summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--clang/lib/Headers/avx2intrin.h2
-rw-r--r--clang/lib/Headers/smmintrin.h55
-rw-r--r--clang/test/CodeGen/sse-builtins.c44
3 files changed, 84 insertions, 17 deletions
diff --git a/clang/lib/Headers/avx2intrin.h b/clang/lib/Headers/avx2intrin.h
index 359c0aad3c3..d165f1fbe2d 100644
--- a/clang/lib/Headers/avx2intrin.h
+++ b/clang/lib/Headers/avx2intrin.h
@@ -157,7 +157,7 @@ _mm256_blendv_epi8(__m256i __V1, __m256i __V2, __m256i __M)
#define _mm256_blend_epi16(V1, V2, M) __extension__ ({ \
__m256i __V1 = (V1); \
__m256i __V2 = (V2); \
- (__m256i)__builtin_ia32_pblendw256((__v16hi)__V1, (__v16hi)__V2, M); })
+ (__m256i)__builtin_ia32_pblendw256((__v16hi)__V1, (__v16hi)__V2, (M)); })
static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
_mm256_cmpeq_epi8(__m256i a, __m256i b)
diff --git a/clang/lib/Headers/smmintrin.h b/clang/lib/Headers/smmintrin.h
index 7e3babd743b..2fab50e4eb5 100644
--- a/clang/lib/Headers/smmintrin.h
+++ b/clang/lib/Headers/smmintrin.h
@@ -57,21 +57,34 @@
#define _mm_floor_ss(X, Y) _mm_round_ss((X), (Y), _MM_FROUND_FLOOR)
#define _mm_floor_sd(X, Y) _mm_round_sd((X), (Y), _MM_FROUND_FLOOR)
-#define _mm_round_ps(X, Y) __builtin_ia32_roundps((X), (Y))
-#define _mm_round_ss(X, Y, M) __builtin_ia32_roundss((X), (Y), (M))
-#define _mm_round_pd(X, M) __builtin_ia32_roundpd((X), (M))
-#define _mm_round_sd(X, Y, M) __builtin_ia32_roundsd((X), (Y), (M))
+#define _mm_round_ps(X, M) __extension__ ({ \
+ __m128 __X = (X); \
+ (__m128) __builtin_ia32_roundps((__v4sf)__X, (M)); })
+
+#define _mm_round_ss(X, Y, M) __extension__ ({ \
+ __m128 __X = (X); \
+ __m128 __Y = (Y); \
+ (__m128) __builtin_ia32_roundss((__v4sf)__X, (__v4sf)__Y, (M)); })
+
+#define _mm_round_pd(X, M) __extension__ ({ \
+ __m128d __X = (X); \
+ (__m128d) __builtin_ia32_roundpd((__v2df)__X, (M)); })
+
+#define _mm_round_sd(X, Y, M) __extension__ ({ \
+ __m128d __X = (X); \
+ __m128d __Y = (Y); \
+ (__m128d) __builtin_ia32_roundsd((__v2df)__X, (__v2df)__Y, (M)); })
/* SSE4 Packed Blending Intrinsics. */
#define _mm_blend_pd(V1, V2, M) __extension__ ({ \
__m128d __V1 = (V1); \
__m128d __V2 = (V2); \
- (__m128d) __builtin_ia32_blendpd ((__v2df)__V1, (__v2df)__V2, M); })
+ (__m128d) __builtin_ia32_blendpd ((__v2df)__V1, (__v2df)__V2, (M)); })
#define _mm_blend_ps(V1, V2, M) __extension__ ({ \
__m128 __V1 = (V1); \
__m128 __V2 = (V2); \
- (__m128) __builtin_ia32_blendps ((__v4sf)__V1, (__v4sf)__V2, M); })
+ (__m128) __builtin_ia32_blendps ((__v4sf)__V1, (__v4sf)__V2, (M)); })
static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
_mm_blendv_pd (__m128d __V1, __m128d __V2, __m128d __M)
@@ -97,7 +110,7 @@ _mm_blendv_epi8 (__m128i __V1, __m128i __V2, __m128i __M)
#define _mm_blend_epi16(V1, V2, M) __extension__ ({ \
__m128i __V1 = (V1); \
__m128i __V2 = (V2); \
- (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__V1, (__v8hi)__V2, M); })
+ (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__V1, (__v8hi)__V2, (M)); })
/* SSE4 Dword Multiply Instructions. */
static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
@@ -113,8 +126,15 @@ _mm_mul_epi32 (__m128i __V1, __m128i __V2)
}
/* SSE4 Floating Point Dot Product Instructions. */
-#define _mm_dp_ps(X, Y, M) __builtin_ia32_dpps ((X), (Y), (M))
-#define _mm_dp_pd(X, Y, M) __builtin_ia32_dppd ((X), (Y), (M))
+#define _mm_dp_ps(X, Y, M) __extension__ ({ \
+ __m128 __X = (X); \
+ __m128 __Y = (Y); \
+ (__m128) __builtin_ia32_dpps((__v4sf)__X, (__v4sf)__Y, (M)); })
+
+#define _mm_dp_pd(X, Y, M) __extension__ ({\
+ __m128d __X = (X); \
+ __m128d __Y = (Y); \
+ (__m128d) __builtin_ia32_dppd((__v2df)__X, (__v2df)__Y, (M)); })
/* SSE4 Streaming Load Hint Instruction. */
static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
@@ -195,14 +215,14 @@ _mm_max_epu32 (__m128i __V1, __m128i __V2)
/* Insert int into packed integer array at index. */
#define _mm_insert_epi8(X, I, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \
- __a[N] = I; \
+ __a[(N)] = (I); \
__a;}))
#define _mm_insert_epi32(X, I, N) (__extension__ ({ __v4si __a = (__v4si)(X); \
- __a[N] = I; \
+ __a[(N)] = (I); \
__a;}))
#ifdef __x86_64__
#define _mm_insert_epi64(X, I, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
- __a[N] = I; \
+ __a[(N)] = (I); \
__a;}))
#endif /* __x86_64__ */
@@ -210,12 +230,12 @@ _mm_max_epu32 (__m128i __V1, __m128i __V2)
* as a zero extended value, so it is unsigned.
