M4RI(E) 20111203
Martin Albrecht
I am planning to release new version of M4RI and M4RIE soon, possibly tomorrow.
== Changes ==
The main changes are:
M4RI: We now support reading/writing of 1-bit PNG images, i.e. you can
import/export your matrices and even look at them.
Preliminary release notes are here:
https://bitbucket.org/malb/m4ri/wiki/M4RI-20111203
M4RIE: Karatsuba-based multiplication was extended up to degree 8
extensions. Hence, elimination, trsm and multiplication got *a lot*
faster.
Preliminary release notes are here:
https://bitbucket.org/malb/m4rie/wiki/M4RIE-20111203
== Testing ==
I've uploaded tarballs to
http://m4ri.sagemath.org/alpha/
and I'd appreciate people testing them. To make it easier, you can use
this script:
https://bitbucket.org/malb/m4ri-scripts/src/6b00b902e122/m4rie-build-and-test.sh
which expects an M4RI and M4RIE tar.gz in the current working
directory and will then get GMP + Givaro and compile everything +
"make check". So, for example:
$ ls
m4ri-20111203.alpha1.tar.gz
m4rie-20111203.beta1.tar.gz
m4rie-build-and-test.sh
$ ./m4rie-build-and-test.sh
will compile & check both M4RI and M4RIE.
== Known Issues ==
I see SIGSEGVs on cicero when the test programs exit. On all other
machines I see no such issue and valgrind reports no address
violations. Hence, I'd particularly be interested in reports on other
32-bit machines.
== Pre-prints ==
There are to new pre-prints available on M4RI(E):
* Martin R. Albrecht, The M4RIE library for dense linear algebra over
small fields with even characteristic, URL
http://arxiv.org/abs/1111.6900
* Martin R. Albrecht, Gregory V. Bard, Clément Pernet, Efficient
Dense Gaussian Elimination over the Finite Field with Two Elements,
URL http://arxiv.org/abs/1111.6549
Comments very much welcome.
Cheers,
Martin
--
name: Martin Albrecht
_pgp: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x8EF0DC99
_otr: 47F43D1A 5D68C36F 468BAEBA 640E8856 D7951CCF
_www: http://www.informatik.uni-bremen.de/~malb
_jab: martinr...@jabber.ccc.de
Martin Albrecht
On Dec 2, 11:29 am, Martin Albrecht <martinralbre...@googlemail.com>
wrote:
> Hi,
>
> I am planning to release new version of M4RI and M4RIE soon, possibly tomorrow.
>
> == Changes ==
>
> The main changes are:
>
> M4RI: We now support reading/writing of 1-bit PNG images, i.e. you can
> import/export your matrices and even look at them.
>
> Preliminary release notes are here:https://bitbucket.org/malb/m4ri/wiki/M4RI-20111203
>
> M4RIE: Karatsuba-based multiplication was extended up to degree 8
> extensions. Hence, elimination, trsm and multiplication got *a lot*
> faster.
>
> Preliminary release notes are here:https://bitbucket.org/malb/m4rie/wiki/M4RIE-20111203
>
> == Testing ==
>
> I've uploaded tarballs to
>
> http://m4ri.sagemath.org/alpha/
>
> and I'd appreciate people testing them. To make it easier, you can use
> this script:
>
> https://bitbucket.org/malb/m4ri-scripts/src/6b00b902e122/m4rie-build-...
>
> which expects an M4RI and M4RIE tar.gz in the current working
> directory and will then get GMP + Givaro and compile everything +
> "make check". So, for example:
>
> $ ls
> m4ri-20111203.alpha1.tar.gz
> m4rie-20111203.beta1.tar.gz
> m4rie-build-and-test.sh
>
> $ ./m4rie-build-and-test.sh
>
> will compile & check both M4RI and M4RIE.
>
> == Known Issues ==
>
> I see SIGSEGVs on cicero when the test programs exit. On all other
> machines I see no such issue and valgrind reports no address
> violations. Hence, I'd particularly be interested in reports on other
> 32-bit machines.
>
> == Pre-prints ==
>
> There are to new pre-prints available on M4RI(E):
>
> * Martin R. Albrecht, The M4RIE library for dense linear algebra over
> small fields with even characteristic, URLhttp://arxiv.org/abs/1111.6900
> * Martin R. Albrecht, Gregory V. Bard, Clément Pernet, Efficient
> Dense Gaussian Elimination over the Finite Field with Two Elements,
> URLhttp://arxiv.org/abs/1111.6549
>
> Comments very much welcome.
