build_files/build_environment/cmake/openimagedenoise.cmake

@@ -14,6 +14,7 @@ if(NOT APPLE)
     ${OIDN_EXTRA_ARGS}
     -DOIDN_DEVICE_SYCL=ON
     -DOIDN_DEVICE_SYCL_AOT=OFF
+    -DOIDN_DEVICE_CUDA=ON
     -DOIDN_DEVICE_HIP=ON
     -DLEVEL_ZERO_ROOT=${LIBDIR}/level-zero
   )

intern/cycles/device/cuda/device.cpp

@@ -10,6 +10,8 @@
 #  include "device/cuda/device_impl.h"
 #  include "device/device.h"
 
+#  include "integrator/denoiser_oidn_gpu.h"
+
 #  include "util/string.h"
 #  include "util/windows.h"
 #endif /* WITH_CUDA */
@@ -162,6 +164,12 @@ void device_cuda_info(vector<DeviceInfo> &devices)
                             (unsigned int)pci_location[1],
                             (unsigned int)pci_location[2]);
 
+#  if defined(WITH_OPENIMAGEDENOISE)
+    if (OIDNDenoiserGPU::is_device_supported(info)) {
+      info.denoisers |= DENOISER_OPENIMAGEDENOISE;
+    }
+#  endif
+
     /* If device has a kernel timeout and no compute preemption, we assume
      * it is connected to a display and will freeze the display while doing
      * computations. */

intern/cycles/device/optix/device.cpp

@@ -8,6 +8,8 @@
 #include "device/cuda/device.h"
 #include "device/optix/device_impl.h"
 
+#include "integrator/denoiser_oidn_gpu.h"
+
 #include "util/log.h"
 
 #ifdef WITH_OSL
@@ -74,6 +76,11 @@ void device_optix_info(const vector<DeviceInfo> &cuda_devices, vector<DeviceInfo
     info.has_osl = true;
 #  endif
     info.denoisers |= DENOISER_OPTIX;
+#  if defined(WITH_OPENIMAGEDENOISE)
+    if (OIDNDenoiserGPU::is_device_supported(info)) {
+      info.denoisers |= DENOISER_OPENIMAGEDENOISE;
+    }
+#  endif
 
     devices.push_back(info);
   }

intern/cycles/integrator/denoiser_oidn_gpu.cpp

@@ -111,6 +111,10 @@ uint OIDNDenoiserGPU::get_device_type_mask() const
 #  ifdef OIDN_DEVICE_SYCL
   device_mask |= DEVICE_MASK_ONEAPI;
 #  endif
+#  ifdef OIDN_DEVICE_CUDA
+  device_mask |= DEVICE_MASK_CUDA;
+  device_mask |= DEVICE_MASK_OPTIX;
+#  endif
 #  ifdef OIDN_DEVICE_HIP
   device_mask |= DEVICE_MASK_HIP;
 #  endif
@@ -155,9 +159,17 @@ bool OIDNDenoiserGPU::denoise_create_if_needed(DenoiseContext &context)
       oidn_device_ = oidnNewSYCLDevice(
           (const sycl::queue *)reinterpret_cast<OneapiDevice *>(denoiser_device_)->sycl_queue(),
           1);
-      denoiser_queue_->init_execution();
       break;
 #  endif
+#  if defined(OIDN_DEVICE_CUDA) && defined(WITH_CUDA)
+    case DEVICE_CUDA:
+    case DEVICE_OPTIX: {
+      /* Directly using the stream from the DeviceQueue returns "invalid resource handle". */
+      cudaStream_t stream = nullptr;
+      oidn_device_ = oidnNewCUDADevice(&denoiser_device_->info.num, &stream, 1);
+      break;
+    }
+#  endif
 #  if defined(OIDN_DEVICE_HIP) && defined(WITH_HIP)
     case DEVICE_HIP: {
       hipStream_t stream = nullptr;