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

fix matrix inverse simd

Browse Source
This commit is contained in:
Recep Aslantas
2016-09-19 00:19:07 +03:00
parent f280d3675e
commit 0ad9992a99
1 changed files with 102 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

167
include/cglm-mat-simd.h
View File

@@ -63,6 +63,7 @@ float
glm_mat4_det_sse2(mat4 mat) { glm_mat4_det_sse2(mat4 mat) {
__m128 v0, dt, t0, t1, t2, t3, t4, r0, r1, r2, r3; __m128 v0, dt, t0, t1, t2, t3, t4, r0, r1, r2, r3;
r0 = _mm_load_ps(mat[0]); r0 = _mm_load_ps(mat[0]);
r1 = _mm_load_ps(mat[1]); r1 = _mm_load_ps(mat[1]);
r2 = _mm_load_ps(mat[2]); r2 = _mm_load_ps(mat[2]);
@@ -97,89 +98,125 @@ glm_mat4_det_sse2(mat4 mat) {
CGLM_INLINE CGLM_INLINE
void void
glm_mat4_inv_see2(mat4 mat, mat4 dest) { glm_mat4_inv_see2(mat4 mat, mat4 dest) {
__m128 v0, v1, v2, v3, dt, t0, t1, t2, t3, t4, r0, r1, r2, r3; __m128 r0, r1, r2, r3,
v0, v1, v2, v3,
t0, t1, t2, t3, t4, t5,
x0, x1, x2, x3, x4, x5, x6, x7;
r0 = _mm_load_ps(mat[0]); /* 127 <- 0 */
r1 = _mm_load_ps(mat[1]); r0 = _mm_load_ps(mat[0]); /* d c b a */
r2 = _mm_load_ps(mat[2]); r1 = _mm_load_ps(mat[1]); /* h g f e */
r3 = _mm_load_ps(mat[3]); r2 = _mm_load_ps(mat[2]); /* l k j i */
r3 = _mm_load_ps(mat[3]); /* p o n m */
t3 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r2, 2, 1, 1, 0), x0 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(3, 2, 3, 2)); /* p o l k */
_mm_shuffle1_ps(r3, 3, 3, 2, 3)), x1 = _mm_shuffle1_ps(x0, 1, 3, 3, 3); /* l p p p */
_mm_mul_ps(_mm_shuffle1_ps(r3, 2, 1, 1, 0), x2 = _mm_shuffle1_ps(x0, 0, 2, 2, 2); /* k o o o */
_mm_shuffle1_ps(r2, 3, 3, 2, 3))); 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 */
t4 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r2, 0, 0, 0, 0), /* t1[0] = k * p - o * l;
_mm_shuffle1_ps(r3, 2, 1, 2, 1)), t1[0] = k * p - o * l;
_mm_mul_ps(_mm_shuffle1_ps(r3, 0, 0, 0, 0), t2[0] = g * p - o * h;
_mm_shuffle1_ps(r2, 2, 1, 2, 1))); t3[0] = g * l - k * h; */
t0 = _mm_sub_ps(_mm_mul_ps(x3, x1), _mm_mul_ps(x2, x0));
t0 = _mm_shuffle1_ps(t3, 3, 3, 2, 1); x4 = _mm_shuffle_ps(r2, r3, _MM_SHUFFLE(2, 1, 2, 1)); /* o n k j */
t1 = _mm_shuffle2_ps(t3, t4, 1, 1, 2, 0, 1, 0, 0, 3); x4 = _mm_shuffle1_ps(x4, 0, 2, 2, 2); /* j n n n */
t2 = _mm_shuffle2_ps(t3, t4, 0, 1, 1, 1, 0, 2, 3, 3); x5 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(1, 1, 1, 1)); /* f f j j */
