sPuReMD: fix memory leak with ICHOLT preconditioner (droptol for OpenMP only) for charge solver. Default to Jacobi preconditioner. Resolve several compiler warnings around C strings (snprintf, etc.). Other code clean-up.

sPuReMD: add functionality to run multiple MD simulations (for usage via integration with external codebases).

PG-PuReMD: in-progress implementation of CUDA-aware MPI for charge solver communications using OpenMPI (default off).

sPuReMD: resolve OpenMP build errors with GCC v10 (behavior changes for function arguments). Suppress possibly unitialized variable warnings for BiCGStab.

PG-PuReMD: fix van der Waals and Coulomb force computations for 1 warp per atom implementation, and enable this implementation by default.

PG-PuReMD: use registers for local arithmetic in CUDA nonbonded calculations and fix shared memory usage in 1 warp per atom implementation.

PG-PuReMD: switch timers to use CUDA events and add additional timers for kernels to match MPI code. Other code cleanup.

This reverts commit 6a1385e9.

PG-PuReMD: align CSR sparse matrix data structure to 128-bit/256-bit segments for memory coalescing (for 1 warp per row SpMV implementation). Change default interaction list and sparse matrix allocations to be multiples of 32 for aforementioned memory coalescing.

PG-PuReMD: switch from using strtok_r to strtok to avoid reliance on GNU extensions with building for the C11/C++11 standard.

PG-PuReMD: only copy host generated grid to device on reneighboring steps (after rebinning atoms and regenerating the grid). Fix small NVE integration bug (reallocation checks before copying new atom info to device).

PG-PuReMD: change CUDA kernel error checking to only make calls to cudaDeviceSynchronize when in debugging mode (non-debug builds thus have better performance with kernel asynchronous launch behavior in single default stream). Change file I/O to only flush when in debugging mode (non-debug builds allow the I/O operations to be grouped for better performance).

PG-PuReMD: add support for symmetric, half stored format (SYM_HALF_MATRIX) of the sparse matrix for the charge model (add initialization routines and fix-up solver).

PG-PuReMD: rework custom reduction functions to use less shared memory and to have the correct thread and block counts. Fix bug in charge matrix initialization (under-allocated space previously). Split charge matrix and bonds/hydrogen bonds memory management routines to mirror the similar splitting of the initialization routines. Code clean-up related to half vs. full list and sparse matrix formats.

PG-PuReMD: improve performance by conditionally re-running initialization kernels within Cuda_Init_Forces based on out-of-memory conditions (for charge matrix, bonds list, hydrogen bonds list).

Tools: add PuReMD custom geometry replication to geo_tool.py. Remove older awk scripts. Fix issue with silica 6000 atom PDB file (CRYST1 lines not 70 characters).

Tools: update run_sim.py to generate more robust Slurm and Torque/PBS job scripts with the MPI+CUDA code (added extra option for specifying additional flags to be used with the command used to invoking the mpi+x code).

PG-PuReMD: fix performance logging code around GPU code. Rearrange header files to allow some preprocessor definitions to be defined via options in configure script. Rework Autoconf and Automake code to allow passing flags directly to nvcc compiler wrapper (for CUDA code). Enable C11 and C++11 standard targets during compilation.

PG-PuReMD: fix compilation error (remove variable). Fix issue with dual CG solver for QEq (local arithmetic in SpMV was incorrect). Revert CG solver convergence criterion to use norm of the preconditioned residual vector.

PG-PuReMD: small fix for initializtion routine performance logging (timers not reset after outputting to log file).

PG-PuReMD: corrections for utilizing a SAI preconditioner in MPI code (errors introduced during previous code merges).

PG-PuReMD: revert CG solver convergance criterion to used preconditioned residual norm. Re-enable dual charge solver for QEq.

PG-PuReMD: rework performance logging code in linear solver routines to avoid excessive MPI communications.

PG-PuReMD: adjust performance logging to compute mean timings across all processors. Other code clean-up.