External/cglm
1
0
Fork 0
mirror of https://github.com/recp/cglm.git synced 2025-10-04 17:09:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

glmm: define fma functions

Browse Source
This commit is contained in:
Recep Aslantas
2021-04-18 13:49:50 +03:00
parent 47e0045015
commit c5655bbd2e
2 changed files with 40 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
include/cglm/simd/arm.h
View File

@@ -79,5 +79,25 @@ glmm_norm_inf(float32x4_t a) {
return glmm_hmax(glmm_abs(a)); return glmm_hmax(glmm_abs(a));
} }
static inline
float32x4_t
glmm_fmadd(float32x4_t a, float32x4_t b, float32x4_t c) {
#if defined(__aarch64__)
return vfmaq_f32(a, b, c);
#else
return vmlaq_f32(a, b, c);
#endif
}
static inline
float32x4_t
glmm_fnmadd(float32x4_t a, float32x4_t b, float32x4_t c) {
#if defined(__aarch64__)
return vfmsq_f32(a, b, c);
#else
return vmlsq_f32(a, b, c);
#endif
}
#endif #endif
#endif /* cglm_simd_arm_h */ #endif /* cglm_simd_arm_h */

20
include/cglm/simd/x86.h
View File

@@ -197,5 +197,25 @@ glmm_store3(float v[3], __m128 vx) {
_mm_store_ss(&v[2], glmm_shuff1(vx, 2, 2, 2, 2)); _mm_store_ss(&v[2], glmm_shuff1(vx, 2, 2, 2, 2));
} }
static inline
__m128
glmm_fmadd(__m128 a, __m128 b, __m128 c) {
#ifdef __FMA__
return _mm_fmadd_ps(a, b, c);
#else
return _mm_add_ps(c, _mm_mul_ps(a, b));
#endif
}
static inline
__m128
glmm_fnmadd(__m128 a, __m128 b, __m128 c) {
#ifdef __FMA__
return _mm_fnmadd_ps(a, b, c);
#else
return _mm_sub_ps(c, _mm_mul_ps(a, b));
#endif
}
#endif #endif
#endif /* cglm_simd_x86_h */ #endif /* cglm_simd_x86_h */