Re: [Blitz-support] Is this a bug, or am I missing something?

[Fernando Perez <fperez@co...>]

Todd Veldhuizen wrote:
> Maybe it's best to visualize an outer product followed by a reduction.
> The indices refer to the dimensions of the outer product.
[snip]
> Your example:
> M(i1,j) * T(i2,i3,i4,j)
> This is a five dimensional array (i1,i2,i3,i4,j)
> 
> sum(M(i1,j) * T(i2,i3,i4,j), j)
> This is a four dimensional array (i1,i2,i3,i4)
> 
> Hope that helps!

Thanks a lot, and it helps indeed.  But I have a bit of a worry now, and it's 
performance.  I am wondering if blitz is making an intermediate rank d+1 array 
above before the reduction, or something.

Here is the problem.  Now that I understood how to fix the code, I went ahead 
and timed the two versions: my hand-rolled loops (horrendous macro code), vs 
the blitz-expanded one (beautiful, elegant one-liners).

The results are:

planck[~/mtinner]> ./mtinner 300
Memory usage requested: 300 MB
Rank    size    dif     t_hand  t_bz    ratio (bz/hand)
1       6270    0       0.49    0.4     0.816327
2       340     0       0.36    0.41    1.13889
3       79      0       0.36    0.53    1.47222
4       33      0       0.4     0.7     1.75
5       18      0       0.35    0.82    2.34286
6       12      0       0.38    1.17    3.07895

Times are in seconds.  I estimate the size at rank 1 to fit in the requested 
memory, and other sizes are computed to keep the total op-count approximately 
constant (memory use goes down with rank).

The hand-rolled loops do indeed remain constant in time, but the 
blitz-expanded code doesn't.  This worries me a lot, since a factor of 3 is 
unfortunately, far more than I can afford to pay once I move to production 
with this project.

I'd really love to be able to use blitz's fancy syntax with abandon, and I 
wonder if I am doing something wrong, or if this is something which can be fixed.

These numbers are on a P-4@..., with 1 GB of RAM, RedHat 9.0. Code compiled 
with gcc 3.2.2 (also tested 3.3.1, no significant difference), and

CXXFLAGS = -I$(BZINCDIR) -DNDEBUG -O3
CFLAGS = -DNDEBUG -O3

Next, I decided to test the Intel compiler (icc 7.0).  Here's what I get, with 
SSE2 optimizations turned on:

planck[~/mtinner]> ./mtinner 300
Memory usage requested: 300 MB
Rank    size    dif     t_hand  t_bz    ratio (bz/hand)
1       6270    0       0.58    0.36    0.62069
2       340     0       0.67    0.37    0.552239
3       79      0       0.79    0.47    0.594937
4       33      0       1.02    0.61    0.598039
5       18      0       0.92    0.7     0.76087
6       12      0       1.11    1.1     0.990991

It's quite interesting:  icc is a bit faster than gcc for the templated code, 
but far worse for the hand-rolled loops.  In the end, the blitz ratio is 
better wiht icc, but for all the wrong reasons :)

The opcount is approximately constant, so what gcc + hand loops gives is the 
'ideal' situation: time _should_ remain constant.

In case I am doing something outright wrong, I've posted a tarball with the 
necessary code so others can run the tests:

http://www-hep.colorado.edu/~fperez/tmp/mtinner.tgz

Just 'make', then './mtinner NNN' will run with NNN Mbytes of memory (if NNN 
is not given, 50 is the default).

In summary:  I'd love to know if it's possible to use the fancy tensor 
notation with no performance penalty, and if so, how to do it.  The project 
I'm working on requires algorithms to be written for d=1..6, and that syntax 
would make my life *much*, *much* easier.

Thanks in advance for any feedback.

Regards,

Fernando.

ps.  And to the Blitz team, many thanks.  Even if the tensor notation can't be 
used for real work, I'm still amazed by blitz, and I'll keep using it with 
loops.  It's a joy to work with :)

Re: [Blitz-support] Is this a bug, or am I missing something?

[Todd Veldhuizen <tveldhui@os...>]

Maybe it's best to visualize an outer product followed by a reduction.
The indices refer to the dimensions of the outer product.

Some examples:

firstIndex i;
secondIndex j;
thirdIndex k;
fourthIndex l;


A(i,j) * B(i,j)
This is a two-dimensional array (i,j)

sum(A(i,j) * B(i,j), j)
This is a one dimensional-array (i)

A(i,j) * B(k,l)
This is a four-dimensional array (i,j,k,l)

A(i,k) * B(k,j)
This is a three-dimensional array (i,j,k)

sum(A(i,k) * B(k,j), k)
This is a two-dimensional array (i,j)


Your example:
M(i1,j) * T(i2,i3,i4,j)
This is a five dimensional array (i1,i2,i3,i4,j)

sum(M(i1,j) * T(i2,i3,i4,j), j)
This is a four dimensional array (i1,i2,i3,i4)

Hope that helps!

