[X86] Replace 128-bit and 256 masked vpermilps/vpermilpd builtins with native IR.

llvm-svn: 274425

4 files changed, 48 insertions, 47 deletions

diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index 0ed9563b7fd..e15462b86a0 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -1800,10 +1800,6 @@ 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_vpermilpd_mask, "V2dV2dIiV2dUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_vpermilpd256_mask, "V4dV4dIiV4dUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_vpermilps_mask, "V4fV4fIiV4fUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_vpermilps256_mask, "V8fV8fIiV8fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_vpermilvarpd_mask, "V2dV2dV2LLiV2dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_vpermilvarpd256_mask, "V4dV4dV4LLiV4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_vpermilvarps_mask, "V4fV4fV4iV4fUc","","avx512vl")
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
index 6bef81c0960..e97ce7179c2 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -6700,43 +6700,44 @@ _mm256_maskz_rcp14_ps (__mmask8 __U, __m256 __A)
 }
 
 #define _mm_mask_permute_pd(W, U, X, C) __extension__ ({ \
-  (__m128d)__builtin_ia32_vpermilpd_mask((__v2df)(__m128d)(X), (int)(C), \
-                                         (__v2df)(__m128d)(W), \
-                                         (__mmask8)(U)); })
+  (__m128d)__builtin_ia32_selectpd_128((__mmask8)(U), \
+                                       (__v2df)_mm_permute_pd((X), (C)), \
+                                       (__v2df)(__m128d)(W)); })
 
 #define _mm_maskz_permute_pd(U, X, C) __extension__ ({ \
-  (__m128d)__builtin_ia32_vpermilpd_mask((__v2df)(__m128d)(X), (int)(C), \
-                                         (__v2df)_mm_setzero_pd(), \
-                                         (__mmask8)(U)); })
+  (__m128d)__builtin_ia32_selectpd_128((__mmask8)(U), \
+                                       (__v2df)_mm_permute_pd((X), (C)), \
+                                       (__v2df)_mm_setzero_pd()); })
 
 #define _mm256_mask_permute_pd(W, U, X, C) __extension__ ({ \
-  (__m256d)__builtin_ia32_vpermilpd256_mask((__v4df)(__m256d)(X), (int)(C), \
-                                            (__v4df)(__m256d)(W), \
-                                            (__mmask8)(U)); })
+  (__m256d)__builtin_ia32_selectpd_256((__mmask8)(U), \
+                                       (__v4df)_mm256_permute_pd((X), (C)), \
+                                       (__v4df)(__m256d)(W)); })
 
 #define _mm256_maskz_permute_pd(U, X, C) __extension__ ({ \
-  (__m256d)__builtin_ia32_vpermilpd256_mask((__v4df)(__m256d)(X), (int)(C), \
-                                            (__v4df)_mm256_setzero_pd(), \
-                                            (__mmask8)(U)); })
+  (__m256d)__builtin_ia32_selectpd_256((__mmask8)(U), \
+                                       (__v4df)_mm256_permute_pd((X), (C)), \
+                                       (__v4df)_mm256_setzero_pd()); })
 
 #define _mm_mask_permute_ps(W, U, X, C) __extension__ ({ \
-  (__m128)__builtin_ia32_vpermilps_mask((__v4sf)(__m128)(X), (int)(C), \
-                                        (__v4sf)(__m128)(W), (__mmask8)(U)); })
+  (__m128)__builtin_ia32_selectps_128((__mmask8)(U), \
+                                      (__v4sf)_mm_permute_ps((X), (C)), \
+                                      (__v4sf)(__m128)(W)); })
 
 #define _mm_maskz_permute_ps(U, X, C) __extension__ ({ \
-  (__m128)__builtin_ia32_vpermilps_mask((__v4sf)(__m128)(X), (int)(C), \
-                                        (__v4sf)_mm_setzero_ps(), \
-                                        (__mmask8)(U)); })
+  (__m128)__builtin_ia32_selectps_128((__mmask8)(U), \
+                                      (__v4sf)_mm_permute_ps((X), (C)), \
+                                      (__v4sf)_mm_setzero_ps()); })
 
 #define _mm256_mask_permute_ps(W, U, X, C) __extension__ ({ \
-  (__m256)__builtin_ia32_vpermilps256_mask((__v8sf)(__m256)(X), (int)(C), \
-                                           (__v8sf)(__m256)(W), \
-                                           (__mmask8)(U)); })
+  (__m256)__builtin_ia32_selectps_256((__mmask8)(U), \
+                                      (__v8sf)_mm256_permute_ps((X), (C)), \
+                                      (__v8sf)(__m256)(W)); })
 
