Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-16 09:45, Simon King wrote: > The question is: Is a missing fortran compiler a good reason to build > all of GCC? In fact, yes. Because building "just" a Fortran compiler implies that most of GCC must be built anyway. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Leif,
On 16 Mrz., 08:51, leif wrote:
> Really? Compiling GCC 4.6.3 ('make -j4') with C, C++, Fortran, Java
> and LTO enabled took exactly 3 hours on my dual-core Netbook(!) (AMD
> Fusion E-450), plus <4 minutes 'make install'. (Bootstrapped with GCC
> 4.4.3, '-march=native -O3 -g'.)
I have just 'make', without '-j', Intel(R) Core(TM)2 Duo CPU
P7450 @ 2.13GHz, bootstrapped with gcc 4.6.2, '-march=native -O3'.
It finished just a minute ago!
real509m19.210s
user420m16.548s
sys 31m48.017s
Successfully installed gcc-4.6.3
When I tried to build it the very first time (bootstrapped with gcc
4.6.2, WITHOUT graphite), it was
real117m26.849s
user105m11.818s
sys 9m39.734s
When I used the resulting gcc 4.6.3 (without graphite) to bootstrap
another gcc 4.6.3 *with* graphite based on GMP (not MPIR), it was
something like
real152m18.098s
user138m19.662s
sys 9m16.678s
Now, I am using cloog-ppl built with MPIR. I wonder if that is the
reason why it takes longer?
Anyway. I will post the install logs to the ticket.
With the gcc that I first built, I could make Sage with '-march=native
-floop* -O3', but the result was inacceptably slow code. But it was
argued that it might be due to GMP. Now, I hope that I will get
amazingly fast code with MPIR...!
> To me it's ok to *include* GCC by default (replacing the Fortran
> package, such that the tarball doesn't grow much, although we could
> offer a tarball without it as well), but to only build it if really
> necessary or requested by the user.
The question is: Is a missing fortran compiler a good reason to build
all of GCC?
> I'm pretty sure we can find ways to avoid rebuilding GCC each time one
> builds a new Sage version, i.e., create scripts (or introduce
> environment variables) to easily recycle an already built GCC.
Sure!
Best regards,
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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Mar 16, 7:20 am, Simon King wrote:
> Hi all,
>
> On 15 Mrz., 14:10, Simon King wrote:
>
> > 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).
>
> It was argued that this was due to replacing MPIR by GMP.
>
> I am now trying again, after I learnt how to build CLooG-PPL with
> MPIR.
>
> However, there is one critical point: Building GCC (with Graphite)
> takes longer than 6 hours on my laptop! That is actually longer than
> building the rest of Sage.
Really? Compiling GCC 4.6.3 ('make -j4') with C, C++, Fortran, Java
and LTO enabled took exactly 3 hours on my dual-core Netbook(!) (AMD
Fusion E-450), plus <4 minutes 'make install'. (Bootstrapped with GCC
4.4.3, '-march=native -O3 -g'.)
Of course this is IMHO still much time for "regular" builds, but as I
understand it, only MacOS users (and others with really "outdated"/
buggy GCC versions) will suffer... XD
To me it's ok to *include* GCC by default (replacing the Fortran
package, such that the tarball doesn't grow much, although we could
offer a tarball without it as well), but to only build it if really
necessary or requested by the user.
I'm pretty sure we can find ways to avoid rebuilding GCC each time one
builds a new Sage version, i.e., create scripts (or introduce
environment variables) to easily recycle an already built GCC.
-leif
> Therefore, I am still +1 for having it an optional package, but -10
> for using it if all what we need is just a Fortran compiler ( I guess
> building a fortran compiler works faster than 6 hours), and -100 for
> building it by default on all platforms.
>
> 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi all, On 15 Mrz., 14:10, Simon King wrote: > 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). It was argued that this was due to replacing MPIR by GMP. I am now trying again, after I learnt how to build CLooG-PPL with MPIR. However, there is one critical point: Building GCC (with Graphite) takes longer than 6 hours on my laptop! That is actually longer than building the rest of Sage. Therefore, I am still +1 for having it an optional package, but -10 for using it if all what we need is just a Fortran compiler ( I guess building a fortran compiler works faster than 6 hours), and -100 for building it by default on all platforms. 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi All!
