[X86] Add __builtin_ia32_selectss_128 and __builtin_ia32_selectsd_128 that is suitabl...

This will convert the i8 mask argument to <8 x i1> and extract an i1 and then emit a select instruction. This replaces the '(__U & 1)" and ternary operator used in some of intrinsics. The old sequence was lowered to a scalar and and compare. The new sequence uses an i1 vector that will interoperate better with other mask intrinsics.

This removes the need to handle div_ss/sd specially in CGBuiltin.cpp. A follow up patch will add the GCCBuiltin name back in llvm and remove the custom handling.

I made some adjustments to legacy move_ss/sd intrinsics which we reused here to do a simpler extract and insert instead of 2 extracts and two inserts or a shuffle.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@336622 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index 0b08e03764933ea2c412c7eed2bf4405ca46bf45..e98f7d612c3c62c7f65eb892299f64624c422fe9 100644 (file)
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -1809,6 +1809,8 @@ TARGET_BUILTIN(__builtin_ia32_selectps_512, "V16fUsV16fV16f", "ncV:512:", "avx51
 TARGET_BUILTIN(__builtin_ia32_selectpd_128, "V2dUcV2dV2d", "ncV:128:", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_selectpd_256, "V4dUcV4dV4d", "ncV:256:", "avx512vl")
 TARGET_BUILTIN(__builtin_ia32_selectpd_512, "V8dUcV8dV8d", "ncV:512:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_selectss_128, "V4fUcV4fV4f", "ncV:128:", "avx512f")
+TARGET_BUILTIN(__builtin_ia32_selectsd_128, "V2dUcV2dV2d", "ncV:128:", "avx512f")
 
 // MONITORX/MWAITX
 TARGET_BUILTIN(__builtin_ia32_monitorx, "vv*UiUi", "n", "mwaitx")

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 6b1198d64e4531317e244cfb95b17e52b35abb26..ba0519b35503f822849131c7f2cdf3921f6100ce 100644 (file)
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -9832,6 +9832,13 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
   case X86::BI__builtin_ia32_selectpd_256:
   case X86::BI__builtin_ia32_selectpd_512:
     return EmitX86Select(*this, Ops[0], Ops[1], Ops[2]);
+  case X86::BI__builtin_ia32_selectss_128:
+  case X86::BI__builtin_ia32_selectsd_128: {
+    Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0);
+    Value *B = Builder.CreateExtractElement(Ops[2], (uint64_t)0);
+    A = EmitX86ScalarSelect(*this, Ops[0], A, B);
+    return Builder.CreateInsertElement(Ops[1], A, (uint64_t)0);
+  }
   case X86::BI__builtin_ia32_cmpb128_mask:
   case X86::BI__builtin_ia32_cmpb256_mask:
   case X86::BI__builtin_ia32_cmpb512_mask:

diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h
index 9024168483600a4b19643597c850dcc74a3dc295..f69620bbcabe81ed41d3022717e80406513c730c 100644 (file)
--- a/lib/Headers/avx512fintrin.h
+++ b/lib/Headers/avx512fintrin.h
@@ -1899,15 +1899,13 @@ _mm512_maskz_abs_epi32 (__mmask16 __U, __m512i __A)
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_mask_add_ss(__m128 __W, __mmask8 __U,__m128 __A, __m128 __B) {
   __A = _mm_add_ss(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : __W[0];
-  return __A;
+  return __builtin_ia32_selectss_128(__U, __A, __W);
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_maskz_add_ss(__mmask8 __U,__m128 __A, __m128 __B) {
   __A = _mm_add_ss(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : 0;
-  return __A;
+  return __builtin_ia32_selectss_128(__U, __A, _mm_setzero_ps());
 }
 
