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

Merge branch 'master' into optimizations

Browse Source
This commit is contained in:
Recep Aslantas
2018-05-10 13:59:10 +03:00
committed by GitHub
parent cfd3600107 48d33c16cb
commit 9fc2ead8ef
64 changed files with 4736 additions and 770 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
include/cglm/simd/avx/affine.h
View File

@@ -21,11 +21,11 @@ glm_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
__m256 y0, y1, y2, y3, y4, y5, y6, y7, y8, y9;
y0 = _mm256_load_ps(m2[0]); /* h g f e d c b a */
y1 = _mm256_load_ps(m2[2]); /* p o n m l k j i */
y0 = glmm_load256(m2[0]); /* h g f e d c b a */
y1 = glmm_load256(m2[2]); /* p o n m l k j i */
y2 = _mm256_load_ps(m1[0]); /* h g f e d c b a */
y3 = _mm256_load_ps(m1[2]); /* p o n m l k j i */
y2 = glmm_load256(m1[0]); /* h g f e d c b a */
y3 = glmm_load256(m1[2]); /* p o n m l k j i */
y4 = _mm256_permute2f128_ps(y2, y2, 0b00000011); /* d c b a h g f e */
y5 = _mm256_permute2f128_ps(y3, y3, 0b00000000); /* l k j i l k j i */
@@ -37,10 +37,10 @@ glm_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
y6 = _mm256_permutevar_ps(y0, _mm256_set_epi32(1, 1, 1, 1, 0, 0, 0, 0));
y8 = _mm256_permutevar_ps(y0, _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1));
_mm256_store_ps(dest[0],
_mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
_mm256_mul_ps(y4, y8)),
_mm256_mul_ps(y5, y7)));
glmm_store256(dest[0],
_mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
_mm256_mul_ps(y4, y8)),
_mm256_mul_ps(y5, y7)));
/* n n n n i i i i */
@@ -52,11 +52,11 @@ glm_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
y8 = _mm256_permutevar_ps(y1, _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1));
y9 = _mm256_permutevar_ps(y1, _mm256_set_epi32(2, 2, 2, 2, 3, 3, 3, 3));
_mm256_store_ps(dest[2],
_mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
_mm256_mul_ps(y3, y7)),
_mm256_add_ps(_mm256_mul_ps(y4, y8),
_mm256_mul_ps(y5, y9))));
glmm_store256(dest[2],
_mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
_mm256_mul_ps(y3, y7)),
_mm256_add_ps(_mm256_mul_ps(y4, y8),
_mm256_mul_ps(y5, y9))));
}
#endif

28
include/cglm/simd/avx/mat4.h
View File

@@ -21,11 +21,11 @@ glm_mat4_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
__m256 y0, y1, y2, y3, y4, y5, y6, y7, y8, y9;
y0 = _mm256_load_ps(m2[0]); /* h g f e d c b a */
y1 = _mm256_load_ps(m2[2]); /* p o n m l k j i */
y0 = glmm_load256(m2[0]); /* h g f e d c b a */
y1 = glmm_load256(m2[2]); /* p o n m l k j i */
y2 = _mm256_load_ps(m1[0]); /* h g f e d c b a */
y3 = _mm256_load_ps(m1[2]); /* p o n m l k j i */
y2 = glmm_load256(m1[0]); /* h g f e d c b a */
y3 = glmm_load256(m1[2]); /* p o n m l k j i */
y4 = _mm256_permute2f128_ps(y2, y2, 0b00000011); /* d c b a h g f e */
y5 = _mm256_permute2f128_ps(y3, y3, 0b00000011); /* l k j i p o n m */
@@ -39,11 +39,11 @@ glm_mat4_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
y8 = _mm256_permutevar_ps(y0, _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1));
y9 = _mm256_permutevar_ps(y0, _mm256_set_epi32(2, 2, 2, 2, 3, 3, 3, 3));
_mm256_store_ps(dest[0],
_mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
_mm256_mul_ps(y3, y7)),
_mm256_add_ps(_mm256_mul_ps(y4, y8),
_mm256_mul_ps(y5, y9))));
glmm_store256(dest[0],
_mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
_mm256_mul_ps(y3, y7)),
_mm256_add_ps(_mm256_mul_ps(y4, y8),
_mm256_mul_ps(y5, y9))));
/* n n n n i i i i */
/* p p p p k k k k */
@@ -54,11 +54,11 @@ glm_mat4_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
y8 = _mm256_permutevar_ps(y1, _mm256_set_epi32(0, 0, 0, 0, 1, 1, 1, 1));
y9 = _mm256_permutevar_ps(y1, _mm256_set_epi32(2, 2, 2, 2, 3, 3, 3, 3));
_mm256_store_ps(dest[2],
_mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
_mm256_mul_ps(y3, y7)),
_mm256_add_ps(_mm256_mul_ps(y4, y8),
_mm256_mul_ps(y5, y9))));
glmm_store256(dest[2],
_mm256_add_ps(_mm256_add_ps(_mm256_mul_ps(y2, y6),
_mm256_mul_ps(y3, y7)),
_mm256_add_ps(_mm256_mul_ps(y4, y8),
_mm256_mul_ps(y5, y9))));
}
#endif

