Module shuffle

Modules§

offset: Offset one scalar per thread. Out-of-block destinations preserve output.
offset_access
rotate: Circular offset one scalar per thread.
rotate_access
shift: Shift a blocked register tile one item. Boundary output is preserved. Returns the block’s discarded prefix (down) or suffix (up).
shift_access: Shift records in blocked order. The unpaired output item is unchanged; return the discarded prefix (down) or suffix (up) to every thread.