|
Mathematical Foundations of Programming
Study of the vectorization efficiency of loop nests with an irregular number of iterations
A. A. Rybakov, S. S. Shumilin Joint Supercomputer Center
Abstract:
Computation vectorization is an important low-level optimization used to create highly efficient parallel code. However, when used in context with an unknown program execution profile, a danger of low effectiveness of the application emerges. This is especially pronounced when vectorizing nests of cycles with an irregular number of iterations of the inner loop. The article discusses a comparison of the theoretical and practical efficiency of vectorization on the example of Shell sorting, since this program code is extremely inconvenient for vectorization.
Key words and phrases:
vectorization, AVX-512, loop sockets with an irregular number of iterations, Shell sorting, theoretical acceleration.
Received: 28.02.2019 Accepted: 30.11.2019
Citation:
A. A. Rybakov, S. S. Shumilin, “Study of the vectorization efficiency of loop nests with an irregular number of iterations”, Program Systems: Theory and Applications, 10:4 (2019), 77–96
Linking options:
https://www.mathnet.ru/eng/ps356 https://www.mathnet.ru/eng/ps/v10/i4/p77
|
Statistics & downloads: |
Abstract page: | 129 | Full-text PDF : | 113 | References: | 10 |
|