Skip to content

CUDA: Don't pass dynamic libraries to device linker

Fixes #18504 (closed)

Previously items that are paths to library files that aren't static would be passed using -Xnvlink to bypass nvcc's front-end. The nvlink tool knows to ignore shared library files. The problem as raised in issue #18504 (closed) is that nvlink mistakenly treats versioned dynamic libraries as static libraries as they don't end in .so but instead .so.1

Merge request reports