changelog

数值计算工具库,C语言编写的,可以直接调用.

- fixed core dump in testsuite/Olaf-Ronneberger-1, found by Wei Ku
- fixed namespace problems found by Martin Reinecke and Wei Ku
- fixed bug found by Wei Ku in loop collapse optimizations
- fixed bug found by Michael Aivazis in the array io; missing
  "return os;" in operator<<(ostream,Array<T,N>)
- fixed bug found by Peter Bienstman where expression involving
  empty arrays would loop forever
- SGI C++ 7.3 now supported (thanks Bill Homer)
- 'make clean' now removes blitz/config.h and config.cache
- fixed problem with polar(x,y) in compiler/compmath.cpp
- added real persistence: operator<< to output N-dimensional arrays
  to ostreams, operator>> to input N-dimensional arrays from istreams
- can now construct arrays from array expressions
- in theory, C++ Builder 4 is now supported.  Oleg Orlov found 2 bugs
  which have now been fixed.
- stencils are expression-templatized!! This means you can use stencils
  in array expressions, e.g. A = Laplacian2D(B) / (h*h);
- updated documentation, index added to postscript version
- new random number generators library: uniform, normal, exponential,
  beta, gamma, chisquare, F, discrete uniform (see random/ and the new 
  chapter in the user guide) 
- added new partial/full 1D reduction: last(X) returns index where
  X is last true
- support for egcs __restrict__
- incorporated Allan Stoke's Intel C++ patches
- new div, grad, curl, difference operators on vector fields; stencil
  patterns shown graphically in documentation
- new expression templates implementation is on by default now
- <blitz/numinquire.h> should work on all platforms now;
  I've included a hacked-up version of <limits> from libstdc++
  (blitz/limits-hack.h)
- Mersenne Twister RNG; see blitz/rand-mt.h; adapted by Allan Stokes
- nicer syntax for constructing Fortran-style arrays:
  Array<int,2> A(3,3,fortranArray);
- fixed reductions minIndex and maxIndex for multidimensional
  arrays; thanks Peter Nordlund
- faster compile times
- ConstArrayIterator, ArrayIterator: usable STL-style forward
  iterators for Arrays
- new math functions for complex arrays: arg, conj, polar
  (real and imag were provided in an earlier release)
- when creating an array from pre-existing data, now have an
  option of duplicateData, deleteDataWhenDone, neverDeleteData
- added user-defined expression template functions: see
  examples/useret.cpp
- added Array::reindex(), reindexSelf() contributed by
  Derrick Bass
- fixed bugs in Array::isStorageContiguous(), thanks Matthias
  Lindstr鰉
- fixed bug in stencils related to automatic determination
  of spatial extent; another due to integer literals
- fixed major bug in reverseSelf()
- fixed memory leak when creating arrays from pre-existing data
  (thanks Matthias Troyer and Pierre-Alain Genilloud)
- removed finite and trunc math functions (these apparently are
  only available under AIX?); thanks Prem Anand Manmohanrao
- incorporated several patches by Tim Brunne: typos in
  benchmarks/acousticf.f and benchmarks/acousticf2.f; guarded
  <fstream> inclusion in benchmarks/acoustic.cpp; removed $^
  in testsuite/Makefile (not recognized by osf make); 
- fixed bug w/ scoping of IEEE math functions
- added -fno-gcse to default options for egcs, will maybe minimize
   memory gobbling bug in egcs
- fixed problems with egcs and OSF related to _XOPEN_SOURCE_EXTENDED
  (thanks to Petter Urkedal for finding this mysterious bug)
- fixed problem with compiler/compmth2.cpp and compiler/compmath.cpp
  (thanks Andy Jewell)
- moved blitz/minmax.h into its own namespace (blitz::minmax) to avoid 
  conflicts in blitz/array/reduce.h; thanks Peter Nordlund
- fixed problem in array/stencilops.h, thanks Dr. Josef Grosch
- fixed problem with Array<T,N> when T does not have trival
  ctors/dtors; patches by Petter Urkedal

Version 0.4, August 6, 1998
- new benchmark results on a variety of platforms; see
  http://monet.uwaterloo.ca/blitz/benchmarks/
- indirection for arrays: A[point-list], A[subdomain-list], 
  A[indexSet(indexlist1,indexlist2,...)]
- new stencil objects, make finite differencing much nicer
- computational fluid dynamics example (currently broken)
- ported and tuned for DECcxx
- central, forward, and backward difference stencil operators
- curl, div, grad, laplacian stencil operators
- support for vector fields/stencil objects (vector versions of curl,
  div, grad)
- BZ_ENUM_CAST kludge
- fixed problems with math functions and namespaces
- implemented RectDomain<N>
- more tidying in blitz/compiler
- loop unrolling is now OFF by default
- incorporated patches from Theo Papadopoulo for setting compiler
  names and flags, and blas path
- vastly improved documentation
- added documentation for numeric inquiry functions

