mirror of
https://gitlab.com/libeigen/eigen.git
synced 2026-04-10 11:34:33 +08:00
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.
296 lines
8.6 KiB
Plaintext
296 lines
8.6 KiB
Plaintext
/**
|
|
\page faq FAQ
|
|
|
|
<p>Contents:</p>
|
|
-# \ref license_tvmet
|
|
-# \ref commercial_tvmet
|
|
-# \ref debug
|
|
-# \ref optimize
|
|
-# \ref ambiguous_overload
|
|
-# \ref conversion_non_scalar
|
|
-# \ref could_not_convert
|
|
-# \ref assign_op
|
|
-# \ref comma_initializer
|
|
-# \ref no_match_operator
|
|
-# \ref dimension_error
|
|
-# \ref rtlx_kernel_crash
|
|
-# \ref regressiontest_failed
|
|
|
|
Certain items on this page are also covered by the \ref usage and \ref notes
|
|
pages!
|
|
|
|
|
|
|
|
\section license_tvmet How is tvmet licensed?
|
|
|
|
tvmet comes with completely free source code. tvmet is available under the
|
|
terms of the GNU Lesser General Public License (LGPL).
|
|
|
|
I have amended the LGPL to explicitly allow statically linking tvmet (or any
|
|
modified version of tvmet) to your software. The LGPL is not clear on this
|
|
and I definitely want to allow it.
|
|
|
|
\sa \ref license
|
|
|
|
|
|
|
|
|
|
\section commercial_tvmet Can I use it in commercial software products?
|
|
|
|
Yes, you can. The LGPL allows you to do this, and you do not need to release
|
|
the source code to your software. You do need to release the source code for
|
|
any modifications you make to tvmet itself. (We would hope you would send
|
|
any improvements like these back to us anyways.)
|
|
|
|
|
|
|
|
|
|
\section debug Debugging the code
|
|
|
|
Not all faults can be caught at compile time. Therefore, there is a need
|
|
for runtime debugging by defining <code>TVMET_DEBUG</code> (e.g. index
|
|
operators of <code>Vector::operator(i)</code> and
|
|
<code>Matrix::operator(i,j)</code>). On defined TVMET_DEBUG a bounds checking
|
|
is involved. Runtime errors are handled throwing an assertion failure using
|
|
std::assert(). <code>TVMET_DEBUG</code> is <b>not</b> enabled by defining
|
|
<code>DEBUG</code>. This behavior differs from release less than 0.6.0
|
|
|
|
If you don't get the expected result on certain operations you can try the
|
|
\ref expr_print feature to check that the expression is being evaluated as
|
|
you expect.
|
|
|
|
|
|
|
|
|
|
|
|
\section optimize Optimizing the code
|
|
|
|
Starting after tvmet 1.5.0 there is a new define <code>TVMET_OPTIMIZE</code>.
|
|
|
|
If this is defined tvmet uses some compiler specific keywords.
|
|
Mainly, this declares the functions using gcc's
|
|
<tt>__attribute__((always_inline))</tt>. This allows the
|
|
compiler to produce high efficient code even on less
|
|
optimization levels, like gcc's -O2 or even -O!
|
|
This is known to work with gcc v3.3.3 (and higher).
|
|
Using icc's v8 gnuc compatibility mode this may work, I've read
|
|
that it's using as an hint, this means you can have static inline
|
|
functions inside left so you can't get the full power of this feature
|
|
for circumstances.
|
|
|
|
|
|
|
|
|
|
\section ambiguous_overload ambiguous overload for ... Compiler Error
|
|
|
|
When using gcc-2.96 you can get a compiler error like:
|
|
|
|
\code
|
|
ambiguous overload for ...
|
|
/usr/include/g++-3/stl_relops.h:42: candidates are: bool operator>
|
|
...
|
|
\endcode
|
|
|
|
I haven't any solution for this--even if I don't use it in the tvmet library.
|
|
A simple define __SGI_STL_INTERNAL_RELOPS in the config header doesn't solve
|
|
the problem. The easiest way (brute force method) is to comment out all
|
|
operators in stl_reops.h. (The better way is to use the gcc-3.0.x and later,
|
|
see \ref compiler.)
|
|
|
|
|
|
|
|
|
|
|
|
\section conversion_non_scalar conversion from ... to non-scalar type ... Compiler Error
|
|
|
|
You get a compiler error like:
|
|
|
|
\code
|
|
conversion from `tvmet::XprVector< ... >, Sz>' to non-scalar type
|
|
`tvmet::Vector<T, Sz>' requested
|
|
\endcode
|
|
|
|
Please read about \ref construct. You probably ignored or forgot the rules
|
|
of using expression templates.
|
|
|
|
|
|
|
|
|
|
|
|
\section could_not_convert could not convert `tvmet::operatorXX(...)' to `bool' ... Compiler Error
|
|
|
|
You get a compiler error like:
|
|
|
|
\code
|
|
could not convert `tvmet::operator==( ... , ... )' to `bool'
|
|
\endcode
|
|
|
|
In the example above, you did try (or hoped) to use a global operator==
|
|
which doesn't return a bool. Please read \ref compare.
