[X86] Remove mask parameter from avx512 pmultishiftqb intrinsics. Use select in IR instead.

Fixes PR40259

llvm-svn: 351036

3 files changed, 55 insertions, 51 deletions

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 3ab814a37868..b1799c0491a6 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -11152,6 +11152,26 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     return EmitX86MaskedCompareResult(*this, Fpclass, NumElts, MaskIn);
   }
 
+  case X86::BI__builtin_ia32_vpmultishiftqb128:
+  case X86::BI__builtin_ia32_vpmultishiftqb256:
+  case X86::BI__builtin_ia32_vpmultishiftqb512: {
+    Intrinsic::ID ID;
+    switch (BuiltinID) {
+    default: llvm_unreachable("Unsupported intrinsic!");
+    case X86::BI__builtin_ia32_vpmultishiftqb128:
+      ID = Intrinsic::x86_avx512_pmultishift_qb_128;
+      break;
+    case X86::BI__builtin_ia32_vpmultishiftqb256:
+      ID = Intrinsic::x86_avx512_pmultishift_qb_256;
+      break;
+    case X86::BI__builtin_ia32_vpmultishiftqb512:
+      ID = Intrinsic::x86_avx512_pmultishift_qb_512;
+      break;
+    }
+
+    return Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
+  }
+
   case X86::BI__builtin_ia32_vpshufbitqmb128_mask:
   case X86::BI__builtin_ia32_vpshufbitqmb256_mask:
   case X86::BI__builtin_ia32_vpshufbitqmb512_mask: {
@@ -11173,8 +11193,8 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
       break;
     }
 
-    Value *Fpclass = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
-    return EmitX86MaskedCompareResult(*this, Fpclass, NumElts, MaskIn);
+    Value *Shufbit = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
+    return EmitX86MaskedCompareResult(*this, Shufbit, NumElts, MaskIn);
   }
 
   // packed comparison intrinsics
diff --git a/clang/lib/Headers/avx512vbmiintrin.h b/clang/lib/Headers/avx512vbmiintrin.h
index b6e93c285871..5463d9015504 100644
--- a/clang/lib/Headers/avx512vbmiintrin.h
+++ b/clang/lib/Headers/avx512vbmiintrin.h
@@ -91,30 +91,26 @@ _mm512_mask_permutexvar_epi8 (__m512i __W, __mmask64 __M, __m512i __A,
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_multishift_epi64_epi8 (__m512i __W, __mmask64 __M, __m512i __X, __m512i __Y)
+_mm512_multishift_epi64_epi8(__m512i __X, __m512i __Y)
 {
-  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v64qi) __X,
-                (__v64qi) __Y,
-                (__v64qi) __W,
-                (__mmask64) __M);
+  return (__m512i)__builtin_ia32_vpmultishiftqb512((__v64qi)__X, (__v64qi) __Y);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_multishift_epi64_epi8 (__mmask64 __M, __m512i __X, __m512i __Y)
+_mm512_mask_multishift_epi64_epi8(__m512i __W, __mmask64 __M, __m512i __X,
+                                  __m512i __Y)
 {
-  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v64qi) __X,
-                (__v64qi) __Y,
-                (__v64qi) _mm512_setzero_si512 (),
-                (__mmask64) __M);
+  return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,
+                                (__v64qi)_mm512_multishift_epi64_epi8(__X, __Y),
+                                (__v64qi)__W);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_multishift_epi64_epi8 (__m512i __X, __m512i __Y)
+_mm512_maskz_multishift_epi64_epi8(__mmask64 __M, __m512i __X, __m512i __Y)
 {
-  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v64qi) __X,
-                (__v64qi) __Y,
-                (__v64qi) _mm512_undefined_epi32 (),
-                (__mmask64) -1);
+  return (__m512i)__builtin_ia32_selectb_512((__mmask64)__M,
+                                (__v64qi)_mm512_multishift_epi64_epi8(__X, __Y),
+                                (__v64qi)_mm512_setzero_si512());
 }
 
 
diff --git a/clang/lib/Headers/avx512vbmivlintrin.h b/clang/lib/Headers/avx512vbmivlintrin.h
index 9a0400b2b5d5..b5d5aa9af523 100644
--- a/clang/lib/Headers/avx512vbmivlintrin.h
+++ b/clang/lib/Headers/avx512vbmivlintrin.h
@@ -150,61 +150,49 @@ _mm256_mask_permutexvar_epi8 (__m256i __W, __mmask32 __M, __m256i __A,
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_mask_multishift_epi64_epi8 (__m128i __W, __mmask16 __M, __m128i __X, __m128i __Y)
+_mm_multishift_epi64_epi8(__m128i __X, __m128i __Y)
 {
-  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X,
-                (__v16qi) __Y,
-                (__v16qi) __W,
-                (__mmask16) __M);
+  return (__m128i)__builtin_ia32_vpmultishiftqb128((__v16qi)__X, (__v16qi)__Y);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_maskz_multishift_epi64_epi8 (__mmask16 __M, __m128i __X, __m128i __Y)
+_mm_mask_multishift_epi64_epi8(__m128i __W, __mmask16 __M, __m128i __X,
+                               __m128i __Y)
 {
-  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X,
-                (__v16qi) __Y,
-                (__v16qi)
-                _mm_setzero_si128 (),
-                (__mmask16) __M);
+  return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
+                                   (__v16qi)_mm_multishift_epi64_epi8(__X, __Y),
+                                   (__v16qi)__W);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_multishift_epi64_epi8 (__m128i __X, __m128i __Y)
+_mm_maskz_multishift_epi64_epi8(__mmask16 __M, __m128i __X, __m128i __Y)
 {
-  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X,
-                (__v16qi) __Y,
-                (__v16qi)
-                _mm_undefined_si128 (),
-                (__mmask16) -1);
+  return (__m128i)__builtin_ia32_selectb_128((__mmask16)__M,
+                                   (__v16qi)_mm_multishift_epi64_epi8(__X, __Y),
+                                   (__v16qi)_mm_setzero_si128());
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_multishift_epi64_epi8 (__m256i __W, __mmask32 __M, __m256i __X, __m256i __Y)
+_mm256_multishift_epi64_epi8(__m256i __X, __m256i __Y)
 {
-  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X,
-                (__v32qi) __Y,
-                (__v32qi) __W,
-                (__mmask32) __M);
+  return (__m256i)__builtin_ia32_vpmultishiftqb256((__v32qi)__X, (__v32qi)__Y);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_maskz_multishift_epi64_epi8 (__mmask32 __M, __m256i __X, __m256i __Y)
+_mm256_mask_multishift_epi64_epi8(__m256i __W, __mmask32 __M, __m256i __X,
+                                  __m256i __Y)
 {
-  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X,
-                (__v32qi) __Y,
-                (__v32qi)
-                _mm256_setzero_si256 (),
-                (__mmask32) __M);
+  return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
+                                (__v32qi)_mm256_multishift_epi64_epi8(__X, __Y),
+                                (__v32qi)__W);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_multishift_epi64_epi8 (__m256i __X, __m256i __Y)
+_mm256_maskz_multishift_epi64_epi8(__mmask32 __M, __m256i __X, __m256i __Y)
 {
-  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X,
-                (__v32qi) __Y,
-                (__v32qi)
-                _mm256_undefined_si256 (),
-                (__mmask32) -1);
+  return (__m256i)__builtin_ia32_selectb_256((__mmask32)__M,
+                                (__v32qi)_mm256_multishift_epi64_epi8(__X, __Y),
+                                (__v32qi)_mm256_setzero_si256());
 }