lapackpp-devel

[Lapackpp-devel] Appreciation and suggestion

[Aosheng R. <ro...@gm...>]

Many thanks go to Dr.Stimming for adding the inverse of complex-value matrix
to lapackpp.  As Dr. Stimming said that  many  times  we could  ignore  the
inverse  of  a matrix  and   re-arrange algorithms for the same  purpose.
But for some  cases, we are lazy to do so and rely on a direct inverse of
the matrix.

LAPACKPP is definitely one of the most useful packages  for engineering
applications.

Right now,  I have a random thought that may increase the significant
efforts of Dr. Stimming. Also I am not sure if such a thought makes sense to
the other people.  My thought is Just for discussion.

Thought:  I wonder If the COMPLEX type in LAPACKPP  is  identical with the
complex<double> in GNU C++.   Right now,  the Complex type in LAPACKPP is
the structure - type, instead of class - type, as in C++.  Within the
LAPACKPP class, Class-Type COMPLEX in C++ is converted to the Structure-Type
COMPLEX, compatible with  Fortran  LAPACK. The direct benifit from such a
definition of Class-Type COMPLEX is that one can do operations on the
product and division using  complex<double> type in GNU C++,  once results
are obtained from the LAPACKPP. With the present definition of COMPLEX, we
have to do term-by-term calculation based  on the decomposition of real and
imaginary parts.

I appreciate the wonderful job of Dr. Stimming who brings us  an easy to
implement  interface  of  LAPACK  with  C++.

Regards,
-------------
Rong

Re: [Lapackpp-devel] Appreciation and suggestion

[Christian S. <sti...@tu...>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Aosheng Rong schrieb:
> Many thanks go to Dr.Stimming for adding the inverse of complex-value
> matrix to lapackpp.  

Did you actually retrieve the code from CVS and test it? If yes, please
confirm so that we can be sure this code is working correctly.


> Thought:  I wonder If the COMPLEX type in LAPACKPP  is  identical with
> the complex<double> in GNU C++.   

No, it is not.
http://lapackpp.sourceforge.net/html/lacomplex_8h.html
http://lapackpp.sourceforge.net/html/classla_1_1complex.html

> Right now,  the Complex type in
> LAPACKPP is the structure - type, instead of class - type, as in C++. 
> Within the LAPACKPP class, Class-Type COMPLEX in C++ is converted to the
> Structure-Type COMPLEX, compatible with  Fortran  LAPACK. The direct
> benifit from such a definition of Class-Type COMPLEX is that one can do
> operations on the product and division using  complex<double> type in
> GNU C++,  once results are obtained from the LAPACKPP. With the present
> definition of COMPLEX, we have to do term-by-term calculation based  on
> the decomposition of real and imaginary parts.

You need to manually convert the COMPLEX values into a LaComplex value,
which will then have the automatic conversion operators to
std::complex<double>, as follows:

  std::complex<double> z;
  z = z * LaComplex( myMatrix(0, 2) );

which will do the automatic conversion for you. (At least I *thought*
this should work.)

Christian
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.1 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iQCVAwUBROV2QmXAi+BfhivFAQJZ7wP+P9fEOrujLVMwi5vDQ+KnGvydX/8sYeQK
sWjDpFgaaNkM0ECQsOhDtJBNC48NETdV57Z9T0nJ0Y1nsLKtbZHGBaTvAFnkPgpc
L6upCNCM+KYOhU4HOD5qO8veuHLuwce+7OXIpsQ5EmsQuiWJyNU05AyL3kqXXLzx
n9fdzjh02mM=
=mk4V
-----END PGP SIGNATURE-----

Re: [Lapackpp-devel] Appreciation and suggestion

[Aosheng R. <ro...@gm...>]

Christaian,

Yes, I VCS lapackpp 2.4.12 with an addition of inverse of the complex-value
matrix. I tested that it works on linux. But I have trouble on windows.

I compiled lapackpp 2.4.12 on windows and run your tGenSolve with Testing
Dir. It works. But when I compile my application that is transplanted into
Windows from Linux, I got the following compiling errors(See below). My
application has been succefully compiled under Linux using lapackpp 2.4.12.

