Changes between ecm-6.2.3 and ecm-6.3:
* New assembly code for 64-bit PowerPC (thanks to Philip McLaughlin)
* Fixed incompatibility with GMP 5.0.0
* Fixed several bugs, and now check return value from malloc() calls
Changes between ecm-6.2.2 and ecm-6.2.3:
* Fixed incompatibility with GMP 4.3.0 when testing version in configure
* SSE2 asm code for Visual C added in stage 2 NTT code
* Small improvement to x86_64 mulredc asm code, slight speedup on Core 2
* Fixed incorrect carry propagation in subquadratic REDC code which
could lead to incorrect arithmetic in rare cases
* Fixed memory leak with -v parameter when factor was found in ECM stage 1
* Fixed bug which caused only one ECM curve to be run in spite of -c
parameter if input line did not end in newline
* Assembler mulredc code enabled by default on x86_64
Changes between ecm-6.2.1 and ecm-6.2.2:
* Updated build project files for Visual C by Brian Gladman, also adds
missing NTT_GFP_TWIDDLE_DI[FT]_BREAKOVER defines in VC parameter file
* Fixed uninitialised parameter to P-1 probability computation
* In tune.c : fixed generation of NTT_GFP_TWIDDLE_DI[FT]_BREAKOVER values,
avoid calling cputime() excessively often when timing short functions,
fixed access to uninitialised memory
* Fixed serious split infinitive in configure script (thanks Paul Leyland)
* Removed unnecessary carry propagation in x86_64 mulredc code, slight
speedup (thanks Philip McLaughlin)
* Fixed non-portable PIC code in x86_64/redc.asm
* Fixed problem with pattern matching host type names in configure.in
* Converted binary constants in spv.c and ntt_gfp.c to hexadecimal,
some assembler do not support binary constants
Changes between ecm-6.2 and ecm-6.2.1:
* Default B2 for new P-1 and P+1 stage 2 increased
* Probabilities for finding factors with P-1 are now printed with -v
* Fixed compilation problem on IA64, EV56, and ARM
* Made threshold between recursive and iterative NTT tunable
Changes between ecm-6.1.3 and ecm-6.2:
* New stage 2 for P-1 and P+1, described in Montgomery and Kruppa,
Improved Stage 2 to P+-1 Factoring Algorithms,
in A. J. van der Poorten and A. Stein (Eds.), ANTS-VIII 2008,
LNCS 5011, pp. 180-195.
* Parallelization in the new P+-1 stage 2 (with --enable-openmp).
* Optimizations to the NTT code by Jason S. Papadopoulos
* Improved mulredc assembly code for Athlon64/Opteron
* Improved modular reduction in the mpzmod range
* Bugfix in P+1 stage 2 which caused incorrect initialisation if
Brent-Suyama polynomial had degree > 1 and i0 was negative (occurs only
with non-standard parameters)
* Bugfix in generation of Lucas chains for P+1 and ECM, causing some stage 1
primes close to 2^32 to be processed incorrectly on 32 bit systems
* Added build project for VC++ by Brian Gladman
* File ecm.h changed from GPL to LGPL: the fact it was under GPL was an
unvoluntary mistake, which has the consequence that applications linking with
libecm for version < 6.2 should be under GPL too.
* Fixed a regression introduced in 6.1.1: the default arithmetic (NTT) for
stage 2 was slower for large inputs. Now defaults to -no-ntt for input
numbers >30 machine words.
Changes between ecm-6.1.2 and ecm-6.1.3:
* fixed incorrect computation of memory use in stage 2, especially for
machines that use Kronecker-Schoenhage multiplication even for large
degrees, such as Core 2.
* fixed -B2scale option whose value hadn't been passed to the factoring
routines
* fixed default B2min for P-1, which could be truncated on 32 bit machines,
causing stage 2 to take a little longer than necessary
* fixed bug for modular multiplication modulo Fermat numbers 2^2^n+1, where
a result of 2^2^n would be truncated to 0.
Changes between ecm-6.1.1 and ecm-6.1.2:
* changed copyright header from sp.h, to recognize the FSF for parts inspired
or taken from gmp-impl.h.
