Opened 10 years ago
Last modified 6 years ago
#9562 closed enhancement
Add M4RIE to Sage — at Version 115
Reported by: | malb | Owned by: | |
---|---|---|---|
Priority: | major | Milestone: | sage-4.8 |
Component: | packages: standard | Keywords: | m4ri, sd32 |
Cc: | mvngu, SimonKing | Merged in: | |
Authors: | Martin Albrecht | Reviewers: | Paul Zimmermann, Simon King |
Report Upstream: | N/A | Work issues: | Fix some docs and fix "randomize()" |
Branch: | Commit: | ||
Dependencies: | #11574 | Stopgaps: |
Description (last modified by )
M4RIE is a library for linear algebra over small extension of GF(2). It is still in an early stage but already offers performance comparable to Magma for many inputs and is more than 1000 times faster than what we have in Sage right now.
Upstream: http://bitbucket.org/malb/m4rie/
Sage Days 24 coding sprint: http://wiki.sagemath.org/days24/projects/gf2e
- Install http://sage.math.washington.edu/home/malb/spkgs/libm4rie-20110821.spkg
- Apply m4rie_for_sage.patch
- Apply m4rie_spkg.patch to Sage root repository
There was a vote on sage-devel, recommending to add this as a standard spkg.
Change History (116)
comment:1 Changed 10 years ago by
- Description modified (diff)
comment:2 Changed 10 years ago by
comment:3 Changed 10 years ago by
The attached patch depends on #9475
comment:4 follow-up: ↓ 8 Changed 10 years ago by
The package compiles on t2. sage-check fails because libstdc++ cannot be found (I believe this is due to a problem in the old Sage I have on t2). I cannot apply my patch against this old version of Sage either.
comment:5 Changed 10 years ago by
This builds a static library only on Cygwin, but segfaults on both of the tests.
comment:6 Changed 10 years ago by
Mike, is there a Sage I can copy on winxp1?
comment:7 Changed 10 years ago by
sage -t devel/sage/sage/modular/modsym/space.py # 1 doctests failed sage -t devel/sage/sage/misc/sagedoc.py # 3 doctests failed sage -t devel/sage/sage/crypto/mq/mpolynomialsystem.py # 19 doctests failed sage -t devel/sage/sage/crypto/mq/sr.py # 7 doctests failed sage -t devel/sage/sage/modular/modsym/modsym.py # 1 doctests failed sage -t devel/sage/sage/rings/polynomial/pbori.pyx # 2 doctests failed sage -t devel/sage/sage/crypto/block_cipher/miniaes.py # 72 doctests failed
comment:8 in reply to: ↑ 4 Changed 10 years ago by
Replying to malb:
The package compiles on t2. sage-check fails because libstdc++ cannot be found (I believe this is due to a problem in the old Sage I have on t2). I cannot apply my patch against this old version of Sage either.
There's a Sage 4.5.1 package in /usr/local.
comment:9 Changed 10 years ago by
After unpacking that I get
21 from numpy.lib import triu ---> 22 from numpy.linalg import lapack_lite 23 from numpy.core.defmatrix import matrix_power 24 ImportError: ld.so.1: python: fatal: libgfortran.so.3: open failed: No such file or directory Error importing ipy_profile_sage - perhaps you should run %upgrade? WARNING: Loading of ipy_profile_sage failed.
Any ideas?
comment:10 Changed 10 years ago by
- Cc mvngu added
- Status changed from new to needs_work
The updated patch fixes all doctest failures.
PS: CCing Minh since I'm touching his code in a potentially non-trivial way/
comment:11 Changed 10 years ago by
It's not passing the tests properly on 64-bit OpenSolaris, and I doubt anywhere where SAGE64 needs to be set to yes. The -m64 flag is not getting passed when running the tests, so whilst it builds a 64-bit library, it looks like it tries to create 32-bit objects and link to that 64-bit library.
I have not investigated this in any detail, but they were my initial observations. I would try building on 't2' with SAGE64 set to yes. Not all of Sage will build 64-bit without some hacks, but it should be fairly easy to get enough of Sage built to build this library.
Successfully installed libm4ri-20100730 Running the test suite. Testing the M4RI library make -j12 test_elimination test_multiplication make[1]: Entering directory `/export/home/drkirkby/sage-4.5/spkg/build/libm4ri-20100730/m4rie' make[1]: warning: -jN forced in submake: disabling jobserver mode. g++ -DHAVE_CONFIG_H -I. -I./src -I/export/home/drkirkby/sage-4.5/local/include -m64 -g -O2 -MT test_elimination.o -MD -MP -MF .deps/test_elimination.Tpo -c -o test_elimination.o `test -f 'tests/test_elimination.cc' || echo './'`tests/test_elimination.cc g++ -DHAVE_CONFIG_H -I. -I./src -I/export/home/drkirkby/sage-4.5/local/include -m64 -g -O2 -MT test_multiplication.o -MD -MP -MF .deps/test_multiplication.Tpo -c -o test_multiplication.o `test -f 'tests/test_multiplication.cc' || echo './'`tests/test_multiplication.cc mv -f .deps/test_elimination.Tpo .deps/test_elimination.Po /bin/sh ./libtool --tag=CXX --mode=link g++ -g -O2 -lm4rie -lm4ri -lgivaro -lntl -lgmpxx -lgmp -lm -lstdc++ -o test_elimination test_elimination.o mv -f .deps/test_multiplication.Tpo .deps/test_multiplication.Po /bin/sh ./libtool --tag=CXX --mode=link g++ -g -O2 -lm4rie -lm4ri -lgivaro -lntl -lgmpxx -lgmp -lm -lstdc++ -o test_multiplication test_multiplication.o libtool: link: g++ -g -O2 -o .libs/test_elimination test_elimination.o /export/home/drkirkby/sage-4.5/spkg/build/libm4ri-20100730/m4rie/.libs/libm4rie.so -L/export/home/drkirkby/sage-4.5/local/lib /export/home/drkirkby/sage-4.5/local/lib/libm4ri.so /export/home/drkirkby/sage-4.5/local/lib/libgivaro.so -L/export/home/drkirkby/sage-4.5/local//lib -lntl /export/home/drkirkby/sage-4.5/local/lib/libgmpxx.so /export/home/drkirkby/sage-4.5/local/lib/libgmp.so /usr/local/gcc-4.4.4-multilib/lib/amd64/libstdc++.so -lm -Wl,-R -Wl,/export/home/drkirkby/sage-4.5/local/lib -Wl,-R -Wl,/usr/local/gcc-4.4.4-multilib/lib/amd64 libtool: link: g++ -g -O2 -o .libs/test_multiplication test_multiplication.o /export/home/drkirkby/sage-4.5/spkg/build/libm4ri-20100730/m4rie/.libs/libm4rie.so -L/export/home/drkirkby/sage-4.5/local/lib /export/home/drkirkby/sage-4.5/local/lib/libm4ri.so /export/home/drkirkby/sage-4.5/local/lib/libgivaro.so -L/export/home/drkirkby/sage-4.5/local//lib -lntl /export/home/drkirkby/sage-4.5/local/lib/libgmpxx.so /export/home/drkirkby/sage-4.5/local/lib/libgmp.so /usr/local/gcc-4.4.4-multilib/lib/amd64/libstdc++.so -lm -Wl,-R -Wl,/export/home/drkirkby/sage-4.5/local/lib -Wl,-R -Wl,/usr/local/gcc-4.4.4-multilib/lib/amd64 ldld:: fatal: filefatal :test_multiplication.o :file wrong test_elimination.o: wrong ELF class:ELF ELFCLASS64 ld: fatal: file processing errors.class No: output ELFCLASS64written to .libs/test_multiplication ld: fatal: file processing errors. No output written to .libs/test_elimination collect2: ld returned 1 exit status collect2: ld returned 1 exit status make[1]: *** [test_multiplication] Error 1 make[1]: *** Waiting for unfinished jobs.... make[1]: *** [test_elimination] Error 1 make[1]: Leaving directory `/export/home/drkirkby/sage-4.5/spkg/build/libm4ri-20100730/m4rie' make: *** [check-am] Error 2 Error testing M4RI ************************************* Error testing package ** libm4ri-20100730 ** ************************************* sage: An error occurred while testing libm4ri-20100730
comment:12 Changed 10 years ago by
- Status changed from needs_work to needs_review
I updated the SPKG linked above
- Building shared libraries on Cygwin now
- Fixed the crashes in spkg-check in Cygwin (this was actually a real bug)
- Fixed flags for SAGE64
comment:13 Changed 10 years ago by
Everything works on my Cygwin install.
comment:14 follow-up: ↓ 15 Changed 10 years ago by
It passed all self-tests on 64-bit OpenSolaris (x64) and 64-bit Solaris 10 (SPARC). Since neither platform has a stable version of Sage yet, running the doctests is pointless.
A few questions:
- Has there been an agreement to add this library? If so, can you provide a link to it.
- Why is it not in another package, rather than added to the libm4ri package?
- Do the self tests pass on Linux?
- Do the doctests pass on Linux?
- Do the self-tests pass on 32-bit SPARC? (Note my point above about there being a 4.5.1 in /usr/local on t2)
- Do the doc tests pass on 32-bit SPARC?
- Do the self-tests pass on OS X?
- Do the doctests pass on OS X?
comment:15 in reply to: ↑ 14 ; follow-up: ↓ 16 Changed 10 years ago by
Replying to drkirkby:
A few questions:
- Has there been an agreement to add this library? If so, can you provide a link to it.
No decision on [sage-devel] has happened yet. However, the Sage developers here at Sage Days 24 seem to be in favour of adding it.
- Why is it not in another package, rather than added to the libm4ri package?
It makes maintaining the thing easier for all sides: I'm the maintainer of both libraries for both upstream and the SPKGs. It isn't even decided yet whether the two libraries might get merged in the future. Finally, William asked me to not add a new SPKG but to add the M4RIe extension to the M4RI package.
- Do the self tests pass on Linux?
Yes.
- Do the doctests pass on Linux?
Yes.
- Do the self-tests pass on 32-bit SPARC? (Note my point above about there being a 4.5.1 in /usr/local on t2)
Note my point above about not being able to use it.
- Do the doc tests pass on 32-bit SPARC?
No clue.
- Do the self-tests pass on OS X?
Yes.
- Do the doctests pass on OS X?
Yes.
comment:16 in reply to: ↑ 15 Changed 10 years ago by
Replying to malb:
Replying to drkirkby:
A few questions:
- Has there been an agreement to add this library? If so, can you provide a link to it.
