PG-PuReMD: complete memory changes. Remove max register per CUDA function option in build. Other misc. changes.

PG-PuReMD: only perform device-side reductions on energy at requested update energy frequency. Fix trajectory file output. Add num. retries of main simulation loop to log file. Other refactoring.

PG-PuReMD: fix nvcc build issue where incorrect C++ compiler was utilized. Fix race condition with three body list computation. Remove debug code. Other misc. refactoring.

PG-PuReMD: begin enhancements. Enable sim. step retries. Enable far neighbors list max per atom. Other general code clean-ups.

PG-PuReMD: fix build system linker argument ordering (LDFLAGS -> LDADD). Fix inline functions. Disable optimized CUDA code generation by default (added Makefile flags to enable along with preprocessor defs to enable relevant code paths).

PG-PuReMD: fix include guards, resolve implicit function declarations, fix inline issues with GCC v5 (C99 std). Other general code clean-up.

PG-PuReMD: fix issue with hydrogen bond cutoff with systems without hydrogen atoms. Remove debugging info for MPI-only version of PG-PuReMD.

PG-PuReMD: include forgotten headers for distribution. Fix CUDA installation check. Additional updates for build system.

sPuReMD: fix incorrect thread parallelism inside bonded force routines. Update default value of CFLAGS in autotools.

Just removing now unnecessary printfs that clutter the code

The MPI-Not-GPU version is able to run through all benchmarks with decent agreement with the other versions now. The test harness has been modified to allow testing between the three functional versions.
Also added a README file in the test_harness directory about how to use the harness. Finally, I have saved one copy of the output from the previous test run to sim_output_082516.txt, for future comparison.

Add minor checks to PG-PuReMD and PuReMD. Major refactoring of sPuReMD: backport much of PuReMD code and add CUSTOM geo format.

Integrating the test bed I have been working on. Also integrates the bug fix in cuda_valence_angles. Also one little issue in cuda_forces has been resolved.