, __m256d __A, __m256i __C) {
   // CHECK-LABEL: @test_mm256_mask_permutevar_pd
-  // CHECK: @llvm.x86.avx512.mask.vpermilvar.pd.256
+  // CHECK: @llvm.x86.avx.vpermilvar.pd.256
+  // CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
   return _mm256_mask_permutevar_pd(__W, __U, __A, __C); 
 }
 
 __m256d test_mm256_maskz_permutevar_pd(__mmask8 __U, __m256d __A, __m256i __C) {
   // CHECK-LABEL: @test_mm256_maskz_permutevar_pd
-  // CHECK: @llvm.x86.avx512.mask.vpermilvar.pd.256
+  // CHECK: @llvm.x86.avx.vpermilvar.pd.256
+  // CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
   return _mm256_maskz_permutevar_pd(__U, __A, __C); 
 }
 
 __m128 test_mm_mask_permutevar_ps(__m128 __W, __mmask8 __U, __m128 __A, __m128i __C) {
   // CHECK-LABEL: @test_mm_mask_permutevar_ps
-  // CHECK: @llvm.x86.avx512.mask.vpermilvar.ps
+  // CHECK: @llvm.x86.avx.vpermilvar.ps
+  // CHECK: select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
   return _mm_mask_permutevar_ps(__W, __U, __A, __C); 
 }
 
 __m128 test_mm_maskz_permutevar_ps(__mmask8 __U, __m128 __A, __m128i __C) {
   // CHECK-LABEL: @test_mm_maskz_permutevar_ps
-  // CHECK: @llvm.x86.avx512.mask.vpermilvar.ps
+  // CHECK: @llvm.x86.avx.vpermilvar.ps
+  // CHECK: select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
   return _mm_maskz_permutevar_ps(__U, __A, __C); 
 }
 
 __m256 test_mm256_mask_permutevar_ps(__m256 __W, __mmask8 __U, __m256 __A, __m256i __C) {
   // CHECK-LABEL: @test_mm256_mask_permutevar_ps
-  // CHECK: @llvm.x86.avx512.mask.vpermilvar.ps.256
+  // CHECK: @llvm.x86.avx.vpermilvar.ps.256
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm256_mask_permutevar_ps(__W, __U, __A, __C); 
 }
 
 __m256 test_mm256_maskz_permutevar_ps(__mmask8 __U, __m256 __A, __m256i __C) {
   // CHECK-LABEL: @test_mm256_maskz_permutevar_ps
-  // CHECK: @llvm.x86.avx512.mask.vpermilvar.ps.256
+  // CHECK: @llvm.x86.avx.vpermilvar.ps.256
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm256_maskz_permutevar_ps(__U, __A, __C); 
 }