Changes between ecm-6.0.1 and ecm-6.1:
* new assembly code contributed by Pierrick Gaudry for combined mul/redc
* new Number Theoretic Transform code contributed by Dave Newman for step 2
* new signal handling and corresponding save files for step 1
* now prints peak memory allocation with -v
* improved and simplified tuning
* command-line options: added -idlecmd -no-ntt -prpcmd -stage1time -maxmem,
removed -prp*
* new configure options --enable-asm-redc, --with-gwnum
* new exit status codes of ecm program
* new interface to George Woltman's GWNUM library (see INSTALL)
* chosen stage 2 bound is now printed right away
* fixed minor memory leak in mpmod.c/isbase2
* fixed invalid Found input number N reported on some numbers
* fixed serious P+1 bug on 64-bit architectures with B1 > 2^32
* fixed wrong detection of divisors of 2^n+1 or 2^n-1
* fixed memory leaks
Changes between ecm-6.0 and ecm-6.0.1:
* now checks for availability of snprintf() during configure
* fixed linking problems with tune and tune2 on PowerPC G5
* fixed segfault in rho.c
* fixed main()'s B2 value being overwritten by callees
* allow both \r and \n for newline (for Apple computers)
* made files compile under Visual C
* fixed bug in listz.c that could leave undefined data
* fixed the -B2scale option
* fixed small error in printed B2' value (with -v)
* added Windows section to INSTALL
* small corrections to ecm.xml and ecm.1
* added curve counter in loop mode again
* fixed segfault when a non-number was in place of B2 on command line
* worked around problem with MinGW/Wine scanf() (value 1 too high for %n)
* free rhotable memory at end of stage 2
* replaced GSL's dilog_series() in rho.c due to licensing (GPL vs. LGPL)
Changes between ecm-5.0.3 and ecm-6.0:
* use of the autotools (configure/make)
* there is now a documentation in "man" format (ecm.1)
* added a set of -prp* command line switches which will use an external
program to perform prp testing of candidate and factors. When the
numbers get large, GMP becomes very non-optimal in PRP testing.
An external program, such as OpenPFGW, can be much faster than
the GMP. (New code from Phil Carmody)
* new parser for symbolic input
* added hex number input into the expression parser. This was needed
due to save files from Prime95 being output in hex. ecm was not
resuming these.
* added some porting code so ecm builds under VC6. VC6 builds an ecm
that is is about 1% to 3% faster than the MinGW build (which is
about 1% to 3% faster than a Cygwin build)
* quiet mode (-q) now prints on stdout all factors found on the same line:
f1 f2 ... fk ccc
where ccc is the remaining composite (contributed by Laurent Fousse).
Example:
$ echo 438573459834757 | ./ecm -sigma 6 -q 1e2
2166151 202466707
* special code for Fermat numbers, which improves both steps, example for F12:
GMP-ECM 5.0.3 [powered by GMP 4.1.4] [ECM]
Input number has 1187 digits
Using B1=100000, B2=31565866, polynomial x^2, sigma=4155936925
Step 1 took 27157ms
Step 2 took 20830ms
GMP-ECM 5.2.0 [powered by GMP 4.1.4] [ECM]
Input number has 1187 digits
Using B1=100000, B2=31565866, polynomial x^2, sigma=535125396
Step 1 took 24264ms
Step 2 took 6605ms
* speed improvement in step 2, especially for large B2 (here with the c155
in the GMP-ECM sources), using Kronecker/Schonhage multiplication:
GMP-ECM 5.0.3 [powered by GMP 4.1.4] [ECM]
Using B1=3000000, B2=4016636514, polynomial Dickson(12), sigma=2383768044
Step 1 took 62305ms
Step 2 took 45322ms
GMP-ECM 5.2.0 [powered by GMP 4.1.4] [ECM]
Using B1=3000000, B2=4016636514, polynomial Dickson(12), sigma=3595368442
Step 1 took 61824ms
Step 2 took 32989ms
* Brent-Suyama's extension now works for P+1 too
* trial division is available (option -t n)
* new options -n (low priority) and -nn (idle priority)
* the looping mode (-c n) now continues to try to factor the composite when
a factor is found (use -one to stop)
* new option -ve n to display only inputs of <= n characters (looping mode)
* new option -treefile which stores product tree of F on disk to save memory
* generation of roots use double sieve idea, increases B2 for given k, dF
* generation of roots for ECM rewritten to reduce number of extgcds
Changes between ecm-5.0 and ecm-5.0.1:
* fixed bug when B1 or B2min is too large, and an overflow occurs in step 2,
making the computations incorrect. The new limit for B1 or B2min is
now around 2^53, and an error occurs when this limit is passed.
* fixed problem with save lines incorrectly written when both sigma and A
are given with -save.
* fixed efficiency problem with -pm1: for B1 > 1e6, use -redc or -mpz_mod
by default.
