Skip to content

Isolate MIOpen cache per xdist worker to avoid lock contention #804

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
psanal35 wants to merge 5 commits into
base: amd-main
Choose a base branch
from
Open

Isolate MIOpen cache per xdist worker to avoid lock contention #804

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
90a9455
[ROCm] Add TheRock 7.13 and TheRock latest to CI test matrices
mminutoli May 21, 2026
1596135
[ROCm] Embed TheRock-aware wheel RUNPATH at link time
gulsumgudukbay Jun 5, 2026
ea0ac83
[ROCm] Put TheRock SDK bin on PATH via so rocminfo/rocm-smi resolve …
gulsumgudukbay Jun 8, 2026
88e42dc
Resolve ROCm workflow conflict markers
psanal35 Jun 18, 2026
b96c0a9
Isolate MIOpen cache per xdist worker to avoid lock contention
psanal35 Jun 18, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 9 additions & 3 deletions .github/workflows/wheel_tests_continuous.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -358,7 +358,9 @@ jobs:
artifact: ["jax-rocm-plugin", "jax-rocm-pjrt"]
python: ["3.12", "3.13", "3.14"]
rocm: [
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", manylinux-image: ""},
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", manylinux-image: ""},
{label: "TheRock 7.13", wheel-version: "7", release-version: "therock-7.13.0", manylinux-image: "ghcr.io/rocm/jax-manylinux_2_28-therock-7.13.0:latest"},
{label: "TheRock latest", wheel-version: "7", release-version: "therock-latest", manylinux-image: "ghcr.io/rocm/jax-manylinux_2_28-therock-latest:latest"},
]
exclude:
- artifact: "jax-rocm-pjrt"
Expand Down Expand Up @@ -391,7 +393,9 @@ jobs:
runner: ["linux-x86-64-1gpu-amd", "linux-x86-64-4gpu-amd", "linux-x86-64-8gpu-amd"]
python: ["3.12"]
rocm: [
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", tag: "rocm720"},
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", tag: "rocm720"},
{label: "TheRock 7.13", wheel-version: "7", release-version: "therock-7.13.0", tag: "therock-7.13.0"},
{label: "TheRock latest", wheel-version: "7", release-version: "therock-latest", tag: "therock-latest"},
]
name: "Pytest ROCm (${{ matrix.rocm.label }})"
with:
Expand All @@ -418,7 +422,9 @@ jobs:
runner: ["linux-x86-64-4gpu-amd", "linux-x86-64-8gpu-amd"]
python: ["3.12"]
rocm: [
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", tag: "rocm720"},
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", tag: "rocm720"},
{label: "TheRock 7.13", wheel-version: "7", release-version: "therock-7.13.0", tag: "therock-7.13.0"},
{label: "TheRock latest", wheel-version: "7", release-version: "therock-latest", tag: "therock-latest"},
]
enable-x64: [0]
name: "Bazel ROCm (${{ matrix.rocm.label }})"
Expand Down
12 changes: 9 additions & 3 deletions .github/workflows/wheel_tests_nightly_release.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -187,7 +187,9 @@ jobs:
artifact: ["jax-rocm-plugin", "jax-rocm-pjrt"]
python: ["3.12", "3.13", "3.14"]
rocm: [
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", manylinux-image: ""},
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", manylinux-image: ""},
{label: "TheRock 7.13", wheel-version: "7", release-version: "therock-7.13.0", manylinux-image: "ghcr.io/rocm/jax-manylinux_2_28-therock-7.13.0:latest"},
{label: "TheRock latest", wheel-version: "7", release-version: "therock-latest", manylinux-image: "ghcr.io/rocm/jax-manylinux_2_28-therock-latest:latest"},
]
exclude:
- artifact: "jax-rocm-pjrt"
Expand Down Expand Up @@ -220,7 +222,9 @@ jobs:
runner: ["linux-x86-64-1gpu-amd", "linux-x86-64-4gpu-amd", "linux-x86-64-8gpu-amd"]
python: ["3.12", "3.13", "3.14"]
rocm: [
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", tag: "rocm720"},
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", tag: "rocm720"},
{label: "TheRock 7.13", wheel-version: "7", release-version: "therock-7.13.0", tag: "therock-7.13.0"},
{label: "TheRock latest", wheel-version: "7", release-version: "therock-latest", tag: "therock-latest"},
]
name: "Pytest ROCm (${{ matrix.rocm.label }})"
with:
Expand All @@ -247,7 +251,9 @@ jobs:
runner: ["linux-x86-64-1gpu-amd"]
python: ["3.12", "3.13", "3.14"]
rocm: [
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", tag: "rocm720"},
{label: "ROCm 7.2.0", wheel-version: "7", release-version: "7.2.0", tag: "rocm720"},
{label: "TheRock 7.13", wheel-version: "7", release-version: "therock-7.13.0", tag: "therock-7.13.0"},
{label: "TheRock latest", wheel-version: "7", release-version: "therock-latest", tag: "therock-latest"},
]
enable-x64: [0]
name: "Bazel ROCm (${{ matrix.rocm.label }})"
Expand Down
9 changes: 9 additions & 0 deletions conftest.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,3 +95,12 @@ def pytest_collection() -> None:
# HIP_VISIBLE_DEVICES to all GPUs; override to "0" so HIP doesn't try to
# enable agents that ROCr just hid.
os.environ["HIP_VISIBLE_DEVICES"] = "0"

# MIOpen serializes kernel lookup through a shared user-DB lock under
# /tmp/miopen-lockfiles. Use a per-worker cache directory to avoid lock
# contention when xdist workers share the same GPU.
miopen_cache_dir = f"/tmp/miopen-{xdist_worker_name}"
os.makedirs(miopen_cache_dir, exist_ok=True)
os.environ["MIOPEN_USER_DB_PATH"] = miopen_cache_dir
os.environ["MIOPEN_CUSTOM_CACHE_DIR"] = miopen_cache_dir

1 change: 0 additions & 1 deletion jaxlib/rocm/rocm_rpath.bzl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,7 +65,6 @@ def _rocm_wheel_rpaths():
where TheRock is in the container's system Python). The legacy `/opt/rocm`
fallback is appended by the caller so it stays last.
"""

# `_rocm_sdk_core/lib` holds the non-arch runtime libs; `_rocm_sdk_libraries
# [...]/lib` holds the math libs + per-arch kernel data. site_libs covers the
# multi-arch + per-family packages; core_libs is the cross-Python/absolute
Expand Down