I was just wondering if there is any plans to release a 2D version of the C++ code or Cuda code, or can this be done using an infinite plane approach along one dimensions? If my understanding is correct the 2D case should be a subset of the 3D, although I worry about making this assumption and do not know if it would be slowed down as compared to a 2d code implementation.
Thanks,
Andrew
Hi Andrew,
We have no immediate plans to release 2D C++ codes, largely because the compute times using MATLAB or MATLAB+GPU are already quite short. Do you have an application in mind where minimising the compute time is critical?
You could certainly run 2D simulations using the 3D C++ code using a small value for Nz (e.g., Nz = 3). You will need to duplicate your source (e.g., source.p0 or source.p_mask) in the third dimension, and turn off the PML using the optional inputs 'PMLSize', [10, 10, 1] and 'PMLAlpha', [2, 2, 0]. However, this will definitely be slower than using an optimised 2D code due to the redundant calculations in the third dimension.
Brad.
Thanks Brad,
For my case it is mostly the number of simulations that need to be run, as there are multiple possible setups and imaging types I want to experiment with. I will likely be upgrading my hardware and just wanted to know if these were being developed before I make decision to purchase something that may not be necessary. Thanks again though.
2D CUDA/C++ would be nice.
I currently run thousands of simulations on 640x480 grid with 1700 time steps and one 2D simulation takes ~35s (gpuArray-single on GTX750Ti) which is fine if you don't need few thousands of them...
Great toolkit anyway! Thanks a lot!
You must log in to post.