On Fri, Feb 13, 2026 at 04:48:15PM +0100, Robin Dapp wrote:
> > I might be completely off here, but a quick idea which came up is
> > whether a secondary reload could be utilized in order to get a scratch
> > register, rendering the initial mov into an ordinary mov with two
> > operands so that LRA would again detect a cycle.
> 
> Yes, Peter tried a secondary reload and it got rid of the ICE.  Just the 
> resulting code was worse.  If nothing else helps, we could still go that
> route.

Ah sorry I should have read the PR before.  Though, I was rather hoping
for a secondary reload instead of a secondary memory as suggested by
Peter, in order to keep things in registers for as long as possible by
basically emitting an insn similar to the current one.  However, in the
meantime I'm not convinced anymore that this would work out.

So ignoring this idea and going back, I still think that the initial
patch is reasonable.

Cheers,
Stefan

Reply via email to