ginac.git
10 years agoindexed::eval: use standard C++ RTTI.
Alexei Sheplyakov [Wed, 15 Oct 2008 11:32:11 +0000 (15:32 +0400)]
indexed::eval: use standard C++ RTTI.

Custom RTTI considered harmful, part 3.

10 years agobasic, expairseq: use standard C++ RTTI in compare(), is_equal(), match(), etc.
Alexei Sheplyakov [Wed, 15 Oct 2008 11:32:11 +0000 (15:32 +0400)]
basic, expairseq: use standard C++ RTTI in compare(), is_equal(), match(), etc.

Custom RTTI considered harmful, part 2.

* basic.cpp: use standard C++ RTTI in compare(), is_equal(), operator=(),
  match().
* expairseq.cpp: use standard C++ RTTI in match(), make_flat(),
  construct_from_2_ex().

10 years agois_exactly_a: use typeid() to check the type of expression.
Alexei Sheplyakov [Wed, 15 Oct 2008 11:32:11 +0000 (15:32 +0400)]
is_exactly_a: use typeid() to check the type of expression.

Custom RTTI considered harmful, part 1.

Custom run time type information (RTTI) system implemented in GiNaC have
several serious drawbacks, such as

1. It makes writing GiNaC classes unnecessary cumbersome.
2. It wastes sizeof(void *) bytes per object, for small objects like
   symbol, numeric, etc. this overhead is considerable.

It turns out that GiNaC's RTTI is not any faster than the standard one
(at least on GNU/Linux and Solaris with GNU C++ library and compiler).
So, GiNaC RTTI have no reasons to exit any more.

10 years agoptr.h: use unsigned int to store reference counts (in order to save memory).
Alexei Sheplyakov [Wed, 15 Oct 2008 06:16:07 +0000 (10:16 +0400)]
ptr.h: use unsigned int to store reference counts (in order to save memory).

We'll hardly ever get more than 2^32 references to the same object. So using
size_t to store reference count only wastes 4 bytes per object (on 64-bit
architecture). Use unsigned int instead.

10 years agoRewritten heuristic and PRS GCD for univariate polynomials, added benchmark.
Alexei Sheplyakov [Sun, 19 Oct 2008 17:27:14 +0000 (21:27 +0400)]
Rewritten heuristic and PRS GCD for univariate polynomials, added benchmark.

Using a better data structure for univariate polynomials makes GCD 10 -- 300
times faster (and less memory hungry).

10 years agopolynomial: introduce a helper function to divide polynomial by ring element.
Alexei Sheplyakov [Tue, 14 Oct 2008 06:44:07 +0000 (10:44 +0400)]
polynomial: introduce a helper function to divide polynomial by ring element.

10 years agomod_gcd: naive hack to chose a 'good' prime (to speed up gcd computation).
Alexei Sheplyakov [Tue, 14 Oct 2008 06:37:42 +0000 (10:37 +0400)]
mod_gcd: naive hack to chose a 'good' prime (to speed up gcd computation).

10 years ago[bugfix] remainder_in_ring: using exact division is plain wrong.
Alexei Sheplyakov [Mon, 29 Sep 2008 05:58:35 +0000 (09:58 +0400)]
[bugfix] remainder_in_ring: using exact division is plain wrong.

10 years ago[bugfix] normalize_in_ring: don't set the leading coefficient to 1.
Alexei Sheplyakov [Mon, 29 Sep 2008 05:53:46 +0000 (09:53 +0400)]
[bugfix] normalize_in_ring: don't set the leading coefficient to 1.

The coefficient ring is not a field, so the leading coefficient don't
have to be 1.

10 years ago[trivial] excompiler.cpp: shut up some silly compiler warnings.
Alexei Sheplyakov [Thu, 25 Sep 2008 09:19:57 +0000 (13:19 +0400)]
[trivial] excompiler.cpp: shut up some silly compiler warnings.

GCC warns about 'comparison between signed and unsigned integer expressions'.
In this case such comparison is harmless, but still it's a bit annoying.

10 years agoMerge branch 'master' of git://ffmssmsc.jinr.ru:443/varg/ginac
Jens Vollinga [Tue, 30 Sep 2008 09:47:25 +0000 (11:47 +0200)]
Merge branch 'master' of git://ffmssmsc.jinr.ru:443/varg/ginac

10 years agoReplaced class UniPoly by cl_UP_MI.
Jens Vollinga [Mon, 29 Sep 2008 15:38:46 +0000 (17:38 +0200)]
Replaced class UniPoly by cl_UP_MI.

10 years ago- Added options argument to factor(). Added flag factor_options::all that lets
Jens Vollinga [Fri, 26 Sep 2008 09:07:02 +0000 (11:07 +0200)]
- Added options argument to factor(). Added flag factor_options::all that lets
  factor() act on all polynomial subexpressions.
- Added more comments.

11 years agoImplemented modular GCD algorithm for univariate polynomials.
Alexei Sheplyakov [Sat, 20 Sep 2008 19:18:46 +0000 (23:18 +0400)]
Implemented modular GCD algorithm for univariate polynomials.

