YourWishes/cglm
1
0
Fork 0
mirror of https://github.com/recp/cglm.git synced 2026-02-17 03:39:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

simd: remove re-load in SSE4 and SSE3

Browse Source
This commit is contained in:
Recep Aslantas
2019-01-25 21:56:17 +03:00
parent 31bb303c55
commit fc7f958167
1 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
include/cglm/simd/x86.h
View File

@@ -72,10 +72,10 @@ static inline
__m128 __m128
glmm_vdots(__m128 a, __m128 b) { glmm_vdots(__m128 a, __m128 b) {
#if (defined(__SSE4_1__) || defined(__SSE4_2__)) && defined(CGLM_SSE4_DOT) #if (defined(__SSE4_1__) || defined(__SSE4_2__)) && defined(CGLM_SSE4_DOT)
return _mm_dp_ps(glmm_load(a), glmm_load(b), 0xFF); return _mm_dp_ps(a, b, 0xFF);
#elif defined(__SSE3__) && defined(CGLM_SSE3_DOT) #elif defined(__SSE3__) && defined(CGLM_SSE3_DOT)
__m128 x0, x1; __m128 x0, x1;
x0 = _mm_mul_ps(glmm_load(a), glmm_load(b)); x0 = _mm_mul_ps(a, b);
x1 = _mm_hadd_ps(x0, x0); x1 = _mm_hadd_ps(x0, x0);
return _mm_hadd_ps(x1, x1); return _mm_hadd_ps(x1, x1);
#else #else
@@ -87,10 +87,10 @@ static inline
__m128 __m128
glmm_vdot(__m128 a, __m128 b) { glmm_vdot(__m128 a, __m128 b) {
#if (defined(__SSE4_1__) || defined(__SSE4_2__)) && defined(CGLM_SSE4_DOT) #if (defined(__SSE4_1__) || defined(__SSE4_2__)) && defined(CGLM_SSE4_DOT)
return _mm_dp_ps(glmm_load(a), glmm_load(b), 0xFF); return _mm_dp_ps(a, b, 0xFF);
#elif defined(__SSE3__) && defined(CGLM_SSE3_DOT) #elif defined(__SSE3__) && defined(CGLM_SSE3_DOT)
__m128 x0, x1; __m128 x0, x1;
x0 = _mm_mul_ps(glmm_load(a), glmm_load(b)); x0 = _mm_mul_ps(a, b);
x1 = _mm_hadd_ps(x0, x0); x1 = _mm_hadd_ps(x0, x0);
return _mm_hadd_ps(x1, x1); return _mm_hadd_ps(x1, x1);
#else #else