Cheers,
Todd
-- 
Todd Veldhuizen  /  tveldhui@...  /  Indiana University Computer Science

On Fri, 29 Aug 2003, Fernando Perez wrote:

> Matthias Schillinger wrote:
> > declare j as thirdIndex and it should work, I think.
> > See also the user guide on page 57.
>
> Thanks a lot!  It does indeed work correctly.  Even though I read the docs
> several times, I just misunderstood something.
>
> However, perhaps more detail on the semantics of these magical index objects
> might be worth having in the manual, since I still find it quite confusing.
>
> The manual says:
>
> "There is a distinct index placeholder type associated with each dimension of
> an array. The types are called firstIndex, secondIndex, thirdIndex, ...,
> tenthIndex, eleventhIndex."
>
> This is fine when all your objects have the same rank.  But for example, the
> 4d version of my code (which thanks to your comments now works correctly), reads:
>
> // 4d version
> void mat_ten_inner0(Array<double,2>& M,
> 		   Array<double,4>& T, Array<double,4>& U ) {
> 	firstIndex i1;
> 	secondIndex i2;
> 	thirdIndex i3;
> 	fourthIndex i4;
>
> 	fifthIndex j;
>
> 	U = sum(M(i1,j)*T(i2,i3,i4,j),j);
> }
>
> j has to be declared as a fifthIndex, even though nothing is 5-dimensional
> here.  And i2,i3,i4 play the role of the 2nd, 3rd and 4th dimension for U, but
> one less each for T.
>
> So exactly what the semantics of these objects is, remains a bit of a mistery
> to me.  I'd love to understand it, though, as they make the code an absolute
> joy to write (once it works ;)
>
> Many thanks again for your help.
>
> Regards,
>
> Fernando.
>
> _______________________________________________
> Blitz-support mailing list
> Blitz-support@...
> http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support
>

Re: [Blitz-support] Is this a bug, or am I missing something?

[Fernando Perez <fperez@co...>]

Matthias Schillinger wrote:
> declare j as thirdIndex and it should work, I think.
> See also the user guide on page 57.

Thanks a lot!  It does indeed work correctly.  Even though I read the docs 
several times, I just misunderstood something.

However, perhaps more detail on the semantics of these magical index objects 
might be worth having in the manual, since I still find it quite confusing.

The manual says:

"There is a distinct index placeholder type associated with each dimension of 
an array. The types are called firstIndex, secondIndex, thirdIndex, ..., 
tenthIndex, eleventhIndex."

This is fine when all your objects have the same rank.  But for example, the 
4d version of my code (which thanks to your comments now works correctly), reads:

// 4d version
void mat_ten_inner0(Array<double,2>& M,
		   Array<double,4>& T, Array<double,4>& U ) {
	firstIndex i1;
	secondIndex i2;
	thirdIndex i3;
	fourthIndex i4;

	fifthIndex j;

	U = sum(M(i1,j)*T(i2,i3,i4,j),j);
}

j has to be declared as a fifthIndex, even though nothing is 5-dimensional 
here.  And i2,i3,i4 play the role of the 2nd, 3rd and 4th dimension for U, but 
one less each for T.

So exactly what the semantics of these objects is, remains a bit of a mistery 
to me.  I'd love to understand it, though, as they make the code an absolute 
joy to write (once it works ;)

Many thanks again for your help.

Regards,

Fernando.

Re: [Blitz-support] Is this a bug, or am I missing something?

[Matthias Schillinger <matthias@tk...>]

Hi Fernando,

declare j as thirdIndex and it should work, I think.
See also the user guide on page 57.

Regards,
Matthias

> // 2d version
> void mat_ten_inner0(Array<double,2>& M,
>                     Array<double,2>& T, Array<double,2>& U ) {
>      firstIndex i0;
>      secondIndex j;
>
>      secondIndex i1;
>      // We want for(i0,i1) U(i0,i1) = sum(...,j): j is the sum index
>      U = sum(M(i0,j)*T(i1,j),j);
> }

[Blitz-support] Is this a bug, or am I missing something?

[Fernando Perez <fperez@co...>]

Hi all,

the strange behavior I was seeing with the old Blitz code is still there, so 
I'd like to know if I am misusing the library in some way.

I am trying to compute the product of a rank-2 array M with a rank-d array T, 
by summing over the 2nd index of M and the last one of T:

U(i1,i2,...,id) = sum(M(i1,j)*T(i2,i3,...,id,j),j),

or in tensor notation

U             = M      T
  i1,i2,...,id     i1,j  i2,i3,...,j

I had initially written code to do this by hand-rolling nested loops with 
preprocessor macros (I need d=1..6), but this is super-ugly.  I later read the 
documentation, and found that Blitz supports essentially the above notation 
unchanged.  So I wrote a new version using this, but it doesn't work correctly 
for d>1 (d=1 is ok).

I'd like to know if I am misusing the library (case in which perhaps the docs 
need some clarification, b/c I read them fairly carefully many times), or if 
this is indeed a problem.

The relevant part of my code is (showing d=1,2,3):

// 1d version
void mat_ten_inner0(Array<double,2>& M,
                    Array<double,1>& T, Array<double,1>& U ) {
     firstIndex i0;
     secondIndex j;

     U = sum(M(i0,j)*T(j),j);
}

// 2d version
void mat_ten_inner0(Array<double,2>& M,
                    Array<double,2>& T, Array<double,2>& U ) {
     firstIndex i0;
     secondIndex j;

     secondIndex i1;
     // We want for(i0,i1) U(i0,i1) = sum(...,j): j is the sum index
     U = sum(M(i0,j)*T(i1,j),j);
}

// 3d version
void mat_ten_inner0(Array<double,2>& M,
                    Array<double,3>& T, Array<double,3>& U ) {
     firstIndex i0;
     secondIndex j;

     secondIndex i1;
     thirdIndex i2;

     // We want for(i0,i1,i2) U(i0,i1,i2) = sum(...)
     U = sum(M(i0,j)*T(i1,i2,j),j);
}

I can post a fully buildable test case if anyone wants to run it without 
having to write anything themselves.