No decision on [sage-devel] has happened yet. However, the Sage developers here at Sage Days 24 seem to be in favour of adding it.
If the packages does get positive review, there should be a note to the release manager(s) not to merge it until there has been an agreement. Though in this case, it looks like getting a vote seems a formality.
- Why is it not in another package, rather than added to the libm4ri package?
It makes maintaining the thing easier for all sides: I'm the maintainer of both libraries for both upstream and the SPKGs. It isn't even decided yet whether the two libraries might get merged in the future. Finally, William asked me to not add a new SPKG but to add the M4RIe extension to the M4RI package.
One obvious disadvantage of that approach is that since one library relies on the other, the first could be built in parallel with some other packages. That could potentially slow parallel builds.
- Do the self tests pass on Linux?
Yes.
- Do the doctests pass on Linux?
Yes.
- Do the self-tests pass on 32-bit SPARC? (Note my point above about there being a 4.5.1 in /usr/local on t2)
Note my point above about not being able to use it.
Your point above says that's probably because you have an old version.
But I said above, there is the latest version on there - (/usr/local/sage-4.5.1-Solaris_10_SPARC-sun4u-SunOS.tar.gz
is a pre-built copy of the latest version of Sage on 't2'). If that does not work, let me know - I'd be very surprised if it does not. Otherwise, you could just build Sage from source.
- Do the doc tests pass on 32-bit SPARC?
No clue.
See point above.
Dave
comment:17 Changed 10 years ago by
Dave, the testuite fails:
/bin/bash ./libtool --tag=CXX --mode=link g++ -g -O2 -lm4rie -lm4ri -lgivaro -lntl -lgmpxx -lgmp -lm -lstdc++ -o test_elimination test_elimination.o libtool: link: warning: library `/home/malb/t2/sage-4.5.1-Solaris_10_SPARC-sun4u-SunOS/local/lib/libstdc++.la' was moved. libtool: link: cannot find the library `/usr/local/gcc-4.4.3/lib/libstdc++.la' or unhandled argument `/usr/local/gcc-4.4.3/lib/libstdc++.la' make[1]: *** [test_elimination] Error 1 make[1]: Leaving directory `/home/malb/t2/sage-4.5.1-Solaris_10_SPARC-sun4u-SunOS/spkg/build/libm4ri-20100730/m4rie'
Any idea why it wouldn't find libstdc++ on t2?
comment:18 Changed 10 years ago by
- Status changed from needs_review to needs_work
comment:19 follow-up: ↓ 20 Changed 10 years ago by
These lines:
libtool: link: warning: library `/home/malb/t2/sage-4.5.1-Solaris_10_SPARC-sun4u-SunOS/local/lib/libstdc++.la' was moved. libtool: link: cannot find the library `/usr/local/gcc-4.4.3/lib/libstdc++.la' or unhandled argument `/usr/local/gcc-4.4.3/lib/libstdc++.la'
make me think it's the Sage binary that is broken? Why is there be a libstdc++ in the Sage tarball ?
comment:20 in reply to: ↑ 19 Changed 10 years ago by
Replying to malb:
These lines:
libtool: link: warning: library `/home/malb/t2/sage-4.5.1-Solaris_10_SPARC-sun4u-SunOS/local/lib/libstdc++.la' was moved. libtool: link: cannot find the library `/usr/local/gcc-4.4.3/lib/libstdc++.la' or unhandled argument `/usr/local/gcc-4.4.3/lib/libstdc++.la'make me think it's the Sage binary that is broken? Why is there be a libstdc++ in the Sage tarball ?
The reason it is there is that the version of gcc shipped with Solaris is 3.4.3, so there are no recent gcc libraries. The compiler is not built with Fortran support, so there is no libgfortran at all. One needs recent run-time libraries, with fortran support, so I added them to the Sage binary.
It may be that deleting (making a copy first) of those .la files will solve the problem. Otherwise, editing them to point at the location of the libraries in $SAGE_LOCAL/lib will almost certainly solve it.
If that does not work, just build Sage from source. It does not take too long if you build packages in parallel.
Dave
comment:21 Changed 10 years ago by
- Status changed from needs_work to needs_review
malb@t2:~/t2/sage-4.5.1$ ./sage -t devel/sage/sage/matrix/matrix_mod2_dense.pyx sage -t "devel/sage/sage/matrix/matrix_mod2_dense.pyx" [92.7 s] ---------------------------------------------------------------------- All tests passed! Total time for all tests: 92.8 seconds malb@t2:~/t2/sage-4.5.1$ ./sage -t devel/sage/sage/matrix/matrix_mod2e_dense.pyx sage -t "devel/sage/sage/matrix/matrix_mod2e_dense.pyx" [50.0 s] ---------------------------------------------------------------------- All tests passed! Total time for all tests: 50.0 seconds
After finally building Sage t2 I can confirm that doctests pass there too
comment:22 Changed 10 years ago by
- Status changed from needs_review to needs_info
It appears to be trying to use autoconf, but autoconf is not a perquisite for Sage. Are you sure the timestamps on all the files are right?
checking for x86 cpuid 0x0 output... unknown checking for the processor vendor... Unknown checking the L1 cache size... 0 Bytes checking the L2 cache size... 0 Bytes checking whether make -j30 sets $(MAKE)... (cached) yes configure: creating ./config.status config.status: creating Makefile config.status: creating src/config.h config.status: executing depfiles commands config.status: executing libtool commands (CDPATH="${ZSH_VERSION+.}:" && cd . && /bin/bash /rootpool2/local/kirkby/t2/64/s age-4.5.3.alpha0/spkg/build/libm4ri-20100730/m4ri/missing --run autoheader) aclocal.m4:16: warning: this file was generated for autoconf 2.65. You have another version of autoconf. It may work, but is not guaranteed to. If you have problems, you may need to regenerate the build system entirely. To do so, use the procedure documented by the package, typically `autoreconf'. rm -f src/stamp-h1 touch src/config.h.in cd . && /bin/bash ./config.status src/config.h config.status: creating src/config.h config.status: src/config.h is unchanged
comment:23 Changed 10 years ago by
I replaced the SPKG with a version where I touched both configure scripts again (I thought I did that before, but apparently I didn't). I tested it on t2 and it doesn't attempt to call autoconf.
comment:24 Changed 10 years ago by
- Status changed from needs_info to needs_review
comment:25 Changed 10 years ago by
Since there doesn't seem to be any movement on this ticket, I took the liberty to update the patch and to prepare a new SPKG:
http://sage.math.washington.edu/home/malb/spkgs/libm4ri-20100817.spkg
Just as before this ticket depends on #9717 which was merged in 4.5.3.alpha1.
I successfully built and doctested the SPKG + the patch on:
- sage.math: 64-bit Linux, Intel CPU, pass
- redhawk: 64-bit Linux, AMD CPU, pass
- bsd: OS X, pass
- t2: Solaris, pass (I failed to build R thus those doctests failed)
I also took a sage-4.5.3.alpha1.tar, replaced the M4RI SPKG and applied the patch. Then I built Sage from scratch on sage.math and ran make ptestlong. All doctests passed.
PS: This new SPKG runs some tests to detect the L1 and L2 cache sizes, thus it compiles a little bit longer than older SPKGs for M4RI. The gained performance is well worth the wait on e.g. modern Intel CPUs where it is better to detect how much memory is fast for random access than to rely on the actual L2 cache size.
comment:26 follow-up: ↓ 27 Changed 10 years ago by
Do you want to ignore m4ri
and m4rie
? Also the dist/
directory can now be removed as per ticket #5903.
comment:27 in reply to: ↑ 26 Changed 10 years ago by
Replying to mvngu:
Do you want to ignore
m4ri
andm4rie
?
