#7375: upgrade M4RI to newest upstream release
---------------------------+------------------------------------------------
Reporter: malb | Owner: tbd
Type: enhancement | Status: needs_work
Priority: major | Milestone: sage-4.3
Component: packages | Keywords: M4RI, linear algebra
Work_issues: | Author: Martin Albrecht
Reviewer: | Merged:
---------------------------+------------------------------------------------
Changes (by drkirkby):
* status: needs_review => needs_work
Old description:
> Improvements in the most recent versions:
> * switched to LQUP instead of PLUQ for better performance
> * because of this and other improvements much better handling of
> sparse-ish matrices
> * overall better performance for elimination
> * better performance for mzd_transpose
> * dropped the check for the numer of CPUs from configure which was
> unused and not cross platform
> * optional tuning code to calculate cache sizes (not enabled by
> default)
> + some refactoring
> + mzd_row_add_offset() fixed a segfault
New description:
Improvements in the most recent versions:
* switched to LQUP instead of PLUQ for better performance
* because of this and other improvements much better handling of sparse-
ish matrices
* overall better performance for elimination
* better performance for mzd_transpose
* dropped the check for the numer of CPUs from configure which was
unused and not cross platform
* optional tuning code to calculate cache sizes (not enabled by default)
+ some refactoring
+ mzd_row_add_offset() fixed a segfault
--
Comment:
Hi,
I have had a few beers tonight, which are a nice benefit of attending the
London Open Solaris User Group (LOSUG), so I hope my comments are
coherent! (FWIW, one of tonights talks at LOSUG was very interesting from
the point of view of Sage).
Anyway, back to the trac ticket...There are still some problems with this,
which means the configure script believes the Sun compiler is broken. I've
also made some other points. Some are ''nit-picking'', but while you
resolve one issue, you might as well resolve the others.
The main problem is that spkg-install sets CFLAGS to include -fPIC and
-Wall, both of which are GNU specific flags. So when called with the Sun
compiler, the compiler fails to build an executable, as its given
erroneous flags. That can be seen in the config.log attached. It's
essential that flags like those are only added on compilers that accept
them.
I've attached a small script called ''testcc'' that I was intending at
some point including in a more general place in Sage, but you can use it
if you wish. It tests the C compiler and will report one of:
* GNU (for any GCC compatible compiler)
* Sun_on_Solaris (for the Sun compiler on Solaris)
* Intel_improved_GCC
* Sun_improved_GCC
* HP_on_Alpha_Linux
* HP_on_HPUX
* IBM_on_AIX
* HP_on_Tru64
* Unknown
The script works by checking what the C pre-processor defines, which is
more reliable than looking for words like GNU or Sun in the output. You
could use that script to decide what flag to add to the compiler - whether
it is -fPIC for GNU compilers, or -Kpic for the Sun compiler. I would not
worry about the others at the minute.
Here is what the script gives as output on both Sun and GNU compilers on a
Sun workstation. It's also been tested on AIX, tru64 and HP-UX too.
{{{
drkir...@swan:[~] $ export CC=/opt/xxxsunstudio12.1/bin/cc
drkir...@swan:[~] $ ./testcc
Sun_on_Solaris
drkir...@swan:[~] $ export CC=gcc
drkir...@swan:[~] $ ./testcc
GNU
}}}
Out of completeness, I'll attach another for the C++ compiler
(''testcxx''), but I do not think you will want the one for C++.
Using that script, you could then add in spkg-install something like the
following, which I have not tested.
{{{
if [ x`testcc` = "xGNU" ] ; then
CFLAGS="$CFLAGS -fPIC -Wall "
elif [ x`testcc` = "xSun_on_Solaris" ] ; then
CLFAGS="$CLFAGS -Kpic "
elif [ x`testcc` = "xHP_on_HPUX" ] ; then
CFLAGS="$CFLAGS + z "
fi
}}}
The Sun compilers are a lot more fussy than the GNU ones, so enabling
extra warnings will probably show too many.
Whilst Sage does not aim to support HP-UX, it would be worth adding that
+z flag, as compiling with other compilers tends to show problems in code.
The library might build on HP-UX with those changes.)
Others issues I see are:
* -m64 is added only on OS X. I suggest a better alternative would be to
have:
{{{
if [ "x$SAGE64" = "xyes" ] ; then
CFLAGS="$CFLAGS -m64 "
fi
}}}
That's far from perfect, but will work with the Sun and GNU compilers, as
both accept -m64.
I do not believe you need to set CPPFLAGS to -m64, as that is not an
option for the C pre-processor. But it would be worth double-checking
that.
* The standard name for the file that creates the configure script was
changed from configure.in to configure.ac. So the autoconf manual says to
use that. See http://www.gnu.org/software/autocon/manual/autoconf.html
#Writing-Autoconf-Input where it says ''"Previous versions of Autoconf
promoted the name configure.in, which is somewhat ambiguous (the tool
needed to process this file is not described by its extension), and
introduces a slight confusion with config.h.in and so on (for which ‘.in’
means “to be processed by configure”). Using configure.ac is now
preferred."''
* The autoconf manual says configure.ac should start with AC_INIT, but
your configure.in does not. This might possibly be the reason the
configure script thinks the compiler is broken, though I doubt it.
* Adding something like ''AC_PREREQ([2.64])'' or whatever version of
autoconf you use to create the configure script will ensure nobody else
can recreate it with an older version. It is recommended practice to do
that. See
http://www.gnu.org/software/autoconf/manual/autoconf.html#Versioning
* I do not think it is a good idea to test for the Sun compiler in
configure.in (or configure.ac), to find what flags the compiler needs for
C99 support, for several reasons
* First, not all Sun compilers need any flag at all.
* Other compilers from HP, IBM, Intel etc might need their own flags.
A better way to add a flag for c99 support ('''if it is needed'''), is to
call the autoconf macro AC_PROG_CC_C99
http://www.gnu.org/software/autoconf/manual/autoconf.html#C-Compiler That
macro was designed for exactly this purpose. As such, you can delete all
the code that checks for the Sun compiler in configure.in and just use the
built-in macro. It would also be wise to exit with an error if the
compiler is not C99 compliant, if that is needed. Test if
$ac_cv_prog_cc_c99 = no, and if so exit with an error that the code needs
to be able to find a C99 compliant compiler.
If you can make those changes, this has a reasonable chance of building on
Solaris with the Sun compiler. I'll also try it on HP-UX for you. The
machine is not running at the minute, but there is no point in even
trying, as I know the GNU flags will break the build.
--
Ticket URL: <http://trac.sagemath.org/sage_trac/ticket/7375#comment:6>
Sage <http://www.sagemath.org>
Sage: Creating a Viable Open Source Alternative to Magma, Maple, Mathematica,
and MATLAB
--
You received this message because you are subscribed to the Google Groups
"sage-trac" group.
To post to this group, send email to [email protected].
For more options, visit this group at
http://groups.google.com/group/sage-trac?hl=.
