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

re-organize folder structure

Browse Source 
* move SIMD instructions (SSE2, AVX) to arch folder
This commit is contained in:
Recep Aslantas
2017-02-13 12:18:52 +03:00
parent f660170497
commit 595ad42cb8
14 changed files with 22 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

63
include/arch/simd/cglm-affine-mat-avx.h Normal file
View File

@@ -0,0 +1,63 @@
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_affine_mat_avx_h
#define cglm_affine_mat_avx_h
#ifdef __AVX__
#include "../../cglm-common.h"
#include "cglm-intrin.h"
#include <immintrin.h>
CGLM_INLINE
void
glm_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
/* D = R * L (Column-Major) */
__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 */
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 */
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 */
/* f f f f a a a a */
/* g g g g c c c c */
/* e e e e b b b b */
y7 = _mm256_permute_ps(y0, 0b10101010);
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)));
/* n n n n i i i i */
/* p p p p k k k k */
/* m m m m j j j j */
/* o o o o l l l l */
y6 = _mm256_permutevar_ps(y1, _mm256_set_epi32(1, 1, 1, 1, 0, 0, 0, 0));
y7 = _mm256_permutevar_ps(y1, _mm256_set_epi32(3, 3, 3, 3, 2, 2, 2, 2));
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))));
}
#endif
#endif /* cglm_affine_mat_avx_h */

79
include/arch/simd/cglm-affine-mat-sse2.h Normal file
View File

@@ -0,0 +1,79 @@
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_affine_mat_sse2_h
#define cglm_affine_mat_sse2_h
#if defined( __SSE__ ) || defined( __SSE2__ )
#include "../../cglm-common.h"
#include "cglm-intrin.h"
CGLM_INLINE
void
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]);
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 = _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 = _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 = _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))));
}
CGLM_INLINE
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);
_MM_TRANSPOSE4_PS(r0, r1, r2, x1);
x0 = _mm_add_ps(_mm_mul_ps(r0, _mm_shuffle1_ps(r3, 0, 0, 0, 0)),
_mm_mul_ps(r1, _mm_shuffle1_ps(r3, 1, 1, 1, 1)));
x0 = _mm_add_ps(x0, _mm_mul_ps(r2, _mm_shuffle1_ps(r3, 2, 2, 2, 2)));
x0 = _mm_xor_ps(x0, _mm_set1_ps(-0.f));
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);
}
#endif
#endif /* cglm_affine_mat_sse2_h */

36
include/arch/simd/cglm-intrin.h Normal file
View File

@@ -0,0 +1,36 @@
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_intrin_h
#define cglm_intrin_h
#if defined( _WIN32 )
# if (defined(_M_AMD64) || defined(_M_X64)) || _M_IX86_FP == 2
# define __SSE2__
# elif _M_IX86_FP == 1
# define __SSE__
# endif
#endif
#if defined( __SSE__ ) || defined( __SSE2__ )
#include <xmmintrin.h>
#include <emmintrin.h>
/* float */
#define _mm_shuffle1_ps(a, z, y, x, w) \
_mm_shuffle_ps(a, a, _MM_SHUFFLE(z, y, x, w))
#define _mm_shuffle1_ps1(a, x) \
_mm_shuffle_ps(a, a, _MM_SHUFFLE(x, x, x, x))
#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)
#endif
#endif /* cglm_intrin_h */

65
include/arch/simd/cglm-mat-simd-avx.h Normal file
View File

@@ -0,0 +1,65 @@
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_mat_simd_avx_h
#define cglm_mat_simd_avx_h
#ifdef __AVX__
#include "../../cglm-common.h"
#include "cglm-intrin.h"
#include <immintrin.h>
CGLM_INLINE
void
glm_mat4_mul_avx(mat4 m1, mat4 m2, mat4 dest) {
/* D = R * L (Column-Major) */
__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 */
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 */
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 */
/* f f f f a a a a */
/* h h h h c c c c */
/* e e e e b b b b */
/* g g g g d d d d */
y6 = _mm256_permutevar_ps(y0, _mm256_set_epi32(1, 1, 1, 1, 0, 0, 0, 0));
y7 = _mm256_permutevar_ps(y0, _mm256_set_epi32(3, 3, 3, 3, 2, 2, 2, 2));
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))));
/* n n n n i i i i */
/* p p p p k k k k */
/* m m m m j j j j */
/* o o o o l l l l */
y6 = _mm256_permutevar_ps(y1, _mm256_set_epi32(1, 1, 1, 1, 0, 0, 0, 0));
y7 = _mm256_permutevar_ps(y1, _mm256_set_epi32(3, 3, 3, 3, 2, 2, 2, 2));
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))));
}
#endif
#endif /* cglm_mat_simd_avx_h */

