Skip to content

Fix miopen cache dir handling #69

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
PatrickRMiles wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix miopen cache dir handling #69

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fe28b02
fix miopen cache dir handling, and ensure torchrun-hpc sets correct c…
May 14, 2026
b425493
add default miopen cache path to gitignore
May 14, 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
1 change: 1 addition & 0 deletions .gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@ htmlcov/
.coverage
.coverage.*
.cache
.scaffold-miopen-cache/
nosetests.xml
coverage.xml
*.cover
Expand Down
6 changes: 6 additions & 0 deletions scripts/scaffold-tuolumne-torchpypi.job
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,12 @@ ml cce/21.0.0 cray-mpich/9.1.0 rocm/7.1.1 rccl/fast-env-slows-mpi
# Use ccl plugin that we manually built with install-rccl.sh
export NCCL_NET_PLUGIN=../aws-ofi-nccl.git/install/lib/librccl-net.so

# Persist MIOpen's find database and compiled-kernel cache across job submissions.
# On Tuolumne, torchrun-hpc derives the MIOpen cache paths from TMPDIR.
export SCAFFOLD_MIOPEN_CACHE_ROOT="${SCAFFOLD_MIOPEN_CACHE_ROOT:-$(pwd)/.scaffold-miopen-cache}"
export TMPDIR="${SCAFFOLD_MIOPEN_TMPDIR:-${SCAFFOLD_MIOPEN_CACHE_ROOT}/tmp}"
mkdir -p "${TMPDIR}/MIOpen_user_db" "${TMPDIR}/MIOpen_custom_cache"

# Disable direct convolution benchmarking (should speedup warmup by a significant amount, does the below three options together)
# export MIOPEN_DEBUG_CONV_DIRECT=0
# Disable direct naive convolution benchmarking (naive_conv_ab_nonpacked_fwd_ndhwc_half_double_half.kd)
Expand Down
6 changes: 6 additions & 0 deletions scripts/scaffold-tuolumne.job
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,12 @@ ml cce/21.0.0 cray-mpich/9.1.0 rocm/7.1.1 rccl/fast-env-slows-mpi
# (2) Removing libmpi may cause segfault on mpi4py import
export LD_PRELOAD="/opt/rocm-7.1.1/llvm/lib/libomp.so /opt/cray/pe/mpich/9.1.0/ofi/gnu/11.2/lib/libmpi_gnu.so.12"

# Persist MIOpen's find database and compiled-kernel cache across job submissions.
# On Tuolumne, torchrun-hpc derives the MIOpen cache paths from TMPDIR.
export SCAFFOLD_MIOPEN_CACHE_ROOT="${SCAFFOLD_MIOPEN_CACHE_ROOT:-$(pwd)/.scaffold-miopen-cache}"
export TMPDIR="${SCAFFOLD_MIOPEN_TMPDIR:-${SCAFFOLD_MIOPEN_CACHE_ROOT}/tmp}"
mkdir -p "${TMPDIR}/MIOpen_user_db" "${TMPDIR}/MIOpen_custom_cache"

# Disable direct convolution benchmarking (should speedup warmup by a significant amount, does the below three options together)
# export MIOPEN_DEBUG_CONV_DIRECT=0
# Disable direct naive convolution benchmarking (naive_conv_ab_nonpacked_fwd_ndhwc_half_double_half.kd)
Expand Down
Loading