sandey/blender
1
0
Fork 0
forked from bartvdbraak/blender
Code Pull Requests Activity
cc95172667
blender/intern/cycles/kernel/bvh/qbvh_volume_all.h

452 lines
14 KiB
C
Raw Normal View History

Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/*
Cycles: Cleanup file headers Some of the files were wrongly attributing code to some other organizations and in few places proper attribution was missing. This is mainly either a copy-paste error (when new file was created from an existing one and header wasn't updated) or due to some refactor which split non-original-BF code with purely BF code. Should solve some confusion around.
2016-09-29 08:11:40 +00:00
* Copyright 2011-2013 Blender Foundation
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/* This is a template BVH traversal function for volumes, where
* various features can be enabled/disabled. This way we can compile optimized
* versions for each case without new features slowing things down.
*
* BVH_INSTANCING: object instancing
* BVH_MOTION: motion blur rendering
*
*/
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
#if BVH_FEATURE(BVH_HAIR)
# define NODE_INTERSECT qbvh_node_intersect
#else
# define NODE_INTERSECT qbvh_aligned_node_intersect
#endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
const Ray *ray,
Intersection *isect_array,
Cycles: Fix wrong camera in volume check when domain is only visible to camera rays
2016-04-04 17:30:38 +00:00
const uint max_hits,
const uint visibility)
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
{
/* TODO(sergey):
* - Test if pushing distance on the stack helps.
* - Likely and unlikely for if() statements.
* - Test restrict attribute for pointers.
*/
/* Traversal stack in CUDA thread-local memory. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
QBVHStackItem traversal_stack[BVH_QSTACK_SIZE];
traversal_stack[0].addr = ENTRYPOINT_SENTINEL;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/* Traversal variables in registers. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
int stack_ptr = 0;
int node_addr = kernel_data.bvh.root;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/* Ray parameters in registers. */
const float tmax = ray->t;
float3 P = ray->P;
float3 dir = bvh_clamp_direction(ray->D);
float3 idir = bvh_inverse_direction(dir);
int object = OBJECT_NONE;
float isect_t = tmax;
#if BVH_FEATURE(BVH_MOTION)
Cycles: Fix wrong intersection with motion blur and degenerate object transform
2015-10-08 16:24:54 +00:00
Transform ob_itfm;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#endif
Fix T45333: Volume Scatter crash blender
2015-07-13 13:53:48 +00:00
uint num_hits = 0;
isect_array->t = tmax;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#ifndef __KERNEL_SSE41__
if(!isfinite(P.x)) {
Cycles: Avoid conversion from bool to uint
2016-09-20 11:00:16 +00:00
return 0;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
}
#endif
#if BVH_FEATURE(BVH_INSTANCING)
int num_hits_in_instance = 0;
#endif
ssef tnear(0.0f), tfar(isect_t);
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
#if BVH_FEATURE(BVH_HAIR)
sse3f dir4(ssef(dir.x), ssef(dir.y), ssef(dir.z));
#endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z));
#ifdef __KERNEL_AVX2__
float3 P_idir = P*idir;
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
sse3f P_idir4(P_idir.x, P_idir.y, P_idir.z);
#endif
#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
sse3f org4(ssef(P.x), ssef(P.y), ssef(P.z));
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#endif
/* Offsets to select the side that becomes the lower or upper bound. */
int near_x, near_y, near_z;
int far_x, far_y, far_z;
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
IsectPrecalc isect_precalc;
triangle_intersect_precalc(dir, &isect_precalc);
/* Traversal loop. */
do {
do {
/* Traverse internal nodes. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
while(node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) {
float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr+0);
Cycles: More tweaks to make specialized BVH traversal matching
2016-09-19 13:29:37 +00:00
#ifdef __VISIBILITY_FLAG__
Cycles: Support visibility check for inner nodes of QBVH It was initially unsupported because initial idea of checking visibility of all children was slowing scenes down a lot. Now the idea has changed and we only perform visibility check of current node. This avoids huge slowdown (from tests here it seems to be withing 1-2%, but more tests would never hurt) and gives nice speedup of ray traversal for complex scenes which utilized ray visibility. Here's timing of koro.blend: Without visibility check With visibility check Original file 4min 20sec 4min 23sec Camera rays only 1min 43 sec 55sec Unfortunately, this doesn't come for free and requires extra data in BVH node, which increases memory usage of BVH nodes by 15%. This we can solve with some future trickery of avoiding __tri_storage created for curve segments.