On 13 Mrz., 10:00, Jeroen Demeyer 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Thursday, 15 March 2012 16:15:09 UTC+8, Simon King wrote: > > Hi Leif, > > On 15 Mrz., 08:17, leif wrote: > > > And I forgot: I also upgraded the optional GMP spkg (GMP is required > > > by CLooG). Our old spkg was broken and needed massive patches. I made > > > an spkg based on unpatched GMP 5.0.4 upstream sources. I have no idea > > > what would happen on other platforms, but it worked on openSuse. > > > > I'm almost sure CLooG works with MPIR (2.x, with '--enable-gmpcompat', > > which we pass) as well. IMHO at least MPIR 1.* does not implement all the GMP functionality, i.e. it is not really a drop-in replacement. E.g. GAP 4.5 does not work with MPIR, even though it does work with GMP 5.0.4. Does MPIR 2.* really implement more GMP calls than MPIR 1.* ? > > How? I mean, what options does one need to pass when one wants to > build CLooG-PPL with MPIR? It does not pick it up automatically - when > I first tried to build CLooG (not CLooG-PPL) then it failed because it > missed GMP. > > 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Leif, On 15 Mrz., 08:17, leif wrote: > > And I forgot: I also upgraded the optional GMP spkg (GMP is required > > by CLooG). Our old spkg was broken and needed massive patches. I made > > an spkg based on unpatched GMP 5.0.4 upstream sources. I have no idea > > what would happen on other platforms, but it worked on openSuse. > > I'm almost sure CLooG works with MPIR (2.x, with '--enable-gmpcompat', > which we pass) as well. How? I mean, what options does one need to pass when one wants to build CLooG-PPL with MPIR? It does not pick it up automatically - when I first tried to build CLooG (not CLooG-PPL) then it failed because it missed GMP. 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Mar 15, 7:15 am, Simon King wrote: > On 15 Mrz., 07:12, Simon King wrote: > > > I installed > > * a modified PPL 0.11.2 spkg, that also installs a C interface, > > * a CLooG-PPL 0.15.11 spkg > > * a modified gcc 4.6.3 spkg, that builds gcc with graphite. > > And I forgot: I also upgraded the optional GMP spkg (GMP is required > by CLooG). Our old spkg was broken and needed massive patches. I made > an spkg based on unpatched GMP 5.0.4 upstream sources. I have no idea > what would happen on other platforms, but it worked on openSuse. I'm almost sure CLooG works with MPIR (2.x, with '--enable-gmpcompat', which we pass) as well. -leif -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On 15 Mrz., 07:12, Simon King wrote: > I installed > * a modified PPL 0.11.2 spkg, that also installs a C interface, > * a CLooG-PPL 0.15.11 spkg > * a modified gcc 4.6.3 spkg, that builds gcc with graphite. And I forgot: I also upgraded the optional GMP spkg (GMP is required by CLooG). Our old spkg was broken and needed massive patches. I made an spkg based on unpatched GMP 5.0.4 upstream sources. I have no idea what would happen on other platforms, but it worked on openSuse. 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Leif, hi all, On 15 Mrz., 04:13, leif wrote: > So what we currently need is CLooG-PPL 0.15.11, and a suitable version > of PPL. (I'm not 100% sure that PPL 0.11.2, the version currently > shipped with Sage, works, but /I think/(tm) it will...) I installed * a modified PPL 0.11.2 spkg, that also installs a C interface, * a CLooG-PPL 0.15.11 spkg * a modified gcc 4.6.3 spkg, that builds gcc with graphite. TODO: Attempt to build graphite only if CLooG-PPL is present. I defined CFLAGS="-O3 -march=native -floop-interchange -floop-strip-mine -floop- block -fno-strict-aliasing" export CFLAGS CXXFLAGS="-O3 -march=native -floop-interchange -floop-strip-mine - floop-block -fno-strict-aliasing" export CXXFLAGS And then I rebuilt all of Sage (not only sage -ba, I rebuilt all spkgs). I did not run the test suite yet, but the compilation in this high optimization went smoothly on my openSuse laptop, and Sage starts. I am looking forward to the total time required for running sage - testall compared with and without optimization! 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Mar 14, 10:25 am, Simon King wrote: > On 14 Mrz., 10:17, Simon King wrote: > > > The first thing I tried was CLooG 0.17.0, not using --with-ppl. That > > has not been enough for your gcc spkg. > > To be precise: > PPL with C interface and CLooG 0.17.0 not using --with-ppl => no > success > Adding CLooG-PPL 0.15.11 not using --with-ppl => no success > Adding CLooG-PPL 0.15.11 using --with-ppl => success > Adding CLooG 0.17.0 using --with-ppl => still success, but I am not > sure if this is an artefact of the previous step. >From the GCC prerequisites page: CLooG-PPL version( 0.15 or CLooG 0.16 Necessary to build GCC with the Graphite loop optimizations. There are two versions available. CLooG-PPL 0.15 as well as CLooG 0.16. The former is the default right now. It can be downloaded from ftp://gcc.gnu.org/pub/gcc/infrastructure/ as cloog-ppl-0.15.tar.gz. CLooG 0.16 support is still in testing stage, but will be the default in future GCC releases. It is also available at ftp://gcc.gnu.org/pub/gcc/infrastructure/ as cloog-0.16.1.tar.gz. To use it add the additional configure option --enable-cloog-backend=isl. Even if CLooG 0.16 does not use PPL, PPL is still required for Graphite. In both cases --with-cloog configure option should be used if CLooG is not installed in your default library search path. Note that there are a couple of 0.15.x and 0.16.x versions floating around; e.g. CLooG-PPL 0.15.9 doesn't work with PPL 0.11 or 0.12 just because its 'configure' script expects *exactly* PPL version 0.10; as mentioned, the latest 0.15.x version of CLooG-PPL (0.15.11) works with both PPL 0.11 and 0.12, provided you specify (at least) '--with-ppl'. (Otherwise 'configure' searches for and finds PPL, but doesn't use it for whatever reason..., but we'd use '--with-ppl=$SAGE_ROOT/local' anyway.) So what we currently need is CLooG-PPL 0.15.11, and a suitable version of PPL. (I'm not 100% sure that PPL 0.11.2, the version currently shipped with Sage, works, but /I think/(tm) it will...) -leif -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On 14 Mrz., 10:17, Simon King wrote: > The first thing I tried was CLooG 0.17.0, not using --with-ppl. That > has not been enough for your gcc spkg. To be precise: PPL with C interface and CLooG 0.17.0 not using --with-ppl => no success Adding CLooG-PPL 0.15.11 not using --with-ppl => no success Adding CLooG-PPL 0.15.11 using --with-ppl => success Adding CLooG 0.17.0 using --with-ppl => still success, but I am not sure if this is an artefact of the previous step. 