CVM Class Library changes
- 5.6.1 April 17, 2008
- 1. New Features:
- Microsoft Visual Studio 2008 supported.
- 1. Bug fixes:
- Cygwin issues resolved.
- DLL linking is now used for all versions due to some issues discovered while calling CVM DLLs from other ones.
- 2. Documentation and package changes:
- New features are documented.
- All required redistributables are now parts of appropriate binary packages.
- 5.6 November 29, 2007
- 1. New Features:
- Symmetry and hermitianity tolerance added (see documentation for more details) as suggested by Claudio Piciarelli (thanks!).
- "Managed C++" compatibility added. Use /clr switch only while building your managed code (reported by Johan Vium, thank you).
- Native LAPACK/BLAS libraries built by Intel Fortran are now supported.
- Upgraded to Intel MKL 10.0.
- 2. Documentation changes:
- New features are documented.
- 5.5.1 July 4, 2007
- 1. Bug fixes:
- AMD-related issues fixed.
- Regression test utility coverage extended.
- 2. Documentation changes:
- Some functions were missed. Added.
- 5.5 June 10, 2007
- 1. New Features:
- QR factorization implementation added (contributed by Greg Reynolds, thank you).
- Upgraded to Intel MKL 9.1.
- 2. Bug fixes:
- Due to its obvious ambiguity, 'reserve' function is now deprecated (reported by Jianguo Lee, thank you).
- Matrices now can be used in STL containers (reported by Jianguo Lee, thank you).
- Missing libguide.so is added to Linux package (reported by Ville Kyrki, thank you).
- 3. Documentation changes:
- New features are documented.
- 5.4.1 January 9, 2007
- 1. New Features:
- AMD ACML 3.6.0 support added.
- Upgraded to Intel MKL 9.0.
- 2. Bug fixes:
- Intel finally fixed in MKL 9.0 their SVD bug causing crash for big non-square matrices.
- 3. Documentation changes:
- CVM documentation discrepancies about SVD algorithm are fixed.
- 5.4 September 23, 2006
- 1. New Features:
- Pseudo (generalized) inversion algorithm implemented for generic rectangular and band matrices (thanks to Greg Reynolds for the idea).
- Unix makefiles reorganized and cleaned up.
- STLport 5.0.2 compatibility added.
- 2. Bug fixes:
- Weird Intel compiler bug about passing complex numbers by referense is worked around.
- Dot product functions returning complex numbers are wrapped by subroutines.
- 3. Documentation changes:
- New functions and makefiles are documented.
- 5.3 July 17, 2006
- 1. New Features:
- Intel Fortran 9.1 and Intel C++ 9.1 compilers for Windows and Linux supported.
- GCC 4.1.x supported.
- MKL 8.1 linked in, no external dependencies.
- EM64T supported.Older compilers are no longer supported - please use the stable version 5.2.
- 5.2 June 3, 2005
- 1. New Features:
- 1.1 New sub-vector and sub-matrix assignment functions added.
- 1.2 MS Visual Studio 2005 support added.
- 2. Bug fixes:
- 2.1 Matrix input operator was missing - added (thank you, Marie).
- 2.2 Linux multithreading issues resolved.
- 2.3 Annoying linker error while mixing debug and non-debug versions is fixed.
- 2.4 svd routines now calculate working array's size automatically.
- 2.5 Minor memory allocation issues for band matrices are fixed.
- 3. Documentation changes:
- 3.1 New functions are documented.
- 5.1.1 March 20, 2005
- 1. New Features:
- 1.1 gcc 3.4 compatibility achieved.
- 2. Bug fixes:
- 2.1 *matrix::transp failed to run when called for itself - fixed.
- 5.1 November 15, 2004
- 1. New features:
- 1.1 Memory manager is no longer used by default.
- 1.2 CVMAllocator is deprecated. Faster std::allocator is used instead.
- 2. Documentation changes:
- 3.1 Memory manager changes are documented.
- 3. Package changes: new Win32 binaries have been generated.
- 5.0.2 October 24, 2004
- 1. Bug fixes:
- LAPACK/MKL documentation bug caused CVM to throw exceptions while calling "svd" function on ?ORGBR call. The problem was about work array size incorrectly defined in the documentation - the exceptions were thrown when either N>M*64 or M>N*64. It's fixed (thank you, Lee).
- G77 has a bug causing
FLOOR and CEILING functions to fail for arguments of type REAL*4 (C type float). It's worked around.
- 2. Package changes:
- New Win32 binaries have been generated.
- 5.0.1 October 8, 2004
- 1. Bug fixes:
std::complex<TR> operator * (std::complex<TR>, const type_proxy<T,TR>&) is added to resolve the issue with multiplying complex number values by type_proxy values.
- Minor issue with the warning generated by KDevelop IDE is fixed.
- 2. Package changes:
- No package changes needed, 5.0 binaries are valid to be used with 5.0.1 source code.