What I mean is this:
[mvngu@sage libm4ri-20100817]$ hg st ? m4ri/.hgtags ? m4ri/AUTHORS ? m4ri/COPYING ? m4ri/ChangeLog ? m4ri/INSTALL ? m4ri/Makefile.am ? m4ri/Makefile.in ? m4ri/NEWS ? m4ri/README ? m4ri/aclocal.m4 ? m4ri/config.guess ? m4ri/config.sub ? m4ri/configure ? m4ri/configure.ac ? m4ri/depcomp ? m4ri/install-sh ? m4ri/ltmain.sh ? m4ri/m4/ax_cache_size.m4 ? m4ri/m4/ax_cache_size_tune.m4 ? m4ri/m4/ax_check_compiler_flags.m4 ? m4ri/m4/ax_cpu_vendor.m4 ? m4ri/m4/ax_ext.m4 ? m4ri/m4/ax_gcc_x86_cpuid.m4 ? m4ri/m4/ax_openmp.m4 ? m4ri/m4/libtool.m4 ? m4ri/m4/ltoptions.m4 ? m4ri/m4/ltsugar.m4 ? m4ri/m4/ltversion.m4 ? m4ri/m4/lt~obsolete.m4 ? m4ri/m4ri ? m4ri/m4ri.sln ? m4ri/m4ri.vcproj ? m4ri/missing ? m4ri/testsuite/.directory ? m4ri/testsuite/Makefile ? m4ri/testsuite/bench_elimination.c ? m4ri/testsuite/bench_multiplication.c ? m4ri/testsuite/bench_pluq.c ? m4ri/testsuite/bench_trsm_lowerleft.c ? m4ri/testsuite/bench_trsm_lowerright.c ? m4ri/testsuite/bench_trsm_upperleft.c ? m4ri/testsuite/bench_trsm_upperright.c ? m4ri/testsuite/cpucycles-20060326/alpha.c ? m4ri/testsuite/cpucycles-20060326/alpha.h ? m4ri/testsuite/cpucycles-20060326/amd64cpuinfo.c ? m4ri/testsuite/cpucycles-20060326/amd64cpuinfo.h ? m4ri/testsuite/cpucycles-20060326/amd64tscfreq.c ? m4ri/testsuite/cpucycles-20060326/amd64tscfreq.h ? m4ri/testsuite/cpucycles-20060326/clockmonotonic.c ? m4ri/testsuite/cpucycles-20060326/clockmonotonic.h ? m4ri/testsuite/cpucycles-20060326/compile ? m4ri/testsuite/cpucycles-20060326/cpucycles.html ? m4ri/testsuite/cpucycles-20060326/do ? m4ri/testsuite/cpucycles-20060326/do.notes ? m4ri/testsuite/cpucycles-20060326/gettimeofday.c ? m4ri/testsuite/cpucycles-20060326/gettimeofday.h ? m4ri/testsuite/cpucycles-20060326/hppapstat.c ? m4ri/testsuite/cpucycles-20060326/hppapstat.h ? m4ri/testsuite/cpucycles-20060326/powerpcaix.c ? m4ri/testsuite/cpucycles-20060326/powerpcaix.h ? m4ri/testsuite/cpucycles-20060326/powerpclinux.c ? m4ri/testsuite/cpucycles-20060326/powerpclinux.h ? m4ri/testsuite/cpucycles-20060326/powerpcmacos.c ? m4ri/testsuite/cpucycles-20060326/powerpcmacos.h ? m4ri/testsuite/cpucycles-20060326/sparc32psrinfo.c ? m4ri/testsuite/cpucycles-20060326/sparc32psrinfo.h ? m4ri/testsuite/cpucycles-20060326/sparcpsrinfo.c ? m4ri/testsuite/cpucycles-20060326/sparcpsrinfo.h ? m4ri/testsuite/cpucycles-20060326/test.c ? m4ri/testsuite/cpucycles-20060326/x86cpuinfo.c ? m4ri/testsuite/cpucycles-20060326/x86cpuinfo.h ? m4ri/testsuite/cpucycles-20060326/x86tscfreq.c ? m4ri/testsuite/cpucycles-20060326/x86tscfreq.h ? m4ri/testsuite/test_elimination.c ? m4ri/testsuite/test_kernel.c ? m4ri/testsuite/test_multiplication.c ? m4ri/testsuite/test_pluq.c ? m4ri/testsuite/test_solve.c ? m4ri/testsuite/test_trsm.c ? m4ri/testsuite/walltime.h ? m4rie/.hgignore ? m4rie/.hgtags ? m4rie/AUTHORS ? m4rie/COPYING ? m4rie/ChangeLog ? m4rie/INSTALL ? m4rie/Makefile.am ? m4rie/Makefile.in ? m4rie/NEWS ? m4rie/README ? m4rie/aclocal.m4 ? m4rie/bench/Makefile.am ? m4rie/bench/Makefile.in ? m4rie/bench/bench_elimination.cc ? m4rie/bench/bench_multiplication.cc ? m4rie/bench/cpucycles-20060326/alpha.c ? m4rie/bench/cpucycles-20060326/alpha.h ? m4rie/bench/cpucycles-20060326/amd64cpuinfo.c ? m4rie/bench/cpucycles-20060326/amd64cpuinfo.h ? m4rie/bench/cpucycles-20060326/amd64tscfreq.c ? m4rie/bench/cpucycles-20060326/amd64tscfreq.h ? m4rie/bench/cpucycles-20060326/clockmonotonic.c ? m4rie/bench/cpucycles-20060326/clockmonotonic.h ? m4rie/bench/cpucycles-20060326/compile ? m4rie/bench/cpucycles-20060326/cpucycles.html ? m4rie/bench/cpucycles-20060326/do ? m4rie/bench/cpucycles-20060326/do.notes ? m4rie/bench/cpucycles-20060326/gettimeofday.c ? m4rie/bench/cpucycles-20060326/gettimeofday.h ? m4rie/bench/cpucycles-20060326/hppapstat.c ? m4rie/bench/cpucycles-20060326/hppapstat.h ? m4rie/bench/cpucycles-20060326/powerpcaix.c ? m4rie/bench/cpucycles-20060326/powerpcaix.h ? m4rie/bench/cpucycles-20060326/powerpclinux.c ? m4rie/bench/cpucycles-20060326/powerpclinux.h ? m4rie/bench/cpucycles-20060326/powerpcmacos.c ? m4rie/bench/cpucycles-20060326/powerpcmacos.h ? m4rie/bench/cpucycles-20060326/sparc32psrinfo.c ? m4rie/bench/cpucycles-20060326/sparc32psrinfo.h ? m4rie/bench/cpucycles-20060326/sparcpsrinfo.c ? m4rie/bench/cpucycles-20060326/sparcpsrinfo.h ? m4rie/bench/cpucycles-20060326/test.c ? m4rie/bench/cpucycles-20060326/x86cpuinfo.c ? m4rie/bench/cpucycles-20060326/x86cpuinfo.h ? m4rie/bench/cpucycles-20060326/x86tscfreq.c ? m4rie/bench/cpucycles-20060326/x86tscfreq.h ? m4rie/bench/walltime.h ? m4rie/config.guess ? m4rie/config.sub ? m4rie/configure ? m4rie/configure.ac ? m4rie/depcomp ? m4rie/gf2e_cxx/finite_field_givaro.h ? m4rie/install-sh ? m4rie/ltmain.sh ? m4rie/m4/ax_cache_size.m4 ? m4rie/m4/ax_cache_size_tune.m4 ? m4rie/m4/ax_check_compiler_flags.m4 ? m4rie/m4/ax_cpu_vendor.m4 ? m4rie/m4/ax_ext.m4 ? m4rie/m4/ax_gcc_x86_cpuid.m4 ? m4rie/m4/ax_openmp.m4 ? m4rie/m4/libtool.m4 ? m4rie/m4/ltoptions.m4 ? m4rie/m4/ltsugar.m4 ? m4rie/m4/ltversion.m4 ? m4rie/m4/lt~obsolete.m4 ? m4rie/missing ? m4rie/tests/Makefile ? m4rie/tests/test_elimination.cc ? m4rie/tests/test_multiplication.cc
comment:28 Changed 10 years ago by
I've updated the SPKG accordingly at
http://sage.math.washington.edu/home/malb/spkgs/libm4ri-20100817.p0.spkg
comment:29 follow-up: ↓ 30 Changed 10 years ago by
Whatever checks are being used to determine the cache size is not working very well. First it reports the L1 cache size is 0, then it spends a couple of minutes on a 3.33 GHz Xeon, to determine the cache size (I thought it had hanged). It's also producing some NaN in the calculation of the cache size - is that not a bug?
The CPU is an Intel Xeon W3580 and the operating system OpenSolaris.
checking for gcc option to accept ISO C99... -std=gnu99 checking for x86 cpuid output... b:756e6547:6c65746e:49656e69 checking for x86 cpuid 0x0 output... b:756e6547:6c65746e:49656e69 checking for the processor vendor... Intel checking for x86 cpuid 0x00000001 output... 106a5:100800:9ce3bd:bfebfbff checking whether mmx is supported... yes checking whether sse is supported... yes checking whether sse2 is supported... yes checking whether sse3 is supported... yes checking whether ssse3 is supported... yes checking whether C compiler accepts -mmmx... yes checking whether C compiler accepts -msse... yes checking whether C compiler accepts -msse2... yes checking whether C compiler accepts -msse3... yes checking mm_malloc.h usability... yes checking mm_malloc.h presence... yes checking for mm_malloc.h... yes checking for x86 cpuid 0x0 output... (cached) b:756e6547:6c65746e:49656e69 checking for the processor vendor... (cached) Intel checking for x86 cpuid 0x80000000 output... 80000008:0:0:0 checking for x86 cpuid 0x80000005 output... 0:0:0:0 checking for x86 cpuid 0x80000006 output... 