This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git
The following commit(s) were added to refs/heads/main by this push:
new 96461a3e18 [CI][CMAKE] Install CUDA driver stub for
libtvm_runtime_cuda (#19886)
96461a3e18 is described below
commit 96461a3e18b79eaa75af8db3b53da4a2ac8086ad
Author: Ruihang Lai <[email protected]>
AuthorDate: Thu Jun 25 07:16:57 2026 -0400
[CI][CMAKE] Install CUDA driver stub for libtvm_runtime_cuda (#19886)
The quay.io/manylinux_cuda image ships the CUDA toolkit but omits the
libcuda driver stub (cuda-driver-devel). Without it CMake's FindCUDA
leaves CUDA_CUDA_LIBRARY empty, and since a shared library links fine
with undefined symbols, libtvm_runtime_cuda built with no libcuda.so.1
dependency -- ldd/readelf did not list it -- breaking downstream use.
Install cuda-driver-devel-13-1 from the CUDA repo already configured in
the image (a ~45 KB package with no dependencies, not the full toolkit
download). The stub lands under /usr/local/cuda/.../stubs where FindCUDA
looks, so the sidecar links libcuda.so.1 again.
Also fail CMake configure with a clear error when CUDA_CUDA_LIBRARY is
unset, so a missing driver stub can never again silently ship a runtime
sidecar without its libcuda dependency.
---
ci/scripts/package/manylinux_build_libtvm_runtime_cuda.sh | 10 +++++++---
cmake/modules/CUDA.cmake | 7 +++++++
2 files changed, 14 insertions(+), 3 deletions(-)
diff --git a/ci/scripts/package/manylinux_build_libtvm_runtime_cuda.sh
b/ci/scripts/package/manylinux_build_libtvm_runtime_cuda.sh
index ea9e563046..d0d847ae83 100755
--- a/ci/scripts/package/manylinux_build_libtvm_runtime_cuda.sh
+++ b/ci/scripts/package/manylinux_build_libtvm_runtime_cuda.sh
@@ -18,9 +18,9 @@
#
# Build libtvm_runtime_cuda.so inside a manylinux CUDA container, run by the
# build_cuda_runtime CI job. The official quay.io/manylinux_cuda images ship
-# the CUDA toolkit preinstalled under /usr/local/cuda, so no toolkit install
-# is needed here. Builds the sidecar into build-wheel-cuda/lib/ for the wheel
-# build to bundle.
+# the CUDA toolkit preinstalled under /usr/local/cuda, but omit the libcuda
+# driver stub, so we install just cuda-driver-devel from the image's CUDA repo.
+# Builds the sidecar into build-wheel-cuda/lib/ for the wheel build to bundle.
#
# Usage: manylinux_build_libtvm_runtime_cuda.sh
set -euxo pipefail
@@ -29,6 +29,10 @@ repo_root="$(cd "$(dirname "${BASH_SOURCE[0]}")/../../.." &&
pwd)"
build_dir="${repo_root}/build-wheel-cuda"
parallel="$(getconf _NPROCESSORS_ONLN 2>/dev/null || echo 4)"
+# The image ships the toolkit but not the libcuda driver stub; install it from
+# the image's CUDA repo so the sidecar links libcuda.so.1 (sync version with
tag).
+dnf -y install cuda-driver-devel-13-1
+
# Build the CUDA runtime sidecar with CUDA on and LLVM off, so it does not need
# the LLVM prefix; the main CPU wheel links LLVM statically. The manylinux CUDA
# image already ships cmake and make, and the build uses the default Makefiles
diff --git a/cmake/modules/CUDA.cmake b/cmake/modules/CUDA.cmake
index 7a383aa0de..e1731e8078 100644
--- a/cmake/modules/CUDA.cmake
+++ b/cmake/modules/CUDA.cmake
@@ -62,6 +62,13 @@ endif(USE_CUDA)
if(USE_CUDA)
message(STATUS "Build cuda device runtime")
+ # tvm_runtime_cuda links libcuda; without it the .so silently drops its
+ # libcuda.so.1 dependency, so fail configure instead.
+ if(NOT CUDA_CUDA_LIBRARY)
+ message(FATAL_ERROR "USE_CUDA is on but libcuda was not found. "
+ "Set -DCUDA_CUDA_LIBRARY=<path to libcuda.so> or make the driver stub
discoverable.")
+ endif()
+
tvm_file_glob(GLOB RUNTIME_CUDA_SRCS src/backend/cuda/runtime/*.cc)
tvm_file_glob(GLOB VM_CUDA_BUILTIN_SRC_CC src/runtime/vm/cuda/*.cc)
