Skip to content

fix: remove unverified Iluvatar graph support#25

Merged
voltjia merged 1 commit into
InfiniTensor:masterfrom
gongchensu:fix/remove-iluvatar-graph-runtime
Jul 10, 2026
Merged

fix: remove unverified Iluvatar graph support#25
voltjia merged 1 commit into
InfiniTensor:masterfrom
gongchensu:fix/remove-iluvatar-graph-runtime

Conversation

@gongchensu

@gongchensu gongchensu commented Jul 10, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Remove Iluvatar from the graph runtime contract until the backend graph capture/replay path is validated.
  • Drop the Iluvatar graph test registration from tests/CMakeLists.txt so CI no longer builds unverified Iluvatar graph APIs.
  • Keep the existing Iluvatar CUDA-style runtime support for device, memory, stream, and event APIs.

Motivation

InfiniTensor/InfiniOps#790 fails in the Iluvatar unit CI after InfiniRT PR #9 because the downstream job installs InfiniRT from master and builds with CoreX 4.3 / CUDA 10.2-compatible headers. That environment exposes the legacy five-argument cudaGraphInstantiate signature, while the new InfiniRT graph runtime dispatch validates and calls GraphInstantiate(GraphExec*, Graph).

The Iluvatar graph backend has not been validated yet. The conservative fix is to stop advertising Iluvatar graph support until hardware validation is available, which keeps the rest of the Iluvatar runtime usable and avoids breaking downstream InfiniOps CI.

Related to InfiniTensor/InfiniOps#790.

Type of Change

  • feat - new feature / new operator / new platform
  • fix - bug fix
  • perf - performance improvement (no behavioral change)
  • refactor - code restructuring without behavior change
  • test - adding or fixing tests only
  • docs - documentation only
  • build / ci - build system or CI configuration
  • chore - tooling, formatting, or other non-code changes
  • Breaking change (requires a ! in the Conventional Commits prefix or a BREAKING CHANGE: footer)

Platforms Affected

  • CPU (WITH_CPU)
  • NVIDIA (WITH_NVIDIA)
  • Iluvatar (WITH_ILUVATAR)
  • MetaX (WITH_METAX)
  • Cambricon (WITH_CAMBRICON)
  • Moore (WITH_MOORE)
  • Ascend (WITH_ASCEND)
  • PyTorch C++ bindings (WITH_TORCH)
  • Build system / CMake / CI
  • Python bindings / user-facing API

Smoke Test Result

GitHub PR checks on InfiniTensor/InfiniRT#25:
- clang-format: passed
- ruff: passed
- Documentation Pages / build: passed

Iluvatar hardware smoke was not run in this PR. The change intentionally removes unverified Iluvatar graph support from the runtime contract and test registration.

Test Results on Supported Platforms

Platform Affected Build / Smoke Result Full Result / Notes
NVIDIA No N/A - not affected N/A
Iluvatar Yes PR CI passed; Iluvatar hardware smoke not run. Graph runtime support is disabled until backend validation is available.
MetaX No N/A - not affected N/A
Cambricon No N/A - not affected N/A
Moore No N/A - not affected N/A
Ascend No N/A - not affected N/A
Full `pytest` output (optional)
N/A - backend pytest was not run for this PR.

Benchmark / Performance Impact

N/A - this only removes unverified graph API exposure for Iluvatar and does not change kernel performance paths.

Notes for Reviewers

  • This is the conservative fix for the InfiniOps Iluvatar CI failure: do not expose graph APIs for Iluvatar until the backend has been validated on CoreX hardware.
  • A follow-up can re-enable Iluvatar graph support once capture/replay behavior and the vendor cudaGraphInstantiate signature are covered by CI or hardware smoke testing.
  • The rest of the Iluvatar runtime remains under CudaRuntime; this PR only removes the graph-specific type aliases, graph methods, and graph test registration.

@gongchensu gongchensu requested a review from voltjia July 10, 2026 03:00
@gongchensu gongchensu self-assigned this Jul 10, 2026
@voltjia voltjia merged commit 3fc9170 into InfiniTensor:master Jul 10, 2026
4 checks passed
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.

2 participants