 #define _mm_add_round_ss(A, B, R) \
@@ -1931,15 +1929,13 @@ _mm_maskz_add_ss(__mmask8 __U,__m128 __A, __m128 __B) {
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_mask_add_sd(__m128d __W, __mmask8 __U,__m128d __A, __m128d __B) {
   __A = _mm_add_sd(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : __W[0];
-  return __A;
+  return __builtin_ia32_selectsd_128(__U, __A, __W);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_maskz_add_sd(__mmask8 __U,__m128d __A, __m128d __B) {
   __A = _mm_add_sd(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : 0;
-  return __A;
+  return __builtin_ia32_selectsd_128(__U, __A, _mm_setzero_pd());
 }
 #define _mm_add_round_sd(A, B, R) \
   (__m128d)__builtin_ia32_addsd_round_mask((__v2df)(__m128d)(A), \
@@ -2018,15 +2014,13 @@ _mm512_maskz_add_ps(__mmask16 __U, __m512 __A, __m512 __B) {
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_mask_sub_ss(__m128 __W, __mmask8 __U,__m128 __A, __m128 __B) {
   __A = _mm_sub_ss(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : __W[0];
-  return __A;
+  return __builtin_ia32_selectss_128(__U, __A, __W);
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_maskz_sub_ss(__mmask8 __U,__m128 __A, __m128 __B) {
   __A = _mm_sub_ss(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : 0;
-  return __A;
+  return __builtin_ia32_selectss_128(__U, __A, _mm_setzero_ps());
 }
 #define _mm_sub_round_ss(A, B, R) \
   (__m128)__builtin_ia32_subss_round_mask((__v4sf)(__m128)(A), \
@@ -2049,15 +2043,13 @@ _mm_maskz_sub_ss(__mmask8 __U,__m128 __A, __m128 __B) {
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_mask_sub_sd(__m128d __W, __mmask8 __U,__m128d __A, __m128d __B) {
   __A = _mm_sub_sd(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : __W[0];
-  return __A;
+  return __builtin_ia32_selectsd_128(__U, __A, __W);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_maskz_sub_sd(__mmask8 __U,__m128d __A, __m128d __B) {
   __A = _mm_sub_sd(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : 0;
-  return __A;
+  return __builtin_ia32_selectsd_128(__U, __A, _mm_setzero_pd());
 }
 
 #define _mm_sub_round_sd(A, B, R) \
@@ -2137,15 +2129,13 @@ _mm512_maskz_sub_ps(__mmask16 __U, __m512 __A, __m512 __B) {
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_mask_mul_ss(__m128 __W, __mmask8 __U,__m128 __A, __m128 __B) {
   __A = _mm_mul_ss(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : __W[0];
-  return __A;
+  return __builtin_ia32_selectss_128(__U, __A, __W);
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_maskz_mul_ss(__mmask8 __U,__m128 __A, __m128 __B) {
   __A = _mm_mul_ss(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : 0;
-  return __A;
+  return __builtin_ia32_selectss_128(__U, __A, _mm_setzero_ps());
 }
 #define _mm_mul_round_ss(A, B, R) \
   (__m128)__builtin_ia32_mulss_round_mask((__v4sf)(__m128)(A), \
@@ -2168,15 +2158,13 @@ _mm_maskz_mul_ss(__mmask8 __U,__m128 __A, __m128 __B) {
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_mask_mul_sd(__m128d __W, __mmask8 __U,__m128d __A, __m128d __B) {
   __A = _mm_mul_sd(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : __W[0];
-  return __A;
+  return __builtin_ia32_selectsd_128(__U, __A, __W);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_maskz_mul_sd(__mmask8 __U,__m128d __A, __m128d __B) {
   __A = _mm_mul_sd(__A, __B);
-  __A[0] = (__U & 1) ? __A[0] : 0;
-  return __A;
+  return __builtin_ia32_selectsd_128(__U, __A, _mm_setzero_pd());
 }
 
 #define _mm_mul_round_sd(A, B, R) \
@@ -2255,20 +2243,14 @@ _mm512_maskz_mul_ps(__mmask16 __U, __m512 __A, __m512 __B) {
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_mask_div_ss(__m128 __W, __mmask8 __U,__m128 __A, __m128 __B) {
-  return (__m128) __builtin_ia32_divss_round_mask ((__v4sf) __A,
-                (__v4sf) __B,
-                (__v4sf) __W,
-                (__mmask8) __U,
-                _MM_FROUND_CUR_DIRECTION);
+  __A = _mm_div_ss(__A, __B);
+  return __builtin_ia32_selectss_128(__U, __A, __W);
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_maskz_div_ss(__mmask8 __U,__m128 __A, __m128 __B) {
-  return (__m128) __builtin_ia32_divss_round_mask ((__v4sf) __A,
-                (__v4sf) __B,
-                (__v4sf)  _mm_setzero_ps (),
-                (__mmask8) __U,
-                _MM_FROUND_CUR_DIRECTION);
+  __A = _mm_div_ss(__A, __B);
+  return __builtin_ia32_selectss_128(__U, __A, _mm_setzero_ps());
 }
 