11 years agoWipe out the old (bison/flex generated) parser.
Alexei Sheplyakov [Sun, 14 Sep 2008 03:13:01 +0000 (07:13 +0400)]
Wipe out the old (bison/flex generated) parser.

Bison generated parser has a number of problems:

1. Bad performance. Parsing a sum seems to be O(N^{2 + a}) (a > 0).
   For example, parsing a sum (actually, a univariate polynomial) of 32768
   terms takes about 90 sec. on my box, parsing a sum of 10^6 terms takes
   "eternity".
2. The user is expected to provide list of all symbols in the expression.
   Often this is very annoying (and useless), sometimes it is not possible
   at all.
3. Parser is not reentrant (bison *can* produce reentrant parsers, but that
   won't solve other problems).
4. Parser is difficult to extend.

Since the new parser handles almost everything (useful) as the old one, let's
remove the latter.

11 years agocheck: time_parser.cpp: don't run the same benchmark twice.
Alexei Sheplyakov [Sun, 14 Sep 2008 02:57:21 +0000 (06:57 +0400)]
check: time_parser.cpp: don't run the same benchmark twice.

Since ex(const string&, lst&) ctor uses the new parser now comparing its
performance (and result) with one of direct invocation of the parser is
pointless.

11 years agoDocument the new parser, provide an example.
Alexei Sheplyakov [Sun, 14 Sep 2008 02:41:12 +0000 (06:41 +0400)]
Document the new parser, provide an example.

11 years agoUse the new parser in the ex(const string&, lst&) ctor.
Alexei Sheplyakov [Sun, 14 Sep 2008 02:24:29 +0000 (06:24 +0400)]
Use the new parser in the ex(const string&, lst&) ctor.

Note: this is certainly not the optimal way to use the parser. It's provided
for backward compatibility only.

11 years agoParser can parse (some) floating point numbers now.
Alexei Sheplyakov [Sun, 14 Sep 2008 02:01:53 +0000 (06:01 +0400)]
Parser can parse (some) floating point numbers now.

11 years agoparser: add necessary checks to operator() to stop accepting nonsense.
Alexei Sheplyakov [Sat, 13 Sep 2008 23:26:33 +0000 (03:26 +0400)]
parser: add necessary checks to operator() to stop accepting nonsense.

Since the parser is recursive parse_* methods (in particular,
parse_binop_rhs()) does NOT check if the last unparsed token is valid.
Thus parse_expression() stops if it encounters an unknown token. That's
why parser::operator() needs to make sure nothing is left in the input
stream.

11 years ago[bugfix] parser::parse_literal_expr(): don't forget to consume the token...
Alexei Sheplyakov [Sat, 13 Sep 2008 22:36:16 +0000 (02:36 +0400)]
[bugfix] parser::parse_literal_expr(): don't forget to consume the token...

... so parser won't process it twice (and get either spurious error or
wrong result).

11 years ago[bugfix]: parser::parse_unary_expr() parses '-a-b' correctly now.
Alexei Sheplyakov [Sat, 13 Sep 2008 23:19:05 +0000 (03:19 +0400)]
[bugfix]: parser::parse_unary_expr() parses '-a-b' correctly now.

Also added regression test.

11 years agoginac.h: include parser.hpp
Alexei Sheplyakov [Sat, 13 Sep 2008 01:12:31 +0000 (05:12 +0400)]
ginac.h: include parser.hpp

11 years agoparser: change order of the constructor optional arguments.
Alexei Sheplyakov [Sat, 13 Sep 2008 00:55:23 +0000 (04:55 +0400)]
parser: change order of the constructor optional arguments.

Functions/methods having multiple optional arguments are not very convenient
to use in C++ (to put it mildly). Shuffle parser ctor arguments so not so
frequently used argument is the last one.

11 years agoparser: allow read/write access to symbol table and strictness.
Alexei Sheplyakov [Sat, 13 Sep 2008 00:40:12 +0000 (04:40 +0400)]
parser: allow read/write access to symbol table and strictness.

Intended usage:

parser reader;
ifstream input_file1, input_file2;
// read the first file...
ex e1 = reader(input_file1);
// ... add extra entry into the symbol table used by parser
symbol x;
parser.get_syms()["x"] = x;
// Disable the parser to introduce new symbols, e.g. to ensure the expression
// in input_file2 contains the same symbols as e1 (read from input_file1).
parser.strict = true;
ex e2;
try {
e2 = reader(input_file2);
} catch (...) {
abort();
}

11 years agoparser: map input strings onto arbitrary expressions (not only symbols).
Alexei Sheplyakov [Sat, 13 Sep 2008 00:30:22 +0000 (04:30 +0400)]
parser: map input strings onto arbitrary expressions (not only symbols).

So it's possible to make abbreviations, e.g.

symbol x;
symtab table;
table["x"] = x + log(x) + 1;
parser reader(table);
ex e = reader("1 + x^2 + 5*x^3");

11 years ago[BUGFIX] parser.hpp: fix include guard so the header is actually usable.
Alexei Sheplyakov [Sat, 13 Sep 2008 00:21:27 +0000 (04:21 +0400)]
[BUGFIX] parser.hpp: fix include guard so the header is actually usable.

