[Clang][Builtin][AVX512] Adding intrinsics for vmovshdup and vmovsldup instruction set

[m_zuckerman]

Ping

[AsafBadouh]

LGTM

cfe/trunk/include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_dbpsadbw512_mask, "V32sV64cV64cIiV32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_psadbw512, "V8LLiV64cV64c","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_compressdf512_mask, "V8dV8dV8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compressdi512_mask, "V8LLiV8LLiV8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compresssf512_mask, "V16fV16fV16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compresssi512_mask, "V16iV16iV16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cmpsd_mask, "UcV2dV2dIiUcIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cmpss_mask, "UcV4fV4fIiUcIi","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_movshdup512_mask, "V16fV16fV16fUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_movsldup512_mask, "V16fV16fV16fUs","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_movshdup128_mask, "V4fV4fV4fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_movshdup256_mask, "V8fV8fV8fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_movsldup128_mask, "V4fV4fV4fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_movsldup256_mask, "V8fV8fV8fUc","","avx512vl")
 
 #undef BUILTIN
 #undef TARGET_BUILTIN

cfe/trunk/lib/Headers/avx512fintrin.h

 
 #define _mm_mask_cmp_sd_mask( __M, __X, __Y, __P) __extension__ ({ \
 __builtin_ia32_cmpsd_mask ((__v2df)( __X),\
                  (__v2df)( __Y),( __P),\
                  (__mmask8)( __M),\
                  _MM_FROUND_CUR_DIRECTION);\
 })
 
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_movehdup_ps (__m512 __A)
+{
+  return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
+               (__v16sf)
+               _mm512_undefined_ps (),
+               (__mmask16) -1);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_mask_movehdup_ps (__m512 __W, __mmask16 __U, __m512 __A)
+{
+  return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
+               (__v16sf) __W,
+               (__mmask16) __U);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_maskz_movehdup_ps (__mmask16 __U, __m512 __A)
+{
+  return (__m512) __builtin_ia32_movshdup512_mask ((__v16sf) __A,
+               (__v16sf)
+               _mm512_setzero_ps (),
+               (__mmask16) __U);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_moveldup_ps (__m512 __A)
+{
+  return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
+               (__v16sf)
+               _mm512_undefined_ps (),
+               (__mmask16) -1);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_mask_moveldup_ps (__m512 __W, __mmask16 __U, __m512 __A)
+{
+  return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
+               (__v16sf) __W,
+               (__mmask16) __U);
+}
+
+static __inline__ __m512 __DEFAULT_FN_ATTRS
+_mm512_maskz_moveldup_ps (__mmask16 __U, __m512 __A)
+{
+  return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
+               (__v16sf)
+               _mm512_setzero_ps (),
+               (__mmask16) __U);
+}
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif // __AVX512FINTRIN_H

cfe/trunk/lib/Headers/avx512vlintrin.h

 
 #define _mm256_maskz_alignr_epi64( __U, __A, __B, __imm) __extension__ ({ \
 __builtin_ia32_alignq256_mask ((__v4di)( __A),\
               (__v4di)( __B),( __imm),\
               (__v4di) _mm256_setzero_si256 (),\
               (__mmask8)( __U));\
 })
 
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_movehdup_ps (__m128 __W, __mmask8 __U, __m128 __A)
+{
+  return (__m128) __builtin_ia32_movshdup128_mask ((__v4sf) __A,
+               (__v4sf) __W,
+               (__mmask8) __U);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_movehdup_ps (__mmask8 __U, __m128 __A)
+{
+  return (__m128) __builtin_ia32_movshdup128_mask ((__v4sf) __A,
+               (__v4sf)
+               _mm_setzero_ps (),
+               (__mmask8) __U);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_mask_movehdup_ps (__m256 __W, __mmask8 __U, __m256 __A)
+{
+  return (__m256) __builtin_ia32_movshdup256_mask ((__v8sf) __A,
+               (__v8sf) __W,
+               (__mmask8) __U);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_maskz_movehdup_ps (__mmask8 __U, __m256 __A)
+{
+  return (__m256) __builtin_ia32_movshdup256_mask ((__v8sf) __A,
+               (__v8sf)
+               _mm256_setzero_ps (),
+               (__mmask8) __U);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_moveldup_ps (__m128 __W, __mmask8 __U, __m128 __A)
+{
+  return (__m128) __builtin_ia32_movsldup128_mask ((__v4sf) __A,
+               (__v4sf) __W,
+               (__mmask8) __U);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_moveldup_ps (__mmask8 __U, __m128 __A)
+{
+  return (__m128) __builtin_ia32_movsldup128_mask ((__v4sf) __A,
+               (__v4sf)
+               _mm_setzero_ps (),
+               (__mmask8) __U);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_mask_moveldup_ps (__m256 __W, __mmask8 __U, __m256 __A)
+{
+  return (__m256) __builtin_ia32_movsldup256_mask ((__v8sf) __A,
+               (__v8sf) __W,
+               (__mmask8) __U);
+}
+
+static __inline__ __m256 __DEFAULT_FN_ATTRS
+_mm256_maskz_moveldup_ps (__mmask8 __U, __m256 __A)
+{
+  return (__m256) __builtin_ia32_movsldup256_mask ((__v8sf) __A,
+               (__v8sf)
+               _mm256_setzero_ps (),
+               (__mmask8) __U);
+}
+
 #undef __DEFAULT_FN_ATTRS
 #undef __DEFAULT_FN_ATTRS_BOTH
 
