Re: [PATCH/RFC] Avoid invalid subregs during reload, call for testers

[Rask Ingemann Lambertsen <rask at sygehus dot dk>]

On Wed, Dec 06, 2006 at 07:59:05AM -0800, Ian Lance Taylor wrote:
[first part of reload1.c patch]
> This part of the patch is OK.

   I'll retest with just this part of the patch.

> I'm not convinced that the assert in subeg_regno_offset is correct.
> As Joseph has pointed out, this is going to have knock-on effects on
> the rest of the compiler.  The call to subreg_regno_offset is not
> necessarily incorrect, even if using it would be incorrect.
 
> In particular, without the assert, I don't think the rest of your
> patch to reload1.c is needed.  The inherited reload should be
> prevented by the call to HARD_REGNO_MODE_OK.

   And also by checking that the register has the right contents
(reg_reloaded_valid and reg_reloaded_contents) -- I didn't intend to imply
that wrong code could be generated. Part of the reason I included the second
change to reload1.c in the patch is that I test changes to
subreg_regno_offset() with a wrapper comparing return values from the old
and the new version:

unsigned int
subreg_regno_offset (unsigned int xregno, enum machine_mode xmode,
                     unsigned int offset, enum machine_mode ymode)
{
  unsigned int old, new;
  old = old_subreg_regno_offset (xregno, xmode, offset, ymode);
  new = new_subreg_regno_offset (xregno, xmode, offset, ymode);
  gcc_assert (old == new);
  return (new);
}

   With invalid parameters, that difficult to achieve. I'll change that
assert to

  gcc_assert (old == new
	      || !subreg_offset_representable_p (xregno, xmode, offset, ymode);

to avoid false positives.

-- 
Rask Ingemann Lambertsen