11 years ago[nitpick] power::expand_add(): don't use int instead of std::size_t.
Alexei Sheplyakov [Fri, 12 Sep 2008 15:55:36 +0000 (19:55 +0400)]
[nitpick] power::expand_add(): don't use int instead of std::size_t.

This shuts a few 'comparison between signed and unsigned integer expressions'
warnings.

11 years ago[nitpick] inifcns_nstdsums: don't use int instead of std::size_t.
Alexei Sheplyakov [Fri, 12 Sep 2008 15:55:36 +0000 (19:55 +0400)]
[nitpick] inifcns_nstdsums: don't use int instead of std::size_t.

This shuts up quite a number of 'comparison between signed and unsigned
integer expressions'  warnings.

11 years ago[nitpick] don't use int instead of std::size_t.
Alexei Sheplyakov [Fri, 12 Sep 2008 15:55:36 +0000 (19:55 +0400)]
[nitpick] don't use int instead of std::size_t.

This shuts up (some of) 'comparison between signed and unsigned integer
expressions'  warnings.

11 years agoG_numeric: use cl_N and int to manipulate numbers (instead of ex).
Alexei Sheplyakov [Thu, 11 Sep 2008 11:16:21 +0000 (15:16 +0400)]
G_numeric: use cl_N and int to manipulate numbers (instead of ex).

Convert helper functions G_do_hoelder and G_do_trafo in the same manner,
update all call sites. This should speed up numerical calculation of
multiple polylogarithms a little bit, and reduce the memory usage.

11 years agoG_numeric: put convergence/acceleration transofrmations into helper functions.
Alexei Sheplyakov [Wed, 10 Sep 2008 11:43:35 +0000 (15:43 +0400)]
G_numeric: put convergence/acceleration transofrmations into helper functions.

This is simple code move, everything else should be considered a bug.
The helper functions (as well as G_numeric itself) will be improved by
subsequent patches.

11 years agomatch() (find()): use exmap (exset) to store matched subexpressions.
Alexei Sheplyakov [Fri, 12 Sep 2008 10:55:42 +0000 (14:55 +0400)]
match() (find()): use exmap (exset) to store matched subexpressions.

There's no need to re-invent associative arrays and wrap them into
a GiNaC class.

11 years agoexpairseq::match(): remove the code which works around basic::match bug.
Alexei Sheplyakov [Thu, 11 Sep 2008 12:59:30 +0000 (16:59 +0400)]
expairseq::match(): remove the code which works around basic::match bug.

basic::match() used to have side effects in a case of a failed match. As
a result of that bug exparsed::match did not work correctly in some cases,
see [1] for more details. These false negatives were worked around by [2].
Now that match() has no unwanted side effects [3] we don't need any work
arounds any more.

Just in a case add a regression test (from [1]).

[1] http://www.ginac.de/pipermail/ginac-devel/2006-April/000942.html
[2] Commit 73f0ce4cf8d91f073f35a45443f5fbe886921c5c ("Fixed bugs in ::match").
[3] Commit 192ed7390b7b2b705ad100e3db0a92eedd2b20ad ("match: don't modify
    subexpression list if expression doesn't match the pattern.").

11 years agoFixed various bugs in multivariate factorization.
Jens Vollinga [Thu, 18 Sep 2008 18:26:37 +0000 (20:26 +0200)]
Fixed various bugs in multivariate factorization.

11 years agoMerge branch 'master' of git://ffmssmsc.jinr.ru:443/varg/ginac
Jens Vollinga [Tue, 9 Sep 2008 20:42:50 +0000 (22:42 +0200)]
Merge branch 'master' of git://ffmssmsc.jinr.ru:443/varg/ginac

11 years agoBug fix related to the usage of int instead of cl_I.
Jens Vollinga [Tue, 9 Sep 2008 20:41:39 +0000 (22:41 +0200)]
Bug fix related to the usage of int instead of cl_I.

11 years agoUpdated NEWS.
Jens Vollinga [Tue, 9 Sep 2008 20:41:24 +0000 (22:41 +0200)]
Updated NEWS.

11 years agobuild: shut up automake warnings, don't use GNU make extensions.
Alexei Sheplyakov [Tue, 9 Sep 2008 10:44:25 +0000 (14:44 +0400)]
build: shut up automake warnings, don't use GNU make extensions.

Not that I really care about non-GNU makes, but those warnings are a bit
annoying and can hide useful ones.

11 years agobuild: put (almost) all auto* tools scripts into the config directory.
Alexei Sheplyakov [Mon, 8 Sep 2008 07:53:30 +0000 (11:53 +0400)]
build: put (almost) all auto* tools scripts into the config directory.

So I can just rm -rf it to clean up the repository.

