[X86][AVX512][Clang] Add support for mask_{move|store|load}_s{s/d} and int2mask/mask2int intrinsics.

[aymanmus]

Summary

Lower _mm_mask{|z}_move_{ss|sd} to generic llvm IR instead of x86 intrinsics.
Add support for missing intrinsics:

• _mm_mask{|z}_load_{ss|sd}
• _mm_mask_store_{ss|sd}
• _mm512_int2mask
• _mm512_mask2int

Clarification:

In the load and store intrinsics the vectors are widened to 512 vectors to stay in legal types for AVX512F (Otherwise illegal types are generated - e.g. v2i2).

• A llvm patch with the patterns for these intrinsics is attached as a son (D26022).

[aymanmus]

Fix spaces and indentations.

[igorb]

LGTM

lib/Headers/avx512fintrin.h

9123

.

9155

.

include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_fixupimmpd128_maskz, "V2dV2dV2dV2LLiIiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fixupimmpd256_mask, "V4dV4dV4dV4LLiIiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fixupimmpd256_maskz, "V4dV4dV4dV4LLiIiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fixupimmps128_mask, "V4fV4fV4fV4iIiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fixupimmps128_maskz, "V4fV4fV4fV4iIiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fixupimmps256_mask, "V8fV8fV8fV8iIiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_fixupimmps256_maskz, "V8fV8fV8fV8iIiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadapd128_mask, "V2dV2d*V2dUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_loadsd128_mask, "V8dV8d*V8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_loadapd256_mask, "V4dV4d*V4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadaps128_mask, "V4fV4f*V4fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_loadss128_mask, "V16fV16f*V16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_loadaps256_mask, "V8fV8f*V8fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loaddqudi128_mask, "V2LLiV2LLi*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loaddqudi256_mask, "V4LLiV4LLi*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loaddqusi128_mask, "V4iV4i*V4iUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_loaddqusi256_mask, "V8iV8i*V8iUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_loadupd128_mask, "V2dV2d*V2dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadupd256_mask, "V4dV4d*V4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadups128_mask, "V4fV4f*V4fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_loadups256_mask, "V8fV8f*V8fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storedquhi512_mask, "vV32s*V32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_storedquqi512_mask, "vV64c*V64cULLi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_storedquhi128_mask, "vV8s*V8sUc","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_storedquhi256_mask, "vV16s*V16sUs","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_storedquqi128_mask, "vV16c*V16cUs","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_storedquqi256_mask, "vV32c*V32cUi","","avx512vl,avx512bw")
 TARGET_BUILTIN(__builtin_ia32_storeapd128_mask, "vV2d*V2dUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storesd128_mask, "vV8d*V8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_storeapd256_mask, "vV4d*V4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storeaps128_mask, "vV4f*V4fUc","","avx512vl")
+TARGET_BUILTIN(__builtin_ia32_storess128_mask, "vV16f*V16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_storeaps256_mask, "vV8f*V8fUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storedqudi128_mask, "vV2LLi*V2LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storedqudi256_mask, "vV4LLi*V4LLiUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storedqusi128_mask, "vV4i*V4iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storedqusi256_mask, "vV8i*V8iUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storeupd128_mask, "vV2d*V2dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storeupd256_mask, "vV4d*V4dUc","","avx512vl")
 TARGET_BUILTIN(__builtin_ia32_storeups128_mask, "vV4f*V4fUc","","avx512vl")

 TARGET_BUILTIN(__builtin_ia32_expanddi512_mask, "V8LLiV8LLiV8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_expandloaddf512_mask, "V8dV8dC*V8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_expandloaddi512_mask, "V8LLiV8LLiC*V8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_expandloadsf512_mask, "V16fV16fC*V16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_expandloadsi512_mask, "V16iV16iC*V16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_expandsf512_mask, "V16fV16fV16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_expandsi512_mask, "V16iV16iV16iUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_cvtps2pd512_mask, "V8dV8fV8dUcIi","","avx512f")
-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/CodeGen/CGBuiltin.cpp

#undef INTRINSIC_X86_XSAVE_ID
   case X86::BI__builtin_ia32_storedqudi512_mask:
   case X86::BI__builtin_ia32_storedqusi512_mask:
   case X86::BI__builtin_ia32_storedquhi512_mask:
   case X86::BI__builtin_ia32_storedquqi512_mask:
   case X86::BI__builtin_ia32_storeupd512_mask:
   case X86::BI__builtin_ia32_storeups512_mask:
     return EmitX86MaskedStore(*this, Ops, 1);
 
