diff options
| -rw-r--r-- | clang/include/clang/Basic/BuiltinsX86.def | 2 | ||||
| -rw-r--r-- | clang/lib/Headers/xopintrin.h | 4 | ||||
| -rw-r--r-- | clang/test/CodeGen/xop-builtins.c | 10 |
3 files changed, 10 insertions, 6 deletions
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def index e147c8fb86b..839122c4bba 100644 --- a/clang/include/clang/Basic/BuiltinsX86.def +++ b/clang/include/clang/Basic/BuiltinsX86.def @@ -832,8 +832,6 @@ TARGET_BUILTIN(__builtin_ia32_vphaddudq, "V2LLiV4i", "", "xop") TARGET_BUILTIN(__builtin_ia32_vphsubbw, "V8sV16c", "", "xop") TARGET_BUILTIN(__builtin_ia32_vphsubwd, "V4iV8s", "", "xop") TARGET_BUILTIN(__builtin_ia32_vphsubdq, "V2LLiV4i", "", "xop") -TARGET_BUILTIN(__builtin_ia32_vpcmov, "V2LLiV2LLiV2LLiV2LLi", "", "xop") -TARGET_BUILTIN(__builtin_ia32_vpcmov_256, "V4LLiV4LLiV4LLiV4LLi", "", "xop") TARGET_BUILTIN(__builtin_ia32_vpperm, "V16cV16cV16cV16c", "", "xop") TARGET_BUILTIN(__builtin_ia32_vprotb, "V16cV16cV16c", "", "xop") TARGET_BUILTIN(__builtin_ia32_vprotw, "V8sV8sV8s", "", "xop") diff --git a/clang/lib/Headers/xopintrin.h b/clang/lib/Headers/xopintrin.h index bdf0cec3264..6104bc66738 100644 --- a/clang/lib/Headers/xopintrin.h +++ b/clang/lib/Headers/xopintrin.h @@ -198,13 +198,13 @@ _mm_hsubq_epi32(__m128i __A) static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_cmov_si128(__m128i __A, __m128i __B, __m128i __C) { - return (__m128i)__builtin_ia32_vpcmov((__v2di)__A, (__v2di)__B, (__v2di)__C); + return (__m128i)((__v2du)__A & (__v2du)__C) | ((__v2du)__B & ~(__v2du)__C); } static __inline__ __m256i __DEFAULT_FN_ATTRS _mm256_cmov_si256(__m256i __A, __m256i __B, __m256i __C) { - return (__m256i)__builtin_ia32_vpcmov_256((__v4di)__A, (__v4di)__B, (__v4di)__C); + return (__m256i)((__v4du)__A & (__v4du)__C) | ((__v4du)__B & ~(__v4du)__C); } static __inline__ __m128i __DEFAULT_FN_ATTRS diff --git a/clang/test/CodeGen/xop-builtins.c b/clang/test/CodeGen/xop-builtins.c index da9a3b925de..5302b9ab8f2 100644 --- a/clang/test/CodeGen/xop-builtins.c +++ b/clang/test/CodeGen/xop-builtins.c @@ -170,13 +170,19 @@ __m128i test_mm_hsubq_epi32(__m128i a) { __m128i test_mm_cmov_si128(__m128i a, __m128i b, __m128i c) { // CHECK-LABEL: test_mm_cmov_si128 - // CHECK: call <2 x i64> @llvm.x86.xop.vpcmov(<2 x i64> %{{.*}}, <2 x i64> %{{.*}}, <2 x i64> %{{.*}}) + // CHECK: [[AND:%.*]] = and <2 x i64> %{{.*}}, %{{.*}} + // CHECK: [[NEG:%.*]] = xor <2 x i64> %{{.*}}, <i64 -1, i64 -1> + // CHECK-NEXT: [[ANDN:%.*]] = and <2 x i64> %{{.*}}, [[NEG]] + // CHECK-NEXT: %{{.*}} = or <2 x i64> [[AND]], [[ANDN]] return _mm_cmov_si128(a, b, c); } __m256i test_mm256_cmov_si256(__m256i a, __m256i b, __m256i c) { // CHECK-LABEL: test_mm256_cmov_si256 - // CHECK: call <4 x i64> @llvm.x86.xop.vpcmov.256(<4 x i64> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}) + // CHECK: [[AND:%.*]] = and <4 x i64> %{{.*}}, %{{.*}} + // CHECK: [[NEG:%.*]] = xor <4 x i64> %{{.*}}, <i64 -1, i64 -1, i64 -1, i64 -1> + // CHECK-NEXT: [[ANDN:%.*]] = and <4 x i64> %{{.*}}, [[NEG]] + // CHECK-NEXT: %{{.*}} = or <4 x i64> [[AND]], [[ANDN]] return _mm256_cmov_si256(a, b, c); } |
