opencv_contrib
opencv_contrib copied to clipboard
[bug][CUDA]multi-gpu errors when using nppsMalloc() functions
I report here a bug that is not an OpenCV bug
It is a bug that I have identified recently in the CUDA run-time (https://forums.developer.nvidia.com/t/bug-current-device-change-after-nppsmalloc-does-nppsmalloc-call-cudasetdevice/324156)
The bug occurs for CUDA runtime > 12.6 Update 3 and so far up to 12.8 (latest at the current day), and involves wrong multi-GPU support when using nppsMalloc() functions
I report it here in the sole purpose of helping OpenCV maintainers not wasting time investigating if related bugs are reported to OpenCV.