https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121882
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|[13/14/15/16 regression] |[13 regression] ICE when
|ICE when building GCC 13 |building GCC 13 with
|with checking and |checking and -mtune=znver5
|-mtune=znver5 |
CC| |jakub at gcc dot gnu.org
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #4)
> (In reply to Andrew Pinski from comment #3)
> > r13-9284-g7392e9e480afe3143e72a99f7b5ac99b2f49c284
> >
> > Yes there is no reg check before checking general registers.
>
> + rtx set1 = PATTERN (mov);
> + if (GET_CODE (set1) != SET
> + || !GENERAL_REG_P (SET_SRC (set1))
> + || !GENERAL_REG_P (SET_DEST (set1)))
> + return false;
>
> There should be 2 checks for reg_p there
??
#define GENERAL_REG_P(X) (REG_P (X) && GENERAL_REGNO_P (REGNO (X)))
>From what I can see, Honza has committed
r15-3420-ge2125a600552bc6e0329e3f1224eea14804db8d3
and very soon after that a fix for a bug introduced in there:
r15-3425-g910e1769a0653ac32bd8c1d6aabb39c797d5d773
To GCC 14 branch then both were backported:
r14-10719-g2c01292411044adbd67f79355c1e24decd2fd3c0
r14-10720-g2eade72b0e2ac9dd18ef517bc3b868157f1ddf48
But to GCC 13 branch only
r13-9284-g7392e9e480afe3143e72a99f7b5ac99b2f49c284
So I think we should just cherry-pick r14-10720 to 13 branch.
