Hi All!
On 13 Mrz., 10:00, Jeroen Demeyer <jdeme...@cage.ugent.be> wrote:
> Try CFLAGS="-O3 -march=native" CXXFLAGS="-O3 -march=native" and see if
> that works.
Meanwhile I managed to build the (slightly modified) gcc with
graphite, based on
* an extension of the standard ppl spkg ("build the ppl C interface",
#12672, needs review)
* an upgrade of the optional gmp spkg (the current version is broken
for me, but 5.0.4 works out of the box. See #12661, needs review.
* a cloog-ppl spkg (note that cloog itself won't suffice). See
#12666, needs review.
Then, building all of Sage (all spkgs) with
CFLAGS="-O3 -march=native -floop-interchange -floop-strip-mine -
floop-block -fno-strict-aliasing"
CXXFLAGS="-O3 -march=native -floop-interchange -floop-strip-mine -
floop-block -fno-strict-aliasing"
went smoothly (of course, some packages override the flags).
And it even seems that the doc tests pass.
And now the bad news: Using the above compiler flags, Sage became
MASSIVELY slower, by factors of 2 or 3. That's why I interrupted sage -
testall after about 54% (there was no error up to that point).
Do you have any insight why optimization has such a big negative
impact?
Even though the optimization thingy seems to be a failure, I still
think that we should upgrade the optional gmp spkg (it is broken, at
least on my machine), and we should have an optional CLooG-PPL package
(which would then require to make our standard PPL spkg build the C
interface, rather than just the C++ interface).
Cheers,
Simon
--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org
