diff options
Diffstat (limited to 'intern/cycles/device')
-rw-r--r-- | intern/cycles/device/device_cpu.cpp | 13 | ||||
-rw-r--r-- | intern/cycles/device/device_cuda.cpp | 13 | ||||
-rw-r--r-- | intern/cycles/device/device_denoising.cpp | 5 | ||||
-rw-r--r-- | intern/cycles/device/device_denoising.h | 1 | ||||
-rw-r--r-- | intern/cycles/device/opencl/opencl.h | 2 | ||||
-rw-r--r-- | intern/cycles/device/opencl/opencl_base.cpp | 48 |
6 files changed, 22 insertions, 60 deletions
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp index c54ac1e5933..be0dcc20755 100644 --- a/intern/cycles/device/device_cpu.cpp +++ b/intern/cycles/device/device_cpu.cpp @@ -459,18 +459,6 @@ public: } }; - bool denoising_set_tile_info(device_ptr *buffers, DenoisingTask *task) - { - TileInfo *tile_info = (TileInfo*) task->tile_info_mem.host_pointer; - for(int i = 0; i < 9; i++) { - tile_info->buffers[i] = buffers[i]; - } - - task->tile_info_mem.copy_to_device(); - - return true; - } - bool denoising_non_local_means(device_ptr image_ptr, device_ptr guide_ptr, device_ptr variance_ptr, device_ptr out_ptr, DenoisingTask *task) { @@ -722,7 +710,6 @@ public: denoising.functions.combine_halves = function_bind(&CPUDevice::denoising_combine_halves, this, _1, _2, _3, _4, _5, _6, &denoising); denoising.functions.get_feature = function_bind(&CPUDevice::denoising_get_feature, this, _1, _2, _3, _4, &denoising); denoising.functions.detect_outliers = function_bind(&CPUDevice::denoising_detect_outliers, this, _1, _2, _3, _4, &denoising); - denoising.functions.set_tile_info = function_bind(&CPUDevice::denoising_set_tile_info, this, _1, &denoising); denoising.filter_area = make_int4(tile.x, tile.y, tile.w, tile.h); denoising.render_buffer.samples = tile.sample; diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp index c8e0ea4d896..8294af340e8 100644 --- a/intern/cycles/device/device_cuda.cpp +++ b/intern/cycles/device/device_cuda.cpp @@ -1251,18 +1251,6 @@ public: } } - bool denoising_set_tile_info(device_ptr *buffers, DenoisingTask *task) - { - TileInfo *tile_info = (TileInfo*) task->tile_info_mem.host_pointer; - for(int i = 0; i < 9; i++) { - tile_info->buffers[i] = buffers[i]; - } - - task->tile_info_mem.copy_to_device(); - - return !have_error(); - } - #define CUDA_GET_BLOCKSIZE(func, w, h) \ int threads_per_block; \ cuda_assert(cuFuncGetAttribute(&threads_per_block, CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK, func)); \ @@ -1622,7 +1610,6 @@ public: denoising.functions.combine_halves = function_bind(&CUDADevice::denoising_combine_halves, this, _1, _2, _3, _4, _5, _6, &denoising); denoising.functions.get_feature = function_bind(&CUDADevice::denoising_get_feature, this, _1, _2, _3, _4, &denoising); denoising.functions.detect_outliers = function_bind(&CUDADevice::denoising_detect_outliers, this, _1, _2, _3, _4, &denoising); - denoising.functions.set_tile_info = function_bind(&CUDADevice::denoising_set_tile_info, this, _1, &denoising); denoising.filter_area = make_int4(rtile.x, rtile.y, rtile.w, rtile.h); denoising.render_buffer.samples = rtile.sample; diff --git a/intern/cycles/device/device_denoising.cpp b/intern/cycles/device/device_denoising.cpp index 318b14ab499..c0d4634262d 100644 --- a/intern/cycles/device/device_denoising.cpp +++ b/intern/cycles/device/device_denoising.cpp @@ -62,11 +62,10 @@ void DenoisingTask::set_render_buffer(RenderTile *rtiles) { tile_info = (TileInfo*) tile_info_mem.alloc(sizeof(TileInfo)/sizeof(int)); - device_ptr buffers[9]; for(int i = 0; i < 9; i++) { - buffers[i] = rtiles[i].buffer; tile_info->offsets[i] = rtiles[i].offset; tile_info->strides[i] = rtiles[i].stride; + tile_info->buffers[i] = rtiles[i].buffer; } tile_info->x[0] = rtiles[3].x; tile_info->x[1] = rtiles[4].x; @@ -81,7 +80,7 @@ void DenoisingTask::set_render_buffer(RenderTile *rtiles) target_buffer.stride = rtiles[9].stride; target_buffer.ptr = rtiles[9].buffer; - functions.set_tile_info(buffers); + tile_info_mem.copy_to_device(); } void DenoisingTask::setup_denoising_buffer() diff --git a/intern/cycles/device/device_denoising.h