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Jeroen, On 14 Mrz., 09:52, Jeroen Demeyer wrote: > The reason they dropped the "-PPL" suffix is that CLooG 0.16 doesn't > need PPL as build-time dependency. CLooG-PPL needs PPL at build-time > (that's why you need the --with-ppl option). GCC with Graphite needs > PPL itself and CLooG(-PPL) 0.15 or later. The first thing I tried was CLooG 0.17.0, not using --with-ppl. That has not been enough for your gcc spkg. 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-14 09:46, Simon King wrote: > cloog is different from cloog-ppl I understand it that CLooG is the logical continuation of CLooG-PPL, so it is the same software. CLooG-PPL 0.15 was followed by CLooG 0.16. The reason they dropped the "-PPL" suffix is that CLooG 0.16 doesn't need PPL as build-time dependency. CLooG-PPL needs PPL at build-time (that's why you need the --with-ppl option). GCC with Graphite needs PPL itself and CLooG(-PPL) 0.15 or later. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Jeroen, On 14 Mrz., 09:27, Jeroen Demeyer wrote: > On 2012-03-14 07:24, Simon King wrote:> * CLooG needs GMP. > > Are you sure it doesn't work with MPIR? When I first tried to install CLooG, then it failed and told me that GMP is missing. That's why I said GLooG needs GMP. However, I could imagine that it would have worked with some configuration option for CLooG, such as --with-mpir. Anyway. It could be that cloog is actually not enough; there also is a cloog-ppl package. What I can say is: If the ppl spkg is built with C interface and the update of the gmp spkg is installed and cloog-ppl is built with the configuration option --with-ppl (yes, cloog-ppl still needs the option --with-ppl, believe it or not), then your gcc package can use CLooG PPL. Perhaps the gcc package would also be happy if one just had cloog with the option --with-ppl (cloog is different from cloog-ppl). Since I now have both cloog and cloog-ppl installed in my version of sage-5.0.beta7, I can't be sure. Unfortunately it seems that sage.math is still down, so that I can't post the update of the standard ppl spkg (enabling C interface, not just C++), the update of my cloog spkg (see #12666) and my new cloog- ppl spkg (will be posted at #12666 as well). 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On Wed, Mar 14, 2012 at 09:27:21AM +0100, Jeroen Demeyer wrote: > On 2012-03-14 07:24, Simon King wrote: > > * CLooG needs GMP. > Are you sure it doesn't work with MPIR? Incidentally, Cédric Bastoul, the author of CLooG is working is the same research lab as me and teaching is the same department as Nicolas. So if close contact is needed we could help. Cheers, Florent -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-14 07:24, Simon King wrote: > * CLooG needs GMP. Are you sure it doesn't work with MPIR? -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-14 02:07, John H Palmieri wrote: > I guess the answer to your question is yes, though: even with OS X 10.6 > or earlier, the proposal is to build gcc. Since Xcode doesn't include a Fortran compiler and the Fortran spkg will be removed, we *have* to build GCC anyway for gfortran. -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-14 01:11, Dima Pasechnik wrote: > You seem to indicate this will not be used to build Sage any more. > Right? Exactly. > (even though this is a working setup, as far as Sage 5.0.beta* are > concerned... Well, is there anything wrong with replacing a working setup by a different working setup? It is likely that GCC-4.6.3 produces better optimized machine code than GCC-4.2.1, so Sage will probably run slightly faster. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Mar 14, 6:44 am, P Purkayastha wrote: > On Wednesday, March 14, 2012 4:54:32 AM UTC+8, François wrote: > > > On Tue, 13 Mar 2012 01:18:41 P Purkayastha wrote: > > > On Tuesday, March 13, 2012 3:41:35 PM UTC+8, Jeroen Demeyer wrote: > > > > If you do this, you certainly should not restrict yourself to the > > > > Graphite command line options, you should optimize gcc's flags in > > > > general (like adding -march for example). > > > > > Jeroen. > > > > If you do set -march, then *-march=native* is a pretty safe option. gcc > > > picks up many optimization flags by itself after detecting the cpu. I > > have > > > used this option for more than 3 years now for my whole Gentoo linux > > setup. > > > > But (of course) binaries made from -march=native can not be distributed. > > > -march=native only works on x86* other archs like ppc(64) are not > > supported. > > > Francois > > Interesting. I didn't know this. It seems -march=native is not present on > arm and -march is itself not available on ppc. IIRC, '-march=native' (on Intel / AMD x86[_64] processors, which have a CPUID insn; likewise '-mtune=native') requires GCC >= 4,2.1; on PPC, you [only] have '-mcpu=...' and '-mtune=...' (while at least '- march=foo' implies '-mtune=foo'). Of course "native" only works as expected (or gives better performance) if the GCC present "knows" your CPU, i.e. has timing parameters for it (and probably knows newly added instructions or ISA extensions). Since OS distros and their packages are supposed to run on a broad variety of processors, their GCCs are configured (and all binary packages supplied are built) with quite generic (and hence suboptimal) settings. Also, tools like valgrind do not always know the latest processors (including all instructions generated by a more recent GCC). -leif -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On Tue, 13 Mar 2012 22:44:03 P Purkayastha wrote: On Wednesday, March 14, 2012 4:54:32 AM UTC+8, François wrote: On Tue, 13 Mar 2012 01:18:41 P Purkayastha wrote: > On Tuesday, March 13, 2012 3:41:35 PM UTC+8, Jeroen Demeyer wrote: > > If you do this, you certainly should not restrict yourself to the > > Graphite command line options, you should optimize gcc's flags in > > general (like adding -march for example). > > > > Jeroen. > > If you do set -march, then *-march=native* is a pretty safe option. gcc > picks up many optimization flags by itself after detecting the cpu. I have > used this option for more than 3 years now for my whole Gentoo linux setup. > > But (of course) binaries made from -march=native can not be distributed. -march=native only works on x86* other archs like ppc(64) are not supported. Francois Interesting. I didn't know this. It seems -march=native is not present on arm and -march is itself not available on ppc. Yep, good old -mcpu like in the good old days: frb15@p2n14-c /hpc/home/frb15 :gcc -v Using built-in specs. Target: powerpc64-suse-linux Configured with: ../configure --prefix=/usr --infodir=/usr/share/info -- mandir=/usr/share/man --libdir=/usr/lib64 --libexecdir=/usr/lib64 --enable- languages=c,c++,objc,fortran,obj-c++,java --enable-checking=release --with- gxx-include-dir=/usr/include/c++/4.3 --enable-ssp --disable-libssp --with- bugurl=http://bugs.opensuse.org/ --with-pkgversion='SUSE Linux' --disable- libgcj --disable-libmudflap --with-slibdir=/lib64 --with-system-zlib --enable- __cxa_atexit --enable-libstdcxx-allocator=new --disable-libstdcxx-pch -- enable-version-specific-runtime-libs --program-suffix=-4.3 --enable-linux- futex --without-system-libunwind --with-cpu=power4 --enable-secureplt --with- long-double-128 --build=powerpc64-suse-linux Thread model: posix gcc version 4.3.4 [gcc-4_3-branch revision 152973] (SUSE Linux) -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Leif, hi all, On 14 Mrz., 05:47, leif wrote: > ... > PPL is also part of Sage But then, GCC can not use CLooG/PPL (I have tried). I was told that it is because of a missing header. The header is easy to get, though: The current (standard) ppl spkg only builds the c++ interface, but not the c interface. In order to build it, one just needs to add two letters to ppl's spkg-install. Here is my experience, so far: * CLooG needs GMP. The optional GMP spkg is broken (at least on openSuse), even though it is massively patched, to address problems on various platforms. I suggest to upgrade the optional GMP spkg to the latest upstream release: For me, it works WITHOUT patches and passes all self test. See #12661. * After installing GMP, installation of CLooG works. I created an spkg, based on unpatched upstream sources. See #12666. * The PPL spkg needs to be slightly changed, so that it provides ppl_c.h (currently, it only provides ppl.hh). I've tried it locally, there is no ticket yet. * Then, the new GCC spkg installs fine -- but it can *still* not use CLooG/PPL!! The install log of the package says: checking for installed CLooG PPL Legacy... no In other words, building the PPL with C interface is not enough. > P.P,S.: I suggested offering an optional GCC spkg (and provided an > experimental 4.5.1 one IIRC) last year; the feedback was less > positive... :-) I guess one important difference is that we have now the trouble with OS X, that might be solvable with a GCC spkg. 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On Wednesday, March 14, 2012 4:54:32 AM UTC+8, François wrote: > > On Tue, 13 Mar 2012 01:18:41 P Purkayastha wrote: > > On Tuesday, March 13, 2012 3:41:35 PM UTC+8, Jeroen Demeyer wrote: > > > If you do this, you certainly should not restrict yourself to the > > > Graphite command line options, you should optimize gcc's flags in > > > general (like adding -march for example). > > > > > > Jeroen. > > > > If you do set -march, then *-march=native* is a pretty safe option. gcc > > picks up many optimization flags by itself after detecting the cpu. I > have > > used this option for more than 3 years now for my whole Gentoo linux > setup. > > > > But (of course) binaries made from -march=native can not be distributed. > > -march=native only works on x86* other archs like ppc(64) are not > supported. > > Francois > Interesting. I didn't know this. It seems -march=native is not present on arm and -march is itself not available on ppc. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Mar 13, 10:27 am, Simon King wrote: > Hi Jeroen, > > On 13 Mrz., 10:07, Jeroen Demeyer wrote: > > > On 2012-03-13 09:59, Simon King wrote:> I have just installed CLooG, and > > you said that Sage provides PPL. > > > Where did you install CLooG? > > By "install", I meant "install with Yast" (my laptop runs under > openSuse). Well, for GCC you'll need a specific version (currently / for GCC 4.6.3 and PPL 0.12: CLooG 0.15.11), and if I'm not completely wrong, you have to explicitly specify '--with-ppl' (optionally giving the path prefix where it is installed, if not [directly] below '/usr') when configuring it, since CLooG supports different backends, while GCC requires PPL. GCC (4.6.3) /should/ use (i.e., get built with support for) PPL+CLooG automatically if its 'configure' happens to find suitable versions of both. -leif IMPORTANT NOTE: If you set CFLAGS when [re]building the Sage library, you always HAVE TO add '-fno-strict-aliasing', since Python's CFLAGS (which contain it) get completely dropped in that case, and with strict aliasing rules GCC produces wrong code for (at least a couple of) Cython modules. