Repository navigation
v2.0.1: download traces from GitHub release assets, and release notes - #568
Merged
Merged
Conversation
…he FTP server get-accel-sim-traces.py now reads manifest.json from a GitHub release, lists the available GPU/workload/argument sets, downloads the selected tar parts (each under 2 GB), checks every part and the whole tar against SHA-256, and extracts them to hw_run/traces/<GPU>/<workload>/<args>/. Private repositories use GITHUB_TOKEN or the gh login. The old FTP trace listing is removed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
publish-trace-release.py packs one tar per GPU/workload/argument set, splits it under GitHub's 2 GiB asset limit, uploads the parts and writes manifest.json plus a README.md with download instructions. ncu_stats.csv now sits beside traces/ so the extracted directory works with both run_simulations.py -T and plot-correlation.py -H. get-accel-sim-traces.py defaults to the v2.0.1 release of accel-sim-framework. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Trace sets now ship as <GPU>/traces/device-0/<cuda>/<workload>/<args>/traces/ plus the Nsight Compute stats in <GPU>/device-0/<cuda>/<workload>/<args>/, extracted under hw_run/, so run_simulations.py -T and plot-correlation.py -H use them directly. Only the kernels listed in each kernelslist.g are packed. The manifest moves to format 2 with per-set trace and hardware directories, and the download checks that every member stays inside them and that the kernel count matches. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Publishing now lives in the publish-trace-release skill of the AALP accel-sim-research plugin; the repository keeps only the download side. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
JRPan
force-pushed
the
trace-release-download
branch
from
October 20, 2026 18:12
75f17b5 to
b8861f2
Compare
main-sass (smoke and full), main-tracer and the weekly jobs launch with --per-kernel --array -M 10G (SIM_PER_KERNEL in lib/common.sh) and verify with --per-kernel-dirs, as the weekly ML job already did. A long kernel no longer serialises the rest of its app or delays its failure: the cutlass_48 warp-specialized GEMMs failed at kernel 4 only after 2-3h of kernels 1-3. main-docker is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
They were copied into accel-sim/H200/hw_run (device-7) from the H200_ml_traces layer, with the same 29 kernels and HW rows left out; the CI no longer reads another user's depot. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…v2.0.1 release notes Co-Authored-By: Claude Opus 5.5 <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
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.
Prepares the v2.0.1 release. Pre-collected traces move from the FTP server to GitHub release assets and ship with the code release they were validated with.
Trace download (
get-accel-sim-traces.py)manifest.jsonfrom a GitHub release (default:accel-sim/accel-sim-framework, tagv2.0.1).-llists the sets.-a GPU[/workload[/args]]downloads a selection, withallallowed at any level.hw_run/in the standard Accel-Sim layout:<GPU>/traces/device-0/<cuda>/<workload>/<args>/traces/is forrun_simulations.py -T hw_run/<GPU>/traces/device-0/<cuda>.<GPU>/device-0/<cuda>/<workload>/<args>/ncu.csv.*is forplot-correlation.py -H hw_run/<GPU>.kernelslist.gmust match the manifest.GITHUB_TOKENor theghlogin.Publishing
Not part of this repository. The release assets are produced by the
publish-trace-releaseskill in the AALP accel-sim-research plugin (purdue-aalp/AALP-claude-plugins#25).Release notes
release.notes.md, covering Revert slurm.sim to the pre-2.0 template (fixes-l locallaunches) #556, Clone gpgpu-sim for issue triage and answer questions instead of only triaging #559, Decode every lane's mbarrier address when dropping unused TRYWAITs #564 and CI redesign: PR smoke tier, correlated dev runs, merged Weekly with H200 ML #566, this PR, and the GPGPU-Sim changes (Issue about using traces generated from 2080Ti-GPU to simulate QV100 #144, accel-framework is not generating stats for all the kernels in rodinia-2.0-ft with simulator #146, Update get-accel-sim-traces.py #147).versioninCITATION.cffto 2.0.1.Release plan
v2.0.1is a pre-release, currently on dev acc3ab9, and the H200 ML traces are being published to it. That is 20 argument sets and 8942 kernels; a few kernels that do not complete in simulation are left out.v2.0.1tag to the merge commit. The release keeps its assets.gpgpu-sim_distributionv2.0.1(dev, d922cf39). Set the release date in the notes, and mark the release official.Testing
v2.0.1pre-release once the upload finishes.run_simulations.pyandplot-correlation.pyare unchanged. Their directory lookups are already covered by CI.🤖 Generated with Claude Code