2008-10-08 Adam Piatyszek <ediap@users.sourceforge.net> * IT++ 4.0.6 release (SVN tag: release-4-0-6) * NEWS, VERSION: Add release notes and update version number for IT++ 4.0.6. * itpp/itpp_acml_tests/Makefile.am, itpp/itpp_mkl_tests/Makefile.am: Add specmat_test.vcproj files to the list of distributed files. 2008-10-05 Erik G. Larsson <erik_g_larsson@users.sourceforge.net> * doc/local/verification.doc: Added Ubuntu 8.04 to list of tested systems * doc/local/installation.doc: Minor fix 2008-09-17 Adam Piatyszek <ediap@users.sourceforge.net> * win32/itpp_acml.sln, win32/itpp_mkl.sln: Use CRLF line endings in MSVC++ solution files. * win32/itpp_acml_tests/itpp_acml_tests.sln, win32/itpp_acml_tests/specmat_test.vcproj, win32/itpp_mkl_tests/itpp_mkl_tests.sln, win32/itpp_mkl_tests/specmat_test.vcproj: Add MSVC++ project files for specmat_test. 2008-09-12 Adam Piatyszek <ediap@users.sourceforge.net> * tests/Makefile.am, tests/specmat_test.cpp, tests/specmat_test.ref: Add specmat_test program for testing special vectors and matrices. Initially only toeplitz() function is tested. * itpp/base/specmat.cpp, itpp/base/specmat.h: Fix broken implementation of toeplitz() function. The previous implementation generated incorrect matrices for two complex-valued input vectors. This patch fixes this problem. BTW, the implementation is now based on a template function, so bmat, smat and imat matrices can now be generated. Thanks to Niklas Johansson for reporting this issue. This closes bug report [2110119]. 2008-08-30 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/itassert.h: Get rid of config.h dependency from itassert.h header file. * itpp/comm/galois.cpp: Fix bug in GF::set_size() method, which was reported by Stephan Ludwig. Because "alphapow" and "logalpha" are static variables, every instantiation of the GF class uses the very same of them. If different instantiations of different Galois fields access this (e.g. 2^8 and 2^6, while 2^6 has been created first), the latter might overwrite/delete the log table of the previous while calling GF::set_size() method (2^8 widens the array to m+1=9, while not copying the 2^6 entries). The solution is to copy the contents of "alphapow" and "logalpha" when resizing them. Thanks to Stephan Ludwig for noticing this problem and also to EndZone for confirming it. * itpp/base/math/trig_hyp.h: Reduce number of divisions in a sinc() function. * itpp/base/sort.h: Make sort() and sort_index() work for vectors with less than two elements. Without this patch, the above two functions resulted in an assertion error. Thanks to Yngve Seln for reporting this problem. 2008-08-18 Erik G. Larsson <erik_g_larsson@users.sourceforge.net> * extras/gdb_macros_for_itpp, extras/Makefile.am: Added GDB debugger display macros for IT++ data types. This closes feature request [1913404]. Thanks to Vasek Smidl for providing an initial version of the macros. 2008-08-11 Adam Piatyszek <ediap@users.sourceforge.net> * win32/itpp_acml.vcproj, win32/itpp_acml_tests/*.vcproj, win32/itpp_mkl.vcproj, win32/itpp_mkl_tests/*.vcproj: Suppress warnings about deprecated "/Wp64" options in MSVC++ project files 2008-08-10 Adam Piatyszek <ediap@users.sourceforge.net> * doc/tutorial/src/ldpc_bersim_awgn.cpp: Add missing "LL" suffix to 64-bit inteager initialisation This eventually fixes integer overflow error in LDPC BER simulation tutorial example. 2008-08-09 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/stat/mog_diag.h: Suppress a minor warning reported by Doxygen * doc/tutorial/src/ldpc_bersim_awgn.cpp: Fix minor problem with integer overflow in expression The initialisation of int64_t variable with a multiplied inteager constants caused an integer overflow of the temporary calculation result. This patch fixes this issue by using one big constant in the initialisation. 