-------------------------------------------------------------Compiling Error
Messages-------------------------------------------------------------------------
1>------ Build started: Project: Test5, Configuration: Debug Win32 ------
1>Compiling...
1>SplitPowerPlane_driver.cc
1>e:\gnu\lapackpp-2.4.12\include\gmc.h(101) : warning C4251:
'LaGenMatComplex::v' : class 'VectorComplex' needs to have dll-interface to
be used by clients of class 'LaGenMatComplex'
1>        e:\gnu\lapackpp-2.4.12\include\vc.h(58) : see declaration of
'VectorComplex'
1>e:\gnu\lapackpp-2.4.12\include\gmc.h(665) : error C2375: 'operator <<' :
redefinition; different linkage
1>        e:\gnu\lapackpp-2.4.12\include\gmc.h(580) : see declaration of
'operator <<'
1>e:\gnu\lapackpp-2.4.12\include\gfqrc.h(54) : warning C4251:
'LaGenQRFactComplex::_tau' : class 'LaVectorComplex' needs to have
dll-interface to be used by clients of class 'LaGenQRFactComplex'
1>        e:\gnu\lapackpp-2.4.12\include\lavc.h(51) : see declaration of
'LaVectorComplex'
1>e:\gnu\lapackpp-2.4.12\include\gfqrc.h(56) : warning C4251:
'LaGenQRFactComplex::_work' : class 'LaVectorComplex' needs to have
dll-interface to be used by clients of class 'LaGenQRFactComplex'
1>        e:\gnu\lapackpp-2.4.12\include\lavc.h(51) : see declaration of
'LaVectorComplex'
1>e:\gnu\lapackpp-2.4.12\include\laexcp.h(51) : warning C4275: non
dll-interface class 'std::runtime_error' used as base for dll-interface
class 'LaException'
1>        d:\microsoft visual studio\vc\include\stdexcept(149) : see
declaration of 'std::runtime_error'
1>        e:\gnu\lapackpp-2.4.12\include\laexcp.h(50) : see declaration of
'LaException'
1>SplitPowerPlane.cc
1>e:\gnu\lapackpp-2.4.12\include\gmc.h(101) : warning C4251:
'LaGenMatComplex::v' : class 'VectorComplex' needs to have dll-interface to
be used by clients of class 'LaGenMatComplex'
1>        e:\gnu\lapackpp-2.4.12\include\vc.h(58) : see declaration of
'VectorComplex'
1>e:\gnu\lapackpp-2.4.12\include\gmc.h(665) : error C2375: 'operator <<' :
redefinition; different linkage
1>        e:\gnu\lapackpp-2.4.12\include\gmc.h(580) : see declaration of
'operator <<'
1>e:\gnu\lapackpp-2.4.12\include\gfqrc.h(54) : warning C4251:
'LaGenQRFactComplex::_tau' : class 'LaVectorComplex' needs to have
dll-interface to be used by clients of class 'LaGenQRFactComplex'
1>        e:\gnu\lapackpp-2.4.12\include\lavc.h(51) : see declaration of
'LaVectorComplex'
1>e:\gnu\lapackpp-2.4.12\include\gfqrc.h(56) : warning C4251:
'LaGenQRFactComplex::_work' : class 'LaVectorComplex' needs to have
dll-interface to be used by clients of class 'LaGenQRFactComplex'
1>        e:\gnu\lapackpp-2.4.12\include\lavc.h(51) : see declaration of
'LaVectorComplex'
1>e:\gnu\lapackpp-2.4.12\include\laexcp.h(51) : warning C4275: non
dll-interface class 'std::runtime_error' used as base for dll-interface
class 'LaException'
1>        d:\microsoft visual studio\vc\include\stdexcept(149) : see
declaration of 'std::runtime_error'
1>        e:\gnu\lapackpp-2.4.12\include\laexcp.h(50) : see declaration of
'LaException'
1>Generating Code...
1>Build log was saved at "
file://f:\SplitPowerPlane\Test5\Test5\Debug\BuildLog.htm<file:///f:/SplitPowerPlane/Test5/Test5/Debug/BuildLog.htm>
"
1>Test5 - 2 error(s), 8 warning(s)
========== Build: 0 succeeded, 1 failed, 0 up-to-date, 0 skipped ==========


--------------------------------------------------------------End of Error
Messages
----------------------------------------------------------------------------

It seems that gmc.h has problem with complex operators. I need more tests to
make sure where the errors come from: lapackpp or my application.


Regards,
-------------
Aosheng

Re: [Lapackpp-devel] Appreciation and suggestion

[Christian S. <sti...@tu...>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Aosheng Rong schrieb:
> I compiled lapackpp 2.4.12 on windows and run your tGenSolve with
> Testing Dir. It works. But when I compile my application that is
> transplanted into Windows from Linux, I got the following compiling
> errors(See below). My application has been succefully compiled under
> Linux using lapackpp 2.4.12.
>  
> 1>e:\gnu\lapackpp-2.4.12\include\gmc.h(665) : error C2375: 'operator <<'
> : redefinition; different linkage
> 1>        e:\gnu\lapackpp- 2.4.12\include\gmc.h(580) : see declaration
> of 'operator <<'

Any error about "linkage" on Windows refers to missing "DLLIMPORT"
macros. In this case the error says that in gmc.h:665 the operator<< is
defined *with* DLLIMPORT whereas in gmc.h:580 it is defined without it.
Solution: Change line 580 so that it reads

      friend DLLIMPORT std::ostream& operator<<(std::ostream&, const
LaGenMatComplex&);

I've changed this in CVS, so if you simply make a "cvs update" in your
existing copy from CVS, then you will get this change already.

Christian
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.1 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iQCVAwUBROXX8WXAi+BfhivFAQJUkQP/a+u+MTkX+lHjk3eLd6KOYT/EKimNiXaZ
EcDA0IsN5lGJM2DWF3RZrH8LVDQsg3DLiSaTFFJSN5p/GcPym/Uu4WEC3BGxso70
FR1LSizTyGlvdx8K7/1tvaE+HqEV1bqm/LT5XfJa4eTmrqEY/8uozRzG6UMsYxTk
Jb4mB8+eJhM=
=yk1S
-----END PGP SIGNATURE-----

Re: [Lapackpp-devel] Appreciation and suggestion

[Aosheng R. <ro...@gm...>]

What I have done with CVS.

Basically I follow your instructions and CVS.

These are what I did:

**********************************************************************
1) Visit  https://sourceforge.net/cvs/?group_id=99696

When I am required to CVS, I choose the packagename as lapackpp.

2)
  ./autogen.sh
  ./configure
  make dist

3)  Move lapackpp 2.4.12.tar.gz from Linux to Windows

4) Unpack .tar.gz and Compile it.

The Compiling of lapackpp package on window  looks successful. I get dynamic
library lapackpp.dll and static library lapackpp.lib.  I compile
tGenSolve.cc under Testing dir and run it. The results look correct.  The
problem right now may be related to complex-value matrix. The error messages
also indicate the errors come from the complex-value matrix in gmc.h.

BTW, I use the same lapackpp package and the same application program for
testing on Linux and Windows.

I would do more tests and try to find the problems. I would report my
testing status.

Regards,