[libm4rie] annotated tag upstream/20150908 created (now bdc523d)
Julien Puydt
julien.puydt at laposte.net
Sat May 21 20:14:36 UTC 2016
This is an automated email from the git hooks/post-receive script.
jpuydt-guest pushed a change to annotated tag upstream/20150908
in repository libm4rie.
at bdc523d (tag)
tagging 7a6cde28c221c96b1423e7984bcb6766bb37c22d (commit)
tagged by Julien Puydt
on Sat May 21 20:47:55 2016 +0200
- Log -----------------------------------------------------------------
Upstream version 20150908
Cédric Boutillier (1):
Imported Upstream version 20140914
Julien Puydt (1):
Imported Upstream version 20150908
Martin Albrecht (190):
starting M4RIE library for linear algebra over GF(2^n) for small n.
added stubs for functions that need to be implemented
implemented two tables in mzed_echelonize_travolta()
implemented 6 grease tables for elimination
added top level echelonize command which picks an algorithm
adding fast code for add_multiple_of_row for 16-bit wide words
adding configure in order to be able to test on systems with old autoconf
same for install-sh
do not enforce autoconf version
using M4RI tuning parameters in mzed_echelonize_travolta
implemented most of the trivial functions
adding Makefile, ignoring configure
using four travolta tables in multiplication now
naive multiplication for comparison
added test code for multiplication, implemented more small functions
implemented add_multiple_of_row for width == 2
implemented cache optimisation and disabled it in most cases since it is slower in general
peform slightly less work in mzed_make_table (avoid 0 and 1)
fixing a bug in r16 where L wasn't updated
fixing memory leaks in test code
adding libm4ri as a dependency for libm4rie
Cygwin requires no-undefined (and we have no undefined symbols)
implemented strassen multiplication
adding tuning parameters for strassen
more robust cache tuning by increasing the number of trials
adapted Strassen cutoff to use cache sizes
implemented Karatsuba multiplication for bitsliced representations of dense matrices over GF(2^2)
some refactoring and making functions more readable (hopefully)
performance improvement: don't slice/cling words which are zero
use Karatsuba for GF(2^2) it is always better
improved speed of cache tuning seems to give good results on prai243,road,eno,iras,redhawk,t2
cleaned up build targets (especially bench/)
new release
Added tag release-20100817 for changeset 27f53a7570aa
fixing two issues with the building of bench_elimination
disable default building of benchmarking binaries, they cause problems
adapted to work with recent M4RI bit order reversal
adapted to work with hg tip of M4RI
imported updated cache tuning from M4RI
remove cache size detection and SSE detection and rely on M4RI
Added tag release-20110715 for changeset 05db9d68bf61
Added tag release-20110601 for changeset f2b76762cd70
mzd_slice_t datatype and first utility functions
mzd_slice4 for slicing degree 3 and 4
revised configure & makefiles to allow --with-m4ri and --with-givaro (needs more testing)
revised benchmarketing code to be easier to build
adding bench_smallops (for now: slicing and clinging)
force rpath to the right M4RIE
10% or so faster slicing
slicing & clinging for k=2,3,4
cosmetic stuff
more cosmetic stuff
and yet more whitespace stuff
improved memory consumption of karatsuba over GF(4)
adapted hierarchy of karatsuba functions in preparation for karatsuba for GF(4) and GF(8)
karatsuba for GF(8)
switching size_t to rci_t
fixed addmul!