66
include/cglm/simd/intrin.h
View File

@@ -8,11 +8,19 @@
#ifndef cglm_intrin_h
#define cglm_intrin_h
#if defined( _WIN32 )
#if defined( _MSC_VER )
# if (defined(_M_AMD64) || defined(_M_X64)) || _M_IX86_FP == 2
# define __SSE2__
# ifndef __SSE2__
# define __SSE2__
# endif
# elif _M_IX86_FP == 1
# define __SSE__
# ifndef __SSE__
# define __SSE__
# endif
# endif
/* do not use alignment for older visual studio versions */
# if _MSC_VER < 1913 /* Visual Studio 2017 version 15.6 */
# define CGLM_ALL_UNALIGNED
# endif
#endif
@@ -36,6 +44,49 @@
# define _mm_shuffle2_ps(a, b, z0, y0, x0, w0, z1, y1, x1, w1) \
_mm_shuffle1_ps(_mm_shuffle_ps(a, b, _MM_SHUFFLE(z0, y0, x0, w0)), \
z1, y1, x1, w1)
static inline
__m128
glmm_dot(__m128 a, __m128 b) {
__m128 x0;
x0 = _mm_mul_ps(a, b);
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 3, 2));
return _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 0, 1));
}
static inline
__m128
glmm_norm(__m128 a) {
return _mm_sqrt_ps(glmm_dot(a, a));
}
static inline
__m128
glmm_load3(float v[3]) {
__m128i xy;
__m128 z;
xy = _mm_loadl_epi64((const __m128i *)v);
z = _mm_load_ss(&v[2]);
return _mm_movelh_ps(_mm_castsi128_ps(xy), z);
}
static inline
void
glmm_store3(__m128 vx, float v[3]) {
_mm_storel_pi((__m64 *)&v[0], vx);
_mm_store_ss(&v[2], _mm_shuffle1_ps(vx, 2, 2, 2, 2));
}
#ifdef CGLM_ALL_UNALIGNED
# define glmm_load(p) _mm_loadu_ps(p)
# define glmm_store(p, a) _mm_storeu_ps(p, a)
#else
# define glmm_load(p) _mm_load_ps(p)
# define glmm_store(p, a) _mm_store_ps(p, a)
#endif
#endif
/* x86, x64 */
@@ -45,6 +96,15 @@
#ifdef __AVX__
# define CGLM_AVX_FP 1
#ifdef CGLM_ALL_UNALIGNED
# define glmm_load256(p) _mm256_loadu_ps(p)
# define glmm_store256(p, a) _mm256_storeu_ps(p, a)
#else
# define glmm_load256(p) _mm256_load_ps(p)
# define glmm_store256(p, a) _mm256_store_ps(p, a)
#endif
#endif
/* ARM Neon */

