newbie question regarding CBLAS

[stockhausen]

Hello, I am working with BLAS and LAPACK in C (using gcc in cygwin) and I have had no problems until now. Since my data types are all doubles, I used the double percision BLAS functions, such as cblas_dsymv (the d for double percision) instead of cblas_ssymv (the s for single percision).

However, right now I'm dealing with very large numbers and as a result I have to use long doubles instead of just double. Unfortunately, it appears that there is no long double equivalent for BLAS.

Is there anyway around this?

Thanks

[Julien Langou]

You are pretty much stucked as far as optimized BLAS libraries are concerned.
I am not aware of any optimized BLAS libraries supporting long double.
(The reason being that long double is not a clearly defined type, it can depends on the
compiler for example)

The only option I see for you is to write your own unoptimized BLAS.

Julien.

[stockhausen]

Thanks for the promt reply. I run the same sort of computations in MATLAB, and I have no problems in that enviornment. My understanding was that MATLAB itself uses BLAS as well and for this reason MATLAB computations are highly optimized. Am I wrong in this matter?

Thanks again.

[Julien Langou]

Well you should ask Mathworks then. I have no idea what they are doing, sorry about that.
Julien.