Skip to content

These codes are used to complete elastic full waveform inversion using temporal fourth-order FD modelling. In order to accelete the algorith, GPU-based versions are privided here. What's more, the GPU shared memory is used in block level and the efficiency improvement is about 10%~15% on GTX 750ti, where efficiency improvement is dependent on th…

Notifications You must be signed in to change notification settings

LongyanU/FWI-using-temporal-fourth-FD-modelling

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FWI-using-temporal-fourth-FD-modelling

These codes are used to complete elastic full waveform inversion using temporal fourth-order FD modelling. In order to accelete the algorith, GPU-based versions are privided here. What's more, the GPU shared memory is used in block level and the efficiency improvement is about 10%~15% on GTX 750ti, where efficiency improvement is dependent on the length of FD orders.

The codes contains two parts: one is the GPU-based elastic FWI using temporal fourth-order Finite-different modelling method, the other is that we use GPU-shared-memory to optimize the first one further. The tested GPU device is GTX 750ti. The FWI section are contains forward section, which can be used to output the forward seismograms.

How to run the code

Firstly, install the MPICH and Cuda, and set the path correctly on your ~/.bashrc file and source it. Then modify the mpicc and nvcc paths in the Makefile in the package. Then compile and generate executable file, run it by sh run.sh.

main files in the package

Seven files are:

  1. *.cpp,main control code for elastic FWI by MPI. It includes the functions of domain decomposition, multi-scale and encoding number parameters, and the MPI data exchange, model parameter update, and etc.
  2. *.cu, the CUDA code and designed for simultaneous source propagation. The main four GPU kernel functions are forward, reconstructed, backward sections.
  3. headmulti.h, used for declaring all the functions to be called in the elastic_fdtd_3d_FWI_ModelCMultiscale.cpp.
  4. Makefile, used for compiling the program with the CUDA’s nvcc, in addition, some shared library functions are listed in the file. One thing to be mentioned is that we should make sure the paths of cuda and mpi are set correctly in this file.
  5. hostgpu, a name list of nodes of GPUs.
  6. run.sh, used to run the elastic FWI on the nodes listed in hostsgpu by mpiexec or mpirun.
  7. nohup.out, a log file used to write the running status. Created automatically by running run.sh. Two folders are:
  8. input/, used to save initial P- and S-wave velocity and density which are binary and the stored along the depth direction.
  9. output/, is used to store the inversion information, including the inversions of different frequency bands, objective function misfit, and temporal results, such as the synthetic data, residual data andgradients.

FWI inversion data

Overthrust model inversion comparsion between FWI with the temporal 4th-order and with the temporal 2th-order FD. "Tradition" folder contains results of FWI with the temporal 2th-order, and "High_order" folder is FWI with the temporal 4th-order FD.

About

These codes are used to complete elastic full waveform inversion using temporal fourth-order FD modelling. In order to accelete the algorith, GPU-based versions are privided here. What's more, the GPU shared memory is used in block level and the efficiency improvement is about 10%~15% on GTX 750ti, where efficiency improvement is dependent on th…

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Cuda 48.1%
  • C++ 26.5%
  • MATLAB 22.6%
  • C 1.8%
  • Other 1.0%