- 5.0 October 3, 2004
- 1. New features:
- Band matrices implementation.
- Symmetric matrices implementation.
- Hermitian matrices implementation.
- User defined allocators support.
- Full support of multithreading applications.
- Submatrices support.
- Numerous optimizations like removal of redundant memory clean-ups etc.
- 2. Bug fixes:
- Memory corruption issues caused by different working array sizes for LAPACK and MKL libraries are fixed.
- 3. Package changes:
- Singular value decomposition routines now return square matrices as sets of eigenvectors.
- Assignment operator is now overloaded for objects of the library only. See the documentation for further details.
- MKL 7.0 included. Now it could be used as sstatically linked library as well as a set of standalone DLLs.
- STLport 4.6.2 included.
- 4.1.1 May 9, 2003
- 1. Bug fixes and optimizations:
- 1.1 Matrix indexing operators threw exceptions for some non-square matrices. It's fixed (thank you Marcin).
- 1.2 MS .NET 2003 compatibility has been achieved.
- 1.3 Intel MKL 6.0 is now supported.
- 2. Package changes:
- 2.1 mkl_support.dll was removed due to library transfer to utilization of MKL 6.0.
- 2.2 MKL 6.0 redistributable dll files are included (in accordance with Intel Corp's License Agreement).
- 3. Documentation changes:
- 3.1 All package changes are documented.
- 4.1 August 25, 2002
- 1. Bug fixes and optimizations:
- 1.1 Sun WorkShop 6 C++ 5.3 and higher compilers are supported now.
- 1.2 cvector::eig caused crash of Borland C++ Builder. It's fixed.
- 1.3 Array<t_type>::norminf and Array<t_type>::norm2 functions are #ifdef'ed now in order to provide compatibility with Borland C++.
- 1.4 Some minor fixes were made in order to provide gcc 3.1 compatibility.
- 2. Package changes:
- 2.1 Some additional libraries for Borland C++ Builder were created and added.
- 2.2 KDevelop 2.1 sample project is created.
- 2.3 Dev-C++ sample project is created.
- 2.4
mkl_support.dll was upgraded to the version 5.2SP1.
- 3. Documentation changes:
- 3.1 All package changes are documented.
- 4.0.3 April 7, 2002
- 1. Bug fixes and optimizations:
- 1.1 Due to MS VC 6 glitches virtual functions
Array<t_type>::stream_in and Array<t_type>::stream_out were replaced by template friend operators >> and <<
- 1.2
srmatrix::det () and scmatrix::det () now catch cvmexception& instead of cvmexception.
- 1.3 Win32 implementation of the
CriticalSection class now contains mutex instead of critical section inside.
- 1.4
gPool global object of the MemoryPool class is no longer exported due to some unexplainable problems with MS VC 6.
- 1.5 Static object of class
CriticalSection is removed now from MemoryPool class to the global scope.
- 1.6 CygWin environment compatibility was achieved.
- 2. Package changes:
- 2.1
mkl_support.dll was upgraded to the version 5.2.
- 3. Documentation changes:
- 3.1 Template friend operators
>> and << are documented
- 4.0.2 February 7, 2002
- 1. Bug fixes:
- 1.1 In constructor
ComplexArray::ComplexArray (const treal*, const treal*, const int nSize, const int nIncr = 1) the nSize parameter was removed because of its uselessness.
- 1.2 Microsoft .NET compiler compatibility was achieved.
- 2. Package changes:
- 2.1
./lib/ftn*.lib and ./lib/lapack_missed*.lib libraries were added into Win32 package. It means that you can rebuild the library without any Fortran compiler.
- 3. Documentation changes:
- 3.1 .NET compatibility and package changes are documented.
- 4.0.1 November 24, 2001
- 1. Bug fixes:
- 1.1
CriticalSection::is_bad() now returns false in single-threaded apps.
- 1.2
MemoryPool::Malloc now catches bad_alloc exception.
- 1.3 Consts's constructor now calls
std::numeric_limits<treal> for MS VC++ as well.
- 1.4 Consts's members are protected now.
- 1.5
CVM_ZERO, CVM_ONE and CVM_TWO #define's are fixed now
- 1.6
MemoryPool::Malloc now begins with cs_enter (); call.
- 1.7 Some missing global utilities were added.
- 1.8
XERBLA callback routine implementation was fixed for Linux.
- 2. Package changes:
- 2.1 MKL 5.1 Service Pack 1 for Win32 was used.
- 2.2
mkl_support.dll was included into the Win32 package according to the Intel Corp's License.
- 3. Documentation changes:
- 3.1 Some previously missed global utilities are documented.
-
4.0 October 23, 2001
- 1. Complex number versions of all algorithms were added.
-
3.0 August 1998
- 1.
LAPACK support was added.
- 2. Memory allocation engine was added.
-
2.0 June 1996
-
1.0 September 1992
Back To Home