shweet/calamity-cobra-2
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

first commit

Browse Source
This commit is contained in:
shweet
2023-08-11 01:51:06 -04:00
commit 68746a654f
194 changed files with 33308 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

115
include/cglm/simd/sse2/affine.h Normal file
View File

@ -0,0 +1,115 @@
/*
* 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 "../../common.h"
#include "../intrin.h"
CGLM_INLINE
void
glm_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
/* D = R * L (Column-Major) */
glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
l = glmm_load(m1[0]);
r0 = glmm_load(m2[0]);
r1 = glmm_load(m2[1]);
r2 = glmm_load(m2[2]);
r3 = glmm_load(m2[3]);
v0 = _mm_mul_ps(glmm_splat_x(r0), l);
v1 = _mm_mul_ps(glmm_splat_x(r1), l);
v2 = _mm_mul_ps(glmm_splat_x(r2), l);
v3 = _mm_mul_ps(glmm_splat_x(r3), l);
l = glmm_load(m1[1]);
v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
v3 = glmm_fmadd(glmm_splat_y(r3), l, v3);
l = glmm_load(m1[2]);
v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
v3 = glmm_fmadd(glmm_splat_z(r3), l, v3);
l = glmm_load(m1[3]);
v3 = glmm_fmadd(glmm_splat_w(r3), l, v3);
glmm_store(dest[0], v0);
glmm_store(dest[1], v1);
glmm_store(dest[2], v2);
glmm_store(dest[3], v3);
}
CGLM_INLINE
void
glm_mul_rot_sse2(mat4 m1, mat4 m2, mat4 dest) {
/* D = R * L (Column-Major) */
glmm_128 l, r0, r1, r2, v0, v1, v2;
l = glmm_load(m1[0]);
r0 = glmm_load(m2[0]);
r1 = glmm_load(m2[1]);
r2 = glmm_load(m2[2]);
v0 = _mm_mul_ps(glmm_splat_x(r0), l);
v1 = _mm_mul_ps(glmm_splat_x(r1), l);
v2 = _mm_mul_ps(glmm_splat_x(r2), l);
l = glmm_load(m1[1]);
v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
l = glmm_load(m1[2]);
v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
glmm_store(dest[0], v0);
glmm_store(dest[1], v1);
glmm_store(dest[2], v2);
glmm_store(dest[3], glmm_load(m1[3]));
}
CGLM_INLINE
void
glm_inv_tr_sse2(mat4 mat) {
__m128 r0, r1, r2, r3, x0, x1, x2, x3, x4, x5;
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);
x2 = glmm_shuff1(r3, 0, 0, 0, 0);
x3 = glmm_shuff1(r3, 1, 1, 1, 1);
x4 = glmm_shuff1(r3, 2, 2, 2, 2);
x5 = _mm_set1_ps(-0.f);
x0 = glmm_fmadd(r0, x2, glmm_fmadd(r1, x3, _mm_mul_ps(r2, x4)));
x0 = _mm_xor_ps(x0, x5);
x0 = _mm_add_ps(x0, x1);
glmm_store(mat[0], r0);
glmm_store(mat[1], r1);
glmm_store(mat[2], r2);
glmm_store(mat[3], x0);
}
#endif
#endif /* cglm_affine_mat_sse2_h */

48
include/cglm/simd/sse2/mat2.h Normal file
View File

@ -0,0 +1,48 @@
/*
* Copyright (c), Recep Aslantas.
*
* MIT License (MIT), http://opensource.org/licenses/MIT
* Full license can be found in the LICENSE file
*/
#ifndef cglm_mat2_sse_h
#define cglm_mat2_sse_h
#if defined( __SSE__ ) || defined( __SSE2__ )
#include "../../common.h"
#include "../intrin.h"
CGLM_INLINE
void
glm_mat2_mul_sse2(mat2 m1, mat2 m2, mat2 dest) {
__m128 x0, x1, x2, x3, x4;
x1 = glmm_load(m1[0]); /* d c b a */
x2 = glmm_load(m2[0]); /* h g f e */
x3 = glmm_shuff1(x2, 2, 2, 0, 0);
x4 = glmm_shuff1(x2, 3, 3, 1, 1);
x0 = _mm_movelh_ps(x1, x1);
x2 = _mm_movehl_ps(x1, x1);
/*
dest[0][0] = a * e + c * f;
dest[0][1] = b * e + d * f;
dest[1][0] = a * g + c * h;
dest[1][1] = b * g + d * h;
*/
x0 = glmm_fmadd(x0, x3, _mm_mul_ps(x2, x4));
glmm_store(dest[0], x0);
}
CGLM_INLINE
void
glm_mat2_transp_sse2(mat2 m, mat2 dest) {
/* d c b a */
/* d b c a */
glmm_store(dest[0], glmm_shuff1(glmm_load(m[0]), 3, 1, 2, 0));
}
#endif
#endif /* cglm_mat2_sse_h */

