Vector Extensions for High-performance Computing

Scalable, vector-length agnostic instruction-set extensions, such as the RISC-V Vector Extension (RVV) and ARM’s Scalable Vector Extension (SVE), have strong potential for high-performance computing (HPC). Realizing that potential depends on compilers vectorizing code well and on applications making use of the vector units.

RISC-V RVV. A TSVC-based instruction-coverage analysis compiles the TSVC benchmarks with GCC and LLVM, in both vector-length agnostic and vector-length specific modes, and checks the emitted instructions against the RVV 1.0 specification. It identifies missed instructions, classifies why vectorization fails, and suggests how benchmark suites could better reflect RVV and guide compiler development.
ARM SVE. A study of SVE performance across HPC applications on the Nvidia Grace CPU.

This work is with Maya Gokhale and Pei-Hung Lin at Lawrence Livermore National Laboratory, Ivy Peng at KTH Royal Institute of Technology, and Jenq-Kuen Lee at National Tsing Hua University.

Publications

SC-W

RISC-V Vectorization Coverage for HPC: A TSVC-Based Analysis

Hung-Ming Lai, Pei-Hung Lin, Maya B. Gokhale, Ivy Bo Peng, Hiren Patel and Jenq-Kuen Lee

In Proceedings of the SC '25 Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis, pp. 1676–1683, November, 2025

DOI

BibTeX

@inproceedings{lai_25_rvv,
  author = {Hung-Ming Lai and Pei-Hung Lin and Maya B. Gokhale and Ivy Bo Peng and Hiren Patel and Jenq-Kuen Lee},
  booktitle = {{Proceedings of the SC '25 Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis}},
  pages = {1676--1683},
  publisher = {ACM},
  month = {nov},
  doi = {10.1145/3731599.3767535},
  title = {{RISC-V Vectorization Coverage for HPC: A TSVC-Based Analysis}},
  year = {2025}
}

Euro-Par

ARM SVE Unleashed: Performance and Insights Across HPC Applications on Nvidia Grace

Ruimin Shi, Gabin Schieffer, Maya B. Gokhale, Pei-Hung Lin, Hiren Patel and Ivy Bo Peng

In Euro-Par 2025: Parallel Processing, pp. 33–47, August, 2025

PDF DOI

BibTeX

@inproceedings{shi_25_sve,
  author = {Ruimin Shi and Gabin Schieffer and Maya B. Gokhale and Pei-Hung Lin and Hiren Patel and Ivy Bo Peng},
  booktitle = {{Euro-Par 2025: Parallel Processing}},
  series = {Lecture Notes in Computer Science},
  pages = {33--47},
  publisher = {Springer},
  month = {aug},
  doi = {10.1007/978-3-031-99857-7_3},
  title = {{ARM SVE Unleashed: Performance and Insights Across HPC Applications on Nvidia Grace}},
  year = {2025}
}