This is the mail archive of the gcc-patches@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: [tree-ssa] Avoid nondeterminism in tree-ssanames


> In message <1070315873.11985.106.camel@p4>, Andrew MacLeod writes:
>  >well yeah, at the beginning of a function I would expect to start at
>  >one, I meant within a function between passes.
>  >
>  >> We do recycle insn #s to cut down on our array/bitmap sizes.  It's flag
>  >> controlled so that you can turn it off for debugging purposes.
>  >
>  >
>  >Yeah, except sometimes by changing your environment like that, the bug
>  >goes away when you try to enable debugging information. aaaaaaaaaaagh! I
>  >hate those ones :-)
> Which is one of the reasons why you can turn off that behavior.  You can
> also turn it off to make dump files easier to diff.

I think that with SSA_NAMEs this can be different given that we do have
very few passes that process instructions in INSN_UID order, but we do
have passes that process SSA_NAMEs in SSA_NAME_VERSIOn order so the
optimization decisions can change.

Honza
> 
> Regardless, the version # reuse currently in the manager is going to have to
> go away.  If we're going to reuse version #s, we're going to need to do it
> differently.
> 
> jeff
> 


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