Re: [Lapackpp-devel] A strange bug with Blas_H_Dot_Prod

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

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

Dear Cheng Guo,

thank you for this detailed question. However, as for your
Blas_H_Dot_Prod problem, I cannot reproduce this here with
lapackpp-2.4.10. The output, as expected, is (3,0) which is the expected
 dot product. This is both with gcc-3.3.5, blas-3.0, lapack-3.0 on SuSE
9.3 and also with gcc-4.0.2, blas-3.0, lapack-3.0 on SuSE 10.0.

Oh, I just see you mentioned x86_64. Stand by - I will check more on
this architecture.

As for the other bug: As you wrote yourself, this doesn't occur on
gcc-3.x, only on the gcc-2.97. So this is obviously an issue due to
different compiler behaviour. I'm actually not really willing to spend a
lot of effort on supporting the old gcc-2.9x compiler, because it is
just so old and there are much newer compilers around for a long time. I
understand you might be forced to use that old version on the True64
computer, but in that case I'm afraid I have to say I cannot really
support this architecture.

I'll look into the x86_64 more closely and report back in some time.

Christian

Cheng Guo schrieb:
> Dear Christian,
> 
> I found the function Blas_H_Dot_Prod doesn't work on my computer. My cpu
> is P4 with EM64t, and my OS is SuSE 10.0 for x86_64.
> The testing program is:
> *****************************************
>  int main()
>  {
>    LaVectorComplex a(3),b(3);
>    a=LaComplex(1,0);
>    b=LaComplex(1,0);
>     cout<<a<<endl;
>   cout<<Blas_H_Dot_Prod(a,b)<<endl;
>    }
> *****************************************
> 
> The output is:
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> (1,0)
> (1,0)
> (1,0)
> 
> (2.31779e-310,4.94066e-323)
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> So the result of Blas_H_Dot_Prod(a,b) is incorrect. The strange thing is
> that when I deleted the line "cout<<a<<endl;"  the result of
> Blas_H_Dot_Prod(a,b) changed to (2,1) which is also wrong.
> 
> 
> By the way, in the email I sent to you on May 29, I also reported a bug.
> I still don't konw how to solve it so I include this bug here:
> 
> My test program is
> ****************************
> #include <cstdlib>
> #include <iostream>
> #include <laindex.h>
> #include <lavd.h>
> #include <gmd.h>
> using namespace std;
> int main()
> {
> LaVectorDouble m,n;
> m.resize(3,1);
> m=2;
> n.resize(2,1);
> n=1;
> m(LaIndex(0,1)).inject(n);
> cout<<m<<endl;
> }
> ***************************
> I expect get the result
> 
> 1
> 1
> 2
> 
> which is what I get on my PC(suse 10,gcc-4.02)
> But I got this
> 
> 2
> 2
> 2
> 
> on the Tru64 UNIX (gcc-2.97)
> I have tested lapackpp2.4.10 and 2.4.8, neither gave the right answer.
> 
> This bug seems not affecting LaGenMatDouble, for the following test code
> works correctly on Tru64
> ****************************
> #include <cstdlib>
> #include <iostream>
> #include <laindex.h>
> #include <lavd.h>
> #include <gmd.h>
> using namespace std;
> int main()
> {
> LaGenMatDouble m,n;
> m.resize(3,3);
> m=2;
> n.resize(2,2);
> n=1;
> m(LaIndex(0,1),LaIndex(0,1)).inject(n);
> cout<<m<<endl;
> }
> ***************************

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

iQCVAwUBRL9Fv2XAi+BfhivFAQLuTwQAqpavSg307G/YqIKMxuGxm5LC1o2EvlLp
aITYrzuG6HK1isuy7V37Tga3BddAP74n7pHomCGrI+szAqARzUNQB/NYj0pT6crv
XiJLtz50loQwe+W0nU6Dkr67m7VzWhBcPJAnqkem7ca5JjxAJAvJkz2KYGAv59Fm
XxEZ+U9LdUw=
=VOQO
-----END PGP SIGNATURE-----