This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: cause of all the mainline Sparc regressions


>    From: Jan Hubicka <jh@suse.cz>
>    Date: Thu, 23 May 2002 16:49:11 +0200
> 
>    > But if you delete unreachable blocks, shouldn't you optimize the
>    > CFG in the hopes that EH will end up doing less work?
>    
>    I think EH ends up with less work only when some receivers become
>    unreachable and this can happen only from unreachable/dead code removal,
>    I guess it does not happen by the optimization cleanup_cfg does except
>    for code removal. We can save some overhead then.
> 
> What if the beginning of EH receiver is a simplejump or a cross jump
> to some other place?

Don't know if this does matter.  I believe it will just produce EH
reciver code that will be later merged to other basic block.
> 
> Isn't this a similar reason why sibcall does a CFG cleanup too?  My
> thinking is that sibcall wants the cfg cleanup done so that the "after
> this call we jump next to epilogue or we tail recurse" condition is
> clearer.

It behaves differently when there is just one jump, I don't think EH
code behaves that way, but RIchard knows definitly more than I do.

Honza


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]