This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Fix WHOPR file name reuse
> On Fri, May 28, 2010 at 1:54 PM, Jan Hubicka <hubicka@ucw.cz> wrote:
> >> > Bootstrapped/regtested x86_64-linux, OK?
> >>
> >> I have a separate patch for the LTRANS naming issue,
> >> see attachment.
> >
> > Haha, so we got a conflict. ?I see you solved the name clash based on dumpbase.
> > I guess you could go ahead and commit it and I will prepare updated patch..
>
> I posted the patch and will wait for Diegos approval.
>
> > We solve bit of ortogonal problems here. ?I still believe we need stable ltrans
> > names for testsuite (i.e. based on the source filename rather than blah.0.o and
> > blah.1.o) and for incremental compilation, so I will handle this incrementally.
> > And I also need the struct ltrans_partition for "proper" partitioning and sorting
> > by unit size.
>
> The names are stable - they are all based on the final linker output
> name, so for gcc -o x t1.c t5.c y.c you get dump files named like
> x.ltrans0.142t.optimized and x.ltrans1.142t.optimized. The
> WPA outputs are x.ltrans0.o, the output from LTRANS stage
> is x.ltrans0.ltrans.o (heh - with -save-temps of course).
Yes, this is fine. For incremental WHOPR I need names to be stable when amount
of partitions change (i.e. you add new file into your project). That is why we
need to use original file names (or symbol names or checksums or whathever)
instead of sequence numbers for in this mode. But incremental whopr is not
implemented at all yet (I plan to start playing with it once WHOPR bootstrap is
working in hope to make it practical for development) and this is all doable
incremntally, of course. So your patch is defintly step in good direction ;)
Honza
>
> Richard.
>
> > Honza
> >