Tensor accelerators now represent a growing share of compute resources in modern CPUs and GPUs. However, they are hard to program, leading developers to use vendor-provided kernel libraries that support tensor accelerators. As a result, the usage of tensor accelerators is limited to the provided interface, mainly designed for traditional ML and scientific computing workloads.In this paper, we show that tensor accelerators can improve the performance of applications beyond simple variants of MatMul. For example, many image processing pipelines are linear transformations over matrices in disguise and can therefore utilize such specialized hardware. This is nonetheless hindered by the difficulties in programming tensor accelerators. We tackle this problem with compiler-based techniques. We use the Halide user-schedulable language and express operations as Halide algorithms succinctly. To this end, we implement a flexible tensor instruction selector based on equality saturation. The tensor instruction selector supports both CPU- and GPU-attached tensor accelerators and works with existing scheduling operations (e.g., producer-consumer fusion). Together, this enables developers to write diverse accelerator-leveraging applications in a few dozen lines.Using our system, we demonstrate the potential of tensor accelerators beyond their traditional domains. We implement several image processing pipelines (e.g., filtering, resampling, and denoising) in our system and evaluate them against non-accelerator-leveraging baselines. We show that these pipelines can achieve significant speedups. For example, a downsampling routine is sped up by 6.1× by utilizing Tensor Cores on an Nvidia RTX 4070 GPU.

Simulating the expansion of a Type II supernova using an adaptive computational fluid dynamics (CFD) engine yields a complex mixture of turbulent flow with dozens of physical properties. The dataset shown in this sketch was initially simulated on iVEC's EPIC supercomputer (a 9600 core Linux cluster) using FLASH [Fryxell et al. 2000] to model the thermonuclear explosion, and later post-processed using a novel integration technique to derive the radio frequency emission spectra of the expanding shock-wave front [Potter et al. 2011]. Model parameters have been chosen to simulate the asymmetric properties of the SN 1987A remnant [Potter et al. 2009].

There are strong indications that the future of interactive graphics involves a more flexible programming model than today's OpenGL/Direct3D pipelines. That means that graphics developers will need a basic understanding of how to combine emerging parallel-programming techniques with the traditional interactive rendering pipeline.