2008-08-07 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/vec.h, tests/vec_test.cpp, tests/vec_test.ref: Fix regression [2041474] in Vec<>::split() method In IT++ 3.10.x releases it was possible to split a vector at its end, e.g.: vec v1 = "1 2 3 4 5"; vec v2 = v1.split(5); In the current stable branch (4.0.x) the above code results in an assertion. This patch fixes this regression and also improves the documentation, so it is clearly explained which elements are returned and which are stored in the original vector. Besides, the vec_test program now includes a new test for splitting a vector at pos = 0 and pos = datasize. Thanks to Yngve Selén for reporting this bug. 2008-08-02 Adam Piatyszek <ediap@users.sourceforge.net> * m4/acx_blas.m4: MKL requires zdotusub Fortran wrapper function on x86_64 architecture This fixes broken blas_test and vector_test on 64-bit linux architectures when IT++ is linked to Intel MKL BLAS library. This fix is equivalent to passing "--with-zdotu=zdotusub" option to configure. ------------------------------------------------------------------------------- 2008-07-18 Adam Piatyszek <ediap@users.sourceforge.net> * IT++ 4.0.5 release (SVN tag: release-4-0-5) * NEWS, VERSION: Add release notes and update version number for IT++ 4.0.5 * tests/vec_test.cpp, tests/vec_test.ref: Add a test-case to reproduce bug [2017948] * itpp/base/vec.cpp: Fix bug [2017948] in parsing error of "a:b:c" ivec initialisation There was a bug in vec.cpp that caused an assertion error when integer vectors were initialized using the "a:b:c" notation. For example: ivec iv = "0:2:4"; raised an exception. This patch fixes this bug by removing an unnecessary "seekg" operation. Thanks to David Hammarwall for reporting the problem and solution to it. 2008-07-11 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/mat.h: Fix bug [2004345] causing incorrect result of Mat::concat_vertical() In case when one of the input matrices to concat_horizontal() or concat_vertical() functions was empty, an error occurred. With this patch the result of concatenating an empty matrix with some other matrix is defined and a copy of the other matrix is returned. This fixes bug [2004345]. Thanks to Carlos Pineda for reporting this problem. * itpp/base/mat.h: Reset number of rows/columns if one of the dimensions given to Mat::set_size() is zero Without this patch, the Mat::set_size() function set number of rows or cols to non zero value, even if the other matrix dimension was zero. This was not compatible with the constructor. Besides, setting one of the dimensions of a matrix to non zero value, if the other one is zero does not makes sense and might cause problems with other functions operating on matrices. This patch fixes this problem. 2008-06-17 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/comm/bch.cpp: Fix bug [1995743] in BCH::decode() function The second argument of if ((delta == GF(n + 1, -1)) || (OldLambda.get_degree() > kk)) takes the degree of OldLambda. But get_degree() does not return the true degree of that polynomial but its allocated space. GFX::get_true_degree() does what we want. If OldLambda's "degree" is larger than its "true degree", a wrong decision is made in the if statement, resulting in a wrong Lambda(x), which in general leads to a decoding failure, while the code word should have been correctable in theory. This patch fixes this bug and also removes a redundant "temp" variable. Signed-off-by: Stephan Ludwig <donludovico@users.sourceforge.net> * itpp/base/mat.h, itpp/base/vec.h: Fix compilation bug caused by SVN commit rev.1578 2008-06-05 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/comm/modulator.cpp, itpp/comm/modulator.h: Remove redundant argument from calculate_softbit_matrices() function Since calculate_softbit_matrices() is a private method of a class, in which "bits2symbols" variable is also declared, there is no need to pass this variable as an argument of this method. * itpp/base/itfile.h, itpp/base/random.h, itpp/comm/bch.h, itpp/comm/rec_syst_conv_code.h, itpp/comm/reedsolomon.h: Suppress MSVC++ warning C4512: "assignment operator could not be generated" MSVC++ warns when an assignment operator can not be automatically generated because of constant variables existing in a class. To suppress this warning an dummy assignment operator