You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Add CUDA architecture to CMake option when building for NVidia devices #2029
Currently, DPCTL_TARGET_CUDA cmake option is binary, and doesn't allow the user to set a CUDA architecture.
This could become problematic in the future and/or for extensions which the compiler may generate code which is unusable on some architectures.
The solution is to use DPCTL_TARGET_CUDA option to allow the user to set an architecture, and if one isn't sett, to fall back to the default (sm_50 is the current default per oneAPI extension)
changed the title [-]Add CUDA architecture parameter when building for NVidia devices[/-][+]Add CUDA architecture to CMake option when building for NVidia devices[/+]on Mar 24, 2025
Activity
[-]Add CUDA architecture parameter when building for NVidia devices[/-][+]Add CUDA architecture to CMake option when building for NVidia devices[/+]