407
include/arch/simd/cglm-mat-simd-sse2.h Normal file
View File

@@ -0,0 +1,407 @@
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_mat_sse_h
#define cglm_mat_sse_h
#if defined( __SSE__ ) || defined( __SSE2__ )
#include "../../cglm-common.h"
#include "cglm-intrin.h"
CGLM_INLINE
void
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));
}
CGLM_INLINE
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]);
_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);
}
CGLM_INLINE
void
glm_mat4_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]);
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 = _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))));
}
CGLM_INLINE
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]),
_mm_shuffle1_ps1(x0, 0)),
_mm_mul_ps(_mm_load_ps(m[1]),
_mm_shuffle1_ps1(x0, 1)));
x2 = _mm_add_ps(_mm_mul_ps(_mm_load_ps(m[2]),
_mm_shuffle1_ps1(x0, 2)),
_mm_mul_ps(_mm_load_ps(m[3]),
_mm_shuffle1_ps1(x0, 3)));
_mm_store_ps(dest, _mm_add_ps(x1, x2));
}
CGLM_INLINE
float
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 */
/*
t[1] = j * p - n * l;
t[2] = j * o - n * k;
t[3] = i * p - m * l;
t[4] = i * o - m * k;
*/
x0 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r2, 0, 0, 1, 1),
_mm_shuffle1_ps(r3, 2, 3, 2, 3)),
_mm_mul_ps(_mm_shuffle1_ps(r3, 0, 0, 1, 1),
_mm_shuffle1_ps(r2, 2, 3, 2, 3)));
/*
t[0] = k * p - o * l;
t[0] = k * p - o * l;
t[5] = i * n - m * j;
t[5] = i * n - m * j;
*/
x1 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r2, 0, 0, 2, 2),
_mm_shuffle1_ps(r3, 1, 1, 3, 3)),
_mm_mul_ps(_mm_shuffle1_ps(r3, 0, 0, 2, 2),
_mm_shuffle1_ps(r2, 1, 1, 3, 3)));
/*
a * (f * t[0] - g * t[1] + h * t[2])
- b * (e * t[0] - g * t[3] + h * t[4])
+ c * (e * t[1] - f * t[3] + h * t[5])
- d * (e * t[2] - f * t[4] + g * t[5])
*/
x2 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r1, 0, 0, 0, 1),
_mm_shuffle_ps(x1, x0, _MM_SHUFFLE(1, 0, 0, 0))),
_mm_mul_ps(_mm_shuffle1_ps(r1, 1, 1, 2, 2),
_mm_shuffle1_ps(x0, 3, 2, 2, 0)));
x2 = _mm_add_ps(x2,
_mm_mul_ps(_mm_shuffle1_ps(r1, 2, 3, 3, 3),
_mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 2, 3, 1))));
x2 = _mm_xor_ps(x2, _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
x0 = _mm_mul_ps(r0, x2);
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 2, 3));
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 3, 3, 1));
return _mm_cvtss_f32(x0);
}
CGLM_INLINE
void
glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
__m128 r0, r1, r2, r3,
v0, v1, v2, v3,
t0, t1, t2, t3, t4, t5,
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 */
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 */
x2 = _mm_shuffle1_ps(x0, 0, 2, 2, 2); /* k o o o */
x0 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(3, 3, 3, 3)); /* h h l l */
x3 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(2, 2, 2, 2)); /* g g k k */
/* t1[0] = k * p - o * l;
t1[0] = k * p - o * l;
t2[0] = g * p - o * h;
t3[0] = g * l - k * h; */
t0 = _mm_sub_ps(_mm_mul_ps(x3, x1), _mm_mul_ps(x2, x0));
x4 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(2, 1, 2, 1)); /* o n k j */
x4 = _mm_shuffle1_ps(x4, 0, 2, 2, 2); /* j n n n */
x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1)); /* f f j j */
/* t1[1] = j * p - n * l;
t1[1] = j * p - n * l;
t2[1] = f * p - n * h;
t3[1] = f * l - j * h; */