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Wednesday, 14 March 2012 09:07:17 UTC+8, John H Palmieri wrote: > > > > On Tuesday, March 13, 2012 5:11:00 PM UTC-7, Dima Pasechnik wrote: >> >> >> >> On Monday, 12 March 2012 06:15:35 UTC+8, Jeroen Demeyer wrote: >>> >>> I have made a spkg for GCC (GNU compiler collection) version 4.6.3 with >>> compilers for C, C++ and Fortran, see Trac #12369. >>> >>> The GCC spkg depends on an MPC (multi-precision complex numbers) spkg. >>> There has been an optional MPC package and a Sage interface for it. GCC >>> would use the upgraded MPC package from #12515. >>> >>> According to (unwritten?) rules, one cannot add a spkg as standard >>> package if it hasn't been an optional spkg first. I would like to ask >>> for an exception for GCC and add it immediately as standard package in >>> sage-5.0. Also MPC would need to become a standard package, but this >>> already has been an optional package and so is less controversial. >>> >>> Let me clarify that GCC would not always be built: if the system already >>> contains sufficiently recent versions of gcc, g++ and gfortran (or g95 >>> or g77), then GCC would not be built. Currently, "sufficiently recent >>> version of gcc" is defined as: at least 4.4.0 but not 4.6.0 nor 4.6.1 >>> (there are known bugs in cliquer and rubiks with these gcc versions). >>> >> $ gcc -v >> Using built-in specs. >> Target: i686-apple-darwin10 >> [...] >> gcc version 4.2.1 (Apple Inc. build 5666) (dot 3) >> >> You seem to indicate this will not be used to build Sage any more. >> Right? >> (even though this is a working setup, as far as Sage 5.0.beta* are >> concerned... >> > > In my experience, it's not a working setup on OS X Lion: pari fails > self-tests (and this also leads to failures in Sage doctests), gsl fails > self-tests, and symmetrica also causes failures in Sage doctests. See also > #12459. (See #11881 for the meta ticket for building on Lion.) > > I guess the answer to your question is yes, though: even with OS X 10.6 or > earlier, the proposal is to build gcc. > how much binary incompatibility is it going to introduce? I imagine it's easy to reach the point when system libraries won't work any more together with binaries produced by such a compiler... Dima > > -- > John > > -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Mar 12, 9:58 pm, François Bissey wrote: > On Mon, 12 Mar 2012 21:44:21 Jeroen Demeyer wrote:> On 2012-03-12 19:48, > Volker Braun wrote: > > > I'm all in favor of adding the gcc spkg. > > > > One thing that I'm not entirely happy with is that we bulid gcc without > > > graphite/ppl/cloog, so I suspect that the code speed isn't as good as it > > > could be. But since we only use the self-compiled gcc if the os-supplied > > > one is crap I think its fine. > > > Well, I believe Graphite is used only when you supply specific command > > line options to gcc (-O3 is not sufficient). Since those options are > > rarely/never supplied by the various SPKGs, I don't think there is much > > to gain by adding CLooG. Well, you can set CFLAGS; meanwhile most spkgs should respect user settings as well. PPL is also part of Sage (although I don't recall right now whether it's a suitable version; 0,11 and 0.12 should be fine for GCC/CLooG), and CLooG[-PPL] itself is quite small (756KB for 0.15.11, which is the version currently needed by GCC). -leif P.S.: In general +1, although IMHO there should also be a Sage tarball without all these potentially unused spkgs. P.P,S.: I suggested offering an optional GCC spkg (and provided an experimental 4.5.1 one IIRC) last year; the feedback was less positive... :-) > Correct. The options you get with graphite are: > -floop-interchange > -floop-strip-mine > -floop-block > there is also > -fgraphite-identity > but I believe it is available when you have both graphite and lto at the same > time. > > I tested the graphite optimization on some lattice QCD code that has many > levels of looping and the benefits were quite significant (around 30%). I was > extremely impressed. > > Francois -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Tuesday, March 13, 2012 5:11:00 PM UTC-7, Dima Pasechnik wrote: > > > > On Monday, 12 March 2012 06:15:35 UTC+8, Jeroen Demeyer wrote: >> >> I have made a spkg for GCC (GNU compiler collection) version 4.6.3 with >> compilers for C, C++ and Fortran, see Trac #12369. >> >> The GCC spkg depends on an MPC (multi-precision complex numbers) spkg. >> There has been an optional MPC package and a Sage interface for it. GCC >> would use the upgraded MPC package from #12515. >> >> According to (unwritten?) rules, one cannot add a spkg as standard >> package if it hasn't been an optional spkg first. I would like to ask >> for an exception for GCC and add it immediately as standard package in >> sage-5.0. Also MPC would need to become a standard package, but this >> already has been an optional package and so is less controversial. >> >> Let me clarify that GCC would not always be built: if the system already >> contains sufficiently recent versions of gcc, g++ and gfortran (or g95 >> or g77), then GCC would not be built. Currently, "sufficiently recent >> version of gcc" is defined as: at least 4.4.0 but not 4.6.0 nor 4.6.1 >> (there are known bugs in cliquer and rubiks with these gcc versions). >> > $ gcc -v > Using built-in specs. > Target: i686-apple-darwin10 > [...] > gcc version 4.2.1 (Apple Inc. build 5666) (dot 3) > > You seem to indicate this will not be used to build Sage any more. > Right? > (even though this is a working setup, as far as Sage 5.0.beta* are > concerned... > In my experience, it's not a working setup on OS X Lion: pari fails self-tests (and this also leads to failures in Sage doctests), gsl fails self-tests, and symmetrica also causes failures in Sage doctests. See also #12459. (See #11881 for the meta ticket for building on Lion.) I guess the answer to your question is yes, though: even with OS X 10.6 or earlier, the proposal is to build gcc. -- John -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Wednesday, 14 March 2012 08:11:00 UTC+8, Dima Pasechnik wrote: > > > > On Monday, 12 March 2012 06:15:35 UTC+8, Jeroen Demeyer wrote: >> >> I have made a spkg for GCC (GNU compiler collection) version 4.6.3 with >> compilers for C, C++ and Fortran, see Trac #12369. >> >> The GCC spkg depends on an MPC (multi-precision complex numbers) spkg. >> There has been an optional MPC package and a Sage interface for it. GCC >> would use the upgraded MPC package from #12515. >> >> According to (unwritten?) rules, one cannot add a spkg as standard >> package if it hasn't been an optional spkg first. I would like to ask >> for an exception for GCC and add it immediately as standard package in >> sage-5.0. Also MPC would need to become a standard package, but this >> already has been an optional package and so is less controversial. >> >> Let me clarify that GCC would not always be