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

simd: rename _mm_ extensions to glmm_

Browse Source
This commit is contained in:
Recep Aslantas
2018-05-10 14:27:53 +03:00
parent 9fc2ead8ef
commit da5ad69863
7 changed files with 115 additions and 123 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
include/cglm/simd/intrin.h
View File

@@ -29,29 +29,27 @@
# include <emmintrin.h>
/* OPTIONAL: You may save some instructions but latency (not sure) */
#ifdef CGLM_USE_INT_DOM_FOR_SHUFF
# define _mm_shuffle1_ps(xmm, z, y, x, w) \
_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(xmm), \
_MM_SHUFFLE(z, y, x, w)))
#ifdef CGLM_USE_INT_DOMAIN
# define glmm_shuff1(xmm, z, y, x, w) \
_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(xmm), \
_MM_SHUFFLE(z, y, x, w)))
#else
# define _mm_shuffle1_ps(xmm, z, y, x, w) \
# define glmm_shuff1(xmm, z, y, x, w) \
_mm_shuffle_ps(xmm, xmm, _MM_SHUFFLE(z, y, x, w))
#endif
# define _mm_shuffle1_ps1(xmm, x) \
_mm_shuffle1_ps(xmm, 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)
#define glmm_shuff1x(xmm, x) glmm_shuff1(xmm, x, x, x, x)
#define glmm_shuff2(a, b, z0, y0, x0, w0, z1, y1, x1, w1) \
glmm_shuff1(_mm_shuffle_ps(a, b, _MM_SHUFFLE(z0, y0, x0, w0)), \
z1, y1, x1, w1)
static inline
__m128
glmm_dot(__m128 a, __m128 b) {
__m128 x0;
x0 = _mm_mul_ps(a, b);
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 3, 2));
return _mm_add_ps(x0, _mm_shuffle1_ps(x0, 0, 1, 0, 1));
x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 3, 2));
return _mm_add_ps(x0, glmm_shuff1(x0, 0, 1, 0, 1));
}
static inline
@@ -76,7 +74,7 @@ static inline
void
glmm_store3(__m128 vx, float v[3]) {
_mm_storel_pi((__m64 *)&v[0], vx);
_mm_store_ss(&v[2], _mm_shuffle1_ps(vx, 2, 2, 2, 2));
_mm_store_ss(&v[2], glmm_shuff1(vx, 2, 2, 2, 2));
}
#ifdef CGLM_ALL_UNALIGNED

50
include/cglm/simd/sse2/affine.h
View File

@@ -25,28 +25,28 @@ glm_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
r = glmm_load(m2[0]);
glmm_store(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)));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_mul_ps(glmm_shuff1x(r, 2), l2)));
r = glmm_load(m2[1]);
glmm_store(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)));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_mul_ps(glmm_shuff1x(r, 2), l2)));
r = glmm_load(m2[2]);
glmm_store(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)));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_mul_ps(glmm_shuff1x(r, 2), l2)));
r = glmm_load(m2[3]);
glmm_store(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))));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
_mm_mul_ps(glmm_shuff1x(r, 3), l3))));
}
CGLM_INLINE
@@ -62,21 +62,21 @@ glm_mul_rot_sse2(mat4 m1, mat4 m2, mat4 dest) {
r = glmm_load(m2[0]);
glmm_store(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)));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_mul_ps(glmm_shuff1x(r, 2), l2)));
r = glmm_load(m2[1]);
glmm_store(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)));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_mul_ps(glmm_shuff1x(r, 2), l2)));
r = glmm_load(m2[2]);
glmm_store(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)));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_mul_ps(glmm_shuff1x(r, 2), l2)));
glmm_store(dest[3], l3);
}
@@ -94,9 +94,9 @@ glm_inv_tr_sse2(mat4 mat) {
_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_add_ps(_mm_mul_ps(r0, glmm_shuff1(r3, 0, 0, 0, 0)),
_mm_mul_ps(r1, glmm_shuff1(r3, 1, 1, 1, 1)));
x0 = _mm_add_ps(x0, _mm_mul_ps(r2, glmm_shuff1(r3, 2, 2, 2, 2)));
x0 = _mm_xor_ps(x0, _mm_set1_ps(-0.f));
x0 = _mm_add_ps(x0, x1);

