External/cglm
1
0
Fork 0
mirror of https://github.com/recp/cglm.git synced 2025-10-04 09:08:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

arm: define CGLM_ARM64 for identify arm64

Browse Source
This commit is contained in:
Recep Aslantas
2021-04-24 00:44:07 +03:00
parent 65292a94a6
commit 059e5010e6
1 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
include/cglm/simd/arm.h
View File

@@ -10,6 +10,10 @@
#include "intrin.h" #include "intrin.h"
#ifdef CGLM_SIMD_ARM #ifdef CGLM_SIMD_ARM
#if defined(_M_ARM64) || defined(_M_HYBRID_X86_ARM64) || defined(_M_ARM64EC) || defined(__aarch64__)
# define CGLM_ARM64 1
#endif
#define glmm_load(p) vld1q_f32(p) #define glmm_load(p) vld1q_f32(p)
#define glmm_store(p, a) vst1q_f32(p, a) #define glmm_store(p, a) vst1q_f32(p, a)
@@ -70,7 +74,7 @@ glmm_abs(float32x4_t v) {
static inline static inline
float float
glmm_hadd(float32x4_t v) { glmm_hadd(float32x4_t v) {
#if defined(__aarch64__) #if CGLM_ARM64
return vaddvq_f32(v); return vaddvq_f32(v);
#else #else
v = vaddq_f32(v, vrev64q_f32(v)); v = vaddq_f32(v, vrev64q_f32(v));
@@ -130,7 +134,7 @@ glmm_norm_inf(float32x4_t a) {
static inline static inline
float32x4_t float32x4_t
glmm_fmadd(float32x4_t a, float32x4_t b, float32x4_t c) { glmm_fmadd(float32x4_t a, float32x4_t b, float32x4_t c) {
#if defined(__aarch64__) #if CGLM_ARM64
return vfmaq_f32(c, a, b); return vfmaq_f32(c, a, b);
#else #else
return vmlaq_f32(c, a, b); return vmlaq_f32(c, a, b);
@@ -140,7 +144,7 @@ glmm_fmadd(float32x4_t a, float32x4_t b, float32x4_t c) {
static inline static inline
float32x4_t float32x4_t
glmm_fnmadd(float32x4_t a, float32x4_t b, float32x4_t c) { glmm_fnmadd(float32x4_t a, float32x4_t b, float32x4_t c) {
#if defined(__aarch64__) #if CGLM_ARM64
return vfmsq_f32(c, a, b); return vfmsq_f32(c, a, b);
#else #else
return vmlsq_f32(c, a, b); return vmlsq_f32(c, a, b);
@@ -150,7 +154,7 @@ glmm_fnmadd(float32x4_t a, float32x4_t b, float32x4_t c) {
static inline static inline
float32x4_t float32x4_t
glmm_fmsub(float32x4_t a, float32x4_t b, float32x4_t c) { glmm_fmsub(float32x4_t a, float32x4_t b, float32x4_t c) {
#if defined(__aarch64__) #if CGLM_ARM64
return vfmsq_f32(c, a, b); return vfmsq_f32(c, a, b);
#else #else
return vmlsq_f32(c, a, b); return vmlsq_f32(c, a, b);