>
> Cheers,
> Martin
>
> --
> name: Martin Albrecht
> _pgp:http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x8EF0DC99
> _otr: 47F43D1A 5D68C36F 468BAEBA 640E8856 D7951CCF
> _www:http://www.informatik.uni-bremen.de/~malb
> _jab: martinralbre...@jabber.ccc.de
Jerry James
I just pushed the button or buttons actually. M4RI(E) 20111203 is out.
I'm finally getting around to packaging M4RIE for Fedora, and have hit a few small issues with 20111203 on Fedora Rawhide (Fedora 17 to be). That platform includes givaro 3.5.0 and what will soon be released as gcc 4.7. First, the test programs would not compile because GFqDOM is in the Givaro namespace, but the references to them in the test programs are from the M4RIE namespace, so an explicit namespace identifier is required.
Second, givaro 3.5.0 includes code in <givaro/givconfig.h> to complain if <stdint.h> is included anywhere without first defining __STDC_LIMIT_MACROS. However, m4ri does exactly that, in <m4ri/misc.h>. It would probably be best to patch m4ri to fix this, but I cheated and defined __STDC_LIMIT_MACROS at the top of each test program to workaround the problem.
Third, the function mzd_slice_canary_is_alive in tests/testing.h has the "return 1" in one inner block too many, resulting in complaints from the compiler.
Finally, gcc gave lots of complaints about comparing signed and unsigned quantities. It appeared to me that most of these were gratuitous, in that the signed quantity could never be negative, so I made those go away with some judicious sprinkling of the "unsigned" keyword in the sources.
The results of all these changes are in the attached patch.
Oh, and FYI, something went wrong with the announcement of 20111203 on http://m4ri.sagemath.org/; it shows up in the News section under a date of "2011/10/104". :-)
Regards,
--
Jerry James
Martin Albrecht
> On Saturday, December 3, 2011 6:52:42 AM UTC-7, Martin Albrecht wrote:
> > I just pushed the button or buttons actually. M4RI(E) 20111203 is out.
>
> I'm finally getting around to packaging M4RIE for Fedora,
Great, thanks!
> and have hit a few small issues with 20111203 on Fedora Rawhide (Fedora 17
> to be). That platform includes givaro 3.5.0 and what will soon be released
> as gcc 4.7.
I've applied your patch except for the Givaro 3.5.0 change (see below).
However, I applied it to M4RIE head which has seen some refactoring, and thus
is not compatible with Sage as it is. I guess I should release this new
version soon-ish + the appropriate Sage patches.
> First, the test programs would not compile because GFqDOM is in the Givaro
> namespace, but the references to them in the test programs are from the
> M4RIE namespace, so an explicit namespace identifier is required.
Sage still ships with Givaro 3.2.13. Until that changes, we're stuck with this
version. However, updating that requires updating LinBox which might require
updating ATLAS. So I didn't include your patch for now. I guess one could do
it conditionally depending on the Givaro version?
> Second, givaro 3.5.0 includes code in <givaro/givconfig.h> to complain if
> <stdint.h> is included anywhere without first defining
> __STDC_LIMIT_MACROS. However, m4ri does exactly that, in <m4ri/misc.h>.
> It would probably be best to patch m4ri to fix this, but I cheated and
> defined __STDC_LIMIT_MACROS at the top of each test program to workaround
> the problem.
I added this to M4RI:
https://bitbucket.org/malb/m4ri/changeset/4a57d73ab61f
> Third, the function mzd_slice_canary_is_alive in tests/testing.h has the
> "return 1" in one inner block too many, resulting in complaints from the
> compiler.
That was already fixed in tip apparently.
> Finally, gcc gave lots of complaints about comparing signed and unsigned
> quantities. It appeared to me that most of these were gratuitous, in that
> the signed quantity could never be negative, so I made those go away with
> some judicious sprinkling of the "unsigned" keyword in the sources.
I applied this to M4RIE, I hope I got everything.
https://bitbucket.org/malb/m4rie/changeset/b81b9649e2d1
> The results of all these changes are in the attached patch.
Thanks again!
> Oh, and FYI, something went wrong with the announcement of 20111203 on
> http://m4ri.sagemath.org/; it shows up in the News section under a date of
> "2011/10/104". :-)
Thanks!
Cheers,
Martin
--
name: Martin Albrecht
_pgp: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x8EF0DC99
_otr: 47F43D1A 5D68C36F 468BAEBA 640E8856 D7951CCF
_www: http://martinralbrecht.wordpress.com/
_jab: martinr...@jabber.ccc.de