kannonier8/blender
1
0
Fork 0
forked from bartvdbraak/blender
Code Pull Requests Activity
bd10b72ad3
blender/intern/cycles/kernel/geom/geom_qbvh_subsurface.h

321 lines
10 KiB
C
Raw Normal View History

Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
/*
* Adapted from code Copyright 2009-2010 NVIDIA Corporation,
* and code copyright 2009-2012 Intel Corporation
*
* Modifications Copyright 2011-2014, Blender Foundation.
*
* 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 subsurface scattering, 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
*
*/
ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
const Ray *ray,
Intersection *isect_array,
int subsurface_object,
uint *lcg_state,
int max_hits)
{
/* TODO(sergey):
* - Test if pushing distance on the stack helps (for non shadow rays).
* - Separate version for shadow rays.
* - Likely and unlikely for if() statements.
* - SSE for hair.
* - Test restrict attribute for pointers.
*/
/* Traversal stack in CUDA thread-local memory. */
Cycles: Bump stack size for QBVH traversal code Traversal now can push up to 2x of nodes to the stack, so need some tweaks to the stack size.
2014-12-29 08:37:18 +00:00
QBVHStackItem traversalStack[BVH_QSTACK_SIZE];
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[0].addr = ENTRYPOINT_SENTINEL;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
/* Traversal variables in registers. */
int stackPtr = 0;
int nodeAddr = kernel_data.bvh.root;
/* Ray parameters in registers. */
float3 P = ray->P;
float3 dir = bvh_clamp_direction(ray->D);
float3 idir = bvh_inverse_direction(dir);
int object = OBJECT_NONE;
float isect_t = ray->t;
uint num_hits = 0;
#if BVH_FEATURE(BVH_MOTION)
Transform ob_tfm;
#endif
ssef tnear(0.0f), tfar(isect_t);
sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z));
#ifdef __KERNEL_AVX2__
float3 P_idir = P*idir;
sse3f P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
#else
sse3f org = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
#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. */
while(nodeAddr >= 0 && nodeAddr != ENTRYPOINT_SENTINEL) {
ssef dist;
int traverseChild = qbvh_node_intersect(kg,
tnear,
tfar,
#ifdef __KERNEL_AVX2__
P_idir4,
#else
org,
#endif
idir4,
near_x, near_y, near_z,
far_x, far_y, far_z,
nodeAddr,
&dist);
if(traverseChild != 0) {
float4 cnodes = kernel_tex_fetch(__bvh_nodes, nodeAddr*BVH_QNODE_SIZE+6);
/* One child is hit, continue with that child. */
int r = __bscf(traverseChild);
if(traverseChild == 0) {
nodeAddr = __float_as_int(cnodes[r]);
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];
r = __bscf(traverseChild);
int c1 = __float_as_int(cnodes[r]);
float d1 = ((float*)&dist)[r];
if(traverseChild == 0) {
if(d1 < d0) {
nodeAddr = c1;
++stackPtr;
Cycles: Add asserts around BVH stack pushes This way we're kind of safer to troubleshoot possible stack overflow issues.
2014-12-29 09:00:25 +00:00
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[stackPtr].addr = c0;
traversalStack[stackPtr].dist = d0;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
continue;
}
else {
nodeAddr = c0;
++stackPtr;
Cycles: Add asserts around BVH stack pushes This way we're kind of safer to troubleshoot possible stack overflow issues.
2014-12-29 09:00:25 +00:00
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[stackPtr].addr = c1;
traversalStack[stackPtr].dist = d1;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +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.
*/
++stackPtr;
Cycles: Add asserts around BVH stack pushes This way we're kind of safer to troubleshoot possible stack overflow issues.
2014-12-29 09:00:25 +00:00
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[stackPtr].addr = c1;
traversalStack[stackPtr].dist = d1;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
++stackPtr;
Cycles: Add asserts around BVH stack pushes This way we're kind of safer to troubleshoot possible stack overflow issues.
2014-12-29 09:00:25 +00:00
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[stackPtr].addr = c0;
traversalStack[stackPtr].dist = d0;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
/* Three children are hit, push all onto stack and sort 3
* stack items, continue with closest child.
*/
r = __bscf(traverseChild);
int c2 = __float_as_int(cnodes[r]);
float d2 = ((float*)&dist)[r];
if(traverseChild == 0) {
++stackPtr;
Cycles: Add asserts around BVH stack pushes This way we're kind of safer to troubleshoot possible stack overflow issues.
2014-12-29 09:00:25 +00:00
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[stackPtr].addr = c2;
traversalStack[stackPtr].dist = d2;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
qbvh_stack_sort(&traversalStack[stackPtr],
&traversalStack[stackPtr - 1],
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
&traversalStack[stackPtr - 2]);
nodeAddr = traversalStack[stackPtr].addr;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
--stackPtr;
continue;
}
/* Four children are hit, push all onto stack and sort 4
* stack items, continue with closest child.
*/
r = __bscf(traverseChild);
int c3 = __float_as_int(cnodes[r]);
float d3 = ((float*)&dist)[r];
++stackPtr;
Cycles: Add asserts around BVH stack pushes This way we're kind of safer to troubleshoot possible stack overflow issues.