76
include/cglm/simd/sse2/mat3.h Normal file
View File

@ -0,0 +1,76 @@
/*
* 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 "../../common.h"
#include "../intrin.h"
CGLM_INLINE
void
glm_mat3_mul_sse2(mat3 m1, mat3 m2, mat3 dest) {
__m128 l0, l1, l2, r0, r1, r2, x0, x1, x2, x3, x4, x5, x6, x7, x8, x9;
l0 = _mm_loadu_ps(m1[0]);
l1 = _mm_loadu_ps(&m1[1][1]);
r0 = _mm_loadu_ps(m2[0]);
r1 = _mm_loadu_ps(&m2[1][1]);
x8 = glmm_shuff1(l0, 0, 2, 1, 0); /* a00 a02 a01 a00 */
x1 = glmm_shuff1(r0, 3, 0, 0, 0); /* b10 b00 b00 b00 */
x2 = _mm_shuffle_ps(l0, l1, _MM_SHUFFLE(1, 0, 3, 3)); /* a12 a11 a10 a10 */
x3 = _mm_shuffle_ps(r0, r1, _MM_SHUFFLE(2, 0, 3, 1)); /* b20 b11 b10 b01 */
x0 = _mm_mul_ps(x8, x1);
x6 = glmm_shuff1(l0, 1, 0, 2, 1); /* a01 a00 a02 a01 */
x7 = glmm_shuff1(x3, 3, 3, 1, 1); /* b20 b20 b10 b10 */
l2 = _mm_load_ss(&m1[2][2]);
r2 = _mm_load_ss(&m2[2][2]);
x1 = _mm_mul_ps(x6, x7);
l2 = glmm_shuff1(l2, 0, 0, 1, 0); /* a22 a22 0.f a22 */
r2 = glmm_shuff1(r2, 0, 0, 1, 0); /* b22 b22 0.f b22 */
x4 = glmm_shuff1(x2, 0, 3, 2, 0); /* a10 a12 a11 a10 */
x5 = glmm_shuff1(x2, 2, 0, 3, 2); /* a11 a10 a12 a11 */
x6 = glmm_shuff1(x3, 2, 0, 0, 0); /* b11 b01 b01 b01 */
x2 = glmm_shuff1(r1, 3, 3, 0, 0); /* b21 b21 b11 b11 */
x8 = _mm_unpackhi_ps(x8, x4); /* a10 a00 a12 a02 */
x9 = _mm_unpackhi_ps(x7, x2); /* b21 b20 b21 b20 */
x0 = glmm_fmadd(x4, x6, x0);
x1 = glmm_fmadd(x5, x2, x1);
x2 = _mm_movehl_ps(l2, l1); /* a22 a22 a21 a20 */
x3 = glmm_shuff1(x2, 0, 2, 1, 0); /* a20 a22 a21 a20 */
x2 = glmm_shuff1(x2, 1, 0, 2, 1); /* a21 a20 a22 a21 */
x4 = _mm_shuffle_ps(r0, r1, _MM_SHUFFLE(1, 1, 2, 2)); /* b12 b12 b02 b02 */
x5 = glmm_shuff1(x4, 3, 0, 0, 0); /* b12 b02 b02 b02 */
x4 = _mm_movehl_ps(r2, x4); /* b22 b22 b12 b12 */
x0 = glmm_fmadd(x3, x5, x0);
x1 = glmm_fmadd(x2, x4, x1);
/*
Dot Product : dest[2][2] = a02 * b20 +
a12 * b21 +
a22 * b22 +
0 * 00 */
x2 = _mm_movelh_ps(x8, l2); /* 0.f a22 a12 a02 */
x3 = _mm_movelh_ps(x9, r2); /* 0.f b22 b21 b20 */
x2 = glmm_vdots(x2, x3);
_mm_storeu_ps(&dest[0][0], x0);
_mm_storeu_ps(&dest[1][1], x1);
_mm_store_ss (&dest[2][2], x2);
}
#endif
#endif /* cglm_mat3_sse_h */

