jcranmer-intel wrote:

> On second thought, why does instcombine need to do this? Don't we already do 
> this late, in CodeGenPrepare or DAGBuilder? The exact min/max operation you 
> want will be target dependent, and this is introducing new constraints to the 
> operation

There is some value in canonicalizing on a particular form for FP min/max 
operations, but given that there are four slightly different definitions and 
that hardware ISAs vary as to which one(s) they implement, I'm not sure I could 
in good conscience advocate one to canonicalize on.

https://github.com/llvm/llvm-project/pull/159652
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to