+  case X86::BI__builtin_ia32_storess128_mask:
+  case X86::BI__builtin_ia32_storesd128_mask: {
+    return EmitX86MaskedStore(*this, Ops, 16);
+  }
   case X86::BI__builtin_ia32_movdqa32store128_mask:
   case X86::BI__builtin_ia32_movdqa64store128_mask:
   case X86::BI__builtin_ia32_storeaps128_mask:
   case X86::BI__builtin_ia32_storeapd128_mask:
   case X86::BI__builtin_ia32_movdqa32store256_mask:
   case X86::BI__builtin_ia32_movdqa64store256_mask:
   case X86::BI__builtin_ia32_storeaps256_mask:
   case X86::BI__builtin_ia32_storeapd256_mask:
#undef INTRINSIC_X86_XSAVE_ID
   case X86::BI__builtin_ia32_loaddqusi128_mask:
   case X86::BI__builtin_ia32_loaddqusi256_mask:
   case X86::BI__builtin_ia32_loaddqusi512_mask:
   case X86::BI__builtin_ia32_loaddqudi128_mask:
   case X86::BI__builtin_ia32_loaddqudi256_mask:
   case X86::BI__builtin_ia32_loaddqudi512_mask:
     return EmitX86MaskedLoad(*this, Ops, 1);
 
+  case X86::BI__builtin_ia32_loadss128_mask:
+  case X86::BI__builtin_ia32_loadsd128_mask:
+    return EmitX86MaskedLoad(*this, Ops, 16);
+
   case X86::BI__builtin_ia32_loadaps128_mask:
   case X86::BI__builtin_ia32_loadaps256_mask:
   case X86::BI__builtin_ia32_loadaps512_mask:
   case X86::BI__builtin_ia32_loadapd128_mask:
   case X86::BI__builtin_ia32_loadapd256_mask:
   case X86::BI__builtin_ia32_loadapd512_mask:
   case X86::BI__builtin_ia32_movdqa32load128_mask:
   case X86::BI__builtin_ia32_movdqa32load256_mask:

lib/Headers/avx512fintrin.h

 }
 
 static __inline __m256i __DEFAULT_FN_ATTRS
 _mm512_castsi512_si256 (__m512i __A)
 {
   return (__m256i)__builtin_shufflevector(__A, __A , 0, 1, 2, 3);
 }
 
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS
+_mm512_int2mask(int __a)
+{
+  return (__mmask16)__a;
+}
+
+static __inline__ int __DEFAULT_FN_ATTRS
+_mm512_mask2int(__mmask16 __a)
+{
+  return (int)__a;
+}
+
 /* Bitwise operators */
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_and_epi32(__m512i __a, __m512i __b)
 {
   return (__m512i)((__v16su)__a & (__v16su)__b);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_moveldup_ps (__mmask16 __U, __m512 __A)
   return (__m512)__builtin_ia32_selectps_512((__mmask16)__U,
                                              (__v16sf)_mm512_moveldup_ps(__A),
                                              (__v16sf)_mm512_setzero_ps());
 }
 
 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);
+  __m128 res = __A; 
+  res[0] = (__U & 1) ? __B[0] : __W[0];
+  return res; 
 }
 
 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);
+  __m128 res = __A; 
+  res[0] = (__U & 1) ? __B[0] : 0; 
+  return res; 
 }
 
