Anton Gorenko

> @peastman : Should we be worried about these test failures on AMD OpenCL? > > ``` > Errors while running CTest > Test project /home/runner/work/openmm/openmm/build > Start 100: TestOpenCLNonbondedForceSingle...

@jchodera BTW, there is an issue with amber20-factorix: https://github.com/openmm/openmm/issues/3391 I've noticed that you encountered it too as your json results don't have results for it. I guess it would be...

According to my measurements there are 589M "potential" interactions (i.e. all interactions we check in inner loops for r < cutoff) and only 160M or them are "positive" interactions (r...

> That seems reasonable to me—was that with a tile size of 32 or 64? 32, these numbers are for CUDA. For CDNA (with hardware warp of 64 and tile...

> > According to my measurements there are 589M "potential" interactions (i.e. all interactions we check in inner loops for r < cutoff) and only 160M or them are "positive"...

> I also found that the single pair list for STMV and a max bits of 4 is 19 768 187 elements long—or about 150MiB. This is too large to...

https://github.com/RadeonOpenCompute/ROCm/issues/1614

@isuruf For the OpenMM-HIP package (https://anaconda.org/streamhpc/openmm-hip) we use this script: [build.sh.txt](https://github.com/conda-forge/conda-forge.github.io/files/12716780/build.sh.txt) It is generated using smithy and mostly based on the openmm package, it uses the regular conda-forge container (`linux_64:...

OpenMM caches compiled binaries so compile time is a problem only for the first time when kernels for a particular simulation are being compiled.

I've checked the latest HCC (`clang_tot_upgrade`) and HIP (`master`) on Ubuntu. Other packages are ROCm 2.0 debs. ``` HCC clang version 9.0.0 (https://github.com/RadeonOpenCompute/hcc-clang-upgrade.git 73b2e4fbb46478ce08d42cc6d9e8bc7b8c7f58da) (https://github.com/RadeonOpenCompute/llvm.git c06b4c8225c8ec11de36cddf5b3d49bdd055c126) (based on HCC 1.3.19045-02e6963-73b2e4f-c06b4c8...