11 years agoAllow user to disable GiNaC::compile_ex (e.g. for security reasons).
Alexei Sheplyakov [Mon, 8 Sep 2008 07:09:20 +0000 (11:09 +0400)]
Allow user to disable GiNaC::compile_ex (e.g. for security reasons).

configure takes --{disable,enable}-excompiler argument now. It can be
used to disable GiNaC::compile_ex (default is to enable it).

acinclude.m4:
GINAC_EXCOMPILER: new macro. Checks for libdl, allows user to disable
GiNaC::compile_ex. Also it doesn't bother to check for libdl on MinGW.

configure.ac: use GINAC_EXCOMPILER to check for libdl.

11 years agoconfigure: run important checks first (and bail out if something is missing).
Alexei Sheplyakov [Sun, 7 Sep 2008 18:53:07 +0000 (22:53 +0400)]
configure: run important checks first (and bail out if something is missing).

Don't bother to check for optional stuff if CLN and/or standard C++ headers
are missing.

11 years agoconfigure: don't bother to run checks which can be done at the compile time.
Alexei Sheplyakov [Sun, 7 Sep 2008 18:47:01 +0000 (22:47 +0400)]
configure: don't bother to run checks which can be done at the compile time.

Don't check for sizeof of various types, this can be done at the compile time.
GCC optimizes away these checks, so the actual code is the same.

11 years agoutils.h: use <stdint.h> (if available) instead of reinventing it.
Alexei Sheplyakov [Sun, 7 Sep 2008 18:17:55 +0000 (22:17 +0400)]
utils.h: use <stdint.h> (if available) instead of reinventing it.

The argument of golden_ratio_hash() is as an integer of the same size as
a void* pointer. Unfortunately ISO C++ 98 does not provide suitable typedef.
Hence

* use <stdint.h> if available and define p_int to uintptr_t. Note: AC_PROG_CC
  already checks for this header, so no extra checks are necessary.
* as a fallback define p_int to be unsigned long, this works on most systems
  I know of (the only exception is woe64).

While at it, stop including "config.h" unconditionally.

11 years agoconfigure: don't check for sizeof(long double), we don't use it.
Alexei Sheplyakov [Sat, 6 Sep 2008 05:21:18 +0000 (09:21 +0400)]
configure: don't check for sizeof(long double), we don't use it.

11 years agobuild: faster check for standard C++ headers.
Alexei Sheplyakov [Sat, 6 Sep 2008 04:51:19 +0000 (08:51 +0400)]
build: faster check for standard C++ headers.

Include them all into a test program and check if it compiles (in order to reduce
the run time of the `configure' script).

11 years agobuild: don't run any ${host} binaries while checking for readline.
Alexei Sheplyakov [Mon, 9 Jun 2008 12:21:30 +0000 (16:21 +0400)]
build: don't run any ${host} binaries while checking for readline.

Now GiNaC (to be more precise, ginsh) can be easily cross compiled.
However, ancient versions of readline (<= 4.2) are not supported any more.

11 years agodon't mention CVS any more, describe how to install from git.
Alexei Sheplyakov [Mon, 8 Sep 2008 07:50:12 +0000 (11:50 +0400)]
don't mention CVS any more, describe how to install from git.

11 years agomatch: don't modify subexpression list if expression doesn't match the pattern.
Alexei Sheplyakov [Tue, 15 Jul 2008 17:08:22 +0000 (21:08 +0400)]
match: don't modify subexpression list if expression doesn't match the pattern.

As of now the match() method modifies the list of matched subexpressions
(its second argument) even if the expression in question does not match
the pattern. Thus, this simple program

 #include <ginac/ginac.h>
 #include <iostream>
using namespace GiNaC;

int main(int argc, char** argv)
{
symbol x;
ex e = pow(x, 5);
ex pattern = pow(wild(), -1);
lst repl;
bool test = e.match(pattern, repl);
std::cout << "repl = " << repl << std::endl;
}

prints

repl = {x}

Such behaviour is a bit unexpected. Sometimes it confuses even GiNaC
developers, see e.g.
http://www.ginac.de/pipermail/ginac-devel/2006-April/000942.html

Hence this patch. Now the above program prints

repl = {}

as expected.

11 years ago[BUGFIX] Reclaiming the memory allocated for static objects *is* necessary.
Alexei Sheplyakov [Tue, 19 Aug 2008 11:39:46 +0000 (15:39 +0400)]
[BUGFIX] Reclaiming the memory allocated for static objects *is* necessary.

GiNaC allocates memory for static objects (i.e. flyweights, remember tables,
etc), but doesn't free it. This is OK if the program lifetime matches libginac
lifetime, since the OS will reclaim that memory anyway.
However, if the program lifetime is different from that of libginac, this
turns into a memory leak. This happens if someone dlopen's libginac.so, and
dlclose's it later on (read: if someone uses GiNaC via scripting language
bindings).

symbol::autoname_prefix(): there's no need for dynamical memory allocation.
remember_table::remember_tables(): likewise.
function::registered_functions(): likewise.
lib_init::~lib_init(): if library usage count drops to 0, reclaim the memory
                       allocated for flyweights.

11 years agosymbol: get rid of assign/unassign (for performance and other reasons).
Alexei Sheplyakov [Tue, 9 Sep 2008 07:14:57 +0000 (11:14 +0400)]
symbol: get rid of assign/unassign (for performance and other reasons).