22
include/cglm/simd/sse2/mat3.h
View File

@@ -27,27 +27,25 @@ glm_mat3_mul_sse2(mat3 m1, mat3 m2, mat3 dest) {
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);
x1 = glmm_shuff2(l0, l1, 1, 0, 3, 3, 0, 3, 2, 0);
x2 = glmm_shuff2(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(_mm_mul_ps(glmm_shuff1(l0, 0, 2, 1, 0),
glmm_shuff1(r0, 3, 0, 0, 0)),
_mm_mul_ps(x1, glmm_shuff2(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_mul_ps(x2, glmm_shuff2(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),
x0 = _mm_add_ps(_mm_mul_ps(glmm_shuff1(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)));
_mm_mul_ps(glmm_shuff1(x1, 1, 0, 2, 1),
glmm_shuff1(r1, 3, 3, 0, 0)));
x0 = _mm_add_ps(x0,
_mm_mul_ps(_mm_shuffle1_ps(x2, 1, 0, 2, 1),
_mm_mul_ps(glmm_shuff1(x2, 1, 0, 2, 1),
_mm_shuffle_ps(r1, r2, _MM_SHUFFLE(0, 0, 1, 1))));
_mm_storeu_ps(&dest[1][1], x0);

112
include/cglm/simd/sse2/mat4.h
View File

@@ -58,29 +58,29 @@ glm_mat4_mul_sse2(mat4 m1, mat4 m2, mat4 dest) {
r = glmm_load(m2[0]);
glmm_store(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))));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
_mm_mul_ps(glmm_shuff1x(r, 3), l3))));
r = glmm_load(m2[1]);
glmm_store(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))));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
_mm_mul_ps(glmm_shuff1x(r, 3), l3))));
r = glmm_load(m2[2]);
glmm_store(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))));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
_mm_mul_ps(glmm_shuff1x(r, 3), l3))));
r = glmm_load(m2[3]);
glmm_store(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))));
_mm_add_ps(_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 0), l0),
_mm_mul_ps(glmm_shuff1x(r, 1), l1)),
_mm_add_ps(_mm_mul_ps(glmm_shuff1x(r, 2), l2),
_mm_mul_ps(glmm_shuff1x(r, 3), l3))));
}
CGLM_INLINE
@@ -89,15 +89,11 @@ glm_mat4_mulv_sse2(mat4 m, vec4 v, vec4 dest) {
__m128 x0, x1, x2;
x0 = glmm_load(v);
x1 = _mm_add_ps(_mm_mul_ps(glmm_load(m[0]),
_mm_shuffle1_ps1(x0, 0)),
_mm_mul_ps(glmm_load(m[1]),
_mm_shuffle1_ps1(x0, 1)));
x1 = _mm_add_ps(_mm_mul_ps(glmm_load(m[0]), glmm_shuff1x(x0, 0)),
_mm_mul_ps(glmm_load(m[1]), glmm_shuff1x(x0, 1)));
x2 = _mm_add_ps(_mm_mul_ps(glmm_load(m[2]),
_mm_shuffle1_ps1(x0, 2)),
_mm_mul_ps(glmm_load(m[3]),
_mm_shuffle1_ps1(x0, 3)));
x2 = _mm_add_ps(_mm_mul_ps(glmm_load(m[2]), glmm_shuff1x(x0, 2)),
_mm_mul_ps(glmm_load(m[3]), glmm_shuff1x(x0, 3)));
glmm_store(dest, _mm_add_ps(x1, x2));
}
@@ -119,20 +115,20 @@ glm_mat4_det_sse2(mat4 mat) {
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)));
x0 = _mm_sub_ps(_mm_mul_ps(glmm_shuff1(r2, 0, 0, 1, 1),
glmm_shuff1(r3, 2, 3, 2, 3)),
_mm_mul_ps(glmm_shuff1(r3, 0, 0, 1, 1),
glmm_shuff1(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)));
x1 = _mm_sub_ps(_mm_mul_ps(glmm_shuff1(r2, 0, 0, 2, 2),
glmm_shuff1(r3, 1, 1, 3, 3)),
_mm_mul_ps(glmm_shuff1(r3, 0, 0, 2, 2),
glmm_shuff1(r2, 1, 1, 3, 3)));
/*
a * (f * t[0] - g * t[1] + h * t[2])
@@ -140,19 +136,19 @@ glm_mat4_det_sse2(mat4 mat) {
+ 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),
