Patch to to fix bad insn generation

Jeffrey A Law
Tue Nov 30 23:59:00 GMT 1999

  In message < >you write:
  > Sure thing.  Here is the patch.  Shall I apply it this time ?
It's a step forward, but it still can't be right.

In particular the change to PREDICATE_CODES is wrong.  general_operand
accepts a lot more than CONST_INT, MEM, REG.  For example, SUBREG, CONST,

PREDICATE_CODES must accept *any* kind of operand that the predicate itself
will accept.  Otherwise bad things happen.

The name valid_psi_mode_address is bogus too.  First we're not dealing with
just addresses here, it could be a REG or SUBREG.  psimode_truncation_operand
or something like that would be better I think since that describes more
correctly what that predicate is used to detect.

Once you fix those two issues you can consider the patch approved.


More information about the Gcc-patches mailing list