[CLANG] [AVX512] [BUILTIN] Adding PSHUF{L|H}W{128|256|512} builtin to clang .

[delena]

One minor fix.

lib/Headers/avx512vlbwintrin.h

36

"{" should be moved one line up

include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_pmovzxwq128_mask, "V2LLiV8sV2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_pmovzxwq256_mask, "V4LLiV8sV4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_prold512_mask, "V16iV16iIiV16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_prolq512_mask, "V8LLiV8LLiIiV8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_prold128_mask, "V4iV4iIiV4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_prold256_mask, "V8iV8iIiV8iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_prolq128_mask, "V2LLiV2LLiIiV2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_prolq256_mask, "V4LLiV4LLiIiV4LLiUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pshufhw512_mask, "V32sV32sIiV32sUi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pshuflw512_mask, "V32sV32sIiV32sUi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_pshufhw128_mask, "V8sV8sIiV8sUc","","avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pshufhw256_mask, "V16sV16sIiV16sUs","","avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pshuflw128_mask, "V8sV8sIiV8sUc","","avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_pshuflw256_mask, "V16sV16sIiV16sUs","","avx512bw,avx512vl")
 
 #undef BUILTIN
 #undef TARGET_BUILTIN

lib/Headers/avx512bwintrin.h

(__mmask16)__builtin_ia32_ucmpw512_mask((__v32hi)(__m512i)(a), \
                                           (__v32hi)(__m512i)(b), \
                                           (p), (__mmask32)-1); })
 
 #define _mm512_mask_cmp_epu16_mask(m, a, b, p) __extension__ ({ \
   (__mmask16)__builtin_ia32_ucmpw512_mask((__v32hi)(__m512i)(a), \
                                           (__v32hi)(__m512i)(b), \
                                           (p), (__mmask32)(m)); })
 
+#define _mm512_shufflehi_epi16( __A, __imm) __extension__ ({ \
+__builtin_ia32_pshufhw512_mask ((__v32hi)( __A),\
+              ( __imm),\
+               (__v32hi)\
+               _mm512_setzero_hi (),\
+               (__mmask32) -1);\
+})
+
+#define _mm512_mask_shufflehi_epi16( __W, __U, __A ,__imm) __extension__ ({ \
+__builtin_ia32_pshufhw512_mask ((__v32hi) (__A),\
+               (__imm),\
+               (__v32hi)( __W),\
+               (__mmask32)( __U));\
+})
+
+
+#define _mm512_maskz_shufflehi_epi16( __U, __A ,__imm) __extension__ ({ \
+__builtin_ia32_pshufhw512_mask ((__v32hi) (__A),\
+               (__imm),\
+               (__v32hi)\
+               _mm512_setzero_hi (),\
+               (__mmask32)( __U));\
+})
+
+#define _mm512_shufflelo_epi16( __A, __imm) __extension__ ({ \
+__builtin_ia32_pshuflw512_mask ((__v32hi)( __A),\
+              ( __imm),\
+               (__v32hi)\
+               _mm512_setzero_hi (),\
+               (__mmask32) -1);\
+})
+
+
+#define _mm512_mask_shufflelo_epi16( __W, __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshuflw512_mask ((__v32hi) (__A),\
+               (__imm),\
+               (__v32hi)( __W),\
+               (__mmask32)( __U));\
+})
+
+
+#define _mm512_maskz_shufflelo_epi16( __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshuflw512_mask ((__v32hi) (__A),\
+               (__imm),\
+               (__v32hi)\
+               _mm512_setzero_hi (),\
+               (__mmask32)( __U));\
+})
 
 #undef __DEFAULT_FN_ATTRS
 
 #endif

lib/Headers/avx512vlbwintrin.h

 #endif
 
 #ifndef __AVX512VLBWINTRIN_H
 #define __AVX512VLBWINTRIN_H
 
 /* Define the default attributes for the functions in this file. */
 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512bw")))
 
