Re: Problems with builtin setjmp receiver getting eliminated - Help

[Andrew Hutchinson <andrewhutchinson at cox dot net>]

With David Daney suggestion I referred to MIPS port. And it looks like I 
figured a similar workaround in target.

builin-setjmp and friends , are seriously messed up. To make matters 
worse, target overrides are not well structured and various optimization 
stages rip out receiver code. My solution is thus very contrived - but 
no more than MIPS.

I  defined a copykat  nonlocal_goto_receiver. I did not need it but its 
presence is the only way to override  a key part of the  randomly 
optimized  builtin_setjmp_receiver code
As  receivers now contains no real instructions , it no longer matters 
if optimizers removes them  or not (MIPS uses unspecs to keep code - but 
I can't trust that to always work)

I then defined TARGET_BUILTIN_SETJMP_FRAME_VALUE - which allows 
expand_builtin_setjmp_setup to store the unadjusted frame_pointer, which 
can  be fully restored in expand_builtin_longjmp - rather than the receiver.

Tada!

So that got the testcase working!

I still have to check all call saved registers are saved by function 
doing setjmp - they appear not to be  at higher optimization levels. 
Either gcc is being smarter than me or there is still a problem.

Lastly,  TWO  receivers still being created - one is setjmp_receiver the 
other is non_local_receiver.  It's not my code, it happens on HEAD - I 
have not figured out why.


Andy




David Daney wrote:
> Andrew Hutchinson wrote:
>   
> > I have realised that  part of the problem is that the receiver block
> > has no incoming edges so cfgcleanup removes it as unreachable block -
> > right?
> >
> > So any target that need a non trivial receiver for builtin_setjmp will
> > not work? That would mean  any that have an offset between stack and
> > pointers?
> >
> > I guess the same problem exists for non-local goto?
> >
> > I am not convinced it could be this wrong. So please comment and
> > suggest solution - I'm sure I can write target handler but it seems so
> > wrong to leave this as  issue open.
> >
> >     
> MIPS uses an unspec_volatile in the nonloca_goto_receiver  That keeps it
> from being removed.
>
> David Daney
>
>