Researchers from around the globe are working using NSIMD for its portability and very easy integration with AMTs like HPX #hpc #research #simd #hpx
October 26 2020

Researchers from India, the US and Germany are working using NSIMD. They needed to test some code on various architectures including the latest Arm SVE SIMD extension. Using NSIMD made their work easier thanks to its portability accross hardware architectures and its ease of integration with existing paradigms such as Asynchronous Many-Task programming models (HPX).

One team consisting of Bine Brank and Dirk Pleiter from the Jülich Research Centre; Nikunj Gupta, Rohit Ashiwal, and Sateesh K. Peddoju from the department of CSE, IIT Roorkee have worked on performance evaluation of ParalleX (HPX) execution model on Arm-based platforms. The advent of Arm-based processors provides an alternative to the existing HPC ecosystem, which is primarily dominated by x86 processors. They have ported an asynchronous many-task runtime system based on the ParalleX model, i.e., High-Performance ParalleX (HPX), to evaluate it on the Arm ecosystem with a suite of benchmarks. They needed NSIMD to write the latter and run them on x86 and Arm CPUs to measure and compare performances. Their very interesting work will soon be available at IEEE, take a look at the Arxiv version.

Another team consisting of Nikunj Gupta from the Department of CSE, IIT Roorkee; Steve R. Brandt, Bibek Wagle, Nanmiao Wu, Alireza Kheirkhahan, Patrick Diehl and Hartmut Kaiser from Center of Computation & Technology, Louisiana State University and Felix W. Baumann from PricewaterhouseCoopers have worked on deploying a task-based runtime system on Raspberry Pi clusters. Although HPC computations are usually done on big servers such as the latest Fujitsu's Fugaku, Raspberry Pis can also be useful on their own and provide an inexpensive platform to become familiar with Arm architecture. They used NSIMD to ease the writing of the algorithms they used to study the impact of explicit vectorization on their work. Their work is available on Arxiv, take a look at this very interesting work.

[1] Jülich Research Centre:
[2] Department of CSE, IIT Roorkee:
[3] HPX:
[4] Center of Computation & Technology, Louisiana State University:
[5] PricewaterhouseCoopers:
[6] Link to 1st research paper:
[7] Link to 2nd research paper: