[SOLVED] FORTRAN, BLAS and matrix multiplication

[nsa_767]

Hi,

I am starting to feel really stupid, but I can't figure out how to use BLAS to multiply a matrix by a vector... I wrote the following Fortran code:

      REAL*8 X(2, 3) /1.D0, 2.D0, 3.D0, 4.D0, 5.D0, 6.D0/
      REAL*8 Y(3) /2.D0, 2.D0, 2.D0/
      REAL*8 Z(2)
      CALL DGEMV('N', 2, 3, 1.D0, X, 2, Y, 1, 0.D0, Z, 1)
      PRINT *, Z
      STOP
      END

DGEMV is a routine from BLAS, which should calculate Z=(0)*Z + (1)*X*Y...

Compiling the above and linking to BLAS,

g77 test.f -o3 -lblas

and then running it, produces:

7.63918485E-313  1.01855798E-312

Which is wrong since the answer should be, 18 and 24???

Any help would be much appreciated!

[WW]

I don't see anything obviously wrong. Your example worked fine for me, on computer running gutsy with g77 and the refblas3-* packages, and on a Mac with blas compiled from source using gfortran.

How did you install blas? From a package, or from source? If from source, did you use g77 to build the library?

[nsa_767]

Hmmm, my memory is somewhat hazy. I know that I didn't compile it... I did install, from the repos, LAPACK and ATLAS, both of which should (I think) install BLAS... My compiler also doesn't complain about the library being missing, so it must be there.

[WW]

Try compiling with gfortran instead of g77.

[nsa_767]

OK, no change when I do that, but when I compile with -llapack instead of -lblas it suddenly spits out the right answer...

Odd, maybe BLAS isn't correctly installed on my system. Anyway, LAPACK works (even with g77), so I am happy (for now).

Thank you for your help... Any further ideas, please post them. (I am not a programmer, just a statistician needing to code something...)

[jgte]

This sort of thing happens when you specify the wrong kind (the 8 in 'real(8)' or 'real*8') of your (input or output) variables. Try another kind. Probably the default kind in BLAS is not 8.