100
include/cglm/simd/sse2/affine.h
View File

@@ -18,35 +18,67 @@ glm_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
/* D = R * L (Column-Major) */
__m128 l0, l1, l2, l3, r;
l0 = _mm_load_ps(m1[0]);
l1 = _mm_load_ps(m1[1]);
l2 = _mm_load_ps(m1[2]);
l3 = _mm_load_ps(m1[3]);
l0 = glmm_load(m1[0]);
l1 = glmm_load(m1[1]);
l2 = glmm_load(m1[2]);
l3 = glmm_load(m1[3]);
r = _mm_load_ps(m2[0]);
_mm_store_ps(dest[0],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
r = glmm_load(m2[0]);
glmm_store(dest[0],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
r = _mm_load_ps(m2[1]);
_mm_store_ps(dest[1],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
r = glmm_load(m2[1]);
glmm_store(dest[1],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
r = _mm_load_ps(m2[2]);
_mm_store_ps(dest[2],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
r = glmm_load(m2[2]);
glmm_store(dest[2],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
r = _mm_load_ps(m2[3]);
_mm_store_ps(dest[3],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
r = glmm_load(m2[3]);
glmm_store(dest[3],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
}
CGLM_INLINE
void
glm_mul_rot_sse2(mat4 m1, mat4 m2, mat4 dest) {
/* D = R * L (Column-Major) */
__m128 l0, l1, l2, l3, r;
l0 = glmm_load(m1[0]);
l1 = glmm_load(m1[1]);
l2 = glmm_load(m1[2]);
l3 = glmm_load(m1[3]);
r = glmm_load(m2[0]);
glmm_store(dest[0],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
r = glmm_load(m2[1]);
glmm_store(dest[1],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
r = glmm_load(m2[2]);
glmm_store(dest[2],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2)));
glmm_store(dest[3], l3);
}
CGLM_INLINE
@@ -54,11 +86,11 @@ void
glm_inv_tr_sse2(mat4 mat) {
__m128 r0, r1, r2, r3, x0, x1;
r0 = _mm_load_ps(mat[0]);
r1 = _mm_load_ps(mat[1]);
r2 = _mm_load_ps(mat[2]);
r3 = _mm_load_ps(mat[3]);
x1 = _mm_set_ps(1.0f, 0.0f, 0.0f, 0.0f);
r0 = glmm_load(mat[0]);
r1 = glmm_load(mat[1]);
r2 = glmm_load(mat[2]);
r3 = glmm_load(mat[3]);
x1 = _mm_set_ps(1.0f, 0.0f, 0.0f, 0.0f);
_MM_TRANSPOSE4_PS(r0, r1, r2, x1);
@@ -69,10 +101,10 @@ glm_inv_tr_sse2(mat4 mat) {
x0 = _mm_add_ps(x0, x1);
_mm_store_ps(mat[0], r0);
_mm_store_ps(mat[1], r1);
_mm_store_ps(mat[2], r2);
_mm_store_ps(mat[3], x0);
glmm_store(mat[0], r0);
glmm_store(mat[1], r1);
glmm_store(mat[2], r2);
glmm_store(mat[3], x0);
}
#endif