I made a debug build, and here are the results (a non-debug build gives 
incorrect results for d=2 and segfaults for d=3):

planck[bug]> ./mtinner
Rank 1 size 3500 dif 4.40307e-12   // Dif is the difference with hand loops
[Blitz++] Precondition failure: Module 
/home/fperez/usr/local/blitz/include/blitz/array/eval.cc line 129
Assigned rank 1 expression to rank 2 array.
mtinner: /home/fperez/usr/local/blitz/include/blitz/array/eval.cc:129: 
blitz::Array<T, N>& blitz::Array<T, N>::evaluate(T_expr, T_update) [with 
T_expr = 
blitz::_bz_ArrayExpr<blitz::_bz_ArrayExprReduce<blitz::_bz_ArrayExpr<blitz::_bz_ArrayExprOp<blitz::_bz_ArrayExpr<blitz::ArrayIndexMapping<double, 
2, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0> >, 
blitz::_bz_ArrayExpr<blitz::ArrayIndexMapping<double, 2, 1, 1, 0, 0, 0, 0, 0, 
0, 0, 0, 0> >, blitz::Multiply<double, double> > >, 1, 
blitz::ReduceSum<double, double> > >, T_update = blitz::_bz_update<double, 
double>, P_numtype = double, int N_rank = 2]: Assertion `0' failed.
Abort

Platform info: gcc 3.2.2, from a stock RedHat 9 install (x86).  Blitz from 
today's CVS.

I don't understand why it says that I'm assigning a rank 1 expression to a 
rank 2 array.  It seems pretty clear to me that in the 2d case, there's 2 free 
indices (i0,i1), so the resulting expression should be rank 2. No?

I'd very much appreciate any advice on this, as I'm rather new to C++ templates.

Regards,

Fernando.

[Blitz-support] New user, some reports

[Fernando Perez <fperez@co...>]

Hi all,

I've been using blitz in the past, from a version included by Eric Jones in 
Sicpy's weave package (python support).  He used a snapshot from 2000/12/13 
for that.  I decided to update to current CVS because I'm having some strange 
problems with that, and would rather not report a bug which has been fixed for 
2 years :)

Here come a few initial comments:

1. First, it might be a good idea to update the webpage 
http://www.oonumerics.org/blitz/download/ at the end, where it describes using 
cvs.  Instead of

*********
NOTE: The CVS repository is mainly intended for developers. It does not 
contain the configure script referred to in the Installation section in the 
User's Guide. You will need additional tools to generate it: be sure to have 
autoconf and automake installed and issue
aclocal; autoheader; automake -a; autoconf

in your favourite shell.
*********

It should say:

*********
NOTE:...

be sure to have at least version 2.57 of autoconf and 1.7.6 of automake 
installed and issue

autoreconf -vif

in your favourite shell.  You can then proceed with the usual 
./configure;make;make install steps.
*********


This would save those willing to use CVS a good deal of frustration and 
mailing list hunting (though truth be told, all the info _was_ in the 
archives, which is good).

2.  The problem reported in 
http://www.oonumerics.org/MailArchives/blitz-support/msg00768.php

is still present.  It can, as was pointed in followups, be handled by adding 
the --disable-shared flag to ./configure.  But it would be worth at least 
adding a note about it to the CVS instructions, if it is not reasonable to 
expect a fix soon.


3.  After a successful build with the above caveats, I tried running the test 
suite, and it fails with:

planck[blitz_cvs]> make check-testsuite

[snip]

Running test suite...
Adnene-Ben-Abdallah-1
Adnene-Ben-Abdallah-2
arrayresize
chris-jeffery-1
chris-jeffery-2
chris-jeffery-3
complex-test
constarray
contiguous
copy
ctors
derrick-bass-1
derrick-bass-3
exprctor
make[1]: *** [check-testsuite] Error 1
make[1]: Leaving directory `/home/fperez/usr/local/blitz_cvs/testsuite'
make: *** [check-testsuite] Error 1

This is running on a stock RedHat 9.0 box, Pentium 4, 1 GB of RAM:

planck[blitz_cvs]> gcc --version
gcc (GCC) 3.2.2 20030222 (Red Hat Linux 3.2.2-5)

planck[blitz_cvs]> uname -a
Linux planck 2.4.20-8 #1 Thu Mar 13 17:54:28 EST 2003 i686 i686 i386 GNU/Linux

With a make -n I got what it wanted to do and ran the same commands as a 
separate script.  The problem is:

...
exprctor
./test: line 6:  7823 Segmentation fault      ( ./$prog >/dev/null 2>&1 )

exprctor is indeed segfaulting.  If I remove it from the list of tests to run, 
the suite passes.  Someone who knows the code might want to look into why 
exprctor is crashing.

I'll be glad to provide more details if it's useful.

Now I'm off to test the problems I was seeing, to find out if they are still 
there or not :)

Best regards,

Fernando.

Re: [Blitz-support] TinyVector unary minus?

[William Gallafent <william@ga...>]

On Tue, 26 Aug 2003, Patrik wrote:

> i even grepped through all the sources and I can't see a definition for a
> unary minus operator anywhere... this is weird. And unless i'm completely
> spacing now, I am using the current cvs code. so, this operator should
> exist?

Seems you're right. On closer inspection I too failed to find it!

I suppose one is expected simply to subtract from 0 to change the sign of each
element of a TinyVector. This compiles for me:

#include <iostream>
#include <blitz/tinyvec-et.h>

int
main (int argc, char* argv[])
{
	blitz::TinyVector<float, 3>  myTV = 1;
	blitz::TinyVector<float, 3>  anotherTV = 0 - myTV;
	std::cout << anotherTV << std::endl;
}

and produces the output

3 [         -1        -1        -1 ]

Hope I haven't misunderstood, and you're actually trying to do something
cleverer or more subtle than that!

-- 
Bill Gallafent.

