--------------------
MAPM Library History
--------------------
V 1.00 - June 1, 1999 Initial Release
V 1.10 - June 20, 1999 Use a stack implementation for local M_APM
variables instead of static variables.
Use a new faster algorithm for SIN & COS.
V 1.20 - July 10, 1999 Improve the algorithm for ARCSIN and ARCCOS
when the input argument is near 1.0.
Added the 'DUP' operator to the 'calc' demo
program.
V 1.30 - July 31, 1999 Added a function to output M_APM values as
integers.
Supply another demo program which generates
prime numbers (and uses the integer output
format).
Improved the 4 iterative routines, dynamically
adjusting the local precision each time through
the loop. Also changed the tolerance test to
use integers (vs a MAPM number).
V 2.00 - Aug 15, 1999 Added a 'fast' multiplication algorithm.
V 2.10 - Sep 24, 1999 Added integer POW function. Compute X^Y when
'Y' is an integer.
Added a random number generator (with a period
of 1.0E+15).
Added function to compute both sin and cos with
one call.
V 2.15 - Feb 10, 2000 Allow caller to easily use their own memory
management functions.
Implement the real exponential algorithm as
originally intended by David Bailey.
Delete the extra recursive function from the
fast multiply routine.
Calculate a few extra digits in the intermediate
results of 'calc' so the final answers are
always consistent.
V 2.20 - Mar 30, 2000 Let log, arc-sin, and arc-cos determine the
max number of iterations that should be required
to calculate a given precision level.
Optimize the raw series expansion for exp, sin,
and cos to use integers instead of MAPM numbers
in the tolerance test to determine loop termination.
V 2.25 - Apr 2, 2000 Added a CBRT function (cube root)
Added the hyperbolic SIN, COS, and TAN functions
plus their inverses.
Alias m_apm_asin to m_apm_arcsin. Also for acos,
atan, and atan2.
Updated CALC and VALIDATE to use the new functions.
V 3.00 - Apr 7, 2000 Added the MAPM C++ wrapper class. Supplied by
Orion Sky Lawlor (olawlor@acm.org)
This C++ class allows 'normal' syntax when using
the library.
V 3.05 - Apr 14, 2000 Optimized the basic divide algorithm. It is now
approx 10 - 15 % faster.
Optimized the fast multiply algorithm. It is now
approx 2X faster. (depends on input numbers,
sometimes I got 50%, other times 3X)
V 3.06 - May 6, 2000 Optimized a few more functions.
In 'CALC', recompute PI on the fly if more digits
are asked for than the default precision.
Changed default compile optimizations to use
'-O2' instead of '-O'. Just making a note.
in case some compilers don't handle this well.
V 3.10 - May 21, 2000 Eliminated all warnings due to internal constants
lacking enough precision for a given calculation.
Any internal constant (like PI, log(2)) lacking the
precison needed for a given calculation will be
re-computed on the fly as necessary. The only
warnings from the library now are legitimate domain
errors, sqrt of a negative number, etc.
V 3.15 - May 26, 2000 Optimize the factorial function. It is now approx
10X faster when N is large (i.e., N > 5000).
Return 1 from pow function when computing 0^0.
V 3.16 - Jun 24, 2000 Added makefile for Borland 5.2 / 5.5 C++
command line compilers for Win NT/9x.
Fixed link problems due to some C libraries not
having native cube root or inverse hyperbolic
functions.
V 3.50 - Jul 4, 2000 Added an FFT (Fast Fourier Transform) based
fast multiplication. The FFT fast multiplication
is O(N * Log2(N)). The FFT used is from the
Numerical Recipes book.
Fixed extern int declare in M_APM.H due to
variable name mangling in MSVC C++ compiler.
V 3.51 - Jul 7, 2000 Changed the FFT. The new FFT is from Takuya OOURA.
email: ooura@mmm.t.u-tokyo.ac.jp
Added build batch file for MSVC C++ compiler.
(Note: I can't verify this works since I don't
have MSVC C++)
V 3.55 - Jul 12, 2000 Optimized the FFT set-up wrapper function. Convert
MAPM base 100 numbers into base 10000 before
invoking the FFT. (2X speed improvement and 1/2
the RAM usage)
Use the AGM (Arithmetic-Geometric Mean) to
calculate PI when a more precise value is needed.
Added a real Makefile for Unix builds (finally!)
V 3.56 - Jul 19, 2000 Decrease the max number of bytes the FFT can
handle. *Worst* case inputs (all 9's) could have
caused the FFT math to overflow.
Added a build batch file for older Borland
compilers. (Works with Turbo C++ 3.00)
V 3.60 - Aug 24, 2000 Added 36 'memory' locations to the calc demo
program.
Added a new function to determine if a MAPM
number is an integer value or not.
When computing X^Y with the pow function,
automatically call integer_pow if 'Y' is
an integer value.
V 3.65 - Oct 1, 2000 Added a reciprocal function using an iterative
method.
