#include <xmmintrin.h>
-void Mod_Skeletal_AnimateVertices_SSE(const dp_model_t * RESTRICT model, const frameblend_t * RESTRICT frameblend, const skeleton_t *skeleton, float * RESTRICT vertex3f, float * RESTRICT normal3f, float * RESTRICT svector3f, float * RESTRICT tvector3f)
+void Mod_Skeletal_AnimateVertices_SSE(const model_t * RESTRICT model, const frameblend_t * RESTRICT frameblend, const skeleton_t *skeleton, float * RESTRICT vertex3f, float * RESTRICT normal3f, float * RESTRICT svector3f, float * RESTRICT tvector3f)
{
// vertex weighted skeletal
int i, k;
int blends;
matrix4x4_t *bonepose;
matrix4x4_t *boneposerelative;
- float m[12];
const blendweights_t * RESTRICT weights;
int num_vertices_minus_one;
const matrix4x4_t * RESTRICT p = &bonepose[model->data_bones[i].parent];
__m128 s0 = _mm_loadu_ps(s->m[0]), s1 = _mm_loadu_ps(s->m[1]), s2 = _mm_loadu_ps(s->m[2]);
#ifdef OPENGLORIENTATION
+ __m128 s3 = _mm_loadu_ps(s->m[3]);
#define SKELETON_MATRIX(r, c) _mm_shuffle_ps(s##c, s##c, _MM_SHUFFLE(r, r, r, r))
#else
#define SKELETON_MATRIX(r, c) _mm_shuffle_ps(s##r, s##r, _MM_SHUFFLE(c, c, c, c))
r1 = _mm_add_ps(r1, _mm_mul_ps(b2, _mm_shuffle_ps(nr, nr, _MM_SHUFFLE(1, 1, 1, 1))));
r2 = _mm_add_ps(r2, _mm_mul_ps(b2, _mm_shuffle_ps(nr, nr, _MM_SHUFFLE(2, 2, 2, 2))));
r3 = _mm_add_ps(r3, _mm_mul_ps(b2, _mm_shuffle_ps(nr, nr, _MM_SHUFFLE(3, 3, 3, 3))));
+ r3 = _mm_add_ps(r3, b3);
_mm_store_ps(r->m[0], r0);
_mm_store_ps(r->m[1], r1);
_mm_store_ps(r->m[2], r2);
{
for (i = 0;i < model->num_bones;i++)
{
- const short * RESTRICT pose7s = model->data_poses7s + 7 * (frameblend[0].subframe * model->num_bones + i);
- float lerp = frameblend[0].lerp,
- tx = pose7s[0], ty = pose7s[1], tz = pose7s[2],
- rx = pose7s[3] * lerp,
- ry = pose7s[4] * lerp,
- rz = pose7s[5] * lerp,
- rw = pose7s[6] * lerp,
- dx = tx*rw + ty*rz - tz*ry,
- dy = -tx*rz + ty*rw + tz*rx,
- dz = tx*ry - ty*rx + tz*rw,
- dw = -tx*rx - ty*ry - tz*rz,
- scale, originscale;
+ float m[12];
+ const short * RESTRICT firstpose7s = model->data_poses7s + 7 * (frameblend[0].subframe * model->num_bones + i);
+ float firstlerp = frameblend[0].lerp,
+ firsttx = firstpose7s[0], firstty = firstpose7s[1], firsttz = firstpose7s[2],
+ rx = firstpose7s[3] * firstlerp,
+ ry = firstpose7s[4] * firstlerp,
+ rz = firstpose7s[5] * firstlerp,
+ rw = firstpose7s[6] * firstlerp,
+ dx = firsttx*rw + firstty*rz - firsttz*ry,
+ dy = -firsttx*rz + firstty*rw + firsttz*rx,
+ dz = firsttx*ry - firstty*rx + firsttz*rw,
+ dw = -firsttx*rx - firstty*ry - firsttz*rz,
+ scale, sx, sy, sz, sw;
