External/cglm
1
0
Fork 0
mirror of https://github.com/recp/cglm.git synced 2025-10-03 16:51:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

wasm: prefer pmin/pmax

Browse Source 
According to [emscripten](https://emscripten.org/docs/porting/simd.html) and [v8](b6520eda5e/src/compiler/backend/x64/code-generator-x64.cc (L2661-L2699)), `[f32x4|f64x2].[min|max]` compiles to much more instructions than `[f32x4|f64x2].[pmin|pmax]`.
It is defined in [spec](https://github.com/WebAssembly/spec/blob/main/proposals/simd/SIMD.md#floating-point-min-and-max) that the difference between pmin/pmax and min/max is NaN-propagating behavior, and the equivalent to the x86 `_mm_min_ps`/`_mm_max_ps` is pmin/pmax in [v8](b6520eda5e/src/compiler/backend/x64/code-generator-x64.cc (L2740-L2747)).
This should make functions with min/max faster on webassembly, and align with the existing behavior with x86 sse.
This commit is contained in:
myfreeer
2023-12-02 09:54:49 +08:00
committed by GitHub
parent 1fdc1c8675
commit 009405adcd
1 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
include/cglm/vec4.h
View File

@@ -649,7 +649,7 @@ glm_vec4_maxadd(vec4 a, vec4 b, vec4 dest) {
#if defined(__wasm__) && defined(__wasm_simd128__)
glmm_store(dest, wasm_f32x4_add(
glmm_load(dest),
wasm_f32x4_max(glmm_load(a), glmm_load(b))));
wasm_f32x4_pmax(glmm_load(a), glmm_load(b))));
#elif defined( __SSE__ ) || defined( __SSE2__ )
glmm_store(dest, _mm_add_ps(glmm_load(dest),
_mm_max_ps(glmm_load(a),
@@ -681,7 +681,7 @@ glm_vec4_minadd(vec4 a, vec4 b, vec4 dest) {
#if defined(__wasm__) && defined(__wasm_simd128__)
glmm_store(dest, wasm_f32x4_add(
glmm_load(dest),
wasm_f32x4_min(glmm_load(a), glmm_load(b))));
wasm_f32x4_pmin(glmm_load(a), glmm_load(b))));
#elif defined( __SSE__ ) || defined( __SSE2__ )
glmm_store(dest, _mm_add_ps(glmm_load(dest),
_mm_min_ps(glmm_load(a),
@@ -854,7 +854,7 @@ CGLM_INLINE
void
glm_vec4_maxv(vec4 a, vec4 b, vec4 dest) {
#if defined(__wasm__) && defined(__wasm_simd128__)
glmm_store(dest, wasm_f32x4_max(glmm_load(a), glmm_load(b)));
glmm_store(dest, wasm_f32x4_pmax(glmm_load(a), glmm_load(b)));
#elif defined( __SSE__ ) || defined( __SSE2__ )
glmm_store(dest, _mm_max_ps(glmm_load(a), glmm_load(b)));
#elif defined(CGLM_NEON_FP)
@@ -878,7 +878,7 @@ CGLM_INLINE
void
glm_vec4_minv(vec4 a, vec4 b, vec4 dest) {
#if defined(__wasm__) && defined(__wasm_simd128__)
glmm_store(dest, wasm_f32x4_min(glmm_load(a), glmm_load(b)));
glmm_store(dest, wasm_f32x4_pmin(glmm_load(a), glmm_load(b)));
#elif defined( __SSE__ ) || defined( __SSE2__ )
glmm_store(dest, _mm_min_ps(glmm_load(a), glmm_load(b)));
#elif defined(CGLM_NEON_FP)
@@ -902,8 +902,8 @@ CGLM_INLINE
void
glm_vec4_clamp(vec4 v, float minVal, float maxVal) {
#if defined(__wasm__) && defined(__wasm_simd128__)
glmm_store(v, wasm_f32x4_min(
wasm_f32x4_max(glmm_load(v), wasm_f32x4_splat(minVal)),
glmm_store(v, wasm_f32x4_pmin(
wasm_f32x4_pmax(glmm_load(v), wasm_f32x4_splat(minVal)),
wasm_f32x4_splat(maxVal)));
#elif defined( __SSE__ ) || defined( __SSE2__ )
glmm_store(v, _mm_min_ps(_mm_max_ps(glmm_load(v), _mm_set1_ps(minVal)),