Flow, Zig build tools, research utilities, and faster-diffbloch: drop-in Metal GPU and optimized CPU acceleration for diffBloch.
Measured at up to 1.80× faster than PyTorch for forward plus backward Bloch-wave electron-diffraction simulation on an Apple M4 Max. It is 1.41×-1.80× faster across the measured 31-579 beam cases.
Install faster-diffbloch from piwheels · Read the source repository