0:0:1006040:0 checking the L1 cache size... 0 Bytes checking the L2 cache size... 262144 Bytes checking for cache sizes... s: 4, rx: 0.03, x: 0.03, wt: 0.03, dx: NaN s: 8, rx: 0.06, x: 0.06, wt: 0.06, dx: 1.01 s: 16, rx: 0.12, x: 0.12, wt: 0.12, dx: 1.00 s: 32, rx: 0.24, x: 0.24, wt: 0.24, dx: 1.00 s: 64, rx: 0.53, x: 0.53, wt: 0.53, dx: 1.10 s: 128, rx: 0.32, x: 1.30, wt: 0.32, dx: 1.23 s: 256, rx: 0.37, x: 2.95, wt: 0.37, dx: 1.14 s: 512, rx: 0.42, x: 6.77, wt: 0.42, dx: 1.15 s: 4, rx: 0.03, x: 0.03, wt: 0.03, dx: NaN s: 8, rx: 0.06, x: 0.06, wt: 0.06, dx: 0.94 s: 16, rx: 0.12, x: 0.12, wt: 0.12, dx: 0.99 s: 32, rx: 0.24, x: 0.24, wt: 0.24, dx: 1.02 s: 64, rx: 0.53, x: 0.53, wt: 0.53, dx: 1.09 s: 128, rx: 0.32, x: 1.29, wt: 0.32, dx: 1.22 s: 256, rx: 0.37, x: 2.97, wt: 0.37, dx: 1.16 s: 512, rx: 0.43, x: 6.80, wt: 0.43, dx: 1.14 s: 4, rx: 0.03, x: 0.03, wt: 0.03, dx: NaN s: 8, rx: 0.06, x: 0.06, wt: 0.06, dx: 0.91 s: 16, rx: 0.12, x: 0.12, wt: 0.12, dx: 1.01 s: 32, rx: 0.24, x: 0.24, wt: 0.24, dx: 1.01 s: 64, rx: 0.52, x: 0.52, wt: 0.52, dx: 1.09 s: 128, rx: 0.32, x: 1.30, wt: 0.32, dx: 1.24 s: 256, rx: 0.37, x: 2.94, wt: 0.37, dx: 1.13 s: 512, rx: 0.41, x: 6.64, wt: 0.42, dx: 1.13 s: 4, rx: 0.03, x: 0.03, wt: 0.03, dx: NaN s: 8, rx: 0.06, x: 0.06, wt: 0.06, dx: 0.92 s: 16, rx: 0.12, x: 0.12, wt: 0.12, dx: 1.02 s: 32, rx: 0.24, x: 0.24, wt: 0.24, dx: 1.00 s: 64, rx: 0.53, x: 0.53, wt: 0.53, dx: 1.11 s: 128, rx: 0.33, x: 1.30, wt: 0.33, dx: 1.23 s: 256, rx: 0.37, x: 2.98, wt: 0.37, dx: 1.14 s: 512, rx: 0.41, x: 6.61, wt: 0.41, dx: 1.11 s: 4, rx: 0.03, x: 0.03, wt: 0.03, dx: NaN s: 8, rx: 0.06, x: 0.06, wt: 0.06, dx: 0.93 s: 16, rx: 0.12, x: 0.12, wt: 0.12, dx: 1.01 s: 32, rx: 0.24, x: 0.24, wt: 0.24, dx: 1.02 s: 64, rx: 0.53, x: 0.53, wt: 0.53, dx: 1.09 s: 128, rx: 0.32, x: 1.30, wt: 0.32, dx: 1.23 s: 256, rx: 0.37, x: 2.94, wt: 0.37, dx: 1.13 s: 512, rx: 0.42, x: 6.75, wt: 0.42, dx: 1.15 s: 512, rx: 0.42, x: 0.42, wt: 0.42, dx: NaN s: 1024, rx: 1.00, x: 1.00, wt: 1.00, dx: 1.18 s: 1536, rx: 0.39, x: 1.57, wt: 0.39, dx: 1.05 s: 2048, rx: 0.27, x: 2.19, wt: 0.27, dx: 1.04 s: 3072, rx: 0.21, x: 3.32, wt: 0.21, dx: 1.01 s: 4096, rx: 0.29, x: 4.60, wt: 0.29, dx: 1.04 s: 6144, rx: 0.28, x: 8.85, wt: 0.28, dx: 1.28 s: 8192, rx: 0.25, x: 15.96, wt: 0.25, dx: 1.35 s: 16384, rx: 0.43, x: 55.40, wt: 0.44, dx: 1.74 s: 32768, rx: 0.61, x: 156.25, wt: 0.62, dx: 1.41 s: 512, rx: 0.43, x: 0.43, wt: 0.43, dx: NaN s: 1024, rx: 0.99, x: 0.99, wt: 0.99, dx: 1.15 s: 1536, rx: 0.39, x: 1.56, wt: 0.39, dx: 1.05 s: 2048, rx: 0.27, x: 2.13, wt: 0.27, dx: 1.03 s: 3072, rx: 0.21, x: 3.32, wt: 0.21, dx: 1.04 s: 4096, rx: 0.28, x: 4.52, wt: 0.28, dx: 1.02 s: 6144, rx: 0.27, x: 8.76, wt: 0.28, dx: 1.29 s: 8192, rx: 0.25, x: 15.87, wt: 0.25, dx: 1.36 s: 16384, rx: 0.42, x: 54.27, wt: 0.43, dx: 1.71 s: 32768, rx: 0.61, x: 156.22, wt: 0.62, dx: 1.44 s: 512, rx: 0.42, x: 0.42, wt: 0.42, dx: NaN s: 1024, rx: 0.99, x: 0.99, wt: 0.99, dx: 1.17 s: 1536, rx: 0.39, x: 1.56, wt: 0.39, dx: 1.05 s: 2048, rx: 0.27, x: 2.14, wt: 0.27, dx: 1.03 s: 3072, rx: 0.21, x: 3.31, wt: 0.21, dx: 1.03 s: 4096, rx: 0.28, x: 4.53, wt: 0.29, dx: 1.03 s: 6144, rx: 0.27, x: 8.73, wt: 0.28, dx: 1.28 s: 8192, rx: 0.25, x: 16.01, wt: 0.25, dx: 1.38 s: 16384, rx: 0.42, x: 54.24, wt: 0.43, dx: 1.69 s: 32768, rx: 0.63, x: 162.00, wt: 0.65, dx: 1.49 s: 512, rx: 0.43, x: 0.43, wt: 0.43, dx: NaN s: 1024, rx: 1.01, x: 1.01, wt: 1.01, dx: 1.19 s: 1536, rx: 0.20, x: 1.58, wt: 0.20, dx: 1.04 s: 2048, rx: 0.28, x: 2.21, wt: 0.28, dx: 1.04 s: 3072, rx: 0.21, x: 3.39, wt: 0.21, dx: 1.02 s: 4096, rx: 0.29, x: 4.63, wt: 0.29, dx: 1.02 s: 6144, rx: 0.28, x: 8.84, wt: 0.28, dx: 1.27 s: 8192, rx: 0.25, x: 16.17, wt: 0.26, dx: 1.37 s: 16384, rx: 0.43, x: 55.01, wt: 0.44, dx: 1.70 s: 32768, rx: 0.61, x: 157.06, wt: 0.63, dx: 1.43 s: 512, rx: 0.43, x: 0.43, wt: 0.43, dx: NaN s: 1024, rx: 1.01, x: 1.01, wt: 1.01, dx: 1.17 s: 1536, rx: 0.20, x: 1.59, wt: 0.20, dx: 1.05 s: 2048, rx: 0.27, x: 2.19, wt: 0.27, dx: 1.03 s: 3072, rx: 0.21, x: 3.40, wt: 0.21, dx: 1.03 s: 4096, rx: 0.29, x: 4.63, wt: 0.29, dx: 1.02 s: 6144, rx: 0.28, x: 8.90, wt: 0.28, dx: 1.28 s: 8192, rx: 0.25, x: 16.12, wt: 0.26, dx: 1.36 s: 16384, rx: 0.43, x: 54.90, wt: 0.44, dx: 1.70 s: 32768, rx: 0.61, x: 157.41, wt: 0.63, dx: 1.43 65536:8388608 checking the L1 cache size... 65536 Bytes checking the L2 cache size... 8388608 Bytes
comment:30 in reply to: ↑ 29 Changed 10 years ago by
Replying to drkirkby:
Whatever checks are being used to determine the cache size is not working very well.
I disagree, it works fine as far as I know but it is slow. For your machine I'd assume that 65536:8388608 indeed gives pretty good performance. If you want to check whether this hunch is correct let me know and I can tell you how to patch and test M4RI for various cache size configurations.
First it reports the L1 cache size is 0,
That's probably because I don't know how to ask Solaris for the right information, however the tuning performed now is the better strategy anyway.
then it spends a couple of minutes on a 3.33 GHz Xeon, to determine the cache size (I thought it had hanged).
Tuning takes a while as described above. Some shells don't seem to print intermediate outputs, I don't know how to fix that. If you do, let me know. Also, I couldn't get reliable information if I lowered the time spent on tuning, if you have any ideas, let me know.
It's also producing some NaN in the calculation of the cache size - is that not a bug?
No, the delta from the first element with respect to the previous element is not defined.
comment:31 Changed 10 years ago by
Minh, do you think you'll have some time to review this?
comment:32 Changed 10 years ago by
the speedup provided by this patch is quite impressive. With vanilla Sage 4.6 on a 2.83Ghz Core 2:
---------------------------------------------------------------------- | Sage Version 4.6, Release Date: 2010-10-30 | | Type notebook() for the GUI, and license() for information. | ---------------------------------------------------------------------- sage: m=matrix(GF(2^8,'x'),1000,1000) sage: m.randomize() sage: time r=m*m CPU times: user 76.33 s, sys: 0.10 s, total: 76.43 s Wall time: 77.63 s
With this patch applied:
sage: m=matrix(GF(2^8,'x'),1000,1000) sage: m.randomize() sage: time r=m*m CPU times: user 0.27 s, sys: 0.00 s, total: 0.27 s Wall time: 0.29 s
Paul Zimmermann
comment:33 Changed 10 years ago by
- Description modified (diff)
- Reviewers set to Paul Zimmermann
comment:34 Changed 9 years ago by
On PPC OS X 10.4, this spkg installation hangs at
checking the L1 cache size... 32768 Bytes checking the L2 cache size... 262144 Bytes checking for cache sizes...
I'm assuming this "cache size" checking isn't supposed to take 10 or more minutes.
comment:35 Changed 9 years ago by
Can you run spkg-install manually, i.e. unpack the spkg and run ./spkg-install in a SAGE shell? It will print intermediate information which might tell us whether it just takes long or whether it really hangs?
comment:36 Changed 9 years ago by
So far it just seems to be extremely slow. Well, this is a machine with a 700MHz processor...
In lines like
s: 512, rx: 18.03, x: 18.03, wt: 18.03, dx: NaN s: 1024, rx: 73.30, x: 73.30, wt: 73.31, dx: 2.03
which column gives the timing, if any?
comment:37 follow-up: ↓ 38 Changed 9 years ago by
Okay, it did finally finish!