 #define _mm256_maskz_permute_ps(U, X, C) __extension__ ({ \
-  (__m256)__builtin_ia32_vpermilps256_mask((__v8sf)(__m256)(X), (int)(C), \
-                                           (__v8sf)_mm256_setzero_ps(), \
-                                           (__mmask8)(U)); })
+  (__m256)__builtin_ia32_selectps_256((__mmask8)(U), \
+                                      (__v8sf)_mm256_permute_ps((X), (C)), \
+                                      (__v8sf)_mm256_setzero_ps()); })
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_mask_permutevar_pd (__m128d __W, __mmask8 __U, __m128d __A,
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 866c25d6acc..89e5e3e5af1 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -1388,7 +1388,6 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI_mm_prefetch:
   case X86::BI__builtin_ia32_extractf32x4_mask:
   case X86::BI__builtin_ia32_extracti32x4_mask:
-  case X86::BI__builtin_ia32_vpermilpd_mask:
   case X86::BI__builtin_ia32_extractf64x2_512_mask:
   case X86::BI__builtin_ia32_extracti64x2_512_mask:
     i = 1; l = 0; u = 3;
@@ -1459,7 +1458,6 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_roundpd:
   case X86::BI__builtin_ia32_roundps256:
   case X86::BI__builtin_ia32_roundpd256:
-  case X86::BI__builtin_ia32_vpermilpd256_mask:
     i = 1; l = 0; u = 15;
     break;
   case X86::BI__builtin_ia32_roundss:
@@ -1563,8 +1561,6 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_fpclasspd512_mask:
   case X86::BI__builtin_ia32_fpclasssd_mask:
   case X86::BI__builtin_ia32_fpclassss_mask:
-  case X86::BI__builtin_ia32_vpermilps_mask:
-  case X86::BI__builtin_ia32_vpermilps256_mask:
   case X86::BI__builtin_ia32_vpermilps512_mask:
     i = 1; l = 0; u = 255;
     break;
diff --git a/clang/test/CodeGen/avx512vl-builtins.c b/clang/test/CodeGen/avx512vl-builtins.c
index 340a9d01002..33fcae285e9 100644
--- a/clang/test/CodeGen/avx512vl-builtins.c
+++ b/clang/test/CodeGen/avx512vl-builtins.c
@@ -4610,50 +4610,58 @@ __m256 test_mm256_maskz_rcp14_ps(__mmask8 __U, __m256 __A) {
 
 __m128d test_mm_mask_permute_pd(__m128d __W, __mmask8 __U, __m128d __X) {
   // CHECK-LABEL: @test_mm_mask_permute_pd
-  // CHECK: @llvm.x86.avx512.mask.vpermil.pd
-  return _mm_mask_permute_pd(__W, __U, __X, 2); 
+  // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x i32> <i32 1, i32 0>
+  // CHECK: select <2 x i1> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}
+  return _mm_mask_permute_pd(__W, __U, __X, 1); 
 }
 
 __m128d test_mm_maskz_permute_pd(__mmask8 __U, __m128d __X) {
   // CHECK-LABEL: @test_mm_maskz_permute_pd
-  // CHECK: @llvm.x86.avx512.mask.vpermil.pd
-  return _mm_maskz_permute_pd(__U, __X, 2); 
+  // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <2 x i32> <i32 1, i32 0>
+  // CHECK: select <2 x i1> %{{.*}}, <2 x double> %{{.*}}, <2 x double> %{{.*}}
+  return _mm_maskz_permute_pd(__U, __X, 1); 
 }
 
 __m256d test_mm256_mask_permute_pd(__m256d __W, __mmask8 __U, __m256d __X) {
   // CHECK-LABEL: @test_mm256_mask_permute_pd
-  // CHECK: @llvm.x86.avx512.mask.vpermil.pd.256
-  return _mm256_mask_permute_pd(__W, __U, __X, 2); 
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
+  return _mm256_mask_permute_pd(__W, __U, __X, 5); 
 }
 
 __m256d test_mm256_maskz_permute_pd(__mmask8 __U, __m256d __X) {
   // CHECK-LABEL: @test_mm256_maskz_permute_pd
-  // CHECK: @llvm.x86.avx512.mask.vpermil.pd.256
-  return _mm256_maskz_permute_pd(__U, __X, 2); 
+  // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
+  return _mm256_maskz_permute_pd(__U, __X, 5); 
 }
 
 __m128 test_mm_mask_permute_ps(__m128 __W, __mmask8 __U, __m128 __X) {
   // CHECK-LABEL: @test_mm_mask_permute_ps
-  // CHECK: @llvm.x86.avx512.mask.vpermil.ps
-  return _mm_mask_permute_ps(__W, __U, __X, 2); 
+  // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
+  return _mm_mask_permute_ps(__W, __U, __X, 0x1b); 
 }
 
 __m128 test_mm_maskz_permute_ps(__mmask8 __U, __m128 __X) {
   // CHECK-LABEL: @test_mm_maskz_permute_ps
-  // CHECK: @llvm.x86.avx512.mask.vpermil.ps
-  return _mm_maskz_permute_ps(__U, __X, 2); 
+  // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+  // CHECK: select <4 x i1> %{{.*}}, <4 x float> %{{.*}}, <4 x float> %{{.*}}
+  return _mm_maskz_permute_ps(__U, __X, 0x1b); 
 }
 
 __m256 test_mm256_mask_permute_ps(__m256 __W, __mmask8 __U, __m256 __X) {
   // CHECK-LABEL: @test_mm256_mask_permute_ps
-  // CHECK: @llvm.x86.avx512.mask.vpermil.ps.256
-  return _mm256_mask_permute_ps(__W, __U, __X, 2); 
+  // CHECK: shufflevector <8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
+  return _mm256_mask_permute_ps(__W, __U, __X, 0x1b); 
 }
 
 __m256 test_mm256_maskz_permute_ps(__mmask8 __U, __m256 __X) {
   // CHECK-LABEL: @test_mm256_maskz_permute_ps
-  // CHECK: @llvm.x86.avx512.mask.vpermil.ps.256
-  return _mm256_maskz_permute_ps(__U, __X, 2); 
+  // CHECK: shufflevector <8 x float> %{{.*}}, <8 x float> %{{.*}}, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
+  return _mm256_maskz_permute_ps(__U, __X, 0x1b); 
 }
 
 __m128d test_mm_mask_permutevar_pd(__m128d __W, __mmask8 __U, __m128d __A, __m128i __C) {