for (blends = 1;blends < MAX_FRAMEBLENDS && frameblend[blends].lerp > 0;blends++)
{
- const short * RESTRICT pose7s = model->data_poses7s + 7 * (frameblend[blends].subframe * model->num_bones + i);
- float lerp = frameblend[blends].lerp,
- tx = pose7s[0], ty = pose7s[1], tz = pose7s[2],
- qx = pose7s[3], qy = pose7s[4], qz = pose7s[5], qw = pose7s[6];
- if(rx*qx + ry*qy + rz*qz + rw*qw < 0) lerp = -lerp;
- qx *= lerp;
- qy *= lerp;
- qz *= lerp;
- qw *= lerp;
+ const short * RESTRICT blendpose7s = model->data_poses7s + 7 * (frameblend[blends].subframe * model->num_bones + i);
+ float blendlerp = frameblend[blends].lerp,
+ blendtx = blendpose7s[0], blendty = blendpose7s[1], blendtz = blendpose7s[2],
+ qx = blendpose7s[3], qy = blendpose7s[4], qz = blendpose7s[5], qw = blendpose7s[6];
+ if(rx*qx + ry*qy + rz*qz + rw*qw < 0) blendlerp = -blendlerp;
+ qx *= blendlerp;
+ qy *= blendlerp;
+ qz *= blendlerp;
+ qw *= blendlerp;
rx += qx;
ry += qy;
rz += qz;
rw += qw;
- dx += tx*qw + ty*qz - tz*qy;
- dy += -tx*qz + ty*qw + tz*qx;
- dz += tx*qy - ty*qx + tz*qw;
- dw += -tx*qx - ty*qy - tz*qz;
+ dx += blendtx*qw + blendty*qz - blendtz*qy;
+ dy += -blendtx*qz + blendty*qw + blendtz*qx;
+ dz += blendtx*qy - blendty*qx + blendtz*qw;
+ dw += -blendtx*qx - blendty*qy - blendtz*qz;
}
scale = 1.0f / (rx*rx + ry*ry + rz*rz + rw*rw);
- originscale = -model->num_posescale * scale;
- m[0] = scale*(rw*rw + rx*rx - ry*ry - rz*rz);
- m[1] = 2*scale*(rx*ry - rw*rz);
- m[2] = 2*scale*(rx*rz + rw*ry);
- m[3] = originscale*(dw*rx - dx*rw + dy*rz - dz*ry);
- m[4] = 2*scale*(rx*ry + rw*rz);
- m[5] = scale*(rw*rw + ry*ry - rx*rx - rz*rz);
- m[6] = 2*scale*(ry*rz - rw*rx);
- m[7] = originscale*(dw*ry - dx*rz - dy*rw + dz*rx);
- m[8] = 2*scale*(rx*rz - rw*ry);
- m[9] = 2*scale*(ry*rz + rw*rx);
- m[10] = scale*(rw*rw + rz*rz - rx*rx - ry*ry);
- m[11] = originscale*(dw*rz + dx*ry - dy*rx - dz*rw);
+ sx = rx * scale;
+ sy = ry * scale;
+ sz = rz * scale;
+ sw = rw * scale;
+ m[0] = sw*rw + sx*rx - sy*ry - sz*rz;
+ m[1] = 2*(sx*ry - sw*rz);
+ m[2] = 2*(sx*rz + sw*ry);
+ m[3] = model->num_posescale*(dx*sw - dy*sz + dz*sy - dw*sx);
+ m[4] = 2*(sx*ry + sw*rz);
+ m[5] = sw*rw + sy*ry - sx*rx - sz*rz;
+ m[6] = 2*(sy*rz - sw*rx);
+ m[7] = model->num_posescale*(dx*sz + dy*sw - dz*sx - dw*sy);
+ m[8] = 2*(sx*rz - sw*ry);
+ m[9] = 2*(sy*rz + sw*rx);
+ m[10] = sw*rw + sz*rz - sx*rx - sy*ry;
+ m[11] = model->num_posescale*(dy*sx + dz*sw - dx*sy - dw*sz);
if (i == r_skeletal_debugbone.integer)
m[r_skeletal_debugbonecomponent.integer % 12] += r_skeletal_debugbonevalue.value;
m[3] *= r_skeletal_debugtranslatex.value;