 #define _mm_div_round_ss(A, B, R) \
@@ -2291,20 +2273,14 @@ _mm_maskz_div_ss(__mmask8 __U,__m128 __A, __m128 __B) {
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_mask_div_sd(__m128d __W, __mmask8 __U,__m128d __A, __m128d __B) {
-  return (__m128d) __builtin_ia32_divsd_round_mask ((__v2df) __A,
-                (__v2df) __B,
-                (__v2df) __W,
-                (__mmask8) __U,
-                _MM_FROUND_CUR_DIRECTION);
+  __A = _mm_div_sd(__A, __B);
+  return __builtin_ia32_selectsd_128(__U, __A, __W);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_maskz_div_sd(__mmask8 __U,__m128d __A, __m128d __B) {
-  return (__m128d) __builtin_ia32_divsd_round_mask ((__v2df) __A,
-                (__v2df) __B,
-                (__v2df)  _mm_setzero_pd (),
-                (__mmask8) __U,
-                _MM_FROUND_CUR_DIRECTION);
+  __A = _mm_div_sd(__A, __B);
+  return __builtin_ia32_selectsd_128(__U, __A, _mm_setzero_pd());
 }
 
 #define _mm_div_round_sd(A, B, R) \
@@ -8639,33 +8615,27 @@ _mm512_maskz_moveldup_ps (__mmask16 __U, __m512 __A)
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_mask_move_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
 {
-  __m128 res = __A;
-  res[0] = (__U & 1) ? __B[0] : __W[0];
-  return res;
+  return __builtin_ia32_selectss_128(__U, _mm_move_ss(__A, __B), __W);
 }
 
 static __inline__ __m128 __DEFAULT_FN_ATTRS128
 _mm_maskz_move_ss (__mmask8 __U, __m128 __A, __m128 __B)
 {
-  __m128 res = __A;
-  res[0] = (__U & 1) ? __B[0] : 0;
-  return res;
+  return __builtin_ia32_selectss_128(__U, _mm_move_ss(__A, __B),
+                                     _mm_setzero_ps());
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_mask_move_sd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
 {
-  __m128d res = __A;
-  res[0] = (__U & 1) ? __B[0] : __W[0];
-  return res;
+  return __builtin_ia32_selectsd_128(__U, _mm_move_sd(__A, __B), __W);
 }
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS128
 _mm_maskz_move_sd (__mmask8 __U, __m128d __A, __m128d __B)
 {
-  __m128d res = __A;
-  res[0] = (__U & 1) ? __B[0] : 0;
-  return res;
+  return __builtin_ia32_selectsd_128(__U, _mm_move_sd(__A, __B),
+                                     _mm_setzero_pd());
 }
 
 static __inline__ void __DEFAULT_FN_ATTRS128

diff --git a/lib/Headers/emmintrin.h b/lib/Headers/emmintrin.h
index be88c65843b50abbf3a4ead80305370c4964f65c..f0ea7cd05c63ca3e726ef278b02d3666415b2973 100644 (file)
--- a/lib/Headers/emmintrin.h
+++ b/lib/Headers/emmintrin.h
@@ -1900,7 +1900,8 @@ _mm_setzero_pd(void)
 static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_move_sd(__m128d __a, __m128d __b)
 {
-  return __extension__ (__m128d){ __b[0], __a[1] };
+  __a[0] = __b[0];
+  return __a;
 }
 
 /// Stores the lower 64 bits of a 128-bit vector of [2 x double] to a

diff --git a/lib/Headers/xmmintrin.h b/lib/Headers/xmmintrin.h
index 0bc765073bfb03442bb870684264fad47725785e..17af17267c83784fe8195f9a87cd75ed6ade6e64 100644 (file)
--- a/lib/Headers/xmmintrin.h
+++ b/lib/Headers/xmmintrin.h
@@ -2671,7 +2671,8 @@ _mm_unpacklo_ps(__m128 __a, __m128 __b)
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_move_ss(__m128 __a, __m128 __b)
 {
-  return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 4, 1, 2, 3);
+  __a[0] = __b[0];
+  return __a;
 }
 