t1 = _mm_sub_ps(_mm_mul_ps(x5, x1), _mm_mul_ps(x4, x0));
/* t1[2] = j * o - n * k
t1[2] = j * o - n * k;
t2[2] = f * o - n * g;
t3[2] = f * k - j * g; */
t2 = _mm_sub_ps(_mm_mul_ps(x5, x2), _mm_mul_ps(x4, x3));
x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0)); /* e e i i */
x7 = _mm_shuffle2_ps(r3, r2, 0, 0, 0, 0, 2, 0, 0, 0); /* i m m m */
/* t1[3] = i * p - m * l;
t1[3] = i * p - m * l;
t2[3] = e * p - m * h;
t3[3] = e * l - i * h; */
t3 = _mm_sub_ps(_mm_mul_ps(x6, x1), _mm_mul_ps(x7, x0));
/* t1[4] = i * o - m * k;
t1[4] = i * o - m * k;
t2[4] = e * o - m * g;
t3[4] = e * k - i * g; */
t4 = _mm_sub_ps(_mm_mul_ps(x6, x2), _mm_mul_ps(x7, x3));
/* t1[5] = i * n - m * j;
t1[5] = i * n - m * j;
t2[5] = e * n - m * f;
t3[5] = e * j - i * f; */
t5 = _mm_sub_ps(_mm_mul_ps(x6, x4), _mm_mul_ps(x7, x5));
x0 = _mm_shuffle2_ps(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
x1 = _mm_shuffle2_ps(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
x2 = _mm_shuffle2_ps(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
x3 = _mm_shuffle2_ps(r1, r0, 3, 3, 3, 3, 2, 2, 2, 0); /* d d d h */
/*
dest[0][0] = f * t1[0] - g * t1[1] + h * t1[2];
dest[0][1] =-(b * t1[0] - c * t1[1] + d * t1[2]);
dest[0][2] = b * t2[0] - c * t2[1] + d * t2[2];
dest[0][3] =-(b * t3[0] - c * t3[1] + d * t3[2]); */
v0 = _mm_add_ps(_mm_mul_ps(x3, t2),
_mm_sub_ps(_mm_mul_ps(x1, t0),
_mm_mul_ps(x2, t1)));
v0 = _mm_xor_ps(v0, _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
/*
dest[1][0] =-(e * t1[0] - g * t1[3] + h * t1[4]);
dest[1][1] = a * t1[0] - c * t1[3] + d * t1[4];
dest[1][2] =-(a * t2[0] - c * t2[3] + d * t2[4]);
dest[1][3] = a * t3[0] - c * t3[3] + d * t3[4]; */
v1 = _mm_add_ps(_mm_mul_ps(x3, t4),
_mm_sub_ps(_mm_mul_ps(x0, t0),
_mm_mul_ps(x2, t3)));
v1 = _mm_xor_ps(v1, _mm_set_ps(0.f, -0.f, 0.f, -0.f));
/*
dest[2][0] = e * t1[1] - f * t1[3] + h * t1[5];
dest[2][1] =-(a * t1[1] - b * t1[3] + d * t1[5]);
dest[2][2] = a * t2[1] - b * t2[3] + d * t2[5];
dest[2][3] =-(a * t3[1] - b * t3[3] + d * t3[5]);*/
v2 = _mm_add_ps(_mm_mul_ps(x3, t5),
_mm_sub_ps(_mm_mul_ps(x0, t1),
_mm_mul_ps(x1, t3)));
v2 = _mm_xor_ps(v2, _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
/*
dest[3][0] =-(e * t1[2] - f * t1[4] + g * t1[5]);
dest[3][1] = a * t1[2] - b * t1[4] + c * t1[5];
dest[3][2] =-(a * t2[2] - b * t2[4] + c * t2[5]);
dest[3][3] = a * t3[2] - b * t3[4] + c * t3[5]; */
v3 = _mm_add_ps(_mm_mul_ps(x2, t5),
_mm_sub_ps(_mm_mul_ps(x0, t2),
_mm_mul_ps(x1, t4)));
v3 = _mm_xor_ps(v3, _mm_set_ps(0.f, -0.f, 0.f, -0.f));
/* determinant */
x0 = _mm_shuffle_ps(v0, v1, _MM_SHUFFLE(0, 0, 0, 0));
x1 = _mm_shuffle_ps(v2, v3, _MM_SHUFFLE(0, 0, 0, 0));
x0 = _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 0, 2, 0));
x0 = _mm_mul_ps(x0, r0);
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 2, 3));
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));
}
CGLM_INLINE
void
glm_mat4_inv_precise_sse2(mat4 mat, mat4 dest) {
__m128 r0, r1, r2, r3,
v0, v1, v2, v3,
t0, t1, t2, t3, t4, t5,
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 */
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 */
x2 = _mm_shuffle1_ps(x0, 0, 2, 2, 2); /* k o o o */
x0 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(3, 3, 3, 3)); /* h h l l */
x3 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(2, 2, 2, 2)); /* g g k k */
/* t1[0] = k * p - o * l;