Used the new reciprocal function to improve the
divide function.
Implemented an improved SQRT alogrithm. The new
one only uses multiplication and subtraction.
(The old algorithm used division which is slow).
V 3.70 - Nov 10, 2000 Use a more efficient 'log' algorithm and also
use a series expansion if the input is very close
to '1'.
Made a minor speed optimization to cube root.
Added build batch file for MSVC C++ compiler which
also uses a real 'makefile'.
V 3.75 - Dec 10, 2000 Use more efficient 'arc' family of functions if
the input arg is very close to zero.
V 3.80 - Mar 1, 2001 Improve the prime number demo program.
A number of minor code tweaks. This was mainly
in preparation for the upcoming 64 bit compilers.
I think everything is ready, but I really won't
know until I can actually get my hands on one ...
V 3.85 - Apr 2, 2001 Added new 'floor' and 'ceil' functions.
Use a new multiplier in the random number generator.
Added a build batch file for the LCC-WIN32 compiler.
V 4.0 - Jul 21, 2001 Added functions to compute GCD (greatest common
divisor) and LCM (least common multiple).
See 'm_apm_gcd' and 'm_apm_lcm'.
Added GCD and LCM to the 'calc' demo program.
Added new utility functions to determine if a MAPM
integer value is EVEN or ODD. See 'm_apm_is_even'
and 'm_apm_is_odd'.
Added function to release ALL memory that MAPM
has allocated. Added similiar function to trim
memory usage of MAPM without impacting any library
operation. See 'm_apm_free_all_mem' and
'm_apm_trim_mem_usage'.
Made a minor speed optimization to SQRT.
Optimized the normalization function.
Optimized the series expansion function for
sin/cos/exp. This has the added benefit of also
speeding up log, log10, pow, arc_###, etc.
V 4.1 - Aug 10, 2001 Fixed a minor problem with m_apm_to_integer_string.
If the input was 'slightly' smaller than an integer,
it would round the value up to the next int.
'Slightly' here means 1.0E-4 or smaller. If the
input was NNN.9999+ then NNN+1 would be returned.
The documented behavior is to truncate the value
to the next smallest integer and it now behaves as
expected.
Note: The previous function always gave the correct
output if the input was already an integer value
(which is why I didn't notice for so long ...).
Optimized the basic multiply function.
Added a build batch file for the Watcom C 11.x
compiler.
Changed the MSVC C++ compiler build from a makefile
back to a batch file (MKALLMSC.BAT). The makefile
was cumbersome and difficult to maintain. If the
user consensus wants the makefile back, I'll put
it back in.
V 4.2 - Oct 3, 2001 Added 3 functions to convert an MAPM number to a
string in fixed point format. Format 1017.829 as
"1017.829" instead of "1.017829E+3".
See the following functions:
m_apm_to_fixpt_string (basic function)
m_apm_to_fixpt_stringex (extended function)
m_apm_to_fixpt_stringexp (extended & returns char *)
V 4.3 - Feb 17, 2002 Include a PDF version of my article which was
first published in the November 2001 issue of
C/C++ Users Journal.
Made a minor optimization to _scale (the speed is
the same, but 2 temporary arrays were eliminated by
processing the scaling operation in reverse).
Added some comments and initial setup changes to
the prime number demo program.
Added some conditional compiler directives to
support the Metrowerks CodeWarrior 7.0 compiler.
(since I don't have it, I don't really know if
this will work ...)
Added a conditional compiler directive so the
user can disable the FFT multiply algorithm.
(the default is to still use the FFT multiply.)
Add a new function to prime the random number
generator with a known seed value (needed for
repeatable random sequences).
V 4.4 - Jun 14, 2002 Added a build batch file for the MINGW-32 compiler.
Change the internal stack used for temporary MAPM
numbers into a dynamically malloc'ed array.
Updated m_apm_cpp_precision so it can be called
prior to a call to m_apm_init.
Updated the documentation to indicate that the
default library is *NOT* thread-safe and problems
could occur in a multi-threaded application unless
certain precautions are taken.
Split mapmutil.c into 2 files since it was getting
too big.
V 4.5 - Nov 10, 2002 Added a makefile for the MAC (OSX).
Changed all library function parameter declarations
to use the modern form instead of the classic form.
Changed the order of the elements in the M_APM
structure. The M_APM structure in m_apm.h is not
compatible with previous versions of the library.
This was done to optimize for future 64 bit
architectures where sizeof(pointer) > sizeof(int).
Added a new integer pow function that does not
perform any rounding of the answer. This is needed
for integer only applications where full precision
must be maintained. See 'm_apm_integer_pow_nr'.
V 4.6 - Apr 7, 2003 Made a minor speed optimization in sqrt by rounding
an intermediate result in the iteration.
Implement a new cubically convergent log algorithm
(each iteration yields 3X more digits).
Compute internal constants log(2) and log(10) with
an AGM algorithm.
Move the output of all error messages into a
stand-alone function. All error messages now call
this function. This allows the user to easily
change the default reaction (which is output to
stderr) by only changing the one function.
