blender
/
blender
136
413
Fork 596
Code Issues 6.3k Pull Requests 770 Projects 19 Wiki Activity

Alternative Upload geometry data in parallel to multiple GPUs using the "Multi-Device" #107552

Open
William Leeson wants to merge 137 commits from leesonw/blender-cluster:upload_changed into main
pull from: leesonw/blender-cluster:upload_changed
merge into: blender:main

When changing the target branch, be careful to rebase the branch in your fork to match. See documentation.
Conversation 2 Commits 137 Files Changed 43 +6 -1
1 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files
Showing only changes of commit a87d9843d7 - Show all commits

7
intern/cycles/device/optix/device_impl.cpp
View File

@ -1007,6 +1007,9 @@ bool OptiXDevice::build_optix_bvh(BVHOptiX *bvh,
* from running out of memory (since both original and compacted acceleration structure memory
* may be allocated at the same time for the duration of this function). The builds would
* otherwise happen on the same CUDA stream anyway. */
static thread_mutex mutex;
thread_scoped_lock lock(mutex);
const CUDAContextScope scope(this);
const bool use_fast_trace_bvh = (bvh->params.bvh_type == BVH_TYPE_STATIC);
/* Compute memory usage. */
@ -1076,10 +1079,12 @@ bool OptiXDevice::build_optix_bvh(BVHOptiX *bvh,
use_fast_trace_bvh ? 1 : 0));
bvh->traversable_handle = static_cast<uint64_t>(out_handle);
/* Wait for all operations to finish. */
cuda_assert(cuStreamSynchronize(NULL));
/* Compact acceleration structure to save memory (do not do this in viewport for faster builds).
*/
if (use_fast_trace_bvh) {
const CUDAContextScope scope(this);
uint64_t compacted_size = sizes.outputSizeInBytes;
cuda_assert(cuMemcpyDtoH(&compacted_size, compacted_size_prop.result, sizeof(compacted_size)));