[llvm] [X86] Recognize POP/ADD/SUB modifying rsp in getSPAdjust. (PR #114265)

[Daniel Zabawa via llvm-commits]

daniel-zabawa wrote:

> > > @daniel-zabawa any tests?
> > 
> > 
> > I am open to suggestions on how to cover this. The potential bug in prolog/epilog is contingent on having a frame object referenced in a call sequence, following a previous call sequence which requires the frame pointer to be spilled.
> > I don't know any reliable way to create the code exposing the issue without MIR, but MIR also does not support serializing the attributes (FPClobberedByCall) required for the FP spill/restore to be generated.
> 
> Can we add this attribute to `llvm/lib/Target/X86/X86MachineFunctionInfo.h` similarly we have it for `amxProgModel` and check with mir tests? It consumes `X86MachineFunctionInfo` that has `getFPClobberedByCall`.

Thanks - I did not notice that it would be simple to add that into serialization. The test itself is still taking some time as it has to satisfy rather particular conditions for the FP to be spilled around a call and for that adjustment to be counted in prologepilog.

https://github.com/llvm/llvm-project/pull/114265