132
include/cglm/simd/sse2/mat4.h
View File

@@ -20,10 +20,10 @@ glm_mat4_scale_sse2(mat4 m, float s){
__m128 x0;
x0 = _mm_set1_ps(s);
_mm_store_ps(m[0], _mm_mul_ps(_mm_load_ps(m[0]), x0));
_mm_store_ps(m[1], _mm_mul_ps(_mm_load_ps(m[1]), x0));
_mm_store_ps(m[2], _mm_mul_ps(_mm_load_ps(m[2]), x0));
_mm_store_ps(m[3], _mm_mul_ps(_mm_load_ps(m[3]), x0));
glmm_store(m[0], _mm_mul_ps(glmm_load(m[0]), x0));
glmm_store(m[1], _mm_mul_ps(glmm_load(m[1]), x0));
glmm_store(m[2], _mm_mul_ps(glmm_load(m[2]), x0));
glmm_store(m[3], _mm_mul_ps(glmm_load(m[3]), x0));
}
CGLM_INLINE
@@ -31,17 +31,17 @@ void
glm_mat4_transp_sse2(mat4 m, mat4 dest){
__m128 r0, r1, r2, r3;
r0 = _mm_load_ps(m[0]);
r1 = _mm_load_ps(m[1]);
r2 = _mm_load_ps(m[2]);
r3 = _mm_load_ps(m[3]);
r0 = glmm_load(m[0]);
r1 = glmm_load(m[1]);
r2 = glmm_load(m[2]);
r3 = glmm_load(m[3]);
_MM_TRANSPOSE4_PS(r0, r1, r2, r3);
_mm_store_ps(dest[0], r0);
_mm_store_ps(dest[1], r1);
_mm_store_ps(dest[2], r2);
_mm_store_ps(dest[3], r3);
glmm_store(dest[0], r0);
glmm_store(dest[1], r1);
glmm_store(dest[2], r2);
glmm_store(dest[3], r3);
}
CGLM_INLINE
@@ -51,36 +51,36 @@ glm_mat4_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
__m128 l0, l1, l2, l3, r;
l0 = _mm_load_ps(m1[0]);
l1 = _mm_load_ps(m1[1]);
l2 = _mm_load_ps(m1[2]);
l3 = _mm_load_ps(m1[3]);
l0 = glmm_load(m1[0]);
l1 = glmm_load(m1[1]);
l2 = glmm_load(m1[2]);
l3 = glmm_load(m1[3]);
r = _mm_load_ps(m2[0]);
_mm_store_ps(dest[0],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
r = _mm_load_ps(m2[1]);
_mm_store_ps(dest[1],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
r = _mm_load_ps(m2[2]);
_mm_store_ps(dest[2],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
r = glmm_load(m2[0]);
glmm_store(dest[0],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
r = glmm_load(m2[1]);
glmm_store(dest[1],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
r = glmm_load(m2[2]);
glmm_store(dest[2],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
r = _mm_load_ps(m2[3]);
_mm_store_ps(dest[3],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
r = glmm_load(m2[3]);
glmm_store(dest[3],
_mm_add_ps(_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 0), l0),
_mm_mul_ps(_mm_shuffle1_ps1(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps1(r, 2), l2),
_mm_mul_ps(_mm_shuffle1_ps1(r, 3), l3))));
}
CGLM_INLINE
@@ -88,18 +88,18 @@ void
glm_mat4_mulv_sse2(mat4 m, vec4 v, vec4 dest) {
__m128 x0, x1, x2;
x0 = _mm_load_ps(v);
x1 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[0]),
x0 = glmm_load(v);
x1 = _mm_add_ps(_mm_mul_ps(glmm_load(m[0]),
_mm_shuffle1_ps1(x0, 0)),
_mm_mul_ps(_mm_load_ps(m[1]),
_mm_mul_ps(glmm_load(m[1]),
_mm_shuffle1_ps1(x0, 1)));
x2 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[2]),
x2 = _mm_add_ps(_mm_mul_ps(glmm_load(m[2]),
_mm_shuffle1_ps1(x0, 2)),
_mm_mul_ps(_mm_load_ps(m[3]),
_mm_mul_ps(glmm_load(m[3]),
_mm_shuffle1_ps1(x0, 3)));
_mm_store_ps(dest, _mm_add_ps(x1, x2));
glmm_store(dest, _mm_add_ps(x1, x2));
}
CGLM_INLINE
@@ -108,10 +108,10 @@ glm_mat4_det_sse2(mat4 mat) {
__m128 r0, r1, r2, r3, x0, x1, x2;
/* 127 <- 0, [square] det(A) = det(At) */
r0 = _mm_load_ps(mat[0]); /* d c b a */
r1 = _mm_load_ps(mat[1]); /* h g f e */
r2 = _mm_load_ps(mat[2]); /* l k j i */
r3 = _mm_load_ps(mat[3]); /* p o n m */
r0 = glmm_load(mat[0]); /* d c b a */
r1 = glmm_load(mat[1]); /* h g f e */
r2 = glmm_load(mat[2]); /* l k j i */
r3 = glmm_load(mat[3]); /* p o n m */
/*
t[1] = j * p - n * l;
@@ -166,10 +166,10 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
x0, x1, x2, x3, x4, x5, x6, x7;
/* 127 <- 0 */
r0 = _mm_load_ps(mat[0]); /* d c b a */
r1 = _mm_load_ps(mat[1]); /* h g f e */
r2 = _mm_load_ps(mat[2]); /* l k j i */
r3 = _mm_load_ps(mat[3]); /* p o n m */
r0 = glmm_load(mat[0]); /* d c b a */
r1 = glmm_load(mat[1]); /* h g f e */
r2 = glmm_load(mat[2]); /* l k j i */
r3 = glmm_load(mat[3]); /* p o n m */
x0 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(3, 2, 3, 2)); /* p o l k */
x1 = _mm_shuffle1_ps(x0, 1, 3, 3, 3); /* l p p p */
@@ -275,10 +275,10 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 0, 1));
x0 = _mm_rcp_ps(x0);
_mm_store_ps(dest[0], _mm_mul_ps(v0, x0));
_mm_store_ps(dest[1], _mm_mul_ps(v1, x0));
_mm_store_ps(dest[2], _mm_mul_ps(v2, x0));
_mm_store_ps(dest[3], _mm_mul_ps(v3, x0));
glmm_store(dest[0], _mm_mul_ps(v0, x0));
glmm_store(dest[1], _mm_mul_ps(v1, x0));
glmm_store(dest[2], _mm_mul_ps(v2, x0));
glmm_store(dest[3], _mm_mul_ps(v3, x0));
}
CGLM_INLINE
@@ -290,10 +290,10 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
x0, x1, x2, x3, x4, x5, x6, x7;
/* 127 <- 0 */
r0 = _mm_load_ps(mat[0]); /* d c b a */
r1 = _mm_load_ps(mat[1]); /* h g f e */
r2 = _mm_load_ps(mat[2]); /* l k j i */
r3 = _mm_load_ps(mat[3]); /* p o n m */
r0 = glmm_load(mat[0]); /* d c b a */
r1 = glmm_load(mat[1]); /* h g f e */
r2 = glmm_load(mat[2]); /* l k j i */
r3 = glmm_load(mat[3]); /* p o n m */
x0 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(3, 2, 3, 2)); /* p o l k */
x1 = _mm_shuffle1_ps(x0, 1, 3, 3, 3); /* l p p p */
@@ -399,10 +399,10 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 0, 1));
x0 = _mm_div_ps(_mm_set1_ps(1.0f), x0);
_mm_store_ps(dest[0], _mm_mul_ps(v0, x0));
_mm_store_ps(dest[1], _mm_mul_ps(v1, x0));
_mm_store_ps(dest[2], _mm_mul_ps(v2, x0));
_mm_store_ps(dest[3], _mm_mul_ps(v3, x0));
glmm_store(dest[0], _mm_mul_ps(v0, x0));
glmm_store(dest[1], _mm_mul_ps(v1, x0));
glmm_store(dest[2], _mm_mul_ps(v2, x0));
glmm_store(dest[3], _mm_mul_ps(v3, x0));
}
#endif