Version 0.3 alpha 01, April 12 1998
- added where(A,B,C) for arrays; as a replacement for ?: operator
- zip(expr1, expr2, T_component) to "zip" a multicomponent expression
  from two scalar-valued expressions
- added real(A) and imag(A) for complex arrays; can be used as lvalues
- added Array<T,N>::operator[](int), and ::chopComponent(), support for
  multicomponent arrays
- Implemented Array<T,N>::reverse()
- Implemented Array<T,N>::resizeAndPreserve()
- fixed various bugs found by aCC (thanks Len Lattanzi)
- beefed up compiler/namespac.cpp; egcs is just good enough to pass
  the old test but still doesn't handle full namespaces
- added section on multicomponent arrays to documentation
- split complex math functions into two parts; some of the ANSI-required
  functions are not provided by KAI C++
- added new array constructor to create an array from pre-existing data,
  with specified strides
- added Array<T,N>::free() to delete an array's data
- updated the documentation: global functions, platforms, fixed many
  errors
- cycleArrays has been changed from a method of Array<T,N> to a
  global function.  This may break existing code.  See manual
  for details (under "Global functions")
- The great source code reorganization has started.  arrayexpr.h,
  arraymap.h have become array/expr.h, array/map.h, etc.  This should
  make no difference to user-level code.
- NB: the semantics of transpose() have been changed!  This may break
  existing code.  Now have transpose() and transposeSelf(); reverse() and 
  reverseSelf().  See manual for details.
- Fixed oversight in type promotion: if one type is user-defined and the
  other is an intrinsic type, then promote to the user-defined type
- Implemented blitz::min(a,b) and blitz::max(a,b), with type promotion
  (blitz/minmax.h)
- Fixed bug with makeUnique() and arrays created from pre-existing data
- Fixed bugs associated with IEEE/SYSV math functions.  Some functions
  were called IEEE in one place and SYSV in another, which caused many
  problems.
- Fixed bug in compiler/instant.cpp; incorrect syntax for explicit
  instantiation request
- Fixed bug in benchmarks/looptest.cpp due to unrolling backwards
  loops incorrectly
- added PhysicalConstants.h and SystemOfUnits.h from CERN, by
  Michel Maire.  These should be supplemented with more constants
  from e.g. CRC handbook.  ** Update: these won't be included until
  a code sharing agreement with CERN is signed. **

Version 0.2 alpha 06, April 12 1998
- serious makeover of the Blitz++ web pages
- Blitz++ development list (blitz-dev) started; see
  http://seurat.uwaterloo.ca/blitz/contribute/
- interlaced arrays seriously implemented: interlaceArrays() and
  allocateArrays().  The first always interlaces; the second only
  interlaces if it's advantageous for the architecture (controlled
  by the BZ_INTERLACE_ARRAYS flag in <blitz/tuning.h>)
- new type promotion mechanism <blitz/promote.h>
- fixed use of bool (instead of _bz_bool) in <blitz/bzdebug.h>
- various minor bug fixes
- included benchmarks/looptest.cpp: tests a variety of C loop styles
  to find which one gives best performance.  This will be a start for
  automatic tuning someday.
- added lots of comments to <blitz/arrayeval.cc> to explain stack
  iteration
- finally, stable compiles with EGCS
- honour BZ_HAVE_NAMESPACES in examples and benchmarks
- will now build into a separate directory (thanks Brendan Kehoe)
- integrated GNU autoconf and configure utilities; many thanks to
  John W. Eaton and Brendan Kehoe for their help
- exciting new benchmark results: see web page
- stable compiles with egcs-980328
- bzconfig script will now run in non-interactive mode
- archive now unpacks into blitz-YYMMDD, instead of just Blitz++
- fixed bug related to globals in <blitz/bzdebug.h>
- fixed bug in unrolling of 1-D array expressions with common, non-unit
  stride <blitz/arrayeval.cc>, added check to test suite
- ColumnMajor changed to ColumnMajorArray to avoid conflict with
  matrix class of the same name

