flow patch to handle clobbers of live registers in PARALLELs

Joern Rennecke amylaar@cygnus.co.uk
Tue Nov 30 23:59:00 GMT 1999

> What has that got to do with anything?  flow2 runs completely
> after reload.

It was flow1 that deleted the insn with the clobber.

reload was happy because it used the global_live_at_start regset, where
the return address register was still marked as dead.
flow2 aborted because when it recalculated global_live_at_start from
global_live_at_end, the result didn't match.

More information about the Gcc-patches mailing list