2016-06-10 08:14:05 +00:00
if((__float_as_uint(inodes.x) & visibility) == 0) {
/* Pop. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
node_addr = traversal_stack[stack_ptr].addr;
--stack_ptr;
Cycles: Support visibility check for inner nodes of QBVH It was initially unsupported because initial idea of checking visibility of all children was slowing scenes down a lot. Now the idea has changed and we only perform visibility check of current node. This avoids huge slowdown (from tests here it seems to be withing 1-2%, but more tests would never hurt) and gives nice speedup of ray traversal for complex scenes which utilized ray visibility. Here's timing of koro.blend: Without visibility check With visibility check Original file 4min 20sec 4min 23sec Camera rays only 1min 43 sec 55sec Unfortunately, this doesn't come for free and requires extra data in BVH node, which increases memory usage of BVH nodes by 15%. This we can solve with some future trickery of avoiding __tri_storage created for curve segments.
2016-06-10 08:14:05 +00:00
continue;
}
#endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
ssef dist;
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
int child_mask = NODE_INTERSECT(kg,
tnear,
tfar,
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#ifdef __KERNEL_AVX2__
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
P_idir4,
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
#endif
#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
org4,
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
#endif
#if BVH_FEATURE(BVH_HAIR)
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
dir4,
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#endif
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
idir4,
near_x, near_y, near_z,
far_x, far_y, far_z,
node_addr,
&dist);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
if(child_mask != 0) {
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
float4 cnodes;
#if BVH_FEATURE(BVH_HAIR)
if(__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) {
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
cnodes = kernel_tex_fetch(__bvh_nodes, node_addr+13);
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
}
else
#endif
{
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
cnodes = kernel_tex_fetch(__bvh_nodes, node_addr+7);
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
}
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/* One child is hit, continue with that child. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
int r = __bscf(child_mask);
if(child_mask == 0) {
node_addr = __float_as_int(cnodes[r]);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
continue;
}
/* Two children are hit, push far child, and continue with
* closer child.
*/
int c0 = __float_as_int(cnodes[r]);
float d0 = ((float*)&dist)[r];
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
r = __bscf(child_mask);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
int c1 = __float_as_int(cnodes[r]);
float d1 = ((float*)&dist)[r];
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
if(child_mask == 0) {
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
if(d1 < d0) {
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
node_addr = c1;
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c0;
traversal_stack[stack_ptr].dist = d0;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
continue;
}
else {
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
node_addr = c0;
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c1;
traversal_stack[stack_ptr].dist = d1;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
continue;
}
}
/* Here starts the slow path for 3 or 4 hit children. We push
* all nodes onto the stack to sort them there.
*/
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c1;
traversal_stack[stack_ptr].dist = d1;
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c0;
traversal_stack[stack_ptr].dist = d0;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/* Three children are hit, push all onto stack and sort 3
* stack items, continue with closest child.
*/
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
r = __bscf(child_mask);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
int c2 = __float_as_int(cnodes[r]);
float d2 = ((float*)&dist)[r];
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
if(child_mask == 0) {
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c2;
traversal_stack[stack_ptr].dist = d2;
qbvh_stack_sort(&traversal_stack[stack_ptr],
&traversal_stack[stack_ptr - 1],
&traversal_stack[stack_ptr - 2]);
node_addr = traversal_stack[stack_ptr].addr;
--stack_ptr;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
continue;
}
/* Four children are hit, push all onto stack and sort 4
* stack items, continue with closest child.
