WG211/M26Bodo

Code Generation in the Mirror --- on Generating ONNX for Scientific Codes

Sven-Bodo Scholz

Many new architectures have been created aiming at improved AI computations. These machines promise faster computations or lower power consumption in comparison to standard hardware. Given that AI computations are mainly based on a small set of linear algebra operations it should be possible to leverage the benefits of these architectures for scientific computation as well. In a first attempt to do so, we look into the feasibility of leveraging Open Neural Network Exchange (ONNX) as an intermediate representation which is supported by a rather wide range of accelerator platforms.

The key challenge here is that ONNX is on a rather high level of abstraction which seems to to turn the code generation task into a code analysis task. This talk presents an idea to address this analysis task by a combination of functional abstraction and ONNX code generation. We present our first experiences when turning scientific applications written in SaC into ONNX programs.