434
include/cglm/simd/sse2/mat4.h Normal file
View File

@ -0,0 +1,434 @@
/*
* 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 "../../common.h"
#include "../intrin.h"
#define glm_mat4_inv_precise_sse2(mat, dest) glm_mat4_inv_sse2(mat, dest)
CGLM_INLINE
void
glm_mat4_scale_sse2(mat4 m, float s) {
__m128 x0;
x0 = _mm_set1_ps(s);
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
void
glm_mat4_transp_sse2(mat4 m, mat4 dest) {
__m128 r0, r1, r2, r3;
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);
glmm_store(dest[0], r0);
glmm_store(dest[1], r1);
glmm_store(dest[2], r2);
glmm_store(dest[3], r3);
}
CGLM_INLINE
void
glm_mat4_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
/* D = R * L (Column-Major) */
glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
l = glmm_load(m1[0]);
r0 = glmm_load(m2[0]);
r1 = glmm_load(m2[1]);
r2 = glmm_load(m2[2]);
r3 = glmm_load(m2[3]);
v0 = _mm_mul_ps(glmm_splat_x(r0), l);
v1 = _mm_mul_ps(glmm_splat_x(r1), l);
v2 = _mm_mul_ps(glmm_splat_x(r2), l);
v3 = _mm_mul_ps(glmm_splat_x(r3), l);
l = glmm_load(m1[1]);
v0 = glmm_fmadd(glmm_splat_y(r0), l, v0);
v1 = glmm_fmadd(glmm_splat_y(r1), l, v1);
v2 = glmm_fmadd(glmm_splat_y(r2), l, v2);
v3 = glmm_fmadd(glmm_splat_y(r3), l, v3);
l = glmm_load(m1[2]);
v0 = glmm_fmadd(glmm_splat_z(r0), l, v0);
v1 = glmm_fmadd(glmm_splat_z(r1), l, v1);
v2 = glmm_fmadd(glmm_splat_z(r2), l, v2);
v3 = glmm_fmadd(glmm_splat_z(r3), l, v3);
l = glmm_load(m1[3]);
v0 = glmm_fmadd(glmm_splat_w(r0), l, v0);
v1 = glmm_fmadd(glmm_splat_w(r1), l, v1);
v2 = glmm_fmadd(glmm_splat_w(r2), l, v2);
v3 = glmm_fmadd(glmm_splat_w(r3), l, v3);
glmm_store(dest[0], v0);
glmm_store(dest[1], v1);
glmm_store(dest[2], v2);
glmm_store(dest[3], v3);
}
CGLM_INLINE
void
glm_mat4_mulv_sse2(mat4 m, vec4 v, vec4 dest) {
__m128 x0, x1, m0, m1, m2, m3, v0, v1, v2, v3;
m0 = glmm_load(m[0]);
m1 = glmm_load(m[1]);
m2 = glmm_load(m[2]);
m3 = glmm_load(m[3]);
x0 = glmm_load(v);
v0 = glmm_splat_x(x0);
v1 = glmm_splat_y(x0);
v2 = glmm_splat_z(x0);
v3 = glmm_splat_w(x0);
x1 = _mm_mul_ps(m3, v3);
x1 = glmm_fmadd(m2, v2, x1);
x1 = glmm_fmadd(m1, v1, x1);
x1 = glmm_fmadd(m0, v0, x1);
glmm_store(dest, x1);
}
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 = 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;
t[2] = j * o - n * k;
t[3] = i * p - m * l;
t[4] = i * o - m * k;
*/
x0 = glmm_fnmadd(glmm_shuff1(r3, 0, 0, 1, 1), glmm_shuff1(r2, 2, 3, 2, 3),
_mm_mul_ps(glmm_shuff1(r2, 0, 0, 1, 1),
glmm_shuff1(r3, 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 = glmm_fnmadd(glmm_shuff1(r3, 0, 0, 2, 2), glmm_shuff1(r2, 1, 1, 3, 3),
_mm_mul_ps(glmm_shuff1(r2, 0, 0, 2, 2),
glmm_shuff1(r3, 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 = glmm_fnmadd(glmm_shuff1(r1, 1, 1, 2, 2), glmm_shuff1(x0, 3, 2, 2, 0),
_mm_mul_ps(glmm_shuff1(r1, 0, 0, 0, 1),
_mm_shuffle_ps(x1, x0, _MM_SHUFFLE(1, 0, 0, 0))));
x2 = glmm_fmadd(glmm_shuff1(r1, 2, 3, 3, 3),
_mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 2, 3, 1)),
x2);
x2 = _mm_xor_ps(x2, _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
return glmm_hadd(_mm_mul_ps(x2, r0));
}
CGLM_INLINE
void
glm_mat4_inv_fast_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, x8, x9;
x8 = _mm_set_ps(-0.f, 0.f, -0.f, 0.f);
x9 = glmm_shuff1(x8, 2, 1, 2, 1);
/* 127 <- 0 */
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_movehl_ps(r3, r2); /* p o l k */
x3 = _mm_movelh_ps(r2, r3); /* n m j i */
x1 = glmm_shuff1(x0, 1, 3, 3 ,3); /* l p p p */
x2 = glmm_shuff1(x0, 0, 2, 2, 2); /* k o o o */
x4 = glmm_shuff1(x3, 1, 3, 3, 3); /* j n n n */
x7 = glmm_shuff1(x3, 0, 2, 2, 2); /* i m m m */
x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0)); /* e e i i */
x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1)); /* f f j j */
x3 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(2, 2, 2, 2)); /* g g k k */
x0 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(3, 3, 3, 3)); /* h h l l */
t0 = _mm_mul_ps(x3, x1);
t1 = _mm_mul_ps(x5, x1);
t2 = _mm_mul_ps(x5, x2);
t3 = _mm_mul_ps(x6, x1);
t4 = _mm_mul_ps(x6, x2);
t5 = _mm_mul_ps(x6, x4);
