[X86] Fold masking into subvector extract builtins.

I'm looking into making the select builtins require avx512f, avx512bw, or avx512vl since masking operations generally require those features.

The extract builtins are funny because the 512-bit versions return a 128 or 256 bit vector with masking even when avx512vl is not supported.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@334330 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index c920b47af07fe6af05feb444909262bdff1f80c3..b694059adca9198b6fb067a55858ca8357e4ca86 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -947,8 +947,8 @@ TARGET_BUILTIN(__builtin_ia32_alignd128, "V4iV4iV4iIi", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_alignd256, "V8iV8iV8iIi", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_alignq128, "V2LLiV2LLiV2LLiIi", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_alignq256, "V4LLiV4LLiV4LLiIi", "nc", "avx512vl")
-TARGET_BUILTIN(__builtin_ia32_extractf64x4, "V4dV8dIi", "nc", "avx512f")
-TARGET_BUILTIN(__builtin_ia32_extractf32x4, "V4fV16fIi", "nc", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_extractf64x4_mask, "V4dV8dIiV4dUc", "nc", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_extractf32x4_mask, "V4fV16fIiV4fUc", "nc", "avx512f")
 
 TARGET_BUILTIN(__builtin_ia32_vpdpbusd128, "V4iV4iV4iV4i", "nc", "avx512vl,avx512vnni")
 TARGET_BUILTIN(__builtin_ia32_vpdpbusd256, "V8iV8iV8iV8i", "nc", "avx512vl,avx512vnni")
@@ -1672,16 +1672,16 @@ TARGET_BUILTIN(__builtin_ia32_pmovqw128_mask, "V8sV2LLiV8sUc", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqw128mem_mask, "vV8s*V2LLiUc", "n", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqw256_mask, "V8sV4LLiV8sUc", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovqw256mem_mask, "vV8s*V4LLiUc", "n", "avx512vl")
-TARGET_BUILTIN(__builtin_ia32_extractf32x8, "V8fV16fIi", "nc", "avx512dq")
-TARGET_BUILTIN(__builtin_ia32_extractf64x2_512, "V2dV8dIi", "nc", "avx512dq")
-TARGET_BUILTIN(__builtin_ia32_extracti32x8, "V8iV16iIi", "nc", "avx512dq")
-TARGET_BUILTIN(__builtin_ia32_extracti64x2_512, "V2LLiV8LLiIi", "nc", "avx512dq")
-TARGET_BUILTIN(__builtin_ia32_extracti32x4, "V4iV16iIi", "nc", "avx512f")
-TARGET_BUILTIN(__builtin_ia32_extracti64x4, "V4LLiV8LLiIi", "nc", "avx512f")
-TARGET_BUILTIN(__builtin_ia32_extractf64x2_256, "V2dV4dIi", "nc", "avx512dq,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_extracti64x2_256, "V2LLiV4LLiIi", "nc", "avx512dq,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_extractf32x4_256, "V4fV8fIi", "nc", "avx512vl")
-TARGET_BUILTIN(__builtin_ia32_extracti32x4_256, "V4iV8iIi", "nc", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_extractf32x8_mask, "V8fV16fIiV8fUc", "nc", "avx512dq")
+TARGET_BUILTIN(__builtin_ia32_extractf64x2_512_mask, "V2dV8dIiV2dUc", "nc", "avx512dq")
+TARGET_BUILTIN(__builtin_ia32_extracti32x8_mask, "V8iV16iIiV8iUc", "nc", "avx512dq")
+TARGET_BUILTIN(__builtin_ia32_extracti64x2_512_mask, "V2LLiV8LLiIiV2LLiUc", "nc", "avx512dq")
+TARGET_BUILTIN(__builtin_ia32_extracti32x4_mask, "V4iV16iIiV4iUc", "nc", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_extracti64x4_mask, "V4LLiV8LLiIiV4LLiUc", "nc", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_extractf64x2_256_mask, "V2dV4dIiV2dUc", "nc", "avx512dq,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_extracti64x2_256_mask, "V2LLiV4LLiIiV2LLiUc", "nc", "avx512dq,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_extractf32x4_256_mask, "V4fV8fIiV4fUc", "nc", "avx512vl")
+TARGET_BUILTIN(__builtin_ia32_extracti32x4_256_mask, "V4iV8iIiV4iUc", "nc", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_insertf32x8, "V16fV16fV8fIi", "nc", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_insertf64x2_512, "V8dV8dV2dIi", "nc", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_inserti32x8, "V16iV16iV8iIi", "nc", "avx512dq")