61
include/cglm/simd/sse2/quat.h
View File

@@ -14,56 +14,33 @@
CGLM_INLINE
void
glm_quat_slerp_sse2(versor q,
versor r,
float t,
versor dest) {
/* https://en.wikipedia.org/wiki/Slerp */
float cosTheta, sinTheta, angle, a, b, c;
glm_quat_mul_sse2(versor p, versor q, versor dest) {
/*
+ (a1 b2 + b1 a2 + c1 d2 − d1 c2)i
+ (a1 c2 − b1 d2 + c1 a2 + d1 b2)j
+ (a1 d2 + b1 c2 − c1 b2 + d1 a2)k
a1 a2 − b1 b2 − c1 c2 − d1 d2
*/
__m128 xmm_q;
__m128 xp, xq, x0, r;
xmm_q = _mm_load_ps(q);
xp = glmm_load(p); /* 3 2 1 0 */
xq = glmm_load(q);
cosTheta = glm_vec4_dot(q, r);
if (cosTheta < 0.0f) {
_mm_store_ps(q,
_mm_xor_ps(xmm_q,
_mm_set1_ps(-0.f))) ;
r = _mm_mul_ps(_mm_shuffle1_ps1(xp, 3), xq);
cosTheta = -cosTheta;
}
x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 0), _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 0, 1, 2, 3)));
if (cosTheta >= 1.0f) {
_mm_store_ps(dest, xmm_q);
return;
}
x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 1), _mm_set_ps(-0.f, -0.f, 0.f, 0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 1, 0, 3, 2)));
sinTheta = sqrtf(1.0f - cosTheta * cosTheta);
x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 2), _mm_set_ps(-0.f, 0.f, 0.f, -0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 2, 3, 0, 1)));
c = 1.0f - t;
/* LERP */
if (sinTheta < 0.001f) {
_mm_store_ps(dest, _mm_add_ps(_mm_mul_ps(_mm_set1_ps(c),
xmm_q),
_mm_mul_ps(_mm_set1_ps(t),
_mm_load_ps(r))));
return;
}
/* SLERP */
angle = acosf(cosTheta);
a = sinf(c * angle);
b = sinf(t * angle);
_mm_store_ps(dest,
_mm_div_ps(_mm_add_ps(_mm_mul_ps(_mm_set1_ps(a),
xmm_q),
_mm_mul_ps(_mm_set1_ps(b),
_mm_load_ps(r))),
_mm_set1_ps(sinTheta)));
glmm_store(dest, r);
}
#endif
#endif /* cglm_quat_simd_h */