Skip to content

[WIP]Feature/warp tracing - #7

Open
frlai wants to merge 39 commits into
developfrom
feature/warp_tracing
Open

[WIP]Feature/warp tracing#7
frlai wants to merge 39 commits into
developfrom
feature/warp_tracing

Conversation

@frlai

@frlai frlai commented Jun 23, 2026

Copy link
Copy Markdown
Collaborator

Summary

Adds experimental NVIDIA Warp tracing/export support to LEAPP, including explicit annotate.warp_op(...) regions, automatic Warp segment detection, APIC capture replay, and ONNX/ExportedProgram export support through a leapp::warp_runner custom operator.

Details

  • Adds Warp tracing datatypes, session state, boundary detection, and two-pass discovery/capture flow.
  • Registers leapp::warp_runner and lowers it to com.nvidia.warp::WrpRunner for ONNX exports.
  • Adds WRPB bundle packing and C++ runtime support for ONNX Runtime and Torch/LibTorch adapters.
  • Adds Warp tracing docs, setup guidance, ONNX custom-op build instructions, and InferenceManager smoke-test guidance.
  • Expands functional/unit coverage for Warp tracing, export validation, runtime behavior, dtype handling, and compatibility paths.

Notes

  • Warp tracing is currently Linux/CUDA-only.
  • Warp ONNX inference requires onnxruntime-gpu, CUDA EP availability, and LEAPP_WARP_ONNX_CUSTOM_OP_LIBRARY pointing at a custom op built against the same ONNX Runtime version in the environment.
  • Boundary dtypes are currently limited to primitive scalar types; vector, quaternion, matrix, and transform dtypes are not supported at Warp region boundaries.

@frlai
frlai force-pushed the feature/warp_tracing branch from 67d95f1 to b83ce24 Compare June 26, 2026 17:56
@frlai
frlai force-pushed the feature/warp_tracing branch from ffbc2a2 to 17f8bc8 Compare July 8, 2026 15:35
@frlai
frlai force-pushed the feature/warp_tracing branch from 5537959 to cb014d7 Compare July 20, 2026 23:58
frlai added 27 commits July 26, 2026 00:56
…matic i/o detection for node based operators
…be registered as part of onnx and torchscript via the script path, export to onnx seems to work
…doesn't support jit-trace, onnx-torchscript or jit-script.
updated input signature to use Tensor[] inputs, str runtime_metadata, Tensor bundle.
all metadata will be stored inside runtime_metadata. this includes the mask and i/o configs

todo:
need to prune runtime_metadata for uneeded fields.
Squashed from two_pass_apic_capture:
- implements two pass tracing
- force ptx sm 75
- updated to use single pass, refactored code ownership
…on and broadened support for automatic reentry validation
…stricter hierarchy, fixed bug with seg fault on reentry because loss of buffer creation ownership

 still need to fix bug with incorrect number of i/o at the boundry
…orch reuse cuda memory so torch memory creation functions sill trip the cupti oracle
@frlai
frlai force-pushed the feature/warp_tracing branch from a2b56d6 to 3dafdbf Compare July 26, 2026 18:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant