Closed wookhyun6165 closed 4 years ago
Manual setup of sigmas and PEC is not supported yet.
To use parallel mode you need to build with it (see https://github.com/zer011b/fdtd3d/blob/master/Docs/Build.md#build-flags). To build with parallel support for 3D mode pass this to cmake:
-DPARALLEL_GRID=ON -DPARALLEL_GRID_DIMENSION=3 -DPARALLEL_BUFFER_DIMENSION=x
Without this parameters, binary has no support for MPI computations and mpiexec will just launch two sequential and unrelated processes. --buffer-size
sets up size of parallel buffers, you can try to tune it for yourself, --manual-topology
allows to setup custom topology instead of automatic one. Automatic one should be the best in most cases.
However, it is strange that simulation time has doubled, it should have been the same in my opinion.
All this stuff is in InternalScheme class. See https://github.com/zer011b/fdtd3d/blob/master/Docs/Design.md#high-level-design for overall design. InternalScheme code is shared between CPU and GPU.
This is the relative path from the directory with source of repository. Release
is manually created dir, I assumed you have named it the same way as in the instruction. fdtd3d
is executable, on Linux you don't need extension for executables.
Feel free to reopen this if you have more questions
Hi. I'm wookhyun. First of all, thank you for your kind reply. I've solved all the previous problems. Thank so much.
Unfortunately, there are four more questions.
mpiexec -n 2 ./Release/Source/fdtd3d --cmd-from-file ./Examples/sphere.txt
As a result, the simulation time has doubled. How can i solve this problem? Will this problem be solved if i add "--buffer-size (default: 1), --manual-topology" to cmd.txt?
Thanks for your interest. Have a good night.