onnxruntime
[Plugin EP] Port graph capture/replay APIs
#27958
Merged

Commits
  • First incorrect draft impl
    adrianlizarraga committed 135 days ago
  • Add IExecutionProvider::GetGraphCaptureNodeAssignmentPolicy() to remove the hardcoded list of EPs that support graph capture from inference_session.cc
    adrianlizarraga committed 134 days ago
  • Minor update to docs. Stills need improvement
    adrianlizarraga committed 134 days ago
  • Add onnxruntime:: namespace qualifier for DML use of enum
    adrianlizarraga committed 134 days ago
  • Merge branch 'main' into adrianl/PluginEp_CudaGraphCaptureReplay
    adrianlizarraga committed 134 days ago
  • Implement new graph capture/replay C APIs in webgpu's OrtEp implementation
    adrianlizarraga committed 134 days ago
  • Add stubs with todos to cuda's plugin EP
    adrianlizarraga committed 134 days ago
  • Use only one enum for the graph capture node assignment policy
    adrianlizarraga committed 134 days ago
  • Update C doc comments
    adrianlizarraga committed 134 days ago
  • First attempt at a webgpu plugin EP test that uses graph capture
    adrianlizarraga committed 133 days ago
  • Add ifdef guards for webgpu plugin EP test
    adrianlizarraga committed 131 days ago
  • Review comments: Update NvTensorRTRTX EP to return false from IExecutionProvider::IsGraphCaptureEnabled() as ORT previously never managed graph capture for this EP; Update use of webgpu name with constant
    adrianlizarraga committed 131 days ago
  • Add InferenceSession::RunImpl() that takes in the recursive run depth
    adrianlizarraga committed 131 days ago
  • Ensure at least one compute node is assigned to the graph capturing EP
    adrianlizarraga committed 131 days ago
  • Add ORT_ENFORCE
    adrianlizarraga committed 131 days ago
  • Merge branch 'main' into adrianl/PluginEp_CudaGraphCaptureReplay
    adrianlizarraga committed 131 days ago
  • Fix graph capture/replay test for the webgpu plugin EP (runs locally)
    adrianlizarraga committed 130 days ago
  • CXX API: add Ort::Env::CopyTensor() to copy only one tensor
    adrianlizarraga committed 130 days ago
  • Update test EP registration to use correct ep lib name for various platforms
    adrianlizarraga committed 130 days ago
  • Update docs regarding the max number of attempts to call IsGraphCaptured
    adrianlizarraga committed 130 days ago
  • Merge branch 'main' into adrianl/PluginEp_CudaGraphCaptureReplay
    adrianlizarraga committed 128 days ago
  • Adjust test ifdef for webgpu EP test that is not really the plugin EP version
    adrianlizarraga committed 128 days ago
  • Address doc comments
    adrianlizarraga committed 128 days ago
  • Check that OrtEp::IsGraphCaptured and OrtEp::ReplayGraph are implemented when necessary
    adrianlizarraga committed 128 days ago
  • Address comments
    adrianlizarraga committed 128 days ago
  • Sync IO binding inputs in test
    adrianlizarraga committed 128 days ago
  • Improved documentation for graph annotation ID
    adrianlizarraga committed 128 days ago
Loading