RE: matrix multiplication and inversions [SEC=UNOFFICIAL]

["STARRITT, Andrew via Tech-talk" <tech-talk@aps.anl.gov>]

Hi all,
We use the aSub record in conjunction with the  eigan3 matrix package, specifically using SVD function
to calculate the storage ring's inverse response matrix from the response matrix. It worked very well.
The array sizes are 98 x 42

Cheers
Andrew


-----Original Message-----
From: Tech-talk <tech-talk-bounces@aps.anl.gov> On Behalf Of Johnson, Andrew N. via Tech-talk
Sent: Thursday, 20 February 2020 2:36 PM
To: Pearson, Matthew R. <pearsonmr@ornl.gov>
Cc: tech-talk <tech-talk@aps.anl.gov>
Subject: Re: matrix multiplication and inversions

Hi Matt,

About 25 years ago (wow, was it really?) John Maclean and I devised the controls for the Gemini primary mirror where the IOC did various matrix calculations before passing the results to the dynamic support system actuators via CANbus. We did the maths in subroutine record sub’s using shared global variables to store the small number of large matrices we had to handle, and only used the sub-records to control when the calculations were executed (and to hold scalar parameters and results). We also used waveform records with “devSymb” device support to read and write the shared global variables, although I don’t remember how we synchronized everything to prevent races. The aSub didn’t exist at the time, although Gemini did soon devise the genSub that was its precursor.

Nowadays I would certainly explore your idea of using aSub’s with a standard matrix library. Copying a 5x5x8=200 byte array across a DB link should be reasonably fast as long as you don’t need to do it too much (our matrices were considerably larger and the CPU much slower back then).

HTH,

- Andrew

-- 
Sent from my iPad

> On Feb 19, 2020, at 4:49 PM, Pearson, Matthew R. via Tech-talk <tech-talk@aps.anl.gov> wrote:
> 
> 
> Hi,
> 
> I’m starting a project involving calculating neutron guide positions where the positions are defined by 5 cam based movers. The position calculations involve 5x5 matrix multiplication and inversions, and some of the matrix elements are trigonometric functions. 
> 
> I’m prototyping the calculations in Python, but does anyone have advice on how to implement the calculations in an EPICS database? My first thought is to use aSub records (or a custom record) linking to GSL libraries. 
> 
> I suspect similar work is routinely done on accelerator controls?
> 
> Cheers,
> Matt
> 
> 
> Data Acquisition and Controls Engineer
> Spallation Neutron Source
> Oak Ridge National Lab
> 
> 
> 
> 
> 
> 
>