Merge branch 'blender-v3.0-release' to bring in D13042:

Fix performance decrease with Scrambling Distance on

5 files changed, 19 insertions, 5 deletions

diff --git a/intern/cycles/integrator/path_trace_work_gpu.cpp b/intern/cycles/integrator/path_trace_work_gpu.cpp
index 956aa6a8c90..05e53f816a0 100644
--- a/intern/cycles/integrator/path_trace_work_gpu.cpp
+++ b/intern/cycles/integrator/path_trace_work_gpu.cpp
@@ -258,7 +258,8 @@ void PathTraceWorkGPU::render_samples(RenderStatistics &statistics,
    * become busy after adding new tiles). This is especially important for the shadow catcher which
    * schedules work in halves of available number of paths. */
   work_tile_scheduler_.set_max_num_path_states(max_num_paths_ / 8);
-
+  work_tile_scheduler_.set_accelerated_rt((device_->get_bvh_layout_mask() & BVH_LAYOUT_OPTIX) !=
+                                          0);
   work_tile_scheduler_.reset(effective_buffer_params_,
                              start_sample,
                              samples_num,
diff --git a/intern/cycles/integrator/tile.cpp b/intern/cycles/integrator/tile.cpp
index 4a1558cce09..e9a3cbd38aa 100644
--- a/intern/cycles/integrator/tile.cpp
+++ b/intern/cycles/integrator/tile.cpp
@@ -46,7 +46,8 @@ ccl_device_inline uint round_up_to_power_of_two(uint x)
   return next_power_of_two(x);
 }
 
-TileSize tile_calculate_best_size(const int2 &image_size,
+TileSize tile_calculate_best_size(const bool accel_rt,
+                                  const int2 &image_size,
                                   const int num_samples,
                                   const int max_num_path_states,
                                   const float scrambling_distance)
@@ -73,7 +74,7 @@ TileSize tile_calculate_best_size(const int2 &image_size,
 
   TileSize tile_size;
   const int num_path_states_per_sample = max_num_path_states / num_samples;
-  if (scrambling_distance < 0.9f) {
+  if (scrambling_distance < 0.9f && accel_rt) {
     /* Prefer large tiles for scrambling distance, bounded by max num path states. */
     tile_size.width = min(image_size.x, max_num_path_states);
     tile_size.height = min(image_size.y, max(max_num_path_states / tile_size.width, 1));
diff --git a/intern/cycles/integrator/tile.h b/intern/cycles/integrator/tile.h
index 61f7d736115..05b1e0af6b1 100644
--- a/intern/cycles/integrator/tile.h
+++ b/intern/cycles/integrator/tile.h
@@ -49,7 +49,8 @@ std::ostream &operator<<(std::ostream &os, const TileSize &tile_size);
  * of active path states.
  * Will attempt to provide best guess to keep path tracing threads of a device as localized as
  * possible, and have as many threads active for every tile as possible. */
-TileSize tile_calculate_best_size(const int2 &image_size,
+TileSize tile_calculate_best_size(const bool accel_rt,
+                                  const int2 &image_size,
                                   const int num_samples,
                                   const int max_num_path_states,
                                   const float scrambling_distance);
diff --git a/intern/cycles/integrator/work_tile_scheduler.cpp b/intern/cycles/integrator/work_tile_scheduler.cpp
index d60f7149bf4..353c357475d 100644
--- a/intern/cycles/integrator/work_tile_scheduler.cpp
+++ b/intern/cycles/integrator/work_tile_scheduler.cpp
@@ -28,6 +28,11 @@ WorkTileScheduler::WorkTileScheduler()
 {
 }
 
+void WorkTileScheduler::set_accelerated_rt(bool accelerated_rt)
+{
+  accelerated_rt_ = accelerated_rt;
+}
+
 void WorkTileScheduler::set_max_num_path_states(int max_num_path_states)
 {
   max_num_path_states_ = max_num_path_states;
@@ -61,7 +66,7 @@ void WorkTileScheduler::reset(const BufferParams &buffer_params,
 void WorkTileScheduler::reset_scheduler_state()
 {
   tile_size_ = tile_calculate_best_size(
-      image_size_px_, samples_num_, max_num_path_states_, scrambling_distance_);
+      accelerated_rt_, image_size_px_, samples_num_, max_num_path_states_, scrambling_distance_);
 
   VLOG(3) << "Will schedule tiles of size " << tile_size_;
 
diff --git a/intern/cycles/integrator/work_tile_scheduler.h b/intern/cycles/integrator/work_tile_scheduler.h
index 2d6395799f7..b9cef7be7c4 100644
--- a/intern/cycles/integrator/work_tile_scheduler.h
+++ b/intern/cycles/integrator/work_tile_scheduler.h
@@ -31,6 +31,9 @@ class WorkTileScheduler {
  public:
   WorkTileScheduler();
 
+  /* To indicate if there is accelerated RT support. */
+  void set_accelerated_rt(bool state);
+
   /* MAximum path states which are allowed to be used by a single scheduled work tile.
    *
    * Affects the scheduled work size: the work size will be as big as possible, but will not exceed
@@ -55,6 +58,9 @@ class WorkTileScheduler {
  protected:
   void reset_scheduler_state();
 
+  /* Used to indicate if there is accelerated ray tracing. */
+  bool accelerated_rt_ = false;
+
   /* Maximum allowed path states to be used.
    *
    * TODO(sergey): Naming can be improved. The fact that this is a limiting factor based on the