* symbol::eval() is trivial now, so compiler can inline it in some cases.
* symbol takes less memory.
* no functionality is lost (as C++ has associative arrays).

11 years agoginsh: use exmap for storing assigned symbols.
Alexei Sheplyakov [Tue, 9 Sep 2008 06:39:26 +0000 (10:39 +0400)]
ginsh: use exmap for storing assigned symbols.

C++ already have associative arrays, there's no need to re-invent them.

11 years agomultiple zeta values: make crandall_Y_loop helper function reentrant.
Alexei Sheplyakov [Fri, 29 Aug 2008 19:44:51 +0000 (23:44 +0400)]
multiple zeta values: make crandall_Y_loop helper function reentrant.

* Move crB and crG variables into initcX function (the only user of these
  variables).
* Pass crX coefficients to crandall_Y_loop instead of using a global variable.
* While at it make crandall_Y_loop and initcX functions static.

11 years agomultiple zeta values: make crandall_Z helper function reentrant.
Alexei Sheplyakov [Fri, 29 Aug 2008 19:44:51 +0000 (23:44 +0400)]
multiple zeta values: make crandall_Z helper function reentrant.

Pass the f_kj coefficients as an arguments to crandall_Z and calc_f
functions instead of using global variables.

11 years agoparser: improve error reporting a little bit.
Alexei Sheplyakov [Mon, 25 Aug 2008 15:08:36 +0000 (19:08 +0400)]
parser: improve error reporting a little bit.

Introduce class 'parse_error' (which contain some info about location of
the error) and throw it on parse errors.

N.B.: the actual info is a bit inaccurate because lexer doesn't track
the location properly yet.

11 years agolexer: when switching to another output stream, clean last read character.
Alexei Sheplyakov [Mon, 25 Aug 2008 13:49:58 +0000 (17:49 +0400)]
lexer: when switching to another output stream, clean last read character.

Otherwise we prepend to the current stream the last character read from
the previous stream, which is obviously incorrect.

11 years agoFixed bug in unvariate factorization. Bound for lifting was using a ordinary
Jens Vollinga [Mon, 8 Sep 2008 17:17:56 +0000 (19:17 +0200)]
Fixed bug in unvariate factorization. Bound for lifting was using a ordinary
int instead of a cl_I.

11 years agoAdded missing code for multivariate factorization.
Jens Vollinga [Mon, 8 Sep 2008 16:32:27 +0000 (18:32 +0200)]
Added missing code for multivariate factorization.

11 years agoFixed bug. G() ignored the imaginary parts of the arguments.
Jens Vollinga [Mon, 8 Sep 2008 16:17:51 +0000 (18:17 +0200)]
Fixed bug. G() ignored the imaginary parts of the arguments.

11 years agoAdded internal code for multivariate factorization.
Jens Vollinga [Wed, 27 Aug 2008 15:25:16 +0000 (17:25 +0200)]
Added internal code for multivariate factorization.

11 years agogcd_pf_pow_pow: deobfuscate a little bit (no functional changes).
Alexei Sheplyakov [Mon, 25 Aug 2008 12:57:38 +0000 (16:57 +0400)]
gcd_pf_pow_pow: deobfuscate a little bit (no functional changes).

Use

if (foo)
return bar();
return baz();

instead of

if (foo) {
return bar();
} else {
return baz();
}

This makes the code a little bit more readable.

11 years agogcd_pf_pow: get rid of duplicate code.
Alexei Sheplyakov [Mon, 25 Aug 2008 12:57:02 +0000 (16:57 +0400)]
gcd_pf_pow: get rid of duplicate code.

This function (which helps gcd() handle partially factored expressions)
contained two copies of the same code. Remove one redundant copy.

11 years agointroduce gcd_pf_pow_pow: gcd helper to handle partially factored expressions.
Alexei Sheplyakov [Mon, 25 Aug 2008 12:56:04 +0000 (16:56 +0400)]
introduce gcd_pf_pow_pow: gcd helper to handle partially factored expressions.

gcd_pf_pow_pow handles the case when both arguments of gcd() are powers.

N.B. the actual code moved from gcd_pf_pow, no functional changes.

11 years agogcd_pf_{pow, mul}: don't check if the arguments are polynomials.
Alexei Sheplyakov [Mon, 25 Aug 2008 12:55:42 +0000 (16:55 +0400)]
gcd_pf_{pow, mul}: don't check if the arguments are polynomials.

These functions gets called only from gcd(), which does this check
on its own.

11 years agogcd_pf_mul: get rid of duplicate code.
Alexei Sheplyakov [Mon, 25 Aug 2008 12:55:13 +0000 (16:55 +0400)]
gcd_pf_mul: get rid of duplicate code.

This function (which helps gcd() handle partially factored expressions)
contained two copies of the same code. Remove one redundant copy.

11 years agogcd(): allow user to override (some of) heuristics.
Alexei Sheplyakov [Mon, 25 Aug 2008 12:54:46 +0000 (16:54 +0400)]
gcd(): allow user to override (some of) heuristics.