can be added as in this patch. * itpp/base/vec.cpp, itpp/fixed/fix.cpp: Suppress MSVC++ warning C4701: "potentially uninitialized local variable..." * itpp/base/gf2mat.cpp, itpp/base/itfile.cpp, itpp/base/mat.h, itpp/base/math/misc.h, itpp/base/vec.h, itpp/comm/galois.cpp, itpp/srccode/audiofile.cpp, itpp/srccode/pnm.cpp: Suppress MSVC++ warning C4244: "conversion from 'int' to 'char'..." Some of the expressions require explicit casts to "char". Where appropriate "int get()" method of iostream class is replaced with "get(char &)" function. Besides, GF::set_size() method is re-factored a little bit, so the "mtemp" integer variable is no longer necessary. * itpp/base/factory.h, itpp/base/mat.h, itpp/base/matfunc.h, itpp/base/timing.cpp, itpp/base/vec.h, itpp/comm/bch.cpp, itpp/comm/channel.cpp, itpp/comm/convcode.cpp, itpp/comm/egolay.cpp, itpp/comm/hammcode.cpp, itpp/comm/ldpc.h, itpp/comm/modulator.cpp, itpp/comm/punct_convcode.cpp, itpp/comm/reedsolomon.cpp, itpp/fixed/cfix.h, itpp/fixed/cfixed.h, itpp/fixed/fix.h, itpp/fixed/fix_functions.h, itpp/protocol/tcp.cpp, itpp/protocol/tcp.h, itpp/stat/misc_stat.cpp, itpp/stat/mog_diag.cpp, itpp/stat/mog_diag.h, itpp/stat/mog_diag_em.cpp: Suppress MSVC++ warning C4100: "unreferenced formal parameter" This warning is suppressed by omitting named parameters in arguments of functions. Only types of parameters are passed. * configure.ac.in: Change default debugging flags to "-g -O1" or "-Wall -ggdb -O1 -pipe" The latter set is only used when GCC compiler is used, so it implicitly assumes that GDB debugger will be used in such a case. 2008-05-21 Adam Piatyszek <ediap@users.sourceforge.net> * doc/local/users_guide.doc: Minor documentation fix in Users Guide There is no such functions as swap_row() and swap_col() in IT++. swap_rows() and swap_cols() are the proper ones. This fixes bug [1968377]. Thanks to Allan Murray for reporting this issue. 2008-05-17 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/bessel/hyperg.cpp, itpp/base/itfile.cpp, itpp/comm/punct_convcode.cpp, itpp/fixed/cfix.cpp, itpp/fixed/fix.cpp, itpp/srccode/pnm.cpp, itpp/stat/mog_diag_em.cpp, itpp/stat/mog_diag_kmeans.cpp, itpp/stat/mog_generic.cpp: Fix GCC 4.3 warnings on "ambiguous else" due to missing braces The latest GCC compiler is much more picky when it comes to if/else statements without explicit braces. This results in a lot of warnings when "-Wall" flag is used. This patch adds missing braces or refactors code for which such warnings were reported by GCC 4.3. 2008-05-15 Adam Piatyszek <ediap@users.sourceforge.net> * configure.ac.in: Set CXXFLAGS_DEBUG to CXXFLAGS when --enable-debug is not used This new behaviour is consistent with setting the same set of libraries for both the debug and optimised cases, when --enable-debug is not used. Besides, the CXXFLAGS_DEBUG variable is now documented when "configure --help" is used. 2008-05-09 Erik G. Larsson <erik_g_larsson@users.sourceforge.net> * doc/tutorial/matlab_itpp.doc: Documentation improvement 2008-04-25 Adam Piatyszek <ediap@users.sourceforge.net> * tests/modulator_nd_test.cpp: Cosmetic fix - remove doubled semicolon * Unify coding style using an astyle source code formatter This huge patch is intended to unify the coding style for all IT++ source and header files. The formatting has been performed with the astyle v1.22 program, using the configuration stored in extras/astylerc file. 2008-04-15 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/vec.h: Fix regression of a missing inline keyword in outer_product() specialisation The inline keyword is necessary for template specialisations that are defined in the header file. This bug resulted in linking problems without any external BLAS library, when this particular specialisation was used. Thanks to Markus Dittrich for investigating this problem and providing a patch for it on Gentoo Bugzilla (bug #217621). 