[Bf-blender-cvs] [0cea9353fdf] master: Fix CUDA out of memory error with OptiX viewport denoising on small GPUs

[Patrick Mours]

Commit: 0cea9353fdfef4ae736d9894e4f6a0a8cef99619
Author: Patrick Mours
Date:   Fri Feb 28 15:57:44 2020 +0100
Branches: master
https://developer.blender.org/rB0cea9353fdfef4ae736d9894e4f6a0a8cef99619

Fix CUDA out of memory error with OptiX viewport denoising on small GPUs

This makes the memory allocation for the denoiser state use the memory allocator in Cycles, which
will evict textures to host memory when there is not enough space on the device. This means the
allocation for the denoiser state won't just fail if there is no more space and instead more space is
made for it to work. Also simplifies code somewhat.

===================================================================

M	intern/cycles/device/device_optix.cpp

===================================================================

diff --git a/intern/cycles/device/device_optix.cpp b/intern/cycles/device/device_optix.cpp
index 2ce8bed3783..0121b89e9d8 100644
--- a/intern/cycles/device/device_optix.cpp
+++ b/intern/cycles/device/device_optix.cpp
@@ -186,14 +186,15 @@ class OptiXDevice : public CUDADevice {
   OptixTraversableHandle tlas_handle = 0;
 
   OptixDenoiser denoiser = NULL;
-  pair<int2, CUdeviceptr> denoiser_state = {};
+  device_only_memory<unsigned char> denoiser_state;
   int denoiser_input_passes = 0;
 
  public:
   OptiXDevice(DeviceInfo &info_, Stats &stats_, Profiler &profiler_, bool background_)
       : CUDADevice(info_, stats_, profiler_, background_),
         sbt_data(this, "__sbt", MEM_READ_ONLY),
-        launch_params(this, "__params")
+        launch_params(this, "__params"),
+        denoiser_state(this, "__denoiser_state")
   {
     // Store number of CUDA streams in device info
     info.cpu_threads = DebugFlags().optix.cuda_streams;
@@ -255,13 +256,10 @@ class OptiXDevice : public CUDADevice {
       cuMemFree(mem);
     }
 
-    if (denoiser_state.second) {
-      cuMemFree(denoiser_state.second);
-    }
-
     sbt_data.free();
     texture_info.free();
     launch_params.free();
+    denoiser_state.free();
 
     // Unload modules
     if (optix_module != NULL)
@@ -813,32 +811,26 @@ class OptiXDevice : public CUDADevice {
       check_result_optix_ret(
           optixDenoiserComputeMemoryResources(denoiser, rect_size.x, rect_size.y, &sizes));
 
-      auto &state = denoiser_state.second;
-      auto &state_size = denoiser_state.first;
       const size_t scratch_size = sizes.recommendedScratchSizeInBytes;
       const size_t scratch_offset = sizes.stateSizeInBytes;
 
       // Allocate denoiser state if tile size has changed since last setup
-      if (state_size.x != rect_size.x || state_size.y != rect_size.y || recreate_denoiser) {
-        // Free existing state before allocating new one
-        if (state) {
-          cuMemFree(state);
-          state = 0;
-        }
-
-        check_result_cuda_ret(cuMemAlloc(&state, scratch_offset + scratch_size));
+      if (recreate_denoiser || (denoiser_state.data_width != rect_size.x ||
+                                denoiser_state.data_height != rect_size.y)) {
+        denoiser_state.alloc_to_device(scratch_offset + scratch_size);
 
         // Initialize denoiser state for the current tile size
         check_result_optix_ret(optixDenoiserSetup(denoiser,
                                                   0,
                                                   rect_size.x,
                                                   rect_size.y,
-                                                  state,
+                                                  denoiser_state.device_pointer,
                                                   scratch_offset,
-                                                  state + scratch_offset,
+                                                  denoiser_state.device_pointer + scratch_offset,
                                                   scratch_size));
 
-        state_size = rect_size;
+        denoiser_state.data_width = rect_size.x;
+        denoiser_state.data_height = rect_size.y;
       }
 
       // Set up input and output layer information
@@ -880,14 +872,14 @@ class OptiXDevice : public CUDADevice {
       check_result_optix_ret(optixDenoiserInvoke(denoiser,
                                                  0,
                                                  &params,
-                                                 state,
+                                                 denoiser_state.device_pointer,
                                                  scratch_offset,
                                                  input_layers,
                                                  task.denoising.optix_input_passes,
                                                  overlap_offset.x,
                                                  overlap_offset.y,
                                                  output_layers,
-                                                 state + scratch_offset,
+                                                 denoiser_state.device_pointer + scratch_offset,
                                                  scratch_size));
 
 #  if OPTIX_DENOISER_NO_PIXEL_STRIDE