built: if the system already >> contains sufficiently recent versions of gcc, g++ and gfortran (or g95 >> or g77), then GCC would not be built. Currently, "sufficiently recent >> version of gcc" is defined as: at least 4.4.0 but not 4.6.0 nor 4.6.1 >> (there are known bugs in cliquer and rubiks with these gcc versions). >> > $ gcc -v > Using built-in specs. > Target: i686-apple-darwin10 > [...] > gcc version 4.2.1 (Apple Inc. build 5666) (dot 3) > > You seem to indicate this will not be used to build Sage any more. > Right? > (even though this is a working setup, as far as Sage 5.0.beta* are > concerned... > your post also seems to say that the effort to port Sage to xcode4 is stopped. As also on darwin 11 the default compiler does not satisfy the version condition that you mention. Please clarify. > > Thanks, > Dima > > > >> An environment variable SAGE_INSTALL_GCC can be used to override this >> default choice (set to "yes" to force installation of GCC or to "no" to >> prevent it). >> >> The Fortran package (which contained gfortran binaries for OS X) would >> be removed, therefore about 4MB would be added to the source tarball >> size (Fortran was 33MB, GCC+MPC is 37MB). >> >> I think there is not much point in making GCC optional, since it's only >> used at build-time. Once Sage has been built, you don't need it anymore. >> >> What are your +1/-1 to making GCC and MPC standard packages? >> >> >> Jeroen. >> >> -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Monday, 12 March 2012 06:15:35 UTC+8, Jeroen Demeyer wrote: > > I have made a spkg for GCC (GNU compiler collection) version 4.6.3 with > compilers for C, C++ and Fortran, see Trac #12369. > > The GCC spkg depends on an MPC (multi-precision complex numbers) spkg. > There has been an optional MPC package and a Sage interface for it. GCC > would use the upgraded MPC package from #12515. > > According to (unwritten?) rules, one cannot add a spkg as standard > package if it hasn't been an optional spkg first. I would like to ask > for an exception for GCC and add it immediately as standard package in > sage-5.0. Also MPC would need to become a standard package, but this > already has been an optional package and so is less controversial. > > Let me clarify that GCC would not always be built: if the system already > contains sufficiently recent versions of gcc, g++ and gfortran (or g95 > or g77), then GCC would not be built. Currently, "sufficiently recent > version of gcc" is defined as: at least 4.4.0 but not 4.6.0 nor 4.6.1 > (there are known bugs in cliquer and rubiks with these gcc versions). > $ gcc -v Using built-in specs. Target: i686-apple-darwin10 [...] gcc version 4.2.1 (Apple Inc. build 5666) (dot 3) You seem to indicate this will not be used to build Sage any more. Right? (even though this is a working setup, as far as Sage 5.0.beta* are concerned... Thanks, Dima > An environment variable SAGE_INSTALL_GCC can be used to override this > default choice (set to "yes" to force installation of GCC or to "no" to > prevent it). > > The Fortran package (which contained gfortran binaries for OS X) would > be removed, therefore about 4MB would be added to the source tarball > size (Fortran was 33MB, GCC+MPC is 37MB). > > I think there is not much point in making GCC optional, since it's only > used at build-time. Once Sage has been built, you don't need it anymore. > > What are your +1/-1 to making GCC and MPC standard packages? > > > Jeroen. > > -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On Tue, 13 Mar 2012 01:18:41 P Purkayastha wrote: > On Tuesday, March 13, 2012 3:41:35 PM UTC+8, Jeroen Demeyer wrote: > > If you do this, you certainly should not restrict yourself to the > > Graphite command line options, you should optimize gcc's flags in > > general (like adding -march for example). > > > > Jeroen. > > If you do set -march, then *-march=native* is a pretty safe option. gcc > picks up many optimization flags by itself after detecting the cpu. I have > used this option for more than 3 years now for my whole Gentoo linux setup. > > But (of course) binaries made from -march=native can not be distributed. -march=native only works on x86* other archs like ppc(64) are not supported. Francois -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Jeroen, On 13 Mrz., 11:43, Jeroen Demeyer wrote: > You should be able to use MPIR instead of GMP. I was told by CLooG that it needs GMP. Independent of that, I wouldn't mind to fix the optional gmp package. I got it to install work on my openSUSE. Unfortunately, spkg-check fails, and also I have to recall how to add a patch to an spkg. Actually I am now trying to upgrade the optional package to gmp-5.0.4 - so far, it seems to build, and it already contains a patch that the old gmp-4.2.1 spkg would need in order to build on openSuse. > But maybe there is no need to install CLooG in $SAGE_LOCAL. We will see later... 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-13 10:39, Simon King wrote: > > > On 13 Mrz., 10:36, Jeroen Demeyer wrote: >> On 2012-03-13 10:27, Simon King wrote:> By "install", I meant "install with >> Yast" (my laptop runs under >>> openSuse). >> >> Does the GCC spkg find it? > > No idea yet. > > I the last minutes, I tried to follow your advice to install CLooG in > $SAGE_LOCAL. However, it requires gmp. So, I tried "sage -i gmp". > However, installation of the gmp package failed. You should be able to use MPIR instead of GMP. But maybe there is no need to install CLooG in $SAGE_LOCAL. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi! I created a ticket for the broken (at least on openSuse) optional gmp-4.2.1 package: #12661. 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On 13 Mrz., 10:36, Jeroen Demeyer wrote: > On 2012-03-13 10:27, Simon King wrote:> By "install", I meant "install with > Yast" (my laptop runs under > > openSuse). > > Does the GCC spkg find it? No idea yet. I the last minutes, I tried to follow your advice to install CLooG in $SAGE_LOCAL. However, it requires gmp. So, I tried "sage -i gmp". However, installation of the gmp package failed. 