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index ac9f46cb449a4e5911c2cc7c44471b44355062b3..012428dbb28545956b87d7b65ae122d8ce4bf1ce 100644 (file)
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -9239,18 +9239,18 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
   case X86::BI__builtin_ia32_vextractf128_ps256:
   case X86::BI__builtin_ia32_vextractf128_si256:
   case X86::BI__builtin_ia32_extract128i256:
-  case X86::BI__builtin_ia32_extractf64x4:
-  case X86::BI__builtin_ia32_extractf32x4:
-  case X86::BI__builtin_ia32_extracti64x4:
-  case X86::BI__builtin_ia32_extracti32x4:
-  case X86::BI__builtin_ia32_extractf32x8:
-  case X86::BI__builtin_ia32_extracti32x8:
-  case X86::BI__builtin_ia32_extractf32x4_256:
-  case X86::BI__builtin_ia32_extracti32x4_256:
-  case X86::BI__builtin_ia32_extractf64x2_256:
-  case X86::BI__builtin_ia32_extracti64x2_256:
-  case X86::BI__builtin_ia32_extractf64x2_512:
-  case X86::BI__builtin_ia32_extracti64x2_512: {
+  case X86::BI__builtin_ia32_extractf64x4_mask:
+  case X86::BI__builtin_ia32_extractf32x4_mask:
+  case X86::BI__builtin_ia32_extracti64x4_mask:
+  case X86::BI__builtin_ia32_extracti32x4_mask:
+  case X86::BI__builtin_ia32_extractf32x8_mask:
+  case X86::BI__builtin_ia32_extracti32x8_mask:
+  case X86::BI__builtin_ia32_extractf32x4_256_mask:
+  case X86::BI__builtin_ia32_extracti32x4_256_mask:
+  case X86::BI__builtin_ia32_extractf64x2_256_mask:
+  case X86::BI__builtin_ia32_extracti64x2_256_mask:
+  case X86::BI__builtin_ia32_extractf64x2_512_mask:
+  case X86::BI__builtin_ia32_extracti64x2_512_mask: {
     llvm::Type *DstTy = ConvertType(E->getType());
     unsigned NumElts = DstTy->getVectorNumElements();
     unsigned Index = cast<ConstantInt>(Ops[1])->getZExtValue() * NumElts;
@@ -9259,10 +9259,15 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     for (unsigned i = 0; i != NumElts; ++i)
       Indices[i] = i + Index;
 
-    return Builder.CreateShuffleVector(Ops[0],
-                                       UndefValue::get(Ops[0]->getType()),
-                                       makeArrayRef(Indices, NumElts),
-                                       "extract");
+    Value *Res = Builder.CreateShuffleVector(Ops[0],
+                                             UndefValue::get(Ops[0]->getType()),
+                                             makeArrayRef(Indices, NumElts),
+                                             "extract");
+
+    if (Ops.size() == 4)
+      Res = EmitX86Select(*this, Ops[3], Res, Ops[2]);
+
+    return Res;
   }
   case X86::BI__builtin_ia32_vinsertf128_pd256:
   case X86::BI__builtin_ia32_vinsertf128_ps256:

diff --git a/lib/Headers/avx512dqintrin.h b/lib/Headers/avx512dqintrin.h
index fbb4bbce8dfdd1423852f7eca2e1c53ea6a76069..76ce32f086d75f7ea304a2c0700afda796382d72 100644 (file)
--- a/lib/Headers/avx512dqintrin.h
+++ b/lib/Headers/avx512dqintrin.h
@@ -1103,56 +1103,70 @@ _mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A)
 }
 
 #define _mm512_extractf32x8_ps(A, imm) \
-  (__m256)__builtin_ia32_extractf32x8((__v16sf)(__m512)(A), (int)(imm))
+  (__m256)__builtin_ia32_extractf32x8_mask((__v16sf)(__m512)(A), (int)(imm), \
+                                           (__v8sf)_mm256_undefined_ps(), \
+                                           (__mmask8)-1);
 
 #define _mm512_mask_extractf32x8_ps(W, U, A, imm) \
-  (__m256)__builtin_ia32_selectps_256((__mmask8)(U), \
-                                   (__v8sf)_mm512_extractf32x8_ps((A), (imm)), \
-                                   (__v8sf)(__m256)(W))
+  (__m256)__builtin_ia32_extractf32x8_mask((__v16sf)(__m512)(A), (int)(imm), \
+                                           (__v8sf)(__m256)(W), \
+                                           (__mmask8)(U))
 
