Skip to content

build(container): install gcc-12 on aarch64 for torch inductor's armv9-a flags #12507

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

Draft
yifjiang wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

build(container): install gcc-12 on aarch64 for torch inductor's armv9-a flags #12507

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
24 changes: 24 additions & 0 deletions container/templates/vllm_runtime.Dockerfile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -265,6 +265,30 @@ RUN set -eux; \
fi; \
rm -rf /var/lib/apt/lists/*

# aarch64/Grace only: the upstream vllm/vllm-openai arm64 image is Ubuntu 22.04
# and ships gcc/g++ 11.4, but torch >= 2.13 detects Grace as armv9-a and its
# inductor CPU backend codegens `-march=armv9-a+sve2+fp16fml+sha3+bf16+i8mm`
# (torch/_inductor/cpu_vec_isa.py: VecSVE.build_arch_flags). GCC only learned
# `armv9-a` in 12, so gcc-11 rejects torch's OWN generated flags:
# cc1plus: error: unknown value 'armv9-a' for '-march'
# That breaks the smoke guard below AND every runtime inductor/Triton CPU JIT on
# a Grace host. gcc-12 is in the jammy archive (gcc-12-base is already pulled in
# as a dependency), so switch the default toolchain to it. amd64 is untouched.
# `cc`/`c++` are their own update-alternatives masters here, so each link is
# registered separately rather than as a slave of gcc.
RUN set -eux; \
if [ "$(uname -m)" = "aarch64" ]; then \
apt-get update; \
DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends gcc-12 g++-12; \
for pair in gcc:gcc-12 g++:g++-12 cc:gcc-12 c++:g++-12; do \
name="${pair%%:*}"; impl="${pair##*:}"; \
update-alternatives --install "/usr/bin/$name" "$name" "/usr/bin/$impl" 120; \
update-alternatives --set "$name" "/usr/bin/$impl"; \
done; \
rm -rf /var/lib/apt/lists/*; \
echo 'int main(){return 0;}' | g++ -march=armv9-a+sve2+fp16fml+sha3+bf16+i8mm -msve-vector-bits=128 -x c++ - -c -o /dev/null; \
fi

# Regression guard for the codec purge above: torch.inductor/Triton JIT shell
# out to a host C/C++ compiler at runtime, so a missing toolchain only surfaces
# on the first compile in production. Reproduce that compile path at build time
Expand Down
Loading