diff options
| author | Sergey Sharybin <sergey.vfx@gmail.com> | 2017-03-23 18:16:05 +0300 |
|---|---|---|
| committer | Sergey Sharybin <sergey.vfx@gmail.com> | 2017-03-23 19:58:03 +0300 |
| commit | ba8c7d2ba120f5ba7fdbaef67fe2e301ec3e5dcb (patch) | |
| tree | f2cb920306275c6608c2b03a40c277b9fa584fbd /intern/cycles/kernel/geom | |
| parent | a1348dde2ed27d0a8a1d62f9e17602857b1f19f1 (diff) | |
Cycles: Use SSE-optimized version of triangle intersection for motion triangles
The title says it all actually. Gives up to 10% speedup on test scenes here
on i7-6800K.
Render times on GPU are unreliable here, but there might be some slowdown
caused by watertight nature of intersections.
Diffstat (limited to 'intern/cycles/kernel/geom')
| -rw-r--r-- | intern/cycles/kernel/geom/geom_motion_triangle_intersect.h | 47 |
1 files changed, 28 insertions, 19 deletions
diff --git a/intern/cycles/kernel/geom/geom_motion_triangle_intersect.h b/intern/cycles/kernel/geom/geom_motion_triangle_intersect.h index 2500228281e..971a34308f1 100644 --- a/intern/cycles/kernel/geom/geom_motion_triangle_intersect.h +++ b/intern/cycles/kernel/geom/geom_motion_triangle_intersect.h @@ -166,14 +166,15 @@ float3 motion_triangle_refine_subsurface(KernelGlobals *kg, * time and do a ray intersection with the resulting triangle. */ -ccl_device_inline bool motion_triangle_intersect(KernelGlobals *kg, - Intersection *isect, - float3 P, - float3 dir, - float time, - uint visibility, - int object, - int prim_addr) +ccl_device_inline bool motion_triangle_intersect( + KernelGlobals *kg, + const TriangleIsectPrecalc *isect_precalc, + Intersection *isect, + float3 P, + float time, + uint visibility, + int object, + int prim_addr) { /* Primitive index for vertex location lookup. */ int prim = kernel_tex_fetch(__prim_index, prim_addr); @@ -185,11 +186,15 @@ ccl_device_inline bool motion_triangle_intersect(KernelGlobals *kg, motion_triangle_vertices(kg, fobject, prim, time, verts); /* Ray-triangle intersection, unoptimized. */ float t, u, v; - if(ray_triangle_intersect_uv(P, - dir, - isect->t, - verts[2], verts[0], verts[1], - &u, &v, &t)) + if(ray_triangle_intersect(isect_precalc, + P, + isect->t, +#if defined(__KERNEL_AVX2__) && defined(__KERNEL_SSE__) + (ssef*)verts, +#else + verts[0], verts[1], verts[2], +#endif + &u, &v, &t)) { #ifdef __VISIBILITY_FLAG__ /* Visibility flag test. we do it here under the assumption @@ -217,9 +222,9 @@ ccl_device_inline bool motion_triangle_intersect(KernelGlobals *kg, #ifdef __SUBSURFACE__ ccl_device_inline void motion_triangle_intersect_subsurface( KernelGlobals *kg, + const TriangleIsectPrecalc *isect_precalc, SubsurfaceIntersection *ss_isect, float3 P, - float3 dir, float time, int object, int prim_addr, @@ -237,11 +242,15 @@ ccl_device_inline void motion_triangle_intersect_subsurface( motion_triangle_vertices(kg, fobject, prim, time, verts); /* Ray-triangle intersection, unoptimized. */ float t, u, v; - if(ray_triangle_intersect_uv(P, - dir, - tmax, - verts[2], verts[0], verts[1], - &u, &v, &t)) + if(ray_triangle_intersect(isect_precalc, + P, + tmax, +#if defined(__KERNEL_AVX2__) && defined(__KERNEL_SSE__) + (ssef*)verts, +#else + verts[0], verts[1], verts[2], +#endif + &u, &v, &t)) { for(int i = min(max_hits, ss_isect->num_hits) - 1; i >= 0; --i) { if(ss_isect->hits[i].t == t) { |