 #define _mm512_maskz_extractf32x8_ps(U, A, imm) \
-  (__m256)__builtin_ia32_selectps_256((__mmask8)(U), \
-                                   (__v8sf)_mm512_extractf32x8_ps((A), (imm)), \
-                                   (__v8sf)_mm256_setzero_ps())
+  (__m256)__builtin_ia32_extractf32x8_mask((__v16sf)(__m512)(A), (int)(imm), \
+                                           (__v8sf)_mm256_setzero_ps(), \
+                                           (__mmask8)(U))
 
 #define _mm512_extractf64x2_pd(A, imm) \
-  (__m128d)__builtin_ia32_extractf64x2_512((__v8df)(__m512d)(A), (int)(imm))
+  (__m128d)__builtin_ia32_extractf64x2_512_mask((__v8df)(__m512d)(A), \
+                                                (int)(imm), \
+                                                (__v2df)_mm_undefined_pd(), \
+                                                (__mmask8)-1)
 
 #define _mm512_mask_extractf64x2_pd(W, U, A, imm) \
-  (__m128d)__builtin_ia32_selectpd_128((__mmask8)(U), \
-                                   (__v2df)_mm512_extractf64x2_pd((A), (imm)), \
-                                   (__v2df)(__m128d)(W))
+  (__m128d)__builtin_ia32_extractf64x2_512_mask((__v8df)(__m512d)(A), \
+                                                (int)(imm), \
+                                                (__v2df)(__m128d)(W), \
+                                                (__mmask8)(U))
 
 #define _mm512_maskz_extractf64x2_pd(U, A, imm) \
-  (__m128d)__builtin_ia32_selectpd_128((__mmask8)(U), \
-                                   (__v2df)_mm512_extractf64x2_pd((A), (imm)), \
-                                   (__v2df)_mm_setzero_pd())
+  (__m128d)__builtin_ia32_extractf64x2_512_mask((__v8df)(__m512d)(A), \
+                                                (int)(imm), \
+                                                (__v2df)_mm_setzero_pd(), \
+                                                (__mmask8)(U))
 
 #define _mm512_extracti32x8_epi32(A, imm) \
-  (__m256i)__builtin_ia32_extracti32x8((__v16si)(__m512i)(A), (int)(imm))
+  (__m256i)__builtin_ia32_extracti32x8_mask((__v16si)(__m512i)(A), (int)(imm), \
+                                            (__v8si)_mm256_undefined_si256(), \
+                                            (__mmask8)-1)
 
 #define _mm512_mask_extracti32x8_epi32(W, U, A, imm) \
-  (__m256i)__builtin_ia32_selectd_256((__mmask8)(U), \
-                                (__v8si)_mm512_extracti32x8_epi32((A), (imm)), \
-                                (__v8si)(__m256i)(W))
+  (__m256i)__builtin_ia32_extracti32x8_mask((__v16si)(__m512i)(A), (int)(imm), \
+                                            (__v8si)(__m256i)(W), \
+                                            (__mmask8)(U))
 
 #define _mm512_maskz_extracti32x8_epi32(U, A, imm) \
-  (__m256i)__builtin_ia32_selectd_256((__mmask8)(U), \
-                                (__v8si)_mm512_extracti32x8_epi32((A), (imm)), \
-                                (__v8si)_mm256_setzero_si256())
+  (__m256i)__builtin_ia32_extracti32x8_mask((__v16si)(__m512i)(A), (int)(imm), \
+                                            (__v8si)_mm256_setzero_si256(), \
+                                            (__mmask8)(U))
 
 #define _mm512_extracti64x2_epi64(A, imm) \
-  (__m128i)__builtin_ia32_extracti64x2_512((__v8di)(__m512i)(A), (int)(imm))
+  (__m128i)__builtin_ia32_extracti64x2_512_mask((__v8di)(__m512i)(A), \
+                                                (int)(imm), \
+                                                (__v2di)_mm_undefined_si128(), \
+                                                (__mmask8)-1)
 
