External/cglm
1
0
Fork 0
mirror of https://github.com/recp/cglm.git synced 2025-10-04 01:00:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix #3

Browse Source 
with caching values, compiler produce better assembly even identical
sse2 implementation
This commit is contained in:
Recep Aslantas
2016-09-27 21:18:52 +03:00
parent dc73cab522
commit 2cb9f7f334
1 changed files with 34 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
include/cglm-mat.h
View File

@@ -38,44 +38,43 @@ glm_mat4_dup(mat4 mat, mat4 dest) {
CGLM_INLINE
void
glm_mat4_mul(mat4 l, mat4 r, mat4 d) {
glm_mat4_mul(mat4 m1, mat4 m2, mat4 dest) {
#ifdef __AVX__
glm_mat4_mul_avx(l, r, d);
glm_mat4_mul_avx(m1, m2, dest);
#elif defined( __SSE__ ) || defined( __SSE2__ )
glm_mat4_mul_sse2(l, r, d);
glm_mat4_mul_sse2(m1, m2, dest);
#else
d[0][0] = l[0][0] * r[0][0] + l[1][0] * r[0][1] +
l[2][0] * r[0][2] + l[3][0] * r[0][3];
d[1][0] = l[0][0] * r[1][0] + l[1][0] * r[1][1] +
l[2][0] * r[1][2] + l[3][0] * r[1][3];
d[2][0] = l[0][0] * r[2][0] + l[1][0] * r[2][1] +
l[2][0] * r[2][2] + l[3][0] * r[2][3];
d[3][0] = l[0][0] * r[3][0] + l[1][0] * r[3][1] +
l[2][0] * r[3][2] + l[3][0] * r[3][3];
d[0][1] = l[0][1] * r[0][0] + l[1][1] * r[0][1] +
l[2][1] * r[0][2] + l[3][1] * r[0][3];
d[1][1] = l[0][1] * r[1][0] + l[1][1] * r[1][1] +
l[2][1] * r[1][2] + l[3][1] * r[1][3];
d[2][1] = l[0][1] * r[2][0] + l[1][1] * r[2][1] +
l[2][1] * r[2][2] + l[3][1] * r[2][3];
d[3][1] = l[0][1] * r[3][0] + l[1][1] * r[3][1] +
l[2][1] * r[3][2] + l[3][1] * r[3][3];
d[0][2] = l[0][2] * r[0][0] + l[1][2] * r[0][1] +
l[2][2] * r[0][2] + l[3][2] * r[0][3];
d[1][2] = l[0][2] * r[1][0] + l[1][2] * r[1][1] +
l[2][2] * r[1][2] + l[3][2] * r[1][3];
d[2][2] = l[0][2] * r[2][0] + l[1][2] * r[2][1] +
l[2][2] * r[2][2] + l[3][2] * r[2][3];
d[3][2] = l[0][2] * r[3][0] + l[1][2] * r[3][1] +
l[2][2] * r[3][2] + l[3][2] * r[3][3];
d[0][3] = l[0][3] * r[0][0] + l[1][3] * r[0][1] +
l[2][3] * r[0][2] + l[3][3] * r[0][3];
d[1][3] = l[0][3] * r[1][0] + l[1][3] * r[1][1] +
l[2][3] * r[1][2] + l[3][3] * r[1][3];
d[2][3] = l[0][3] * r[2][0] + l[1][3] * r[2][1] +
l[2][3] * r[2][2] + l[3][3] * r[2][3];
d[3][3] = l[0][3] * r[3][0] + l[1][3] * r[3][1] +
l[2][3] * r[3][2] + l[3][3] * r[3][3];
float a00, a01, a02, a03, b00, b01, b02, b03,
a10, a11, a12, a13, b10, b11, b12, b13,
a20, a21, a22, a23, b20, b21, b22, b23,
a30, a31, a32, a33, b30, b31, b32, b33;
a00 = m1[0][0], a01 = m1[0][1], a02 = m1[0][2], a03 = m1[0][3],
a10 = m1[1][0], a11 = m1[1][1], a12 = m1[1][2], a13 = m1[1][3],
a20 = m1[2][0], a21 = m1[2][1], a22 = m1[2][2], a23 = m1[2][3],
a30 = m1[3][0], a31 = m1[3][1], a32 = m1[3][2], a33 = m1[3][3];
b00 = m2[0][0], b01 = m2[0][1], b02 = m2[0][2], b03 = m2[0][3],
b10 = m2[1][0], b11 = m2[1][1], b12 = m2[1][2], b13 = m2[1][3],
b20 = m2[2][0], b21 = m2[2][1], b22 = m2[2][2], b23 = m2[2][3],
b30 = m2[3][0], b31 = m2[3][1], b32 = m2[3][2], b33 = m2[3][3];
dest[0][0] = a00 * b00 + a10 * b01 + a20 * b02 + a30 * b03;
dest[1][0] = a00 * b10 + a10 * b11 + a20 * b12 + a30 * b13;
dest[2][0] = a00 * b20 + a10 * b21 + a20 * b22 + a30 * b23;
dest[3][0] = a00 * b30 + a10 * b31 + a20 * b32 + a30 * b33;
dest[0][1] = a01 * b00 + a11 * b01 + a21 * b02 + a31 * b03;
dest[1][1] = a01 * b10 + a11 * b11 + a21 * b12 + a31 * b13;
dest[2][1] = a01 * b20 + a11 * b21 + a21 * b22 + a31 * b23;
dest[3][1] = a01 * b30 + a11 * b31 + a21 * b32 + a31 * b33;
dest[0][2] = a02 * b00 + a12 * b01 + a22 * b02 + a32 * b03;
dest[1][2] = a02 * b10 + a12 * b11 + a22 * b12 + a32 * b13;
dest[2][2] = a02 * b20 + a12 * b21 + a22 * b22 + a32 * b23;
dest[3][2] = a02 * b30 + a12 * b31 + a22 * b32 + a32 * b33;
dest[0][3] = a03 * b00 + a13 * b01 + a23 * b02 + a33 * b03;
dest[1][3] = a03 * b10 + a13 * b11 + a23 * b12 + a33 * b13;
dest[2][3] = a03 * b20 + a13 * b21 + a23 * b22 + a33 * b23;
dest[3][3] = a03 * b30 + a13 * b31 + a23 * b32 + a33 * b33;
#endif
}