Skip to content

intel-pti: Add version 1.1.0 - #2938

Draft
luhenry wants to merge 4 commits into
mainfrom
intel-pti
Draft

luhenry wants to merge 4 commits into
mainfrom
intel-pti

Conversation

@luhenry

@luhenry luhenry commented Oct 13, 2026 •

Copy link
Copy Markdown
Member

Compiles the MIT-licensed PTI SDK (libpti_view.so and libpti.so) from the pti-1.1.0 tag, the exact commit embedded in Intel's x86_64 wheel. Upstream publishes no riscv64 wheel.

Mirrors upstream's sdk_build_and_test.yml and builds with the sdk's own scikit-build-core pyproject.toml (PTI_RELEASE_WHL=1).

Differs from upstream

  • GCC (upstream's linux-gxx-release preset) instead of icpx - no oneAPI compiler on riscv64
  • libze_loader.so.1 not bundled - same as Intel's x86_64 wheel
  • Wheel is py3-none; Intel's is py2.py3-none

Testing

  • Only the unit and libload ctest labels run - every other label needs an Intel GPU
  • Smoke test loads libpti_view.so.1 and checks ptiVersionString()

License: Wheel statically links spdlog and fmt (MIT); the repo's third-party-programs.txt ships alongside PTI's own LICENSE.

Built once (py3-none); CI results pending.

luhenry added a commit that referenced this pull request Oct 13, 2026
@luhenry
luhenry marked this pull request as draft October 13, 2026 09:27
@github-actions

Copy link
Copy Markdown
Contributor
PR Preview Action v1.8.1

QR code for preview link

🚀 View preview at
https://riseproject-dev.github.io/python-wheels/pr-preview/pr-2938/

Built to branch gh-pages at 2026-10-13 10:07 UTC.
Preview will be ready when the GitHub Pages deployment is complete.

The manylinux image's glibc warns when _FORTIFY_SOURCE is set without
optimisation, and upstream's wheel flags add -Werror, so the unoptimised
C++17 filesystem probe failed and configure stopped with "C++17 filesystem
support is required".
pip runs the build backend from ./sdk, so build-dir=build landed in
sdk/build and the ctest step that follows found no /work/build.
ze_event_manager_tests, ze_events_pools_observer_test and
pc_sampling_internal_test are labelled "unit" but call
zeInit(ZE_INIT_FLAG_GPU_ONLY) in their fixtures, which returns
ZE_RESULT_ERROR_UNINITIALIZED without an Intel GPU, so 73 of 136 tests
failed. Exclude hw-metrics as upstream's skip_hw_metrics does, and the
two event pool suites by name.

This branch has not been deployed

No deployments
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