diff options
| author | eschnett <eschnett@105869f7-3296-0410-a4ea-f4349344b45a> | 2011-08-25 17:40:05 +0000 |
|---|---|---|
| committer | eschnett <eschnett@105869f7-3296-0410-a4ea-f4349344b45a> | 2011-08-25 17:40:05 +0000 |
| commit | 802b82837b5b37b7c76ee807939bbffe76f17fdd (patch) | |
| tree | cc18cef877eb8067bdbd202a696112cf2d6e0431 | |
| parent | b075a3dfcf5aaa72c086d7896ab0c975f42d04c2 (diff) | |
Suggest asm statements to support SSE4a with Intel compilers.
Indent vector architecture definitions.
git-svn-id: https://svn.cct.lsu.edu/repos/numrel/LSUThorns/Vectors/trunk@30 105869f7-3296-0410-a4ea-f4349344b45a
| -rw-r--r-- | src/vectors-4-SSE.h | 6 | ||||
| -rw-r--r-- | src/vectors-8-SSE2.h | 16 |
2 files changed, 16 insertions, 6 deletions
diff --git a/src/vectors-4-SSE.h b/src/vectors-4-SSE.h index 2f55183..dbf0cce 100644 --- a/src/vectors-4-SSE.h +++ b/src/vectors-4-SSE.h @@ -21,11 +21,11 @@ #ifdef __SSE4_1__ -#define vec4_architecture "SSE4.1 (32-bit precision)" +# define vec4_architecture "SSE4.1 (32-bit precision)" #elif defined(__SSE4A__) -#define vec4_architecture "SSE4A (32-bit precision)" +# define vec4_architecture "SSE4A (32-bit precision)" #else -#define vec4_architecture "SSE (32-bit precision)" +# define vec4_architecture "SSE (32-bit precision)" #endif // Vector type corresponding to CCTK_REAL diff --git a/src/vectors-8-SSE2.h b/src/vectors-8-SSE2.h index 56c614d..3b11990 100644 --- a/src/vectors-8-SSE2.h +++ b/src/vectors-8-SSE2.h @@ -16,16 +16,26 @@ #ifdef __SSE4A__ // AMD's SSE 4a # include <ammintrin.h> + +// Intel compilers don't support SSE 4a. Here is how we can implement +// these instructions in assembler instead: + +// inline void __attribute__((__always_inline__)) +// _mm_stream_sd (double *p, __m128d x) +// { +// asm ("movntsd %[x],%[p]" : "=m" (*p) : [p] "m" (*p), [x] "x" (x)); +// } + #endif #ifdef __SSE4_1__ -#define vec8_architecture "SSE4.1 (64-bit precision)" +# define vec8_architecture "SSE4.1 (64-bit precision)" #elif defined(__SSE4A__) -#define vec8_architecture "SSE4A (64-bit precision)" +# define vec8_architecture "SSE4A (64-bit precision)" #else -#define vec8_architecture "SSE2 (64-bit precision)" +# define vec8_architecture "SSE2 (64-bit precision)" #endif // Vector type corresponding to CCTK_REAL |