Re: [PATCH 27/46] tcg/optimize: Use fold_masks_zs in fold_qemu_ld

[Richard Henderson]

On 12/17/24 14:35, Pierrick Bouvier wrote:
> > @@ -3001,11 +3010,18 @@ void tcg_optimize(TCGContext *s)
> >               break;
> >           case INDEX_op_qemu_ld_a32_i32:
> >           case INDEX_op_qemu_ld_a64_i32:
> > +            done = fold_qemu_ld_1reg(&ctx, op);
> > +            break;
> >           case INDEX_op_qemu_ld_a32_i64:
> >           case INDEX_op_qemu_ld_a64_i64:
> > +            if (TCG_TARGET_REG_BITS == 64) {
> > +                done = fold_qemu_ld_1reg(&ctx, op);
> > +                break;
> > +            }
> > +            QEMU_FALLTHROUGH;
> >           case INDEX_op_qemu_ld_a32_i128:
> >           case INDEX_op_qemu_ld_a64_i128:
> > -            done = fold_qemu_ld(&ctx, op);
> > +            done = fold_qemu_ld_2reg(&ctx, op);
> >               break;
> >           case INDEX_op_qemu_st8_a32_i32:
> >           case INDEX_op_qemu_st8_a64_i32:
>
> Couldn't we handle this case in fold_masks instead (at least the 64 bits store on 32 bits 
> guest case)?

No, not with the assertion that the TCGOp passed to fold_masks have a single 
output.


r~