void thread_run(DeviceTask *task)
{
flush_texture_buffers();
if(task->type == DeviceTask::FILM_CONVERT) {
film_convert(*task, task->buffer, task->rgba_byte, task->rgba_half);
}
else if(task->type == DeviceTask::SHADER) {
shader(*task);
}
else if(task->type == DeviceTask::RENDER) {
RenderTile tile;
DenoisingTask denoising(this);
/* Allocate buffer for kernel globals */
device_only_memory<KernelGlobalsDummy> kgbuffer(this, "kernel_globals");
kgbuffer.alloc_to_device(1);
/* Keep rendering tiles until done. */
while(task->acquire_tile(this, tile)) {
if(tile.task == RenderTile::PATH_TRACE) {
assert(tile.task == RenderTile::PATH_TRACE);
scoped_timer timer(&tile.buffers->render_time);
split_kernel->path_trace(task,
tile,
kgbuffer,
*const_mem_map["__data"]);
/* Complete kernel execution before release tile. */
/* This helps in multi-device render;
* The device that reaches the critical-section function
* release_tile waits (stalling other devices from entering
* release_tile) for all kernels to complete. If device1 (a
* slow-render device) reaches release_tile first then it would
* stall device2 (a fast-render device) from proceeding to render
* next tile.
*/
clFinish(cqCommandQueue);
}
else if(tile.task == RenderTile::DENOISE) {
tile.sample = tile.start_sample + tile.num_samples;
denoise(tile, denoising, *task);
task->update_progress(&tile, tile.w*tile.h);
}
task->release_tile(tile);
}
kgbuffer.free();
}
}
void thread_render(DeviceTask &task)
{
if (task_pool.canceled()) {
if (task.need_finish_queue == false)
return;
}
/* allocate buffer for kernel globals */
device_only_memory<KernelGlobals> kgbuffer(this, "kernel_globals");
kgbuffer.alloc_to_device(1);
KernelGlobals *kg = new ((void *)kgbuffer.device_pointer)
KernelGlobals(thread_kernel_globals_init());
profiler.add_state(&kg->profiler);
CPUSplitKernel *split_kernel = NULL;
if (use_split_kernel) {
split_kernel = new CPUSplitKernel(this);
if (!split_kernel->load_kernels(requested_features)) {
thread_kernel_globals_free((KernelGlobals *)kgbuffer.device_pointer);
kgbuffer.free();
delete split_kernel;
return;
}
}
RenderTile tile;
DenoisingTask denoising(this, task);
denoising.profiler = &kg->profiler;
while (task.acquire_tile(this, tile)) {
if (tile.task == RenderTile::PATH_TRACE) {
if (use_split_kernel) {
device_only_memory<uchar> void_buffer(this, "void_buffer");
split_kernel->path_trace(&task, tile, kgbuffer, void_buffer);
}
else {
path_trace(task, tile, kg);
}
}
else if (tile.task == RenderTile::DENOISE) {
denoise(denoising, tile);
task.update_progress(&tile, tile.w * tile.h);
}
task.release_tile(tile);
if (task_pool.canceled()) {
if (task.need_finish_queue == false)
break;
}
}
profiler.remove_state(&kg->profiler);
thread_kernel_globals_free((KernelGlobals *)kgbuffer.device_pointer);
kg->~KernelGlobals();
kgbuffer.free();
delete split_kernel;
}