Re: [Blitz-support] TinyVector unary minus?

[Todd Veldhuizen <tveldhui@os...>]

It does look like this is missing, I see this comment at the bottom of 
blitz/vecexpr.h:

// NEEDS_WORK: implement operator- for Range, VectorPick, TinyVector

Sorry about that, looks like I never got around to implementing those.
As a kludge, "0-x" should work as well as "-x".

Todd

On Tue, 26 Aug 2003, Patrik wrote:
> At 12:54 PM 8/26/2003 +0100, you wrote:
> >On Thu, 21 Aug 2003, Patrik wrote:
> >
> > > I don't know if I've never tried this before, but it seems there is no 
> > unary
> > > minus for TinyVector.  Is there a reason for this?  Or am I missing
> > > something?
> >
> >I think you just need to:
> >
> >#include <blitz/tinyvec-et.h>
> >
> >to get these and all the other tinyvector operators. I think this isn't
> >included automatically from tinyvec.h since it slows down compilation quite a
> >bit (vecbops.cc, for example, is 17000 lines of templated operators!)
> 
> well, trouble is that I am...
> 
> i even grepped through all the sources and I can't see a definition for a 
> unary minus operator anywhere... this is weird. And unless i'm completely 
> spacing now, I am using the current cvs code. so, this operator should exist?
> 
> /patrik
> 
> 
> _______________________________________________
> Blitz-support mailing list
> Blitz-support@...
> http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support
> 


-- 
Todd Veldhuizen  /  tveldhui@...  /  Indiana University Computer Science

Re: [Blitz-support] TinyVector unary minus?

[Patrik <patrik@uc...>]

At 12:54 PM 8/26/2003 +0100, you wrote:
>On Thu, 21 Aug 2003, Patrik wrote:
>
> > I don't know if I've never tried this before, but it seems there is no 
> unary
> > minus for TinyVector.  Is there a reason for this?  Or am I missing
> > something?
>
>I think you just need to:
>
>#include <blitz/tinyvec-et.h>
>
>to get these and all the other tinyvector operators. I think this isn't
>included automatically from tinyvec.h since it slows down compilation quite a
>bit (vecbops.cc, for example, is 17000 lines of templated operators!)

well, trouble is that I am...

i even grepped through all the sources and I can't see a definition for a 
unary minus operator anywhere... this is weird. And unless i'm completely 
spacing now, I am using the current cvs code. so, this operator should exist?

/patrik

Re: [Blitz-support] TinyVector unary minus?

[William Gallafent <william.gallafent@vi...>]

On Thu, 21 Aug 2003, Patrik wrote:

> I don't know if I've never tried this before, but it seems there is no unary
> minus for TinyVector.  Is there a reason for this?  Or am I missing
> something?

I think you just need to:

#include <blitz/tinyvec-et.h>

to get these and all the other tinyvector operators. I think this isn't
included automatically from tinyvec.h since it slows down compilation quite a
bit (vecbops.cc, for example, is 17000 lines of templated operators!)

-- 
Bill Gallafent.

[Blitz-support] TinyVector unary minus?

[Patrik <patrik@uc...>]

Hi everyone,

I don't know if I've never tried this before, but it seems there is no 
unary minus for TinyVector.  Is there a reason for this?  Or am I missing 
something?

Thanks,

/Patrik

[Blitz-support] Problem with TinyVector and conj()

[Matthias Schillinger <matthias@tk...>]

Hi,

if I try to compile this small program with gcc-3.2
/******************************************/
#include <iostream>
#include <complex>
#include <blitz/array.h>
using namespace blitz;
typedef TinyVector< std::complex<double> ,2> cVector2D;

int main()
{
  cVector2D a( std::complex<double>(1.0,1.0), std::complex<double>(2.0,2.0));
  cVector2D b;
  b=conj(a);
  return 0;
};
/*****************************************/

I get the following error message:

tinytest.cpp: In function `int main()':
tinytest.cpp:11: cannot convert `cVector2D' to `__complex__ double' for
   argument `1' to `__complex__ double std::conj(__complex__ double)'

For Arrays conj() works fine and following the blitz++ user guide it should 
also work for TinyVector. 

Thanks for any suggestions!

Matthias

Re: [Blitz-support] Help with compiler warning

[Rob Lipshutz <rob_lipshutz@mi...>]

Julian
Thanks.  I misspoke.  I am using VS.NET.  I will see if I can suppress them.
Rob
----- Original Message ----- 
From: "Julian Cummings" <cummings@...>
To: "Support list for Blitz++" <blitz-support@...>
Sent: Wednesday, August 06, 2003 11:25 AM
Subject: Re: [Blitz-support] Help with compiler warning


