Parallelized the /examples/performance_test_voxels.cpp example - now every process updates its central voxel. A bit unfair for the parallel version
but the major work done is in comparison and not writing a single voxel substrate density. To let a single process (the appropriate process which conatins the
central voxel would be unfair as ALL processes must execute i.e. the ENTIRE DOMAIN must be covered).
Further, it is very difficult to change the signature of the supply function and target function pointers - this will need a change in the core files of BioFVM which would break the other
examples. This is also UNDESIRABLE as this is "just" a performance test and not a "real" example.
The output is produced in a scaling_*.txt file.

The README file updated (1) Text improvement Gaurav (2) Markdown Arnau

Added script to plot it plotting_microenv_tumor.py (by Arnau Montagud) in ./matlab
Added script file to run this example as script_3DtumorX.sh
To compile this GS_BioFVM_X/GS_BioFVM> make 3DtumorX and then run using sbatch script_3DtumorX.sh
The main Makefile was modified to incorporate this example.
The executable will need to be deleted manually from the examples directory (before committing to Git)

Possibly it is only the README file that's been updated, hence merging.

function, when we calculate the coordinates of the MPI process
corresponding to the (x,y,z) position of the Basic Agent, we were not
subtracting the initial starting coordinates of the physical domain from
the tempPoint[] array. This will not work when the physical domain
starts from a non-zero value. Thus added
tempPoint[0] = tempPoint[0] - lower_x_boundary_coordinate_of_domain;
similary for the y, z dimensions.
Have not compiled it but committing it.

BioFVM with X decomposition - obtaining much better timings than the Z decomposition. Also remember to use the ppr notation in the script file. Best results being obtained from 2 processes per node i.e. one per NUMA region and then nodes can be increased.

Working version till 4 processes. It doesn't work with 8 processes because of (99.99% sure) divisibility problem. The reason: domain size in Z-dimension: (1000-0)=1000, since voxel size=10 then voxels in Z-direction = 1000/10 = 100. Since there are 8 processes we have to divide 100 by 8 --> not perfectly divisible. Hence change domain size. Further the order in which the SERIAL program was generating sources and sinks was it was alternating between sources and sinks but the PARALLEL version was genearting all sources first then generating all sinks. Although random numbers generated were the same but they were alternating between sources and sinks in SERIAL version but going all to sources first then all sinks next in the PARALLEL version. Repairing this made the output the same !

Parallelized Code -- output is 'slightly' different than the serial version of the code. Remember decomposition is 1-D in Z direction. To check try with : 1 process to see if output is the same...then with 2 processes then with 4...

Added parallel code for : initialization of solver, forward elimination and backward substitution. Needs to be checked...this is just the first attempt at adding parallel code. DO NOT FORGET to comment out the sequential code for (1) forward elimination (2) backward substitution. The part of the code to be commented out is clearly marked with comments.