remaining libjava/verify_local_live_at_start failures

[Jan Hubicka]

> 
> mmix-knuth-mmixware:
> g77.f-torture/compile/970915-0.f: In subroutine `tstsig11':
> g77.f-torture/compile/970915-0.f:19: Internal compiler error in verify_local_live_at_start, at flow.c:607
>  with -Os

This looks like latent bug in dead code removal.  There is register used only by dead store.
Dead store is removed, but register is not marged as dead.  I am not quite sure why...

Honza
> 
> g77.f-torture/execute/f90-intrinsic-bit.f: In program `MAIN__':
> g77.f-torture/execute/f90-intrinsic-bit.f:191: Internal compiler error in fixup_memory_subreg, at function.c:2516
>  with -O1, -O2, -O3 -fomit-frame-pointer, -O3
> -fomit-frame-pointer -funroll-loops, -O3 -fomit-frame-pointer
> -funroll-all-loops -finline-functions, -O3 -g, -Os.
> No, it doesn't look like this bug is "yours", I just mention it
> for completeness since this test fails with cris-axis-elf too
> in "your" code area.
> 
> I haven't checked other target's test-results, but I see the
> i960-coff which built before dies, correlated in time with your
> changes.  (Objc doesn't build;  configured with
> --enable-languages=c++,f77,java.)  It dies with:
> In file included from
> /home/hp/cvs_areas/combined/cvs_write/gcc/unwind-sjlj.c:278:
> /home/hp/cvs_areas/combined/cvs_write/gcc/unwind.inc: In
> function `_Unwind_RaiseException_Phase2':
> /home/hp/cvs_areas/combined/cvs_write/gcc/unwind.inc:67:
> warning: implicit declaration of function `abort'
> In file included from
> /home/hp/cvs_areas/combined/cvs_write/gcc/unwind-sjlj.c:278:
> /home/hp/cvs_areas/combined/cvs_write/gcc/unwind.inc: In
> function `_Unwind_SjLj_RaiseException':
> /home/hp/cvs_areas/combined/cvs_write/gcc/unwind.inc:129: Wrong
> amount of branch edges after unconditional jump 27
> /home/hp/cvs_areas/combined/cvs_write/gcc/unwind.inc:129:
> verify_flow_info failed
> 
> Thanks.
> 
> brgds, H-P