Skip to content

Manual device control

Jonathan Endres requested to merge better_devices into main

This merge changes the GPU / CPU selection from a global flag (util.use_gpu) to manual control of sequence, phantom etc. All relevant MRzero objects should have a .cpu() and .cuda() method as well as a .device property exactly how torch does it itself. Functions like simulation and reconstruction then compute on the same device as their inputs.

Merge request reports