 /// Constructs a 128-bit floating-point vector of [4 x float]. The lower

diff --git a/test/CodeGen/avx512f-builtins.c b/test/CodeGen/avx512f-builtins.c
index be5fcb207c595fa764eac52b601c485b11e78277..2beae24e45ee168e4f8b7a85a9a86a46cfa024dc 100644 (file)
--- a/test/CodeGen/avx512f-builtins.c
+++ b/test/CodeGen/avx512f-builtins.c
@@ -3150,11 +3150,12 @@ __m128 test_mm_mask_add_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
   // CHECK: extractelement <4 x float> %{{.*}}, i32 0
   // CHECK: fadd float %{{.*}}, %{{.*}}
   // CHECK: insertelement <4 x float> %{{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
-  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, float %{{.*}}, float %{{.*}}
+  // CHECK-NEXT: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
   return _mm_mask_add_ss(__W,__U,__A,__B); 
 }
 __m128 test_mm_maskz_add_ss(__mmask8 __U, __m128 __A, __m128 __B) {
@@ -3164,10 +3165,12 @@ __m128 test_mm_maskz_add_ss(__mmask8 __U, __m128 __A, __m128 __B) {
   // CHECK: extractelement <4 x float> %{{.*}}, i32 0
   // CHECK: fadd float %{{.*}}, %{{.*}}
   // CHECK: insertelement <4 x float> %{{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, float %{{.*}}, float %{{.*}}
+  // CHECK-NEXT: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
   return _mm_maskz_add_ss(__U,__A,__B); 
 }
 __m128d test_mm_add_round_sd(__m128d __A, __m128d __B) {
@@ -3192,11 +3195,12 @@ __m128d test_mm_mask_add_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
   // CHECK: extractelement <2 x double> %{{.*}}, i32 0
   // CHECK: fadd double %{{.*}}, %{{.*}}
   // CHECK: insertelement <2 x double> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: extractelement <2 x double> %{{.*}}, i32 0
-  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, double %{{.*}}, double %{{.*}}
+  // CHECK-NEXT: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
   return _mm_mask_add_sd(__W,__U,__A,__B); 
 }
 __m128d test_mm_maskz_add_sd(__mmask8 __U, __m128d __A, __m128d __B) {
@@ -3206,10 +3210,12 @@ __m128d test_mm_maskz_add_sd(__mmask8 __U, __m128d __A, __m128d __B) {
   // CHECK: extractelement <2 x double> %{{.*}}, i32 0
   // CHECK: fadd double %{{.*}}, %{{.*}}
   // CHECK: insertelement <2 x double> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, double %{{.*}}, double %{{.*}}
+  // CHECK-NEXT: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
   return _mm_maskz_add_sd(__U,__A,__B); 
 }
 __m512d test_mm512_sub_round_pd(__m512d __A, __m512d __B) {
@@ -3292,11 +3298,12 @@ __m128 test_mm_mask_sub_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
   // CHECK: extractelement <4 x float> %{{.*}}, i32 0
   // CHECK: fsub float %{{.*}}, %{{.*}}
   // CHECK: insertelement <4 x float> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
-  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, float %{{.*}}, float %{{.*}}
+  // CHECK-NEXT: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
   return _mm_mask_sub_ss(__W,__U,__A,__B); 
 }
 __m128 test_mm_maskz_sub_ss(__mmask8 __U, __m128 __A, __m128 __B) {
@@ -3306,10 +3313,12 @@ __m128 test_mm_maskz_sub_ss(__mmask8 __U, __m128 __A, __m128 __B) {
   // CHECK: extractelement <4 x float> %{{.*}}, i32 0
   // CHECK: fsub float %{{.*}}, %{{.*}}
   // CHECK: insertelement <4 x float> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, float %{{.*}}, float %{{.*}}
+  // CHECK-NEXT: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
   return _mm_maskz_sub_ss(__U,__A,__B); 
 }
 __m128d test_mm_sub_round_sd(__m128d __A, __m128d __B) {
@@ -3334,11 +3343,12 @@ __m128d test_mm_mask_sub_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
   // CHECK: extractelement <2 x double> %{{.*}}, i32 0
   // CHECK: fsub double %{{.*}}, %{{.*}}
   // CHECK: insertelement <2 x double> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: extractelement <2 x double> %{{.*}}, i32 0
-  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, double %{{.*}}, double %{{.*}}
+  // CHECK-NEXT: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
   return _mm_mask_sub_sd(__W,__U,__A,__B); 
 }
 __m128d test_mm_maskz_sub_sd(__mmask8 __U, __m128d __A, __m128d __B) {
@@ -3348,10 +3358,12 @@ __m128d test_mm_maskz_sub_sd(__mmask8 __U, __m128d __A, __m128d __B) {
   // CHECK: extractelement <2 x double> %{{.*}}, i32 0
   // CHECK: fsub double %{{.*}}, %{{.*}}
   // CHECK: insertelement <2 x double> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, double %{{.*}}, double %{{.*}}
+  // CHECK-NEXT: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
   return _mm_maskz_sub_sd(__U,__A,__B); 
 }
 __m512d test_mm512_mul_round_pd(__m512d __A, __m512d __B) {
@@ -3434,11 +3446,12 @@ __m128 test_mm_mask_mul_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
   // CHECK: extractelement <4 x float> %{{.*}}, i32 0
   // CHECK: fmul float %{{.*}}, %{{.*}}
   // CHECK: insertelement <4 x float> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
-  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, float %{{.*}}, float %{{.*}}
+  // CHECK-NEXT: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
   return _mm_mask_mul_ss(__W,__U,__A,__B); 
 }
 __m128 test_mm_maskz_mul_ss(__mmask8 __U, __m128 __A, __m128 __B) {
@@ -3448,10 +3461,12 @@ __m128 test_mm_maskz_mul_ss(__mmask8 __U, __m128 __A, __m128 __B) {
   // CHECK: extractelement <4 x float> %{{.*}}, i32 0
   // CHECK: fmul float %{{.*}}, %{{.*}}
   // CHECK: insertelement <4 x float> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, float %{{.*}}, float %{{.*}}
+  // CHECK-NEXT: insertelement <4 x float> %{{.*}}, float %{{.*}}, i64 0
   return _mm_maskz_mul_ss(__U,__A,__B); 
 }
 __m128d test_mm_mul_round_sd(__m128d __A, __m128d __B) {
@@ -3476,11 +3491,12 @@ __m128d test_mm_mask_mul_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
   // CHECK: extractelement <2 x double> %{{.*}}, i32 0
   // CHECK: fmul double %{{.*}}, %{{.*}}