 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);
+  __m128d res = __A; 
+  res[0] = (__U & 1) ? __B[0] : __W[0];
+  return res; 
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_maskz_move_sd (__mmask8 __U, __m128d __A, __m128d __B)
 {
-  return (__m128d) __builtin_ia32_movsd_mask ((__v2df) __A, (__v2df) __B,
-               (__v2df)
-               _mm_setzero_pd (),
-               (__mmask8) __U);
+  __m128d res = __A; 
+  res[0] = (__U & 1) ? __B[0] : 0; 
+  return res; 
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_store_ss (float * __W, __mmask8 __U, __m128 __A)
+{
+  __builtin_ia32_storess128_mask ((__v16sf *)__W, 
+                (__v16sf) _mm512_castps128_ps512(__A),
+                (__mmask16) __U & (__mmask16)1);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS
+_mm_mask_store_sd (double * __W, __mmask8 __U, __m128d __A)
+{
+  __builtin_ia32_storesd128_mask ((__v8df *)__W, 
+                (__v8df) _mm512_castpd128_pd512(__A),
+                (__mmask8) __U & 1);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_load_ss (__m128 __W, __mmask8 __U, const float* __A)
+{
+  __m128 src = (__v4sf) __builtin_shufflevector((__v4sf) __W,
+                                                (__v4sf) {0.0, 0.0, 0.0, 0.0},
+                                                0, 4, 4, 4);
+
+  return (__m128) __builtin_shufflevector(
+                           __builtin_ia32_loadss128_mask ((__v16sf *) __A,
+                                      (__v16sf) _mm512_castps128_ps512(src),
+                                      (__mmask16) __U & 1),
+                           _mm512_undefined_ps(), 0, 1, 2, 3);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_load_ss (__mmask8 __U, const float* __A)
+{
+  return (__m128) __builtin_shufflevector(
+                           __builtin_ia32_loadss128_mask ((__v16sf *) __A,
+                                      (__v16sf) _mm512_setzero_ps(),
+                                      (__mmask16) __U & 1),
+                           _mm512_undefined_ps(), 0, 1, 2, 3);
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_mask_load_sd (__m128d __W, __mmask8 __U, const double* __A)
+{
+  __m128d src = (__v2df) __builtin_shufflevector((__v2df) __W,
+                                                 (__v2df) {0.0, 0.0}, 0, 2);
+
+  return (__m128d) __builtin_shufflevector(
+                            __builtin_ia32_loadsd128_mask ((__v8df *) __A,
+                                      (__v8df) _mm512_castpd128_pd512(src),
+                                      (__mmask8) __U & 1),
+                            _mm512_undefined_pd(), 0, 1);
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_maskz_load_sd (__mmask8 __U, const double* __A)
+{
+  return (__m128d) __builtin_shufflevector(
+                            __builtin_ia32_loadsd128_mask ((__v8df *) __A,
+                                      (__v8df) _mm512_setzero_pd(),
+                                      (__mmask8) __U & 1),
+                            _mm512_undefined_pd(), 0, 1);
 }
 
 #define _mm512_shuffle_epi32(A, I) __extension__ ({ \
   (__m512i)__builtin_shufflevector((__v16si)(__m512i)(A), \
                                    (__v16si)_mm512_undefined_epi32(), \
                                    0  + (((I) >> 0) & 0x3), \
                                    0  + (((I) >> 2) & 0x3), \
                                    0  + (((I) >> 4) & 0x3), \

test/CodeGen/avx512f-builtins.c

 // RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512f -emit-llvm -o - -Wall -Werror | FileCheck %s
+// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512f -O2 -emit-llvm -o - -Wall -Werror | FileCheck %s -check-prefix=O2
 
 #include <immintrin.h>
 
 __m512d test_mm512_sqrt_pd(__m512d a)
 {
   // CHECK-LABEL: @test_mm512_sqrt_pd
   // CHECK: @llvm.x86.avx512.mask.sqrt.pd.512
   return _mm512_sqrt_pd(a);

 
 __m512d test_mm512_setzero_pd()
 {
   // CHECK-LABEL: @test_mm512_setzero_pd
   // CHECK: zeroinitializer
   return _mm512_setzero_pd();
 }
 
+__mmask16 test_mm512_int2mask(int __a)
+{
+  // O2-LABEL: test_mm512_int2mask
+  // O2: trunc i32 %__a to i16
+  return _mm512_int2mask(__a);
+}
+
+int test_mm512_mask2int(__mmask16 __a)
+{
+  // O2-LABEL: test_mm512_mask2int
+  // O2: zext i16 %__a to i32
+  return _mm512_mask2int(__a);
+}
+
 __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
+  // O2-LABEL: @test_mm_mask_move_ss
+  // O2: %[[M:.*]] = and i8 %__U, 1
+  // O2: %[[M2:.*]] = icmp ne i8 %[[M]], 0
+  // O2: %[[ELM1:.*]] = extractelement <4 x float> %__B, i32 0
+  // O2: %[[ELM2:.*]] = extractelement <4 x float> %__W, i32 0
+  // O2: %[[SEL:.*]] = select i1 %[[M2]], float %[[ELM1]], float %[[ELM2]]
+  // O2: %[[RES:.*]] = insertelement <4 x float> %__A, float %[[SEL]], i32 0
+  // O2: ret <4 x float> %[[RES]]
   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
+  // O2-LABEL: @test_mm_maskz_move_ss
+  // O2: %[[M:.*]] = and i8 %__U, 1
+  // O2: %[[M2:.*]] = icmp ne i8 %[[M]], 0
+  // O2: %[[ELM1:.*]] = extractelement <4 x float> %__B, i32 0
+  // O2: %[[SEL:.*]] = select i1 %[[M2]], float %[[ELM1]], float 0.0 
+  // O2: %[[RES:.*]] = insertelement <4 x float> %__A, float %[[SEL]], i32 0
+  // O2: ret <4 x float> %[[RES]]
   return _mm_maskz_move_ss (__U, __A, __B);
 }
 
