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

simd128: inline _mm_sqrt_ss

Browse Source
This commit is contained in:
myfreeer
2023-03-29 20:22:03 +08:00
parent e27f80b0bb
commit 5c7cd42407
1 changed files with 3 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
include/cglm/simd/wasm.h
View File

@@ -34,12 +34,6 @@ _mm_movelh_ps(glmm_128 __a, glmm_128 __b)
return wasm_i32x4_shuffle(__a, __b, 0, 1, 4, 5); return wasm_i32x4_shuffle(__a, __b, 0, 1, 4, 5);
} }
static inline glmm_128 __attribute__((__always_inline__, __nodebug__))
_mm_sqrt_ss(glmm_128 __a)
{
return wasm_i32x4_shuffle(__a, wasm_f32x4_sqrt(__a), 4, 1, 2, 3);
}
#define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \ #define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \
do { \ do { \
glmm_128 __row0 = (row0); \ glmm_128 __row0 = (row0); \
@@ -145,7 +139,9 @@ glmm_dot(glmm_128 a, glmm_128 b) {
static inline static inline
float float
glmm_norm(glmm_128 a) { glmm_norm(glmm_128 a) {
return _mm_cvtss_f32(_mm_sqrt_ss(glmm_vhadds(wasm_f32x4_mul(a, a)))); glmm_128 x0;
x0 = glmm_vhadds(wasm_f32x4_mul(a, a));
return _mm_cvtss_f32(wasm_i32x4_shuffle(x0, wasm_f32x4_sqrt(x0),4, 1, 2, 3));
} }
static inline static inline