Automating the AI Vecotorization Workflow

We extended our AI Vectorization Workflow by integrating the open-source project onnx2c with the emmtrix transformation pipeline to automatically generate optimized C code for supported platforms.

onnx2c converts ONNX models into generic sequential C code.
Based on this, our transformation flow within emmtrix Code Vectorizer applies a series of optimizations – Inline, PtrResolve, DeadStmtRemoval, IdiomRecognizer, LoopFusion, and ArrayToScalar – followed by AutoVectorize, which automatically vectorizes all compatible code sections.

These optimizations represent the current state of development, and additional optimization passes are already in progress.

This end-to-end automation accelerates the conversion from ML models to optimized, vectorized C code, enabling faster and more efficient hardware execution.