CCL_NAMESPACE_BEGIN
+DenoisingTask::DenoisingTask(Device *device)
+: tiles_mem(device, "denoising tiles_mem", MEM_READ_WRITE),
+ storage(device),
+ buffer(device),
+ device(device)
+{
+}
+
+DenoisingTask::~DenoisingTask()
+{
+ storage.XtWX.free();
+ storage.XtWY.free();
+ storage.transform.free();
+ storage.rank.free();
+ storage.temporary_1.free();
+ storage.temporary_2.free();
+ storage.temporary_color.free();
+ buffer.mem.free();
+ tiles_mem.free();
+}
+
void DenoisingTask::init_from_devicetask(const DeviceTask &task)
{
radius = task.denoising_radius;
render_buffer.denoising_clean_offset = task.pass_denoising_clean;
/* Expand filter_area by radius pixels and clamp the result to the extent of the neighboring tiles */
- rect = make_int4(max(tiles->x[0], filter_area.x - radius),
- max(tiles->y[0], filter_area.y - radius),
- min(tiles->x[3], filter_area.x + filter_area.z + radius),
- min(tiles->y[3], filter_area.y + filter_area.w + radius));
+ rect = rect_from_shape(filter_area.x, filter_area.y, filter_area.z, filter_area.w);
+ rect = rect_expand(rect, radius);
+ rect = rect_clip(rect, make_int4(tiles->x[0], tiles->y[0], tiles->x[3], tiles->y[3]));
}
void DenoisingTask::tiles_from_rendertiles(RenderTile *rtiles)
{
- tiles = (TilesInfo*) tiles_mem.resize(sizeof(TilesInfo)/sizeof(int));
+ tiles = (TilesInfo*) tiles_mem.alloc(sizeof(TilesInfo)/sizeof(int));
device_ptr buffers[9];
for(int i = 0; i < 9; i++) {
{
/* Allocate denoising buffer. */
buffer.passes = 14;
- buffer.w = align_up(rect.z - rect.x, 4);
+ buffer.width = rect.z - rect.x;
+ buffer.stride = align_up(buffer.width, 4);
buffer.h = rect.w - rect.y;
- buffer.pass_stride = align_up(buffer.w * buffer.h, divide_up(device->mem_address_alignment(), sizeof(float)));
- buffer.mem.resize(buffer.pass_stride * buffer.passes);
- device->mem_alloc("Denoising Pixel Buffer", buffer.mem, MEM_READ_WRITE);
+ buffer.pass_stride = align_up(buffer.stride * buffer.h, divide_up(device->mem_address_alignment(), sizeof(float)));
+ buffer.mem.alloc_to_device(buffer.pass_stride * buffer.passes, false);
device_ptr null_ptr = (device_ptr) 0;
/* Prefilter shadow feature. */
{
- device_sub_ptr unfiltered_a (device, buffer.mem, 0, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr unfiltered_b (device, buffer.mem, 1*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr sample_var (device, buffer.mem, 2*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr sample_var_var (device, buffer.mem, 3*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr buffer_var (device, buffer.mem, 5*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr filtered_var (device, buffer.mem, 6*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr nlm_temporary_1(device, buffer.mem, 7*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr nlm_temporary_2(device, buffer.mem, 8*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr nlm_temporary_3(device, buffer.mem, 9*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
+ device_sub_ptr unfiltered_a (buffer.mem, 0, buffer.pass_stride);
+ device_sub_ptr unfiltered_b (buffer.mem, 1*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr sample_var (buffer.mem, 2*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr sample_var_var (buffer.mem, 3*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr buffer_var (buffer.mem, 5*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr filtered_var (buffer.mem, 6*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr nlm_temporary_1(buffer.mem, 7*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr nlm_temporary_2(buffer.mem, 8*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr nlm_temporary_3(buffer.mem, 9*buffer.pass_stride, buffer.pass_stride);
nlm_state.temporary_1_ptr = *nlm_temporary_1;
nlm_state.temporary_2_ptr = *nlm_temporary_2;
functions.non_local_means(filtered_b, filtered_a, residual_var, final_b);
/* Combine the two double-filtered halves to a final shadow feature. */
- device_sub_ptr shadow_pass(device, buffer.mem, 4*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
+ device_sub_ptr shadow_pass(buffer.mem, 4*buffer.pass_stride, buffer.pass_stride);
functions.combine_halves(final_a, final_b, *shadow_pass, null_ptr, 0, rect);
}
/* Prefilter general features. */
{
- device_sub_ptr unfiltered (device, buffer.mem, 8*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr variance (device, buffer.mem, 9*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr nlm_temporary_1(device, buffer.mem, 10*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr nlm_temporary_2(device, buffer.mem, 11*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr nlm_temporary_3(device, buffer.mem, 12*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
+ device_sub_ptr unfiltered (buffer.mem, 8*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr variance (buffer.mem, 9*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr nlm_temporary_1(buffer.mem, 10*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr nlm_temporary_2(buffer.mem, 11*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr nlm_temporary_3(buffer.mem, 12*buffer.pass_stride, buffer.pass_stride);
