[Bug target/114978] [14/15 regression] 548.exchange2_r 14%-28% regressions on Loongarch64 after gcc 14 snapshot 20240317

[chz0808 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114978

--- Comment #42 from Chen Chen <chz0808 at gmail dot com> ---
(In reply to Xi Ruoyao from comment #39)
> (In reply to Chen Chen from comment #38)
> > (In reply to Xi Ruoyao from comment #37)
> > > So if we revert r15-7525 now, would things work normally with just r15-6657?
> > > If so I'd suggest to revert r15-7525 (now or when GCC 16 stage 1 starts) and
> > > close this as a PR114932 dup.
> > 
> > Why revert r15-7525? In its description it never mentions 548.exchange2_r.
> 
> See comment 19.  r15-7525 was intended for work this around.
> 
> > And it is beneficial to have an option to adjust addr-reg-reg cost. Anyway
> > one can set "-maddr-reg-reg-cost=1" if exaggerating the cost is not needed
> > or prefered.
> 
> Because on LA464 and LA664 the cost of LDX/STX isn't really so high.
> 
> Or we should keep the option but change the default of -maddr-reg-reg-cost
> back to 1.

I think it is better to keep the option in case we need it. I do not mind
setting the default of -maddr-reg-reg-cost to 1 or 3.