-__m128d test_mm_mask_move_sd (__m128 __W, __mmask8 __U, __m128d __A, __m128d __B)
+__m128d test_mm_mask_move_sd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
 {
-  // CHECK-LABEL: @test_mm_mask_move_sd
-  // CHECK: @llvm.x86.avx512.mask.move.sd
+  // O2-LABEL: @test_mm_mask_move_sd
+  // O2: %[[M:.*]] = and i8 %__U, 1
+  // O2: %[[M2:.*]] = icmp ne i8 %[[M]], 0
+  // O2: %[[ELM1:.*]] = extractelement <2 x double> %__B, i32 0
+  // O2: %[[ELM2:.*]] = extractelement <2 x double> %__W, i32 0
+  // O2: %[[SEL:.*]] = select i1 %[[M2]], double %[[ELM1]], double %[[ELM2]]
+  // O2: %[[RES:.*]] = insertelement <2 x double> %__A, double %[[SEL]], i32 0
+  // O2: ret <2 x double> %[[RES]]
   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
+  // O2-LABEL: @test_mm_maskz_move_sd
+  // O2: %[[M:.*]] = and i8 %__U, 1
+  // O2: %[[M2:.*]] = icmp ne i8 %[[M]], 0
+  // O2: %[[ELM1:.*]] = extractelement <2 x double> %__B, i32 0
+  // O2: %[[SEL:.*]] = select i1 %[[M2]], double %[[ELM1]], double 0.0
+  // O2: %[[RES:.*]] = insertelement <2 x double> %__A, double %[[SEL]], i32 0
+  // O2: ret <2 x double> %[[RES]]
   return _mm_maskz_move_sd (__U, __A, __B);
 }
 
