Skip to content

Changes to choose a launch configuration based on maximizing occupancy.

James Sutherland requested to merge sr-cuda-max-occupancy into master

Using cudaOccupancyMaxPotentialBlockSize we can calculate a configuration (grid size and block size) based on which the kernel is launched with a view of maximizing occupancy. cudaOccupancyMaxPotentialBlockSize returns a potential block size, which we then divide up into a 2D configuration.

Merge request reports