|
|
|
|
|
|
|
|
|
|
\section assign_op no match for ?= operator
|
|
|
|
You get a compiler error like:
|
|
|
|
\code
|
|
no match for `tvmet::Vector<...>& /= tvmet::Vector<...>&' operator
|
|
\endcode
|
|
|
|
For element-wise operations, you need to use the element_wise namespace
|
|
(found within tvmet). As the name suggests, all these operators are
|
|
performed in an element wise fashion. (e.g. division of two vectors--not
|
|
allowed from a mathematical perspective--element by element.) The following
|
|
code snippet shows how to use it:
|
|
|
|
\code
|
|
using namespace tvmet;
|
|
Vector<double,3> v1, v2;
|
|
|
|
v1 = 1,2,3;
|
|
v2 = v1;
|
|
|
|
cout << v1 << endl;
|
|
{
|
|
using namespace tvmet::element_wise;
|
|
v1 /= v2;
|
|
}
|
|
cout << v1 << endl;
|
|
\endcode
|
|
|
|
\sa \ref operators
|
|
\sa \ref alias
|
|
|
|
|
|
|
|
|
|
\section comma_initializer storage size of `ERROR_CommaInitializerList_is_too_long' isn't known
|
|
|
|
You get a compiler error like:
|
|
|
|
\code
|
|
In member function `tvmet::CommaInitializer<Obj, LEN>::Initializer<T, (N + 1)> ....
|
|
storage size of `ERROR_CommaInitializerList_is_too_long' isn't known
|
|
\endcode
|
|
|
|
You have caused a forced compile time error. tvmet prevents you from
|
|
overwriting foreign memory! In other words, your comma separated initializer
|
|
list is too long, e.g. you wrote something like:
|
|
|
|
\code
|
|
using namespace tvmet;
|
|
Vector<double,3> v0;
|
|
|
|
v0 = 1,2,3,4,5,6,7,8,9;
|
|
\endcode
|
|
|
|
You just tried to fill a %Vector of length 3 with 9 values, which would
|
|
normally result in values being written off the end of the vector. tvmet
|
|
prevents this with the compile time error you just saw.
|
|
|
|
|
|
|
|
|
|
\section no_match_operator no match for `...' operator
|
|
|
|
You get a compiler error like:
|
|
|
|
\code
|
|
no match for `tvmet::Matrix<T, Sz, Sz>& * tvmet::XprVector<tvmet::MatrixColVectorReference<T, Sz, Sz>, Sz>' operator
|
|
include/tvmet/xpr/VectorOperators.h:123: candidates are:
|
|
// a lot of candidates....
|
|
\endcode
|
|
|
|
Perhaps you wrote code like:
|
|
|
|
\code
|
|
Matrix<float,3,3> eigenvecs;
|
|
Matrix<float,3,3> M;
|
|
...
|
|
|
|
Vector<float,3> ev0( M * col(eigenvecs, 0) );
|
|
\endcode
|
|
|
|
Using %tvmet prior to release 1.2.0 in this case: Obviously an operator
|
|
is missing... but which one? Well, all arithmetic operators have a
|
|
functional equivalent, in this case, prod(). Rewriting the code above
|
|
using the function syntax will give you a more descriptive error message:
|
|
|
|
\code
|
|
no matching function for call to
|
|
`prod(tvmet::Matrix<float, 3, 3>&, tvmet::XprVector<tvmet::MatrixColVectorReference<float, 3, 3>, 3>)'
|
|
\endcode
|
|
|
|
This says, I forgot to write a function with the appropriate operator.
|
|
In this case, please give me a hint by writing a short example, used compiler
|
|
and tvmet release.
|
|
|
|
\sa \ref dimension_error.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
\section dimension_error no match for `tvmet::Matrix<double, R, C>& * tvmet::Vector<double, Sz>&' operator
|
|
|
|
You get a compiler error like:
|
|
|
|
\code
|
|
no match for `tvmet::Matrix<double, 4, 4>& * tvmet::Vector<double, 3>&' operator
|
|
candidates are: T tvmet::operator*(const T&, ...) ...
|
|
\endcode
|
|
|
|
This is a feature of tvmet. The compiler detects a mathematical problem:
|
|
You tried call an operator between tvmet objects which had incompatible
|
|
dimensions. For example, you attempted to multiply a matrix of dimension
|
|
4x4 with a vector of size 3 - this isn't mathematically feasible.
|
|
|
|
|
|
|
|
|
|
|
|
\section rtlx_kernel_crash Using with Linux Real-time Extensions crashes in kernel mode
|
|
|
|
Normally there should not be a problem on using tvmet inside Linux kernel
|
|
space as a C++ kernel module. Unfortunately, code working in normal
|
|
environment (e.g. Linux user space) crashes the Linux box by using linux
|
|
real-time extensions, definitely on <a href=http://www.rtai.org>RTAI</a>.
|
|
|
|
I haven't found the reason yet, my knowledge of assembler and debugging is
|
|
too limited.
|
|
|
|
|
|
|
|
|
|
|
|
\section regressiontest_failed Failed regression tests
|
|
|
|
Well, this is a strange world. It can happen that some (especially the
|
|
tan) regression test can fail. This is not a problem with the %tvmet library.
|
|
After some hours I reduce the problem to:
|
|
|
|
\code
|
|
cout << (std::tan(1.0) - std::tan(1.0)) << endl;
|
|
\endcode
|
|
|
|
got <tt>6.17995e-17</tt> on my Linux Mandrake 8.2/9.1 box. It makes no
|
|
difference if I take the tan function from namespace std or from the global
|
|
namespace. This is especially a problem for g++ v3.2 and prior. The Intel
|
|
compiler v7.0 isn't affected.
|
|
|
|
Especially on VC++ 7.1 you will get 2 failed tests for complex
|
|
%Matrix and %Vector. The error is about <tt>8.88178e-016</tt> and
|
|
<tt>1.77636e-015</tt>.
|
|
|
|
*/
|
|
|
|
|
|
|
|
// Local Variables:
|
|
// mode:c++
|
|
// End:
|
|
// LocalWords: RTAI
|