External/cglm
1
0
Fork 0
mirror of https://github.com/recp/cglm.git synced 2025-12-24 12:32:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

sse: optimize affine with sse

Browse Source 
* re-oder instructions for ILP
This commit is contained in:
Recep Aslantas
2021-05-01 02:58:14 +03:00
parent 376cf31ee7
commit 5b7bc522ac
1 changed files with 64 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

121
include/cglm/simd/sse2/affine.h
View File

@@ -16,76 +16,80 @@ CGLM_INLINE
void
glm_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
/* D = R * L (Column-Major) */
__m128 l0, l1, l2, l3, r;
glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
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],
glmm_fmadd(glmm_splat(r, 0), l0,
glmm_fmadd(glmm_splat(r, 1), l1,
_mm_mul_ps(glmm_splat(r, 2), l2))));
r = glmm_load(m2[1]);
glmm_store(dest[1],
glmm_fmadd(glmm_splat(r, 0), l0,
glmm_fmadd(glmm_splat(r, 1), l1,
_mm_mul_ps(glmm_splat(r, 2), l2))));
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]);
r = glmm_load(m2[2]);
glmm_store(dest[2],
glmm_fmadd(glmm_splat(r, 0), l0,
glmm_fmadd(glmm_splat(r, 1), l1,
_mm_mul_ps(glmm_splat(r, 2), l2))));
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);
r = glmm_load(m2[3]);
glmm_store(dest[3],
glmm_fmadd(glmm_splat(r, 0), l0,
glmm_fmadd(glmm_splat(r, 1), l1,
glmm_fmadd(glmm_splat(r, 2), l2,
_mm_mul_ps(glmm_splat(r, 3), l3)))));
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) */
__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]);
glmm_128 l, r0, r1, r2, r3, v0, v1, v2, v3;
r = glmm_load(m2[0]);
glmm_store(dest[0],
glmm_fmadd(glmm_splat(r, 0), l0,
glmm_fmadd(glmm_splat(r, 1), l1,
_mm_mul_ps(glmm_splat(r, 2), l2))));
r = glmm_load(m2[1]);
glmm_store(dest[1],
glmm_fmadd(glmm_splat(r, 0), l0,
glmm_fmadd(glmm_splat(r, 1), l1,
_mm_mul_ps(glmm_splat(r, 2), l2))));
r = glmm_load(m2[2]);
glmm_store(dest[2],
glmm_fmadd(glmm_splat(r, 0), l0,
glmm_fmadd(glmm_splat(r, 1), l1,
_mm_mul_ps(glmm_splat(r, 2), l2))));
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]);
glmm_store(dest[3], l3);
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);
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;
__m128 r0, r1, r2, r3, x0, x1, x2, x3, x4, x5;
r0 = glmm_load(mat[0]);
r1 = glmm_load(mat[1]);
@@ -95,10 +99,13 @@ glm_inv_tr_sse2(mat4 mat) {
_MM_TRANSPOSE4_PS(r0, r1, r2, x1);
x0 = glmm_fmadd(r0, glmm_shuff1(r3, 0, 0, 0, 0),
glmm_fmadd(r1, glmm_shuff1(r3, 1, 1, 1, 1),
_mm_mul_ps(r2, glmm_shuff1(r3, 2, 2, 2, 2))));
x0 = _mm_xor_ps(x0, _mm_set1_ps(-0.f));
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);