v0 = _mm_mul_ps(_mm_shuffle1_ps(r1, 1, 0, 0, 0), t0); /* t1[1] = j * p - n * l;
v0 = _mm_sub_ps(v0, _mm_mul_ps(_mm_shuffle1_ps(r1, 2, 2, 1, 1), t1)); t1[1] = j * p - n * l;
v0 = _mm_add_ps(v0, _mm_mul_ps(_mm_shuffle1_ps(r1, 3, 3, 3, 2), t2)); t2[1] = f * p - n * h;
v0 = _mm_xor_ps(v0, _mm_set_ps(0.f, -0.f, 0.f, -0.f)); t3[1] = f * l - j * h; */
t1 = _mm_sub_ps(_mm_mul_ps(x5, x1), _mm_mul_ps(x4, x0));
v1 = _mm_mul_ps(_mm_shuffle1_ps(r0, 1, 0, 0, 0), t0); /* t1[2] = j * o - n * k
v1 = _mm_sub_ps(v1, _mm_mul_ps(_mm_shuffle1_ps(r0, 2, 2, 1, 1), t1)); t1[2] = j * o - n * k;
v1 = _mm_add_ps(v1, _mm_mul_ps(_mm_shuffle1_ps(r0, 3, 3, 3, 2), t2)); t2[2] = f * o - n * g;
v1 = _mm_xor_ps(v1, _mm_set_ps(-0.f, 0.f, -0.f, 0.f)); t3[2] = f * k - j * g; */
t2 = _mm_sub_ps(_mm_mul_ps(x5, x2), _mm_mul_ps(x4, x3));
t3 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r1, 2, 1, 1, 0), x6 = _mm_shuffle_ps(r2, r1, _MM_SHUFFLE(0, 0, 0, 0)); /* e e i i */
_mm_shuffle1_ps(r3, 3, 3, 2, 3)), x7 = _mm_shuffle2_ps(r3, r2, 0, 0, 0, 0, 2, 0, 0, 0); /* i m m m */
_mm_mul_ps(_mm_shuffle1_ps(r3, 2, 1, 1, 0),
_mm_shuffle1_ps(r1, 3, 3, 2, 3)));
t4 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r1, 0, 0, 0, 0), /* t1[3] = i * p - m * l;
_mm_shuffle1_ps(r3, 2, 1, 2, 1)), t1[3] = i * p - m * l;
_mm_mul_ps(_mm_shuffle1_ps(r3, 0, 0, 0, 0), t2[3] = e * p - m * h;
_mm_shuffle1_ps(r1, 2, 1, 2, 1))); t3[3] = e * l - i * h; */
t3 = _mm_sub_ps(_mm_mul_ps(x6, x1), _mm_mul_ps(x7, x0));
t0 = _mm_shuffle1_ps(t3, 3, 3, 2, 1); /* t1[4] = i * o - m * k;
t1 = _mm_shuffle2_ps(t3, t4, 1, 1, 2, 0, 1, 0, 0, 3); t1[4] = i * o - m * k;
t2 = _mm_shuffle2_ps(t3, t4, 0, 1, 1, 1, 0, 2, 3, 3); 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));
v2 = _mm_mul_ps(_mm_shuffle1_ps(r0, 1, 0, 0, 0), t0); /* t1[5] = i * n - m * j;
v2 = _mm_sub_ps(v2, _mm_mul_ps(_mm_shuffle1_ps(r0, 2, 2, 1, 1), t1)); t1[5] = i * n - m * j;
v2 = _mm_add_ps(v2, _mm_mul_ps(_mm_shuffle1_ps(r0, 3, 3, 3, 2), t2)); t2[5] = e * n - m * f;
v2 = _mm_xor_ps(v2, _mm_set_ps(0.f, -0.f, 0.f, -0.f)); t3[5] = e * j - i * f; */
t5 = _mm_sub_ps(_mm_mul_ps(x6, x4), _mm_mul_ps(x7, x5));
t3 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r1, 2, 1, 1, 0), x0 = _mm_shuffle2_ps(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
