3036eeca0a
release of tvmet (inactive for 2 years and developer unreachable) as the basis for eigen2, because it provides seemingly good expression template mechanisms, we want that, and it would take years to reinvent that wheel. We'll see. So this commit imports the last tvmet release.
191 lines
6.9 KiB
Plaintext
191 lines
6.9 KiB
Plaintext
* release 1.7.1
|
|
|
|
- major and minor bug fix release.
|
|
- element wise operations on non square matrix expressions.
|
|
does work as expected now.
|
|
- tvmet compiles with gcc 3.4.3.
|
|
- extended docs.
|
|
- new regression tests.
|
|
|
|
* release 1.7.0
|
|
|
|
- support for Microsoft VC++ 7.1
|
|
- several critical bug fixes.
|
|
- docs enhanced and corrected; new chapter for VC++ 7.1.
|
|
- conj for signed types disabled, since an unary minus operator
|
|
applied to unsigned type will result unsigned type.
|
|
- drem, hypot, jn, yn, cbrt and rint are inside the global namespace
|
|
now.
|
|
- some fixes for Makefile on creating pdf documentation.
|
|
- header depencies limited.
|
|
|
|
* release 1.6.0
|
|
|
|
- major bug fix: tvmet did crash on zero matrix print (introduced
|
|
with release 1.5.0, there was only a patch available).
|
|
- corrected and extended docs.
|
|
- new regressions added for traits.
|
|
- some compiler specific optimizations for gcc and icc
|
|
(compatibility mode) introduced, use it by defined TVMET_OPTIMIZE.
|
|
This uses the gcc's __attribute__((always_inline)) wrapped by a
|
|
macro - this allows to produce high optimized and inline code
|
|
even on lower optimization levels, e.g. -O. This is enabled by
|
|
configure process by --enable-optimize.
|
|
- For pgCC 5.1 the TVMET_HAVE_LONG_DOUBLE has been disabled, since
|
|
the support is incomplete imo. For icc some pragma are used
|
|
to avoid warnings about 'operands are evaluated in unspecified
|
|
order' since it concerns ostream printing only.
|
|
- traits prepared to be an entry point of several functions
|
|
used in the feature.
|
|
- new functions c{matrix,vector}_ref to handle C style vectors
|
|
and arrays as expressions - using inside the tvmet Matrix/Vector
|
|
expressions as usual.
|
|
- tvmet's syntax/naming convention more cleaner, code should be
|
|
more readable now. Further more unused template parameter has
|
|
been removed.
|
|
- new function identity<>() to create identity matrices.
|
|
- some cosmetic changes.
|
|
|
|
* release 1.5.0
|
|
|
|
- the build process is more compliant to the GNU standard.
|
|
- alias function extended.
|
|
- Matrix and Vector data printing improved, gives better output
|
|
even by negative values/contents.
|
|
- configure script supports different compiler better; users
|
|
can specify --enable-optimize, --enable-debug and
|
|
--enable-warnings for their needs.
|
|
- the configure and build process has been improved generally;
|
|
tvmet can be build in a separate build directory and passes
|
|
the distcheck target successfully.
|
|
|
|
* release 1.4.1
|
|
|
|
- new function alias to solve the aliasing problem.
|
|
- major bug fix: forgotten header tvmet/loop/Vector.h added
|
|
and benchmark shows AtA and AAt graph with matrices
|
|
less than 10x10 too - the BTL patch is updated.
|
|
|
|
* release 1.4.0
|
|
|
|
- major and minor bug fixes
|
|
- Matrix/Vector and XprMatrix/XprVector uses an assign_to()
|
|
function - this could be a way for pre-evaluate/caching
|
|
nodes of the sub expressions for further releases.
|
|
- an alternate way of computing matrix-matrix and matrix-vector
|
|
products added: using meta templates or loops. Therefore some
|
|
trigger are introduced which are not tuned yet. Using
|
|
matrix-matrix products less than Rows*Cols 8*8 or
|
|
matrix-vector products less than Sz=8 uses meta templates
|
|
else loops.
|
|
- regression tests added for loop products, as well as bugs
|
|
fixed.
|
|
- enums Rows, Cols and Size for Matrix and Vector are
|
|
introduced.
|
|
- expression level printing improved - more info's are shown
|
|
- output format has been changed, the output is compatible
|
|
to octave/matlab. To get the type and size info as before
|
|
a new member function info() has been introduced which can
|
|
be streamed to ostreams.
|
|
- improved data printing, aligns output regarding of maximum
|
|
element.
|
|
- different members for expression and data printing.
|
|
- BTL benchmark results added.
|
|
|
|
* release 1.3.0
|
|
|
|
- major and minor bug fixes - all users should update !
|
|
- RVO problem/bug removed, chaining/composed expressions are
|
|
working now. The problem was related on use of temporaries
|
|
on composed functions, which gone out of scope.
|
|
- sources compiled without optimizations, by -O0, doesn't crash
|
|
any more. The Problem was related by holding expressions by
|
|
references instead by value. At higher optimization levels
|
|
there were no faults than.
|
|
- better english documentation (spelling and grammar) as well
|
|
some corrections and extensions.
|
|
- benchmark removed from sources.
|
|
- interface changes.
|
|
- prodTrans renamed to trans_prod to follow the naming convention
|
|
- boolean version of eval removed to avoid confusion, comparing
|
|
is done using all_elements and any_elements.
|
|
- support for complex<> slightly improved.
|
|
|
|
* release 1.2.1
|
|
|
|
- major and minor bug fixes.
|
|
- corrected and extended docs.
|
|
- tvmet's const correctness improved.
|
|
|
|
* release 1.2.0
|
|
|
|
- major and minor bug fixes
|
|
- better gcc 2.95.3 support but, still problems
|
|
- function naming convention from product to prod, transpose
|
|
to trans etc. changed.
|
|
- new functions add,sub,mul and div on element wise functions.
|
|
- expressions used for row/col and diag vectors on matrices,
|
|
prevents use of temporaries on expressions for pre-evaluation.
|
|
- Vector access using braces () is default on internal use,
|
|
brackets [] are still valid and supported.
|
|
- better debug support.
|
|
- regression test improved and more tests added, especially
|
|
on expressions.
|
|
- more functions and operators on functions.
|
|
- better header file layout.
|
|
|
|
* release 1.1.0
|
|
|
|
- major and minor bug fixes
|
|
- new meta templated functions for matrix operations for
|
|
product(transpose(M), M), product(M, transpose(M)) and
|
|
transpose(product(M,M)), namely MtM_product(), MMt_product
|
|
as well as productTransposed().
|
|
- productTranspose(M,V) renamed to Mtx_product(). In this
|
|
function is a return dimension bug removed (working for non-
|
|
square matrices too now).
|
|
- Therefore, old productTransposed() functions doesn't have
|
|
the old meaning any more!
|
|
- a matrix expression can be transposed too.
|
|
- new functions and operators for products of expression of
|
|
matrices and vectors added - chaining of this should work
|
|
now.
|
|
- better support for non-square matrices and non-square
|
|
matrix-matrix and matrix-vector functions/operations.
|
|
- regression test improved and more tests added.
|
|
- file and directory structure extended.
|
|
|
|
* release 1.0.1:
|
|
|
|
- minor bug fixes
|
|
|
|
* release 1.0.0:
|
|
|
|
- corrected and improved documents
|
|
- new product(XprMatrix, Vector) and operator*(XprMatrix, Vector)
|
|
as well as product(XprMatrix<>, XprVector<>),
|
|
- more regression tests
|
|
- better intel compiler support
|
|
- extended include directory structure
|
|
|
|
* release 0.9.0:
|
|
|
|
- the benchmark is delayed.
|
|
- new function product(XprMatrix<>,XprMatrix<>) and operator*
|
|
(XprMatrix<>,XprMatrix<>), therefore concating of mathematical
|
|
expressions are better supported now.
|
|
- Addendum to the LGPL, static linking is explicit allowed.
|
|
- regression tests for element wise operations added.
|
|
- docs FAQ enhanced.
|
|
- new function diag(Matrix<>) added.
|
|
- autoconf 2.5x ready.
|
|
- default distribution is tar.bz2 now.
|
|
- tvmet-doc-xxx.prm target removed, docs are inside the rpm
|
|
package self.
|
|
|
|
|
|
|
|
|
|
|
|
%%% Local IspellDict: "english"
|