llvm dev - Oct 2020 - MachineCSE ignores COPY

Hello,

I see that MachienCSE ignores COPY instructions
bool MachineCSE::isCSECandidate(MachineInstr *MI) {
...
  // Ignore copies.
  if (MI->isCopyLike())
    return false;

Are there any correctness concerns?

Thanks,
Vivek
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20201019/a68d6d0f/attachment.html>

Hi Vivek,

My guess is because copies are transient, so we expect them to disappear anyway.

Maybe it has to do as well with the fact that chances are the are dealing with
physical reg copies, e.g., used for ABI.

Cheers,
-Quentin
> On Oct 18, 2020, at 11:07 PM, vivek pandya via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> Hello,
> 
> I see that MachienCSE ignores COPY instructions 
> bool MachineCSE::isCSECandidate(MachineInstr *MI) {
> ...
>   // Ignore copies.
>   if (MI->isCopyLike())
>     return false;
> 
> Are there any correctness concerns?
> 
> Thanks,
> Vivek
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

On Mon, Oct 19, 2020 at 10:53 PM Quentin Colombet <qcolombet at apple.com>
wrote:
> Hi Vivek,
>
> My guess is because copies are transient, so we expect them to disappear
> anyway.
>
That may prevent CSE in certain cases.
>
> Maybe it has to do as well with the fact that chances are the are dealing
> with physical reg copies, e.g., used for ABI.
>
If an instruction has any physical register shouldn't that will be handled
in hasLivePhysRegDefUses() and such instruction will not be CSE.

-Vivek
>
> Cheers,
> -Quentin
>
> > On Oct 18, 2020, at 11:07 PM, vivek pandya via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
> >
> > Hello,
> >
> > I see that MachienCSE ignores COPY instructions
> > bool MachineCSE::isCSECandidate(MachineInstr *MI) {
> > ...
> >   // Ignore copies.
> >   if (MI->isCopyLike())
> >     return false;
> >
> > Are there any correctness concerns?
> >
> > Thanks,
> > Vivek
> > _______________________________________________
> > LLVM Developers mailing list
> > llvm-dev at lists.llvm.org
> > https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20201020/1c9b542a/attachment.html>