+static  __inline __m128i __DEFAULT_FN_ATTRS
+_mm_setzero_hi(void)
+{
+    return (__m128i){ 0LL, 0LL };
+}
+
 /* Integer compare */
 
 static __inline__ __mmask16 __DEFAULT_FN_ATTRS
 _mm_cmpeq_epi8_mask(__m128i __a, __m128i __b) {
   return (__mmask16)__builtin_ia32_pcmpeqb128_mask((__v16qi)__a, (__v16qi)__b,
                                                    (__mmask16)-1);
 }
 
(__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)(__m256i)(a), \
                                           (__v16hi)(__m256i)(b), \
                                           (p), (__mmask16)-1); })
 
 #define _mm256_mask_cmp_epu16_mask(m, a, b, p) __extension__ ({ \
   (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)(__m256i)(a), \
                                           (__v16hi)(__m256i)(b), \
                                           (p), (__mmask16)(m)); })
 
+#define _mm_mask_shufflehi_epi16( __W, __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshufhw128_mask ((__v8hi) __A, (__imm),\
+               (__v8hi)( __W),\
+               (__mmask8)( __U));\
+})
+
+#define _mm_maskz_shufflehi_epi16( __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshufhw128_mask ((__v8hi)( __A),( __imm),\
+               (__v8hi)\
+               _mm_setzero_hi (),\
+               (__mmask8)( __U));\
+})
+
+
+#define _mm256_mask_shufflehi_epi16( __W, __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshufhw256_mask ((__v16hi) (__A),\
+               (__imm),\
+               (__v16hi)( __W),\
+               (__mmask16)( __U));\
+})
+
+
+#define _mm256_maskz_shufflehi_epi16( __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshufhw256_mask ((__v16hi) (__A),\
+               (__imm),\
+               (__v16hi)\
+               _mm256_setzero_si256 (),\
+               (__mmask16)( __U));\
+})
+
+
+#define _mm_mask_shufflelo_epi16( __W, __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshuflw128_mask ((__v8hi) __A, (__imm),\
+               (__v8hi)( __W),\
+               (__mmask8)( __U));\
+})
+
+#define _mm_maskz_shufflelo_epi16( __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshuflw128_mask ((__v8hi)( __A),( __imm),\
+               (__v8hi)\
+               _mm_setzero_hi (),\
+               (__mmask8)( __U));\
+})
+
+
+#define _mm256_mask_shufflelo_epi16( __W, __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshuflw256_mask ((__v16hi) (__A),\
+               (__imm),\
+               (__v16hi)( __W),\
+               (__mmask16)( __U));\
+})
+
+
+#define _mm256_maskz_shufflelo_epi16( __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_pshuflw256_mask ((__v16hi) (__A),\
+               (__imm),\
+               (__v16hi)\
+               _mm256_setzero_si256 (),\
+               (__mmask16)( __U));\
+})
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif /* __AVX512VLBWINTRIN_H */

test/CodeGen/avx512bw-builtins.c

__m512i test_mm512_mask_cvtepu8_epi16(__m512i __W, __mmask32 __U, __m256i __A) {
   return _mm512_mask_cvtepu8_epi16(__W, __U, __A); 
 }
 
