Update Docker environment for MI355X benchmark reproducibility#202
Open
amd-andycha wants to merge 2 commits intofeat/reducescatter-sdmafrom
Open
Update Docker environment for MI355X benchmark reproducibility#202amd-andycha wants to merge 2 commits intofeat/reducescatter-sdmafrom
amd-andycha wants to merge 2 commits intofeat/reducescatter-sdmafrom
Conversation
- Update base image from ROCm 6.4 to ROCm 7.1.1 (PyTorch 2.10.0) to support MI355X gfx950 architecture - Add Python dependencies: amd-aiter (INT8 GEMM overlap tests), ninja, prettytable, pytest-assume, setuptools-scm - Add MORI build step: COPY source, compile with MORI_GPU_ARCHS=gfx950, use --no-build-isolation to leverage base image PyTorch - Set required environment variables: HSA_NO_SCRATCH_RECLAIM=1, PYTHONPATH - Add .dockerignore to exclude .git/, build/, *.csv (reduces build context from 235MB to ~19MB) - Add docker/run_benchmark.sh helper script with GPU passthrough flags - Update README with Docker usage, benchmark commands, and non-Docker installation instructions
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Update
docker/Dockerfile.devto produce a self-contained Docker image that allows users to reproduce all CCL benchmark experiments (AllReduce, AllGather, ReduceScatter, and GEMM overlap tests) on any MI355X machine with a singledocker run. Also update README with comprehensive Docker and installation instructions.Changes
docker/Dockerfile.dev.dockerignore.git/,build/,*.csv, etc. from build contextdocker/run_benchmark.shREADME.mdKey changes in Dockerfile
rocm/pytorch:rocm6.4.3_ubuntu22.04_py3.10torocm/pytorch:rocm7.1.1_ubuntu22.04_py3.10_pytorch_release_2.10.0amd-aiter(INT8 GEMM for overlap tests),ninja,prettytable,pytest-assume,setuptools-scmCOPYsource, compile withMORI_GPU_ARCHS=gfx950+--no-build-isolationHSA_NO_SCRATCH_RECLAIM=1,PYTHONPATH,SETUPTOOLS_SCM_PRETEND_VERSIONUsage
Build:
cd mori
docker build -t rocm/mori:benchmark -f docker/Dockerfile.dev .
Run:
bash docker/run_benchmark.sh
Inside container:
bash tests/python/ccl/bench_allgather_sweep.sh
bash tests/python/ccl/bench_reducescatter_sweep.sh
bash tests/python/ccl/bench_ag_overlap_sweep.sh
bash tests/python/ccl/bench_rs_overlap_sweep.sh
Test plan
docker buildsucceeds from cleangit clone(~1 min with cached base image)torch,mori.shmem,mori.ccl.AllreduceSdma,mori.ccl.ReduceScatterSdma,aiter