Teflon has been merged into Mesa 24.1 as a Gallium3D front-end that TensorFlow can load for delegating the execution of operations in a neural network model. Teflon was created initially for the Etnaviv Gallium3D driver for being able to run AI inferencing on Vivante NPUs.
Tomeu Vizoso of Collabora has been spending much time in recent months working on enabling Vivante NPU support within the Mesa Etnaviv driver. Kernel-side back in Linux 6.3 Vivante NPU cores were added to the Etnaviv DRM driver while the Mesa enablement effort has been ongoing and finally today was the initial code upstreamed for Mesa 24.1.
This merge adds Teflon as a TensorFlow Lite delegate and Vivante NPU driver support.
“These changes allow to run inference on a NPU from VeriSilicon that is closely based on the Vivante GPU IP.”
There’s over eleven thousand lines of new code as part of this merge for enabling the TensorFlow Lite offloading to Gallium3D initially for Vivante NPU cores. It’s quite an interesting addition to Mesa and will be interesting to see what other users of the Gallium3D Teflon front-end may come about in time given the growing AI popularity.
Be the first to comment