article · 2025

Kernel-level Access-based Memory Transfer Optimization Framework (KAMTOF): Parallel Incremental GPU Porting of CONVERGE CFD using SYCL

Sidarth Narayanan D, Josin Tom, Joshua Duncan, Mohsen Moghadasi Barazandeh, Sparsh Ganju, Raju Mandhapati, Yogiraj Mantri, Cooper Burns D

Cited by 0Abstract and details only

Abstract

Porting a large-scale computational code from CPUs to GPUs is a complex and time-intensive process, likely performed incrementally by multiple developers. This requires working on code sections without full knowledge of whether adjacent computations have been, or will be, ported. A minimally invasive framework is proposed that automatically minimizes memory transfers to aid the conversion of a legacy CPU code to GPUs. SYCL is leveraged to obtain hardware agnosticism. Developers focus solely on the computational characteristics of assigned code sections, while the framework optimizes data movement using lightweight metadata that tracks variable states and dynamically determines whether data transfers between CPU and GPU are necessary. Transfers of unmodified data are avoided and single-machine data ownership is enforced to guarantee data validity. This framework was utilized to port the incompressible solver of CONVERGE CFD [1], a large-scale computational fluid dynamics (CFD) code, achieving an efficient GPU implementation with minimal data transfer overhead, in a short time.

Cite this paper

APA 7

D, S. N., Tom, J., Duncan, J., Barazandeh, M. M., Ganju, S., Mandhapati, R., Mantri, Y., & D, C. B. (2025). Kernel-level Access-based Memory Transfer Optimization Framework (KAMTOF): Parallel Incremental GPU Porting of CONVERGE CFD using SYCL. https://doi.org/10.1109/sci68648.2025.11333834

MLA 9

D, Sidarth Narayanan, et al. “Kernel-level Access-based Memory Transfer Optimization Framework (KAMTOF): Parallel Incremental GPU Porting of CONVERGE CFD using SYCL.” 2025, pp. 1–10, https://doi.org/10.1109/sci68648.2025.11333834.

Chicago

D, Sidarth Narayanan, Josin Tom, Joshua Duncan, Mohsen Moghadasi Barazandeh, Sparsh Ganju, Raju Mandhapati, Yogiraj Mantri, and Cooper Burns D. 2025. “Kernel-level Access-based Memory Transfer Optimization Framework (KAMTOF): Parallel Incremental GPU Porting of CONVERGE CFD using SYCL.” https://doi.org/10.1109/sci68648.2025.11333834.

Harvard

D, S.N. et al. (2025) ‘Kernel-level Access-based Memory Transfer Optimization Framework (KAMTOF): Parallel Incremental GPU Porting of CONVERGE CFD using SYCL’. Available at: https://doi.org/10.1109/sci68648.2025.11333834.

BibTeX

@article{d2025kernel,
  title = {{Kernel-level Access-based Memory Transfer Optimization Framework (KAMTOF): Parallel Incremental GPU Porting of CONVERGE CFD using SYCL}},
  author = {D, Sidarth Narayanan and Tom, Josin and Duncan, Joshua and Barazandeh, Mohsen Moghadasi and Ganju, Sparsh and Mandhapati, Raju and Mantri, Yogiraj and D, Cooper Burns},
  pages = {1--10},
  year = {2025},
  doi = {10.1109/sci68648.2025.11333834}
}

Made from the index record. Check names, capitalisation and page numbers against the paper. Need another style? AI for Research has over 10,000.