GiNaC tries to avoid expanding expressions while computing GCDs and applies
a number of heuristics. Usually this improves performance, but in some cases
it doesn't. Allow user to switch off heuristics.

Part 5:

* gcd(): don't use heuristic GCD algorithm if gcd_options::no_heur_gcd
  flag is set.
* gcd(): don't handle partially factored expressions in a special way
  if gcd_options::no_part_factored flag is set.

11 years agorefactor gcd() a little bit (no functional changes).
Alexei Sheplyakov [Mon, 25 Aug 2008 12:54:10 +0000 (16:54 +0400)]
refactor gcd() a little bit (no functional changes).

GiNaC tries to avoid expanding expressions while computing GCDs and applies
a number of heuristics. Usually this improves performance, but in some cases
it doesn't. Allow user to switch off heuristics.

Part 4:

refactor gcd() a little bit, so subsequent patch(es) won't be so big and ugly.

11 years agointroduce gcd_pf_mul: gcd helper to handle partially factored expressions.
Alexei Sheplyakov [Mon, 25 Aug 2008 12:53:47 +0000 (16:53 +0400)]
introduce gcd_pf_mul: gcd helper to handle partially factored expressions.

GiNaC tries to avoid expanding expressions while computing GCDs and applies
a number of heuristics. Usually this improves performance, but in some cases
it doesn't. Allow user to switch off heuristics.

Part 3:

Move the code handling products from gcd() into a separate function. This
is *really* only code move, everything else should be considered a bug.

11 years agointroduce gcd_pf_pow: gcd helper to handle partially factored expressions.
Alexei Sheplyakov [Mon, 25 Aug 2008 12:53:07 +0000 (16:53 +0400)]
introduce gcd_pf_pow: gcd helper to handle partially factored expressions.

GiNaC tries to avoid expanding expressions while computing GCDs and applies
a number of heuristics. Usually this improves performance, but in some cases
it doesn't. Allow user to switch off heuristics.

Part 2:

Move the code handling powers from gcd() into a separate function. This
is *really* only code move, everything else should be considered a bug.

11 years agogcd: allow user to override (some of) heuristics.
Alexei Sheplyakov [Mon, 25 Aug 2008 12:52:45 +0000 (16:52 +0400)]
gcd: allow user to override (some of) heuristics.

GiNaC tries to avoid expanding expressions while computing GCDs and applies
a number of heuristics. Usually this improves performance, but in some cases
it doesn't. Allow user to switch off heuristics.

Part 1:

* add new (optional) argument to gcd() to control its behaviour.
* introduce gcd_options structure.

N.B. No actual code changes so far, the actual handling of newly introduced
options is the subject of further patches.

11 years agoinifscn_nstdsums: make functions handling Li/G transformations reentrant.
Alexei Sheplyakov [Sun, 24 Aug 2008 11:25:24 +0000 (15:25 +0400)]
inifscn_nstdsums: make functions handling Li/G transformations reentrant.

Explicitly pass the dummy symbols instead of using a global variable.
As a side effect, the code is more clear now (that's a bit subjective, but
anyway).

11 years agoBail out if both autogen and autogen'erated file(s) are missing.
Alexei Sheplyakov [Thu, 21 Aug 2008 22:42:17 +0000 (02:42 +0400)]
Bail out if both autogen and autogen'erated file(s) are missing.

This makes the error message(s) more helpful.

11 years agoFaster, better (recursive descent) expression parser.
Alexei Sheplyakov [Tue, 19 Aug 2008 19:48:03 +0000 (23:48 +0400)]
Faster, better (recursive descent) expression parser.

Ouch, it fails to compile. Here is a correct version:

From: Alexei Sheplyakov <varg@theor.jinr.ru>
Date: Tue, 19 Aug 2008 21:50:03 +0400
Subject: [PATCH] Faster, better (recursive descent) expression parser.

bison generated parser has a number of problems:

1. Bad performance. Parsing a sum seems to be O(N^{2 + a}) (a > 0) [1].
   For example, parsing a sum (actually, a univariate polynomial) of 32768
   terms takes about 90 sec. on my box, parsing a sum of 10^6 terms takes
   "eternity".
2. The user is expected to provide list of all symbols in the expression.
   Often this is very annoying (and useless).
3. Parser is not reentrant (bison *can* produce reentrant parsers, but that
   won't solve other problems).
4. Parser is difficult to extend.

Hence the new parser.

Features:

1. Parsing large sums and products is O(N).
2. Parser is reentrant (well, almost).
3. It's possible to insert (shell style) comments inside the expressions.
4. matrices, lists, FAIL are NOT handled. Yes, this *is* a feature :-)

Limitations:

1. Error handling is a bit terse: on error exception is thrown, and that's it.
2. Binary, octal, and hexadecimal numbers can not be parsed (yet).
3. Tensors, noncommutative products, etc. can not be parsed.

Other notes:

1. ex ctor still uses the old parser.
2. ginsh still uses the old parser.