checking for cache sizes... s: 4, rx: 0.20, x: 0.20, wt: 0.20, dx: NaN s: 8, rx: 0.38, x: 0.38, wt: 0.38, dx: 0.96 s: 16, rx: 0.39, x: 0.79, wt: 0.39, dx: 1.03 s: 32, rx: 0.46, x: 1.83, wt: 0.46, dx: 1.16 s: 64, rx: 0.49, x: 3.92, wt: 0.49, dx: 1.07 s: 128, rx: 0.57, x: 9.17, wt: 0.57, dx: 1.17 s: 256, rx: 0.63, x: 40.57, wt: 0.64, dx: 2.21 s: 512, rx: 1.17, x: 300.03, wt: 1.18, dx: 3.70 s: 4, rx: 0.19, x: 0.19, wt: 0.19, dx: NaN s: 8, rx: 0.38, x: 0.38, wt: 0.38, dx: 1.01 s: 16, rx: 0.42, x: 0.83, wt: 0.42, dx: 1.10 s: 32, rx: 0.44, x: 1.78, wt: 0.44, dx: 1.06 s: 64, rx: 0.50, x: 4.00, wt: 0.50, dx: 1.13 s: 128, rx: 0.46, x: 14.83, wt: 0.46, dx: 1.85 s: 256, rx: 0.99, x: 63.40, wt: 0.99, dx: 2.14 s: 512, rx: 1.21, x: 310.35, wt: 1.22, dx: 2.45 s: 4, rx: 0.17, x: 0.17, wt: 0.17, dx: NaN s: 8, rx: 0.39, x: 0.39, wt: 0.39, dx: 1.14 s: 16, rx: 0.43, x: 0.87, wt: 0.43, dx: 1.11 s: 32, rx: 0.43, x: 1.73, wt: 0.43, dx: 1.00 s: 64, rx: 0.47, x: 3.74, wt: 0.47, dx: 1.08 s: 128, rx: 0.60, x: 9.58, wt: 0.60, dx: 1.28 s: 256, rx: 0.56, x: 36.13, wt: 0.57, dx: 1.89 s: 512, rx: 1.25, x: 320.55, wt: 1.26, dx: 4.44 s: 4, rx: 0.21, x: 0.21, wt: 0.21, dx: NaN s: 8, rx: 0.39, x: 0.39, wt: 0.39, dx: 0.94 s: 16, rx: 0.43, x: 0.85, wt: 0.43, dx: 1.10 s: 32, rx: 0.44, x: 1.74, wt: 0.44, dx: 1.02 s: 64, rx: 0.48, x: 3.87, wt: 0.48, dx: 1.11 s: 128, rx: 0.59, x: 9.39, wt: 0.59, dx: 1.21 s: 256, rx: 0.63, x: 40.42, wt: 0.63, dx: 2.15 s: 512, rx: 1.16, x: 297.41, wt: 1.17, dx: 3.68 s: 4, rx: 0.19, x: 0.19, wt: 0.19, dx: NaN s: 8, rx: 0.39, x: 0.39, wt: 0.39, dx: 1.01 s: 16, rx: 0.41, x: 0.82, wt: 0.41, dx: 1.04 s: 32, rx: 0.42, x: 1.69, wt: 0.42, dx: 1.04 s: 64, rx: 0.48, x: 3.81, wt: 0.48, dx: 1.13 s: 128, rx: 0.59, x: 9.40, wt: 0.59, dx: 1.23 s: 256, rx: 0.73, x: 46.75, wt: 0.73, dx: 2.49 s: 512, rx: 1.20, x: 306.20, wt: 1.20, dx: 3.27 s: 512, rx: 20.54, x: 20.54, wt: 20.54, dx: NaN s: 1024, rx: 68.61, x: 68.61, wt: 68.63, dx: 1.67 s: 1536, rx: 0.24, x: 124.21, wt: 0.26, dx: 1.21 s: 2048, rx: 0.18, x: 182.39, wt: 0.20, dx: 1.10 s: 3072, rx: 0.30, x: 303.95, wt: 0.32, dx: 1.11 s: 4096, rx: 0.24, x: 487.92, wt: 0.28, dx: 1.20 s: 6144, rx: 0.19, x: 793.79, wt: 0.25, dx: 1.08 s: 8192, rx: 0.00, x: 0.38, wt: 0.08, dx: 0.00 s: 16384, rx: 0.00, x: 0.31, wt: 0.19, dx: 0.41 s: 32768, rx: 0.00, x: 0.37, wt: 0.32, dx: 0.59 s: 512, rx: 18.06, x: 18.06, wt: 18.07, dx: NaN s: 1024, rx: 68.43, x: 68.43, wt: 68.43, dx: 1.89 s: 1536, rx: 0.24, x: 125.17, wt: 0.26, dx: 1.22 s: 2048, rx: 0.17, x: 169.91, wt: 0.18, dx: 1.02 s: 3072, rx: 0.31, x: 318.15, wt: 0.34, dx: 1.25 s: 4096, rx: 0.22, x: 444.11, wt: 0.26, dx: 1.05 s: 6144, rx: 0.20, x: 803.39, wt: 0.26, dx: 1.21 s: 8192, rx: 0.00, x: 0.30, wt: 0.08, dx: 0.00 s: 16384, rx: 0.00, x: 0.38, wt: 0.15, dx: 0.62 s: 32768, rx: 0.00, x: 0.31, wt: 0.31, dx: 0.41 s: 512, rx: 17.85, x: 17.85, wt: 17.86, dx: NaN s: 1024, rx: 68.59, x: 68.59, wt: 68.60, dx: 1.92 s: 1536, rx: 0.26, x: 131.07, wt: 0.27, dx: 1.27 s: 2048, rx: 0.18, x: 182.17, wt: 0.20, dx: 1.04 s: 3072, rx: 0.32, x: 323.67, wt: 0.34, dx: 1.18 s: 4096, rx: 0.22, x: 444.65, wt: 0.26, dx: 1.03 s: 6144, rx: 0.19, x: 788.59, wt: 0.25, dx: 1.18 s: 8192, rx: 0.29, x: 1203.62, wt: 0.38, dx: 1.14 s: 16384, rx: 0.00, x: 0.35, wt: 0.16, dx: 0.00 s: 32768, rx: 0.00, x: 0.35, wt: 0.33, dx: 0.50 s: 512, rx: 17.39, x: 17.39, wt: 17.40, dx: NaN s: 1024, rx: 68.02, x: 68.02, wt: 68.03, dx: 1.96 s: 1536, rx: 0.24, x: 123.60, wt: 0.26, dx: 1.21 s: 2048, rx: 0.17, x: 178.99, wt: 0.19, dx: 1.09 s: 3072, rx: 0.29, x: 301.11, wt: 0.32, dx: 1.12 s: 4096, rx: 0.24, x: 501.66, wt: 0.28, dx: 1.25 s: 6144, rx: 0.19, x: 796.99, wt: 0.25, dx: 1.06 s: 8192, rx: 0.28, x: 1141.88, wt: 0.36, dx: 1.07 s: 16384, rx: 0.00, x: 0.36, wt: 0.15, dx: 0.00 s: 32768, rx: 0.00, x: 0.37, wt: 0.31, dx: 0.51 s: 512, rx: 19.65, x: 19.65, wt: 19.66, dx: NaN s: 1024, rx: 68.37, x: 68.37, wt: 68.38, dx: 1.74 s: 1536, rx: 0.23, x: 119.48, wt: 0.26, dx: 1.16 s: 2048, rx: 0.20, x: 202.14, wt: 0.22, dx: 1.27 s: 3072, rx: 0.29, x: 299.73, wt: 0.32, dx: 0.99 s: 4096, rx: 0.22, x: 441.61, wt: 0.26, dx: 1.11 s: 6144, rx: 0.19, x: 795.29, wt: 0.25, dx: 1.20 s: 8192, rx: 0.28, x: 1141.73, wt: 0.35, dx: 1.08 s: 16384, rx: 0.00, x: 0.35, wt: 0.16, dx: 0.00 s: 32768, rx: 0.00, x: 0.35, wt: 0.32, dx: 0.50 262144:524288
Doctest failure:
sage: A.pivots() # indirect doctest Expected: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] Got: (0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14)
Series of failures, for all such doctests:
AttributeError: 'sage.matrix.matrix_mod2e_dense.Matrix_mod2e_dense' object has no attribute '_multiply_classical'
Is something not inheriting properly? This method seems to be defined in Sage, and similar doctests in the rest of the matrix/ folder pass (in fact, all other tests in crypto and matrix pass other than one unrelated one from something I did on this installation).
comment:38 in reply to: ↑ 37 ; follow-up: ↓ 39 Changed 9 years ago by
Replying to kcrisman:
Okay, it did finally finish!
Okay, good ... well, I'm still not sure what I should do about this. The easiest might be to disable cache tuning on PPC?
Doctest failure:
sage: A.pivots() # indirect doctest Expected: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] Got: (0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14)
I can reproduce this one.
Series of failures, for all such doctests:
AttributeError: 'sage.matrix.matrix_mod2e_dense.Matrix_mod2e_dense' object has no attribute '_multiply_classical'Is something not inheriting properly? This method seems to be defined in Sage, and similar doctests in the rest of the matrix/ folder pass (in fact, all other tests in crypto and matrix pass other than one unrelated one from something I did on this installation).
I cannot seem to reproduce these (in the matrix/ folder). Which file does give this?
PS: Thank you so much for taking an interest in this ticket!
comment:39 in reply to: ↑ 38 Changed 9 years ago by
Okay, good ... well, I'm still not sure what I should do about this. The easiest might be to disable cache tuning on PPC?
What I would say is to disable this on any machine that is probably old and slow - this is probably not a PPC thing per se. That would be OS X 10.4, probably older versions of Ubuntu, ... I don't know how one would do this, though.
Series of failures, for all such doctests:
AttributeError: 'sage.matrix.matrix_mod2e_dense.Matrix_mod2e_dense' object has no attribute '_multiply_classical'I cannot seem to reproduce these (in the matrix/ folder). Which file does give this?
This is the new file and class 'sage.matrix.matrix_mod2e_dense.Matrix_mod2e_dense' as indicated above. Since the tests for _multiply_classical
work in other files, something isn't working about the inheritance.
This could be related to my having installed numpy 1.6 before testing. But that would seem strange, since neither this nor the other multiplication rely on this, they are .pyx files... and all the other ones work...
PS: Thank you so much for taking an interest in this ticket!
I just have an interest in making sure older systems can still use Sage. There is too much planned obsolescence in computers already.
comment:40 Changed 9 years ago by
- Status changed from needs_review to needs_work
Hi, can you give
http://sage.math.washington.edu/home/malb/spkgs/libm4ri-20110701.alpha.spkg
a try? It doesn't fix the doctest failures but compilation should be quicker and provide more feedback.
comment:41 Changed 9 years ago by
Sorry, I won't have access to that machine for a while now.
comment:42 Changed 9 years ago by
- Dependencies set to #11574
comment:43 Changed 9 years ago by
I'm a little confused. The spkgs here and at #11574 are both libm4ri, and seem to be numbered backwards (the "earlier" one in the dependency is from 2011).
comment:44 follow-up: ↓ 45 Changed 9 years ago by
- Owner changed from tbd to (none)
Okay, I figured it out - sorry for the noise. I agree that it is very annoying to have two things inside the same spkg. There is, for instance, an open ticket to remove rpy2 from the r spkg.
comment:45 in reply to: ↑ 44 ; follow-ups: ↓ 46 ↓ 90 Changed 9 years ago by
Replying to kcrisman:
Okay, I figured it out - sorry for the noise.
NP, I should have explained things better.
I agree that it is very annoying to have two things inside the same spkg. There is, for instance, an open ticket to remove rpy2 from the r spkg.
Well, it makes it easier to re-use tuning results. For example, I plan to run the cache tuning only once, i.e. for M4RI and M4RIE would re-use the results.
comment:46 in reply to: ↑ 45 Changed 9 years ago by
Replying to malb:
Replying to kcrisman:
I agree that it is very annoying to have two things inside the same spkg. There is, for instance, an open ticket to remove rpy2 from the r spkg.
Well, it makes it easier to re-use tuning results. For example, I plan to run the cache tuning only once, i.e. for M4RI and M4RIE would re-use the results.
That makes a lot of sense, especially since that takes a while even on speedy machines.
By the way, I tried #11574 but ran into trouble with an undefined symbol about m4ri_swap_bits
, perhaps due to my having installed #9562 first. I did apply the patch at #11574. Anyway, I'm reverting to the vanilla Sage version, and then going back to #11574 (I already had done the PolyBoRi upgrade).
comment:47 follow-up: ↓ 48 Changed 9 years ago by
Hi, I updated the patch for #11574 21h ago, did you try it? It fixes the missing symbol stuff.
comment:48 in reply to: ↑ 47 Changed 9 years ago by
comment:49 Changed 9 years ago by
- Description modified (diff)
comment:50 Changed 9 years ago by
- Status changed from needs_work to needs_review
comment:51 Changed 9 years ago by
- Description modified (diff)
comment:52 Changed 9 years ago by
- Report Upstream changed from None of the above - read trac for reasoning. to N/A
comment:53 Changed 9 years ago by
- Status changed from needs_review to needs_work
Please no
depends = [SAGE_ROOT + "/local/include/m4rie/m4rie.h"],
in module_list.py I have just cleaned up (#11377) do
depends = [SAGE_INC + "m4rie/m4rie.h"],
instead.
comment:55 Changed 9 years ago by
- Description modified (diff)
comment:56 Changed 9 years ago by
If I could just squeeze an extra request (not counting for review): could you tag a corresponding release of m4rie on bitbucket? That way I could make a gentoo ebuild from the bitbucket release rather than pulling the spkg.