 #define _mm512_mask_extracti64x2_epi64(W, U, A, imm) \
-  (__m128d)__builtin_ia32_selectq_128((__mmask8)(U), \
-                                (__v2di)_mm512_extracti64x2_epi64((A), (imm)), \
-                                (__v2di)(__m128i)(W))
+  (__m128i)__builtin_ia32_extracti64x2_512_mask((__v8di)(__m512i)(A), \
+                                                (int)(imm), \
+                                                (__v2di)(__m128i)(W), \
+                                                (__mmask8)(U))
 
 #define _mm512_maskz_extracti64x2_epi64(U, A, imm) \
-  (__m128d)__builtin_ia32_selectq_128((__mmask8)(U), \
-                                (__v2di)_mm512_extracti64x2_epi64((A), (imm)), \
-                                (__v2di)_mm_setzero_si128())
+  (__m128i)__builtin_ia32_extracti64x2_512_mask((__v8di)(__m512i)(A), \
+                                                (int)(imm), \
+                                                (__v2di)_mm_setzero_si128(), \
+                                                (__mmask8)(U))
 
 #define _mm512_insertf32x8(A, B, imm) \
   (__m512)__builtin_ia32_insertf32x8((__v16sf)(__m512)(A), \

diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h
index 381011ed3921a172da53df5f88ac8577321b29a3..dbac414fff964a29cb14c1ae2b605300c5d8afa0 100644 (file)
--- a/lib/Headers/avx512fintrin.h
+++ b/lib/Headers/avx512fintrin.h
@@ -3494,30 +3494,34 @@ _mm512_maskz_permutex2var_epi64(__mmask8 __U, __m512i __A, __m512i __I,
 /* Vector Extract */
 
 #define _mm512_extractf64x4_pd(A, I) \
-  (__m256d)__builtin_ia32_extractf64x4((__v8df)(__m512d)(A), (int)(I))
+  (__m256d)__builtin_ia32_extractf64x4_mask((__v8df)(__m512d)(A), (int)(I), \
+                                            (__v4df)_mm256_undefined_si256(), \
+                                            (__mmask8)-1)
 
 #define _mm512_mask_extractf64x4_pd(W, U, A, imm) \
-  (__m256d)__builtin_ia32_selectpd_256((__mmask8)(U), \
-                                   (__v4df)_mm512_extractf64x4_pd((A), (imm)), \
-                                   (__v4df)(__m256d)(W))
+  (__m256d)__builtin_ia32_extractf64x4_mask((__v8df)(__m512d)(A), (int)(imm), \
+                                            (__v4df)(__m256d)(W), \
+                                            (__mmask8)(U))
 
 #define _mm512_maskz_extractf64x4_pd(U, A, imm) \
-  (__m256d)__builtin_ia32_selectpd_256((__mmask8)(U), \
-                                   (__v4df)_mm512_extractf64x4_pd((A), (imm)), \
-                                   (__v4df)_mm256_setzero_pd())
+  (__m256d)__builtin_ia32_extractf64x4_mask((__v8df)(__m512d)(A), (int)(imm), \
+                                            (__v4df)_mm256_setzero_pd(), \
+                                            (__mmask8)(U))
 
 #define _mm512_extractf32x4_ps(A, I) \
-  (__m128)__builtin_ia32_extractf32x4((__v16sf)(__m512)(A), (int)(I))
+  (__m128)__builtin_ia32_extractf32x4_mask((__v16sf)(__m512)(A), (int)(I), \
+                                           (__v4sf)_mm_undefined_ps(), \
+                                           (__mmask8)-1)
 
 #define _mm512_mask_extractf32x4_ps(W, U, A, imm) \
-  (__m128)__builtin_ia32_selectps_128((__mmask8)(U), \
-                                   (__v4sf)_mm512_extractf32x4_ps((A), (imm)), \
-                                   (__v4sf)(__m128)(W))
+  (__m128)__builtin_ia32_extractf32x4_mask((__v16sf)(__m512)(A), (int)(imm), \
+                                           (__v4sf)(__m128)(W), \
+                                           (__mmask8)(U))
 
 #define _mm512_maskz_extractf32x4_ps(U, A, imm) \
-  (__m128)__builtin_ia32_selectps_128((__mmask8)(U), \
-                                   (__v4sf)_mm512_extractf32x4_ps((A), (imm)), \
-                                   (__v4sf)_mm_setzero_ps())
+  (__m128)__builtin_ia32_extractf32x4_mask((__v16sf)(__m512)(A), (int)(imm), \
+                                           (__v4sf)_mm_setzero_ps(), \
+                                           (__mmask8)(U))
 
