software-stack

Research software map from model/kernel entry to silicon evidence

Compilation flow

Kernel or model → MLIR transformations → arithmetic generation → CIRCT / RTL → physical implementation → evaluation.

Components

Front-end and ingestion

  • MLModelMLIRGEN: exports model artifacts from PyTorch to MLIR.
  • Faust and tensor kernels as entry points (DSP + AI use cases).

Arithmetic-aware compilation

  • Emeraude-MLIR: custom dialects and lowering passes.
  • HAriCo/FloPoCo direction for explicit arithmetic materialization.

HDL and interop

  • VH2V: VHDL-to-Verilog bridge where needed for OSS flow compatibility.

Physical implementation and campaign orchestration

  • SUF: reproducible campaign orchestration over open EDA flows.
  • OpenROAD-GPL Experiments: placement-centric experiments and visualization.

Layout post-processing and communication

  • gdsiistl: GDS to STL pipeline.
  • Codez: annotated code visuals for talks/posters.

Evaluation

Generated circuits are evaluated with physical design tools to test arithmetic and compiler cost models.