UoB Optimisation (Work in progress)

[Lancasterg]

Hi all,

I have been working on my MSc thesis in computer science over the summer, entitled 'Porting and optimisting the Isca climate model on Intel and Arm processors'. This project involved porting, bechmarking and optimising Isca on a few different HPC systems including BlueCrystal phase 3 at Bristol, Blue Pebble at Bristol, and the Isambard supercomputer as part of the GW4 alliance. I am currently proof reading my thesis, however I can make it public to you if it is something you would be interested in reading.

One of the optimisations I have made is to use the FFTW library rather than the bespoke FFT supplied with the FMS that was written in 1981. This change has improved the performance of the code!

I have mostly been working from the Grey Mars branch, which doesn't appear to have been merged with the master branch yet. However, I can merge only my changes if required, as they do not depend on any of the code in the Grey Mars branch. Please let me know how best to merge my changes.

Many thanks,

George

[gkvallis]

Hi George,

That’s great! And using FFTW instead of the old FFTs is something I’ve long wished for. Does this, or could it in future, allow parallelisation in the zonal direction also? In any case, it is certainly an improvement. We'd love to read your thesis whenever it is ready — congratulations on that and on your work.

As regards how best to merge your changes, we’ll need to discuss this and get back to you. We also may need to discuss this with you offline, so if you wouldn’t mind sending an email address that would help. (I see one at brighton.ac.uk , but that may be out of date.?)

Geoff

[Lancasterg]

Hi Geoff,

Thanks! I have been using the serial version of FFTW, however it would be trivial to plug in the MPI/OpenMP version instead.

Yes, that email address is out of date. I've found your exeter.ac.uk email address, so I'll shoot you an email to discuss this further.

Cheers

[sit23]

Thanks very much for this George. Looks like really excellent work. Certainly very keen to read your thesis once it's available.

As you say, the grey_mars branch has not been merged into the master branch. This is probably something I'll look to do soon anyway, which will make this p/r a bit cleaner. But, as Geoff says, we will discuss it and let you know how we think we should proceed.

[Lancasterg]

Brilliant, thanks Stephen!

I'm still proof reading at the moment, but the document will be finished by the 9th of September at the latest. Let me know when you've decided how best to proceed.