x2 = _mm_sub_ps(_mm_mul_ps(glmm_shuff1(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)));
_mm_mul_ps(glmm_shuff1(r1, 1, 1, 2, 2),
glmm_shuff1(x0, 3, 2, 2, 0)));
x2 = _mm_add_ps(x2,
_mm_mul_ps(_mm_shuffle1_ps(r1, 2, 3, 3, 3),
_mm_mul_ps(glmm_shuff1(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));
x0 = _mm_add_ps(x0, glmm_shuff1(x0, 0, 1, 2, 3));
x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 3, 3, 1));
return _mm_cvtss_f32(x0);
}
@@ -172,8 +168,8 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
r3 = glmm_load(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 */
x1 = glmm_shuff1(x0, 1, 3, 3, 3); /* l p p p */
x2 = glmm_shuff1(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 */
@@ -184,7 +180,7 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
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 */
x4 = glmm_shuff1(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;
@@ -200,7 +196,7 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
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 */
x7 = glmm_shuff2(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;
@@ -220,10 +216,10 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
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 */
x0 = glmm_shuff2(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
x1 = glmm_shuff2(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
x2 = glmm_shuff2(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
x3 = glmm_shuff2(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];
@@ -271,8 +267,8 @@ glm_mat4_inv_fast_sse2(mat4 mat, mat4 dest) {
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_add_ps(x0, glmm_shuff1(x0, 0, 1, 2, 3));
x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 0, 1));
x0 = _mm_rcp_ps(x0);
glmm_store(dest[0], _mm_mul_ps(v0, x0));
@@ -296,8 +292,8 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
r3 = glmm_load(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 */
x1 = glmm_shuff1(x0, 1, 3, 3, 3); /* l p p p */
x2 = glmm_shuff1(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 */
@@ -308,7 +304,7 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
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 */
x4 = glmm_shuff1(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;
@@ -324,7 +320,7 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
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 */
x7 = glmm_shuff2(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;
@@ -344,10 +340,10 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
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 */
x0 = glmm_shuff2(r1, r0, 0, 0, 0, 0, 2, 2, 2, 0); /* a a a e */
x1 = glmm_shuff2(r1, r0, 1, 1, 1, 1, 2, 2, 2, 0); /* b b b f */
x2 = glmm_shuff2(r1, r0, 2, 2, 2, 2, 2, 2, 2, 0); /* c c c g */
x3 = glmm_shuff2(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];
@@ -395,8 +391,8 @@ glm_mat4_inv_sse2(mat4 mat, mat4 dest) {
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_add_ps(x0, glmm_shuff1(x0, 0, 1, 2, 3));
x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 0, 1));
x0 = _mm_div_ps(_mm_set1_ps(1.0f), x0);
glmm_store(dest[0], _mm_mul_ps(v0, x0));