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-13 10:27, Simon King wrote: > By "install", I meant "install with Yast" (my laptop runs under > openSuse). Does the GCC spkg find it? -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Jeroen, On 13 Mrz., 10:07, Jeroen Demeyer wrote: > On 2012-03-13 09:59, Simon King wrote:> I have just installed CLooG, and you > said that Sage provides PPL. > > Where did you install CLooG? By "install", I meant "install with Yast" (my laptop runs under openSuse). 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-13 09:59, Simon King wrote: > I have just installed CLooG, and you said that Sage provides PPL. Where did you install CLooG? If you install it in $SAGE_ROOT/local, then it might "just work" if you compile the GCC spkg from #12369. Check the "configure" part of spkg/logs/gcc-4.6.3 for CLooG. -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-13 09:50, Simon King wrote: > So, can I assume that when I install PPL and CLooG on my machine, then > my system's gcc would be able to use the loop optimizer? Or would I > then also need to rebuild gcc? Yes, you need to rebuild GCC. > Or: If the user happens to have CLooG, would our gcc spkg > automatically support the loop optimization? No. I guess we could support an optional CLooG spkg in Sage. But I first want to concentrate on having a basic GCC spkg which works (it still needs_review at #12369). Extra features like this can still be added later. > And if I'd like to test something like "-march=native" mentioned by > Purkayastha, can I set some environment variable such that building a > Cython extension would be built with -march=native? Try CFLAGS="-O3 -march=native" CXXFLAGS="-O3 -march=native" and see if that works. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Jeroen, On 13 Mrz., 09:50, Simon King wrote: > And if I'd like to test something like "-march=native" mentioned by > Purkayastha, can I set some environment variable such that building a > Cython extension would be built with -march=native? Got it: With CFLAGS="-march=native", it seems to work. I have just installed CLooG, and you said that Sage provides PPL. So, can I expect that it would make sense to add -march=native and -floop- interchange -floop-strip-mine -floop-block to CFLAGS, and run sage -ba, in order to see whether Sage would become noticeably faster? 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi Jeroen, On 13 Mrz., 09:45, Jeroen Demeyer wrote: > Graphite depends on PPL (which is in Sage) and on CLooG (which is NOT in > Sage). So, while the Graphite loop optimizer is in GCC, it will not be > compiled in without these dependencies. Thank you! So, can I assume that when I install PPL and CLooG on my machine, then my system's gcc would be able to use the loop optimizer? Or would I then also need to rebuild gcc? Or: If the user happens to have CLooG, would our gcc spkg automatically support the loop optimization? And if I'd like to test something like "-march=native" mentioned by Purkayastha, can I set some environment variable such that building a Cython extension would be built with -march=native? 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-13 09:40, Simon King wrote: > So, could some expert please elaborate why it is claimed that the gcc > spkg does *not* come with graphite? I suppose gcc 4.6.3 (provided by > the spkg) contains graphite, since gcc 4.4 does. Graphite depends on PPL (which is in Sage) and on CLooG (which is NOT in Sage). So, while the Graphite loop optimizer is in GCC, it will not be compiled in without these dependencies. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi! On 13 Mrz., 07:33, Simon King wrote: > That sounds like > - We should add graphite/ppl/cloog to the new gcc spkg Google told me that graphite is in gcc since version 4.4. But from the comments in this thread I got the impression that graphite is something that one may or may not have on top of gcc. So, could some expert please elaborate why it is claimed that the gcc spkg does *not* come with graphite? I suppose gcc 4.6.3 (provided by the spkg) contains graphite, since gcc 4.4 does. 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On Tuesday, March 13, 2012 3:41:35 PM UTC+8, Jeroen Demeyer wrote: > > If you do this, you certainly should not restrict yourself to the > Graphite command line options, you should optimize gcc's flags in > general (like adding -march for example). > > Jeroen. > If you do set -march, then *-march=native* is a pretty safe option. gcc picks up many optimization flags by itself after detecting the cpu. I have used this option for more than 3 years now for my whole Gentoo linux setup. But (of course) binaries made from -march=native can not be distributed. -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-13 07:33, Simon King wrote: > - We should see what spkgs would benefit from the additional > optimization, and let their spkg-install / their makefiles provide the > relevant gcc command line options And who is volunteering to do this? Not me. This sounds like something which can be done by Volker's compilerwrapper. If you do this, you certainly should not restrict yourself to the Graphite command line options, you should optimize gcc's flags in general (like adding -march for example). Jeroen. -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
