forked from bartvdbraak/blender
ecdfb465cc
On user level this fixes dead-lock of OpenCL render on Intel Iris GPUs.
Note that this patch does not include change in the logic which allows
or disallows OpenCL platforms to be used, that will happen after the
kernel fix is known to be fine for the currently officially supported
platforms.
The dead-lock was caused by wrong usage of memory barriers: as per the
OpenCL specification the barrier is to be executed by the entire work
group. This means, that the following code is invalid:
void foo() {
if (some_condition) {
return;
}
barrier(CLK_LOCAL_MEM_FENCE);
}
void bar() {
foo();
}
The Cycles code was mentioning this as an invalid code on CPU, while in
fact this is invalid as per specification. From the implementation side
this change removes the ifdefs around the CPU-only barrier logic, and
is implementing similar logic in the shader setup kernel.
Tested on NUC8i7HVK NUC.
The root cause of the dead-lock was identified by Max Dmitrichenko.
There is no measurable difference in performance of currently supported
OpenCL platforms.
Differential Revision: https://developer.blender.org/D9039
155 lines
6.3 KiB
C
155 lines
6.3 KiB
C
/*
|
|
* 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.
|
|
*/
|
|
|
|
CCL_NAMESPACE_BEGIN
|
|
|
|
/* This kernel takes care of rays that hit the background (sceneintersect
|
|
* kernel), and for the rays of state RAY_UPDATE_BUFFER it updates the ray's
|
|
* accumulated radiance in the output buffer. This kernel also takes care of
|
|
* rays that have been determined to-be-regenerated.
|
|
*
|
|
* We will empty QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS queue in this kernel.
|
|
*
|
|
* Typically all rays that are in state RAY_HIT_BACKGROUND, RAY_UPDATE_BUFFER
|
|
* will be eventually set to RAY_TO_REGENERATE state in this kernel.
|
|
* Finally all rays of ray_state RAY_TO_REGENERATE will be regenerated and put
|
|
* in queue QUEUE_ACTIVE_AND_REGENERATED_RAYS.
|
|
*
|
|
* State of queues when this kernel is called:
|
|
* At entry,
|
|
* - QUEUE_ACTIVE_AND_REGENERATED_RAYS will be filled with RAY_ACTIVE rays.
|
|
* - QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS will be filled with
|
|
* RAY_UPDATE_BUFFER, RAY_HIT_BACKGROUND, RAY_TO_REGENERATE rays.
|
|
* At exit,
|
|
* - QUEUE_ACTIVE_AND_REGENERATED_RAYS will be filled with RAY_ACTIVE and
|
|
* RAY_REGENERATED rays.
|
|
* - QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS will be empty.
|
|
*/
|
|
ccl_device void kernel_buffer_update(KernelGlobals *kg,
|
|
ccl_local_param unsigned int *local_queue_atomics)
|
|
{
|
|
if (ccl_local_id(0) == 0 && ccl_local_id(1) == 0) {
|
|
*local_queue_atomics = 0;
|
|
}
|
|
ccl_barrier(CCL_LOCAL_MEM_FENCE);
|
|
|
|
int ray_index = ccl_global_id(1) * ccl_global_size(0) + ccl_global_id(0);
|
|
if (ray_index == 0) {
|
|
/* We will empty this queue in this kernel. */
|
|
kernel_split_params.queue_index[QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS] = 0;
|
|
}
|
|
char enqueue_flag = 0;
|
|
ray_index = get_ray_index(kg,
|
|
ray_index,
|
|
QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS,
|
|
kernel_split_state.queue_data,
|
|
kernel_split_params.queue_size,
|
|
1);
|
|
|
|
if (ray_index != QUEUE_EMPTY_SLOT) {
|
|
ccl_global char *ray_state = kernel_split_state.ray_state;
|
|
ccl_global PathState *state = &kernel_split_state.path_state[ray_index];
|
|
