[Bf-blender-cvs] [6e268a749fe] master: Fix adaptive sampling artifacts on tile boundaries
Sergey Sharybin
noreply at git.blender.org
Tue Oct 5 16:19:18 CEST 2021
Commit: 6e268a749fee16b442bcb3fba6cb6e08850d8389
Author: Sergey Sharybin
Date: Tue Sep 21 17:03:22 2021 +0200
Branches: master
https://developer.blender.org/rB6e268a749fee16b442bcb3fba6cb6e08850d8389
Fix adaptive sampling artifacts on tile boundaries
Implement an overscan support for tiles, so that adaptive sampling can
rely on the pixels neighbourhood.
Differential Revision: https://developer.blender.org/D12599
===================================================================
M intern/cycles/blender/blender_camera.cpp
M intern/cycles/device/optix/device_impl.cpp
M intern/cycles/integrator/denoiser_oidn.cpp
M intern/cycles/integrator/pass_accessor_cpu.cpp
M intern/cycles/integrator/pass_accessor_gpu.cpp
M intern/cycles/integrator/path_trace.cpp
M intern/cycles/integrator/path_trace_work.cpp
M intern/cycles/integrator/path_trace_work_gpu.cpp
M intern/cycles/kernel/device/gpu/kernel.h
M intern/cycles/render/buffers.cpp
M intern/cycles/render/buffers.h
M intern/cycles/render/session.cpp
M intern/cycles/render/tile.cpp
M intern/cycles/render/tile.h
===================================================================
diff --git a/intern/cycles/blender/blender_camera.cpp b/intern/cycles/blender/blender_camera.cpp
index 4e8df5a99a6..93f19b73f53 100644
--- a/intern/cycles/blender/blender_camera.cpp
+++ b/intern/cycles/blender/blender_camera.cpp
@@ -927,6 +927,9 @@ BufferParams BlenderSync::get_buffer_params(
params.height = height;
}
+ params.window_width = params.width;
+ params.window_height = params.height;
+
return params;
}
diff --git a/intern/cycles/device/optix/device_impl.cpp b/intern/cycles/device/optix/device_impl.cpp
index 49d4e22143f..f9a15553aa9 100644
--- a/intern/cycles/device/optix/device_impl.cpp
+++ b/intern/cycles/device/optix/device_impl.cpp
@@ -768,7 +768,13 @@ void OptiXDevice::denoise_color_read(DenoiseContext &context, const DenoisePass
destination.num_components = 3;
destination.pixel_stride = context.buffer_params.pass_stride;
- pass_accessor.get_render_tile_pixels(context.render_buffers, context.buffer_params, destination);
+ BufferParams buffer_params = context.buffer_params;
+ buffer_params.window_x = 0;
+ buffer_params.window_y = 0;
+ buffer_params.window_width = buffer_params.width;
+ buffer_params.window_height = buffer_params.height;
+
+ pass_accessor.get_render_tile_pixels(context.render_buffers, buffer_params, destination);
}
bool OptiXDevice::denoise_filter_color_preprocess(DenoiseContext &context, const DenoisePass &pass)
diff --git a/intern/cycles/integrator/denoiser_oidn.cpp b/intern/cycles/integrator/denoiser_oidn.cpp
index 7fc2b2b1892..ee3b62668a7 100644
--- a/intern/cycles/integrator/denoiser_oidn.cpp
+++ b/intern/cycles/integrator/denoiser_oidn.cpp
@@ -289,7 +289,13 @@ class OIDNDenoiseContext {
* pixels. */
const PassAccessorCPU pass_accessor(pass_access_info, 1.0f, num_samples_);
- pass_accessor.get_render_tile_pixels(render_buffers_, buffer_params_, destination);
+ BufferParams buffer_params = buffer_params_;
+ buffer_params.window_x = 0;
+ buffer_params.window_y = 0;
+ buffer_params.window_width = buffer_params.width;
+ buffer_params.window_height = buffer_params.height;
+
+ pass_accessor.get_render_tile_pixels(render_buffers_, buffer_params, destination);
}
/* Read pass pixels using PassAccessor into a temporary buffer which is owned by the pass.. */