 /* Vector Blend */
 
@@ -7534,30 +7538,34 @@ _mm512_mask_cvtepi64_storeu_epi16 (void *__P, __mmask8 __M, __m512i __A)
 }
 
 #define _mm512_extracti32x4_epi32(A, imm) \
-  (__m128i)__builtin_ia32_extracti32x4((__v16si)(__m512i)(A), (int)(imm))
+  (__m128i)__builtin_ia32_extracti32x4_mask((__v16si)(__m512i)(A), (int)(imm), \
+                                            (__v4si)_mm_undefined_si128(), \
+                                            (__mmask8)-1)
 
 #define _mm512_mask_extracti32x4_epi32(W, U, A, imm) \
-  (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
-                                (__v4si)_mm512_extracti32x4_epi32((A), (imm)), \
-                                (__v4si)(__m128i)(W))
+  (__m128i)__builtin_ia32_extracti32x4_mask((__v16si)(__m512i)(A), (int)(imm), \
+                                            (__v4si)(__m128i)(W), \
+                                            (__mmask8)(U))
 
 #define _mm512_maskz_extracti32x4_epi32(U, A, imm) \
-  (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
-                                (__v4si)_mm512_extracti32x4_epi32((A), (imm)), \
-                                (__v4si)_mm_setzero_si128())
+  (__m128i)__builtin_ia32_extracti32x4_mask((__v16si)(__m512i)(A), (int)(imm), \
+                                            (__v4si)_mm_setzero_si128(), \
+                                            (__mmask8)(U))
 
 #define _mm512_extracti64x4_epi64(A, imm) \
-  (__m256i)__builtin_ia32_extracti64x4((__v8di)(__m512i)(A), (int)(imm))
+  (__m256i)__builtin_ia32_extracti64x4_mask((__v8di)(__m512i)(A), (int)(imm), \
+                                            (__v4di)_mm256_undefined_si256(), \
+                                            (__mmask8)-1)
 
 #define _mm512_mask_extracti64x4_epi64(W, U, A, imm) \
-  (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
-                                (__v4di)_mm512_extracti64x4_epi64((A), (imm)), \
-                                (__v4di)(__m256i)(W))
+  (__m256i)__builtin_ia32_extracti64x4_mask((__v8di)(__m512i)(A), (int)(imm), \
+                                            (__v4di)(__m256i)(W), \
+                                            (__mmask8)(U))
 
 #define _mm512_maskz_extracti64x4_epi64(U, A, imm) \
-  (__m256i)__builtin_ia32_selectq_256((__mmask8)(U), \
-                                (__v4di)_mm512_extracti64x4_epi64((A), (imm)), \
-                                (__v4di)_mm256_setzero_si256())
+  (__m256i)__builtin_ia32_extracti64x4_mask((__v8di)(__m512i)(A), (int)(imm), \
+                                            (__v4di)_mm256_setzero_si256(), \
+                                            (__mmask8)(U))
 
 #define _mm512_insertf64x4(A, B, imm) \
   (__m512d)__builtin_ia32_insertf64x4((__v8df)(__m512d)(A), \

diff --git a/lib/Headers/avx512vldqintrin.h b/lib/Headers/avx512vldqintrin.h
index 8d6ff3ec8d9774d5bd767441e0dbd38189cd16ea..62c0fd7bd1d2bbf9bb692a17a922977cedc47f5d 100644 (file)
--- a/lib/Headers/avx512vldqintrin.h
+++ b/lib/Headers/avx512vldqintrin.h
@@ -1083,30 +1083,40 @@ _mm256_maskz_broadcast_i64x2 (__mmask8 __M, __m128i __A)
 }
 
 #define _mm256_extractf64x2_pd(A, imm) \
-  (__m128d)__builtin_ia32_extractf64x2_256((__v4df)(__m256d)(A), (int)(imm))
+  (__m128d)__builtin_ia32_extractf64x2_256_mask((__v4df)(__m256d)(A), \
+                                                (int)(imm), \
+                                                (__v2df)_mm_undefined_pd(), \
+                                                (__mmask8)-1)
 
 #define _mm256_mask_extractf64x2_pd(W, U, A, imm) \
-  (__m128d)__builtin_ia32_selectpd_128((__mmask8)(U), \
-                                   (__v2df)_mm256_extractf64x2_pd((A), (imm)), \
-                                   (__v2df)(__m128d)(W))
+  (__m128d)__builtin_ia32_extractf64x2_256_mask((__v4df)(__m256d)(A), \
+                                                (int)(imm), \
+                                                (__v2df)(__m128d)(W), \
+                                                (__mmask8)(U))
 