2008-04-14 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/signal/fastica.cpp: Fix broken initialisation of Fast_ICA::set_init_guess() method The set_init_guess() now changes initState, which is checked in separate(), so that the provided guess is passed into the solver, rather than a zero matrix as before. This patch fixes bug report [1941219]. Signed-off-by: Matthew Chaudhuri <mchaudhuri@users.sourceforge.net> 2008-04-08 Adam Piatyszek <ediap@users.sourceforge.net> * README, doc/local/index.doc.in: Reworded and rearranged the brief description of the IT++ library Thanks to Kumar Appaiah for some language suggestions. ------------------------------------------------------------------------------- 2008-03-31 Adam Piatyszek <ediap@users.sourceforge.net> * IT++ 4.0.4 release (SVN tag: release-4-0-4) * NEWS, VERSION: Add release notes and update version number for IT++ 4.0.4 * win32/itpp_acml_tests/Makefile.am, win32/itpp_mkl_tests/Makefile.am: Add converters_test.vcproj to EXTRA_DISTS variable of Makefiles Without this patch, the project files for converters_test were not included in the distribution package. 2008-03-26 Adam Piatyszek <ediap@users.sourceforge.net> * tests/itfile_test.cpp: Remove setf(ios::scientific) and setf(ios::fixed) from itfile_test.cpp These two formatting options causes broken output of floating-point data when using MSVC++ Express 2008. In spite of the fact that this seems to be a bug in MSVC++ Express 2008, removal of these formatting options fixes this problem. 2008-03-25 Adam Piatyszek <ediap@users.sourceforge.net> * tests/itfile_test.cpp, tests/itfile_test_data.it: Improve itfile_test by using a non-continuous reference data file Such a data file is prepared by first saving a longer data structure (ivec in this case), then deleting it and finally saving a shorter data structure in the same place. * itpp/base/itfile.cpp: Do not duplicate seek operation after reading data_header fields When reading each field of a data_header structure of an IT++ file, the reading position is automatically incremented. So, the additional seekg() operation afterwards is redundant. This small modification fixes a problem with endless loop while reading an itfile in programs compiled with MSVC++ 2008. Previous MSVC++ version (2005) does not show this strange behaviour. * tests/Makefile.am, tests/converters_test.cpp, tests/converters_test.ref, win32/itpp_acml_tests/converters_test.vcproj, win32/itpp_acml_tests/itpp_acml_tests.sln, win32/itpp_mkl_tests/converters_test.vcproj, win32/itpp_mkl_tests/itpp_mkl_tests.sln: Create a simple test program of various conversion functions This test should catch possible incompatible implementation of the rint() function, which is not available under MSVC++ compiler. * itpp/base/converters.cpp, itpp/base/converters.h: Fix incorrect implementation of rint() function used under MSVC++ The rint() function is not available under MSVC++ compiler. Thus, the local implementation has to be conditionally enabled for this compiler. This patch fixes the incorrect implementation of rint() function, which behaved like a similar round() function. It also moves the "double rint(double)" definition out of the itpp namespace. 2008-03-20 Adam Piatyszek <ediap@users.sourceforge.net> * win32/itpp_acml.vcproj, win32/itpp_mkl.vcproj: Do not treat warnings as errors in MSVC++ projects This is a workaround for MS Visual C++ Express 2008 problem in compiling a correct piece of templated code. * itpp/base/algebra/cholesky.h, itpp/base/algebra/det.h, itpp/base/algebra/eigen.h, itpp/base/algebra/ls_solve.h, itpp/base/algebra/lu.h, itpp/base/algebra/qr.h, itpp/base/binary.h, itpp/base/binfile.h, itpp/base/copy_vector.h, itpp/base/gf2mat.h, itpp/base/itassert.h, itpp/base/itfile.h, itpp/base/ittypes.h, itpp/base/mat.h, itpp/base/sort.h, itpp/base/svec.h, itpp/base/vec.h: Minor spelling fixes in IT++ base module documentation This commit fixes several little spelling mistakes in the IT++ base module documentation. Signed-off-by: Kumar Appaiah <akumar@ee.iitm.ac.in> 2008-03-18 Adam Piatyszek <ediap@users.sourceforge.net> * tests/ldpc_test.cpp, tests/ldpc_test.ref: Minor workaround for a failing ldpc_test on win32 platforms MinGW and MSVC++ produce win32 executables, which use