7#include "device/cpu/kernel.h"
10#ifdef WITH_CYCLES_DEBUG
40 const int thread_index = tbb::this_task_arena::current_thread_index();
42 DCHECK_LE(thread_index, kernel_thread_globals.size());
44 return &kernel_thread_globals[thread_index];
50 const bool *cancel_requested_flag)
51 :
PathTraceWork(device, film, device_scene, cancel_requested_flag),
70 const int start_sample,
71 const int samples_num,
72 const int sample_offset)
76 const int64_t total_pixels_num = image_width * image_height;
80 kernel_globals.start_profiling();
85 local_arena.execute([&]() {
86 parallel_for(
int64_t(0), total_pixels_num, [&](
int64_t work_index) {
91 const int y = work_index / image_width;
92 const int x = work_index -
y * image_width;
112 kernel_globals.stop_profiling();
121 const int samples_num)
131 shadow_catcher_state = &integrator_states[1];
160#if defined(WITH_PATH_GUIDING)
161 if (kernel_globals->
data.integrator.train_guiding) {
162 assert(kernel_globals->opgl_path_segment_storage);
163 assert(kernel_globals->opgl_path_segment_storage->GetNumSegments() == 0);
171 if (shadow_catcher_state) {
172 kernel_globals->
data.integrator.train_guiding =
false;
174 kernel_globals->
data.integrator.train_guiding =
true;
181 if (shadow_catcher_state) {
188 if (render_timer.
lap(time)) {
190 kernel_globals->
data.film.pass_stride;
191 *(buffer + kernel_globals->
data.film.pass_render_time) +=
float(time);
200 const int num_samples)
220 const PassAccessorCPU pass_accessor(pass_access_info, kfilm.exposure, num_samples);
226 local_arena.execute([&]() {
237 return buffers_->copy_from_device();
264 uint num_active_pixels = 0;
269 local_arena.execute([&]() {
270 parallel_for(full_y, full_y + height, [&](
int y) {
273 bool row_converged =
true;
274 uint num_row_pixels_active = 0;
275 for (
int x = 0;
x < width; ++
x) {
279 ++num_row_pixels_active;
280 row_converged =
false;
286 if (!row_converged) {
293 if (num_active_pixels) {
294 local_arena.execute([&]() {
295 parallel_for(full_x, full_x + width, [&](
int x) {
298 kernel_globals,
render_buffer,
x, full_y, height, offset, stride);
303 return num_active_pixels;
316 local_arena.execute([&]() {
317 parallel_for(0, height, [&](
int y) {
319 int pixel_index =
y * width;
321 for (
int x = 0;
x < width; ++
x, ++pixel_index) {
341 const blocked_range2d<int> range(min_x, max_x, min_y, max_y);
344 local_arena.execute([&]() {
345 parallel_for(range, [&](
const blocked_range2d<int> r) {
347 for (
int y = r.cols().begin();
y < r.cols().end(); ++
y) {
348 for (
int x = r.rows().begin();
x < r.rows().end(); ++
x) {
358 local_arena.execute([&]() {
359 parallel_for(min_x, max_x, [&](
int x) {
367#if defined(WITH_PATH_GUIDING)
370void PathTraceWorkCPU::guiding_init_kernel_globals(
void *guiding_field,
371 void *sample_data_storage,
378 openpgl::cpp::Field *field = (openpgl::cpp::Field *)guiding_field;
381 kg.opgl_guiding_field = field;
383# if PATH_GUIDING_LEVEL >= 4
384 if (kg.opgl_surface_sampling_distribution) {
385 kg.opgl_surface_sampling_distribution.reset();
387 if (kg.opgl_volume_sampling_distribution) {
388 kg.opgl_volume_sampling_distribution.reset();
392 kg.opgl_surface_sampling_distribution =
393 make_unique<openpgl::cpp::SurfaceSamplingDistribution>(field);
394 kg.opgl_volume_sampling_distribution = make_unique<openpgl::cpp::VolumeSamplingDistribution>(
400 kg.
data.integrator.train_guiding = train;
401 kg.opgl_sample_data_storage = (openpgl::cpp::SampleStorage *)sample_data_storage;
404 kg.opgl_path_segment_storage_size = kg.
data.integrator.transparent_max_bounce +
405 kg.
