version 2.2.1 as can be downloaded elsewhere

CHANGES

@@ -0,0 +1,92 @@
+version 2.2.1 -- 11/8/01
+
+     Alter the QR rotation routines supporting singular value
+     decomposition qrbdv.c, qrbdu1.c, and qrbdi.c to avoid a
+     rare numerical instability.
+
+     Modify the installation script makelibs.sh to ensure that
+     the new QR rotation code is not compiled with optimization.
+     The QR fix does not survive the GNU compiler's optimization.
+
+     Alter include in sany.c to read: #include "../ccmath.h".
+
+
+version 2.2.0 -- 12/13/00
+
+     Convert all function definitions to the modern ASCII style.
+
+     Revise test programs to support automatic checking of their
+     output against the sample output appended as a comment to
+     the test code source. Modifications were made to most of
+     the test code to support this.
+
+     Supply shell code to permit the automatic checks of standard
+     test outputs.
+
+     Corect the manual entry for the values returned by the
+     solv function (revision in C-01).
+
+
+version 2.1.2 -- 10/05/00
+
+     Modify installation scripts makelibs.sh and install.sh
+     to eliminate an undefined linker option.
+
+     Correct structure initializations in ccmath.h. (missing
+     braces)
+
+     Use <stdlib.h> to define all references to allocation
+     functions calloc() and malloc(). This header is now
+     included by ccmath.h.
+
+     Correct failures to free allocated storage in some eigenvalue
+     routines.
+
+     Change declarations of unsigned long to unsigned int
+     to correct problems on the DEC Alpha where unsigned long
+     has a length of 64 bits.
+
+     Revise ccmath.h and manual to reflect the altered
+     declarations of unsigned quantities. (Revisions in
+     C-07, C-10, and C-14.)
+
+     Fixed a bug in the singular value distribution code
+     that could lead to an attempt to operate with NANs. 
+
+
+version 2.1.1 -- 5/19/00
+
+     Revise hash storage programs to allow use of multiple
+     hash schemes in a program. ( Programs that use the
+     functions hashins, hashdel, hfind, and hval must be
+     revised to use this new form.)
+
+     Revised manual chapter C10-sort to reflect hash function
+     changes.
+
+     Minor correction in manual chapter C01-matrix.
+
+     Clean up anachronistic K&R type declarations of calloc
+     and malloc. (stdlib.h is now used)
+
+     Alter index incrementation code in time series functions
+     that resulted in warnings from DEC ALPHA compiler.
+
+
+version 2.1.0 -- 5/1/00
+
+     Add section containing benchmark test code and output.
+     ('benchmk' directory )
+
+     Specify use of '/bin/sh' in installation scripts
+
+     Change license policy to LPGL the GNU library public license.
+
+
+
+version 2.0.1 --  2/27/00 
+
+     Correct spelling error in script 'install.sh'.
+
+     3/19/00
+     Correct function name in '~/simu/supp/lranb.c' :  "lrana() -> lranb()".

INSTALL

@@ -0,0 +1,48 @@
+
+                          INSTALL CCMATH
+
+     The CCMATH Mathematics Library is installed using three shell scripts
+ located in the distribution directory containing this INSTALL file:
+
+          makelibs.sh ,    non_intel.sh ,   and      install.sh
+
+ These scripts should be run from the distribution directory.
+
+ NOTE: The installation scripts assume a shell that accepts Borne shell
+       syntax commands, such as 'bash' or 'ksh'. I They current specify
+       use of '/bin.sh'. You may alter this another compatible shell
+       by altering the first line in the script files.
+
+     1) SKIP THIS STEP IF YOUR TARGET IS INTEL COMPATIBLE!!
+        If you are installing the library on a platform that does
+        not use an Intel 80x86 compatible processor, run the script
+        non_intel.sh first. The script will query you on the byte
+        order employed by the target processor (little-endian for
+        INTEL and big-endian for many UNIX workstations). You
+        should know the answer before running non_intel.sh.
+        [ To test the byte order of your processor, compile
+          and run the program byteord.c , and compare the
+          output with sample output for a little-endian
+          processor appended as a comment to the byteord
+          source. If they differ, your processor is
+          big-endian. ]
+
+     2) Run the makelibs.sh script to compile the source code.
+        The script will query you on whether or not the target
+        processor is Intel compatible. If it is, respond by
+        entering 'y'. The script will compile both a static
+        library 'ccm.a' and a shared library 'lccm.so',
+        stored in the subdirectory 'tmp'.
+
+        Optimization flags and/or the compiler command used
+        can be altered by editing the script. This distribution
+        was tested using gcc version 2.7.2.3.
+
+    3) Run the install.sh script as the ROOT user to install the
+       header file and the libraries. The default locations are:
+
+          header:        /usr/include/ccmath.h
+          static lib:    /usr/lib/libccm.a
+          shared lib:    /lib/libccm.so
+
+       These locations can be altered by editing the script.