slightly different rounding approach for real numbers output through cout or cerr I/O streams. This patch just changes the precision of the output values, so the problematic case does not occur any more. * itpp/base/operators.cpp: Fix complex constructors in operators.cpp to build with g++ 4.3 It appears that g++ 4.3 is a bit more strict with respect to type checking when we attempt to overload the std::complex constructor for use with non-double types. Use of an explicit static_cast is needed to overcome this restriction. This patch does the needful by converting those constructors to use static_casts. Signed-off-by: Kumar Appaiah <akumar@ee.iitm.ac.in> 2008-03-13 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/vec.h: Remove checks for different sizes before invoking set_size() function set_size() methods first checks if the requested new size differs from the current one. Therefore, there is no need to check this condition explicitly before invoking set_size() method. * itpp/base/mat.h, itpp/base/vec.h: Decrease the number of inline functions in the Vec and Mat classes Overusing the inline keyword usually causes unnecessary code bloat, which results in a rather slower performance of the resulting code. Therefore, this patch reduces the number of inline functions and lefts only the ones which are really trivial or frequently reused. * itpp/base/factory.cpp, itpp/base/factory.h, itpp/base/sources.mk, win32/itpp_acml.vcproj, win32/itpp_mkl.vcproj: Mark all create_elements() functions as inline ones These functions have 2-4 lines and are only used by alloc() and free() methods only, so we might gain more by inlining them instead of more general functions of Vec, Mat and Array classes. * itpp/base/mat.cpp, itpp/base/mat.h: Add in_range(r,c), col_in_range(c) and row_in_range(r) inline methods This is a pure clean-up patch which simplifies it_assert_debug() checks. BTW, it also includes a few cosmetic changes to other assertion checks. * itpp/base/mat.cpp, itpp/base/mat.h: Make consistent naming of variables used in the Mat class interface This patch unifies the variable names in the Mat class interface. The following names are now commonly used: - m, m1, m2, etc.: matrix - v: vector - t: simple numeric type Num_T, e.g. bin, int, double - r, r1, r2: row index - c, c1, c2: column index - i: generic index * itpp/base/mat.cpp, itpp/base/mat.h: Clean-up redundant use of "const" keyword in the Mat class There is no need for declaring the returned by value types with an additional "const" keyword. Moreover, Num_T is assumed a simple or built in type for the Mat class, i.e. bin, short int, int, double and complex<double>. Therefore it might be more efficient to pass arguments of this type by value, in which case the "const" keyword is also redundant. * itpp/base/vec.cpp, itpp/base/vec.h: Clean-up redundant use of "const" keyword in the Vec class There is no need for declaring the returned by value types with an additional "const" keyword. Moreover, Num_T is assumed a simple or built in type for the Vec class, i.e. bin, short int, int, double and complex<double>. Therefore it might be more efficient to pass arguments of this type by value, in which case the "const" keyword is also redundant. Logical operators (==, !=, >, >=, < and <=) no longer create a local copy of the vector contents. By the way, use consistent name "t" for Num_T variables and wrap long lines before 80 column. * itpp/base/random.cpp, itpp/base/random.h: Move the implementation of Normal_RNG::sample() to the source file The implementation of Normal_RNG::sample() function is too complicated for inlining, so it should not be included inside the class declaration. This patch moves the actual code from the header file to the source file. * itpp/base/mat.h, itpp/base/vec.h: Fix it_assert_debug() conditions in elem_mult_out() methods of the Vec and Mat classes The checks for equal sizes of three and more variables can not be written like this: "a.size == b.size == c.size". The proper form is: "(a.size == b.size) && (a.size == c.size)". This patch fixes this