Added Missing HOST_DEVICE...HOST_DEVICE_END decorations around function definitions that previously only had HOST_DEVICE_CUDA decorations. (#52)

* use C++11 instead of POSIX clock()

Signed-off-by: Jeff Hammond <jeff.r.hammond@intel.com>

Fix two minor memory leaks.  Neither of these are likely to ever cause 
a problem, but the clang leak sanitizer now reports no errors (at least
not for the default problem).

Add comment that Quicksilver assumes unified memory

* Fix number of particles in 1 rank example problem

* Add CTS2 Benchmark Problem
* Fix warning about incorrect printf formats
* Fix hardcoded loop bound in energy spectrum printing
* Remove unnecessary #include
* Minor cosmetic changes

Last Cycle timers where only being cleared if the print last cycle timers paramter was set.
Fixed this by making sure last cycle timers are cleared at the end of each cycle regardless
of if we want more verbose timer printing.

Modified the initialization of the GPUD to be more generically rank%num_gpus.
This replaced coding that was specifically desiged for power8 cpu systems.

* Increase tolerance for fluence test from 5% to 6%.

This allows these parameters to be set on the command line.

* Added meaningful timers to cycle tracking to distinguich between the kernel and mpi. Fixed last cycle timer logic to be a cumulation of each timer call for each cycle when used.

* Fixing mistakenly commited makefile

Units now correctly reporting segments per second.  

Improved organization of CORAL2 problems and adjusted problem sizes to facilitate scaling studies.

Changed Coral2 correctness tests to use hard coded values based on the
coralBenchmark input parameter, which now uses values 0, 1, or 2. We
know exactly what values Problem1 and Problem2 should provide and
theirfore can check the correctness of the runs based on these values.

* 1) Added functions to compute the energy spectrum of particles that reach census
2) Added function to print hisogram of energy spectrum to a file
3) Made computing/printing energy spectrumn an input deck argument
   Turn on plotting energy spectrum by providing a file name in the input deck
   or command line

* Refactor Energy Spectrum Addition

It now uses std::vector instead of allocated pointer. Additionally it is stored as a tally

* Removed setter functions for energy spectrum
Now set up in the constructor only.

Removed Tallies default constructor that was left and unused.

Made the creation of the decomposition object based on direct computation for mode 0.

Added assert for mode 1 to be less then 1000k processors.

Seperated mode 0 and mode 1 so that mode 0 is as simple
and direct as possible.

Removed memory allocation hold over from qs_vector to std::vector change.
This asssumed worst case size and becomes a large over allocation for large problems.

* Adding Coral2 Second Problem with non flat cross sections and an energy spectrum set to be meaningful

* Moved Coral 2 Problems into their own subdirectories. Added srun lines for Problem 2 in the example script

1) Added printing cross sections as an input parameter. 
2) Also cross sections are printed to a file instead of the screen
   Added filename based on input deck or command line argument.
   Command line argument overides input deck arguments

BugFix: Cross section caches weren't initialized. Caused rare crashes. 
We also now re-initialized the cross section caches before every time step in cycleInit.

Bug: Added missing declare macro to updateTrajectory function

Add README file

Fix typo

Added a test check that number of collisions and number of
facet crossings are equal for the coral 2 problem.

Revisions to CORAL2 benchmark problem

An example with cross sections that aren't flat.

Also some minor clean ups in code related to handling and debugging cross sections

* Fixing bug when a particle undergoes multiple fissions in a cycle and
produces more than estimated number of particles

* Changing collision event to update particle trajectories without un-needed copies.

1) Work on improving the cross sections. This converts the interpretation of the polynomials representing cross sections so that they are in log-log space. We also change the normalization of the cross sections so that you specify the total cross section and cross section ratios at 1MeV instead of averages. The former is much easier to obtain from cross section plots. These changes do not alter the behavior of the default problem since it uses flat cross sections. Flat cross sections are not changed by these modifications.

2) Make more sensible use of the atomFraction and cellNumberDensity quantities when computing macroscopic cross sections. Now both need to be retrieved (atomFraction was set locally). Also, atomFraction is (sensibly) 1/nIsotopes and cellNumberDensity is (somewhat arbitrarily) 1.0