GitLab

* Compile with -munsafe-fp-atomics to enable fast hardware f32 atomic
  add on global memory on pre-MI100 GPUs;
* Use fixed point charge spreading on other GPUs, otherwise float atomic
  add will be compiled as a slow CAS loop;
* Tune block sizes, use executeKernelFlat;
* Tune launch bounds of PME grid-related kernels: force the compiler to
  use all registers by limiting max waves per EU to 1.

Fix SegFault in HipCalcHippoNonbondedForceKernel

    HipSort was created using a temporary ref. Adding `HipContext& cu`
    field to HipCalcHippoNonbondedForceKernel fixes the issue;