Port Cannons to gpu
Initial port of Cannon's algorithm to GPU by offloading all GEMMs to GPU.
The old module-image for the shared runners will be discontinued on October 31. All users still referencing gitlab-registry.mpcdf.mpg.de/mpcdf/module-image in their CI pipelines need to switch to the new CI images now, see instructions here.
Initial port of Cannon's algorithm to GPU by offloading all GEMMs to GPU.