+void test_mm_mask_store_ss(float * __P, __mmask8 __U, __m128 __A)
+{
+  // O2-LABEL: @test_mm_mask_store_ss
+  // O2: %[[CAST:.*]] = bitcast float* %__P to <16 x float>*
+  // O2: %[[SHUFFLE:.*]] = shufflevector <4 x float> %__A, <4 x float> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  // O2: %[[MASK1:.*]] = and i8 %__U, 1
+  // O2: %[[MASK2:.*]] = zext i8 %[[MASK1]] to i16
+  // O2: %[[MASK3:.*]] = bitcast i16 %[[MASK2]] to <16 x i1>
+  // O2: tail call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> %[[SHUFFLE]], <16 x float>* %[[CAST]], i32 16, <16 x i1> %[[MASK3]])
+  _mm_mask_store_ss(__P, __U, __A);
+}
+
+void test_mm_mask_store_sd(double * __P, __mmask8 __U, __m128d __A)
+{
+  // O2-LABEL: @test_mm_mask_store_sd
+  // O2: %[[CAST:.*]] = bitcast double* %__P to <8 x double>*
+  // O2: %[[SHUFFLE:.*]] = shufflevector <2 x double> %__A, <2 x double> undef, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  // O2: %[[MASK1:.*]] = and i8 %__U, 1
+  // O2: %[[MASK2:.*]] = bitcast i8 %[[MASK1]] to <8 x i1>
+  // O2: tail call void @llvm.masked.store.v8f64.p0v8f64(<8 x double> %[[SHUFFLE]], <8 x double>* %[[CAST]], i32 16, <8 x i1> %[[MASK2]])
+  _mm_mask_store_sd(__P, __U, __A);
+}
+
+__m128 test_mm_mask_load_ss(__m128 __A, __mmask8 __U, const float* __W)
+{
+  // O2-LABEL: @test_mm_mask_load_ss
+  // O2: %[[SHUF:.*]] = shufflevector <4 x float> %__A, <4 x float> <float 0.000000e+00, float undef, float undef, float undef>, <4 x i32> <i32 0, i32 4, i32 4, i32 4>
+  // O2: %[[PTR:.*]] = bitcast float* %__W to <16 x float>*
+  // O2: %[[SHUF2:.*]] = shufflevector <4 x float> %[[SHUF]], <4 x float> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  // O2: %[[AND:.*]] = and i8 %__U, 1
+  // O2: %[[MASK:.*]] = zext i8 %[[AND]] to i16
+  // O2: %[[MASK2:.*]] = bitcast i16 %[[MASK]] to <16 x i1>
+  // O2: %[[RES:.*]] = tail call <16 x float> @llvm.masked.load.v16f32.p0v16f32(<16 x float>* %[[PTR]], i32 16, <16 x i1> %[[MASK2]], <16 x float> %[[SHUF2]]) 
+  // O2: shufflevector <16 x float> %[[RES]], <16 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  return _mm_mask_load_ss(__A, __U, __W);
+}
+
+__m128 test_mm_maskz_load_ss (__mmask8 __U, const float * __W)
+{
+  // O2-LABEL: @test_mm_maskz_load_ss
+  // O2: %[[PTR:.*]] = bitcast float* %__W to <16 x float>*
+  // O2: %[[AND:.*]] = and i8 %__U, 1
+  // O2: %[[MASK:.*]] = zext i8 %[[AND]] to i16
+  // O2: %[[MASK2:.*]] = bitcast i16 %[[MASK]] to <16 x i1>
+  // O2: %[[RES:.*]] = tail call <16 x float> @llvm.masked.load.v16f32.p0v16f32(<16 x float>* %[[PTR]], i32 16, <16 x i1> %[[MASK2]], <16 x float> zeroinitializer) 
+  // O2: shufflevector <16 x float> %[[RES]], <16 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  return _mm_maskz_load_ss (__U, __W);
+}
+
+__m128d test_mm_mask_load_sd (__m128d __A, __mmask8 __U, const double * __W)
+{
+  // O2-LABEL: @test_mm_mask_load_sd
+  // O2: %[[SHUF:.*]] = insertelement <2 x double> %__A, double 0.000000e+00, i32 1
+  // O2: %[[PTR:.*]] = bitcast double* %__W to <8 x double>*
+  // O2: %[[SHUF2:.*]] = shufflevector <2 x double> %[[SHUF]], <2 x double> undef, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+  // O2: %[[AND:.*]] = and i8 %__U, 1
+  // O2: %[[MASK:.*]] = bitcast i8 %[[AND]] to <8 x i1>
+  // O2: %[[RES:.*]] = tail call <8 x double> @llvm.masked.load.v8f64.p0v8f64(<8 x double>* %[[PTR]], i32 16, <8 x i1> %[[MASK]], <8 x double> %[[SHUF2]]) 
+  // O2: shufflevector <8 x double> %[[RES]], <8 x double> undef, <2 x i32> <i32 0, i32 1>
+  return _mm_mask_load_sd (__A, __U, __W);
+}
+
+__m128d test_mm_maskz_load_sd (__mmask8 __U, const double * __W)
+{
+  // O2-LABEL: @test_mm_maskz_load_sd
+  // O2: %[[PTR:.*]] = bitcast double* %__W to <8 x double>*
+  // O2: %[[AND:.*]] = and i8 %__U, 1
+  // O2: %[[MASK:.*]] = bitcast i8 %[[AND]] to <8 x i1>
+  // O2: %[[RES:.*]] = tail call <8 x double> @llvm.masked.load.v8f64.p0v8f64(<8 x double>* %[[PTR]], i32 16, <8 x i1> %[[MASK]], <8 x double> zeroinitializer) 
+  // O2: shufflevector <8 x double> %[[RES]], <8 x double> undef, <2 x i32> <i32 0, i32 1>
+  return _mm_maskz_load_sd (__U, __W);
+}
+
 __m512d test_mm512_abs_pd(__m512d a){
   // CHECK-LABEL: @test_mm512_abs_pd
   // CHECK: and <8 x i64> 
   return _mm512_abs_pd(a);
 }
 
 __m512d test_mm512_mask_abs_pd (__m512d __W, __mmask8 __U, __m512d __A){
   // CHECK-LABEL: @test_mm512_mask_abs_pd