I also see that suddenly we have split libm4rie from libm4ri (which I am happy about) but without apparent warnings.
comment:57 Changed 9 years ago by
- Status changed from needs_review to needs_work
Yeah, sorry I was a bit lazy:
1) There will be an official 20110715 release of both M4RI and M4RIE. I put together the SPKGs to test whether they work with Sage. So perhaps I shouldn't have set "needs review"
2) Yes, I gave in to the demand to split them. I should have mentioned it.
comment:58 Changed 9 years ago by
I must say it only occurred to me yesterday while looking at tags on bitbucket that 20110715 is in the future.
comment:59 Changed 9 years ago by
- Status changed from needs_work to needs_review
It's now officially released.
comment:60 Changed 9 years ago by
- Description modified (diff)
comment:61 Changed 9 years ago by
- Cc SimonKing added
comment:62 Changed 9 years ago by
- Status changed from needs_review to needs_work
I get numerous errors when trying to install the spkg. The following is just a small selection of errors:
... /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:248: error: too many arguments to function 'mzed_make_table' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:249: warning: passing argument 2 of 'mzed_make_table' makes pointer from integer without a cast /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:145: note: expected 'struct mzed_t *' but argument is of type 'size_t' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:249: warning: passing argument 3 of 'mzed_make_table' makes pointer from integer without a cast /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:145: note: expected 'struct gf2e *' but argument is of type 'size_t' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:249: error: too many arguments to function 'mzed_make_table' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:251: error: 'mzed_t' has no member named 'nrows' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:251: warning: passing argument 2 of 'mzed_process_rows3' makes pointer from integer without a cast /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.h:172: note: expected 'struct mzed_t *' but argument is of type 'size_t' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:251: warning: passing argument 4 of 'mzed_process_rows3' makes pointer from integer without a cast /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.h:172: note: expected 'struct mzed_t *' but argument is of type 'size_t' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:251: error: too many arguments to function 'mzed_process_rows3' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:253: warning: passing argument 3 of 'mzed_process_rows3' makes pointer from integer without a cast /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.h:172: note: expected 'struct mzed_t *' but argument is of type 'size_t' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:253: warning: passing argument 4 of 'mzed_process_rows3' makes pointer from integer without a cast /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.h:172: note: expected 'struct mzed_t *' but argument is of type 'size_t' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:253: error: too many arguments to function 'mzed_process_rows3' /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:256: warning: passing argument 2 of 'mzed_make_table' makes pointer from integer without a cast /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/travolta.c:145: note: expected 'struct mzed_t *' but argument is of type 'size_t' make[1]: libtool: compile: gcc -std=gnu99 -DHAVE_CONFIG_H -I. -I/mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie -I./src -I/mnt/local/king/SAGE/sage-4.7.1.rc1/local/include -I/mnt/local/king/SAGE/sage-4.7.1.rc1/local/include -g -fPIC -Wall -pedantic -O2 -MT finite_field.lo -MD -MP -MF .deps/finite_field.Tpo -c /mnt/local/king/SAGE/sage-4.7.1.rc1/spkg/build/libm4rie-20110715/m4rie/src/finite_field.c -o finite_field.o >/dev/null 2>&1 *** [gf2e_matrix.lo] Fehler 1 ...
What may be the problem? Is it needed to apply the patches first (in contrast to what is said in the ticket description)?
comment:63 Changed 9 years ago by
Ouch. Sorry. Right when hitting the "submit" button, I saw that the ticket has a dependency that I forgot to apply first.
comment:64 Changed 9 years ago by
- Status changed from needs_work to needs_review
comment:65 Changed 9 years ago by
I got doctest failures in 4 files. But I have to admit that I applied your patch on top of various other patches, in particular #9138 (which is quite invasive). So, it could be that the problems are ultimately due to other patches.
comment:66 Changed 9 years ago by
I'll build 4.7.1.rc2 (on sage.math and locally) and test with that.
comment:67 follow-up: ↓ 69 Changed 9 years ago by
Simon, did you install #11261 as well and rebuild pbori.pyx? It's a dependency of the M4RI update and easy to overlook.
comment:68 Changed 9 years ago by
With everything applied + 4.7.1.rc2 I get
The following tests failed: sage -t -long -force_lib devel/sage/sage/tests/cmdline.py # 1 doctests failed
which is normal on sage.math.
comment:69 in reply to: ↑ 67 Changed 9 years ago by
comment:70 follow-up: ↓ 72 Changed 9 years ago by
- Reviewers changed from Paul Zimmermann to Paul Zimmermann, Simon King
- Status changed from needs_review to needs_work
The doc tests pass, but I think the documentation must be put into the reference manual.
comment:71 Changed 9 years ago by
- Work issues set to Improve scalar multiplication; add docs to the references
I found one benchmark that you should improve:
sage: MS = MatrixSpace(GF(64,'a'),5000,5000) sage: K = MS.base_ring() sage: c = K.random_element() sage: %time A*c CPU times: user 1.33 s, sys: 0.02 s, total: 1.35 s Wall time: 1.35 s 5000 x 5000 dense matrix over Finite Field in a of size 2^6 sage: MS1 = MatrixSpace(GF(64,'a'),1,5000) sage: B = MS1.random_element() sage: %timeit B*c 625 loops, best of 3: 291 µs per loop
The reason is that "A*c" relies on a slow generic implementation.
I am sure that your library has a fast way to multiply a row respectively a matrix with a scalar. But you should overload _lmul_
(please not _rmul_
, according to the documentation), so that the user can benefit from it.
comment:72 in reply to: ↑ 70 Changed 9 years ago by
Replying to SimonKing:
The doc tests pass, but I think the documentation must be put into the reference manual.
I'm not sure that's desirable: most functions are cdefs and hence won't show up. Other functions begin with underscores (_foo
) and hence won't show up. Thus, it looks pretty weird and as if very few actual functionality was implemented.
comment:73 follow-up: ↓ 74 Changed 9 years ago by
- Status changed from needs_work to needs_review
The updated patch includes _lmul_
and includes this class in the reference manual. However, there isn't much content any way so it might be better not to include it?
comment:74 in reply to: ↑ 73 Changed 9 years ago by
- Work issues changed from Improve scalar multiplication; add docs to the references to Improve scalar multiplication
Replying to malb:
The updated patch includes
_lmul_
and includes this class in the reference manual. However, there isn't much content any way so it might be better not to include it?
Actually it seems you are right. ALL attributes of sage.matrix.matrix_mod2_dense.Matrix_mod2_dense
either start with "_" or override a method that is documented in sage.matrix.matrix_dense.Matrix_dense
.
comment:75 Changed 9 years ago by
I am puzzled. I see that your new patch version contains _lmul_. When I downloaded it and read it in an editor, I find _lmul_. When I qdelete the old patch and qimport the new one, then _lmul_ is missing.
Do you have any explanation for what I did wrong?
comment:76 follow-up: ↓ 77 Changed 9 years ago by
qpush
missing?
comment:77 in reply to: ↑ 76 Changed 9 years ago by
Replying to malb:
qpush
missing?
No. When I tried first, I simply did qpush
after the qimport, and sage -br
, but found that _lmul_
was missing.
When one qimports a patch, then it is stored in the folder .hg/patches
. When one qdeletes it, it is also removed from the folder.
And when I qimported your new patch, a file with the given patch name was created in .hg/patches
, but that file did not contain the string "lmul".
That never happened to me before today.
comment:78 follow-up: ↓ 79 Changed 9 years ago by
I worked around the problem: I downloaded the new patch, saved it locally, and qimported it from the local file. I wonder, though, why the old patch version was not used when qimporting it with the http-address?
comment:79 in reply to: ↑ 78 Changed 9 years ago by
Replying to SimonKing:
I wonder, though, why the old patch version was not used when qimporting it with the http-address?
One negation to many: "... why the old patch version was used when qimporting it via http".
comment:80 Changed 9 years ago by
The timings have clearly improved:
sage: MS = MatrixSpace(GF(64,'a'),5000,5000) sage: K = MS.base_ring() sage: c = K.random_element() sage: A = MS.random_element() sage: %time A*c CPU times: user 0.68 s, sys: 0.02 s, total: 0.70 s Wall time: 0.71 s 5000 x 5000 dense matrix over Finite Field in a of size 2^6 sage: MS1 = MatrixSpace(GF(64,'a'),1,5000) sage: B = MS1.random_element() sage: %timeit B*c 625 loops, best of 3: 88.3 µs per loop
That is good enough, I'd say.
Personally, I am still not happy, since scalar multiplication will very frequently occur in my application. My Meataxe fork does the first example in 39.4 ms and the second one in 20 µs. On the other hand, the time that I can gain by using M4RIE in echelon computation will probably be more than the time lost in scalar multiplication...
comment:81 Changed 9 years ago by
comment:82 Changed 9 years ago by
the updated patch avoids adding this class to the reference manual.
comment:83 follow-up: ↓ 84 Changed 9 years ago by
With the old version of the M4RI and M4RIE patches, I got some doctest errors on mark (which is solaris):
sage -t -force_lib "devel/sage/sage/rings/polynomial/pbori.pyx" sage -t -force_lib "devel/sage/sage/rings/polynomial/multi_polynomial_sequence.py" sage -t -force_lib "devel/sage/sage/plot/plot.py" # Time out sage -t -force_lib "devel/sage/sage/crypto/mq/mpolynomialsystem.py" sage -t -force_lib "devel/sage/sage/crypto/mq/sr.py" sage -t -force_lib "devel/sage/sage/schemes/elliptic_curves/ell_rational_field.py" # Time out sage -t -force_lib "devel/sage/sage/schemes/elliptic_curves/heegner.py" # Time out
I need to investigate it further.
comment:84 in reply to: ↑ 83 ; follow-up: ↓ 85 Changed 9 years ago by
Replying to SimonKing:
With the old version of the M4RI and M4RIE patches, I got some doctest errors on mark (which is solaris):
And that's since I am stupid!!! Again I forgot to install the dependencies!
comment:85 in reply to: ↑ 84 Changed 9 years ago by
Replying to SimonKing:
And that's since I am stupid!!! Again I forgot to install the dependencies!
Or perhaps not. The new polybori spkg from #11261 is installed, and so is the new M4RI spkg and the corresponding patches.
