You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Move QDP Python tests from qdp/qdp-python/tests/ to testing/qdp_python/ so all Python tests live under the top-level testing/ tree and are run by make test_python / root pytest.
Change the default Python test run to single-process (no -n auto) to avoid CUDA out-of-memory and flakiness when multiple workers use the GPU; document that users with sufficient compute can optionally use -n auto or -n N to speed up.
Why
Test location: QDP Python tests live under qdp/qdp-python/tests/, while the rest of the Python suite is under testing/ (e.g. testing/qumat, testing/qdp). Root pytest uses testpaths = ["testing"], so the qdp-python tests were not discovered by the default make test_python run unless invoked from the right place or with an extra path.
Parallelism: Running the full suite with -n auto can cause CUDA OOM or instability when several workers use the GPU. A single-process default is safer; users who have enough GPU memory and want faster runs should be able to opt in to parallelism via documentation.
What
qdp/qdp-python/tests/totesting/qdp_python/so all Python tests live under the top-leveltesting/tree and are run bymake test_python/ root pytest.-n auto) to avoid CUDA out-of-memory and flakiness when multiple workers use the GPU; document that users with sufficient compute can optionally use-n autoor-n Nto speed up.Why
qdp/qdp-python/tests/, while the rest of the Python suite is undertesting/(e.g.testing/qumat,testing/qdp). Rootpytestusestestpaths = ["testing"], so the qdp-python tests were not discovered by the defaultmake test_pythonrun unless invoked from the right place or with an extra path.-n autocan cause CUDA OOM or instability when several workers use the GPU. A single-process default is safer; users who have enough GPU memory and want faster runs should be able to opt in to parallelism via documentation.