Skip to content

[TEST][CI] Restrict CI to run only Adapters/max_malloc.cpp - #23141

Draft
ldorau wants to merge 5 commits into
intel:syclfrom
ldorau:Enable_sycl_test-e2e_Adapters_max_malloc.cpp_test
Draft

ldorau wants to merge 5 commits into
intel:syclfrom
ldorau:Enable_sycl_test-e2e_Adapters_max_malloc.cpp_test

Conversation

@ldorau

@ldorau ldorau commented Sep 10, 2026

Copy link
Copy Markdown
Contributor

No description provided.

Temporarily restrict sycl-linux-precommit.yml to a single job chain
(detect_changes -> build -> E2E) and a single E2E matrix entry
(GEN 12 Integrated), with LIT_FILTER set to only run
sycl/test-e2e/Adapters/max_malloc.cpp. This isolates the test re-enabled
in the previous commit so its result can be checked in CI without the
noise/cost of running the full E2E matrix.

This is a temporary, debug-only change and should be reverted afterwards.
@ldorau
ldorau force-pushed the Enable_sycl_test-e2e_Adapters_max_malloc.cpp_test branch 3 times, most recently from 5b3e5a6 to 9a135b2 Compare September 14, 2026 09:18
ldorau and others added 3 commits September 14, 2026 10:05
sycl/test-e2e/Adapters/max_malloc.cpp is being reproduced/verified on a
temporary CI configuration that runs only this test (LIT_FILTER
restriction in sycl-linux-precommit.yml). Its lit per-test watchdog
(config.maxIndividualTestTime, 300s in run-only mode) may be too short
for this test on some devices, contributing to the timeouts tracked in
issue intel#22227.

lit has no per-file timeout directive, and maxIndividualTestTime is
resolved per-directory (TestingConfig), not per single test file, so it
cannot be scoped to only max_malloc.cpp without moving the test into
its own subdirectory. Since CI is currently restricted to run only this
one test anyway, bump the shared run-only value from 300 to 400 seconds
here instead.

Temporary/debug-only change, to be reverted once the reproduction is
complete, along with the other DEBUG-marked commits on this branch.
….cpp

ZE_DEBUG=-1, ZE_ENABLE_VALIDATION_LAYER=1, and SYCL_UR_TRACE=-1 were
added to Adapters/max_malloc.cpp's RUN line for Linux CI diagnostics,
but the RUN line isn't OS-scoped, so it also applied on Windows.

On the Windows BMG (Arc B580, Level-Zero V2) CI job, this caused
max_malloc.cpp to crash with 0xC0000409 (STATUS_STACK_BUFFER_OVERRUN)
during zeInit/driver-extension enumeration, whereas the same test on
the same hardware passes cleanly without these env vars (confirmed via
a concurrent unrelated PR run). ZE_ENABLE_VALIDATION_LAYER=1 in
particular appears to interact badly with this driver/adapter
combination on Windows.

Wrap the debug env vars in a `%if linux %{ ... %}` lit conditional so
they only apply on Linux, restoring the original (working) Windows
behavior while keeping the Linux diagnostics intact.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Add the "Intel / Ponte Vecchio GPU" runner to the temporarily-restricted
E2E matrix in sycl-linux-precommit.yml, alongside the existing GEN12
entry, both scoped via LIT_FILTER to only run Adapters/max_malloc.cpp.

PVC is where max_malloc.cpp was observed to genuinely fail
(malloc_device() returning nullptr for a 44Gb allocation on a 48Gb
device, in an unrestricted full-suite CI run on PR intel#23142), so it needs
to be included in this isolated reproduction to investigate/verify that
failure directly.

Temporary/debug-only change, to be reverted once the reproduction is
complete, along with the other DEBUG-marked commits on this branch.
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