14
include/cglm/simd/sse2/quat.h
View File

@@ -27,16 +27,16 @@ glm_quat_mul_sse2(versor p, versor q, versor dest) {
xp = glmm_load(p); /* 3 2 1 0 */
xq = glmm_load(q);
r = _mm_mul_ps(_mm_shuffle1_ps1(xp, 3), xq);
r = _mm_mul_ps(glmm_shuff1x(xp, 3), xq);
x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 0), _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 0, 1, 2, 3)));
x0 = _mm_xor_ps(glmm_shuff1x(xp, 0), _mm_set_ps(-0.f, 0.f, -0.f, 0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, glmm_shuff1(xq, 0, 1, 2, 3)));
x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 1), _mm_set_ps(-0.f, -0.f, 0.f, 0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 1, 0, 3, 2)));
x0 = _mm_xor_ps(glmm_shuff1x(xp, 1), _mm_set_ps(-0.f, -0.f, 0.f, 0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, glmm_shuff1(xq, 1, 0, 3, 2)));
x0 = _mm_xor_ps(_mm_shuffle1_ps1(xp, 2), _mm_set_ps(-0.f, 0.f, 0.f, -0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, _mm_shuffle1_ps(xq, 2, 3, 0, 1)));
x0 = _mm_xor_ps(glmm_shuff1x(xp, 2), _mm_set_ps(-0.f, 0.f, 0.f, -0.f));
r = _mm_add_ps(r, _mm_mul_ps(x0, glmm_shuff1(xq, 2, 3, 0, 1)));
glmm_store(dest, r);
}

6
include/cglm/vec4-ext.h
View File

@@ -225,10 +225,10 @@ glm_vec4_sign(vec4 v, vec4 dest) {
x0 = glmm_load(v);
x1 = _mm_set_ps(0.0f, 0.0f, 1.0f, -1.0f);
x2 = _mm_shuffle1_ps1(x1, 2);
x2 = glmm_shuff1x(x1, 2);
x3 = _mm_and_ps(_mm_cmpgt_ps(x0, x2), _mm_shuffle1_ps1(x1, 1));
x4 = _mm_and_ps(_mm_cmplt_ps(x0, x2), _mm_shuffle1_ps1(x1, 0));
x3 = _mm_and_ps(_mm_cmpgt_ps(x0, x2), glmm_shuff1x(x1, 1));
x4 = _mm_and_ps(_mm_cmplt_ps(x0, x2), glmm_shuff1x(x1, 0));
glmm_store(dest, _mm_or_ps(x3, x4));
#else

8
include/cglm/vec4.h
View File

@@ -170,8 +170,8 @@ glm_vec4_dot(vec4 a, vec4 b) {
#if defined( __SSE__ ) || defined( __SSE2__ )
__m128 x0;
x0 = _mm_mul_ps(glmm_load(a), glmm_load(b));
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 3, 2));
return _mm_cvtss_f32(_mm_add_ss(x0, _mm_shuffle1_ps(x0, 0, 1, 0, 1)));
x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 3, 2));
return _mm_cvtss_f32(_mm_add_ss(x0, glmm_shuff1(x0, 0, 1, 0, 1)));
#else
return a[0] * b[0] + a[1] * b[1] + a[2] * b[2] + a[3] * b[3];
#endif
@@ -195,8 +195,8 @@ glm_vec4_norm2(vec4 v) {
__m128 x0;
x0 = glmm_load(v);
x0 = _mm_mul_ps(x0, x0);
x0 = _mm_add_ps(x0, _mm_shuffle1_ps(x0, 1, 0, 3, 2));
return _mm_cvtss_f32(_mm_add_ss(x0, _mm_shuffle1_ps(x0, 0, 1, 0, 1)));
x0 = _mm_add_ps(x0, glmm_shuff1(x0, 1, 0, 3, 2));
return _mm_cvtss_f32(_mm_add_ss(x0, glmm_shuff1(x0, 0, 1, 0, 1)));
#else
return v[0] * v[0] + v[1] * v[1] + v[2] * v[2] + v[3] * v[3];
#endif