2014-12-29 09:00:25 +00:00
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[stackPtr].addr = c3;
traversalStack[stackPtr].dist = d3;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
++stackPtr;
Cycles: Add asserts around BVH stack pushes This way we're kind of safer to troubleshoot possible stack overflow issues.
2014-12-29 09:00:25 +00:00
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[stackPtr].addr = c2;
traversalStack[stackPtr].dist = d2;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
qbvh_stack_sort(&traversalStack[stackPtr],
&traversalStack[stackPtr - 1],
&traversalStack[stackPtr - 2],
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
&traversalStack[stackPtr - 3]);
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
}
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
nodeAddr = traversalStack[stackPtr].addr;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
--stackPtr;
}
/* If node is leaf, fetch triangle list. */
if(nodeAddr < 0) {
float4 leaf = kernel_tex_fetch(__bvh_nodes, (-nodeAddr-1)*BVH_QNODE_SIZE+6);
int primAddr = __float_as_int(leaf.x);
#if BVH_FEATURE(BVH_INSTANCING)
if(primAddr >= 0) {
#endif
int primAddr2 = __float_as_int(leaf.y);
Cycles: Enable leaf split by primitive type and adopt BVH traversal for this This commit enables BVH leaf nodes split by the primitive type and makes it so BVH traversal code is now aware and benefits from this. As was mentioned in original commit, this change is crucial to be able to do single ray to multiple triangle intersection. But it also appears to give barely visible speedup in some scene. In any case there should be no noticeable slowdown, and this change is what we need to have anyway.
2015-01-10 17:41:49 +00:00
const uint type = __float_as_int(leaf.w);
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
/* Pop. */
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
nodeAddr = traversalStack[stackPtr].addr;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
--stackPtr;
/* Primitive intersection. */
Cycles: Enable leaf split by primitive type and adopt BVH traversal for this This commit enables BVH leaf nodes split by the primitive type and makes it so BVH traversal code is now aware and benefits from this. As was mentioned in original commit, this change is crucial to be able to do single ray to multiple triangle intersection. But it also appears to give barely visible speedup in some scene. In any case there should be no noticeable slowdown, and this change is what we need to have anyway.
2015-01-10 17:41:49 +00:00
switch(type & PRIMITIVE_ALL) {
case PRIMITIVE_TRIANGLE: {
/* Intersect ray against primitive, */
for(; primAddr < primAddr2; primAddr++) {
Cycles: Add leaf primitives sanity check asserts to the kernel This way we'll notice that leaf splitting didn't happen correct pretty easily in debug builds. There'll be absolutely no impact on release builds.
2015-01-11 21:04:35 +00:00
kernel_assert(kernel_tex_fetch(__prim_type, primAddr) == type);
Cycles: Enable leaf split by primitive type and adopt BVH traversal for this This commit enables BVH leaf nodes split by the primitive type and makes it so BVH traversal code is now aware and benefits from this. As was mentioned in original commit, this change is crucial to be able to do single ray to multiple triangle intersection. But it also appears to give barely visible speedup in some scene. In any case there should be no noticeable slowdown, and this change is what we need to have anyway.
2015-01-10 17:41:49 +00:00
/* Only primitives from the same object. */
uint tri_object = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, primAddr): object;
if(tri_object != subsurface_object) {
continue;
}
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
triangle_intersect_subsurface(kg, &isect_precalc, isect_array, P, dir, object, primAddr, isect_t, &num_hits, lcg_state, max_hits);
}
Cycles: Enable leaf split by primitive type and adopt BVH traversal for this This commit enables BVH leaf nodes split by the primitive type and makes it so BVH traversal code is now aware and benefits from this. As was mentioned in original commit, this change is crucial to be able to do single ray to multiple triangle intersection. But it also appears to give barely visible speedup in some scene. In any case there should be no noticeable slowdown, and this change is what we need to have anyway.
2015-01-10 17:41:49 +00:00
break;
}
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
#if BVH_FEATURE(BVH_MOTION)
Cycles: Enable leaf split by primitive type and adopt BVH traversal for this This commit enables BVH leaf nodes split by the primitive type and makes it so BVH traversal code is now aware and benefits from this. As was mentioned in original commit, this change is crucial to be able to do single ray to multiple triangle intersection. But it also appears to give barely visible speedup in some scene. In any case there should be no noticeable slowdown, and this change is what we need to have anyway.
2015-01-10 17:41:49 +00:00
case PRIMITIVE_MOTION_TRIANGLE: {
/* Intersect ray against primitive. */
for(; primAddr < primAddr2; primAddr++) {
Cycles: Add leaf primitives sanity check asserts to the kernel This way we'll notice that leaf splitting didn't happen correct pretty easily in debug builds. There'll be absolutely no impact on release builds.
2015-01-11 21:04:35 +00:00
kernel_assert(kernel_tex_fetch(__prim_type, primAddr) == type);
Cycles: Enable leaf split by primitive type and adopt BVH traversal for this This commit enables BVH leaf nodes split by the primitive type and makes it so BVH traversal code is now aware and benefits from this. As was mentioned in original commit, this change is crucial to be able to do single ray to multiple triangle intersection. But it also appears to give barely visible speedup in some scene. In any case there should be no noticeable slowdown, and this change is what we need to have anyway.
2015-01-10 17:41:49 +00:00
/* Only primitives from the same object. */
uint tri_object = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, primAddr): object;
if(tri_object != subsurface_object) {
continue;
}
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
motion_triangle_intersect_subsurface(kg, isect_array, P, dir, ray->time, object, primAddr, isect_t, &num_hits, lcg_state, max_hits);
}
Cycles: Enable leaf split by primitive type and adopt BVH traversal for this This commit enables BVH leaf nodes split by the primitive type and makes it so BVH traversal code is now aware and benefits from this. As was mentioned in original commit, this change is crucial to be able to do single ray to multiple triangle intersection. But it also appears to give barely visible speedup in some scene. In any case there should be no noticeable slowdown, and this change is what we need to have anyway.
2015-01-10 17:41:49 +00:00
break;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
}
Cycles: Enable leaf split by primitive type and adopt BVH traversal for this This commit enables BVH leaf nodes split by the primitive type and makes it so BVH traversal code is now aware and benefits from this. As was mentioned in original commit, this change is crucial to be able to do single ray to multiple triangle intersection. But it also appears to give barely visible speedup in some scene. In any case there should be no noticeable slowdown, and this change is what we need to have anyway.
2015-01-10 17:41:49 +00:00
#endif
default:
break;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
}
}
#if BVH_FEATURE(BVH_INSTANCING)
else {
/* Instance push. */
if(subsurface_object == kernel_tex_fetch(__prim_object, -primAddr-1)) {
object = subsurface_object;
#if BVH_FEATURE(BVH_MOTION)
bvh_instance_motion_push(kg, object, ray, &P, &dir, &idir, &isect_t, &ob_tfm);
#else
bvh_instance_push(kg, object, ray, &P, &dir, &idir, &isect_t);
#endif
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));
#ifdef __KERNEL_AVX2__
P_idir = P*idir;
P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
#else
org = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
#endif
triangle_intersect_precalc(dir, &isect_precalc);
++stackPtr;
Cycles: Add asserts around BVH stack pushes This way we're kind of safer to troubleshoot possible stack overflow issues.
2014-12-29 09:00:25 +00:00
kernel_assert(stackPtr < BVH_QSTACK_SIZE);
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
traversalStack[stackPtr].addr = ENTRYPOINT_SENTINEL;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
nodeAddr = kernel_tex_fetch(__object_node, object);
}
else {
/* Pop. */
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
nodeAddr = traversalStack[stackPtr].addr;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
--stackPtr;
}
}
}
#endif /* FEATURE(BVH_INSTANCING) */
} while(nodeAddr != ENTRYPOINT_SENTINEL);
#if BVH_FEATURE(BVH_INSTANCING)
if(stackPtr >= 0) {
kernel_assert(object != OBJECT_NONE);
/* Instance pop. */
#if BVH_FEATURE(BVH_MOTION)
bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, &isect_t, &ob_tfm);
#else
bvh_instance_pop(kg, object, ray, &P, &dir, &idir, &isect_t);
#endif
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));
#ifdef __KERNEL_AVX2__
P_idir = P*idir;
P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z);
#else
org = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
#endif
triangle_intersect_precalc(dir, &isect_precalc);
object = OBJECT_NONE;
Cycles: Distance optimization for QBVH This commit implements heuristic which allows to skip nodes pushed to the stack from intersection if distance to them is larger than the distance to the current intersection. This should solve speed regression which i didn't notice in the original QBVH commit (which could have because i had WIP version of this patch applied in my local branch). From quick tests speed seems to be much closer to what is was with regular BVH. There's still some possible code cleanup, but they'll need a bit of assembly code check and now i want to make it so artists can happily use Cycles over the holidays.
2014-12-25 17:40:02 +00:00
nodeAddr = traversalStack[stackPtr].addr;
Cycles: Implement QBVH tree traversal This commit implements traversal for QBVH tree, which is based on the old loop code for traversal itself and Embree for node intersection. This commit also does some changes to the loop inspired by Embree: - Visibility flags are only checked for primitives. Doing visibility check for every node cost quite reasonable amount of time and in most cases those checks are true-positive. Other idea here would be to do visibility checks for leaf nodes only, but this would need to be investigated further. - For minimum hair width we extend all the nodes' bounding boxes. Again doing curve visibility check is quite costly for each of the nodes and those checks returns truth for most of the hierarchy anyway. There are number of possible optimization still, but current state is good enough in terms it makes rendering faster a little bit after recent watertight commit. Currently QBVH is only implemented for CPU with SSE2 support at least. All other devices would need to be supported later (if that'd make sense from performance point of view). The code is enabled for compilation in kernel. but blender wouldn't use it still.
2014-12-16 15:11:37 +00:00
--stackPtr;
}
#endif /* FEATURE(BVH_INSTANCING) */
} while(nodeAddr != ENTRYPOINT_SENTINEL);
return num_hits;
}
Copy Permalink