[AVX512][clang] Amending vpmultishiftqb

[coby]

Summary

The 'vpmultishiftqb' instruction was implemented falsely, this patch amend it.
This is the clang part, llvm side is accessible here:
https://reviews.llvm.org/D26257

[coby]

deprecated.

include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_cvtsi2sd64, "V2dV2dLLiIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtsi2ss32, "V4fV4fiIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtsi2ss64, "V4fV4fLLiIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtss2sd_round_mask, "V2dV2dV4fV2dUcIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtusi2sd32, "V2dV2dUi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtusi2sd64, "V2dV2dULLiIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtusi2ss32, "V4fV4fUiIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtusi2ss64, "V4fV4fULLiIi","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_vpmultishiftqb512_mask, "V64cV64cV64cV64cULLi","","avx512vbmi")
-TARGET_BUILTIN(__builtin_ia32_vpmultishiftqb128_mask, "V16cV16cV16cV16cUs","","avx512vbmi,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_vpmultishiftqb256_mask, "V32cV32cV32cV32cUi","","avx512vbmi,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_vpmultishiftqb512_mask, "V8LLiV8LLiV8LLiV8LLiULLi","","avx512vbmi")
+TARGET_BUILTIN(__builtin_ia32_vpmultishiftqb128_mask, "V2LLiV2LLiV2LLiV2LLiUs","","avx512vbmi,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_vpmultishiftqb256_mask, "V4LLiV4LLiV4LLiV4LLiUi","","avx512vbmi,avx512vl")
 
 // generic select intrinsics
 TARGET_BUILTIN(__builtin_ia32_selectb_128, "V16cUsV16cV16c", "", "")
 TARGET_BUILTIN(__builtin_ia32_selectb_256, "V32cUiV32cV32c", "", "")
 TARGET_BUILTIN(__builtin_ia32_selectb_512, "V64cULLiV64cV64c", "", "")
 TARGET_BUILTIN(__builtin_ia32_selectw_128, "V8sUcV8sV8s", "", "")
 TARGET_BUILTIN(__builtin_ia32_selectw_256, "V16sUsV16sV16s", "", "")
 TARGET_BUILTIN(__builtin_ia32_selectw_512, "V32sUiV32sV32s", "", "")

lib/Headers/avx512vbmiintrin.h

return (__m512i) __builtin_ia32_permvarqi512_mask ((__v64qi) __B,
                  (__v64qi) __A,
                  (__v64qi) __W,
                  (__mmask64) __M);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_mask_multishift_epi64_epi8 (__m512i __W, __mmask64 __M, __m512i __X, __m512i __Y)
 {
-  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v64qi) __X,
-                (__v64qi) __Y,
-                (__v64qi) __W,
+  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v8di) __X,
+                (__v8di) __Y,
+                (__v8di) __W,
                 (__mmask64) __M);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_maskz_multishift_epi64_epi8 (__mmask64 __M, __m512i __X, __m512i __Y)
 {
-  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v64qi) __X,
-                (__v64qi) __Y,
-                (__v64qi) _mm512_setzero_si512 (),
+  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v8di) __X,
+                (__v8di) __Y,
+                (__v8di) _mm512_setzero_si512 (),
                 (__mmask64) __M);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_multishift_epi64_epi8 (__m512i __X, __m512i __Y)
 {
-  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v64qi) __X,
-                (__v64qi) __Y,
-                (__v64qi) _mm512_undefined_epi32 (),
+  return (__m512i) __builtin_ia32_vpmultishiftqb512_mask ((__v8di) __X,
+                (__v8di) __Y,
+                (__v8di) _mm512_undefined_epi32 (),
                 (__mmask64) -1);
 }
 
 
 #undef __DEFAULT_FN_ATTRS
 
 #endif

lib/Headers/avx512vbmivlintrin.h

return (__m256i) __builtin_ia32_permvarqi256_mask ((__v32qi) __B,
                  (__v32qi) __A,
                  (__v32qi) __W,
                  (__mmask32) __M);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_mask_multishift_epi64_epi8 (__m128i __W, __mmask16 __M, __m128i __X, __m128i __Y)
 {
-  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X,
-                (__v16qi) __Y,
-                (__v16qi) __W,
+  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v2di) __X,
+                (__v2di) __Y,
+                (__v2di) __W,
                 (__mmask16) __M);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_maskz_multishift_epi64_epi8 (__mmask16 __M, __m128i __X, __m128i __Y)
 {
-  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X,
-                (__v16qi) __Y,
-                (__v16qi)
+  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v2di) __X,
+                (__v2di) __Y,
+                (__v2di)
                 _mm_setzero_si128 (),
                 (__mmask16) __M);
 }
 
 static __inline__ __m128i __DEFAULT_FN_ATTRS
 _mm_multishift_epi64_epi8 (__m128i __X, __m128i __Y)
 {
-  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v16qi) __X,
-                (__v16qi) __Y,
-                (__v16qi)
+  return (__m128i) __builtin_ia32_vpmultishiftqb128_mask ((__v2di) __X,
+                (__v2di) __Y,
+                (__v2di)
                 _mm_undefined_si128 (),
                 (__mmask16) -1);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm256_mask_multishift_epi64_epi8 (__m256i __W, __mmask32 __M, __m256i __X, __m256i __Y)
 {
-  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X,
-                (__v32qi) __Y,
-                (__v32qi) __W,
+  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v4di) __X,
+                (__v4di) __Y,
+                (__v4di) __W,
                 (__mmask32) __M);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm256_maskz_multishift_epi64_epi8 (__mmask32 __M, __m256i __X, __m256i __Y)
 {
-  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X,
-                (__v32qi) __Y,
-                (__v32qi)
+  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v4di) __X,
+                (__v4di) __Y,
+                (__v4di)
                 _mm256_setzero_si256 (),
                 (__mmask32) __M);
 }
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS
 _mm256_multishift_epi64_epi8 (__m256i __X, __m256i __Y)
 {
-  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v32qi) __X,
-                (__v32qi) __Y,
-                (__v32qi)
+  return (__m256i) __builtin_ia32_vpmultishiftqb256_mask ((__v4di) __X,
+                (__v4di) __Y,
+                (__v4di)
                 _mm256_undefined_si256 (),
                 (__mmask32) -1);
 }
 
 
 #undef __DEFAULT_FN_ATTRS
 
 #endif