saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-05-14 20:48:00 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Use compute queues by default in DML EP (#20438)

Browse source 
### Description
We originally only use compute queues for compute-only devices; this
change sets the default for DX12 devices to use compute queues as well.



### Motivation and Context
There have been issues with TDRs occurring when using the current
default queues, which doesn't happen on compute queues.
This commit is contained in:
maggie1059 2024-04-24 10:44:16 -07:00 committed by GitHub
parent f78215adad
commit dfd4bce36e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 5 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
onnxruntime/core/providers/dml/dml_provider_factory.cc
View file

@ -542,7 +542,11 @@ static D3D12_COMMAND_LIST_TYPE CalculateCommandListType(ID3D12Device* d3d12_devi
sizeof(feature_levels)
));
auto use_compute_command_list = (feature_levels.MaxSupportedFeatureLevel <= D3D_FEATURE_LEVEL_1_0_CORE);
// Use compute queue whenever possible on supported hardware to avoid TDR and maintain UI QoS
// Core and generic devices only have compute queues, DX11 has "immediate" submission, DX12 has both
auto use_compute_command_list = (feature_levels.MaxSupportedFeatureLevel <= D3D_FEATURE_LEVEL_1_0_CORE) ||
(feature_levels.MaxSupportedFeatureLevel >= D3D_FEATURE_LEVEL_12_0);
if (use_compute_command_list)
{
return D3D12_COMMAND_LIST_TYPE_COMPUTE;