Le lundi 12 mars, Simon King a écrit: > - The spkg should be used not only if the system's gcc has the > "wrong" version, but also if the system's gcc does not provide > graphite/ppl/cloog Is sage a distribution of mathematical software or a distribution? It's already replicating a good chunk of my ARM setup (it mostly lacks the kernel, emacs and latex)! I voted +1 for gcc as a standard non-mandatory (wink to Keshav) package ; I vote -1 for yet another standard mandatory package. Snark on #sagemath -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On Mon, Mar 12, 2012 at 11:33 PM, Simon King wrote: > Hi François, Jeroen and Volker, > > On 12 Mrz., 21:58, François Bissey > wrote: >> On Mon, 12 Mar 2012 21:44:21 Jeroen Demeyer wrote:> On 2012-03-12 19:48, >> Volker Braun wrote: >> > > One thing that I'm not entirely happy with is that we bulid gcc without >> > > graphite/ppl/cloog, ... >> >> > Well, I believe Graphite is used only when you supply specific command >> > line options to gcc (-O3 is not sufficient). Since those options are >> > rarely/never supplied by the various SPKGs, I don't think there is much >> > to gain by adding CLooG. >> >> ... >> I tested the graphite optimization on some lattice QCD code that has many >> levels of looping and the benefits were quite significant (around 30%). I was >> extremely impressed. > > That sounds like > - We should add graphite/ppl/cloog to the new gcc spkg We could also wait to do this until later, but keep it in mind... > - The spkg should be used not only if the system's gcc has the > "wrong" version, but also if the system's gcc does not provide > graphite/ppl/cloog > - We should see what spkgs would benefit from the additional > optimization, and let their spkg-install / their makefiles provide the > relevant gcc command line options > - We should see what Cython modules may benefit from the additional > optimization, and add gcc command line options in module_list.py as > appropriate. > > Isn't it? > > 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 -- William Stein Professor of Mathematics University of Washington http://wstein.org -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Hi François, Jeroen and Volker, On 12 Mrz., 21:58, François Bissey wrote: > On Mon, 12 Mar 2012 21:44:21 Jeroen Demeyer wrote:> On 2012-03-12 19:48, > Volker Braun wrote: > > > One thing that I'm not entirely happy with is that we bulid gcc without > > > graphite/ppl/cloog, ... > > > Well, I believe Graphite is used only when you supply specific command > > line options to gcc (-O3 is not sufficient). Since those options are > > rarely/never supplied by the various SPKGs, I don't think there is much > > to gain by adding CLooG. > > ... > I tested the graphite optimization on some lattice QCD code that has many > levels of looping and the benefits were quite significant (around 30%). I was > extremely impressed. That sounds like - We should add graphite/ppl/cloog to the new gcc spkg - The spkg should be used not only if the system's gcc has the "wrong" version, but also if the system's gcc does not provide graphite/ppl/cloog - We should see what spkgs would benefit from the additional optimization, and let their spkg-install / their makefiles provide the relevant gcc command line options - We should see what Cython modules may benefit from the additional optimization, and add gcc command line options in module_list.py as appropriate. Isn't it? 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On Mon, 12 Mar 2012 21:44:21 Jeroen Demeyer wrote: > On 2012-03-12 19:48, Volker Braun wrote: > > I'm all in favor of adding the gcc spkg. > > > > One thing that I'm not entirely happy with is that we bulid gcc without > > graphite/ppl/cloog, so I suspect that the code speed isn't as good as it > > could be. But since we only use the self-compiled gcc if the os-supplied > > one is crap I think its fine. > > Well, I believe Graphite is used only when you supply specific command > line options to gcc (-O3 is not sufficient). Since those options are > rarely/never supplied by the various SPKGs, I don't think there is much > to gain by adding CLooG. Correct. The options you get with graphite are: -floop-interchange -floop-strip-mine -floop-block there is also -fgraphite-identity but I believe it is available when you have both graphite and lto at the same time. I tested the graphite optimization on some lattice QCD code that has many levels of looping and the benefits were quite significant (around 30%). I was extremely impressed. Francois -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
On 2012-03-12 19:48, Volker Braun wrote: > I'm all in favor of adding the gcc spkg. > > One thing that I'm not entirely happy with is that we bulid gcc without > graphite/ppl/cloog, so I suspect that the code speed isn't as good as it > could be. But since we only use the self-compiled gcc if the os-supplied > one is crap I think its fine. Well, I believe Graphite is used only when you supply specific command line options to gcc (-O3 is not sufficient). Since those options are rarely/never supplied by the various SPKGs, I don't think there is much to gain by adding CLooG. -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
On Sunday, March 11, 2012 3:15:35 PM UTC-7, Jeroen Demeyer wrote: > > What are your +1/-1 to making GCC and MPC standard packages? > > +1 (especially since it allows us to build Sage on Lion without making other compromises) -- John -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
I'm all in favor of adding the gcc spkg. One thing that I'm not entirely happy with is that we bulid gcc without graphite/ppl/cloog, so I suspect that the code speed isn't as good as it could be. But since we only use the self-compiled gcc if the os-supplied one is crap I think its fine. -- 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
Re: [sage-devel] Re: Adding GCC and MPC as standard packages
Le lundi 12 mars, Keshav Kini a écrit: > Julien Puydt writes: > > As it's optional, it's not really added: +1 too. > > But it is not optional - Jeroen is proposing to make this a standard > package. Or did I misunderstand you? Poor choice of words on my part : it is not "optional" since it's "standard", but it's "optional" since it's "not mandatory". It is that second meaning I had in mind. It will be like the fortran package, which checks for the presence of a convenient fortran compiler and sets things up to use it if found. which makes me think it might be interesting if packages had an uniform way to pass such information around : fortran has SAGE_FORTRAN and makes hackish symlinks ; the atlas package also has magical environment variables and symlinks tricks... Snark on #sagemath -- 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
[sage-devel] Re: Adding GCC and MPC as standard packages
Julien Puydt writes: > As it's optional, it's not really added: +1 too. But it is not optional - Jeroen is proposing to make this a standard package. Or did I misunderstand you? -Keshav Join us in #sagemath on irc.freenode.net ! -- 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