bug. It also removes the redundant checks for different dimensions of a matrix before invoking the set_size(rows, cols) method. Thanks to Martin Senst for reporting this bug. 2008-03-12 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/comm/ldpc.cpp, itpp/comm/ldpc.h: Simplify BLDPC_Parity::calculate_base_matrix() function The previous implementation used to create a temporary dense parity check matrix and then extracted Z by Z submatrices from it for further processing. With this patch, the temporary Z by Z submatrices are sparse and extracted directly from the sparse parity check matrix H. Besides, the checks for cyclic-shifted identity matrices are simplified. Therefore, an additional method circular_eye_b() is no longer needed. * itpp/comm/ldpc.cpp: Simplify BLDPC_Parity::expand_base() implementation The previous implementation used to create a temporary dense parity check matrix and then copied its ones to the final sparse matrix. This patch changes the code so the sparse parity check matrix is constructed directly from the input base matrix. * tests/ldpc_test.cpp, tests/ldpc_test.ref: Add simple tests of BLDPC_Parity and BLDPC_Generator classes 2008-03-06 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/comm/modulator.h, tests/modulator_test.cpp, tests/modulator_test.ref: Fix bug [1908644] in Modulator<>::set(symbols, bits2symbols) function The bitmap matrix used in the demodulate_bits() function of the Modulator<> base class was improperly constructed from the input bits2symbols mapping vector. This resulted in demodulation errors for some constellations, e.g. 64-QAM or 8-PSK. This patch fixes this problem. Thanks to Yann Cocheril for reporting this problem. 2008-03-04 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/comm/llr.cpp, itpp/comm/llr.h: In case of an overflow saturate QLLR values instead of aborting An overflow may occur when converting real-valued LLR values into QLLR ones and also when calculating the Hagenauer's "boxplus" operator. This patch implements a conditional QLLR saturation in such cases. Moreover, the Boxplus() function is no longer an inline function, because this actually caused performance loss of the LDPC codes decoder. Finally, make the argument of to_double(QLLR) and to_qllr(double) pass-by-value instead of const-reference, since that's faster for POD like double and int. Thanks to Martin Senst for providing the initial patch. 2008-03-02 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/vec.h: Replace ((i < datasize) && (i >= 0)) with in_range(i) inline method This is a pure clean-up patch which simplifies it_assert_debug() checks. BTW, it also includes a few cosmetic changes to other assertion checks. 2008-02-28 Adam Piatyszek <ediap@users.sourceforge.net> * AUTHORS, doc/local/authors.doc: Update list of contributors with Martin Senst * win32/itpp_acml_tests/Makefile.am, win32/itpp_acml_tests/error_count_test.vcproj, win32/itpp_acml_tests/itpp_acml_tests.sln, win32/itpp_mkl_tests/Makefile.am, win32/itpp_mkl_tests/error_count_test.vcproj, win32/itpp_mkl_tests/itpp_mkl_tests.sln: Add MSVC++ project files for test program of BERC and BLERC classes * tests/error_count_test.cpp: Clean-up redundant casts to short integer * tests/error_count_test.cpp: Use constructor's initialisation list in BERC and BLERC constructors * tests/Makefile.am, tests/error_count_test.cpp, tests/error_count_test.ref: Add simple test program for BERC and BLERC error counters * itpp/comm/error_counters.cpp, itpp/comm/error_counters.h: Make const member functions actually const Introduce the keyword const for member functions like get_errorrate() that do not change any member variables. Patch submitted by Martin Senst. * itpp/comm/channel.h: Unhide Fading_Generator::generate(int) method for derived classes The "cvec Fading_Generator::generate(int no_samples)" function used to be hidden. This change makes it accessible from derived classes. Patch submitted by Martin Senst. * itpp/base/converters.h: Add specialisations of to_cvec(cvec) and to_cmat(cmat) These specialisations make it possible