b/intern/cycles/device/device_denoising.h index 1063d43d039..e0da52867f1 100644 --- a/intern/cycles/device/device_denoising.h +++ b/intern/cycles/device/device_denoising.h @@ -89,7 +89,6 @@ public: device_ptr depth_ptr, device_ptr output_ptr )> detect_outliers; - function<bool(device_ptr*)> set_tile_info; function<void(RenderTile *rtiles)> map_neighbor_tiles; function<void(RenderTile *rtiles)> unmap_neighbor_tiles; } functions; diff --git a/intern/cycles/device/opencl/opencl.h b/intern/cycles/device/opencl/opencl.h index c550d738bd5..22e0503365c 100644 --- a/intern/cycles/device/opencl/opencl.h +++ b/intern/cycles/device/opencl/opencl.h @@ -436,8 +436,6 @@ protected: device_ptr depth_ptr, device_ptr output_ptr, DenoisingTask *task); - bool denoising_set_tile_info(device_ptr *buffers, - DenoisingTask *task); device_ptr mem_alloc_sub_ptr(device_memory& mem, int offset, int size); void mem_free_sub_ptr(device_ptr ptr); diff --git a/intern/cycles/device/opencl/opencl_base.cpp b/intern/cycles/device/opencl/opencl_base.cpp index 50d371e4c55..9a50d217321 100644 --- a/intern/cycles/device/opencl/opencl_base.cpp +++ b/intern/cycles/device/opencl/opencl_base.cpp @@ -246,7 +246,6 @@ bool OpenCLDeviceBase::load_kernels(const DeviceRequestedFeatures& requested_fea denoising_program.add_kernel(ustring("filter_nlm_normalize")); denoising_program.add_kernel(ustring("filter_nlm_construct_gramian")); denoising_program.add_kernel(ustring("filter_finalize")); - denoising_program.add_kernel(ustring("filter_set_tile_info")); vector<OpenCLProgram*> programs; programs.push_back(&base_program); @@ -981,9 +980,16 @@ bool OpenCLDeviceBase::denoising_divide_shadow(device_ptr a_ptr, cl_kernel ckFilterDivideShadow = denoising_program(ustring("filter_divide_shadow")); - kernel_set_args(ckFilterDivideShadow, 0, - task->render_buffer.samples, - tile_info_mem, + int arg_ofs = kernel_set_args(ckFilterDivideShadow, 0, + task->render_buffer.samples, + tile_info_mem); + cl_mem buffers[9]; + for(int i = 0; i < 9; i++) { + buffers[i] = CL_MEM_PTR(task->tile_info->buffers[i]); + arg_ofs += kernel_set_args(ckFilterDivideShadow, arg_ofs, + buffers[i]); + } + kernel_set_args(ckFilterDivideShadow, arg_ofs, a_mem, b_mem, sample_variance_mem, @@ -1012,9 +1018,16 @@ bool OpenCLDeviceBase::denoising_get_feature(int mean_offset, cl_kernel ckFilterGetFeature = denoising_program(ustring("filter_get_feature")); - kernel_set_args(ckFilterGetFeature, 0, - task->render_buffer.samples, - tile_info_mem, + int arg_ofs = kernel_set_args(ckFilterGetFeature, 0, + task->render_buffer.samples, + tile_info_mem); + cl_mem buffers[9]; + for(int i = 0; i < 9; i++) { + buffers[i] = CL_MEM_PTR(task->tile_info->buffers[i]); + arg_ofs += kernel_set_args(ckFilterGetFeature, arg_ofs, + buffers[i]); + } + kernel_set_args(ckFilterGetFeature, arg_ofs, mean_offset, variance_offset, mean_mem, @@ -1056,29 +1069,8 @@ bool OpenCLDeviceBase::denoising_detect_outliers(device_ptr image_ptr, return true; } -bool OpenCLDeviceBase::denoising_set_tile_info(device_ptr *buffers, - DenoisingTask *task) -{ - task->tile_info_mem.copy_to_device(); - - cl_mem tile_info_mem = CL_MEM_PTR(task->tile_info_mem.device_pointer); - - cl_kernel ckFilterSetTileInfo = denoising_program(ustring("filter_set_tile_info")); - - kernel_set_args(ckFilterSetTileInfo, 0, tile_info_mem); - for(int i = 0; i < 9; i++) { - cl_mem buffer_mem = CL_MEM_PTR(buffers[i]); - kernel_set_args(ckFilterSetTileInfo, i+1, buffer_mem); - } - - enqueue_kernel(ckFilterSetTileInfo, 1, 1); - - return true; -} - void OpenCLDeviceBase::denoise(RenderTile &rtile, DenoisingTask& denoising) { - denoising.functions.set_tile_info = function_bind(&OpenCLDeviceBase::denoising_set_tile_info, this, _1, &denoising); denoising.functions.construct_transform = function_bind(&OpenCLDeviceBase::denoising_construct_transform, this, &denoising); denoising.functions.reconstruct = function_bind(&OpenCLDeviceBase::denoising_reconstruct, this, _1, _2, _3, &denoising); denoising.functions.divide_shadow = function_bind(&OpenCLDeviceBase::denoising_divide_shadow, this, _1, _2, _3, _4, _5, &denoising); |