This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH 00/65] Merge from the Graphite branch to trunk
On Mon, Aug 2, 2010 at 21:55, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> On Mon, Aug 02, 2010 at 10:52:16PM -0400, Jack Howarth wrote:
>>
>> nolto = -ffast-math -funroll-loops -O3 -fno-protect-parens
>> nolto+graphite = -ffast-math -funroll-loops -O3 -fno-protect-parens -fgraphite-identity
>> lto = -ffast-math -funroll-loops -O3 -flto -fwhole-program -fno-protect-parens
>> lto+graphite = -ffast-math -funroll-loops -O3 -flto -fwhole-program -fno-protect-parens -fgraphite-identity
>>
>
> Opps. Minor correction. The actual runs where -fwhopr rather than -flto.
>
Thanks for running these measurements.
It looks like the run time regressions that we saw with -O3 and
graphite identity are almost gone, except for induct.
I have not merged the last patches improving the graphite code
generation with more signed types (that would enable some more loops
to be vectorized). These patches are still miscompiling some SPEC
benchmarks. I will work a bit more on that front and submit them
separately. Hopefully they will stabilize faster.
Sebastian