 #define _mm256_maskz_extractf64x2_pd(U, A, imm) \
-  (__m128d)__builtin_ia32_selectpd_128((__mmask8)(U), \
-                                   (__v2df)_mm256_extractf64x2_pd((A), (imm)), \
-                                   (__v2df)_mm_setzero_pd())
+  (__m128d)__builtin_ia32_extractf64x2_256_mask((__v4df)(__m256d)(A), \
+                                                (int)(imm), \
+                                                (__v2df)_mm_setzero_pd(), \
+                                                (__mmask8)(U))
 
 #define _mm256_extracti64x2_epi64(A, imm) \
-  (__m128i)__builtin_ia32_extracti64x2_256((__v4di)(__m256i)(A), (int)(imm))
+  (__m128i)__builtin_ia32_extracti64x2_256_mask((__v4di)(__m256i)(A), \
+                                                (int)(imm), \
+                                                (__v2di)_mm_undefined_si128(), \
+                                                (__mmask8)-1)
 
 #define _mm256_mask_extracti64x2_epi64(W, U, A, imm) \
-  (__m128i)__builtin_ia32_selectq_128((__mmask8)(U), \
-                                (__v2di)_mm256_extracti64x2_epi64((A), (imm)), \
-                                (__v2di)(__m128i)(W))
+  (__m128i)__builtin_ia32_extracti64x2_256_mask((__v4di)(__m256i)(A), \
+                                                (int)(imm), \
+                                                (__v2di)(__m128i)(W), \
+                                                (__mmask8)(U))
 
 #define _mm256_maskz_extracti64x2_epi64(U, A, imm) \
-  (__m128i)__builtin_ia32_selectq_128((__mmask8)(U), \
-                                (__v2di)_mm256_extracti64x2_epi64((A), (imm)), \
-                                (__v2di)_mm_setzero_si128())
+  (__m128i)__builtin_ia32_extracti64x2_256_mask((__v4di)(__m256i)(A), \
+                                                (int)(imm), \
+                                                (__v2di)_mm_setzero_si128(), \
+                                                (__mmask8)(U))
 
 #define _mm256_insertf64x2(A, B, imm) \
   (__m256d)__builtin_ia32_insertf64x2_256((__v4df)(__m256d)(A), \

diff --git a/lib/Headers/avx512vlintrin.h b/lib/Headers/avx512vlintrin.h
index 22a1fcb41e329e4b5d81683f943899e4e2931953..fea88e27c87402de6f17d4697f6a64de71d0a330 100644 (file)
--- a/lib/Headers/avx512vlintrin.h
+++ b/lib/Headers/avx512vlintrin.h
@@ -7699,30 +7699,40 @@ _mm256_mask_cvtepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
 }
 
 #define _mm256_extractf32x4_ps(A, imm) \
-  (__m128)__builtin_ia32_extractf32x4_256((__v8sf)(__m256)(A), (int)(imm))
+  (__m128)__builtin_ia32_extractf32x4_256_mask((__v8sf)(__m256)(A), \
+                                               (int)(imm), \
+                                               (__v4sf)_mm_undefined_ps(), \
+                                               (__mmask8)-1)
 
 #define _mm256_mask_extractf32x4_ps(W, U, A, imm) \
-  (__m128)__builtin_ia32_selectps_128((__mmask8)(U), \
-                                   (__v4sf)_mm256_extractf32x4_ps((A), (imm)), \
-                                   (__v4sf)(__m128)(W))
+  (__m128)__builtin_ia32_extractf32x4_256_mask((__v8sf)(__m256)(A), \
+                                               (int)(imm), \
+                                               (__v4sf)(__m128)(W), \
+                                               (__mmask8)(U))
 
 #define _mm256_maskz_extractf32x4_ps(U, A, imm) \
-  (__m128)__builtin_ia32_selectps_128((__mmask8)(U), \
-                                   (__v4sf)_mm256_extractf32x4_ps((A), (imm)), \
-                                   (__v4sf)_mm_setzero_ps())
+  (__m128)__builtin_ia32_extractf32x4_256_mask((__v8sf)(__m256)(A), \
+                                               (int)(imm), \
+                                               (__v4sf)_mm_setzero_ps(), \
+                                               (__mmask8)(U))
 