to call to_cvec() and to_cmat() with complex arguments. The functions then simply return their argument. This fixes the problem with instantiating the Freq_Filt class with std::complex<double> type. Patch submitted by Martin Senst. * itpp/base/math/elem_math.h: Documentation improvements Clarify that sqr(cvec) and sqr(cmat) compute the absolute square, not simply the square. Patch submitted by Martin Senst. * itpp/comm/ldpc.cpp: Correct the length of the result of decode() The decode() function used to return the first "ncheck" bits of the decoded codeword, but we want the first "nvar-ncheck" bits. Thanks to Martin Senst for submitting this patch. 2008-02-25 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/vec.h: Add support for "-1" indexing in Vec<>::del(i1, i2) method The indexing operator(i1, i2) of Vec class supports "-1" indexes denoting the last element of a vector. This feature was missing in the del(i1, i2) method, and this patch fixes this. It also fixes the incorrect range check (i1 <= i2). By the way, remove the redundant local variables ii1 and ii2 from the operator(i1, i2) method. ------------------------------------------------------------------------------ 2008-02-21 Adam Piatyszek <ediap@users.sourceforge.net> * IT++ 4.0.3 released (SVN tag: release-4-0-3) * NEWS, VERSION: Release notes added and version number updated for IT++ 4.0.3 * itpp/base/vec.h: Minor documentation fixes to prevent warnings of the Doxygen 2008-02-20 Adam Piatyszek <ediap@users.sourceforge.net> * INSTALL, doc/local/installation.doc, itpp/base/blas.h, itpp/base/vec.cpp, itpp/base/vec.h, itpp/config_msvc.h, m4/acx_blas.m4: Removed non-portable "--with-zdotu=complex" method It seems that there is no portable solution for calling from C++ the Fortran functions that returns a complex value. C99 introduces "double _Complex" type, which is compatible with Fortran COMPLEX. But this C99 type is not a valid C++ type. Therefore this patch removes the "complex" method from "--with-zdotu" option. The valid methods are now: - "zdotusub" - use a zdotusub_ Fortran wrapper function - "void" - compatible with BLAS libraries built with g77 compiler - "none" - do not use zdotu_ function at all 2008-02-19 Adam Piatyszek <ediap@users.sourceforge.net> * INSTALL, configure.ac.in, doc/local/installation.doc, itpp/base/blas.h, itpp/base/vec.cpp, itpp/base/vec.h, itpp/config_msvc.h, m4/acx_blas.m4: Add "--with-zdotu" option to select zdotu_ calling convention The "--with-zdotu=<method>" option can be used to change the default calling convention of BLAS zdotu_ Fortran function. The BLAS libraries built with g77 compiler do not return the complex result of a function, but pass it via the first argument of the function. The implementations compiled with gfortran and other vendor compilers return complex result as a typical function. By using a "void" method with the "--with-zdotu=<method>" option, the former approach is used (g77 compatible). To use the zdotu_ call that returns the result by value, a "complex" method should be passed. Without setting any of the two methods explicitly, IT++ tries to guess the proper method based on the detected BLAS implementation. Otherwise, a Fortran wrapper function zdotusub_ is used, which requires a working Fortran compiler. Finally, if this is not the case, the zdotu_ BLAS method is not used at all and a relevant warning message is displayed during the configuration step. 2008-02-18 Adam Piatyszek <ediap@users.sourceforge.net> * autogen.sh: Cosmetic improvements Replace back-ticks `...