   // CHECK: insertelement <2 x double> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: extractelement <2 x double> %{{.*}}, i32 0
-  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, double %{{.*}}, double %{{.*}}
+  // CHECK-NEXT: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
   return _mm_mask_mul_sd(__W,__U,__A,__B); 
 }
 __m128d test_mm_maskz_mul_sd(__mmask8 __U, __m128d __A, __m128d __B) {
@@ -3490,10 +3506,12 @@ __m128d test_mm_maskz_mul_sd(__mmask8 __U, __m128d __A, __m128d __B) {
   // CHECK: extractelement <2 x double> %{{.*}}, i32 0
   // CHECK: fmul double %{{.*}}, %{{.*}}
   // CHECK: insertelement <2 x double> {{.*}}, i32 0
-  // CHECK: and i32 {{.*}}, 1
-  // CHECK: icmp ne i32 %{{.*}}, 0
-  // CHECK: br {{.*}}, {{.*}}, {{.*}}
-  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: select i1 %{{.*}}, double %{{.*}}, double %{{.*}}
+  // CHECK-NEXT: insertelement <2 x double> %{{.*}}, double %{{.*}}, i64 0
   return _mm_maskz_mul_sd(__U,__A,__B); 
 }
 __m512d test_mm512_div_round_pd(__m512d __A, __m512d __B) {
@@ -3581,10 +3599,12 @@ __m128 test_mm_maskz_div_round_ss(__mmask8 __U, __m128 __A, __m128 __B) {
 }
 __m128 test_mm_mask_div_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
   // CHECK-LABEL: @test_mm_mask_div_ss
+  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
+  // CHECK: fdiv float %{{.*}}, %{{.*}}
+  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
   // CHECK: extractelement <4 x float> %{{.*}}, i64 0
   // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
-  // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
-  // CHECK-NEXT: fdiv float %{{.*}}, %{{.*}}
   // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
   // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
   // CHECK-NEXT: select i1 %{{.*}}, float %{{.*}}, float %{{.*}}
@@ -3593,10 +3613,12 @@ __m128 test_mm_mask_div_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
 }
 __m128 test_mm_maskz_div_ss(__mmask8 __U, __m128 __A, __m128 __B) {
   // CHECK-LABEL: @test_mm_maskz_div_ss
+  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
+  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
+  // CHECK: fdiv float %{{.*}}, %{{.*}}
+  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
   // CHECK: extractelement <4 x float> %{{.*}}, i64 0
   // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
-  // CHECK-NEXT: extractelement <4 x float> %{{.*}}, i64 0
-  // CHECK-NEXT: fdiv float %{{.*}}, %{{.*}}
   // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
   // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
   // CHECK-NEXT: select i1 %{{.*}}, float %{{.*}}, float %{{.*}}
@@ -3620,10 +3642,12 @@ __m128d test_mm_maskz_div_round_sd(__mmask8 __U, __m128d __A, __m128d __B) {
 }
 __m128d test_mm_mask_div_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) {
   // CHECK-LABEL: @test_mm_mask_div_sd
+  // CHECK: extractelement <2 x double> %{{.*}}, i32 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i32 0
+  // CHECK: fdiv double %{{.*}}, %{{.*}}
+  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
   // CHECK: extractelement <2 x double> %{{.*}}, i64 0
   // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
-  // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
-  // CHECK-NEXT: fdiv double %{{.*}}, %{{.*}}
   // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
   // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
   // CHECK-NEXT: select i1 %{{.*}}, double %{{.*}}, double %{{.*}}
@@ -3632,10 +3656,12 @@ __m128d test_mm_mask_div_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
 }
 __m128d test_mm_maskz_div_sd(__mmask8 __U, __m128d __A, __m128d __B) {
   // CHECK-LABEL: @test_mm_maskz_div_sd
+  // CHECK: extractelement <2 x double> %{{.*}}, i32 0
+  // CHECK: extractelement <2 x double> %{{.*}}, i32 0
+  // CHECK: fdiv double %{{.*}}, %{{.*}}
+  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
   // CHECK: extractelement <2 x double> %{{.*}}, i64 0
   // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
-  // CHECK-NEXT: extractelement <2 x double> %{{.*}}, i64 0
-  // CHECK-NEXT: fdiv double %{{.*}}, %{{.*}}
   // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
   // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
   // CHECK-NEXT: select i1 %{{.*}}, double %{{.*}}, double %{{.*}}
@@ -10531,38 +10557,56 @@ int test_mm512_mask2int(__mmask16 __a)
 __m128 test_mm_mask_move_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
 {
   // CHECK-LABEL: @test_mm_mask_move_ss
-  // CHECK:  extractelement <4 x float> %{{.*}}, i32 0
-  // CHECK:  extractelement <4 x float> %{{.*}}, i32 0
-  // CHECK:  phi float [ %{{.*}}, %{{.*}} ], [ %{{.*}}, %{{.*}} ]
-  // CHECK:  insertelement <4 x float> %{{.*}}, float %cond.i, i32 0
+  // CHECK: [[EXT:%.*]] = extractelement <4 x float> %{{.*}}, i32 0
+  // CHECK: insertelement <4 x float> %{{.*}}, float [[EXT]], i32 0
+  // CHECK: [[A:%.*]] = extractelement <4 x float> [[VEC:%.*]], i64 0
+  // CHECK-NEXT: [[B:%.*]] = extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: [[SEL:%.*]] = select i1 %{{.*}}, float [[A]], float [[B]]
+  // CHECK-NEXT: insertelement <4 x float> [[VEC]], float [[SEL]], i64 0
   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:  extractelement <4 x float> %{{.*}}, i32 0
-  // CHECK:  phi float [ %{{.*}}, %{{.*}} ], [ 0.000000e+00, %{{.*}} ]
-  // CHECK:  insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  // CHECK: [[EXT:%.*]] = extractelement <4 x float> %{{.*}}, i32 0
+  // CHECK: insertelement <4 x float> %{{.*}}, float [[EXT]], i32 0
+  // CHECK: [[A:%.*]] = extractelement <4 x float> [[VEC:%.*]], i64 0
+  // CHECK-NEXT: [[B:%.*]] = extractelement <4 x float> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: [[SEL:%.*]] = select i1 %{{.*}}, float [[A]], float [[B]]
+  // CHECK-NEXT: insertelement <4 x float> [[VEC]], float [[SEL]], i64 0
   return _mm_maskz_move_ss (__U, __A, __B);
 }
 