> Hi Rob,
>
> I don't know if there is a compiler-specific flag that will disable
> those warnings
> for you.  I remember seeing warnings like those back when I was still
> trying
> to use VSC++ 7.  The newer MS compiler with VS .NET does not issue these
> bogus warnings.  But they are definitely harmless.
>
> Regards, Julian C.
>
>
> Rob Lipshutz wrote:
>
> > I am getting the following error message when compiling in VS 7.0 -
> > while it is warning and doesn't seem to affect what I am doing - it
> > would be nice to avoid it.
> >
> > Any help would be appreciated
> >
>
> --------------------------------------------------------------------------
-------------------------------------
> >
> > D:\Documents and Settings\Rob\My Documents\My
> > Project\blitz-0.6\blitz\vector.h(501) : warning C4521:
> > 'blitz::Vector<P_numtype>' : multiple copy constructors specified with
> >
> > [
> >
> > P_numtype=int
> >
> > ]
> >
> > D:\Documents and Settings\Rob\My Documents\My
> > Project\blitz-0.6\blitz\vecpick.h(314) : see reference to class
> > template instantiation 'blitz::Vector<P_numtype>' being compiled with
> >
> > [
> >
> > P_numtype=int
> >
> > ]
> >
> > D:\Documents and Settings\Rob\My Documents\My
> > Project\blitz-0.6\blitz\vecpick.h(315) : see reference to class
> > template instantiation 'blitz::VectorPick<P_numtype>' being compiled
> >
>
> --------------------------------------------------------------------------
--------------------------------------------------
> >
> >------------------------------------------------------------------------
> >
> >_______________________________________________
> >Blitz-support mailing list
> >Blitz-support@...
> >http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support
> >
> >
>
> -- 
> Dr. Julian C. Cummings                       E-mail:
cummings@...
> California Institute of Technology           Phone:  626-395-2543
> 1200 E. California Blvd., Mail Code 158-79   Fax:    626-584-5917
> Pasadena, CA 91125
>
>
>
> _______________________________________________
> Blitz-support mailing list
> Blitz-support@...
> http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support
>

Re: [Blitz-support] Help with compiler warning

[Julian Cummings <cummings@ca...>]

Hi Rob,

I don't know if there is a compiler-specific flag that will disable 
those warnings
for you.  I remember seeing warnings like those back when I was still 
trying
to use VSC++ 7.  The newer MS compiler with VS .NET does not issue these
bogus warnings.  But they are definitely harmless.

Regards, Julian C.


Rob Lipshutz wrote:

> I am getting the following error message when compiling in VS 7.0 - 
> while it is warning and doesn't seem to affect what I am doing - it 
> would be nice to avoid it.
>
> Any help would be appreciated
>
> ---------------------------------------------------------------------------------------------------------------
>
> D:\Documents and Settings\Rob\My Documents\My 
> Project\blitz-0.6\blitz\vector.h(501) : warning C4521: 
> 'blitz::Vector<P_numtype>' : multiple copy constructors specified with
>
> [
>
> P_numtype=int
>
> ]
>
> D:\Documents and Settings\Rob\My Documents\My 
> Project\blitz-0.6\blitz\vecpick.h(314) : see reference to class 
> template instantiation 'blitz::Vector<P_numtype>' being compiled with
>
> [
>
> P_numtype=int
>
> ]
>
> D:\Documents and Settings\Rob\My Documents\My 
> Project\blitz-0.6\blitz\vecpick.h(315) : see reference to class 
> template instantiation 'blitz::VectorPick<P_numtype>' being compiled
>
> ----------------------------------------------------------------------------------------------------------------------------
>
>------------------------------------------------------------------------
>
>_______________________________________________
>Blitz-support mailing list
>Blitz-support@...
>http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support
>  
>

-- 
Dr. Julian C. Cummings                       E-mail: cummings@...
California Institute of Technology           Phone:  626-395-2543
1200 E. California Blvd., Mail Code 158-79   Fax:    626-584-5917
Pasadena, CA 91125

Re: [Blitz-support] interpolation / sorting multidimensional arrays

[Jonathan Stickel <jjstickel@my...>]

Thanks for suggestions!  I was able to implement a sorting function by 
using a vector of pairs.  It turns out that boolean operators on pairs 
operate on the first element, so I didn't need to define a class to do 
the sorting.  I have attached the code below.  It can sort the rows of 
any m x n Blitz array by the specified column.  It would be trivial to 
modify the function to sort columns by specified row number.

I wonder if there might be interest in making this a member function of 
the Array class?

Regards,
Jonathan


/*Sort the rows of a 2D array by a single column (colNumb). The column
   to be sorted is copied to vector of pairs where the first pair
   element is the value and the second is the row index number.  This
   pair vector is sorted by the values, and the the ordered indexes are
   used to reorder the table accordingly.

   Could also write functions to sort the columns (and even pages of 3D
   arrays).
*/

template <class Type>
void sortRows(Array<Type,2> & table, const int colNumb)
{
   int i, j, m=table.rows(), n=table.columns();

   vector< pair<Type,int> > valIndPair(m);
   for (i=0; i<m; ++i) {
     valIndPair[i].first = table(i,colNumb);
     valIndPair[i].second=i;
   }

   sort(valIndPair.begin(), valIndPair.end());

   Array<Type,1> colReord(m);

   for (j=0; j<n; ++j) {
     if (j==colNumb)
       for (i=0; i<m; ++i)
	table(i,j)=valIndPair[i].first;
     else {
       colReord=table(rAll,j);
       for (i=0; i<m; ++i)
	table(i,j) = colReord(valIndPair[i].second);
     }
   }
}

RE: [Blitz-support] Too many warnings...

[Julian C. Cummings <cummings@ca...>]

Hello Heberto,

A lot of these warnings about unused parameters and so forth have been
cleaned up in the current blitz code base that is available by anonymous
cvs.  Please try checking out blitz from the cvs repository and installing
it, and then send another message if you continue to see lots of compiler
warnings.  Clearly, it has been too long since we last created a tarball of
the blitz distribution.  I will try to address that soon.  By the way, you
would need reasonably current versions of the GNU autoconf, automake and
libtool packages in order to configure the repository version of blitz.

Regards, Julian C.


Dr. Julian C. Cummings
Staff Scientist, CACR/Caltech
(626) 395-2543
cummings@...
 

