[PATCH] D36795: [SystemZ] Increase number of LOCRs emitted by passing regalloc hints

[Quentin Colombet via Phabricator via llvm-commits]

qcolombet accepted this revision.
qcolombet added a comment.
This revision is now accepted and ready to land.

> Do you think that looking at the number of spilled live ranges is a good way to consider the trade-off for using hard hints, or do you perhaps have anything to add?

To me, it sounds like we should tell RA that the non-hard-hints will be expensive to expand. Right now, copies are expected to be cheap regardless of what, where and how they are done and this already does not play very nice with shrink-wrapping for instance. I believe this is yet another instance of that problem.

Short term the hard hints sound fine. Long term, I don't know, I haven't thought about it.

> It would have been nice to *guarantee* that the pseudo will get either high or low parts, like gcc will let you specify combinations of legal register operands. If that could be done, SystemZ could even remove its custom pass that handles any rare cases of mixed operands. But I suppose this would not be easy to do, or?

I didn't get what you would like to do.


https://reviews.llvm.org/D36795