Version 0.2 alpha 05  March 13 1998
- Blitz++ is now distributed under the terms of the GNU General Public
  License
- solid EGCS support (see http://egcs.cygnus.com/)
- unfortunately, Cray C++ support broken by requiring <string>; this can
  be fixed if anyone really wants it.
- libblitz.a now contains global instances (so far, just tensor index 
  objects)
- one-step installation if using KAI C++ (cd Blitz++; make)
- various problems with multiple module programs fixed (I hope)
- partial integration with Tau profiling tools
  http://www.cs.uoregon.edu/research/paracomp/tau/
- total reductions
- arrays with different index sets may no longer be used in the
  same expression (e.g. adding a base-1 array to a base-0 array).
  This causes ambiguities for expressions such as sum(A+B+i)-- from
  which array should the index i take its values?
- shape checking for array expressions (in debug mode only)
- added pretty printing for array expressions; will be used for
  tracing and profiling, shape checking
- fixed bug in loop collapse optimizations which broke examples/storage.cpp
- minor changes to documentation
- added ColumnMajor<N_rank> array storage order
- full implementation of Array<T,N>::resize()
- Array<T,N>::permute() renamed to transpose()
- duplicate Array member function deprecated: length()
- start of automated test suite

Version 0.2 alpha 04  September 1997
- EGCS port (see http://www.cygnus.com/egcs/)
- many new performance benchmarks for vector & array operations
- more tuning of vector & array operations
- several bug fixes

Version 0.2 alpha 03  September 1997
- Intel C++ port (Windows'95 and NT).  Compiler test suite won't work;
  just copy <blitz/config-ICL.h> to <blitz/config.h>
- inlining groups (BZ_INLINE_GROUP1, BZ_INLINE_GROUP2) in <blitz/tuning.h>
  can be used to disable inlining of certain operations
- in evaluating array expressions, multidimensional loops are collapsed 
  to 1D when possible
- automatic tiling for two-dimensional array stencils
- Array ctor to create arrays from pre-existing data (e.g. Fortran arrays)
- new benchmark: stencil.cpp (3D, 7-pt stencil)
- new benchmark: loops.cpp (measures performance of various C loops)
- new benchmark: acou3d.cpp (3-D PDE)
- bug fix for subarrays of subarrays
- common & unit stride optimizations for array expressions
- new example: transform.cpp
- indirection now used for innermost loop of array expressions; is faster
  than pointer increments
- new benchmark: acoustic.cpp (2-D PDE)
- list initializers for Vector<T>
- fixed bug in Vector<T>::reverse()
- optimizations for PDEs: allocateArrays() (for interlaced arrays),
  cycleArrays() (for convenient array relabelling)
- compiler test suite now generates a log file

Version 0.2 alpha 02 August 1997
- Several additions to compiler test suite
- Cray T3E port (Cray C++ 3.0.0.0)
- Honour absence of <limits>, <set>, <complex>; needed for Cray & SGi
- Compatibility with old for scoping rules (needed for Cray C++)
- Array<T,N>::permute(..) implemented
- cross products for TinyVector<T,3>
- New example: <examples/fixed.cpp> creating arrays of user types
- Updated documentation
- New reduction: first(expr, index) returns the first index value for
  which expr evaluates true

Version 0.2 alpha 01  July 1997
- New to this release: Array<T,N>
- added <blitz/numinquire.h>
- fixed bug in TinyVector<T,N>::length()
- added TinyVector<T,N>(x1,x2,x3,...,xn) constructors

Version 0.1 alpha 04b
- fixed bug with index() in VectorPick
- fixed fortran compatability problems in benchmarks
- added qcd and haney benchmarks to distribution
- added check in examples/erf.cpp; this example won't compile
  without BZ_HAVE_IEEE_MATH

Version 0.1 alpha 04
- sum(), product() metaprograms for TinyVector
- min(), minValue(), minIndex(), max(), maxValue(), maxIndex()
  for TinyVector (loops not unravelled)
- norm(), norm1() for TinyVector (loop not unravelled)
- Added any(), all(), and count() for vector expressions
- Fixed bug in Vector<T>::makeUnique()
- Revised random number generators to take a template parameter
  for the uniform generator 
- Added discrete uniform generator, in <blitz/rand-dunif.h>

Version 0.1 alpha 03 February 1, 1997
- Added TinyVector class & expression templates support
- Added preconditions for mean(..) in <blitz/vecsum.cc>
- Tidied many #include directives to include header files only
  if necessary (#ifndef BZ_xxx ... #include ... #endif)
  Should reduce preprocessing time.

Version 0.1 alpha 02 January 27, 1997
- improved documentation
- complex<T> operands in expression templates are templated, rather than
  providing different specializations for float, double, long double
- full implementation of where(X,Y,Z) for vectors
- wrote new Benchmark class with external control model; in <blitz/benchext.h>
- fixed error in return type of mean() in <blitz/vecsum.cc>
- changed debug flag in <blitz/debug.h> to BZ_DEBUG (from just DEBUG)
- added some comments

Version 0.1 alpha 01 January 24, 1997
- alpha release: Vector<T>, expression templates, VectorPick<T>,
  Random<Normal>, Random<Uniform>, Benchmark & Timer