Linux Based Assignment Help On GCC compiler
The Linux based Assignment Help On GCC compiler is based on grid is going to be split two part one is middle the biggest part can stay together. The most important are rows on outskirt as they need neighbors to be completely calculated. This is only idea, if better approach is known or possible to apply please feel free to do so. The most important is to implement “HALO exchange” (MPI_SEND, MPI_ISEND, MPI_RECEIVE, MPI_IRECEIVE etc.). This particular problem has to scale between 4 nodes and each node includes 16 processors you can see how it work
You can testing and final result will be mainly for 1024×1024 and 700×500 grid size
700×500 grid size was deliberately chosen so that it does NOT divide evenly by 64. This means that the code has to be able to handle cases where you need to distribute different amounts of work to different processes.
In code comments, could you please make a brief description of why the particular code was
The final implementation on Lattice Boltzmann serial code d2q9.c needs to pass “check_result”
verification. I will be able to test it on actual cluster. Or if there is a way, script is provided in files.
I got VTune profiler available to profile code if needed during the process. I know Allinea profiler can
be used but I have no resources to do so. Down below is more information to understand different files and their purpose. No changes or coding is going to be on these files. “cfgs folder” –> includes particular problems which main “d2q9-bgk.c” needs to produce output. No need to change anything there as that is only for reference or testing.
“final_state.plt” –> is script to produce the picture after the “d2q9-bgk.c” finishes. This is only for
reference and testing. No need to change the code.
“final_state.png” –> is the example which is produced after the program runs. This one is I think for
“Makefile” –> is for creating the executable file from “d2q9-bgk.c”
“mpi_submit” –> is script for submitting job to a cluster
“check_results” –> is for checking the result after the main program runs. Folder “refs” holds the
Kod nemusi byt perfektne implementnuty. Najdolezitejsie je ze program compilne a prejde
verifikaciou. Taktiez ze cas bude rychlejsi ako predosla OpenMP implementacia co bolo 8.3 sekundy
na 128×256 grid size.