LVC21F-212 Development of Deep Learning Library for AArch64 CPU

Session Abstract

Level: Intermediate This session introduces development of oneDNN for Armv8-A + SVE architecture. oneDNN is one of the de facto deep learning processing library, which is originally developed for Intel architecture. oneDNN optimizes the execution code by writing at the x86_64 instruction level with the Just-In-Time assembler Xbyak. The porting work of oneDNN for AArch64 requires to rewrite the x86_64 instruction-level code into that of AArch64 instruction-level with the Just-In-Time assembler for AArch64, Xbyak_aarch64. It is very time-consuming work. To eliminate this work, we developed the binary translator Xbyak_translator_aarch64 which is also introduced in this session.

Session Speakers

Kentaro Kawakami

Senior Researcher (Fujitsu Limited)

He joined Fujitsu Laboratories Ltd. in 2007. He has been involved in R&D of image codec LSIs and wireless sensor nodes, and is currently engaged in R&D of AI software for Arm HPC. His GitHub account name is "kawakami-k".

Level: Intermediate This session introduces development of oneDNN for Armv8-A + SVE architecture. oneDNN is one of the de facto deep learning processing library, which is originally developed for Intel architecture. oneDNN optimizes the execution code by writing at the x86_64 instruction level with the Just-In-Time assembler “Xbyak”. The porting work of oneDNN for AArch64 requires to rewrite the x86_64 instruction-level code into that of AArch64 instruction-level with the Just-In-Time assembler for AArch64, “Xbyak_aarch64”. It is very time-consuming work. To eliminate this work, we developed the binary translator “Xbyak_translator_aarch64” which is also introduced in this session.

comments powered by Disqus

Other Posts

Sign up. Receive Updates. Stay informed.

Sign up to our mailing list to receive updates on the latest Linaro Connect news!