 __m128d test_mm_mask_move_sd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B)
 {
   // CHECK-LABEL: @test_mm_mask_move_sd
-  // CHECK:  extractelement <2 x double> %{{.*}}, i32 0
-  // CHECK:  extractelement <2 x double> %{{.*}}, i32 0
-  // CHECK:  phi double [ %{{.*}}, %{{.*}} ], [ %{{.*}}, %{{.*}} ]
-  // CHECK:  insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  // CHECK: [[EXT:%.*]] = extractelement <2 x double> %{{.*}}, i32 0
+  // CHECK: insertelement <2 x double> %{{.*}}, double [[EXT]], i32 0
+  // CHECK: [[A:%.*]] = extractelement <2 x double> [[VEC:%.*]], i64 0
+  // CHECK-NEXT: [[B:%.*]] = extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: [[SEL:%.*]] = select i1 %{{.*}}, double [[A]], double [[B]]
+  // CHECK-NEXT: insertelement <2 x double> [[VEC]], double [[SEL]], i64 0
   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:  extractelement <2 x double> %{{.*}}, i32 0
-  // CHECK:  phi double [ %{{.*}}, %{{.*}} ], [ 0.000000e+00, %{{.*}} ]
-  // CHECK:  insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  // CHECK: [[EXT:%.*]] = extractelement <2 x double> %{{.*}}, i32 0
+  // CHECK: insertelement <2 x double> %{{.*}}, double [[EXT]], i32 0
+  // CHECK: [[A:%.*]] = extractelement <2 x double> [[VEC:%.*]], i64 0
+  // CHECK-NEXT: [[B:%.*]] = extractelement <2 x double> %{{.*}}, i64 0
+  // CHECK-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // CHECK-NEXT: [[SEL:%.*]] = select i1 %13, double [[A]], double [[B]]
+  // CHECK-NEXT: insertelement <2 x double> [[VEC]], double [[SEL]], i64 0
   return _mm_maskz_move_sd (__U, __A, __B);
 }
 