*/
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
r = __bscf(child_mask);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
int c3 = __float_as_int(cnodes[r]);
float d3 = ((float*)&dist)[r];
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c3;
traversal_stack[stack_ptr].dist = d3;
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = c2;
traversal_stack[stack_ptr].dist = d2;
qbvh_stack_sort(&traversal_stack[stack_ptr],
&traversal_stack[stack_ptr - 1],
&traversal_stack[stack_ptr - 2],
&traversal_stack[stack_ptr - 3]);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
}
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
node_addr = traversal_stack[stack_ptr].addr;
--stack_ptr;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
}
/* If node is leaf, fetch triangle list. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
if(node_addr < 0) {
float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr-1));
Cycles: Implement proper visibility check in QBVH volume all intersection
2016-08-25 09:09:06 +00:00
if((__float_as_uint(leaf.z) & visibility) == 0) {
/* Pop. */
node_addr = traversal_stack[stack_ptr].addr;
--stack_ptr;
continue;
}
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
int prim_addr = __float_as_int(leaf.x);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#if BVH_FEATURE(BVH_INSTANCING)
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
if(prim_addr >= 0) {
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#endif
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
int prim_addr2 = __float_as_int(leaf.y);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
const uint type = __float_as_int(leaf.w);
const uint p_type = type & PRIMITIVE_ALL;
bool hit;
/* Pop. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
node_addr = traversal_stack[stack_ptr].addr;
--stack_ptr;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/* Primitive intersection. */
switch(p_type) {
case PRIMITIVE_TRIANGLE: {
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
for(; prim_addr < prim_addr2; prim_addr++) {
kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/* Only primitives from volume object. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
uint tri_object = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, prim_addr): object;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
int object_flag = kernel_tex_fetch(__object_flag, tri_object);
if((object_flag & SD_OBJECT_HAS_VOLUME) == 0) {
continue;
}
/* Intersect ray against primitive. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
hit = triangle_intersect(kg, &isect_precalc, isect_array, P, visibility, object, prim_addr);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
if(hit) {
Cycles: Revert previous fixes to intersect_all functions While they prevent legit write past the array boundary error those fixes introduced regression in behavior when having exact max_hits transparent intersections and nothing else. Previous code would have considered such case a totally opaque, but it's not correct. Fixes T48941: Some materials don't get transparent shadows anymore
2016-07-26 14:30:13 +00:00
/* Move on to next entry in intersections array. */
isect_array++;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
num_hits++;
#if BVH_FEATURE(BVH_INSTANCING)
num_hits_in_instance++;
#endif
Cycles: Revert previous fixes to intersect_all functions While they prevent legit write past the array boundary error those fixes introduced regression in behavior when having exact max_hits transparent intersections and nothing else. Previous code would have considered such case a totally opaque, but it's not correct. Fixes T48941: Some materials don't get transparent shadows anymore
2016-07-26 14:30:13 +00:00
isect_array->t = isect_t;
if(num_hits == max_hits) {
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#if BVH_FEATURE(BVH_INSTANCING)
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# if BVH_FEATURE(BVH_MOTION)
Cycles: Fix wrong intersection with motion blur and degenerate object transform
2015-10-08 16:24:54 +00:00
float t_fac = 1.0f / len(transform_direction(&ob_itfm, dir));
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: Fix intersection issues caused by degenerate instance matrix Issue was caused by wrong intersection distance scaling on instance pop, which could cause intersection distance to become zero, confusing following intersection checks.