 __m512i test_mm512_maskz_cvtepu8_epi16(__mmask32 __U, __m256i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvtepu8_epi16
   // CHECK: @llvm.x86.avx512.mask.pmovzxb.w.512
   return _mm512_maskz_cvtepu8_epi16(__U, __A); 
 }
+
+__m512i test_mm512_shufflehi_epi16(__m512i __A) {
+  // CHECK-LABEL: @test_mm512_shufflehi_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufh.w.512
+  return _mm512_shufflehi_epi16(__A, 5); 
+}
+
+__m512i test_mm512_mask_shufflehi_epi16(__m512i __W, __mmask32 __U, __m512i __A) {
+  // CHECK-LABEL: @test_mm512_mask_shufflehi_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufh.w.512
+  return _mm512_mask_shufflehi_epi16(__W, __U, __A, 5); 
+}
+
+__m512i test_mm512_maskz_shufflehi_epi16(__mmask32 __U, __m512i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_shufflehi_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufh.w.512
+  return _mm512_maskz_shufflehi_epi16(__U, __A, 5); 
+}
+
+__m512i test_mm512_shufflelo_epi16(__m512i __A) {
+  // CHECK-LABEL: @test_mm512_shufflelo_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufl.w.512
+  return _mm512_shufflelo_epi16(__A, 5); 
+}
+
+__m512i test_mm512_mask_shufflelo_epi16(__m512i __W, __mmask32 __U, __m512i __A) {
+  // CHECK-LABEL: @test_mm512_mask_shufflelo_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufl.w.512
+  return _mm512_mask_shufflelo_epi16(__W, __U, __A, 5); 
+}
+
+__m512i test_mm512_maskz_shufflelo_epi16(__mmask32 __U, __m512i __A) {
+  // CHECK-LABEL: @test_mm512_maskz_shufflelo_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufl.w.512
+  return _mm512_maskz_shufflelo_epi16(__U, __A, 5); 
+}
+
+

test/CodeGen/avx512vlbw-builtins.c

__m256i test_mm256_mask_cvtepu8_epi16(__m256i __W, __mmask32 __U, __m128i __A) {
   // CHECK: @llvm.x86.avx512.mask.pmovzxb.w.256
   return _mm256_mask_cvtepu8_epi16(__W, __U, __A); 
 }
 
 __m256i test_mm256_maskz_cvtepu8_epi16(__mmask16 __U, __m128i __A) {
   // CHECK-LABEL: @test_mm256_maskz_cvtepu8_epi16
   // CHECK: @llvm.x86.avx512.mask.pmovzxb.w.256
   return _mm256_maskz_cvtepu8_epi16(__U, __A); 
-}
+}
+
+__m128i test_mm_mask_shufflehi_epi16(__m128i __W, __mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_shufflehi_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufh.w.128
+  return _mm_mask_shufflehi_epi16(__W, __U, __A, 5); 
+}
+
+__m128i test_mm_maskz_shufflehi_epi16(__mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_maskz_shufflehi_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufh.w.128
+  return _mm_maskz_shufflehi_epi16(__U, __A, 5); 
+}
+
+__m256i test_mm256_mask_shufflehi_epi16(__m256i __W, __mmask16 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_mask_shufflehi_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufh.w.256
+  return _mm256_mask_shufflehi_epi16(__W, __U, __A, 5); 
+}
+
+__m256i test_mm256_maskz_shufflehi_epi16(__mmask16 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_shufflehi_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufh.w.256
+  return _mm256_maskz_shufflehi_epi16(__U, __A, 5); 
+}
+
+__m128i test_mm_mask_shufflelo_epi16(__m128i __W, __mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_mask_shufflelo_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufl.w.128
+  return _mm_mask_shufflelo_epi16(__W, __U, __A, 5); 
+}
+
+__m128i test_mm_maskz_shufflelo_epi16(__mmask8 __U, __m128i __A) {
+  // CHECK-LABEL: @test_mm_maskz_shufflelo_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufl.w.128
+  return _mm_maskz_shufflelo_epi16(__U, __A, 5); 
+}
+
+__m256i test_mm256_mask_shufflelo_epi16(__m256i __W, __mmask16 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_mask_shufflelo_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufl.w.256
+  return _mm256_mask_shufflelo_epi16(__W, __U, __A, 5); 
+}
+
+__m256i test_mm256_maskz_shufflelo_epi16(__mmask16 __U, __m256i __A) {
+  // CHECK-LABEL: @test_mm256_maskz_shufflelo_epi16
+  // CHECK: @llvm.x86.avx512.mask.pshufl.w.256
+  return _mm256_maskz_shufflelo_epi16(__U, __A, 5); 
+}
+