Evan Pretti

I'm finally getting around to taking a look at this... sorry for the delay. It looks like the hard-coding in `DummySystemGenerator` is intentional, or at least not a problem. See...

I recall being confused by implicit solvent cutoffs a long while ago when dealing with Amber. It seems like Amber employs an additional cutoff for computing the effective Born radii...

I see; I don't know if Amber does anything special to handle the discontinuity in the electrostatic energy at all. I don't see anything specific in the documentation. NAMD's documentation...

Closing this issue since it is over a year old with no additional response, and it doesn't appear to involve an outstanding problem. This issue can be reopened if needed.

@peastman The CPU platform and preconditioning flag should be ready now if you want to review.

@peastman I have the GPU implementation working locally for OpenCL single, double, and mixed precision using the Intel CPU implementation, and OpenCL single precision on my GPU. OpenCL and CUDA...

I wanted to mention one of the problems that I found because it might affect another part of OpenMM. I was using a (single thread block only) reduction identical to...

Thanks for doing this! I'll try the faster kernel for the matrix solver, and I think your suggestions for the CG solver should all be possible to implement. Both convergence...

It would be good if this can be tested on HIP since I can only check OpenCL and CUDA.

Current status of this feature: 1. I found a way to accelerate the matrix solver: by processing 32x32 blocks on the diagonal followed by 32xN slices to the right of...