Edit
kc3-lang/libtommath/changes.txt
Tom St Denis
- changes.txt
-
February 12th, 2005 v0.34 -- Fixed two more small errors in mp_prime_random_ex() -- Fixed overflow in mp_mul_d() [Kevin Kenny] -- Added mp_to_(un)signed_bin_n() functions which do bounds checking for ya [and report the size] -- Added "large" diminished radix support. Speeds up things like DSA where the moduli is of the form 2^k - P for some P < 2^(k/2) or so Actually is faster than Montgomery on my AMD64 (and probably much faster on a P4) -- Updated the manual a bit -- Ok so I haven't done the textbook work yet... My current freelance gig has landed me in France till the end of Feb/05. Once I get back I'll have tons of free time and I plan to go to town on the book. As of this release the API will freeze. At least until the book catches up with all the changes. I welcome bug reports but new algorithms will have to wait. December 23rd, 2004 v0.33 -- Fixed "small" variant for mp_div() which would munge with negative dividends... -- Fixed bug in mp_prime_random_ex() which would set the most significant byte to zero when no special flags were set -- Fixed overflow [minor] bug in fast_s_mp_sqr() -- Made the makefiles easier to configure the group/user that ltm will install as -- Fixed "final carry" bug in comba multipliers. (Volkan Ceylan) -- Matt Johnston pointed out a missing semi-colon in mp_exptmod October 29th, 2004 v0.32 -- Added "makefile.shared" for shared object support -- Added more to the build options/configs in the manual -- Started the Depends framework, wrote dep.pl to scan deps and produce "callgraph.txt" ;-) -- Wrote SC_RSA_1 which will enable close to the minimum required to perform RSA on 32-bit [or 64-bit] platforms with LibTomCrypt -- Merged in the small/slower mp_div replacement. You can now toggle which you want to use as your mp_div() at build time. Saves roughly 8KB or so. -- Renamed a few files and changed some comments to make depends system work better. (No changes to function names) -- Merged in new Combas that perform 2 reads per inner loop instead of the older 3reads/2writes per inner loop of the old code. Really though if you want speed learn to use TomsFastMath ;-) August 9th, 2004 v0.31 -- "profiled" builds now :-) new timings for Intel Northwoods -- Added "pretty" build target -- Update mp_init() to actually assign 0's instead of relying on calloc() -- "Wolfgang Ehrhardt" <Wolfgang.Ehrhardt@munich.netsurf.de> found a bug in mp_mul() where if you multiply a negative by zero you get negative zero as the result. Oops. -- J Harper from PeerSec let me toy with his AMD64 and I got 60-bit digits working properly [this also means that I fixed a bug where if sizeof(int) < sizeof(mp_digit) it would bug] April 11th, 2004 v0.30 -- Added "mp_toradix_n" which stores upto "n-1" least significant digits of an mp_int -- Johan Lindh sent a patch so MSVC wouldn't whine about redefining malloc [in weird dll modes] -- Henrik Goldman spotted a missing OPT_CAST in mp_fwrite() -- Tuned tommath.h so that when MP_LOW_MEM is defined MP_PREC shall be reduced. [I also allow MP_PREC to be externally defined now] -- Sped up mp_cnt_lsb() by using a 4x4 table [e.g. 4x speedup] -- Added mp_prime_random_ex() which is a more versatile prime generator accurate to exact bit lengths (unlike the deprecated but still available mp_prime_random() which is only accurate to byte lengths). See the new LTM_PRIME_* flags ;-) -- Alex Polushin contributed an optimized mp_sqrt() as well as mp_get_int() and mp_is_square(). I've cleaned them all up to be a little more consistent [along with one bug fix] for this release. -- Added mp_init_set and mp_init_set_int to initialize and set small constants with one function call. -- Removed /etclib directory [um LibTomPoly deprecates this]. -- Fixed mp_mod() so the sign of the result agrees with the sign of the modulus. ++ N.B. My semester is almost up so expect updates to the textbook to be posted to the libtomcrypt.org website. Jan 25th, 2004 v0.29 ++ Note: "Henrik" from the v0.28 changelog refers to Henrik Goldman ;-) -- Added fix to mp_shrink to prevent a realloc when used == 0 [e.g. realloc zero bytes???] -- Made the mp_prime_rabin_miller_trials() function internal table smaller and also set the minimum number of tests to two (sounds a bit safer). -- Added a mp_exteuclid() which computes the extended euclidean algorithm. -- Fixed a memory leak in s_mp_exptmod() [called when Barrett reduction is to be used] which would arise if a multiplication or subsequent reduction failed [would not free the temp result]. -- Made an API change to mp_radix_size(). It now returns an error code and stores the required size through an "int star" passed to it. Dec 24th, 2003 v0.28 -- Henrik Goldman suggested I add casts to the montomgery code [stores into mu...] so compilers wouldn't spew [erroneous] diagnostics... fixed. -- Henrik Goldman also spotted two typos. One in mp_radix_size() and another in mp_toradix(). -- Added fix to mp_shrink() to avoid a memory leak. -- Added mp_prime_random() which requires a callback to make truly random primes of a given nature (idea from chat with Niels Ferguson at Crypto'03) -- Picked up a second wind. I'm filled with Gooo. Mission Gooo! -- Removed divisions from mp_reduce_is_2k() -- Sped up mp_div_d() [general case] to use only one division per digit instead of two. -- Added the heap macros from LTC to LTM. Now you can easily [by editing four lines of tommath.h] change the name of the heap functions used in LTM [also compatible with LTC via MPI mode] -- Added bn_prime_rabin_miller_trials() which gives the number of Rabin-Miller trials to achieve a failure rate of less than 2^-96 -- fixed bug in fast_mp_invmod(). The initial testing logic was wrong. An invalid input is not when "a" and "b" are even it's when "b" is even [the algo is for odd moduli only]. -- Started a new manual [finally]. It is incomplete and will be finished as time goes on. I had to stop adding full demos around half way in chapter three so I could at least get a good portion of the manual done. If you really need help using the library you can always email me! -- My Textbook is now included as part of the package [all Public Domain] Sept 19th, 2003 v0.27 -- Removed changes.txt~ which was made by accident since "kate" decided it was a good time to re-enable backups... [kde is fun!] -- In mp_grow() "a->dp" is not overwritten by realloc call [re: memory leak] Now if mp_grow() fails the mp_int is still valid and can be cleared via mp_clear() to reclaim the memory. -- Henrik Goldman found a buffer overflow bug in mp_add_d(). Fixed. -- Cleaned up mp_mul_d() to be much easier to read and follow. Aug 29th, 2003 v0.26 -- Fixed typo that caused warning with GCC 3.2 -- Martin Marcel noticed a bug in mp_neg() that allowed negative zeroes. Also, Martin is the fellow who noted the bugs in mp_gcd() of 0.24/0.25. -- Martin Marcel noticed an optimization [and slight bug] in mp_lcm(). -- Added fix to mp_read_unsigned_bin to prevent a buffer overflow. -- Beefed up the comments in the baseline multipliers [and montgomery] -- Added "mont" demo to the makefile.msvc in etc/ -- Optimized sign compares in mp_cmp from 4 to 2 cases. Aug 4th, 2003 v0.25 -- Fix to mp_gcd again... oops (0,-a) == (-a, 0) == a -- Fix to mp_clear which didn't reset the sign [Greg Rose] -- Added mp_error_to_string() to convert return codes to strings. [Greg Rose] -- Optimized fast_mp_invmod() to do the test for invalid inputs [both even] first so temps don't have to be initialized if it's going to fail. -- Optimized mp_gcd() by removing mp_div_2d calls for when one of the inputs is odd. -- Tons of new comments, some indentation fixups, etc. -- mp_jacobi() returns MP_VAL if the modulus is less