t1[0] = k * p - o * l;
t2[0] = g * p - o * h;
t3[0] = g * l - k * h; */
t0 = _mm_sub_ps(_mm_mul_ps(x3, x1), _mm_mul_ps(x2, x0));
x4 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(2, 1, 2, 1)); /* o n k j */
x4 = _mm_shuffle1_ps(x4, 0, 2, 2, 2); /* j n n n */
x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1)); /* f f j j */
/* t1[1] = j * p - n * l;
t1[1] = j * p - n * l;
t2[1] = f * p - n * h;
t3[1] = f * l - j * h; */
t1 = _mm_sub_ps(_mm_mul_ps(x5, x1), _mm_mul_ps(x4, x0));
/* t1[2] = j * o - n * k
t1[2] = j * o - n * k;
t2[2] = f * o - n * g;
t3[2] = f * k - j * g; */
t2 = _mm_sub_ps(_mm_mul_ps(x5, x2), _mm_mul_ps(x4, x3));
x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0)); /* e e i i */
x7 = _mm_shuffle2_ps(r3, r2, 0, 0, 0, 0, 2, 0, 0, 0); /* i m m m */
/* t1[3] = i * p - m * l;
t1[3] = i * p - m * l;
t2[3] = e * p - m * h;
t3[3] = e * l - i * h; */
t3 = _mm_sub_ps(_mm_mul_ps(x6, x1), _mm_mul_ps(x7, x0));
/* t1[4] = i * o - m * k;
t1[4] = i * o - m * k;
t2[4] = e * o - m * g;
t3[4] = e * k - i * g; */
t4 = _mm_sub_ps(_mm_mul_ps(x6, x2), _mm_mul_ps(x7, x3));
/* t1[5] = i * n - m * j;
t1[5] = i * n - m * j;
t2[5] = e * n - m * f;
t3[5] = e * j - i * f; */
t5 = _mm_sub_ps(_mm_mul_ps(x6, x4), _mm_mul_ps(x7, x5));
x0 = _mm_shuffle2_ps(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
x1 = _mm_shuffle2_ps(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
x2 = _mm_shuffle2_ps(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
x3 = _mm_shuffle2_ps(r1, r0, 3, 3, 3, 3, 2, 2, 2, 0); /* d d d h */
/*
dest[0][0] = f * t1[0] - g * t1[1] + h * t1[2];
dest[0][1] =-(b * t1[0] - c * t1[1] + d * t1[2]);
dest[0][2] = b * t2[0] - c * t2[1] + d * t2[2];
dest[0][3] =-(b * t3[0] - c * t3[1] + d * t3[2]); */
v0 = _mm_add_ps(_mm_mul_ps(x3, t2),
_mm_sub_ps(_mm_mul_ps(x1, t0),
_mm_mul_ps(x2, t1)));
v0 = _mm_xor_ps(v0, _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
/*
dest[1][0] =-(e * t1[0] - g * t1[3] + h * t1[4]);
dest[1][1] = a * t1[0] - c * t1[3] + d * t1[4];
dest[1][2] =-(a * t2[0] - c * t2[3] + d * t2[4]);
dest[1][3] = a * t3[0] - c * t3[3] + d * t3[4]; */
v1 = _mm_add_ps(_mm_mul_ps(x3, t4),
_mm_sub_ps(_mm_mul_ps(x0, t0),
_mm_mul_ps(x2, t3)));
v1 = _mm_xor_ps(v1, _mm_set_ps(0.f, -0.f, 0.f, -0.f));
/*
dest[2][0] = e * t1[1] - f * t1[3] + h * t1[5];
dest[2][1] =-(a * t1[1] - b * t1[3] + d * t1[5]);
dest[2][2] = a * t2[1] - b * t2[3] + d * t2[5];
dest[2][3] =-(a * t3[1] - b * t3[3] + d * t3[5]);*/
v2 = _mm_add_ps(_mm_mul_ps(x3, t5),
_mm_sub_ps(_mm_mul_ps(x0, t1),
_mm_mul_ps(x1, t3)));
v2 = _mm_xor_ps(v2, _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
/*
dest[3][0] =-(e * t1[2] - f * t1[4] + g * t1[5]);
dest[3][1] = a * t1[2] - b * t1[4] + c * t1[5];
dest[3][2] =-(a * t2[2] - b * t2[4] + c * t2[5]);
dest[3][3] = a * t3[2] - b * t3[4] + c * t3[5]; */
v3 = _mm_add_ps(_mm_mul_ps(x2, t5),
_mm_sub_ps(_mm_mul_ps(x0, t2),
_mm_mul_ps(x1, t4)));
v3 = _mm_xor_ps(v3, _mm_set_ps(0.f, -0.f, 0.f, -0.f));
/* determinant */
x0 = _mm_shuffle_ps(v0, v1, _MM_SHUFFLE(0, 0, 0, 0));
x1 = _mm_shuffle_ps(v2, v3, _MM_SHUFFLE(0, 0, 0, 0));
x0 = _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 0, 2, 0));
x0 = _mm_mul_ps(x0, r0);
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 2, 3));
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));
}
#endif
#endif /* cglm_mat_sse_h */