> -----Original Message-----
> From: blitz-support-bounces@... 
> [mailto:blitz-support-bounces@...] On Behalf Of 
> Heberto del Rio
> Sent: Monday, August 04, 2003 3:22 PM
> To: blitz-support@...
> Subject: [Blitz-support] Too many warnings...
> 
> 
> Hi, I have just downloaded and installed blitz-0.6 to my 
> computer, it is 
> running. Although I have a question:
> 
> Every time that I try to input values into the arrays, my 
> compiler produces 
> plenty of WARNINGS, I have tried everything but I cannot get 
> rid of them.
> 
> blits is intalled into the following directories:
> 
> /usr/include/blitz
> /usr/include/random
> and
> /usr/lib
> 
> I have used the following to compile: -lblitz
> 
> the program is the following:
> 
> 
> #ifdef HAVE_CONFIG_H
> #include <config.h>
> #endif
> 
> #include <iostream>
> #include <stdlib.h>
> 
> #include <blitz/array.h>
> 
> using namespace blitz;
> 
> int main(void)
> {
>   cout << "Hello, World!" << endl;
> 
> 
>   int n = 10;
>   Array<double, 1> A(n);
>   A=0;
>   cout<< A <<endl;
>   return 1;
> }
> 
> As you can see this is a really very easy program. I have 
> included the 
> output of the compilation
> 
> I hope someone can help me to figure out how to fix it.
> 
> Heberto
> 
> 
> +-------------------------------------------------------------
> -------------+
> |                                                             
>              |
> | I am in the process of changing ISP, therefore my email 
> address will     |
> |                                                             
>              |
> | change too. Please use the following email address in the 
> future:        |
> |                                                             
>              |
> |       hebertodelrio@...                             
>              |
> |                                                             
>              
> | |
> +-------------------------------------------------------------
> -------------+
> 

[Blitz-support] Too many warnings...

[Heberto del Rio <hebertodelrio@fa...>]

Hi, I have just downloaded and installed blitz-0.6 to my computer, it is 
running. Although I have a question:

Every time that I try to input values into the arrays, my compiler produces 
plenty of WARNINGS, I have tried everything but I cannot get rid of them.

blits is intalled into the following directories:

/usr/include/blitz
/usr/include/random
and
/usr/lib

I have used the following to compile: -lblitz

the program is the following:


#ifdef HAVE_CONFIG_H
#include <config.h>
#endif

#include <iostream>
#include <stdlib.h>

#include <blitz/array.h>

using namespace blitz;

int main(void)
{
  cout << "Hello, World!" << endl;


  int n = 10;
  Array<double, 1> A(n);
  A=0;
  cout<< A <<endl;
  return 1;
}

As you can see this is a really very easy program. I have included the 
output of the compilation

I hope someone can help me to figure out how to fix it.

Heberto


+--------------------------------------------------------------------------+
|                                                                          |
| I am in the process of changing ISP, therefore my email address will     |
|                                                                          |
| change too. Please use the following email address in the future:        |
|                                                                          |
|       hebertodelrio@...                                          |
|                                                                          |
+--------------------------------------------------------------------------+

Re: [Blitz-support] libtool problem (was: Problem setting up cvs checkout with autoreconf)

[Patrik <patrik@uc...>]

(I apologize for the previous message that Eudora decided to send before I 
had even started typing it... ;-)

I managed to get one step further in building my cvs snapshot, by making 
the following change to the configure.in:

diff -r1.29 configure.in
248a249
 >         LDFLAGS="-XCClinker -q64"

The output now looks like this:

/bin/sh ../libtool --mode=link KCC  -q64 --restrict --longlong 
--strict_warnings  -XCClinker -q64 -o libblitz.la -rpath 
/u7/patrik/lib64  globals.lo  -lm
rm -fr  .libs/libblitz.exp .libs/libblitz.so.0
generating symbol list for `libblitz.la'
/usr/common/usg/gnu/bin/nm -B -Bpg  .libs/globals.o  | awk '{ if ((($2 == 
"T") || ($2 == "D") || ($2 == "B")) && (substr($3,1,1) != ".")) { print $3 
} }' | sort -u > .libs/libblitz.exp
/usr/common/usg/gnu/bin/nm: .libs/globals.o: File format not recognized
KCC -Wl,-bM:SRE -o .libs/libblitz.so.0  .libs/globals.o  -lm -lc  -q64 
-Wl,-bE:.libs/libblitz.exp -Wl,-bnoentry ${wl}-berok
$(CXX) -q64 -o .libs/libblitz.a .libs/libblitz.so.0
../libtool[4137]: CXX:  not found
../libtool[4137]: -q64:  not found
make[1]: *** [libblitz.la] Error 127

I don't know what's going on now.  CXX is certainly set to KCC in the 
Makefile. It looks like libtool is actually trying to invoke "$(cxx)" 
literally?  Moreover, if I try to help libtool out with this step by doing 
what I assume it wants to, the following happens:

[37] patrik@...% KCC  -q64 -o .libs/libblitz.a .libs/libblitz.so.0
KCC: Option .libs/libblitz.so.0 not recognized.

Apparently, the ".0" suffix on the file confuses KCC.