than or equal to zero. -- fixed two typos in the header of each file :-) -- LibTomMath is officially Public Domain [see LICENSE] July 15th, 2003 v0.24 -- Optimized mp_add_d and mp_sub_d to not allocate temporary variables -- Fixed mp_gcd() so the gcd of 0,0 is 0. Allows the gcd operation to be chained e.g. (0,0,a) == a [instead of 1] -- Should be one of the last release for a while. Working on LibTomMath book now. -- optimized the pprime demo [/etc/pprime.c] to first make a huge table of single digit primes then it reads them randomly instead of randomly choosing/testing single digit primes. July 12th, 2003 v0.23 -- Optimized mp_prime_next_prime() to not use mp_mod [via is_divisible()] in each iteration. Instead now a smaller table is kept of the residues which can be updated without division. -- Fixed a bug in next_prime() where an input of zero would be treated as odd and have two added to it [to move to the next odd]. -- fixed a bug in prime_fermat() and prime_miller_rabin() which allowed the base to be negative, zero or one. Normally the test is only valid if the base is greater than one. -- changed the next_prime() prototype to accept a new parameter "bbs_style" which will find the next prime congruent to 3 mod 4. The default [bbs_style==0] will make primes which are either congruent to 1 or 3 mod 4. -- fixed mp_read_unsigned_bin() so that it doesn't include both code for the case DIGIT_BIT < 8 and >= 8 -- optimized div_d() to easy out on division by 1 [or if a == 0] and use logical shifts if the divisor is a power of two. -- the default DIGIT_BIT type was not int for non-default builds. Fixed. July 2nd, 2003 v0.22 -- Fixed up mp_invmod so the result is properly in range now [was always congruent to the inverse...] -- Fixed up s_mp_exptmod and mp_exptmod_fast so the lower half of the pre-computed table isn't allocated which makes the algorithm use half as much ram. -- Fixed the install script not to make the book :-) [which isn't included anyways] -- added mp_cnt_lsb() which counts how many of the lsbs are zero -- optimized mp_gcd() to use the new mp_cnt_lsb() to replace multiple divisions by two by a single division. -- applied similar optimization to mp_prime_miller_rabin(). -- Fixed a bug in both mp_invmod() and fast_mp_invmod() which tested for odd via "mp_iseven() == 0" which is not valid [since zero is not even either]. June 19th, 2003 v0.21 -- Fixed bug in mp_mul_d which would not handle sign correctly [would not always forward it] -- Removed the #line lines from gen.pl [was in violation of ISO C] June 8th, 2003 v0.20 -- Removed the book from the package. Added the TDCAL license document. -- This release is officially pure-bred TDCAL again [last officially TDCAL based release was v0.16] June 6th, 2003 v0.19 -- Fixed a bug in mp_montgomery_reduce() which was introduced when I tweaked mp_rshd() in the previous release. Essentially the digits were not trimmed before the compare which cause a subtraction to occur all the time. -- Fixed up etc/tune.c a bit to stop testing new cutoffs after 16 failures [to find more optimal points]. Brute force ho! May 29th, 2003 v0.18 -- Fixed a bug in s_mp_sqr which would handle carries properly just not very elegantly. (e.g. correct result, just bad looking code) -- Fixed bug in mp_sqr which still had a 512 constant instead of MP_WARRAY -- Added Toom-Cook multipliers [needs tuning!] -- Added efficient divide by 3 algorithm mp_div_3 -- Re-wrote mp_div_d to be faster than calling mp_div -- Added in a donated BCC makefile and a single page LTM poster (ahalhabsi@sbcglobal.net) -- Added mp_reduce_2k which reduces an input modulo n = 2**p - k for any single digit k -- Made the exptmod system be aware of the 2k reduction algorithms. -- Rewrote mp_dr_reduce to be smaller, simpler and easier to understand. May 17th, 2003 v0.17 -- Benjamin Goldberg submitted optimized mp_add