Page MenuHome
Differential D656 Diff 2127 intern/cycles/device/device_cuda.cpp

Changeset View

Standalone View

View Options

intern/cycles/device/device_cuda.cpp

Show First 20 Lines • Show All 770 Lines • ▼ Show 20 Lines void shader(DeviceTask& task)
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 shader_x = start; shader_x < end; shader_x += shader_chunk_size) {
if(task.get_cancel()) if(task.get_cancel())
break; break;
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);
int shader_eval_type = task.shader_eval_type; int shader_eval_type = task.shader_eval_type;
offset = align_up(offset, __alignof(shader_eval_type)); offset = align_up(offset, __alignof(shader_eval_type));
cuda_assert(cuParamSeti(cuShader, offset, task.shader_eval_type)); cuda_assert(cuParamSeti(cuShader, offset, task.shader_eval_type));
offset += sizeof(task.shader_eval_type); offset += sizeof(task.shader_eval_type);
cuda_assert(cuParamSeti(cuShader, offset, shader_x)); cuda_assert(cuParamSeti(cuShader, offset, shader_x));
offset += sizeof(shader_x); offset += sizeof(shader_x);
cuda_assert(cuParamSeti(cuShader, offset, shader_w)); cuda_assert(cuParamSeti(cuShader, offset, shader_w));
offset += sizeof(shader_w); offset += sizeof(shader_w);
cuda_assert(cuParamSeti(cuShader, offset, sample)); cuda_assert(cuParamSeti(cuShader, offset, task.sample));
offset += sizeof(sample); offset += sizeof(task.sample);
cuda_assert(cuParamSetSize(cuShader, offset)); cuda_assert(cuParamSetSize(cuShader, offset));
/* launch kernel */ /* launch kernel */
int threads_per_block; int threads_per_block;
cuda_assert(cuFuncGetAttribute(&threads_per_block, CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK, cuShader)); cuda_assert(cuFuncGetAttribute(&threads_per_block, CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK, cuShader));
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());
} }
}
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…
cuda_pop_context(); cuda_pop_context();
} }
CUdeviceptr map_pixels(device_ptr mem) CUdeviceptr map_pixels(device_ptr mem)
{ {
if(!background) { if(!background) {
PixelMem pmem = pixel_mem_map[mem]; PixelMem pmem = pixel_mem_map[mem];
CUdeviceptr buffer; CUdeviceptr buffer;
▲ Show 20 Lines • Show All 329 Lines • Show Last 20 Lines