Re: [PATCH] ppc: Fix 'calli' when floating-point arguments are passed

[Paul Cercueil]

Hi Paulo,

Le sam., sept. 10 2022 at 08:07:28 -0300, Paulo César Pereira de 
Andrade <paulo.cesar.pereira.de.andrade@gmail.com> a écrit :
> Em sáb., 10 de set. de 2022 às 03:53, Paul Cercueil
> <paul@crapouillou.net> escreveu:
>
>   Hi Paul,
>
> [snip]
>
> >  >   Please confirm the workaround works as expected.
> >
> >  Yes, everything works fine with the workaround in place.
>
> [snip]
>
>   Please try the last git head. I did a new patch that forces a second
> pass when the instructions are modified after the computation of live
> state at the start of blocks.
>   Also added a patch to use relative unconditional jumps to unknown
> addresses. Only the powerpc port was doing it in the expensive way,
> but it was kept that way until we understood the previous problem
> already corrected.

Looks like it works!
I will do extensive tests next weeks to see if I can find another 
regression.

>   The patch to do a second pass might be expensive in jit generation
> time for very large inputs with very complex branches, so, depending
> on feedback, it might be better to revert to the "workaround" state, 
> and
> not generate more clear code (with redundancy removal) but generate
> jit faster.

Yes, if it increases the generation time by a large margin it might not 
be worth it. Thanksfully most of my code generation consists in very 
simple blocks.

Cheers,
-Paul