Skip to content

Support Nvidia build on theRock for HIP-tests#3335

Open
litvaOo wants to merge 31 commits intodevelopfrom
dev/amd/dlytovch/therock-nvidia
Open

Support Nvidia build on theRock for HIP-tests#3335
litvaOo wants to merge 31 commits intodevelopfrom
dev/amd/dlytovch/therock-nvidia

Conversation

@litvaOo
Copy link
Contributor

@litvaOo litvaOo commented Feb 18, 2026

Motivation

The current implementation of hip-tests CMakeLists.txt does not support build of hip-tests on Nvidia. This PR adds corresponding support

Technical Details

This PR mainly does two things:

  1. Links to required CUDA libraries.
  2. Updates the custom targets to use nvcc directly, as previous version relied on the hipcc converting the arguments to nvcc.

Additionally, some fixes for hipother to fix compatibility with multiple CUDA version are added: namely workarounds for mismatch between driver and runtime memory location types and Nvidia guards for some tests.
NOTE: some tests are disabled due to currently unresolvable issues with CUDA compatibility, e.g lack of hipStreamGetCaptureInfo_V2 in latest CUDA. These fixes are out of scope for this PR and require separate PR discussion

JIRA ID

NA

Test Plan

HIP-tests should be built successfully as-is on the current TheRock setup, and successfully built for NVidia

Test Result

Nvidia built passing locally, not running in CI yet due to lack of the infrastructure.
AMD build passing locally, waiting for PSDB to pass

Submission Checklist

Copilot AI review requested due to automatic review settings February 18, 2026 15:36
@litvaOo litvaOo requested review from a team as code owners February 18, 2026 15:36
@litvaOo litvaOo force-pushed the dev/amd/dlytovch/therock-nvidia branch from 4e15013 to 86fef42 Compare February 18, 2026 15:37
@litvaOo litvaOo force-pushed the dev/amd/dlytovch/therock-nvidia branch from 86fef42 to e4a966e Compare February 18, 2026 15:47
Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

@litvaOo litvaOo force-pushed the dev/amd/dlytovch/therock-nvidia branch 2 times, most recently from 3cc1bbf to 9a96157 Compare February 18, 2026 16:52
@litvaOo litvaOo force-pushed the dev/amd/dlytovch/therock-nvidia branch 6 times, most recently from b6f1dba to 0c2eb76 Compare February 26, 2026 07:54
@mangupta mangupta requested a review from emankov February 26, 2026 16:11
Copy link

@emankov emankov left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is a good start on getting the NVIDIA CI green, but there are a few critical issues here - including segfault risks, race conditions, and some heavy CMake redundancies - that need to be addressed before this can be merged.

@emankov
Copy link

emankov commented Feb 26, 2026

Sorry, @chrispaquot, I don't think this PR is ready to merge due to critical issues.

@litvaOo litvaOo force-pushed the dev/amd/dlytovch/therock-nvidia branch from ca84502 to 426065f Compare February 27, 2026 08:49
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants