diff options
| author | Jukka Ojanen <jukka.ojanen@linkotec.net> | 2014-11-20 16:54:24 +0200 |
|---|---|---|
| committer | Jukka Ojanen <jukka.ojanen@linkotec.net> | 2014-11-20 16:54:24 +0200 |
| commit | d896a265d5c75a15333b0998cb7eaa465ecb8419 (patch) | |
| tree | dbbf20d73c7ebf826a11a3cab72333e6a91543b9 /src | |
| parent | 71f1f4dae77c2f6b335c3e06c13a3ecedf73ccda (diff) | |
| download | ffts-d896a265d5c75a15333b0998cb7eaa465ecb8419.zip ffts-d896a265d5c75a15333b0998cb7eaa465ecb8419.tar.gz | |
Use _mm_movelh_ps instead of _mm_shuffle_ps to implement VUNPACKLO
Diffstat (limited to 'src')
| -rw-r--r-- | src/macros-sse.h | 6 |
1 files changed, 2 insertions, 4 deletions
diff --git a/src/macros-sse.h b/src/macros-sse.h index 85cd02d..cab822c 100644 --- a/src/macros-sse.h +++ b/src/macros-sse.h @@ -43,7 +43,7 @@ typedef __m128 V; #define VADD _mm_add_ps #define VSUB _mm_sub_ps #define VMUL _mm_mul_ps -//#define VLIT4 _mm_set_ps +#define VLIT4 _mm_set_ps #define VXOR _mm_xor_ps #define VST _mm_store_ps #define VLD _mm_load_ps @@ -51,12 +51,10 @@ typedef __m128 V; #define VSWAPPAIRS(x) (_mm_shuffle_ps(x,x,_MM_SHUFFLE(2,3,0,1))) #define VUNPACKHI(x,y) (_mm_shuffle_ps(x,y,_MM_SHUFFLE(3,2,3,2))) -#define VUNPACKLO(x,y) (_mm_shuffle_ps(x,y,_MM_SHUFFLE(1,0,1,0))) +#define VUNPACKLO(x,y) (_mm_movelh_ps(x,y)) #define VBLEND(x,y) (_mm_shuffle_ps(x,y,_MM_SHUFFLE(3,2,1,0))) -#define VLIT4 _mm_set_ps - #define VDUPRE(r) (_mm_shuffle_ps(r,r,_MM_SHUFFLE(2,2,0,0))) #define VDUPIM(r) (_mm_shuffle_ps(r,r,_MM_SHUFFLE(3,3,1,1))) |
