Add single-precision and MI300X variants to docker-publish workflow#132
Merged
Merged
Conversation
Extend the GitHub Actions docker-publish matrix so each published image
is available in both single and double precision, and add MI300X
(gfx942) as a new published target.
- docker/x86_sm100/Dockerfile, docker/x86_gfx942/Dockerfile: add a
SELF_DOUBLE_PRECISION build-arg (default ON) wired to the
SELF_ENABLE_DOUBLE_PRECISION CMake option, so the same Dockerfile
builds either precision.
- docker/x86_gfx942/Dockerfile: new, modeled on the gfx90a Dockerfile,
targeting gfx942 via the latest-x86-rocm643-gfx942 selfish base image.
- docker-publish.yml: pass SELF_DOUBLE_PRECISION through build-args and
add matrix entries. Published tags:
latest-x86-cuda130-sm100 (double, unchanged)
latest-x86-cuda130-sm100-sp (single, new)
latest-x86-rocm643-gfx942 (double, new)
latest-x86-rocm643-gfx942-sp (single, new)
Existing tags are unchanged for backward compatibility; single-precision
images use the -sp suffix. Each variant keeps a separate buildcache tag.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
fluidnumericsJoe
added a commit
that referenced
this pull request
Jun 13, 2026
Add double- and single-precision MI210 (gfx90a) variants to the docker-publish matrix, and wire the SELF_DOUBLE_PRECISION build arg into the x86_gfx90a Dockerfile so the single-precision variant builds correctly. The MI300X (gfx942) and NVIDIA variants previously included on this branch are already on main via #132, so only the new gfx90a builds remain here. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
fluidnumericsJoe
added a commit
that referenced
this pull request
Jun 13, 2026
Add double- and single-precision MI210 (gfx90a) variants to the docker-publish matrix, and wire the SELF_DOUBLE_PRECISION build arg into the x86_gfx90a Dockerfile so the single-precision variant builds correctly. The MI300X (gfx942) and NVIDIA variants previously included on this branch are already on main via #132, so only the new gfx90a builds remain here. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
fluidnumericsJoe
added a commit
that referenced
this pull request
Jun 13, 2026
Add double- and single-precision MI210 (gfx90a) variants to the docker-publish matrix, and wire the SELF_DOUBLE_PRECISION build arg into the x86_gfx90a Dockerfile so the single-precision variant builds correctly. The MI300X (gfx942) and NVIDIA variants previously included on this branch are already on main via #132, so only the new gfx90a builds remain here. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Extends the GitHub Actions
docker-publishworkflow so each published image is offered in both single and double precision, and adds MI300X (gfx942) as a new published target.The published-tag matrix becomes:
x86-cuda130-sm100latest-x86-cuda130-sm100(unchanged)x86-cuda130-sm100-splatest-x86-cuda130-sm100-sp(new)x86-rocm643-gfx942latest-x86-rocm643-gfx942(new)x86-rocm643-gfx942-splatest-x86-rocm643-gfx942-sp(new)(Each also gets the corresponding
<sha>-…tag.)Changes
docker/x86_sm100/Dockerfile,docker/x86_gfx942/Dockerfile— add aSELF_DOUBLE_PRECISIONbuild-arg (defaultON) wired to the existingSELF_ENABLE_DOUBLE_PRECISIONCMake option, so one Dockerfile builds either precision.docker/x86_gfx942/Dockerfile— new file, modeled on the existingx86_gfx90a(MI210) Dockerfile, targetinggfx942via thelatest-x86-rocm643-gfx942selfish base image (verified present on Docker Hub)..github/workflows/docker-publish.yml— passSELF_DOUBLE_PRECISIONthroughbuild-argsand add the four matrix entries.Notes / backward compatibility
latest-x86-cuda130-sm100tag keeps its name and stays double precision — purely additive. Single precision uses the-spsuffix.buildcache-<tag_suffix>cache tag, so no cache collisions; all four build in parallel (fail-fast: false).ubuntu-22.04runner. The Buildkite pipeline remains the path that build+tests AMD images on real hardware (MI210). The gfx942 image is built but not exercised on an MI300X here.x86-rocm643-gfx90aconvention rather than anmi300xalias, for consistency across the GHA and Buildkite pipelines.Verification
The workflow only triggers on push to
main/workflow_dispatch, so this PR won't exercise it. After merge, recommend a manualworkflow_dispatchrun to confirm the two new AMD builds pull the gfx942 base and compile.🤖 Generated with Claude Code