 #define _mm256_extracti32x4_epi32(A, imm) \
-  (__m128i)__builtin_ia32_extracti32x4_256((__v8si)(__m256i)(A), (int)(imm))
+  (__m128i)__builtin_ia32_extracti32x4_256_mask((__v8si)(__m256i)(A), \
+                                                (int)(imm), \
+                                                (__v4si)_mm_undefined_si128(), \
+                                                (__mmask8)-1)
 
 #define _mm256_mask_extracti32x4_epi32(W, U, A, imm) \
-  (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
-                                (__v4si)_mm256_extracti32x4_epi32((A), (imm)), \
-                                (__v4si)(__m128i)(W))
+  (__m128i)__builtin_ia32_extracti32x4_256_mask((__v8si)(__m256i)(A), \
+                                                (int)(imm), \
+                                                (__v4si)(__m128i)(W), \
+                                                (__mmask8)(U))
 
 #define _mm256_maskz_extracti32x4_epi32(U, A, imm) \
-  (__m128i)__builtin_ia32_selectd_128((__mmask8)(U), \
-                                (__v4si)_mm256_extracti32x4_epi32((A), (imm)), \
-                                (__v4si)_mm_setzero_si128())
+  (__m128i)__builtin_ia32_extracti32x4_256_mask((__v8si)(__m256i)(A), \
+                                                (int)(imm), \
+                                                (__v4si)_mm_setzero_si128(), \
+                                                (__mmask8)(U))
 
 #define _mm256_insertf32x4(A, B, imm) \
   (__m256)__builtin_ia32_insertf32x4_256((__v8sf)(__m256)(A), \

diff --git a/lib/Sema/SemaChecking.cpp b/lib/Sema/SemaChecking.cpp
index 3e0d36b384fa13342e8b41cd015181282d03888d..ae9c2f021fb4103c3cd65ab76bbe2809300cb1c8 100644 (file)
--- a/lib/Sema/SemaChecking.cpp
+++ b/lib/Sema/SemaChecking.cpp
@@ -2611,14 +2611,14 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_vextractf128_ps256:
   case X86::BI__builtin_ia32_vextractf128_si256:
   case X86::BI__builtin_ia32_extract128i256:
-  case X86::BI__builtin_ia32_extractf64x4:
-  case X86::BI__builtin_ia32_extracti64x4:
-  case X86::BI__builtin_ia32_extractf32x8:
-  case X86::BI__builtin_ia32_extracti32x8:
-  case X86::BI__builtin_ia32_extractf64x2_256:
-  case X86::BI__builtin_ia32_extracti64x2_256:
-  case X86::BI__builtin_ia32_extractf32x4_256:
-  case X86::BI__builtin_ia32_extracti32x4_256:
+  case X86::BI__builtin_ia32_extractf64x4_mask:
+  case X86::BI__builtin_ia32_extracti64x4_mask:
+  case X86::BI__builtin_ia32_extractf32x8_mask:
+  case X86::BI__builtin_ia32_extracti32x8_mask:
+  case X86::BI__builtin_ia32_extractf64x2_256_mask:
+  case X86::BI__builtin_ia32_extracti64x2_256_mask:
+  case X86::BI__builtin_ia32_extractf32x4_256_mask:
+  case X86::BI__builtin_ia32_extracti32x4_256_mask:
     i = 1; l = 0; u = 1;
     break;
   case X86::BI__builtin_ia32_vec_set_v2di:
@@ -2641,10 +2641,10 @@ bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   case X86::BI__builtin_ia32_vec_ext_v4si:
   case X86::BI__builtin_ia32_vec_ext_v4sf:
   case X86::BI__builtin_ia32_vec_ext_v4di:
-  case X86::BI__builtin_ia32_extractf32x4:
-  case X86::BI__builtin_ia32_extracti32x4:
-  case X86::BI__builtin_ia32_extractf64x2_512:
-  case X86::BI__builtin_ia32_extracti64x2_512:
+  case X86::BI__builtin_ia32_extractf32x4_mask:
+  case X86::BI__builtin_ia32_extracti32x4_mask:
+  case X86::BI__builtin_ia32_extractf64x2_512_mask:
+  case X86::BI__builtin_ia32_extracti64x2_512_mask:
     i = 1; l = 0; u = 3;
     break;
   case X86::BI_mm_prefetch: