A few changes improve the consistency of outputs between running uninterrupted from t=0 vs interrupting and restarting a run.
During any given timestep, dti is calculated -- the simulation now computes dti consistently before the first loop, whether for init=Read_Grid or for a normal run.
For Gravity, the potential and timestep of the previous timestep are needed to extrapolate a half-timestep value. Since this requires saving and reading more data during restarts, I've wrapped this in the GRAVITY_RESTART flag.
Other minor changes:
Write_HDF5_Dataset replaces HDF5_Dataset, I finally pay penance for my sins
OMP_NUM_THREADS for frontier set to 7
kernel_printf function provided for easily identifying and disabling printfs within kernels... The utility of this may be questionable.
Extra H5open and H5close deleted
Variety of feedback_CIC_gpu.cu changes that I've made @ojwg aware of. Orlando has access to a branch where I've made further refactoring changes but I'm not pushing that to dev for now.
A few changes improve the consistency of outputs between running uninterrupted from t=0 vs interrupting and restarting a run.
GRAVITY_RESTARTflag.Other minor changes:
feedback_CIC_gpu.cuchanges that I've made @ojwg aware of. Orlando has access to a branch where I've made further refactoring changes but I'm not pushing that to dev for now.