Open mdboom opened 3 months ago
pyperformance currently very much by design only runs on a single core.
Running benchmarks in parallel would really help with overall runtime, but it may be tricky to only run codespeed-runner on the subprocesses and not the orchestration process. The codspeed action seems to always run everything inside of valgrind.
We could also use multiple Github Action jobs to create parallelism, but for CPython where compilation time is rather significant, we would have to pass the built artifacts from one job to another to avoid duplicating that work.
pyperformance currently very much by design only runs on a single core.
Running benchmarks in parallel would really help with overall runtime, but it may be tricky to only run codespeed-runner on the subprocesses and not the orchestration process. The codspeed action seems to always run everything inside of valgrind.
We could also use multiple Github Action jobs to create parallelism, but for CPython where compilation time is rather significant, we would have to pass the built artifacts from one job to another to avoid duplicating that work.