README

@@ -0,0 +1,83 @@
+                    CCMATH: A Mathematics Library
+                    -----------------------------
+                          Version 2.2.0
+
+                    developed and maintained by
+                        Daniel A. Atkinson
+                         <[email protected]>
+
+      This package contains the source and test code for the CCMATH
+ mathematics library. The library is a collection of functions, coded in
+ the C-language, that perform computations frequently encountered in
+ scientific, mathematical, and statistical applications. Library functions
+ are designed to exploit the efficiency and portability of C in such
+ applications. The following areas are covered:
+
+                       o Linear Algebra
+                       o Numerical Integration
+                       o Geometry and Trigonometry
+                       o Curve Fitting
+                       o Roots and Optimization
+                       o Fourier Analysis
+                       o Simulation Support
+                       o Statistical Functions
+                       o Special Functions
+                       o Sorts and Searches
+                       o Time Series Modeling
+                       o Complex Arithmetic
+                       o High Precision Arithmetic
+
+     The library code may be freely redistributed under the terms of the
+ GNU library public license (LGPL). The distribution directory contains a
+ copy of this license. Note that CCMATH is provides AS IS, with NO WARRANTY,
+ either expressed or implied.
+
+     Version 2.2.0 incorporates some major upgrades of the library designed
+ to facilitate its usability. First all function definitions have been
+ upgraded to the modern ASCII standard. This was already used in the
+ function declarations of the header file 'ccmath.h'. It enables compile
+ time checking of the types of function arguments. The test functions
+ have also been upgraded to support automatic comparison with sample
+ outputs. This a simple test of each function does not require tedious
+ comparison of printed numerical values. This is explained further in the
+ TESTING file in the current directory.
+
+     The 'manual' subdirectory contains documentation for each of the
+ functions in the library. The introductory Chapter 'C00-intro' is
+ recommended reading for all users. Each function is covered by a detailed
+ function synopsis. Chapters of the manual are indexed with the name of
+ the subdirectory containing the source code covered in the chapter.
+
+     Compilation and installation are covered in the INSTALL file in the
+ current directory. This should be fairly simple, but you will need to know
+ the answers to:
+
+             1) Is the target processor Intel compatible?
+             2) If not, is the byte order big-endian or not?
+
+     Each source code directory has a 'test' subdirectory with library
+ test code. This test code serves dual roles, since it permits tests of
+ library functions and it provides an example of how the function is used.
+ I strongly recommend inspection of the test code for functions you intend
+ to use. The modules tested by a test function are identified in the test
+ function's header, and a sample of test output is given in a trailing
+ comment.
+
+     The 'benchmk' directory contains drivers for timing the execution
+ of many library functions. The file 'linux_p2.tim' in this directory
+ presents results that you can compare with times obtained on your
+ hardware.
+
+     Problems encountered with CCMATH, comments on your use of the
+ package, or timing data you would like to share can be sent to me at
+ <[email protected]>. To properly address problems I will need the following
+ information:
+
+     a. The compiler used  to compile the package, and the target
+        processor.
+     b. Compiler options used if they differ from the defaults in
+        'mklibs.sh'.
+     c. Output from runs of test code for the suspect functions, if it
+        differs from the sample output in this distribution.
+     d. If possible, source code for the program where the problem was
+        discovered.

TESTING

@@ -0,0 +1,39 @@
+                        TESTING CCMATH
+
+     Two script files, runtest (a shell script), and testex (a perl script)
+ are supplied to facilitate comparison of test program output with the
+ sample output appended as a comment to each test source file. These
+ scripts should be placed in some directory on your executable path. You
+ should also check the first line of each script to and modify it if
+ the locations assumed /bin/sh and /usr/bin/perl are not correct.
+
+     To run a test and check its output you simply execute runtest in
+ the 'test' subdirectory containing the test source code.
+
+          runtest testsource.c test-command-line-parameters
+
+ The standard command line parameters for each test are tabulated in the
+ README file of the test subdirectory. This tabulation also specifies
+ the sequence of prompt responses (if any) required to generate the
+ standard test output. The output of the run is then compared, using
+ 'diff' to the standard output appended as a comment to the test source.
+ Output of a successful test will look like the sample below.
+
+                    1,2d0
+                    < /*  Test output
+                    < 
+                    36d33
+                    < */
+
+ It includes only the enclosing comment flags and possibly some blank lines.
+ Any differences in numerical values that appear may indicate a failed test.
+
+     Note that some compilers generate slightly different formats for
+ scientific (%e) notation, so differences in such numbers, flagged by 'diff',
+ should be inspected to see if they actually differ in value as well as
+ format.
+
+     This test capability simplifies and speeds up initial testing of
+ library functions. I urge you to inspect the test code and run some
+ alternative tests for any function that will play a critical role in
+ your application.