diff --git a/intern/cycles/integrator/pass_accessor_cpu.cpp b/intern/cycles/integrator/pass_accessor_cpu.cpp
index 3c6691f6d43..80908271ff6 100644
--- a/intern/cycles/integrator/pass_accessor_cpu.cpp
+++ b/intern/cycles/integrator/pass_accessor_cpu.cpp
@@ -99,17 +99,22 @@ inline void PassAccessorCPU::run_get_pass_kernel_processor_float(
{
DCHECK_EQ(destination.stride, 0) << "Custom stride for float destination is not implemented.";
- const float *buffer_data = render_buffers->buffer.data();
+ const int64_t pass_stride = buffer_params.pass_stride;
+ const int64_t buffer_row_stride = buffer_params.stride * buffer_params.pass_stride;
+
+ const float *window_data = render_buffers->buffer.data() + buffer_params.window_x * pass_stride +
+ buffer_params.window_y * buffer_row_stride;
+
const int pixel_stride = destination.pixel_stride ? destination.pixel_stride :
destination.num_components;
- tbb::parallel_for(0, buffer_params.height, [&](int64_t y) {
- int64_t pixel_index = y * buffer_params.width;
- for (int64_t x = 0; x < buffer_params.width; ++x, ++pixel_index) {
- const int64_t input_pixel_offset = pixel_index * buffer_params.pass_stride;
- const float *buffer = buffer_data + input_pixel_offset;
- float *pixel = destination.pixels + (pixel_index + destination.offset) * pixel_stride;
+ tbb::parallel_for(0, buffer_params.window_height, [&](int64_t y) {
+ const float *buffer = window_data + y * buffer_row_stride;
+ float *pixel = destination.pixels +
+ (y * buffer_params.width + destination.offset) * pixel_stride;
+ for (int64_t x = 0; x < buffer_params.window_width;
+ ++x, buffer += pass_stride, pixel += pixel_stride) {
processor(kfilm_convert, buffer, pixel);
}
});
@@ -123,26 +128,28 @@ inline void PassAccessorCPU::run_get_pass_kernel_processor_half_rgba(
const Destination &destination,
const Processor &processor) const
{
- const float *buffer_data = render_buffers->buffer.data();
+ const int64_t pass_stride = buffer_params.pass_stride;
+ const int64_t buffer_row_stride = buffer_params.stride * buffer_params.pass_stride;
+
+ const float *window_data = render_buffers->buffer.data() + buffer_params.window_x * pass_stride +
+ buffer_params.window_y * buffer_row_stride;
half4 *dst_start = destination.pixels_half_rgba + destination.offset;
const int destination_stride = destination.stride != 0 ? destination.stride :
buffer_params.width;
- tbb::parallel_for(0, buffer_params.height, [&](int64_t y) {
- int64_t pixel_index = y * buffer_params.width;
- half4 *dst_row_start = dst_start + y * destination_stride;
- for (int64_t x = 0; x < buffer_params.width; ++x, ++pixel_index) {
- const int64_t input_pixel_offset = pixel_index * buffer_params.pass_stride;
- const float *buffer = buffer_data + input_pixel_offset;
+ tbb::parallel_for(0, buffer_params.window_height, [&](int64_t y) {
+ const float *buffer = window_data + y * buffer_row_stride;
+ half4 *pixel = dst_start + y * destination_stride;
+ for (int64_t x = 0; x < buffer_params.window_width; ++x, buffer += pass_stride, ++pixel) {
- float pixel[4];
- processor(kfilm_convert, buffer, pixel);
+ float pixel_rgba[4];
+ processor(kfilm_convert, buffer, pixel_rgba);
- film_apply_pass_pixel_overlays_rgba(kfilm_convert, buffer, pixel);
+ film_apply_pass_pixel_overlays_rgba(kfilm_convert, buffer, pixel_rgba);
- half4 *pixel_half_rgba = dst_row_start + x;
- float4_store_half(&pixel_half_rgba->x, make_float4(pixel[0], pixel[1], pixel[2], pixel[3]));
+ float4_store_half(&pixel->x,
