flow patch to handle clobbers of live registers in PARALLELs

Joern Rennecke amylaar@cygnus.co.uk
Tue Nov 30 09:12:00 GMT 1999


> On Fri, Nov 26, 1999 at 03:07:32AM +0000, Joern Rennecke wrote:
> > The final call propagate_block found out that the insn that clobbered the
> > return value was dead, and thus removed it.  However, propagate_block
> > does not update the live_at_start information - it is supposed to
> > be stable in the last call.
> 
> Yes it's supposed to be stable.  It's supposed to have taken the
> fact that it's going to be deleting the insn into consideration
> when computing live_at_start.

It doesn't, and it must not.  The return address register would be live
throughout the function, and it is needed for spills.


More information about the Gcc-patches mailing list