BLAS Routines in C/C++
BLAS Routines in C/C++
Hi, Will.
Some time ago I think you mentioned adding C translations of some BLAS, LAPACK, etc. routines to CodeCogs.
Has any progress been made toward this goal?
I am planning to start translating and posting some these translations; however, they are not totally true translations. How complete do you want translations to be? For example, DNRM2 computes the Euclidean norm of an input real vector. One of the parameters passed to DNRM2 is INCX, which tells the routine how much to increment the vector index during the iteration (i.e. - include every second, third, etc. entry in the computations). I have never used DNRM2 with INCX anything but 1, so my translation is written as if INCX is 1 and INCX is not passed into the routine--one passed parameter eliminated.
Would such code be appropriate on CodeCogs? Or would you prefer one version of a routine--a true transliteration?
Regards,
David
7 May 08, 9:47PM
Hi,
I've not made any progress on this I'm afraid, so your contributions on this front would be very useful.
In terms of how you translate the functions, I don't think you need to go down the 100% pure route, so missing out INCX as in your example is fine. My only thought is that we should try to ensure that someone can easily convert from using the CodeCogs BLAS routine to the real library (i.e. when they want max performance etc).
However, so long as you document your work and detail any amendments to the interface / how it differs from the real library then I think it'll be a great and very valuable contribution. You could perhaps also show example using your function and the real BLAS,LAPACK functions.
Let me know if you need any help with this effort. Fairly busy ring now trying to get CodeCogs adopted by some universities and improving the embedded calculators. However this is important work, so I'll support you as much as I can.
Will
Login