******* WHAT YOU HAVE TO DO ********
- Install following dependencies
	- MPI compiler (openmpi, intel mpi, etc.)
	- Open MP 
	- Make utility
- Make desired changes to the Makefile
	- Change the compiler. Default is 'mpicc'. 
	- Set code optimization levels using -ON flag, where N is the optimization level
	- Make more changes if necessary
- Run 'make' in the command line

******** WHAT IT DOES ********
- Generates a library ../lib/libmbir4d.a
- This library after linking can be used to call the following reconstruction function :
	- Function Name : reconstruct
	- Inputs (in order)
		- float** object : Address of the pointer to the reconstructed object.
			*object is a pointer to a 1D array in raster order of size recon_num x proj_rows x proj_cols x proj_cols where 'recon_num' is the number of time samples in the reconstruction, 'proj_rows' is the number of projection slices (along the axis of rotation), and 'proj_cols' is the number columns in the projection (or number of pixels along a row of detector bins)
		- float* projections : Pointer to the projection data. 
			'projections' is a pointer to a 1D array in raster order (row-major order) of size proj_num x proj_cols x proj_rows. A projection is typically computed as the logarithm of the ratio of the measurement in the absence of the sample to the measurement with the sample..
		- float* weights : Pointer to the weight data.
			'weights' is a pointer to a 1D array in raster order of size proj_num x proj_cols x proj_rows. Every entry of 'weights' is used to appropriately weigh each term of the 'projections' array in the likelihood term of MBIR. 'weights' is typically set to be equal to the detector measurement data. 
		- float* proj_angles : Pointer to the list of angles at which the projections are acquired (in radians).
		- float* proj_times : Pointer to the list of times at which the projections are acquired.
		- float* recon_times : Pointer to a array of reconstruction times. The reconstruction is assumed to be peicewise constant with fixed/varying step-sizes. Thus, the array 'recon_times' contains the times at which the steps occur. For example, to do a single 3D reconstruction, we use a array of two elements, first element being the time of the first projection and the second element being the time of the last projection.  
		- int32_t proj_rows : Number of rows in the projection data i.e., the number of projection slices (along the axis of rotation)
		- int32_t proj_cols : Number of columns in the projection data i.e., the number of pixels of the detector perpendicular to the axis of rotation.
		- int32_t proj_num : Total number of projections used for reconstruction
		- int32_t recon_num : Total number of reconstruction time steps. Note that the number of elements in the array pointed by 'recon_times' is 'recon_num + 1'.
		- float vox_wid : Side length of each cubic voxel. The unit of reconstruction is the inverse unit of 'vox_wid'.
		- float rot_center : Center of rotation. For example, if the center of rotation coincides with the center of the detector then rot_center = proj_rows/2. 
		- float sig_s : Spatial regularization parameter to be varied to achieve the optimum reconstruction quality. Reducing 'sig_s' will make the reconstruction smoother and increasing it will make it sharper but also noisier.
		- float sig_t : Temporal regularization parameter to be varied to achieve best quality. Reducing it will increase temporal smoothness which can improve quality. However, excessive smoothing along time might introduce artifacts.
		- float c_s : Parameter of the spatial qGGMRF prior. It should be chosen such that c_s < 0.01*D/sig_s where 'D' is a rough estimate for the maximum change in value of the reconstruction along an edge in space.
		- float c_t : Parameter of the temporal qGGMRF prior. It should be chosen such that c_t < 0.01*D/sig_t where 'D' is a rough estimate for the maximum change in value of the reconstruction along a temporal edge.
		- float convg_thresh : Convergence threshold expressed as a percentage (chosen in the range of 0 to 100).
		- float remove_rings : Legal values are '0', '1', '2' and '3'. If '0', then the algorithm does not do any ring correction. If '1', it does ring correction by estimating an offset error in the projection data. If it is equal to or greater than '1', it does ring correction. '1' does a uncontrained optimization. '2' enforces a zero mean constraint on the offset errors. '3' enforces a zero constraint on the weighted average of the offset errors over overlapping rectangular patches. 
		- int32_t quad_convex : Legal values are '0' and '1'. If '1', then the algorithm uses a convex quadratic forward model. This model does not account for the zinger measurements which causes streak artifacts in the reconstruction. If '0', then the algorithm uses a generalized Huber function which models the effect of zingers. This reduces streak artifacts in the reconstruction. Also, using '1' disables estimation of variance parameter 'sigma' and '0' enables it.
		- float huber_delta : The parameter \delta of the generalized Huber function which models the effect of zingers. Legal values are in the range 0 to 1.
		- float huber_T : The threshold parameter T of the generalized Huber function. All positive values are legal values.
		- uint8_t restart : Legal values are '0' and '1'. It is typically used if the reconstruction gets killed for any reason. If '1', the reconstruction starts from the previously run multi-resolution stage.
		- FILE *debug_msg_ptr : Pointer to the file to which the debug messages should be directed. Use 'stdout' if you do not want to direct messages to a file on disk.
	- Outputs (Return value) : '0' implies a safe return. 
- Useful Compiler Macros
	- NO_DEBUG : Reduces the number of debug messages in the file debug.log.
	- NO_COST_CALCULATE : Improves computation time by not calculating the optimization cost function in every iteration.
 