nlm_state.temporary_1_ptr = *nlm_temporary_1;
nlm_state.temporary_2_ptr = *nlm_temporary_2;
nlm_state.temporary_3_ptr = *nlm_temporary_3;
- int mean_from[] = { 0, 1, 2, 6, 7, 8, 12 };
- int variance_from[] = { 3, 4, 5, 9, 10, 11, 13 };
- int pass_to[] = { 1, 2, 3, 0, 5, 6, 7 };
+ int mean_from[] = { 0, 1, 2, 12, 6, 7, 8 };
+ int variance_from[] = { 3, 4, 5, 13, 9, 10, 11};
+ int pass_to[] = { 1, 2, 3, 0, 5, 6, 7};
for(int pass = 0; pass < 7; pass++) {
- device_sub_ptr feature_pass(device, buffer.mem, pass_to[pass]*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
+ device_sub_ptr feature_pass(buffer.mem, pass_to[pass]*buffer.pass_stride, buffer.pass_stride);
/* Get the unfiltered pass and its variance from the RenderBuffers. */
functions.get_feature(mean_from[pass], variance_from[pass], *unfiltered, *variance);
/* Smooth the pass and store the result in the denoising buffers. */
int mean_to[] = { 8, 9, 10};
int variance_to[] = {11, 12, 13};
int num_color_passes = 3;
+
+ storage.temporary_color.alloc_to_device(3*buffer.pass_stride, false);
+
for(int pass = 0; pass < num_color_passes; pass++) {
- device_sub_ptr color_pass (device, buffer.mem, mean_to[pass]*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr color_var_pass(device, buffer.mem, variance_to[pass]*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
+ device_sub_ptr color_pass(storage.temporary_color, pass*buffer.pass_stride, buffer.pass_stride);
+ device_sub_ptr color_var_pass(buffer.mem, variance_to[pass]*buffer.pass_stride, buffer.pass_stride);
functions.get_feature(mean_from[pass], variance_from[pass], *color_pass, *color_var_pass);
}
+
+ {
+ device_sub_ptr depth_pass (buffer.mem, 0, buffer.pass_stride);
+ device_sub_ptr color_var_pass(buffer.mem, variance_to[0]*buffer.pass_stride, 3*buffer.pass_stride);
+ device_sub_ptr output_pass (buffer.mem, mean_to[0]*buffer.pass_stride, 3*buffer.pass_stride);
+ functions.detect_outliers(storage.temporary_color.device_pointer, *color_var_pass, *depth_pass, *output_pass);
+ }
}
storage.w = filter_area.z;
storage.h = filter_area.w;
- storage.transform.resize(storage.w*storage.h*TRANSFORM_SIZE);
- storage.rank.resize(storage.w*storage.h);
- device->mem_alloc("Denoising Transform", storage.transform, MEM_READ_WRITE);
- device->mem_alloc("Denoising Rank", storage.rank, MEM_READ_WRITE);
+ storage.transform.alloc_to_device(storage.w*storage.h*TRANSFORM_SIZE, false);
+ storage.rank.alloc_to_device(storage.w*storage.h, false);
functions.construct_transform();
- device_only_memory<float> temporary_1;
- device_only_memory<float> temporary_2;
- temporary_1.resize(buffer.w*buffer.h);
- temporary_2.resize(buffer.w*buffer.h);
- device->mem_alloc("Denoising NLM temporary 1", temporary_1, MEM_READ_WRITE);
- device->mem_alloc("Denoising NLM temporary 2", temporary_2, MEM_READ_WRITE);
+ device_only_memory<float> temporary_1(device, "Denoising NLM temporary 1");
+ device_only_memory<float> temporary_2(device, "Denoising NLM temporary 2");
+ temporary_1.alloc_to_device(buffer.pass_stride, false);
+ temporary_2.alloc_to_device(buffer.pass_stride, false);
reconstruction_state.temporary_1_ptr = temporary_1.device_pointer;
reconstruction_state.temporary_2_ptr = temporary_2.device_pointer;
- storage.XtWX.resize(storage.w*storage.h*XTWX_SIZE);
- storage.XtWY.resize(storage.w*storage.h*XTWY_SIZE);
- device->mem_alloc("Denoising XtWX", storage.XtWX, MEM_READ_WRITE);
- device->mem_alloc("Denoising XtWY", storage.XtWY, MEM_READ_WRITE);
+ storage.XtWX.alloc_to_device(storage.w*storage.h*XTWX_SIZE, false);
+ storage.XtWY.alloc_to_device(storage.w*storage.h*XTWY_SIZE, false);
- reconstruction_state.filter_rect = make_int4(filter_area.x-rect.x, filter_area.y-rect.y, storage.w, storage.h);
+ reconstruction_state.filter_window = rect_from_shape(filter_area.x-rect.x, filter_area.y-rect.y, storage.w, storage.h);
int tile_coordinate_offset = filter_area.y*render_buffer.stride + filter_area.x;
reconstruction_state.buffer_params = make_int4(render_buffer.offset + tile_coordinate_offset,
render_buffer.stride,
reconstruction_state.source_h = rect.w-rect.y;
{
- device_sub_ptr color_ptr (device, buffer.mem, 8*buffer.pass_stride, 3*buffer.pass_stride, MEM_READ_WRITE);
- device_sub_ptr color_var_ptr(device, buffer.mem, 11*buffer.pass_stride, 3*buffer.pass_stride, MEM_READ_WRITE);
- functions.reconstruct(*color_ptr, *color_var_ptr, *color_ptr, *color_var_ptr, render_buffer.ptr);
+ device_sub_ptr color_ptr (buffer.mem, 8*buffer.pass_stride, 3*buffer.pass_stride);
+ device_sub_ptr color_var_ptr(buffer.mem, 11*buffer.pass_stride, 3*buffer.pass_stride);
+ functions.reconstruct(*color_ptr, *color_var_ptr, render_buffer.ptr);
}
- device->mem_free(storage.XtWX);
- device->mem_free(storage.XtWY);
- device->mem_free(storage.transform);
- device->mem_free(storage.rank);
- device->mem_free(temporary_1);
- device->mem_free(temporary_2);
- device->mem_free(buffer.mem);
- device->mem_free(tiles_mem);
return true;
}
projects / blender.git / blobdiff