data.integrator.max_bounce + 3;
406 kg.opgl_path_segment_storage->Reserve(kg.opgl_path_segment_storage_size);
407 kg.opgl_path_segment_storage->Clear();
417# ifdef WITH_CYCLES_DEBUG
420 const bool validSegments = kg->opgl_path_segment_storage->ValidateSegments();
421 if (!validSegments) {
422 LOG_DEBUG <<
"Guiding: invalid path segments!";
427 pgl_vec3f pgl_final_color = kg->opgl_path_segment_storage->CalculatePixelEstimate(
false);
429 float3 final_color =
make_float3(pgl_final_color.x, pgl_final_color.y, pgl_final_color.z);
439# if PATH_GUIDING_LEVEL >= 2
440 const bool use_direct_light =
kernel_data.integrator.use_guiding_direct_light;
441 const bool use_mis_weights =
kernel_data.integrator.use_guiding_mis_weights;
442 kg->opgl_path_segment_storage->PrepareSamples(use_mis_weights, use_direct_light,
false);
445# ifdef WITH_CYCLES_DEBUG
448 const bool validSamples = kg->opgl_path_segment_storage->ValidateSamples();
451 <<
"Guiding: path segment storage generated/contains invalid radiance/training samples!";
456# if PATH_GUIDING_LEVEL >= 3
458 size_t num_samples = 0;
459 const openpgl::cpp::SampleData *samples = kg->opgl_path_segment_storage->GetSamples(num_samples);
460 kg->opgl_sample_data_storage->AddSamples(samples, num_samples);
464 kg->opgl_path_segment_storage->Clear();
ATOMIC_INLINE uint32_t atomic_fetch_and_add_uint32(uint32_t *p, uint32_t x)
unsigned long long int uint64_t
FilterYFunction volume_guiding_filter_y
AdaptiveSamplingConvergenceCheckFunction adaptive_sampling_convergence_check
CryptomattePostprocessFunction cryptomatte_postprocess
CPUKernelFunction< void(*)(const ThreadKernelGlobalsCPU *kg, ccl_global float *render_buffer, const int y, const int center_x, const int min_x, const int max_x, const int offset, int stride)> volume_guiding_filter_x
FilterYFunction adaptive_sampling_filter_y
FilterXFunction adaptive_sampling_filter_x
IntegratorShadeFunction integrator_megakernel
IntegratorInitFunction integrator_init_from_bake
IntegratorInitFunction integrator_init_from_camera
virtual vector< ThreadKernelGlobalsCPU > * acquire_cpu_kernel_thread_globals()
virtual void release_cpu_kernel_thread_globals()
bool get_render_tile_pixels(const RenderBuffers *render_buffers, const Destination &destination) const
half4 * map_texture_buffer()
void unmap_texture_buffer()
bool copy_render_buffers_to_device() override
vector< ThreadKernelGlobalsCPU > * kernel_thread_globals_
void deinit_execution() override
void render_samples(RenderStatistics &statistics, const int start_sample, const int samples_num, const int sample_offset) override
void init_execution() override
bool copy_render_buffers_from_device() override
int adaptive_sampling_converge_filter_count_active(const float threshold, bool reset) override
void copy_to_display(PathTraceDisplay *display, PassMode pass_mode, const int num_samples) override
void destroy_gpu_resources(PathTraceDisplay *display) override
void cryptomatte_postproces() override
const CPUKernels & kernels_
void render_samples_full_pipeline(ThreadKernelGlobalsCPU *kernel_globals, const KernelWorkTile &work_tile, const int samples_num)
PathTraceWorkCPU(Device *device, Film *film, DeviceScene *device_scene, const bool *cancel_requested_flag)
bool zero_render_buffers() override
void denoise_volume_guiding_buffers() override
BufferParams effective_denoised_buffer_params_
PassAccessor::Destination get_display_destination_template(const PathTraceDisplay *display, const PassMode mode) const
unique_ptr< RenderBuffers > buffers_
PassAccessor::PassAccessInfo get_display_pass_access_info(PassMode pass_mode) const
BufferParams effective_buffer_params_
DeviceScene * device_scene_
PathTraceWork(Device *device, Film *film, DeviceScene *device_scene, const bool *cancel_requested_flag)
bool is_cancel_requested() const
bool lap(uint64_t &delta)
#define CCL_NAMESPACE_END
#define assert(assertion)
VecBase< float, 3 > float3
ccl_gpu_kernel_postfix ccl_global KernelWorkTile const int ccl_global float * render_buffer
CCL_NAMESPACE_BEGIN ccl_device_inline void path_state_init_queues(IntegratorState state)
static ThreadKernelGlobalsCPU * kernel_thread_globals_get(vector< ThreadKernelGlobalsCPU > &kernel_thread_globals)
static CCL_NAMESPACE_BEGIN tbb::task_arena local_tbb_arena_create(const Device *device)
CCL_NAMESPACE_END KernelData data
ccl_device_inline void film_write_pass_float3(ccl_global float *ccl_restrict buffer, const float3 value)
CCL_NAMESPACE_BEGIN ccl_device_forceinline ccl_global float * film_pass_pixel_render_buffer(KernelGlobals kg, ConstIntegratorState state, ccl_global float *ccl_restrict render_buffer)