and mp_sub routines. A new gen.pl as well as several smaller suggestions. Thanks! -- removed call to mp_cmp in inner loop of mp_div and put mp_cmp_mag in its place :-) -- Fixed bug in mp_exptmod that would cause it to fail for odd moduli when DIGIT_BIT != 28 -- mp_exptmod now also returns errors if the modulus is negative and will handle negative exponents -- mp_prime_is_prime will now return true if the input is one of the primes in the prime table -- Damian M Gryski (dgryski@uwaterloo.ca) found a index out of bounds error in the mp_fast_s_mp_mul_high_digs function which didn't come up before. (fixed) -- Refactored the DR reduction code so there is only one function per file. -- Fixed bug in the mp_mul() which would erroneously avoid the faster multiplier [comba] when it was allowed. The bug would not cause the incorrect value to be produced just less efficient (fixed) -- Fixed similar bug in the Montgomery reduction code. -- Added tons of (mp_digit) casts so the 7/15/28/31 bit digit code will work flawlessly out of the box. Also added limited support for 64-bit machines with a 60-bit digit. Both thanks to Tom Wu (tom@arcot.com) -- Added new comments here and there, cleaned up some code [style stuff] -- Fixed a lingering typo in mp_exptmod* that would set bitcnt to zero then one. Very silly stuff :-) -- Fixed up mp_exptmod_fast so it would set "redux" to the comba Montgomery reduction if allowed. This saves quite a few calls and if statements. -- Added etc/mont.c a test of the Montgomery reduction [assuming all else works :-| ] -- Fixed up etc/tune.c to use a wider test range [more appropriate] also added a x86 based addition which uses RDTSC for high precision timing. -- Updated demo/demo.c to remove MPI stuff [won't work anyways], made the tests run for 2 seconds each so its not so insanely slow. Also made the output space delimited [and fixed up various errors] -- Added logs directory, logs/graph.dem which will use gnuplot to make a series of PNG files that go with the pre-made index.html. You have to build [via make timing] and run ltmtest first in the root of the package. -- Fixed a bug in mp_sub and mp_add where "-a - -a" or "-a + a" would produce -0 as the result [obviously invalid]. -- Fixed a bug in mp_rshd. If the count == a.used it should zero/return [instead of shifting] -- Fixed a "off-by-one" bug in mp_mul2d. The initial size check on alloc would be off by one if the residue shifting caused a carry. -- Fixed a bug where s_mp_mul_digs() would not call the Comba based routine if allowed. This made Barrett reduction slower than it had to be. Mar 29th, 2003 v0.16 -- Sped up mp_div by making normalization one shift call -- Sped up mp_mul_2d/mp_div_2d by aliasing pointers :-) -- Cleaned up mp_gcd to use the macros for odd/even detection -- Added comments here and there, mostly there but occasionally here too. Mar 22nd, 2003 v0.15 -- Added series of prime testing routines to lib -- Fixed up etc/tune.c -- Added DR reduction algorithm -- Beefed up the manual more. -- Fixed up demo/demo.c so it doesn't have so many warnings and it does the full series of tests -- Added "pre-gen" directory which will hold a "gen.pl"'ed copy of the entire lib [done at zipup time so its always the latest] -- Added conditional casts for C++ users [boo!] Mar 15th, 2003 v0.14 -- Tons of manual updates -- cleaned up the directory -- added MSVC makefiles -- source changes [that I don't recall] -- Fixed up the lshd/rshd code to use pointer aliasing -- Fixed up the mul_2d and div_2d to not call rshd/lshd unless needed -- Fixed up etc/tune.c a tad -- fixed up demo/demo.c to output comma-delimited results of timing also fixed up timing demo to use a finer granularity for various functions -- fixed up demo/demo.c testing to pause during testing so my Duron won't catch on fire [stays around 31-35C during testing :-)] Feb 13th, 2003 v0.13 -- tons of minor speed-ups in low level add, sub, mul_2 and div_2 which propagate to other functions like mp_invmod, mp_div, etc... -- Sped up mp_exptmod_fast by using new code to find R mod m [e.g. B^n mod m] -- minor fixes Jan 17th, 2003 v0.12 -- re-wrote the majority of the makefile so its more portable and will install via "make install" on most *nix platforms -- Re-packaged all the source as seperate files. Means the library a single file packagage any more. Instead of just adding "bn.c" you have to add libtommath.a -- Renamed "bn.h" to "tommath.h" -- Changes to the manual to reflect all of this -- Used GNU Indent to clean up the source Jan 15th, 2003 v0.11 -- More subtle fixes -- Moved to gentoo linux [hurrah!] so made *nix specific fixes to the make process -- Sped up the montgomery reduction code quite a bit -- fixed up demo so when building timing for the x86 it assumes ELF format now Jan 9th, 2003 v0.10 -- Pekka Riikonen suggested fixes to the radix conversion code. -- Added baseline montgomery and comba montgomery reductions, sped up exptmods [to a point, see bn.h for MONTGOMERY_EXPT_CUTOFF] Jan 6th, 2003 v0.09 -- Updated the manual to reflect recent changes. :-) -- Added Jacobi function (mp_jacobi) to supplement the number theory side of the lib -- Added a Mersenne prime finder demo in ./etc/mersenne.c Jan 2nd, 2003 v0.08 -- Sped up the multipliers by moving the inner loop variables into a smaller scope -- Corrected a bunch of small "warnings" -- Added more comments -- Made "mtest" be able to use /dev/random, /dev/urandom or stdin for RNG data -- Corrected some bugs where error messages were potentially ignored -- add etc/pprime.c program which makes numbers which are provably prime. Jan 1st, 2003 v0.07 -- Removed alot of heap operations from core functions to speed them up -- Added a root finding function [and mp_sqrt macro like from MPI] -- Added more to manual Dec 31st, 2002 v0.06 -- Sped up the s_mp_add, s_mp_sub which inturn sped up mp_invmod, mp_exptmod, etc... -- Cleaned up the header a bit more Dec 30th, 2002 v0.05 -- Builds with MSVC out of the box -- Fixed a bug in mp_invmod w.r.t. even moduli -- Made mp_toradix and mp_read_radix use char instead of unsigned char arrays -- Fixed up exptmod to use fewer multiplications -- Fixed up mp_init_size to use only one heap operation -- Note there is a slight "off-by-one" bug in the library somewhere without the padding (see the source for comment) the library crashes in libtomcrypt. Anyways a reasonable workaround is to pad the numbers which will always correct it since as the numbers grow the padding will still be beyond the end of the number -- Added more to the manual Dec 29th, 2002 v0.04 -- Fixed a memory leak in mp_to_unsigned_bin -- optimized invmod code -- Fixed bug in mp_div -- use exchange instead of copy for results -- added a bit more to the manual Dec 27th, 2002 v0.03 -- Sped up s_mp_mul_high_digs by not computing the carries of the lower digits -- Fixed a bug where mp_set_int wouldn't zero the value first and set the used member. -- fixed a bug in s_mp_mul_high_digs where the limit placed on the result digits was not calculated properly -- fixed bugs in add/sub/mul/sqr_mod functions where if the modulus and dest were the same it wouldn't work -- fixed a bug in mp_mod and mp_mod_d concerning negative inputs -- mp_mul_d didn't preserve sign -- Many many many many fixes -- Works in LibTomCrypt now :-) -- Added iterations to the timing demos... more accurate. -- Tom needs a job. Dec 26th, 2002 v0.02 -- Fixed a few "slips" in the manual. This is "LibTomMath" afterall :-) -- Added mp_cmp_mag, mp_neg, mp_abs and mp_radix_size that were missing. -- Sped up the fast [comba] multipliers more [yahoo!] Dec 25th,2002 v0.01 -- Initial release. Gimme a break. -- Todo list, add details to manual [e.g. algorithms] more comments in code example programs