V 4.6.1 - Apr 12, 2003 Fix a bug where specially contrived test input
could result in the new log algorithm terminating
too soon.
Applied lessons learned from the log bug to make
the asin & acos functions more efficient.
V 4.7 - Jun 10, 2003 Include the thread safe function wrappers in this
release. The necessary files are in the directory
'multi_thread' for the unix distribution and in
'multithd' for the Win/Dos distribution.
Added #define to obtain a clean compile with
Open Watcom 1.0 (using mkallwat.bat).
Add 2 functions to get the version of the compiled
library. See 'm_apm_lib_version' and
'm_apm_lib_short_version'. These could be very
useful when MAPM is used as a shared library/DLL.
(Long Overdue ....)
Implement faster cbrt algorithm.
Optimize the raw sin algorithm.
Implement another log algorithm. See details in
'algorithms.used'. Testing on my system has shown
that it is faster than an AGM algorithm, up to
~100,000 digits anyway. An AGM _should_ eventually
be faster, but I didn't find that cross-over point.
Move the generic error handling function into it's
own separate module. This allows the user to easily
replace the function (with their own version)
without changing the base library.
V 4.8 - Jul 27, 2003 Fix a bug in arctan. Very large inputs (> 1e16)
could cause incorrect answers in the last few
digits of the result and only under certain decimal
place accuracies. PI/2 was being returned (instead
of PI/2 - epsilon) due to a rounding problem. Thanks
to Ivano Primi for bringing this to my attention.
Fix arcsinh when the input was a very large negative
number. Under certain combinations of large input
values with certain decimal place accuracy would
result in a log(0). Thanks again to Ivano Primi.
Fix m_apm_set_double. An invalid input (NAN or
+/- INF) could cause MAPM to crash. Lesson here is
to always check return codes of function calls, even
when they cannot possibly fail. Thanks to Luiz H. de
Figueiredo (of the Lua project) for this report.
Did a complete re-write of validate. The intent of
this is to use the 'gcov' code coverage tool to get
as much code coverage of the library that is
practical. 100% code coverage is not obtained (nor
practical), but it's quite close. (Needs more work
in fixed point formatting.)
Convert all build scripts/makefiles to use a common
naming convention. All .a libraries will use the
base name 'libmapm' (libmapm.a). All .lib libraries
will use the basename 'mapm' (mapm.lib).
Include a short script to build a Linux shared
library. Other Unix systems should have similiar
commands. Most users won't need this.
Modify all error/warning messages so they have a
consistent format.
V 4.8.1 - Nov 30, 2003 Added support for National Instruments LabWindows
CVI. Some minimal help instructions on how to set up
various project files can be found in 'labwindw.cvi'.
This is the only update, so users of version 4.8
will have no compelling reason to upgrade.
V 4.9 - Jan 3, 2004 Simplify logic and make minor optimization in AGM
log algorithm.
Simplify logic in the exp algorithm. This also
resulted in the exp algorithm being more 'well
behaved'.
Include a new file (commentary.txt). This is just a
brief discussion on a design decision of why there
is (one) 'exit' call in the library.
Include 4 sample programs to compute PI (in the
directory 'pi_demo'). These are not meant to be
the fastest programs for PI, they are here simply
to demonstrate how to use the library. These could
be used as a starting template for a new program.
V 4.9.1 - Apr 2, 2004 Added a warning to the exp function if the input is
too large. If the input is too large, overflow can
occur internally so a warning will be output and
the result will be 0.
Made minor optimizations to the basic multiply and
the divide & conquer multiply.
Checked out the library with a new free compiler,
DEV-CPP (DEV-C++) 5.0 Beta 8 release 4.9.8.0. This
environment uses the mingw gcc compiler under the
hood, so everything worked fine using mkallmgw.bat
to compile the library.
Document a known COMPILER BUG with Microsoft's
Visual C++ 7.x (VS.NET 2003). This compiler bug
outputs error C2676 in a C++ application. (C apps
are OK). See the README file for a workaround.
Document the numerical limitations of the library.
(max/min significant digits, exponents, etc).
See the README file for more details.
V 4.9.2 - Jun 10, 2004 Added the % operator (and %=) to the C++ wrapper.
Added a MOD operator to 'calc' demo program.
Fixed a potential buffer overflow in the 'exp'
function.
V 4.9.5 - Dec 10, 2007 Updated software license, allowing distribution
of modified source code.
The actual library source code is identical to
version 4.9.2 (except the obvious library version
identifier, and Digital Mars #ifdef).
Added support for the Digital Mars Compiler.
Added build script for Intel Linux Compiler.
V 4.9.5a - Feb 21, 2010 Updated authors email address.
Note that there are no changes to the library
source code - so the library will still identify
itself as "4.9.5".
Added Makefile for Microsoft Visual Studio 2005.
Added another Makefile for MAC OSX (for 10.6).
**************************************************************************