> Hello,
> I am trying to call the optimized (assembler-coded) BLAS routines
> written by Kazushigi Goto from a C Program. He writes they are
> entirely compatible to the Fortran originals, but I didn't manage to
> call sdot_. All what I get out is zero. I tried also to compare with
> f2c output but I don't know Fortran :-(.
> I was trying this:
> tdot.c:
> --------------------
> float s;
> int n=5,inc=1;
> float x[]= { .....};
> float y[]= { .....};
> s = sdot_(&n, x, &inc, y, &inc);
> ---------------------
> gcc -o tdot tdot.c -ldot
> All documentation I found regarding the BLAS Routines was like:
> S - REAL
> D - DOUBLE PRECISION
> C - COMPLEX
> Z - COMPLEX*16
> FUNCTION xDOT (N, X, INCX, Y, INCY) dot <- X^T*Y x=S,D,DS
> .......
> anyone could help me?
> BTW., is it possible to link this to object files compiled with ccc?
> Does someone has suggestions for very fast vectorized signal
> processing routines? Does there exist a standard library?
> Johannes
Johannes,
I have some C code that I wrote to interface to Goto-san's BLAS
it uses SGEMM but maybe it will be helpful.
Compaq has a whole signal processing library already written,
obtain the library. It used used to be easily found through their
website, but since the fortran compiler went commercial, it is
hard to find.
(from their online docs)
CXML includes subroutines covering the areas of:
* Basic Linear Algebra (BLAS) - includes BLAS 1, BLAS 1E, BLAS 1S,
BLAS 2 and BLAS 3.
* Linear and Eigen System Solvers (LAPACK)
* Sparse Linear System Solvers: includes direct and iterative solvers
* Signal Processing: includes FFTs, cos/sin transforms, convolution,
correlations and digital filters
So, you might not have much work to do to implement signal processing
on you alpha box. Then again, these routines aren't exactly portable
to another architecture so YMMV.
I've never seen any performance numbers for CXML. I know Goto-san's
code is extremely fast, maybe all those high priced programmers at
Compaq can come close =-)
Regards,
Tony Hammitt
by