Repository navigation
[BUG]: cublasmp dependencies are not reflected in supported_nvidia_libs.py #1116
Description
Discovered by chance while testing on a workstation that did not have the CUDA driver installed:
libcublasmp.so.0is the only supported lib that requireslibcuda.so.1, which led to atest_load_nvidia_dynamic_lib.pyfailure when the driver was not installed.- To double-check I ran the
ldd, which made it obvious that we are missing all dependencies:cublas,cublasLt,nvshmem_host,nccl
The dependency on libcuda.so.1 is unusual: not sure if we want to do something about it. But the other dependencies should be added to supported_nvidia_libs.py.
mgx-c2g2-pvt-66.cl1u1.colossus.nvidia.com:/wrk/forked/cuda-python/cuda_pathfinder $ ldd /wrk/forked/cuda-python/cuda_pathfinder/.pixi/envs/cu12-linux-aarch64/lib/python3.14/site-packages/nvidia/cublasmp/cu12/lib/libcublasmp.so.0
linux-vdso.so.1 (0x0000fb3f18b6a000)
libcuda.so.1 => /lib/aarch64-linux-gnu/libcuda.so.1 (0x0000fb3f11a00000)
libcublas.so.12 => /wrk/forked/cuda-python/cuda_pathfinder/.pixi/envs/cu12-linux-aarch64/lib/python3.14/site-packages/nvidia/cublasmp/cu12/lib/../../../cublas/lib/libcublas.so.12 (0x0000fb3f0b600000)
libcublasLt.so.12 => /wrk/forked/cuda-python/cuda_pathfinder/.pixi/envs/cu12-linux-aarch64/lib/python3.14/site-packages/nvidia/cublasmp/cu12/lib/../../../cublas/lib/libcublasLt.so.12 (0x0000fb3edb800000)
libnvshmem_host.so.3 => /wrk/forked/cuda-python/cuda_pathfinder/.pixi/envs/cu12-linux-aarch64/lib/python3.14/site-packages/nvidia/cublasmp/cu12/lib/../../../nvshmem/lib/libnvshmem_host.so.3 (0x0000fb3ed2000000)
libnccl.so.2 => /wrk/forked/cuda-python/cuda_pathfinder/.pixi/envs/cu12-linux-aarch64/lib/python3.14/site-packages/nvidia/cublasmp/cu12/lib/../../../nccl/lib/libnccl.so.2 (0x0000fb3ebb800000)
librt.so.1 => /lib/aarch64-linux-gnu/librt.so.1 (0x0000fb3f18af0000)
libpthread.so.0 => /lib/aarch64-linux-gnu/libpthread.so.0 (0x0000fb3f18ac0000)
libdl.so.2 => /lib/aarch64-linux-gnu/libdl.so.2 (0x0000fb3f18a90000)
libstdc++.so.6 => /lib/aarch64-linux-gnu/libstdc++.so.6 (0x0000fb3ebb400000)
libm.so.6 => /lib/aarch64-linux-gnu/libm.so.6 (0x0000fb3f17750000)
libgcc_s.so.1 => /lib/aarch64-linux-gnu/libgcc_s.so.1 (0x0000fb3f18a50000)
libc.so.6 => /lib/aarch64-linux-gnu/libc.so.6 (0x0000fb3ebb240000)
/lib/ld-linux-aarch64.so.1 (0x0000fb3f18b2d000)
Activity
Add cublasmp DIRECT_DEPENDENCIES (closes NVIDIA#1116)
Add cuTENSOR support & bug fixes discovered while working on conda te…