better tests + bugfixes for multiplication
new function mzed_mul_scalar which performs scalar scaling of a matrix (fixes #9)
faster smallops tests
Added tag release-20110821 for changeset fe6dab8a9bb4
Removed tag release-20110821
fix a typo mzed_mul_scalar (fixes #10)
Added tag release-20110821 for changeset dd06c345824f
fixed compiler warnings about wrong types
karatsuba multiplication for 2^4 and less memory consumption for 2^3
improved memory consumption for GF(2^4) Karatsuba from 9 to 7 temporaries
better benchmarketing
TRSM lower left for mzed_t: X = ~L * B
trsm lower_left templated (for mzd_slice_t and mzed_t)
added mzed_add to bench_smallops
better benchmarkteting for trsm
refactored templating stuff for easier re-use
fixed but in add multiple of rows
more fixes to mzed_add_multiple_of_row()
better test_elimination
implemented naive PLE decomposition
documentation updated
adding canary (cf., stack protection) to detect buffer overfows (i.e., algorithms writing past ncols)
asymptotically fast PLE decomposition + travolta PLE
faster mzed_rescale_row
better PLE tests
merge
documentation update
bench_ple added
better help for bench_ple/bench_trsm
better complexity feedback + fixed bug in _mzed_mul_karatsuba
gf2e_t16_init()/free() for construction/destruction of 16-bit multiplication tables
TRSM upper left
PLUQ, better support for offsets, much faster TRSM
faster trsm upper left + better PLE crossovers
slightly faster mzed_make_table
performance enhancements for ple
fixed ple cutoff
slightly more sensible benchmarking Makefile
use m4ri_mm_malloc/free instead of malloc/free
TRSM lower left with Travolta tables
bugfix in mzed_add_multiple_of_row
mzed_echelonize_ple() implemented and passes tests
travolta table creation in k MUL + 2^k ADD instead of 2^k MUL
use ple in echelonform
don't use bitmask begin in clinging/slicing: it didn't work properly, we don't need it, it costs cycles
ignore more stuff like Makefiles
slicing that isn't totally naive (clinging yet to do)
matrix_trsm_upper_left_travolta
allow type in bench_elimination
define mzd_slice_echelonize (alias for mzd_slice_echelonize_ple)
allow type argument in bench_multiplication
removed old cold & renamed some internal functions for slicing
clinging (2-4) that doesn't totally suck
DNDEBUG in bench
whitespace changes for more readability
more compact slicing/clinging
copy less in echelonform.c
pass M4RI's SIMD and OpenMP flags to compiler
add a test for one particular if branch + asserts
changing version number for upcoming release
simple re-arrangement
allow benching for trsm_upper_left
played with the cutoffs for nicer scaling
Added tag release-20111004 for changeset d1d724d9545a
table creation got cheaper and hence blocks make sense now
slicing/clinging for e<=8
refactored: clinging/slicing in conversion
naive polynomial multiplication for bitsliced matrices
Karatsuba for e=5
Karatsuba for deg 8
unified strassen cutoff
use rpath for test_* to make sure the right version of the library is used
Karatsuba for GF(2^6)
saving memory in karatsuba multiplication
Karatsuba for GF(2^7), i.e. for all e <= 8
using less temporary memory for Karatsuba for GF(2^4)
added more documentation and first steps toward Travolta => Newton-John
wrote a lot of documentation
mzd_slice_mul_scalar
adding a softlink m4rie -> src
remove -lntl, it's not needed
replaced nonstandard __STRING(x) with #x
degrees are small, hence ints
avoid new/delete in testing code (caused problems on cicero)
fixed memleak in test_smallops
bitmasks should be unsigned
fixed a memleak
make sure the canary does not set bits which are expected to be zero (e.g. in lockup tables)
this prevents a segfault on cicero with GCC 4.6.1 (and I'm too lazy right now to properly debug this)
changed library version to 20111203
Added tag release-20111203 for changeset e6bd45516c69
refactoring: renamed files to match names of main structs in them
more compact docstrings & fixed single include directives
refactored: mzd_poly added (for now internal-ish stuff)
fixed includes to work with development version of M4RI
use unsigned ints for degrees everywhere
removed a bit of dead code scan-build reported
Added tag release-20120415 for changeset 8245cd046d86
reserve enough pointers for finite fields in testsuite
Removed tag release-20120415
Added tag release-20120415 for changeset 83b1a73e827a
Removed tag release-20120415
Added tag release-20120415 for changeset 4c2e04ec11a4
don't divide by arbitrary constants in cycle count
be more indifferent about Givaro versions
changed all includes to <m4rie/blah.h> so that -I$INCLUDEDIR is enough
continuation of previous patch (forgot to save files)
library version updated to 20120415
I am stupid and I don't close by brackets
preparation for next (bugfix) release
use autotools' release mechanisms to avoid release fuck-up's
don't use rpath in TESTS on Makefile.am, Solaris chokes on this.
"make dist" should produce something that works now
"make check" now works with "make dist" created tarballs
added gf2e_init(minpoly)
small bugfixes
removed dependency on Givaro for finite field creation
dropping inversion tables from gf2e
added gf2e_mul() and started moving functions over
moved to ff->mul from explicit multiplication tables
added first tests for degrees > 10
more tests for degrees > 10 and <= 16
avoid mzd_set_ui() in table creation
conversion mzed_t <=> mzd_slice_t for up to degree 16
fixed typos
fixed typo in comment
added naive polynomial multiplication
updating library version 20121224
Added tag release-20121224 for changeset 621cc82cee2a
split conversion into several files to make it easier to compile M4RIE on machines with low RAM
preparing for upcoming release
Added tag release-20130416 for changeset 749b26c9affd
-----------------------------------------------------------------------
No new revisions were added by this update.
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/debian-science/packages/libm4rie.git
More information about the debian-science-commits
mailing list