Re: [PATCH] Fix PR37053: Move tweaks of commutative precedence to target hook

[Paolo Bonzini <bonzini at gnu dot org>]

> Notice how we're doing a final bit of validation -- constrain_operands --
> here rather than in the validate routines.  We're assuming that if this
> extra validation fails, we can simply call validate_replace_rtx again
> to revert the change.  But this isn't always true, because the
> validation routines can "canonicalise" the resulting insn.
> It would be more robust (and cleaner) to treat this replacement
> in the same way as post-reload ones, and do the constrain_operands
> in recog.c:insn_invalid_p.

Note that it is already being done if reload_completed, just not if 
reload_in_progress.  No matter how pleasant, it would be a surprise if 
the attached patch worked without this change it includes:

-      if (! constrain_operands (1))
+      if (! constrain_operands (reload_completed))

(Note: I haven't tested it with the change, either).  On the other hand, 
with the change there is a subtle difference in reload_as_needed now 
(I'm not sure it matters).

Maxim, can you try it on m68k?

Paolo

Attachment: pr37053.patch
Description: Text document