HayterMSAsq erratic results at very small Q and needs test on Mac

[richardh]

HayterMSAsq gives poor reproducibility on different cpu at very small Q.
It likely needs an asymptotic limit in the code. (RichardH will later try to add HayterRMSA as used in FISH, which is in any case better.)
It also needs to be tested on a Mac cpu.

Paul K said: can run a number of different configurations:

intel-cpu-single-dll
intel-cpu-single-ocl
intel-gpu-single-ocl
nvidia-gpu-single-ocl

intel-cpu-double-dll
intel-cpu-double-ocl
intel-gpu-double-ocl Not supported
nvidia-gpu-double-ocl

It fails completely on intel-gpu-single-ocl and it gives invalid results for low q on the other single precision platforms (gpu and cpu). The double precision platforms work fine.

[butler]

This is probably part of the core sasmodels work .. or at least very close. Think we should make this blocker for now — though if it "works" but refinement is still desired we can downgrade this a bit?

[richardh]

I now have a better version, having fixed a bug in my Taylor expansion at small Q, this has not been uploaded to |GIT as I want to change its name also, will need help to sort ths at code camp.
Richard

[butler]

This is as finished as it will be. Main conclusion is it will have to ALWAYS use double precision. May add a FISH version later which may be more stable. Closing this ticket.