+ make_float4(pixel_rgba[0], pixel_rgba[1], pixel_rgba[2], pixel_rgba[3]));
}
});
}
diff --git a/intern/cycles/integrator/pass_accessor_gpu.cpp b/intern/cycles/integrator/pass_accessor_gpu.cpp
index eb80ba99655..7b01d061708 100644
--- a/intern/cycles/integrator/pass_accessor_gpu.cpp
+++ b/intern/cycles/integrator/pass_accessor_gpu.cpp
@@ -43,10 +43,13 @@ void PassAccessorGPU::run_film_convert_kernels(DeviceKernel kernel,
KernelFilmConvert kfilm_convert;
init_kernel_film_convert(&kfilm_convert, buffer_params, destination);
- const int work_size = buffer_params.width * buffer_params.height;
+ const int work_size = buffer_params.window_width * buffer_params.window_height;
const int destination_stride = destination.stride != 0 ? destination.stride :
- buffer_params.width;
+ buffer_params.window_width;
+
+ const int offset = buffer_params.window_x * buffer_params.pass_stride +
+ buffer_params.window_y * buffer_params.stride * buffer_params.pass_stride;
if (destination.d_pixels) {
DCHECK_EQ(destination.stride, 0) << "Custom stride for float destination is not implemented.";
@@ -55,8 +58,8 @@ void PassAccessorGPU::run_film_convert_kernels(DeviceKernel kernel,
const_cast<device_ptr *>(&destination.d_pixels),
const_cast<device_ptr *>(&render_buffers->buffer.device_pointer),
const_cast<int *>(&work_size),
- const_cast<int *>(&buffer_params.width),
- const_cast<int *>(&buffer_params.offset),
+ const_cast<int *>(&buffer_params.window_width),
+ const_cast<int *>(&offset),
const_cast<int *>(&buffer_params.stride),
const_cast<int *>(&destination.offset),
const_cast<int *>(&destination_stride)};
@@ -70,8 +73,8 @@ void PassAccessorGPU::run_film_convert_kernels(DeviceKernel kernel,
const_cast<device_ptr *>(&destination.d_pixels_half_rgba),
const_cast<device_ptr *>(&render_buffers->buffer.device_pointer),
const_cast<int *>(&work_size),
- const_cast<int *>(&buffer_params.width),
- const_cast<int *>(&buffer_params.offset),
+ const_cast<int *>(&buffer_params.window_width),
+ const_cast<int *>(&offset),
const_cast<int *>(&buffer_params.stride),
const_cast<int *>(&destination.offset),
const_cast<int *>(&destination_stride)};
diff --git a/intern/cycles/integrator/path_trace.cpp b/intern/cycles/integrator/path_trace.cpp
index 7624b244175..3ea5c3c64b8 100644
--- a/intern/cycles/integrator/path_trace.cpp
+++ b/intern/cycles/integrator/path_trace.cpp
@@ -282,6 +282,12 @@ static BufferParams scale_buffer_params(const BufferParams ¶ms, int resoluti
scaled_params.width = max(1, params.width / resolution_divider);
scaled_params.height = max(1, params.height / resolution_divider);
+
+ scaled_params.window_x = params.window_x / resolution_divider;
+ scaled_params.window_y = params.window_y / resolution_divider;
+ scaled_params.window_width = params.window_width / resolution_divider;
+ scaled_params.window_height = params.window_height / resolution_divider;
+
scaled_params.full_x = params.full_x / resolution_divider;
scaled_params.full_y = params.full_y / resolution_divider;
scaled_params.full_width = params.full_width / resolution_divider;
@@ -1005,12 +1011,12 @@ bool PathTrace::set_render_tile_pixels(PassAccessor &pass_accessor,
int2 PathTrace::get_render_tile_size() const
{
if (full_frame_state_.render_buffers) {
- return make_int2(full_frame_state_.render_buffers->params.width,
- full_frame_state_.render_buffers->params.height);
+ return make_int2(full_frame_state_.render_buffers->params.window_width,
+
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list