Our lab has just started analyzing Apple’s M4 chip. It turns out that M4 supports the Scalable Matrix Extension (SME) of the Arm Architecture. This opens the way for open source developments that support M4’s matrix accelerator(s). We plan to add SME support to the JITter LIBXSMM in the next few weeks, which will allow us to integrate SME into upstream software. We are documenting these efforts on a dedicated homepage. Check it out!

Image of the iPad Pro on which our M4 work is done.