2015-10-08 15:30:51 +00:00
Transform itfm = object_fetch_transform(kg, object, OBJECT_INVERSE_TRANSFORM);
float t_fac = 1.0f / len(transform_direction(&itfm, dir));
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
for(int i = 0; i < num_hits_in_instance; i++) {
(isect_array-i-1)->t *= t_fac;
}
#endif /* BVH_FEATURE(BVH_INSTANCING) */
return num_hits;
}
}
}
break;
}
#if BVH_FEATURE(BVH_MOTION)
case PRIMITIVE_MOTION_TRIANGLE: {
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
for(; prim_addr < prim_addr2; prim_addr++) {
kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/* Only primitives from volume object. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
uint tri_object = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, prim_addr): object;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
int object_flag = kernel_tex_fetch(__object_flag, tri_object);
if((object_flag & SD_OBJECT_HAS_VOLUME) == 0) {
continue;
}
/* Intersect ray against primitive. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
hit = motion_triangle_intersect(kg, isect_array, P, dir, ray->time, visibility, object, prim_addr);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
if(hit) {
Cycles: Revert previous fixes to intersect_all functions While they prevent legit write past the array boundary error those fixes introduced regression in behavior when having exact max_hits transparent intersections and nothing else. Previous code would have considered such case a totally opaque, but it's not correct. Fixes T48941: Some materials don't get transparent shadows anymore
2016-07-26 14:30:13 +00:00
/* Move on to next entry in intersections array. */
isect_array++;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
num_hits++;
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# if BVH_FEATURE(BVH_INSTANCING)
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
num_hits_in_instance++;
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: Revert previous fixes to intersect_all functions While they prevent legit write past the array boundary error those fixes introduced regression in behavior when having exact max_hits transparent intersections and nothing else. Previous code would have considered such case a totally opaque, but it's not correct. Fixes T48941: Some materials don't get transparent shadows anymore
2016-07-26 14:30:13 +00:00
isect_array->t = isect_t;
if(num_hits == max_hits) {
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# if BVH_FEATURE(BVH_INSTANCING)
# if BVH_FEATURE(BVH_MOTION)
Cycles: Fix wrong intersection with motion blur and degenerate object transform
2015-10-08 16:24:54 +00:00
float t_fac = 1.0f / len(transform_direction(&ob_itfm, dir));
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: Fix intersection issues caused by degenerate instance matrix Issue was caused by wrong intersection distance scaling on instance pop, which could cause intersection distance to become zero, confusing following intersection checks.
2015-10-08 15:30:51 +00:00
Transform itfm = object_fetch_transform(kg, object, OBJECT_INVERSE_TRANSFORM);
float t_fac = 1.0f / len(transform_direction(&itfm, dir));
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
for(int i = 0; i < num_hits_in_instance; i++) {
(isect_array-i-1)->t *= t_fac;
}
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif /* BVH_FEATURE(BVH_INSTANCING) */
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
return num_hits;
}
}
}
break;
}
#endif
}
}
#if BVH_FEATURE(BVH_INSTANCING)
else {
/* Instance push. */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
object = kernel_tex_fetch(__prim_object, -prim_addr-1);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
int object_flag = kernel_tex_fetch(__object_flag, object);
if(object_flag & SD_OBJECT_HAS_VOLUME) {
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# if BVH_FEATURE(BVH_MOTION)
Cycles: Fix wrong intersection with motion blur and degenerate object transform
2015-10-08 16:24:54 +00:00
bvh_instance_motion_push(kg, object, ray, &P, &dir, &idir, &isect_t, &ob_itfm);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
bvh_instance_push(kg, object, ray, &P, &dir, &idir, &isect_t);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
tfar = ssef(isect_t);
idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z));
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
# if BVH_FEATURE(BVH_HAIR)
dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z));
# endif
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# ifdef __KERNEL_AVX2__
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
P_idir = P*idir;
P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
# endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
triangle_intersect_precalc(dir, &isect_precalc);
num_hits_in_instance = 0;
isect_array->t = isect_t;
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
++stack_ptr;
kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
node_addr = kernel_tex_fetch(__object_node, object);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
}
else {
/* Pop. */
object = OBJECT_NONE;
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
node_addr = traversal_stack[stack_ptr].addr;
--stack_ptr;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
}
}
}
#endif /* FEATURE(BVH_INSTANCING) */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
} while(node_addr != ENTRYPOINT_SENTINEL);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
#if BVH_FEATURE(BVH_INSTANCING)
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
if(stack_ptr >= 0) {
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
kernel_assert(object != OBJECT_NONE);
/* Instance pop. */
if(num_hits_in_instance) {
float t_fac;
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# if BVH_FEATURE(BVH_MOTION)
Cycles: Fix wrong intersection with motion blur and degenerate object transform
2015-10-08 16:24:54 +00:00
bvh_instance_motion_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac, &ob_itfm);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
bvh_instance_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
/* Scale isect->t to adjust for instancing. */
for(int i = 0; i < num_hits_in_instance; i++) {
(isect_array-i-1)->t *= t_fac;
}
}
else {
float ignore_t = FLT_MAX;
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# if BVH_FEATURE(BVH_MOTION)
Cycles: Fix wrong intersection with motion blur and degenerate object transform
2015-10-08 16:24:54 +00:00
bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, &ignore_t, &ob_itfm);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# else
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
bvh_instance_pop(kg, object, ray, &P, &dir, &idir, &ignore_t);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
}
Cycles: More tweaks to make specialized BVH traversal matching
2016-09-19 13:29:37 +00:00
isect_t = tmax;
isect_array->t = isect_t;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
if(idir.x >= 0.0f) { near_x = 0; far_x = 1; } else { near_x = 1; far_x = 0; }
if(idir.y >= 0.0f) { near_y = 2; far_y = 3; } else { near_y = 3; far_y = 2; }
if(idir.z >= 0.0f) { near_z = 4; far_z = 5; } else { near_z = 5; far_z = 4; }
tfar = ssef(isect_t);
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
# if BVH_FEATURE(BVH_HAIR)
dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z));
# endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z));
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# ifdef __KERNEL_AVX2__
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
P_idir = P*idir;
P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
Cycles: Cleanup, indent nested preprocessor directives Quite straightforward, main trick is happening in path_source_replace_includes(). Reviewers: brecht, dingto, lukasstockner97, juicyfruit Differential Revision: https://developer.blender.org/D1794
2016-02-12 17:33:43 +00:00
# endif
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__)
org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
# endif
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
triangle_intersect_precalc(dir, &isect_precalc);
object = OBJECT_NONE;
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
node_addr = traversal_stack[stack_ptr].addr;
--stack_ptr;
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
}
#endif /* FEATURE(BVH_INSTANCING) */
Cycles: Cleanup, variables name Using camel case for variables is something what didn't came from our original code, but rather from third party libraries. Let's avoid those as much as possible.
2016-07-11 11:44:19 +00:00
} while(node_addr != ENTRYPOINT_SENTINEL);
Cycles: Record all possible volume intersections for SSS and camera checks This replaces sequential ray moving followed with scene intersection with single BVH traversal, which gives us all possible intersections. Only implemented for CPU, due to qsort and a bigger memory usage on GPU which we rather avoid. GPU still uses the regular bvh volume intersection code, while CPU now uses the new code. This improves render performance for scenes with: a) Camera inside volume mesh b) SSS mesh intersecting a volume mesh/domain In simple volume files (not much geometry) performance is roughly the same (slightly faster). In files with a lot of geometry, the performance increase is larger. bmps.blend with a volume shader and camera inside the mesh, it renders ~10% faster here. Patch by Sergey and myself. Differential Revision: https://developer.blender.org/D1264
2015-04-29 21:21:05 +00:00
return num_hits;
}
Cycles: Implement unaligned nodes BVH traversal This commit implements traversal of unaligned BVH nodes. QBVH traversal is fully SIMD optimized and calculates orientation for all 4 children at a time, regular BVH might probably be optimized a bit more.
2016-07-07 10:23:13 +00:00
#undef NODE_INTERSECT
Copy Permalink