Anyone have any idea what the @<#@& problem is??  I'm getting pretty 
frustrated with this "portable" way of building the library, when a simple 
"KCC -o libblitz.a globals.o" would do it and I would have had time to type 
that many thousands of times in the time I've spent trying to get this to 
work... :-(

Regards,

/Patrik

At 10:43 PM 8/1/2003 -0700, you wrote:
>Thanks to Julian and Frank who helped me with the autoreconf problem!  Now 
>I have another one.  I'm doing
>
>./configure --with-cxx=KCC --enable-64bit --libdir=/u7/patrik/lib64 
>--includedir=/u7/patrik/include
>
>and the making fails with:
>
>/usr/common/usg/gnu/bin/nm: .libs/globals.o: File format not recognized
>KCC -bM:SRE -o .libs/libblitz.so.0  .libs/globals.o  -lm 
>-lc  -bE:.libs/libblitz.exp -bnoentry ${wl}-berok
>dump: .libs/globals.o: 0654-108 file is not valid in the current object 
>file mode.
>         Use the -X option to specify the desired object mode.
>ld: 0711-736 ERROR: Input file .libs/globals.o:
>         XCOFF64 object files are not allowed in 32-bit mode.
>
>This is what happens when the "-q64" flag is not specified for the linking 
>invocation of KCC.  Can anyone tell me what the problem is this time?  (I 
>believe what I did before was just create libblitz.a from globals.o by 
>hand, but it's sort of defies the entire idea of this super fancy build 
>system... ;-)
>
>Thanks,
>
>/Patrik
>
>At 01:47 PM 8/1/2003 +0200, you wrote:
>>It seems to me that aclocal does not find libtool.
>>Do you have libtool installed?
>>
>>If you do, aclocal does not find the file `libtool.m4' which defines
>>`AC_PROG_LIBTOOL' in its m4 directory. Try `aclocal --print-ac-dir'
>>and symlink or copy `libtool.m4' there or use the -I option of aclocal.
>>
>>-Frank
>>_______________________________________________
>>Blitz-support mailing list
>>Blitz-support@...
>>http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support
>
>
>_______________________________________________
>Blitz-support mailing list
>Blitz-support@...
>http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support

Re: [Blitz-support] KCC/libtool problem (was: Problem setting up cvs checkout with autoreconf

[Patrik <patrik@uc...>]

At 10:43 PM 8/1/2003 -0700, you wrote:
>Thanks to Julian and Frank who helped me with the autoreconf problem!  Now 
>I have another one.  I'm doing
>
>./configure --with-cxx=KCC --enable-64bit --libdir=/u7/patrik/lib64 
>--includedir=/u7/patrik/include
>
>and the making fails with:
>
>/usr/common/usg/gnu/bin/nm: .libs/globals.o: File format not recognized
>KCC -bM:SRE -o .libs/libblitz.so.0  .libs/globals.o  -lm 
>-lc  -bE:.libs/libblitz.exp -bnoentry ${wl}-berok
>dump: .libs/globals.o: 0654-108 file is not valid in the current object 
>file mode.
>         Use the -X option to specify the desired object mode.
>ld: 0711-736 ERROR: Input file .libs/globals.o:
>         XCOFF64 object files are not allowed in 32-bit mode.
>
>This is what happens when the "-q64" flag is not specified for the linking 
>invocation of KCC.  Can anyone tell me what the problem is this time?  (I 
>believe what I did before was just create libblitz.a from globals.o by 
>hand, but it's sort of defies the entire idea of this super fancy build 
>system... ;-)
>
>Thanks,
>
>/Patrik
>
>At 01:47 PM 8/1/2003 +0200, you wrote:
>>It seems to me that aclocal does not find libtool.
>>Do you have libtool installed?
>>
>>If you do, aclocal does not find the file `libtool.m4' which defines
>>`AC_PROG_LIBTOOL' in its m4 directory. Try `aclocal --print-ac-dir'
>>and symlink or copy `libtool.m4' there or use the -I option of aclocal.
>>
>>-Frank
>>_______________________________________________
>>Blitz-support mailing list
>>Blitz-support@...
>>http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support
>
>
>_______________________________________________
>Blitz-support mailing list
>Blitz-support@...
>http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support

[Blitz-support] Help with compiler warning

[Rob Lipshutz <rob_lipshutz@mi...>]

I am getting the following error message when compiling in VS 7.0 - =
while it is warning and doesn't seem to affect what I am doing - it =
would be nice to avoid it.

Any help would be appreciated

-------------------------------------------------------------------------=
--------------------------------------

D:\Documents and Settings\Rob\My Documents\My =
Project\blitz-0.6\blitz\vector.h(501) : warning C4521: =
'blitz::Vector<P_numtype>' : multiple copy constructors specified with

[

P_numtype=3Dint

]

D:\Documents and Settings\Rob\My Documents\My =
Project\blitz-0.6\blitz\vecpick.h(314) : see reference to class template =
instantiation 'blitz::Vector<P_numtype>' being compiled with

[

P_numtype=3Dint

]

D:\Documents and Settings\Rob\My Documents\My =
Project\blitz-0.6\blitz\vecpick.h(315) : see reference to class template =
instantiation 'blitz::VectorPick<P_numtype>' being compiled

-------------------------------------------------------------------------=
---------------------------------------------------

Re: [Blitz-support] Problem setting up cvs checkout with autoreconf

[Patrik <patrik@uc...>]

Thanks to Julian and Frank who helped me with the autoreconf problem!  Now 
I have another one.  I'm doing

./configure --with-cxx=KCC --enable-64bit --libdir=/u7/patrik/lib64 
--includedir=/u7/patrik/include

and the making fails with:

/usr/common/usg/gnu/bin/nm: .libs/globals.o: File format not recognized
KCC -bM:SRE -o .libs/libblitz.so.0  .libs/globals.o  -lm 
-lc  -bE:.libs/libblitz.exp -bnoentry ${wl}-berok
dump: .libs/globals.o: 0654-108 file is not valid in the current object 
file mode.
         Use the -X option to specify the desired object mode.
ld: 0711-736 ERROR: Input file .libs/globals.o:
         XCOFF64 object files are not allowed in 32-bit mode.

