Porting Math Libraries Across Heterogeneous Architectures

June 18, 2020

Existing math kernel libraries have lacked portability across heterogeneous platforms—until now. A unifying programming model—and availability of standard library interfaces—enables development of performance-portable libraries among diverse hardware architectures. In this episode, we talk about a collaborative project to make this portability possible, and dive into Mehdi’s work to enable the first math library implementation for oneAPI on Nvidia GPUs.

Julia Sukharina

Senior Engineering Manager

Mehdi Goli

Principal Software Engineer