*/
#define _mm_extract_epi8(X, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \
- (unsigned char)__a[N];}))
+ (unsigned char)__a[(N)];}))
#define _mm_extract_epi32(X, N) (__extension__ ({ __v4si __a = (__v4si)(X); \
- (unsigned)__a[N];}))
+ (unsigned)__a[(N)];}))
#ifdef __x86_64__
#define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \
- __a[N];}))
+ __a[(N)];}))
#endif /* __x86_64 */
/* SSE4 128-bit Packed Integer Comparisons. */
@@ -330,7 +350,10 @@ _mm_packus_epi32(__m128i __V1, __m128i __V2)
}
/* SSE4 Multiple Packed Sums of Absolute Difference. */
-#define _mm_mpsadbw_epu8(X, Y, M) __builtin_ia32_mpsadbw128((X), (Y), (M))
+#define _mm_mpsadbw_epu8(X, Y, M) __extension__ ({ \
+ __m128i __X = (X); \
+ __m128i __Y = (Y); \
+ (__m128i) __builtin_ia32_mpsadbw128((__v16qi)__X, (__v16qi)__Y, (M)); })
static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
_mm_minpos_epu16(__m128i __V)
diff --git a/clang/test/CodeGen/sse-builtins.c b/clang/test/CodeGen/sse-builtins.c
index 07c33e34359..2d574251555 100644
--- a/clang/test/CodeGen/sse-builtins.c
+++ b/clang/test/CodeGen/sse-builtins.c
@@ -105,5 +105,49 @@ __m128i test_loadl_epi64(void* y) {
}
__m128i test_mm_minpos_epu16(__m128i x) {
+ // CHECK: define {{.*}} @test_mm_minpos_epu16
+ // CHECK: @llvm.x86.sse41.phminposuw
return _mm_minpos_epu16(x);
}
+
+__m128i test_mm_mpsadbw_epu8(__m128i x, __m128i y) {
+ // CHECK: define {{.*}} @test_mm_mpsadbw_epu8
+ // CHECK: @llvm.x86.sse41.mpsadbw
+ return _mm_mpsadbw_epu8(x, y, 1);
+}
+
+__m128 test_mm_dp_ps(__m128 x, __m128 y) {
+ // CHECK: define {{.*}} @test_mm_dp_ps
+ // CHECK: @llvm.x86.sse41.dpps
+ return _mm_dp_ps(x, y, 2);
+}
+
+__m128d test_mm_dp_pd(__m128d x, __m128d y) {
+ // CHECK: define {{.*}} @test_mm_dp_pd
+ // CHECK: @llvm.x86.sse41.dppd
+ return _mm_dp_pd(x, y, 2);
+}
+
+__m128 test_mm_round_ps(__m128 x) {
+ // CHECK: define {{.*}} @test_mm_round_ps
+ // CHECK: @llvm.x86.sse41.round.ps
+ return _mm_round_ps(x, 2);
+}
+
+__m128 test_mm_round_ss(__m128 x, __m128 y) {
+ // CHECK: define {{.*}} @test_mm_round_ss
+ // CHECK: @llvm.x86.sse41.round.ss
+ return _mm_round_ss(x, y, 2);
+}
+
+__m128d test_mm_round_pd(__m128d x) {
+ // CHECK: define {{.*}} @test_mm_round_pd
+ // CHECK: @llvm.x86.sse41.round.pd
+ return _mm_round_pd(x, 2);
+}
+
+__m128d test_mm_round_sd(__m128d x, __m128d y) {
+ // CHECK: define {{.*}} @test_mm_round_sd
+ // CHECK: @llvm.x86.sse41.round.sd
+ return _mm_round_sd(x, y, 2);
+}
OpenPOWER on IntegriCloud