On Fri, Mar 1, 2024 at 4:07 AM Robin Dapp <rdapp....@gmail.com> wrote: > > Hi Han, > > in addition to what Juzhe mentioned (and that late-combine is going > to handle such cases) it should be noted that register pressure > should not be the only consideration here. Many uarchs have a higher > latency for register-file-crossing moves. At least without spilling > the vv variant is preferable, with spilling it very much depends.
And of course there are uarches for which this is not the case (e.g. post-commit decoupled vector unit), in which case the .vx and .vf versions are preferable to the .vv form regardless of vector register pressure, because they reduce vector regfile access energy (especially if a splat can be avoided). So it's a job for -mtune. > > > Regards > Robin >