This is what happens when the "-q64" flag is not specified for the linking 
invocation of KCC.  Can anyone tell me what the problem is this time?  (I 
believe what I did before was just create libblitz.a from globals.o by 
hand, but it's sort of defies the entire idea of this super fancy build 
system... ;-)

Thanks,

/Patrik

At 01:47 PM 8/1/2003 +0200, you wrote:
>It seems to me that aclocal does not find libtool.
>Do you have libtool installed?
>
>If you do, aclocal does not find the file `libtool.m4' which defines
>`AC_PROG_LIBTOOL' in its m4 directory. Try `aclocal --print-ac-dir'
>and symlink or copy `libtool.m4' there or use the -I option of aclocal.
>
>-Frank
>_______________________________________________
>Blitz-support mailing list
>Blitz-support@...
>http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support

Re: [Blitz-support] Problem setting up cvs checkout with autoreconf

[Frank Schimmel <schimmel@dk...>]

It seems to me that aclocal does not find libtool.
Do you have libtool installed?

If you do, aclocal does not find the file `libtool.m4' which defines
`AC_PROG_LIBTOOL' in its m4 directory. Try `aclocal --print-ac-dir'
and symlink or copy `libtool.m4' there or use the -I option of aclocal.

-Frank

Re: [Blitz-support] interpolation / sorting multidimensional arrays

[Patrik <patrik@uc...>]

hi Jonathan,

I think the way I would do it would be to define the data points as a 
"class point : private std::pair < double, double >", and define operator< 
for this class to operate on the x-value.  That way, you can sort a vector 
of these objects the same way as below, like

Sort (points .begin (), points.end ()), and your y values will just tag 
along for the ride.

There might be efficiency impacts using this type to do the interpolation, 
so it's probably best to convert this to 2 blitz arrays before starting to 
interpolate (if performance is critical).  Maybe someone can come up with 
an even fancier solution?  ;-)

Good luck,

/Patrik

At 04:59 PM 7/31/2003 -0700, you wrote:
>Hi
>
>I am relatively new to Blitz++ (and c++ programming in general), but I am 
>very impressed.  I have recently translated a good deal of my own Matlab 
>code to c++ by using Blitz arrays.
>
>I would like to interpolate within a set of data points given by 
>one-dimensional arrays (e.g. x and y for 1D interpolation) to find a value 
>at an arbitrary point within the range of the data.  Not being able to 
>find an existing function to do this, I planned to write a simple linear 
>interpolation function.  However, I am stuck at sorting the data 
>points.  Essentially, I need to sort x from least to greatest and then 
>reorder y so that there is 1-1 correspondence with the original data points.
>
>I can sort x by:
>
>sort( &x(0), &x(x.size())+1);
>
>but I am at a loss at how to reorder y accordingly.  Other than writing a 
>sorting function from scratch, the methods I can think of would be 
>extremely inefficient (O(N^2)).  Does anyone have a suggestion?
>
>Regards,
>Jonathan
>
>_______________________________________________
>Blitz-support mailing list
>Blitz-support@...
>http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support

RE: [Blitz-support] interpolation / sorting multidimensional arrays

[Bob Uva <bobu21@co...>]

You can create a Point object that contains both the x and y coordinates,
and then have a comparison function in the form of an overloaded less than
(<) operator. The sort function will use that 'less-than' operator function
to do the comparison. You said that you're new to c++ as well. If you don't
know
how to write an overloaded operator member function, you can find good
tutorials
in many elementary c++ texts. I strongly suggest going through an elementary
text
before attempting any serious c++ programming.

-Bob

-----Original Message-----
From: blitz-support-bounces@...
[mailto:blitz-support-bounces@...]On Behalf Of Jonathan
Stickel
Sent: Thursday, July 31, 2003 5:00 PM
To: blitz-support@...
Subject: [Blitz-support] interpolation / sorting multidimensional arrays


Hi

I am relatively new to Blitz++ (and c++ programming in general), but I
am very impressed.  I have recently translated a good deal of my own
Matlab code to c++ by using Blitz arrays.

I would like to interpolate within a set of data points given by
one-dimensional arrays (e.g. x and y for 1D interpolation) to find a
value at an arbitrary point within the range of the data.  Not being
able to find an existing function to do this, I planned to write a
simple linear interpolation function.  However, I am stuck at sorting
the data points.  Essentially, I need to sort x from least to greatest
and then reorder y so that there is 1-1 correspondence with the original
data points.

I can sort x by:

sort( &x(0), &x(x.size())+1);

but I am at a loss at how to reorder y accordingly.  Other than writing
a sorting function from scratch, the methods I can think of would be
extremely inefficient (O(N^2)).  Does anyone have a suggestion?

Regards,
Jonathan

_______________________________________________
Blitz-support mailing list
Blitz-support@...
http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support

[Blitz-support] interpolation / sorting multidimensional arrays

[Jonathan Stickel <jjstickel@my...>]

Hi

I am relatively new to Blitz++ (and c++ programming in general), but I 
am very impressed.  I have recently translated a good deal of my own 
Matlab code to c++ by using Blitz arrays.

I would like to interpolate within a set of data points given by 
one-dimensional arrays (e.g. x and y for 1D interpolation) to find a 
value at an arbitrary point within the range of the data.  Not being 
able to find an existing function to do this, I planned to write a 
simple linear interpolation function.  However, I am stuck at sorting 
the data points.  Essentially, I need to sort x from least to greatest 
and then reorder y so that there is 1-1 correspondence with the original 
data points.

I can sort x by:

sort( &x(0), &x(x.size())+1);

but I am at a loss at how to reorder y accordingly.  Other than writing 
a sorting function from scratch, the methods I can think of would be 
extremely inefficient (O(N^2)).  Does anyone have a suggestion?

Regards,
Jonathan