I have to leave office now. But it seems that either #11574 or this ticket are causing the problem.
comment:86 follow-up: ↓ 87 Changed 9 years ago by
Can you touch sage/libs/polybori/decl.pxd
, run sage -b
and try those again?
comment:87 in reply to: ↑ 86 Changed 9 years ago by
comment:88 follow-up: ↓ 95 Changed 9 years ago by
- Status changed from needs_review to needs_work
Finally, the tests on 32 bit solaris (mark on skynet) are finshed. They passed, modulo the usual timeouts. The long tests pass on my machine. If I understand the discussion above, it builds on 64 bit solaris and on Cygwin as well, and of course on linux and os x.
Moreover, it provides a very impressive speedup compared with old Sage matrices over GF(2^{e}). If I am not mistaken, it is even faster than Magma.
So, it is almost a positive review.
But there remain a few things to do.
- David Kirkby noted that we are adding a new standard spkg here. So, there should be a voting on sage-devel. So, that's "needs info".
- I found some issues with the "randomize" methods.
i) The randomize method in sage/libs/ntl/ntl_mat_GF2E.pyx is not documented (thus, also has no tests), and it lacks the usual optional arguments
density
andnonzero
.
ii) The doc of Matrix_mod2e_dense.randomize gives no use cases for the optional arguments. Actually, the behaviour when passing the optional arguments is clearly not what we want:
sage: MS = MatrixSpace(GF(64,'a'),100,100) sage: A = MS.random_element(densitiy=0.3) sage: A.density() 4913/5000 sage: B = MS.random_element(nonzero=True) sage: B.density() 2469/2500 sage: C = MS.random_element(nonzero=True, density=1) sage: C.density() 2461/2500 sage: D = MS.random_element(density=0) sage: D.density() 983/1000
- The doc of many methods needs some polishing.
- It would be nice to have a reference to research articles. I had never heard of Travolta tables before. Strassen-Winograd and Karatsuba may be better known, but still, a reference could help.
- _matrix_times_matrix is not documented.
__init__
,__cinit__
do not specify their arguments.
_lmul_
,__neg__
,__richcmp__
,__invert__
,__reduce__
,set_unsafe
andget_unsafe
do not state what they do, and the method names do not occur in the example (so, it should be marked as an indirect doctest).
- cdef'd methods such as rescale_row_c or add_multiple_of_row_c or swap_rows_c and so on may not be as easily visible by the user than Python methods. However, as a courtesy to developers who actually read the source file, the arguments of those methods should be specified.
- "echelonize" should state what it does and what its optional arguments are. If I am not mistaken, it changes the matrix inplace, and that should be documented.
- When I read "Classical cubic matrix multiplication.", I first understood that the matrix is cubic. But perhaps I'm a bit square here...
So, that's "needs work" for now.
Since there should be a vote on sage-devel anyway, we might also ask whether it should be included in the docs. I am still not sure: On the one hand, I think a reference manual should be thorough. On the other hand, all "new" methods that are no cdef'd methods and do not start with an underscore overwrite methods from super-classes that are documented elsewhere.
comment:89 Changed 9 years ago by
For the record: I asked for a vote on sage-devel.
comment:90 in reply to: ↑ 45 Changed 9 years ago by
Replying to malb:
Replying to kcrisman:
I agree that it is very annoying to have two things inside the same spkg. There is, for instance, an open ticket to remove rpy2 from the r spkg.
Well, it makes it easier to re-use tuning results. For example, I plan to run the cache tuning only once, i.e. for M4RI and M4RIE would re-use the results.
Save them to some file in, say, $SAGE_ROOT/local/share/m4ri/
?
comment:91 follow-up: ↓ 93 Changed 9 years ago by
P.S.: This would perhaps even allow specifying cache sizes manually, though the proper way would be something like
$ export M4RI_EXTRA_OPTS="--L1-cache-size=32 --L2-cache-size=1024" # passed to M4RI[E]'s 'configure' $ sage -i m4rie
(And similar options for other cache parameters like the cache line sizes.)
Specifying these may at least speed up the tuning (by omitting a lot of tries); don't know if you could skip tuning, given these, in whole.
Storing and using tuning parameters analoguous to GMP's gmp-mparams.h
(including defaults for a couple of platforms / processors) wouldn't be bad either; then [optionally] bypassing self-tuning would really make sense.
[Disclaimer: Haven't looked at the spkg at all, so there might be better or other ways to achieve this.]
comment:92 Changed 9 years ago by
- Work issues changed from Improve scalar multiplication to Fix some docs and fix "randomize()"
comment:93 in reply to: ↑ 91 Changed 9 years ago by
Replying to leif:
[Disclaimer: Haven't looked at the spkg at all, so there might be better or other ways to achieve this.]
Looks like you're currently not trying to query any of
_SC_LEVEL1_ICACHE_SIZE _SC_LEVEL1_ICACHE_ASSOC _SC_LEVEL1_ICACHE_LINESIZE _SC_LEVEL1_DCACHE_SIZE _SC_LEVEL1_DCACHE_ASSOC _SC_LEVEL1_DCACHE_LINESIZE _SC_LEVEL2_CACHE_SIZE _SC_LEVEL2_CACHE_ASSOC _SC_LEVEL2_CACHE_LINESIZE _SC_LEVEL3_CACHE_SIZE _SC_LEVEL3_CACHE_ASSOC _SC_LEVEL3_CACHE_LINESIZE _SC_LEVEL4_CACHE_SIZE _SC_LEVEL4_CACHE_ASSOC _SC_LEVEL4_CACHE_LINESIZE
(with sysconf()
). At least some of these values aren't available on every platform, but if they are, you could perhaps use them.
AFAIK you'd have to set some feature test macro before including unistd.h
on Solaris, although I doubt it supports any of the above.
comment:94 Changed 9 years ago by
- there's an optional parameter
--with-cachesize
for M4RI's configure which allows to specify L1 and L2. - However, this is not exported to Sage, i.e. M4RI_EXTRA_OPTS does not exist yet.
- M4RIE does not re-tune any more but re-uses the data from M4RI.
- In my experience (on i7s for example) tuning is much better than using the data reported by the CPU. That is, if one trusts L1 and L2 as reported by the CPU (which are correct) the code is much slower than tuning which essentially uses L3 instead of L2.
comment:95 in reply to: ↑ 88 Changed 9 years ago by
Replying to SimonKing:
- I found some issues with the "randomize" methods.
i) The randomize method in sage/libs/ntl/ntl_mat_GF2E.pyx is not documented (thus, also has no tests), and it lacks the usual optional arguments
density
andnonzero
.
Okay, I'll take a look. Note that ntl_mat_GF2E
does not have the same interface as normal matrices though. But it makes sense to make it consistent where possible.
ii) The doc of Matrix_mod2e_dense.randomize gives no use cases for the optional arguments. Actually, the behaviour when passing the optional arguments is clearly not what we want:
Okay, I'll take a look.
- The doc of many methods needs some polishing.
- It would be nice to have a reference to research articles. I had never heard of Travolta tables before. Strassen-Winograd and Karatsuba may be better known, but still, a reference could help.
There is no research article on Travolta tables yet since I made them up for M4RIE. But for the other ones I can add references.
- _matrix_times_matrix is not documented.
This is an internal method which is part of the standard matrix interface. I'd say it is hence understood what it does.
__init__
,__cinit__
do not specify their arguments.
Okay.
_lmul_
,__neg__
,__richcmp__
,__invert__
,__reduce__
,set_unsafe
andget_unsafe
do not state what they do, and the method names do not occur in the example (so, it should be marked as an indirect doctest).
These are special methods where either Python or our Matrix classes define what they do. I'd say it is hence understood what they do.
- cdef'd methods such as rescale_row_c or add_multiple_of_row_c or swap_rows_c and so on may not be as easily visible by the user than Python methods. However, as a courtesy to developers who actually read the source file, the arguments of those methods should be specified.
Okay.
- "echelonize" should state what it does and what its optional arguments are. If I am not mistaken, it changes the matrix inplace, and that should be documented.
Okay.
- When I read "Classical cubic matrix multiplication.", I first understood that the matrix is cubic. But perhaps I'm a bit square here...
So, that's "needs work" for now.
Since there should be a vote on sage-devel anyway, we might also ask whether it should be included in the docs. I am still not sure: On the one hand, I think a reference manual should be thorough. On the other hand, all "new" methods that are no cdef'd methods and do not start with an underscore overwrite methods from super-classes that are documented elsewhere.
I'd much rather add a note to the reference manual which lists which library drives which base field?
comment:96 Changed 9 years ago by
The dist/
(Debian) directory can and should be deleted (see #5903).
comment:97 Changed 9 years ago by
Agreed!
comment:98 Changed 9 years ago by
- Description modified (diff)
- Status changed from needs_work to needs_review
The updated patch + updated SPKG should address the comments above. Simon, as an added bonus the new SPKG also contains the faster scalar product.
comment:99 Changed 9 years ago by
doctests pass on sage.math btw.
comment:100 Changed 9 years ago by
- Status changed from needs_review to needs_info
I did sage -f http://sage.math.washington.edu/home/malb/spkgs/libm4rie-20110821.spkg
, I removed the old version of m4rie_for_sage.patch
, qimported the new one, checked that my computer was really downloading the new version, did qpush, and sage -br
.
However, I get
sage: MS = MatrixSpace(GF(64,'a'),800,800) sage: A = MS.random_element(densitiy=1) sage: RR(A.density()) 0.984492187500000 sage: A = MS.random_element(densitiy=0.001) sage: RR(A.density()) 0.984167187500000
I verified (by A.randomize?
) that the new patch is applied.
Do you have any clue why that happened?
comment:101 Changed 9 years ago by
Strange enough, the doc tests for matrix_mod2e_dense pass.
