nickdesaulniers wrote:

Sorry, my code review queue is totally overloaded at the moment.

I'm still hoping @nikic or @arsenm can help answer my question in 
https://github.com/llvm/llvm-project/pull/197798#discussion_r3375604758.

I'm still curious if we can keep support for this feature isolated to fast reg 
alloc?  Like this PR asserts that fast regalloc is deficient and thus all this 
is necessary to work around fast regalloc.  But that _still_ feels like a false 
premise to me.  fast regalloc is infalliable when it discovers its exhausted 
registers, but _I think_ we can still "pre-filter" spilling "rm" as "m".  
Surely we can do something better than 
https://github.com/llvm/llvm-project/pull/74344 ? @MatzeB 

https://github.com/llvm/llvm-project/pull/197798
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to