single precision matrix multiplication

[GerdW]

Hi,

another approach:

This creates one more memory buffer, but works without case structure


Message Edited by GerdW on 04-08-2008 05:22 PM

[altenbach]

Well, we don't need 2D for the vector. 😄

Message Edited by altenbach on 04-08-2008 08:36 AM

[LukasN]

Hello Altenbach,

thank you very much for the algorithm. I have tried it and obviously it works, but unfortunately it seems not to be useful in real-time. I have verified the performance and it is about 5-times slower than the standard matrix multiplication in double precision (see attached file)  

Best regards,

Lukas

[LukasN]

Thank you GerdW, but it is still much slower than LV built-in matrix multiplication function. I am afraid that there exist the only one solution. Program it in C, compile in to DLL an link it into LabVIEW (for now, I can't do it...)

Lukas

[DFGray]

Most modern processors do double precision arithmetic about as fast as single precision (both are single-cycle or multiple operation per cycle operations).  Your only advantage in using single precision is probably in memory handling.  Modern processors can also do a lot of the matrix math in single operations (SSE2 and above capable processors can do this in double precision, MMX is integer, SSE is single precision).

Take home message - you probably will not get much speed out of this exercise.

One final note. Although I am an NI employee, I do not work with the math/analysis team so have no knowledge of the internal structure of the matrix operation primitives/VIs.  I develop applications in LabVIEW.

[LukasN]

I have to try it, because this multiplication must run faster. I can't see any other solution (except buing a new processor for my PXI)

[DSPGuy]

Lukas,

What is your LabVIEW version and OS?  I may have a solution for you.

-Jim