This project defines tooling and tests comprising IREE's model regression testing suite. It aims operate at the compiler-input level and choreographs the normal tool flows of iree-compile
, iree-run-module
, iree-benchmark-module
, etc.
If you have IREE tools on your path or have a virtual environment setup:
pip install -e experimental/regression_suite PATH=../iree-build/tools:$PATH \ pytest experimental/regression_suite
Useful options:
-s
: Stream all test output.-m MARKEXPR
: Select subsets of the test suite.Common marker selections:
-m "plat_host_cpu and presubmit"
: Run the host-CPU tests configured for presubmit on the CI.-m "plat_rdna3_vulkan and presubmit"
: Run the host-CPU tests configured for presubmit on the CI.You can display all markers with pytest experimental/regression_suite --markers
NOTE: For this to work, you must previously have installed GitHub command line tools and authenticated (gh auth
). See https://cli.github.com/.
The test suite doesn't care how you get tools on your path, but a common case is to run the regression suite from built compilers and tools from a GitHub presubmit or postsubmit run. This can be done in one step by setting up a venv:
deactivate # If have any venv active. python ./build_tools/pkgci/setup_venv.py \ /tmp/iree_gh_venv \ --fetch-gh-workflow=${RUN_ID} \ [--compiler-variant=asserts] [--runtime-variant=asserts] source /tmp/iree_gh_venv/bin/activate # Then install the regression suite and run pytest as usual: pip install -e experimental/regression_suite
In the above, <<RUN_ID>>
is the value in any GitHub action presubmit/postsubmit workflow which has built package artifacts. As an example, if looking at a GitHub Actions status page on: https://github.com/iree-org/iree/actions/runs/5957351746/job/16159877442
, then the run id is the first number in the URL (5957351746).
Running the above will allow you to run pytest
and you will have tools as built at the commit from which the workflow run originated.