Skip to content

fixing fbgemm on Milan #845

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
ahmadelyoussef wants to merge 1 commit into
base: v2-beta
Choose a base branch
from
Open

fixing fbgemm on Milan #845

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
27 changes: 27 additions & 0 deletions packages/ai_wdl/fbgemm/install_fbgemm_bench.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -911,6 +911,31 @@ install_pytorch() {
echo "[SETUP] PyTorch installation complete."
}

# Apply the upstream PR 5846 fix so AVX variants are selected by compiler
# support rather than by executing them on the build host. Runtime checks omit
# AVX-512 objects on hosts such as T1_MLN and leave fbgemm.so with unresolved
# symbols required by fbgemm_gpu_tbe_inference.so.
apply_avx_compile_check_fix() {
local find_avx="fbgemm_${FBGEMM_VERSION}/cmake/modules/FindAVX.cmake"

if ! grep -Fqx 'INCLUDE(CheckSourceRuns)' "${find_avx}" || \
! grep -Fq 'CHECK_SOURCE_RUNS' "${find_avx}"; then
echo "[ERROR] FindAVX.cmake does not contain the expected PR 5846 preimage." >&2
return 1
fi

sed -i \
-e 's/INCLUDE(CheckSourceRuns)/INCLUDE(CheckSourceCompiles)/' \
-e 's/CHECK_SOURCE_RUNS(/CHECK_SOURCE_COMPILES(/' \
"${find_avx}"

if ! grep -Fqx 'INCLUDE(CheckSourceCompiles)' "${find_avx}" || \
! grep -Fq 'CHECK_SOURCE_COMPILES' "${find_avx}"; then
echo "[ERROR] Failed to apply the PR 5846 FindAVX.cmake substitutions." >&2
return 1
fi
}

# Function to clone the FBGEMM repository
# This downloads the FBGEMM source code from GitHub
clone_fbgemm_repo() {
Expand All @@ -929,6 +954,8 @@ clone_fbgemm_repo() {
# Cherry-pick the latest commit from the FBGEMM main branch to fix issue https://github.com/pytorch/FBGEMM/pull/5037
git -C fbgemm_${FBGEMM_VERSION} cherry-pick 9df97a7090c2c5edecea4fd08bad11ab8a23284c

apply_avx_compile_check_fix || return 1

# Disable the postbuild script to prevent race conditions during linking
# This is a workaround for a known issue in the build process
echo "[SETUP] Disabling postbuild script..."
Expand Down