[patch] Autoparallelization

Daniel Berlin dannyb@google.com
Thu Sep 13 21:10:00 GMT 2007


On Sep 13, 2007 3:01 PM, Diego Novillo <dnovillo@google.com> wrote:
> On 9/13/07, Zdenek Dvorak <rakdver@kam.mff.cuni.cz> wrote:
>
> > Well, I think that these messages are typically read by someone hunting
> > for a bug (frankly -- no user without a thorough knowledge of gcc
> > internals can interpret any of -fdump-tree-all-details dumps).
>
> Yes, in general this is true.  However, autoparallelization and
> vectorization have historically been special.  Power users look at the
> compiler dumps to find out whether the compiler could parallelize the
> loop they were hoping to parallelize and that kind of thing.
>
>
> > Actually, thinking about it again, we should probably rerun alias
> > analysis here anyway, as otherwise we might lose virtual operands
> > for non-gimple_reg variables that were not call clobbered, but are
> > accessed in the parallelized loop.  It is a bit annoying that this
> > will pessimize virtual operands throughout the whole function, though;
> > it would be nicer if we were able to specify which variables may be
> > accessed by a particular call.
>
> ISTR Dan talking about adding this feature.  Dan, am I misremembering?
>
No, you aren't.  But I put it off until LTO is ready.  It's really not
useful on small parts of the program.

Basically, you want to do context-sensitive call clobbering analysis.

> > I will send an updated patch as soon as possible,
>
> The changes are pre-approved, BTW.  If you want, it may be easier to
> commit this patch with the cosmetic changes and send the aliasing and
> --param changes separately.
>



More information about the Gcc-patches mailing list