_mm_shuffle1_ps(r2, 3, 3, 2, 3)), x1 = _mm_shuffle2_ps(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
_mm_mul_ps(_mm_shuffle1_ps(r2, 2, 1, 1, 0), x2 = _mm_shuffle2_ps(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
_mm_shuffle1_ps(r1, 3, 3, 2, 3))); x3 = _mm_shuffle2_ps(r1, r0, 3, 3, 3, 3, 2, 2, 2, 0); /* d d d h */
t4 = _mm_sub_ps(_mm_mul_ps(_mm_shuffle1_ps(r1, 0, 0, 0, 0), /*
_mm_shuffle1_ps(r2, 2, 1, 2, 1)), dest[0][0] = f * t1[0] - g * t1[1] + h * t1[2];
_mm_mul_ps(_mm_shuffle1_ps(r2, 0, 0, 0, 0), dest[0][1] =-(b * t1[0] - c * t1[1] + d * t1[2]);
_mm_shuffle1_ps(r1, 2, 1, 2, 1))); 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));
t0 = _mm_shuffle1_ps(t3, 3, 3, 2, 1); /*
t1 = _mm_shuffle2_ps(t3, t4, 1, 1, 2, 0, 1, 0, 0, 3); dest[1][0] =-(e * t1[0] - g * t1[3] + h * t1[4]);
t2 = _mm_shuffle2_ps(t3, t4, 0, 1, 1, 1, 0, 2, 3, 3); 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));
v3 = _mm_mul_ps(_mm_shuffle1_ps(r0, 1, 0, 0, 0), t0); /*
v3 = _mm_sub_ps(v3, _mm_mul_ps(_mm_shuffle1_ps(r0, 2, 2, 1, 1), t1)); dest[2][0] = e * t1[1] - f * t1[3] + h * t1[5];
v3 = _mm_add_ps(v3, _mm_mul_ps(_mm_shuffle1_ps(r0, 3, 3, 3, 2), t2)); 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)); v3 = _mm_xor_ps(v3, _mm_set_ps(0.f, -0.f, 0.f, -0.f));
dt = _mm_mul_ps(_mm_shuffle1_ps(r0, 0, 1, 2, 3), v0); /* determinant */
dt = _mm_add_ps(dt, _mm_shuffle1_ps(dt, 0, 1, 2, 3)); x0 = _mm_shuffle_ps(v0, v1, _MM_SHUFFLE(0, 0, 0, 0));
dt = _mm_add_ps(dt, _mm_shuffle1_ps(dt, 1, 3, 3, 1)); x1 = _mm_shuffle_ps(v2, v3, _MM_SHUFFLE(0, 0, 0, 0));
dt = _mm_rcp_ps(dt); x0 = _mm_shuffle_ps(x0, x1, _MM_SHUFFLE(2, 0, 2, 0));
v0 = _mm_shuffle1_ps(v0, 0, 1, 2, 3); x0 = _mm_mul_ps(x0, r0);
v1 = _mm_shuffle1_ps(v1, 0, 1, 2, 3); x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 2, 3));
v2 = _mm_shuffle1_ps(v2, 0, 1, 2, 3); x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 0, 1));
v3 = _mm_shuffle1_ps(v3, 0, 1, 2, 3); x0 = _mm_rcp_ps(x0);
_mm_store_ps(dest[0], _mm_mul_ps(v0, dt)); _mm_store_ps(dest[0], _mm_mul_ps(v0, x0));
_mm_store_ps(dest[1], _mm_mul_ps(v1, dt)); _mm_store_ps(dest[1], _mm_mul_ps(v1, x0));
_mm_store_ps(dest[2], _mm_mul_ps(v2, dt)); _mm_store_ps(dest[2], _mm_mul_ps(v2, x0));
_mm_store_ps(dest[3], _mm_mul_ps(v3, dt)); _mm_store_ps(dest[3], _mm_mul_ps(v3, x0));
} }
#endif /* cglm_mat_sse_h */ #endif /* cglm_mat_sse_h */