Re: [PATCH, rs6000]: mark clobber for registers changed by untpyed_call

[Segher Boessenkool <segher@kernel.crashing.org>]

On Fri, Feb 14, 2020 at 02:58:49PM +0800, Jiufu Guo wrote:
> Segher Boessenkool <segher@kernel.crashing.org> writes:
> 
> > On Sat, Feb 08, 2020 at 10:17:42AM -0600, Segher Boessenkool wrote:
> >> And we do not know which of the register will be used for the return, in
> >> untyped_call (only untyped-return knows).  But we can add clobbers of all
> >> registers that *might* be used for the return, we do know that here, see
> >> operands[2] of untyped_call.
> >
> > Clobbers in parallel to the call, I mean, not as separate insns later.
> >
> Thanks Segher, as discussed, we may refine the patch by adding a
> 'barrier' to avoid instructions mis-scheduled.  This improvement patch
> maybe fine for practice.
> 
> Below is the updated patch: bootstrap and regtest pass on powerpc64le.
> Is this ok to submit to trunk?

This is okay for trunk, and for backports after a week or so.  Thanks!

(It doesn't fully solve the problem, as we discussed; because the return
registers are not seen (by df, etc.) as live immediately after the call,
new instructions using those registers can be created between the call
and the later clobber.  The barrier does nothing to prevent that.)

Don't forget to do a changelog entry, mentioning the PR as well :-)


Segher