[Clang][AVX512][BuiltIn]Adding intrinsics move_{sd|ss} to clang

[delena]

lib/Headers/avx512fintrin.h

9124

please try the following:
if (__U)

return __builtin_shuffle(A, B, (0, 5, 6, 7)); // may be you need to swap A and B

return W;

I know that the immediate code will be less optimal, but we can optimize it later.

[RKSimon]

lib/Headers/avx512fintrin.h

9124

Any update on this? I currently have a patch (D24653) looking at removing the movss/movsd mask intrinsics as we should be able to do this with purely generic shuffles. I can help with the optimization if necessary.

include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_expandsi512_mask, "V16iV16iV16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtps2pd512_mask, "V8dV8fV8dUcIi","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_movapd512_mask, "V8dV8dV8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_movaps512_mask, "V16fV16fV16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_movapd128_mask, "V2dV2dV2dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_movapd256_mask, "V4dV4dV4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_movaps128_mask, "V4fV4fV4fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_movaps256_mask, "V8fV8fV8fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_movss_mask, "V4fV4fV4fV4fUc","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_movsd_mask, "V2dV2dV2dV2dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compressstoredf512_mask, "vV8d*V8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compressstoredi512_mask, "vV8LLi*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compressstoresf512_mask, "vV16f*V16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compressstoresi512_mask, "vV16i*V16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_vcvtph2ps_mask, "V4fV8sV4fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_vcvtph2ps256_mask, "V8fV8sV8fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_vcvtps2ph_mask, "V8sV4fIiV8sUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_vcvtps2ph256_mask, "V8sV8fIiV8sUc","","avx512vl")

lib/Headers/avx512fintrin.h

 _mm512_maskz_moveldup_ps (__mmask16 __U, __m512 __A)
 {
   return (__m512) __builtin_ia32_movsldup512_mask ((__v16sf) __A,
                (__v16sf)
                _mm512_setzero_ps (),
                (__mmask16) __U);
 }
 
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_move_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
+{
+  return (__m128) __builtin_ia32_movss_mask ((__v4sf) __A, (__v4sf) __B,
+               (__v4sf) __W,
+               (__mmask8) __U);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_move_ss (__mmask8 __U, __m128 __A, __m128 __B)
+{
+  return (__m128) __builtin_ia32_movss_mask ((__v4sf) __A, (__v4sf) __B,
+               (__v4sf)
+               _mm_setzero_si128(),
+               (__mmask8) __U);
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_mask_move_sd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
+{
+  return (__m128d) __builtin_ia32_movsd_mask ((__v2df) __A, (__v2df) __B,
+               (__v2df) __W,
+               (__mmask8) __U);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_move_sd (__mmask8 __U, __m128d __A, __m128d __B)
+{
+  return (__m128d) __builtin_ia32_movsd_mask ((__v2df) __A, (__v2df) __B,
+               (__v2di)
+               _mm_setzero_pd (),
+               (__mmask8) __U);
+}
+
 #define _mm512_shuffle_epi32(A, I) __extension__ ({ \
   (__m512i)__builtin_ia32_pshufd512_mask((__v16si)(__m512i)(A), (int)(I), \
                                          (__v16si)_mm512_undefined_epi32(), \
                                          (__mmask16)-1); })
 
 #define _mm512_mask_shuffle_epi32(W, U, A, I) __extension__ ({ \
   (__m512i)__builtin_ia32_pshufd512_mask((__v16si)(__m512i)(A), (int)(I), \
                                          (__v16si)(__m512i)(W), \

test/CodeGen/avx512f-builtins.c

__m512i test_mm512_setzero_ps()
   return _mm512_setzero_ps();
 }
 
 __m512d test_mm512_setzero_pd()
 {
   // CHECK-LABEL: @test_mm512_setzero_pd
   // CHECK: zeroinitializer
   return _mm512_setzero_pd();
-}
+}
+
+__m128 test_mm_mask_move_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
+{
+  // CHECK-LABEL: @test_mm_mask_move_ss
+  // CHECK: @llvm.x86.avx512.mask.move.ss
+  return _mm_mask_move_ss ( __W,  __U,  __A,  __B);
+}
+
+__m128 test_mm_maskz_move_ss (__mmask8 __U, __m128 __A, __m128 __B)
+{
+  // CHECK-LABEL: @test_mm_maskz_move_ss
+  // CHECK: @llvm.x86.avx512.mask.move.ss
+  return _mm_maskz_move_ss (__U, __A, __B);
+}
+
+__m128d test_mm_mask_move_sd (__m128 __W, __mmask8 __U, __m128d __A, __m128d __B)
+{
+  // CHECK-LABEL: @test_mm_mask_move_sd
+  // CHECK: @llvm.x86.avx512.mask.move.sd
+  return _mm_mask_move_sd ( __W,  __U,  __A,  __B);
+}
+
+__m128d test_mm_maskz_move_sd (__mmask8 __U, __m128d __A, __m128d __B)
+{
+  // CHECK-LABEL: @test_mm_maskz_move_sd
+  // CHECK: @llvm.x86.avx512.mask.move.sd
+  return _mm_maskz_move_sd (__U, __A, __B);
+}