` with POSIX shell syntax $(...). BTW, improve generation of a snapshot version string. * itpp/base/blas.h, itpp/base/vec.cpp, itpp/base/vec.h, itpp/config_msvc.h, m4/acx_blas.m4, m4/ax_func_zdotu.m4: Use zdotusub_ Fortran wrapper by default This patch removes the run-time checks for correct zdotu_ calling conventions, which caused a lot of portability problems in IT++ 4.0.1 and 4.0.2 releases. The preferred method is now to use the locally provided zdotusub_ Fortran wrapper (as in 4.0.0 release). This requires the availability of a Fortran compiler, unless an Intel MKL library is used, in which zdotu_ can be called directly from C++ without any problems. If no Fortran compiler is available and Intel MKL is not used, the zdotu_ function is not called at all. In such case, a relevant warning message is printed during the configuration step. * configure.ac.in, itpp/base/Makefile.am, itpp/base/sources.mk, itpp/base/zdotusub.f, m4/acx_blas.m4: Revert "Removed redundant zdotusub_ Fortran wrapper to zdotu_ function" 2008-02-15 Adam Piatyszek <ediap@users.sourceforge.net> * VERSION: Updated for SVN snapshot releases ------------------------------------------------------------------------------ 2008-02-15 Adam Piatyszek <ediap@users.sourceforge.net> * IT++ 4.0.2 released (SVN tag: release-4-0-2) * NEWS, VERSION: Release notes added and version number updated for IT++ 4.0.2 * ChangeLog-2007: Archive ChangeLog entries from year 2007 2008-02-14 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/parser.cpp, tests/parser_test.cpp, tests/parser_test.ref: Fix wrong verbose output of Parser's get() function for int and bool types When Parser was used to scan an already defined int or bool variable and the parsed variable was not found, the verbose output was wrong. Here is a minimum example: Parser p(argc, argv); int i = 5; bool b = true; p.get(a, "a"); p.get(b, "b"); The wrong output of this program was: i = []; b = ; instead of: i = 5; b = 1; This patch fixes this problem and also improves the parser_test program, to detect such issues in future. Thanks to Jia-Yin for reporting this problem. * itpp/comm/convcode.h: Documentation improvements related to encoder state. This patch improves the documentation of the encoding and decoding functions. Especially, it clarifies why set_start_state() and init_encoder() methods have no effect on encode_tail() and encode_tailbite() functions. * itpp/comm/modulator.h: Improve the documentation of Modulator::get_bits2symbols() function 2008-01-31 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/gf2mat.cpp: Fix GF2mat_sparse_alist::from_sparse() conversion function The conversion function from GF2mat_sparse format to "alist" text file format could incorrectly create the "mlist" and "nlist" matrices with indexes of non-zero entries. This patch fixes this bug and also improves the conversion speed by using the get_nz_indices() function from Sparse_Vec class and set_row() instead of append_row() where possible. 2008-01-23 Adam Piatyszek <ediap@users.sourceforge.net> * Makefile.am, VERSION, autogen.sh: Optionally append SVN revision to IT++ package version When bootstrapping IT++ sources from SVN or Git repositories append the SVN revision number to the package version string. This feature is triggered by including the additional "svn" keyword in at the end of the first line of the VERSION file. BTW, remove snapshot and snapshot-html top Makefile's targets. 2008-01-17 Adam Piatyszek <ediap@users.sourceforge.net> * configure.ac.in, itpp/Makefile.am: Fix improper default shared and static settings Besides, do not add "-no-undefined" libtool switch to global LDFLAGS. Use NO_UNDEFINED substituted variable instead. 2008-01-06 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/math/log_exp.h: Fix bug [1863940] in log_add() function for infinite arguments When both arguments of log_add() function were either inf or -inf, the result was NaN instead of +/-inf. This patch fixes this issue. 2008-01-04 Adam Piatyszek <ediap@users.sourceforge.net> * itpp/base/math/misc.h: Add missing itpp namespace description Without this patch, Doxygen 1.5.4 did not generate any documentation for the itpp namespace. 2008-01-01 Adam Piatyszek <ediap@users.sourceforge.net> * configure.ac.in, extras/*.m, itpp-config.1.in, itpp/**/*.h, itpp/**/*.cpp, tests/*.cpp: Copyright dates updated to `1995-2008' * configure.ac.in: Add support for building a DLL library on Cygwin When libblas and liblapack are installed from Cygwin repository, it is possible to build IT++ as a DLL library. This requires "-no-undefined" flag to be passed to the linker.