/* 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 = glmm_fnmadd(x2, x0, t0);
/* 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 = glmm_fnmadd(x4, x0, t1);
/* 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 = glmm_fnmadd(x4, x3, t2);
/* 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 = glmm_fnmadd(x7, x0, t3);
/* 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 = glmm_fnmadd(x7, x3, t4);
/* 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 = glmm_fnmadd(x7, x5, t5);
x4 = _mm_movelh_ps(r0, r1); /* f e b a */
x5 = _mm_movehl_ps(r1, r0); /* h g d c */
x0 = glmm_shuff1(x4, 0, 0, 0, 2); /* a a a e */
x1 = glmm_shuff1(x4, 1, 1, 1, 3); /* b b b f */
x2 = glmm_shuff1(x5, 0, 0, 0, 2); /* c c c g */
x3 = glmm_shuff1(x5, 1, 1, 1, 3); /* d d d h */
v2 = _mm_mul_ps(x0, t1);
v1 = _mm_mul_ps(x0, t0);
v3 = _mm_mul_ps(x0, t2);
v0 = _mm_mul_ps(x1, t0);
v2 = glmm_fnmadd(x1, t3, v2);
v3 = glmm_fnmadd(x1, t4, v3);
v0 = glmm_fnmadd(x2, t1, v0);
v1 = glmm_fnmadd(x2, t3, v1);
v3 = glmm_fmadd(x2, t5, v3);
v0 = glmm_fmadd(x3, t2, v0);
v2 = glmm_fmadd(x3, t5, v2);
v1 = glmm_fmadd(x3, t4, v1);
/*
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_xor_ps(v0, x8);
/*
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_xor_ps(v2, x8);
/*
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_xor_ps(v1, x9);
/*
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_xor_ps(v3, x9);
/* 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_rcp_ps(glmm_vhadd(_mm_mul_ps(x0, r0)));
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
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, x8, x9;
x8 = _mm_set_ps(-0.f, 0.f, -0.f, 0.f);
x9 = glmm_shuff1(x8, 2, 1, 2, 1);
/* 127 <- 0 */
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_movehl_ps(r3, r2); /* p o l k */
x3 = _mm_movelh_ps(r2, r3); /* n m j i */
x1 = glmm_shuff1(x0, 1, 3, 3 ,3); /* l p p p */
x2 = glmm_shuff1(x0, 0, 2, 2, 2); /* k o o o */
x4 = glmm_shuff1(x3, 1, 3, 3, 3); /* j n n n */
x7 = glmm_shuff1(x3, 0, 2, 2, 2); /* i m m m */
x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0)); /* e e i i */
x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1)); /* f f j j */
x3 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(2, 2, 2, 2)); /* g g k k */
x0 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(3, 3, 3, 3)); /* h h l l */
t0 = _mm_mul_ps(x3, x1);
t1 = _mm_mul_ps(x5, x1);
t2 = _mm_mul_ps(x5, x2);
t3 = _mm_mul_ps(x6, x1);
t4 = _mm_mul_ps(x6, x2);
t5 = _mm_mul_ps(x6, x4);
/* 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 = glmm_fnmadd(x2, x0, t0);
/* 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 = glmm_fnmadd(x4, x0, t1);
/* 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 = glmm_fnmadd(x4, x3, t2);
/* 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 = glmm_fnmadd(x7, x0, t3);
/* 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 = glmm_fnmadd(x7, x3, t4);
/* 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 = glmm_fnmadd(x7, x5, t5);
x4 = _mm_movelh_ps(r0, r1); /* f e b a */
x5 = _mm_movehl_ps(r1, r0); /* h g d c */
x0 = glmm_shuff1(x4, 0, 0, 0, 2); /* a a a e */
x1 = glmm_shuff1(x4, 1, 1, 1, 3); /* b b b f */
x2 = glmm_shuff1(x5, 0, 0, 0, 2); /* c c c g */
x3 = glmm_shuff1(x5, 1, 1, 1, 3); /* d d d h */
v2 = _mm_mul_ps(x0, t1);
v1 = _mm_mul_ps(x0, t0);
v3 = _mm_mul_ps(x0, t2);
v0 = _mm_mul_ps(x1, t0);
v2 = glmm_fnmadd(x1, t3, v2);
v3 = glmm_fnmadd(x1, t4, v3);
v0 = glmm_fnmadd(x2, t1, v0);
v1 = glmm_fnmadd(x2, t3, v1);
v3 = glmm_fmadd(x2, t5, v3);
v0 = glmm_fmadd(x3, t2, v0);
v2 = glmm_fmadd(x3, t5, v2);
v1 = glmm_fmadd(x3, t4, v1);
/*
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_xor_ps(v0, x8);
/*
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_xor_ps(v2, x8);
/*
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_xor_ps(v1, x9);
/*
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_xor_ps(v3, x9);
/* 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_div_ps(_mm_set1_ps(1.0f), glmm_vhadd(_mm_mul_ps(x0, r0)));
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
#endif /* cglm_mat_sse_h */

