This is the mail archive of the
mailing list for the GCC project.
Re: [RFA:] Fix PR target/18701, address adjustment error withparadoxical mem subregs. (Disallow them, I say!)
On Wed, 29 Dec 2004, Hans-Peter Nilsson wrote:
> My recollection is that
> besides the issue of having two identical move patterns for the
> same insn, one specifying *_extend, it would mean I have to
> allow for register operands for the *_extend case as well, and
> that doesn't exist on MMIX, so that seemed worse.
FWIW, I hope to revisit whether to have sign/zero_extend
patterns in the MMIX port for 4.1. Nah, the patterns wouldn't
be identical, for example the extending variants should not
allow register source or memory destination as early as in the
predicates. GCC should still be able to synthesize those
operations for register source by shifts or ands (and not, for
example, introduce stack pseudos). I hope I can count on help
getting middle-end performance and ICE issues resolved, i.e.
Still, whether those patterns are present is tangential. Let's
focus on our differences wrt. paradoxical mem subregs and/or