Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
/*
|
|
|
|
* Copyright 2011-2015 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.
|
|
|
|
*/
|
|
|
|
|
2015-05-21 12:40:04 +00:00
|
|
|
#include "kernel_split_common.h"
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
|
2015-05-26 14:12:49 +00:00
|
|
|
/* Note on kernel_direct_lighting kernel.
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
* This is the eighth kernel in the ray tracing logic. This is the seventh
|
|
|
|
* of the path iteration kernels. This kernel takes care of direct lighting
|
|
|
|
* logic. However, the "shadow ray cast" part of direct lighting is handled
|
|
|
|
* in the next kernel.
|
|
|
|
*
|
|
|
|
* This kernels determines the rays for which a shadow_blocked() function associated with direct lighting should be executed.
|
|
|
|
* Those rays for which a shadow_blocked() function for direct-lighting must be executed, are marked with flag RAY_SHADOW_RAY_CAST_DL and
|
|
|
|
* enqueued into the queue QUEUE_SHADOW_RAY_CAST_DL_RAYS
|
|
|
|
*
|
|
|
|
* The input and output are as follows,
|
|
|
|
*
|
2015-05-21 12:40:04 +00:00
|
|
|
* rng_coop -----------------------------------------|--- kernel_direct_lighting --|--- BSDFEval_coop
|
|
|
|
* PathState_coop -----------------------------------| |--- ISLamp_coop
|
2015-10-29 16:44:36 +00:00
|
|
|
* sd -----------------------------------------------| |--- LightRay_coop
|
2015-05-21 12:40:04 +00:00
|
|
|
* ray_state ----------------------------------------| |--- ray_state
|
|
|
|
* Queue_data (QUEUE_ACTIVE_AND_REGENERATED_RAYS) ---| |
|
2015-11-01 11:51:32 +00:00
|
|
|
* kg (globals) -------------------------------------| |
|
2015-05-21 12:40:04 +00:00
|
|
|
* queuesize ----------------------------------------| |
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
*
|
|
|
|
* Note on Queues :
|
|
|
|
* This kernel only reads from the QUEUE_ACTIVE_AND_REGENERATED_RAYS queue and processes
|
|
|
|
* only the rays of state RAY_ACTIVE; If a ray needs to execute the corresponding shadow_blocked
|
|
|
|
* part, after direct lighting, the ray is marked with RAY_SHADOW_RAY_CAST_DL flag.
|
|
|
|
*
|
|
|
|
* State of queues when this kernel is called :
|
|
|
|
* state of queues QUEUE_ACTIVE_AND_REGENERATED_RAYS and QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS will be same
|
|
|
|
* before and after this kernel call.
|
|
|
|
* QUEUE_SHADOW_RAY_CAST_DL_RAYS queue will be filled with rays for which a shadow_blocked function must be executed, after this
|
|
|
|
* kernel call. Before this kernel call the QUEUE_SHADOW_RAY_CAST_DL_RAYS will be empty.
|
|
|
|
*/
|
2015-05-26 14:12:49 +00:00
|
|
|
ccl_device char kernel_direct_lighting(
|
2015-10-29 16:44:36 +00:00
|
|
|
KernelGlobals *kg,
|
|
|
|
ShaderData *sd, /* Required for direct lighting */
|
2015-05-26 14:12:49 +00:00
|
|
|
ccl_global uint *rng_coop, /* Required for direct lighting */
|
|
|
|
ccl_global PathState *PathState_coop, /* Required for direct lighting */
|
|
|
|
ccl_global int *ISLamp_coop, /* Required for direct lighting */
|
|
|
|
ccl_global Ray *LightRay_coop, /* Required for direct lighting */
|
|
|
|
ccl_global BsdfEval *BSDFEval_coop, /* Required for direct lighting */
|
|
|
|
ccl_global char *ray_state, /* Denotes the state of each ray */
|
|
|
|
int ray_index)
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
{
|
|
|
|
char enqueue_flag = 0;
|
2015-05-26 14:12:49 +00:00
|
|
|
if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) {
|
|
|
|
ccl_global PathState *state = &PathState_coop[ray_index];
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
|
2015-05-26 14:12:49 +00:00
|
|
|
/* direct lighting */
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
#ifdef __EMISSION__
|
2015-05-26 19:13:32 +00:00
|
|
|
if((kernel_data.integrator.use_direct_light &&
|
|
|
|
(ccl_fetch(sd, flag) & SD_BSDF_HAS_EVAL)))
|
|
|
|
{
|
|
|
|
/* Sample illumination from lights to find path contribution. */
|
2015-05-26 14:12:49 +00:00
|
|
|
ccl_global RNG* rng = &rng_coop[ray_index];
|
|
|
|
float light_t = path_state_rng_1D(kg, rng, state, PRNG_LIGHT);
|
|
|
|
float light_u, light_v;
|
|
|
|
path_state_rng_2D(kg, rng, state, PRNG_LIGHT_U, &light_u, &light_v);
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
|
2015-05-26 14:12:49 +00:00
|
|
|
LightSample ls;
|
2015-05-26 19:13:32 +00:00
|
|
|
light_sample(kg,
|
|
|
|
light_t, light_u, light_v,
|
|
|
|
ccl_fetch(sd, time),
|
|
|
|
ccl_fetch(sd, P),
|
|
|
|
state->bounce,
|
|
|
|
&ls);
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
|
2015-05-26 14:12:49 +00:00
|
|
|
Ray light_ray;
|
2015-05-14 13:46:26 +00:00
|
|
|
#ifdef __OBJECT_MOTION__
|
2015-05-26 14:12:49 +00:00
|
|
|
light_ray.time = ccl_fetch(sd, time);
|
2015-05-14 13:46:26 +00:00
|
|
|
#endif
|
|
|
|
|
2015-05-26 14:12:49 +00:00
|
|
|
BsdfEval L_light;
|
|
|
|
bool is_lamp;
|
2016-05-22 20:35:47 +00:00
|
|
|
if(direct_emission(kg, sd, kg->sd_input, &ls, state, &light_ray, &L_light, &is_lamp)) {
|
2015-05-26 19:13:32 +00:00
|
|
|
/* Write intermediate data to global memory to access from
|
|
|
|
* the next kernel.
|
|
|
|
*/
|
2015-05-26 14:12:49 +00:00
|
|
|
LightRay_coop[ray_index] = light_ray;
|
|
|
|
BSDFEval_coop[ray_index] = L_light;
|
|
|
|
ISLamp_coop[ray_index] = is_lamp;
|
2015-05-26 19:13:32 +00:00
|
|
|
/* Mark ray state for next shadow kernel. */
|
2015-05-26 14:12:49 +00:00
|
|
|
ADD_RAY_FLAG(ray_state, ray_index, RAY_SHADOW_RAY_CAST_DL);
|
|
|
|
enqueue_flag = 1;
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
}
|
|
|
|
}
|
2015-05-26 19:13:32 +00:00
|
|
|
#endif /* __EMISSION__ */
|
2015-05-26 14:12:49 +00:00
|
|
|
}
|
|
|
|
return enqueue_flag;
|
Cycles: OpenCL kernel split
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.
Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.
Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.
This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.
More feature will be enabled once they're ported to the split kernel and
tested.
Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.
Based on the research paper:
https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf
Here's the documentation:
https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit
Design discussion of the patch:
https://developer.blender.org/T44197
Differential Revision: https://developer.blender.org/D1200
2015-05-09 14:34:30 +00:00
|
|
|
}
|