const float * RESTRICT n = model->surfmesh.data_normal3f;
if (svector3f && tvector3f)
{
- const float * RESTRICT sv = model->surfmesh.data_svector3f;
- const float * RESTRICT tv = model->surfmesh.data_tvector3f;
+ const float * RESTRICT svec = model->surfmesh.data_svector3f;
+ const float * RESTRICT tvec = model->surfmesh.data_tvector3f;
// Note that for SSE each iteration stores one element past end, so we break one vertex short
// and handle that with scalars in that case
- for (i = 0; i < num_vertices_minus_one; i++, v += 3, n += 3, sv += 3, tv += 3, b++,
+ for (i = 0; i < num_vertices_minus_one; i++, v += 3, n += 3, svec += 3, tvec += 3, b++,
vertex3f += 3, normal3f += 3, svector3f += 3, tvector3f += 3)
{
LOAD_MATRIX4();
TRANSFORM_POSITION(v, vertex3f);
TRANSFORM_VECTOR(n, normal3f);
- TRANSFORM_VECTOR(sv, svector3f);
- TRANSFORM_VECTOR(tv, tvector3f);
+ TRANSFORM_VECTOR(svec, svector3f);
+ TRANSFORM_VECTOR(tvec, tvector3f);
}
// Last vertex needs to be done with scalars to avoid reading/writing 1 word past end of arrays
LOAD_MATRIX_SCALAR();
TRANSFORM_POSITION_SCALAR(v, vertex3f);
TRANSFORM_VECTOR_SCALAR(n, normal3f);
- TRANSFORM_VECTOR_SCALAR(sv, svector3f);
- TRANSFORM_VECTOR_SCALAR(tv, tvector3f);
+ TRANSFORM_VECTOR_SCALAR(svec, svector3f);
+ TRANSFORM_VECTOR_SCALAR(tvec, tvector3f);
}
//printf("elapsed ticks: %llu\n", rdtsc() - ts); // XXX
return;
if (svector3f)
{
- const float * RESTRICT sv = model->surfmesh.data_svector3f;
+ const float * RESTRICT svec = model->surfmesh.data_svector3f;
const unsigned short * RESTRICT b = model->surfmesh.blends;
- for (i = 0; i < num_vertices_minus_one; i++, sv += 3, b++, svector3f += 3)
+ for (i = 0; i < num_vertices_minus_one; i++, svec += 3, b++, svector3f += 3)
{
LOAD_MATRIX3();
- TRANSFORM_VECTOR(sv, svector3f);
+ TRANSFORM_VECTOR(svec, svector3f);
}
{
LOAD_MATRIX_SCALAR();
- TRANSFORM_VECTOR_SCALAR(sv, svector3f);
+ TRANSFORM_VECTOR_SCALAR(svec, svector3f);
}
}
if (tvector3f)
{
- const float * RESTRICT tv = model->surfmesh.data_tvector3f;
+ const float * RESTRICT tvec = model->surfmesh.data_tvector3f;
const unsigned short * RESTRICT b = model->surfmesh.blends;
- for (i = 0; i < num_vertices_minus_one; i++, tv += 3, b++, tvector3f += 3)
+ for (i = 0; i < num_vertices_minus_one; i++, tvec += 3, b++, tvector3f += 3)
{
LOAD_MATRIX3();
- TRANSFORM_VECTOR(tv, tvector3f);
+ TRANSFORM_VECTOR(tvec, tvector3f);
}
{
LOAD_MATRIX_SCALAR();
- TRANSFORM_VECTOR_SCALAR(tv, tvector3f);
+ TRANSFORM_VECTOR_SCALAR(tvec, tvector3f);
}
}
projects / xonotic / darkplaces.git / blobdiff