[1] Mesured by this script (requires gnuplot):

make_expression_string ()
{
printf "1 + x"
local n=2
while test $n -le $1; do
printf " + %s*x^%s" $n $n
n=$(expr $n + 1)
done
}

single_test ()
{
printf "$1 \t"
( printf "time("; make_expression_string $1; printf ");" ) | \
ginsh | sed -e 's/s$//'
}

benchmark ()
{
local N=$1
while test $N -le $2; do
single_test $N
N=$(expr $N \* 2)
done

}

gnuplot_header ()
{
  echo "set logscale xy"
echo "set xlabel 'degree (# of terms)'"
echo "set ylabel 'time, sec.'"
echo "set xrange [${1}:${2}]"
echo "plot '-' using 1:2 with lines title '1+x+2*x^2+...+n*x^n'"
}

gnuplot_footer ()
{
echo "e"
}

benchmark_and_plot ()
{
( gnuplot_header $1 $2
  benchmark $1 $2 | tee $3.txt
  gnuplot_footer ) | \
gnuplot -persist '-'
}

N_ini=${1:-1024}
N_fin=${2:-32768}
out_base=${3:-parser_benchmark}

benchmark_and_plot $N_ini $N_fin $out_base

11 years agoDaily bugfix in the polynomial factorization (code didn't catch polynomial "x"
Jens Vollinga [Thu, 21 Aug 2008 20:37:09 +0000 (22:37 +0200)]
Daily bugfix in the polynomial factorization (code didn't catch polynomial "x"
and cln::sqrt conversion was buggy).

11 years agoFixed bug in the Q matrix calculation and the univariate factorization function.
Jens Vollinga [Wed, 20 Aug 2008 20:16:13 +0000 (22:16 +0200)]
Fixed bug in the Q matrix calculation and the univariate factorization function.

11 years agoFixed bug in modular square-free factorization.
Jens Vollinga [Mon, 18 Aug 2008 19:10:48 +0000 (21:10 +0200)]
Fixed bug in modular square-free factorization.

11 years agoAdded polynomial factorization (univariate case).
Jens Vollinga [Sat, 9 Aug 2008 08:14:02 +0000 (10:14 +0200)]
Added polynomial factorization (univariate case).

11 years agoAdded polynomial factorization (univariate case).
Jens Vollinga [Sat, 9 Aug 2008 08:11:40 +0000 (10:11 +0200)]
Added polynomial factorization (univariate case).

11 years agoAny complex number can be (un)archived properly.
Alexei Sheplyakov [Wed, 30 Jul 2008 16:32:46 +0000 (20:32 +0400)]
Any complex number can be (un)archived properly.

A complex number can have an exact real part and inexact (floating point)
imaginary part, and vice a versa. Handle these cases properly in the archiving
code instead of bailing out with a bizzare error message.

Thanks to Chris Bouchard for reporting the bug.

NOTE: this fix changes the format of GiNaC archives, so formally it breaks
the binary compatibility. However, "mixed" complex numbers (i.e.  ones with
exact real part and inexact imaginary part) can not be archived with previous
versions of GiNaC, so there's nothing to break.

11 years agodefine lgamma and tgamma taking cl_N as an argument.
Alexei Sheplyakov [Wed, 19 Mar 2008 09:28:34 +0000 (12:28 +0300)]
define lgamma and tgamma taking cl_N as an argument.

The actual code is the same. These functions are OK since cl_N is not
automatically converted to numeric any more.

11 years agoImplicit conversion from cl_N to numeric considered harmful.
Alexei Sheplyakov [Wed, 19 Mar 2008 09:28:10 +0000 (12:28 +0300)]
Implicit conversion from cl_N to numeric considered harmful.

Finally, mark the numeric(const cl_N&) ctor as explicit.
This allows one to mix the code using GiNaC and CLN, i.e.

cl_N x, y;
// initialize them
cl_N z = sin(x) +  y*exp(y);
symbol a("a");
ex e = exp(a);

11 years agobernoulli, fibonacci, iquo: explicitly convert return values to numeric.
Alexei Sheplyakov [Wed, 19 Mar 2008 09:27:27 +0000 (12:27 +0300)]
bernoulli, fibonacci, iquo: explicitly convert return values to numeric.

Implicit conversion from cl_N to numeric considered harmful, part 6.

11 years agotgamma, lgamma: convert to take cl_N as an argument; provide wrappers for GiNaC:...
Alexei Sheplyakov [Wed, 19 Mar 2008 09:27:07 +0000 (12:27 +0300)]
tgamma, lgamma: convert to take cl_N as an argument; provide wrappers for GiNaC::numeric

Implicit conversion from cl_N to numeric considered harmful, part 5.

11 years agoLi2, zeta, sqrt, abs, gcd, etc.: explicitly convert return value to numeric.
Alexei Sheplyakov [Wed, 19 Mar 2008 09:26:52 +0000 (12:26 +0300)]
Li2, zeta, sqrt, abs, gcd, etc.: explicitly convert return value to numeric.

Implicit conversion from cl_N to numeric considered harmful, part 4.

