Cycles: Improve denoiser update performance when rendering with multiple GPUs

This patch causes the render buffers to be copied to the denoiser
device only once before denoising and output/display is then fed
from that single buffer on the denoiser device. That way usually all
but one copy (from all the render devices to the denoiser device)
can be eliminated, provided that the denoiser device is also the
display device (in which case interop is used to update the display).
As such this patch also adds some logic that tries to ensure the
chosen denoiser device is the same as the display device.

Differential Revision: https://developer.blender.org/D15657

1 files changed, 9 insertions, 6 deletions

diff --git a/intern/cycles/session/session.cpp b/intern/cycles/session/session.cpp
index e7de82a6e1b..c94b53535a7 100644
--- a/intern/cycles/session/session.cpp
+++ b/intern/cycles/session/session.cpp
@@ -370,6 +370,14 @@ RenderWork Session::run_update_for_next_iteration()
     if (update_scene(width, height)) {
       profiler.reset(scene->shaders.size(), scene->objects.size());
     }
+
+    /* Unlock scene mutex before loading denoiser kernels, since that may attempt to activate
+     * graphics interop, which can deadlock when the scene mutex is still being held. */
+    scene_lock.unlock();
+
+    path_trace_->load_kernels();
+    path_trace_->alloc_work_memory();
+
     progress.add_skip_time(update_timer, params.background);
   }
 
@@ -618,12 +626,7 @@ bool Session::update_scene(int width, int height)
   Camera *cam = scene->camera;
   cam->set_screen_size(width, height);
 
-  const bool scene_update_result = scene->update(progress);
-
-  path_trace_->load_kernels();
-  path_trace_->alloc_work_memory();
-
-  return scene_update_result;
+  return scene->update(progress);
 }
 
 static string status_append(const string &status, const string &suffix)