54
include/cglm/simd/sse2/quat.h Normal file
View File

@ -0,0 +1,54 @@
/*
* 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
#if defined( __SSE__ ) || defined( __SSE2__ )
#include "../../common.h"
#include "../intrin.h"
CGLM_INLINE
void
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 xp, xq, x1, x2, x3, r, x, y, z;
xp = glmm_load(p); /* 3 2 1 0 */
xq = glmm_load(q);
x1 = _mm_set_ps(-0.f, 0.f, -0.f, 0.f); /* TODO: _mm_set1_ss() + shuff ? */
r = _mm_mul_ps(glmm_splat_w(xp), xq);
x2 = _mm_unpackhi_ps(x1, x1);
x3 = glmm_shuff1(x1, 3, 2, 0, 1);
x = glmm_splat_x(xp);
y = glmm_splat_y(xp);
z = glmm_splat_z(xp);
x = _mm_xor_ps(x, x1);
y = _mm_xor_ps(y, x2);
z = _mm_xor_ps(z, x3);
x1 = glmm_shuff1(xq, 0, 1, 2, 3);
x2 = glmm_shuff1(xq, 1, 0, 3, 2);
x3 = glmm_shuff1(xq, 2, 3, 0, 1);
r = glmm_fmadd(x, x1, r);
r = glmm_fmadd(y, x2, r);
r = glmm_fmadd(z, x3, r);
glmm_store(dest, r);
}
#endif
#endif /* cglm_quat_simd_h */