On Wed, Jul 22, 2020 at 09:33:47AM +0100, Richard Sandiford wrote:
> Jozef Lawrynowicz <joze...@mittosystems.com> writes:
> > is_int_mode does not allow MODE_PARTIAL_INT modes, so convert_modes was
> > not allowing a constant value to be converted to a MODE_PARTIAL_INT for
> > use as operand 2 in patterns such as ashlpsi3. The constant had
> > to be copied into a register before it could be used, but now can be
> > used directly as an operand without any copying.
>
> Yeah. I guess this dates back to when MODE_PARTIAL_INTs didn't have
> a known precision.
Is that what the section on MODE_PARTIAL_INT in the description for the
"subreg" RTX refers to? From "14.8 Registers and Memory" of gccint:
A MODE_PARTIAL_INT mode behaves as if it were as wide as the corresponding
MODE_INT mode, except that it has an unknown number of undefined bits.
If so, that whole section seems out of date. I can work on getting it
fixed up.
>
> > diff --git a/gcc/expr.c b/gcc/expr.c
> > index c7c3e9fd655..5a252f0935f 100644
> > --- a/gcc/expr.c
> > +++ b/gcc/expr.c
> > @@ -696,7 +696,7 @@ convert_modes (machine_mode mode, machine_mode oldmode,
> > rtx x, int unsignedp)
> > return x;
> >
> > if (CONST_SCALAR_INT_P (x)
> > - && is_int_mode (mode, &int_mode))
> > + && is_a <scalar_int_mode> (mode, &int_mode))
> > {
> > /* If the caller did not tell us the old mode, then there is not
> > much to do with respect to canonicalization. We have to
>
> I think we also need to change the condition in:
>
> /* If the caller did not tell us the old mode, then there is not
> much to do with respect to canonicalization. We have to
> assume that all the bits are significant. */
> if (GET_MODE_CLASS (oldmode) != MODE_INT)
>
> to is_a <scalar_int_mode> (old_mode)
>
> OK with that, thanks.
Thanks, I'll regtest that change and apply if all looks good.
Jozef
>
> Richard
