A 2D LBM-tho phase fortran-based code is presented here. 
It is not a production code. It is not bug-free at all. It is intended to show a possible optimiziation procedure, 
starting from cleaning a "test" code, perform serial optimization at different level (inside and between subroutines), 
Shared memory parallelization using OpenMP and at the end porting it to GPU using OpenACC. 
Only a fortran compilere is needed.

It is not the best performing code, neither it was intended to be. 

You can write (Giorgio Amati - g.amati@gmail.com, or Giacomo Falcucci - giacomo.falcucci@uniroma2.it) and we'll try to reply but no support is provided.
Good Luck ;-) 

* Directory structure  (each directory contains a README file with some details, and some png images to seed the flow)

DOC: this directory contains the documentation
ORIG: this directory contains the original code as it was. It must be used only as a reference
STEP_0: Code cleaned. No optimization performed (now empity)
STEP_1: Single subroutine level optimization
STEP_2: Code improvement (F90,allocatable vectors). No optimization performed
STEP_3: Multi subroutine level optimization
STEP_4: Changing data allocation 
STEP_5: Fine tuning, merging (big) loops, fusing further subroutine
STEP_6: OpenMP version
STEP_7: OpenACC version + some fix (e.g. inithydro1.F90)



All the time reported here refers to a 

CPU: Intel(R) Xeon(R) Platinum 8260 CPU @ 2.40GHz with the following compilers
GPU: Nvidia V100 and Nvidia A100

* gfortran: gcc version 8.4.1 20200928 (Red Hat 8.4.1-1) (GCC) 
* ifort: fort version 2021.5.0
* nvfortran: nvfortran 21.9-0 64-bit target on x86-64 Linux -tp skylake