Accelerating Electromagnetic Field Simulations Based on Memory-Optimized CPML-FDTD with OpenACC

Round 1

Reviewer 1 Report

Review of the manuscript entitled “Accelerating electromagnetic propagation based on

memory-optimized CPML-FDTD with OpenACC”.

In this manuscript, GPU computation of FETD scheme is implemented through OpenACC, and the effect of acceleration is achieved. But the manuscript needs a heavy revision, I am not convinced that it can be published in Applied Sciences. 

1. As stated by the authors, this manuscript is just an original code transfer to OpenACC, I don't see innovative work.

2. The title of the paper seems to expand the research results, and the content of the paper is limited to the study of a two-dimensional simulation of high-frequency electromagnetic propagation problems. And the title of the article is not correct. To speed up the calculation of electromagnetic propagation simulations, not to speed up electromagnetic propagation.

3. The introduction of this article is not properly structured. Most of the introduction introduces the advantages of OpenACC, without clarifying what scientific problem needs to be solved, the importance of the problem, the meaning of the problem, etc.

4. In particular, most of the manuscript is the same as the reference [15], for example, all of the content of "2. Propagation and Algorithm", most of "3. Algorithm and Flux Diagram", and the numerical experimental model.

5. In the numerical simulation section, the model discretization case is an important factor in the computational memory consumption and computational time already simulation accuracy, but I do not see this information.

6. Rearranging the conclusions, other works have demonstrated that OpenACC can accelerate the simulation of electromagnetic propagation, for example, reference [25].

7. Does the author need to explain the limitations of the proposed method and the possible future trends?

Author Response

Please find attached a PDF file of our responses.

Best Regards

Dr. Carlos Couder

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper aims at accelerating electromagnetic propagation based on memory-optimized CPML-FDTD with OpenACC.

1. My main concern is the novelty of this manuscript. The novelty of the proposed approach is poor or not properly emphasized, and the contributions are inadequate. It is suggested to add more new technical contents and comparative experiments. 

2. Furthermore, the authors should clarify the motivation of the proposed work and difference compared to the other related work. Moreover, the writting and the layout are really poor. 

3. The Abstract should be improved greatly, the authors should put more contents regarding contributions instead of background.

4. The key words should be in alphabetical order.

5. Some important related work regarding high-performance computing is missing, they should be added and discussed in the manuscript. The current references are low-quality and outdated. Such as the work below:

fgSpMSpV: A Fine-grained Parallel SpMSpV Framework on HPC Platforms. ACM Trans. Parallel Comput. 9(2): 8:1-8:29 (2022)

GPU-Accelerated Parallel Finite-Difference Time-Domain Method for Electromagnetic Waves Propagation in Unmagnetized Plasma Media. CoRR abs/1709.00821 (2017)

CASpMV: A Customized and Accelerative SpMV Framework for the Sunway TaihuLight. IEEE Trans. Parallel Distributed Syst. 32(1): 131-146 (2021)

6. The authors need to add and conduct more experiments and analysis in the manuscript. The present experimental results are not convincing.

7. The language has some typos and errors (it is very important to use the correct English!).

I am happy to review it again after a major revision.

Author Response

Please find attached a PDF file of our responses.

Best Regards

Dr. Carlos Couder 

Author Response File: Author Response.pdf

Reviewer 3 Report

In this work, the authors accelerate a Fortran code for simulating the propagation of electromagnetic waves that solves the Maxwell curl equations. The acceleration is done with the OpenACC library, which, in my opinion, is the best option when working with previously written codes with Fortran and C. I appreciate the work done by the authors and recommend it for publishing with minor revisions. The following comments should be addressed before publishing.

1) It would be nice to see a comparison to other packages. The problem is well-known and there are other packages solving it.

2) Why one version of the program is much faster than another? Is it related to efficient data caching by GPU?

3) Could you provide a profile of the program with NVidia Visual Profiler or other tools? It would be nice to see potential parts for further optimization.

4) I don't understand why OpenMP didnt work well. Could you try to interpret OpenACC instructions as OpenMP? 

5) As I understand, computations are done in single precision. I though that such problems require double. Could you comment on that?

6) Have you tried to adapt the code for using several GPUs? What are possible bottlenecks in this case?

7) What are data saving costs? Is data saving done asynchronously? 

8) Line 134, I dont understand $ after word data.

Author Response

Dear reviewer, 

We apologize if the answers file was not uploaded correctly or was a wrong file. Here is the new version.

Best Regards.

Dr. Carlos-Couder. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I would like to suggesti the authors have more effcitive review of the current method to make the manuscript better. Such as, 

CNCSU-FDTD,CNDS-FDTD

Sun G L, Trueman C W. Efficient implementations of the crank-nicolson scheme for the finite-difference time-domain method[J]. IEEE Transactions On Microwave Theory and Techniques, 2006,54(5):2275-2284.

CNDS-FDTD

Jiang H L, Wu L T, Zhang X G, et al. Computationally efficient cn-pml for EM simulations[J]. IEEE Transactions On Microwave Theory and Techniques, 2019,67(121):4646-4655.

CNAFS-FDTD

Sun G L, Trueman C W. Unconditionally-stable fdtd method based on crank-nicolson scheme for solving three-dimensional maxwell equations[J]. Electronics Letters, 2004,40(10):589-590.

BEDS:  

Liu, Shangbin et al. Fast 3D transient electromagnetic forward modeling using BEDS-FDTD algorithm and GPU parallelization, GEOPHYSICS (2022),87(5): E359

Author Response

We addressed the references to improve the manuscript.

We thank you for your comments.

Reviewer 2 Report

Thanks for addressing the reviewers' comments & concerns, I recommand to accept it after polishing.

Author Response

We thank you for your comments on improving the manuscript.