PathRadiance *L = &kernel_split_state.path_radiance[ray_index];
|
|
ccl_global Ray *ray = &kernel_split_state.ray[ray_index];
|
|
ccl_global float3 *throughput = &kernel_split_state.throughput[ray_index];
|
|
bool ray_was_updated = false;
|
|
|
|
if (IS_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER)) {
|
|
ray_was_updated = true;
|
|
uint sample = state->sample;
|
|
uint buffer_offset = kernel_split_state.buffer_offset[ray_index];
|
|
ccl_global float *buffer = kernel_split_params.tile.buffer + buffer_offset;
|
|
|
|
/* accumulate result in output buffer */
|
|
kernel_write_result(kg, buffer, sample, L);
|
|
|
|
ASSIGN_RAY_STATE(ray_state, ray_index, RAY_TO_REGENERATE);
|
|
}
|
|
|
|
if (kernel_data.film.cryptomatte_passes) {
|
|
/* Make sure no thread is writing to the buffers. */
|
|
ccl_barrier(CCL_LOCAL_MEM_FENCE);
|
|
if (ray_was_updated && state->sample - 1 == kernel_data.integrator.aa_samples) {
|
|
uint buffer_offset = kernel_split_state.buffer_offset[ray_index];
|
|
ccl_global float *buffer = kernel_split_params.tile.buffer + buffer_offset;
|
|
ccl_global float *cryptomatte_buffer = buffer + kernel_data.film.pass_cryptomatte;
|
|
kernel_sort_id_slots(cryptomatte_buffer, 2 * kernel_data.film.cryptomatte_depth);
|
|
}
|
|
}
|
|
|
|
if (IS_STATE(ray_state, ray_index, RAY_TO_REGENERATE)) {
|
|
/* We have completed current work; So get next work */
|
|
ccl_global uint *work_pools = kernel_split_params.work_pools;
|
|
uint total_work_size = kernel_split_params.total_work_size;
|
|
uint work_index;
|
|
|
|
if (!get_next_work(kg, work_pools, total_work_size, ray_index, &work_index)) {
|
|
/* If work is invalid, this means no more work is available and the thread may exit */
|
|
ASSIGN_RAY_STATE(ray_state, ray_index, RAY_INACTIVE);
|
|
}
|
|
|
|
if (IS_STATE(ray_state, ray_index, RAY_TO_REGENERATE)) {
|
|
ccl_global WorkTile *tile = &kernel_split_params.tile;
|
|
uint x, y, sample;
|
|
get_work_pixel(tile, work_index, &x, &y, &sample);
|
|
|
|
/* Store buffer offset for writing to passes. */
|
|
uint buffer_offset = (tile->offset + x + y * tile->stride) * kernel_data.film.pass_stride;
|
|
kernel_split_state.buffer_offset[ray_index] = buffer_offset;
|
|
|
|
/* Initialize random numbers and ray. */
|
|
uint rng_hash;
|
|
kernel_path_trace_setup(kg, sample, x, y, &rng_hash, ray);
|
|
|
|
if (ray->t != 0.0f) {
|
|
/* Initialize throughput, path radiance, Ray, PathState;
|
|
* These rays proceed with path-iteration.
|
|
*/
|
|
*throughput = make_float3(1.0f, 1.0f, 1.0f);
|
|
path_radiance_init(kg, L);
|
|
path_state_init(kg,
|
|
AS_SHADER_DATA(&kernel_split_state.sd_DL_shadow[ray_index]),
|
|
state,
|
|
rng_hash,
|
|
sample,
|
|
ray);
|
|
#ifdef __SUBSURFACE__
|
|
kernel_path_subsurface_init_indirect(&kernel_split_state.ss_rays[ray_index]);
|
|
#endif
|
|
ASSIGN_RAY_STATE(ray_state, ray_index, RAY_REGENERATED);
|
|
enqueue_flag = 1;
|
|
}
|
|
else {
|
|
ASSIGN_RAY_STATE(ray_state, ray_index, RAY_TO_REGENERATE);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/* Enqueue RAY_REGENERATED rays into QUEUE_ACTIVE_AND_REGENERATED_RAYS;
|
|
* These rays will be made active during next SceneIntersectkernel.
|
|
*/
|
|
enqueue_ray_index_local(ray_index,
|
|
QUEUE_ACTIVE_AND_REGENERATED_RAYS,
|
|
enqueue_flag,
|
|
kernel_split_params.queue_size,
|
|
local_queue_atomics,
|
|
kernel_split_state.queue_data,
|
|
kernel_split_params.queue_index);
|
|
}
|
|
|
|
CCL_NAMESPACE_END
|