[llvm-dev] globalisel: cross-bank constant propagation?

[Matt Arsenault via llvm-dev]

> On Mar 29, 2021, at 06:14, Jay Foad <jay.foad at gmail.com> wrote:
> 
> To be honest I am confused about the role of RegBankSelect when you
> have instruction selection patterns that match more than one GMIR
> node. How can RegBankSelect possibly tell you what alternative
> mappings (i.e. instructions) are available, without predicting what
> patterns will be matched by instruction selection?

There is a table of costs for alternative bank mappings. For the AMDGPU case, I don’t think this cost table makes much sense. The condition is mostly divergence based. The exception would be for some VGPR vs. AGPR cases. 

-Matt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20210329/f142e461/attachment.html>