Your Jan 15 change broke the x86

Jeffrey A Law law@cygnus.com
Sat Jan 30 14:50:00 GMT 1999


  In message < 19990130142632.A27028@cygnus.com >you write:
  > On Sat, Jan 30, 1999 at 11:08:50AM -0700, Jeffrey A Law wrote:
  > Got it.  The immediate problem was that we weren't finding the matching
  > constraint on "%0" and so failing to OK anything.  The underlying problem
  > is to aggressive checking with too little information -- if a matching
  > constraint is buried in alternatives ("r,0") we have scarce hope of
  > doing the right thing before reload.
  > 
  > The attached patch modifies asm_operand_ok to recognize when results are
  > inconclusive.  Its two uses then take the conservative approach. 
  > 
  > In addition, "%" handling is improved so that we don't have to throw up
  > our hands quite so often.  I don't really check swapping of operands,
  > but it makes sure at least one of the two operands will be close.
  > 
  > > It also appears that they're causing the PA to blow up with a similar
  > > problem, except in divdi3, which also uses asms...
  > 
  > Would you check to see that this solves the PA problem as well?
It doesn't fix the PA problem.  Give me about 5 minutes to get a testcase
ready.

jeff



More information about the Gcc-patches mailing list