Machine Learning with NIR - Extensions and Optimization Techniques

Explore extensions to NIR (New Intermediate Representation) for machine learning applications in this 49-minute conference talk from XDC 2025. Learn about new tensor operations and intrinsics that enable ML workloads within the NIR framework. Discover optimization techniques through hardware-independent and hardware-specific compiler passes, and examine machine code generation strategies for both GPGPU units and dedicated fixed-function accelerator hardware. Gain insights into how NIR can be leveraged to bridge the gap between high-level machine learning operations and efficient hardware execution across different computing architectures.