61
include/arch/simd/cglm-mat3-simd-sse2.h Normal file
View File

@@ -0,0 +1,61 @@
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_mat3_sse_h
#define cglm_mat3_sse_h
#if defined( __SSE__ ) || defined( __SSE2__ )
#include "../../cglm-common.h"
#include "cglm-intrin.h"
CGLM_INLINE
void
glm_mat3_mul_sse2(mat3 m1, mat3 m2, mat3 dest) {
__m128 l0, l1, l2;
__m128 r0, r1, r2;
__m128 x0, x1, x2;
l0 = _mm_loadu_ps(m1[0]);
l1 = _mm_loadu_ps(&m1[1][1]);
l2 = _mm_set1_ps(m1[2][2]);
r0 = _mm_loadu_ps(m2[0]);
r1 = _mm_loadu_ps(&m2[1][1]);
r2 = _mm_set1_ps(m2[2][2]);
x1 = _mm_shuffle2_ps(l0, l1, 1, 0, 3, 3, 0, 3, 2, 0);
x2 = _mm_shuffle2_ps(l1, l2, 0, 0, 3, 2, 0, 2, 1, 0);
x0 = _mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps(l0, 0, 2, 1, 0),
_mm_shuffle1_ps(r0, 3, 0, 0, 0)),
_mm_mul_ps(x1,
_mm_shuffle2_ps(r0, r1, 0, 0, 1, 1, 2, 0, 0, 0)));
x0 = _mm_add_ps(x0,
_mm_mul_ps(x2,
_mm_shuffle2_ps(r0, r1, 1, 1, 2, 2, 2, 0, 0, 0)));
_mm_storeu_ps(dest[0], x0);
x0 = _mm_add_ps(_mm_mul_ps(_mm_shuffle1_ps(l0, 1, 0, 2, 1),
_mm_shuffle_ps(r0, r1, _MM_SHUFFLE(2, 2, 3, 3))),
_mm_mul_ps(_mm_shuffle1_ps(x1, 1, 0, 2, 1),
_mm_shuffle1_ps(r1, 3, 3, 0, 0)));
x0 = _mm_add_ps(x0,
_mm_mul_ps(_mm_shuffle1_ps(x2, 1, 0, 2, 1),
_mm_shuffle_ps(r1, r2, _MM_SHUFFLE(0, 0, 1, 1))));
_mm_storeu_ps(&dest[1][1], x0);
dest[2][2] = m1[0][2] * m2[2][0]
+ m1[1][2] * m2[2][1]
+ m1[2][2] * m2[2][2];
}
#endif
#endif /* cglm_mat3_sse_h */

67
include/arch/simd/cglm-quat-simd.h Normal file
View File

@@ -0,0 +1,67 @@
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_quat_simd_h
#define cglm_quat_simd_h
#include "../../cglm-common.h"
#include "cglm-intrin.h"
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;
__m128 xmm_q;
xmm_q = _mm_load_ps(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))) ;
cosTheta = -cosTheta;
}
if (cosTheta >= 1.0f) {
_mm_store_ps(dest, xmm_q);
return;
}
sinTheta = sqrtf(1.0f - cosTheta * cosTheta);
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)));
}
#endif /* cglm_quat_simd_h */