So, perhaps MS.random_element(...)
is calling randomize(...)
in a wrong way.
comment:102 Changed 9 years ago by
- Status changed from needs_info to needs_review
Yes, it is passing the wrong arguments, due to a typo on my end: I wrote densitiy, not density). Sorry.
comment:103 follow-ups: ↓ 104 ↓ 105 Changed 9 years ago by
- Status changed from needs_review to needs_work
Questions / To do
- The hg repository of the spkg needs being updated. Even though the last log entry is of today (August 22), the status report is
$ hg status ? config.log ? m4rie/.hgignore ? m4rie/.hgtags ? m4rie/AUTHORS ? m4rie/COPYING ? m4rie/ChangeLog ? m4rie/INSTALL ? m4rie/Makefile.am ? m4rie/Makefile.in ? m4rie/NEWS ? m4rie/README ? m4rie/aclocal.m4 ? m4rie/bench/Makefile.am ? m4rie/bench/Makefile.in ? m4rie/bench/bench_elimination.cc ? m4rie/bench/bench_multiplication.cc ? m4rie/bench/bench_smallops.cc ? m4rie/bench/benchmarking.cc ? m4rie/bench/benchmarking.h ? m4rie/bench/cpucycles-20060326/alpha.c ? m4rie/bench/cpucycles-20060326/alpha.h ? m4rie/bench/cpucycles-20060326/amd64cpuinfo.c ? m4rie/bench/cpucycles-20060326/amd64cpuinfo.h ? m4rie/bench/cpucycles-20060326/amd64tscfreq.c ? m4rie/bench/cpucycles-20060326/amd64tscfreq.h ? m4rie/bench/cpucycles-20060326/clockmonotonic.c ? m4rie/bench/cpucycles-20060326/clockmonotonic.h ? m4rie/bench/cpucycles-20060326/compile ? m4rie/bench/cpucycles-20060326/cpucycles.html ? m4rie/bench/cpucycles-20060326/do ? m4rie/bench/cpucycles-20060326/do.notes ? m4rie/bench/cpucycles-20060326/gettimeofday.c ? m4rie/bench/cpucycles-20060326/gettimeofday.h ? m4rie/bench/cpucycles-20060326/hppapstat.c ? m4rie/bench/cpucycles-20060326/hppapstat.h ? m4rie/bench/cpucycles-20060326/powerpcaix.c ? m4rie/bench/cpucycles-20060326/powerpcaix.h ? m4rie/bench/cpucycles-20060326/powerpclinux.c ? m4rie/bench/cpucycles-20060326/powerpclinux.h ? m4rie/bench/cpucycles-20060326/powerpcmacos.c ? m4rie/bench/cpucycles-20060326/powerpcmacos.h ? m4rie/bench/cpucycles-20060326/sparc32psrinfo.c ? m4rie/bench/cpucycles-20060326/sparc32psrinfo.h ? m4rie/bench/cpucycles-20060326/sparcpsrinfo.c ? m4rie/bench/cpucycles-20060326/sparcpsrinfo.h ? m4rie/bench/cpucycles-20060326/test.c ? m4rie/bench/cpucycles-20060326/x86cpuinfo.c ? m4rie/bench/cpucycles-20060326/x86cpuinfo.h ? m4rie/bench/cpucycles-20060326/x86tscfreq.c ? m4rie/bench/cpucycles-20060326/x86tscfreq.h ? m4rie/config.guess ? m4rie/config.sub ? m4rie/configure ? m4rie/configure.ac ? m4rie/depcomp ? m4rie/gf2e_cxx/finite_field_givaro.h ? m4rie/install-sh ? m4rie/ltmain.sh ? m4rie/m4/ax_check_compiler_flags.m4 ? m4rie/m4/ax_openmp.m4 ? m4rie/m4/libtool.m4 ? m4rie/m4/ltoptions.m4 ? m4rie/m4/ltsugar.m4 ? m4rie/m4/ltversion.m4 ? m4rie/m4/lt~obsolete.m4 ? m4rie/missing ? m4rie/tests/test_elimination.cc ? m4rie/tests/test_multiplication.cc ? m4rie/tests/test_smallops.cc ? m4rie/tests/testing.h
- spkg-check does not seem to work. After opening the package and working in a sage shell, I get
$ ./spkg-check ./spkg-check: Zeile 30: cd: build/m4ri: Datei oder Verzeichnis nicht gefunden Testing the M4RI library make: *** Keine Regel, um »check« zu erstellen. Schluss. Error testing M4RI
- Have you already implemented the ideas for storing the cache size information, so that installation of the package does not take so long? Or will that be in a future release?
- You said you'd much rather add a note to the reference manual which lists which library drives which base field. Did you do so? I can not find it in the patch.
Done
- The long doctests in doc/ and sage/ pass on my machine. The doctests in sage/matrix pass on 32 bit solaris.
- Randomize seems to work fine.
- Scalar multiplication has improved:
sage: MS = MatrixSpace(GF(64,'a'),5000,5000) sage: K = MS.base_ring() sage: c = K.random_element() sage: A = MS.random_element() sage: %time A*c CPU times: user 0.05 s, sys: 0.01 s, total: 0.06 s Wall time: 0.06 s 5000 x 5000 dense matrix over Finite Field in a of size 2^6 sage: MS1 = MatrixSpace(GF(64,'a'),1,5000) sage: B = MS1.random_element() sage: %timeit B*c 625 loops, best of 3: 51.5 µs per loop
- The docs are now almost fine from my point of view. We only have
$ sage -coverage sage/matrix/matrix_mod2e_dense.pyx ---------------------------------------------------------------------- sage/matrix/matrix_mod2e_dense.pyx SCORE sage/matrix/matrix_mod2e_dense.pyx: 100% (27 of 27) Possibly wrong (function name doesn't occur in doctests): * Matrix_mod2e_dense _multiply_travolta(Matrix_mod2e_dense self, Matrix_mod2e_dense right): * Matrix_mod2e_dense _multiply_karatsuba(Matrix_mod2e_dense self, Matrix_mod2e_dense right): * Matrix_mod2e_dense _multiply_strassen(Matrix_mod2e_dense self, Matrix_mod2e_dense right, cutoff=0): * ModuleElement _lmul_(self, RingElement right): ----------------------------------------------------------------------
and I suggest that I add a referee patch that simply adds a "indirect doctest" in the appropriate places.
Conclusion
I put it "needs work", but that's only since I'd like to create a referee patch. If you did not add the information on which library drives which base field yet, I can do so as well.
Please check in the changes to the hg repository, or "hg ignore" them. Please fix the self tests (or tell me why my attempt to call spkg-check did not work).
comment:104 in reply to: ↑ 103 Changed 9 years ago by
Replying to SimonKing:
I put it "needs work", but that's only since I'd like to create a referee patch.
Sorry, I did a mistake when editing. I wrote that line before noticing the spkg-check and hg status issues. These need fixes as well.
comment:105 in reply to: ↑ 103 ; follow-up: ↓ 106 Changed 9 years ago by
Replying to SimonKing:
- spkg-check does not seem to work. After opening the package and working in a sage shell, I get
$ ./spkg-check ./spkg-check: Zeile 30: cd: build/m4ri: Datei oder Verzeichnis nicht gefunden Testing the M4RI library make: *** Keine Regel, um »check« zu erstellen. Schluss. Error testing M4RI
I guess you'll first have to build, which also creates the directory and the Makefile there.
Did you try env SAGE_CHECK=yes sage -f ...
?
If you've run sage -i
also with -s
, entering the build directory and doing ./spkg-check
should also work.
comment:106 in reply to: ↑ 105 ; follow-up: ↓ 107 Changed 9 years ago by
Replying to leif:
I guess you'll first have to build, which also creates the directory and the Makefile there.
Did you try
env SAGE_CHECK=yes sage -f ...
?
I tried now, but it did not work.
If you've run
sage -i
also with-s
, entering the build directory and doing./spkg-check
should also work.
Well, I learnt that ./spkg-check
should work if you have simply have opened the package and use a sage shell.
comment:107 in reply to: ↑ 106 Changed 9 years ago by
Replying to SimonKing:
Well, I learnt that
./spkg-check
should work if you have simply have opened the package and use a sage shell.
And export SAGE_CHECK="yes"
followed by sage -f
should work as well. But it doesn't
comment:108 Changed 9 years ago by
- Description modified (diff)
comment:109 Changed 9 years ago by
Replying to SimonKing:
- The hg repository of the spkg needs being updated. Even though the last log entry is of today (August 22), the status report is
Agreed & fixed in the SPKG I'll update in a minute.
- spkg-check does not seem to work. After opening the package and working in a sage shell, I get
Fixed.
- Have you already implemented the ideas for storing the cache size information, so that installation of the package does not take so long? Or will that be in a future release?
I only implemented that M4RIE avoids detecting the cache size. It has nothing to do with cache size detection any more it just uses M4RI's results.
- You said you'd much rather add a note to the reference manual which lists which library drives which base field. Did you do so? I can not find it in the patch.
We should open a new ticket for this?
- The docs are now almost fine from my point of view. We only have
and I suggest that I add a referee patch that simply adds a "indirect doctest" in the appropriate places.
I'll do that.
comment:110 Changed 9 years ago by
- Status changed from needs_work to needs_review
SPKG + patch updated.
comment:111 Changed 9 years ago by
- Keywords sd32 added
comment:112 follow-up: ↓ 113 Changed 9 years ago by
- Status changed from needs_review to needs_info
m4rie_for_sage.patch did not apply on top of my private patch chain. The conflict is with #4260. So, one should be made dependent on the other.
Martin, what seems to be more stable? This M4RIE patches? Or the Linbox patches?
comment:113 in reply to: ↑ 112 ; follow-up: ↓ 114 Changed 9 years ago by
- Status changed from needs_info to needs_review
Replying to SimonKing:
m4rie_for_sage.patch did not apply on top of my private patch chain.
Ouch, how unfortunate. I guess it's in matrix_space.py
?
The conflict is with #4260. So, one should be made dependent on the other.
I'm not so sure about this. They are fairly independent. I suggest each is reviewed independently and whichever gets in later is re-based accordingly.
Martin, what seems to be more stable? This M4RIE patches? Or the Linbox patches?
I'd say M4RIE:
- the M4RIE interface has been around for months and months
- I'm willing to debug any issue that might come up because I'm quite invested in this ticket.
- Many people have looked at M4RIE over the time.
- The LinBox? switch-over has not received a single review yet.
- You reported some speed issues with the new LinBox? interface
- It is unclearly whether the new LinBox? interface works/builds on anything besides 64-bit Linux.
comment:114 in reply to: ↑ 113 Changed 9 years ago by
Replying to malb:
Replying to SimonKing:
m4rie_for_sage.patch did not apply on top of my private patch chain.
Ouch, how unfortunate. I guess it's in
matrix_space.py
?
Sure. It is when the __matrix_class
is determined. And of course I have already modified the patches in my private patch chain, so that I can continue to work...
I'm not so sure about this. They are fairly independent. I suggest each is reviewed independently and whichever gets in later is re-based accordingly.
Makes sense.
comment:115 Changed 9 years ago by
- Description modified (diff)
The SPKG is here:
http://sage.math.washington.edu/home/malb/spkgs/libm4ri-20100730.spkg