11 years agoelementary functions: explicitly convert return values from cl_N to numeric.
Alexei Sheplyakov [Wed, 19 Mar 2008 09:26:23 +0000 (12:26 +0300)]
elementary functions: explicitly convert return values from cl_N to numeric.

Implicit conversion from cl_N to numeric considered harmful, part 3.

11 years agoinifcns_nstdsums.cpp: Lin_numeric takes cl_N as an argument instead of numeric.
Alexei Sheplyakov [Wed, 19 Mar 2008 09:25:55 +0000 (12:25 +0300)]
inifcns_nstdsums.cpp: Lin_numeric takes cl_N as an argument instead of numeric.

Implicit conversion from cl_N to numeric considered harmful, part 2.

11 years agoinifcns_nstdsums.cpp: S_num takes cl_N as an argument instead of numeric.
Alexei Sheplyakov [Wed, 19 Mar 2008 09:24:36 +0000 (12:24 +0300)]
inifcns_nstdsums.cpp: S_num takes cl_N as an argument instead of numeric.

Implicit conversion from cl_N to numeric considered harmful.

Using GiNaC::numeric for numerical computations incurs significant
overhead, so one might want to do these computations using proper CLN
types. Unfortunately, it's not easy due to automatic conversion from
cln::cl_N to GiNaC::numeric. Here is a simple example:

cl_N x, y;
// initialize them
return sin(x) +  y*exp(y);

The compiler complains about ambigously overloaded of functions, i.e.
cl_N cln::sin(const cl_N&) versus numeric GiNaC::sin(const numeric&).

Hence, I propose to disable *implicit* conversion from cl_N to numeric
(this can be done by marking the numeric ctor as `explicit').

However, this change happens to be a bit nontrivial, because that evil
conversion is used in GiNaC itself. So, I decided to rewrite those fragments
of code.

11 years agodocumentation: it is get_dim(), not get_dimension()
Richard B. Kreckel [Tue, 15 Jul 2008 21:57:43 +0000 (23:57 +0200)]
documentation: it is get_dim(), not get_dimension()

This was misdocumented ever since GiNaC-0.8.0.

11 years agoclifford_unit: fix possible bugs due to wrong operator[!=]= usage.
Alexei Sheplyakov [Mon, 14 Jul 2008 13:46:25 +0000 (17:46 +0400)]
clifford_unit: fix possible bugs due to wrong operator[!=]= usage.

11 years agoRemoved left-over debugging line.
Jens Vollinga [Wed, 25 Jun 2008 09:28:11 +0000 (11:28 +0200)]
Removed left-over debugging line.

11 years agoMake the behaviour of class function more consistent with respect to
Jens Vollinga [Wed, 25 Jun 2008 09:17:55 +0000 (11:17 +0200)]
Make the behaviour of class function more consistent with respect to
ncmul::eval(). [V.Kisil]

11 years agoImprove heur_gcd() so it can handle rational polynomials, and add a test case.
Jens Vollinga [Tue, 24 Jun 2008 19:26:58 +0000 (21:26 +0200)]
Improve heur_gcd() so it can handle rational polynomials, and add a test case.

Previously heur_gcd() worked only with integer polynomials, and did not check
if the inputs are indeed integer polynomials. That lead to an endless loop or
a miscomputed gcd.

[A.Sheplyakov]

11 years agoFixed bug in mLi summation causing premature drop-out and made Nielsen polylog
Jens Vollinga [Fri, 4 Apr 2008 12:48:19 +0000 (14:48 +0200)]
Fixed bug in mLi summation causing premature drop-out and made Nielsen polylog
invalidate its lookup tables if precision has been changed.

11 years agoAdded legacy tests for zeta and multiple polylog.
Jens Vollinga [Fri, 4 Apr 2008 12:47:07 +0000 (14:47 +0200)]
Added legacy tests for zeta and multiple polylog.

11 years agoCheck for fixed bug in multiple polylogs.
Jens Vollinga [Wed, 13 Dec 2006 20:04:23 +0000 (20:04 +0000)]
Check for fixed bug in multiple polylogs.

11 years agoFixed bugs found by Jianqiang Zhao.
Jens Vollinga [Thu, 3 Apr 2008 09:56:57 +0000 (11:56 +0200)]
Fixed bugs found by Jianqiang Zhao.

11 years agoAdjust for GiNaC 1.5.
Richard B. Kreckel [Thu, 27 Mar 2008 22:56:32 +0000 (23:56 +0100)]
Adjust for GiNaC 1.5.

11 years agoMerge commit 'origin/master'
Jens Vollinga [Thu, 27 Mar 2008 10:16:11 +0000 (11:16 +0100)]
Merge commit 'origin/master'

11 years agoFixed make distcheck issues.
Jens Vollinga [Thu, 27 Mar 2008 10:08:09 +0000 (11:08 +0100)]
Fixed make distcheck issues.

11 years agoUpdate debian/changelog in anticipation of release.
Richard B. Kreckel [Wed, 26 Mar 2008 23:50:19 +0000 (00:50 +0100)]
Update debian/changelog in anticipation of release.

In reality, I'm testing the post-receive hook.