Page MenuHome
Differential D656 Diff 2192 intern/cycles/device/device_cuda.cpp

Changeset View

Standalone View

View Options

intern/cycles/device/device_cuda.cpp

Show First 20 Lines • Show All 764 Lines • ▼ Show 20 Lines else {
cuda_assert(cuModuleGetFunction(&cuShader, cuModule, "kernel_cuda_shader")); cuda_assert(cuModuleGetFunction(&cuShader, cuModule, "kernel_cuda_shader"));
} }
/* do tasks in smaller chunks, so we can cancel it */ /* do tasks in smaller chunks, so we can cancel it */
const int shader_chunk_size = 65536; const int shader_chunk_size = 65536;
const int start = task.shader_x; const int start = task.shader_x;
const int end = task.shader_x + task.shader_w; const int end = task.shader_x + task.shader_w;
for(int shader_x = start; shader_x < end; shader_x += shader_chunk_size) { for(int sample = 0; sample < task.num_samples; sample++) {
if(task.get_cancel()) if(task.get_cancel())
break; break;
for(int shader_x = start; shader_x < end; shader_x += shader_chunk_size) {
int shader_w = min(shader_chunk_size, end - shader_x); int shader_w = min(shader_chunk_size, end - shader_x);
for(int sample = 0; sample < task.num_samples; sample++) {
/* pass in parameters */ /* pass in parameters */
int offset = 0; int offset = 0;
cuda_assert(cuParamSetv(cuShader, offset, &d_input, sizeof(d_input))); cuda_assert(cuParamSetv(cuShader, offset, &d_input, sizeof(d_input)));
offset += sizeof(d_input); offset += sizeof(d_input);
cuda_assert(cuParamSetv(cuShader, offset, &d_output, sizeof(d_output))); cuda_assert(cuParamSetv(cuShader, offset, &d_output, sizeof(d_output)));
offset += sizeof(d_output); offset += sizeof(d_output);
Show All 22 Lines for(int sample = 0; sample < task.num_samples; sample++) {
int xblocks = (shader_w + threads_per_block - 1)/threads_per_block; int xblocks = (shader_w + threads_per_block - 1)/threads_per_block;
cuda_assert(cuFuncSetCacheConfig(cuShader, CU_FUNC_CACHE_PREFER_L1)); cuda_assert(cuFuncSetCacheConfig(cuShader, CU_FUNC_CACHE_PREFER_L1));
cuda_assert(cuFuncSetBlockShape(cuShader, threads_per_block, 1, 1)); cuda_assert(cuFuncSetBlockShape(cuShader, threads_per_block, 1, 1));
cuda_assert(cuLaunchGrid(cuShader, xblocks, 1)); cuda_assert(cuLaunchGrid(cuShader, xblocks, 1));
cuda_assert(cuCtxSynchronize()); cuda_assert(cuCtxSynchronize());
} }
if(task.update_progress_sample)
sergeyUnsubmitted
Not Done
Inline Actions

Suggestion: Make DeviceTask::update_prgoress to use a pointer to render tile insteado f a reference and call it from here with NULL tile. This way you wouldn't worry about calling update to often.

sergey: Suggestion: Make DeviceTask::update_prgoress to use a pointer to render tile insteado f a…
task.update_progress_sample();
} }
cuda_pop_context(); cuda_pop_context();
} }
CUdeviceptr map_pixels(device_ptr mem) CUdeviceptr map_pixels(device_ptr mem)
{ {
if(!background) { if(!background) {
▲ Show 20 Lines • Show All 237 Lines • ▼ Show 20 Lines#endif
public: public:
CUDADeviceTask(CUDADevice *device, DeviceTask& task) CUDADeviceTask(CUDADevice *device, DeviceTask& task)
: DeviceTask(task) : DeviceTask(task)
{ {
run = function_bind(&CUDADevice::thread_run, device, this); run = function_bind(&CUDADevice::thread_run, device, this);
} }
}; };
int task_count(DeviceTask& task)
{
return 1;
}
void task_add(DeviceTask& task) void task_add(DeviceTask& task)
{ {
if(task.type == DeviceTask::FILM_CONVERT) { if(task.type == DeviceTask::FILM_CONVERT) {
/* must be done in main thread due to opengl access */ /* must be done in main thread due to opengl access */
film_convert(task, task.buffer, task.rgba_byte, task.rgba_half); film_convert(task, task.buffer, task.rgba_byte, task.rgba_half);
cuda_push_context(); cuda_push_context();
cuda_assert(cuCtxSynchronize()); cuda_assert(cuCtxSynchronize());
▲ Show 20 Lines • Show All 78 Lines • Show Last 20 Lines