External/cglm
1
0
Fork 0
mirror of https://github.com/recp/cglm.git synced 2026-01-02 21:59:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

arm, neon: optimize glmm_vhadd and add glmm_vdot

Browse Source
This commit is contained in:
Recep Aslantas
2024-04-02 02:36:16 +03:00
parent 8366e51b47
commit f0e09776d7
1 changed files with 15 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
include/cglm/simd/arm.h
View File

@@ -63,8 +63,17 @@ static inline float32x4_t glmm_max(float32x4_t a, float32x4_t b) { return vmaxq_
static inline
float32x4_t
glmm_vhadd(float32x4_t v) {
float32x4_t p;
p = vpaddq_f32(v, v); /* [a+b, c+d, a+b, c+d] */
return vpaddq_f32(p, p); /* [t, t, t, t] */;
/* TODO: measure speed of this compare to above */
/* return vdupq_n_f32(vaddvq_f32(v)); */
/*
return vaddq_f32(vaddq_f32(glmm_splat_x(v), glmm_splat_y(v)),
vaddq_f32(glmm_splat_z(v), glmm_splat_w(v)));
*/
/*
this seems slower:
v = vaddq_f32(v, vrev64q_f32(v));
@@ -108,6 +117,12 @@ glmm_dot(float32x4_t a, float32x4_t b) {
return glmm_hadd(vmulq_f32(a, b));
}
static inline
float32x4_t
glmm_vdot(float32x4_t a, float32x4_t b) {
return glmm_vhadd(vmulq_f32(a, b));
}
static inline
float
glmm_norm(float32x4_t a) {