Our researchers Marc Casas, Alexandre Santana and Adrià Armejach will be presenting a paper about “Efficient Direct Convolution Using Long SIMD Instructions” on February 28 from 16:20h to 16:40h during the main PPoPP 2023 Conference.
The paper demonstrates the state-of-the-art proposals to compute convolutions on architectures paper demonstrates that state-of-the-art proposals to compute convolutions on architectures with CPUs supporting SIMD instructions deliver poor performance for long SIMD lengths due to frequent cache conflict misses.
About PPoPP
PPoPP is the premier forum for leading work on all aspects of parallel programming, including theoretical foundations, techniques, languages, compilers, runtime systems, tools, and practical experience. In the context of the symposium, “parallel programming” encompasses work on concurrent and parallel systems (multicore, multi-threaded, heterogeneous, clustered, and distributed systems; grids; datacenters; clouds; and large scale machines). Given the rise of parallel architectures in the consumer market (desktops, laptops, and mobile devices) and data centers, PPoPP is particularly interested in work that addresses new parallel workloads and issues that arise out of extreme-scale applications or cloud platforms, as well as techniques and tools that improve the productivity of parallel programming or work towards improved synergy with such emerging architectures.