diff --git a/test/CodeGen/sse-builtins.c b/test/CodeGen/sse-builtins.c
index b7c7a7fc7e497015a2e0139bd3d8b874051066eb..e9801487be0dfd0158b312244383cb312029645d 100644 (file)
--- a/test/CodeGen/sse-builtins.c
+++ b/test/CodeGen/sse-builtins.c
@@ -450,7 +450,8 @@ __m128 test_mm_min_ss(__m128 A, __m128 B) {
 
 __m128 test_mm_move_ss(__m128 A, __m128 B) {
   // CHECK-LABEL: test_mm_move_ss
-  // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <4 x i32> <i32 4, i32 1, i32 2, i32 3>
+  // CHECK: extractelement <4 x float> %{{.*}}, i32 0
+  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
   return _mm_move_ss(A, B);
 }
 

diff --git a/test/CodeGen/sse2-builtins.c b/test/CodeGen/sse2-builtins.c
index dbf79c47367e9326ca02bef98f44938b15a7655c..fe7f7ccf83eeaf20681cb8e4b14a3922278dc784 100644 (file)
--- a/test/CodeGen/sse2-builtins.c
+++ b/test/CodeGen/sse2-builtins.c
@@ -794,9 +794,7 @@ __m128i test_mm_move_epi64(__m128i A) {
 __m128d test_mm_move_sd(__m128d A, __m128d B) {
   // CHECK-LABEL: test_mm_move_sd
   // CHECK: extractelement <2 x double> %{{.*}}, i32 0
-  // CHECK: insertelement <2 x double> undef, double %{{.*}}, i32 0
-  // CHECK: extractelement <2 x double> %{{.*}}, i32 1
-  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 1
+  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
   return _mm_move_sd(A, B);
 }