benchmk/README

@@ -0,0 +1,51 @@
+                        Benchmark Timing
+
+     This directory contains the source code for drivers that time the
+execution of important CCMATH functions. The header comments in each of
+the source files describes the input command line parameters required
+by the function. The file 'linux_2p.tim' contains timing data I obtained
+on a 233 MHz Pentium II.
+
+     The quoted times were obtained, after instalation of the ccmath
+ library, with programs compiled using
+
+     cc -O3 "bench".c -lccm -lm ,
+
+ where "bench " is one of the following drivers.
+
+
+     timbes.c ----- time the execution of Bessel functions
+     timeigv.c ---- time the computation of the eigenvalues
+                    and eigenvectors of a real symmetric matrix
+     timeval.c ---- time the computation of the eigenvalues of
+                    a real symmetric matrix
+     timfft2.c ---- time a radix-2 Fast Fourier Transform (FFT)
+     timfftgc.c --- time a general radix complex FFT
+     timheval.c --- time the computation of the eigenvalues of a
+                    hermitian matrix
+     timhevec.c --- time the computation of the eigenvalues and
+                    eigenvectors of a hermitian matrix
+     timhsrt.c ---- time the heap sort of a real array
+     timintg.c ---- time a numerical integration of complete
+                    elliptic integrals
+     timlpac.c ---- time the solution of a real linear system (LINPAC)
+     timmiv.c ----- time the inversion of a real matrix
+     timmsrt.c ---- time the merge sort of a real array
+     timmul.c ----- time the multiplication of real matrices
+     timnrml.c ---- time the generation of pseudorandom normals
+     timqsrt.c ---- time the 'quicksort' of a real array
+     timslpac.c --- time the solution of 'small' linear systems
+                    (dimension < 400)
+     timssrt.c ---- time the "Shell' sort of a real array
+
+     The following functions generate the binary matrix data files used
+     as input to the eigensystem timing routines.
+
+     hmatgen.c ---- generate a hermitian matrix with the specified
+                    dimension
+     smatgen.c ---- generate a real symmetric matrix with the specified
+                    dimension
+
+     The eigenvalues employed in these generators have a simple equal
+     spacing. This portion of the generator code can easily be changed
+     to generate matrices with more complex patterns of eigenvalues.

benchmk/hmatgen.c

@@ -0,0 +1,31 @@
+/*  hmatgen.c    CCMATH mathematics library source code.
+ *
+ *  Copyright (C)  2000   Daniel A. Atkinson    All rights reserved.
+ *  This code may be redistributed under the terms of the GNU library
+ *  public license (LGPL). ( See the lgpl.license file for details.)
+ * ------------------------------------------------------------------------
+ */
+/*
+    Create a Hermitian matrix with specified eigenvalues.
+
+    Input parameters:  size -> n  for an n by n matrix
+                       o_file -> name of output file
+*/
+#include <time.h>
+#include "ccmath.h"
+void main(int na,char **av)
+{ Cpx *h,*u; FILE *fp;
+  int i,n,m; double *e,s;
+  unsigned int seed;
+  if(na!=3){ printf("para: size o_file\n"); exit(1);}
+  n=atoi(*++av); m=n*n;
+  fp=fopen(*++av,"wb");
+  fwrite((void *)&n,sizeof(int),1,fp);
+  h=(Cpx *)calloc(2*m,sizeof(Cpx)); u=h+m;
+  e=(double *)calloc(n,sizeof(double));
+  for(i=0,s=1.; i<n ;++i,s+=.5) e[i]=s;
+  seed=(unsigned int)time(NULL); setunfl(seed);
+  unitary(u,n);
+  hmgen(h,e,u,n);
+  fwrite((void *)h,sizeof(Cpx),m,fp);
+}