 #endif /* __AVX512VLINTRIN_H */

cfe/trunk/test/CodeGen/avx512f-builtins.c

__mmask8 test_mm_cmp_sd_mask(__m128d __X, __m128d __Y) {
   return _mm_cmp_sd_mask(__X, __Y, 5); 
 }
 
 __mmask8 test_mm_mask_cmp_sd_mask(__mmask8 __M, __m128d __X, __m128d __Y) {
   // CHECK-LABEL: @test_mm_mask_cmp_sd_mask
   // CHECK: @llvm.x86.avx512.mask.cmp
   return _mm_mask_cmp_sd_mask(__M, __X, __Y, 5); 
 }
+
+__m512 test_mm512_movehdup_ps(__m512 __A) {
+  // CHECK-LABEL: @test_mm512_movehdup_ps
+  // CHECK: @llvm.x86.avx512.mask.movshdup.512
+  return _mm512_movehdup_ps(__A); 
+}
+
+__m512 test_mm512_mask_movehdup_ps(__m512 __W, __mmask16 __U, __m512 __A) {
+  // CHECK-LABEL: @test_mm512_mask_movehdup_ps
+  // CHECK: @llvm.x86.avx512.mask.movshdup.512
+  return _mm512_mask_movehdup_ps(__W, __U, __A); 
+}
+
+__m512 test_mm512_maskz_movehdup_ps(__mmask16 __U, __m512 __A) {
+  // CHECK-LABEL: @test_mm512_maskz_movehdup_ps
+  // CHECK: @llvm.x86.avx512.mask.movshdup.512
+  return _mm512_maskz_movehdup_ps(__U, __A); 
+}
+
+__m512 test_mm512_moveldup_ps(__m512 __A) {
+  // CHECK-LABEL: @test_mm512_moveldup_ps
+  // CHECK: @llvm.x86.avx512.mask.movsldup.512
+  return _mm512_moveldup_ps(__A); 
+}
+
+__m512 test_mm512_mask_moveldup_ps(__m512 __W, __mmask16 __U, __m512 __A) {
+  // CHECK-LABEL: @test_mm512_mask_moveldup_ps
+  // CHECK: @llvm.x86.avx512.mask.movsldup.512
+  return _mm512_mask_moveldup_ps(__W, __U, __A); 
+}
+
+__m512 test_mm512_maskz_moveldup_ps(__mmask16 __U, __m512 __A) {
+  // CHECK-LABEL: @test_mm512_maskz_moveldup_ps
+  // CHECK: @llvm.x86.avx512.mask.movsldup.512
+  return _mm512_maskz_moveldup_ps(__U, __A); 
+}

cfe/trunk/test/CodeGen/avx512vl-builtins.c

__m256i test_mm256_mask_alignr_epi64(__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
   return _mm256_mask_alignr_epi64(__W, __U, __A, __B, 1); 
 }
 
 __m256i test_mm256_maskz_alignr_epi64(__mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: @test_mm256_maskz_alignr_epi64
   // CHECK: @llvm.x86.avx512.mask.valign.q.256
   return _mm256_maskz_alignr_epi64(__U, __A, __B, 1); 
 }
+
+__m128 test_mm_mask_movehdup_ps(__m128 __W, __mmask8 __U, __m128 __A) {
+  // CHECK-LABEL: @test_mm_mask_movehdup_ps
+  // CHECK: @llvm.x86.avx512.mask.movshdup.128
+  return _mm_mask_movehdup_ps(__W, __U, __A); 
+}
+
+__m128 test_mm_maskz_movehdup_ps(__mmask8 __U, __m128 __A) {
+  // CHECK-LABEL: @test_mm_maskz_movehdup_ps
+  // CHECK: @llvm.x86.avx512.mask.movshdup.128
+  return _mm_maskz_movehdup_ps(__U, __A); 
+}
+
+__m256 test_mm256_mask_movehdup_ps(__m256 __W, __mmask8 __U, __m256 __A) {
+  // CHECK-LABEL: @test_mm256_mask_movehdup_ps
+  // CHECK: @llvm.x86.avx512.mask.movshdup.256
+  return _mm256_mask_movehdup_ps(__W, __U, __A); 
+}
+
+__m256 test_mm256_maskz_movehdup_ps(__mmask8 __U, __m256 __A) {
+  // CHECK-LABEL: @test_mm256_maskz_movehdup_ps
+  // CHECK: @llvm.x86.avx512.mask.movshdup.256
+  return _mm256_maskz_movehdup_ps(__U, __A); 
+}
+
+__m128 test_mm_mask_moveldup_ps(__m128 __W, __mmask8 __U, __m128 __A) {
+  // CHECK-LABEL: @test_mm_mask_moveldup_ps
+  // CHECK: @llvm.x86.avx512.mask.movsldup.128
+  return _mm_mask_moveldup_ps(__W, __U, __A); 
+}
+
+__m128 test_mm_maskz_moveldup_ps(__mmask8 __U, __m128 __A) {
+  // CHECK-LABEL: @test_mm_maskz_moveldup_ps
+  // CHECK: @llvm.x86.avx512.mask.movsldup.128
+  return _mm_maskz_moveldup_ps(__U, __A); 
+}
+
+__m256 test_mm256_mask_moveldup_ps(__m256 __W, __mmask8 __U, __m256 __A) {
+  // CHECK-LABEL: @test_mm256_mask_moveldup_ps
+  // CHECK: @llvm.x86.avx512.mask.movsldup.256
+  return _mm256_mask_moveldup_ps(__W, __U, __A); 
+}
+
+__m256 test_mm256_maskz_moveldup_ps(__mmask8 __U, __m256 __A) {
+  // CHECK-LABEL: @test_mm256_maskz_moveldup_ps
+  // CHECK: @llvm.x86.avx512.mask.movsldup.256
+  return _mm256_maskz_moveldup_ps(__U, __A); 
+}