MultiVic Benchmarks

This repository contains benchmarks for evaluating the performance of the MultiVic architecture. It offers python scripts and makefiles for easy generation of benchmarks for different configurations. The matmul benchmark has been used in the evaluation of the MultiVic architecture:

@InProceedings{platzer_et_al:LIPIcs.ECRTS.2021.1,
  author =  {Kirschner, Maximilian and Dudzik, Konstantin and Krusekamp, Ben and Becker, Juergen},
  title =   {{MultiVic: A Time-Predictable RISC-V Multi-Core Processor Optimized for Neural Network Inference}},
  booktitle = {ITEM Workshop 2025 – IoT, Edge, and Mobile for Embedded Machine Learning},
  year =    {2025},
}

Compilation

To generate the binaries for a a specific benchmark, you can use the Justfile. For example to generate the matmul benchmark execute:

just compile test_matmul

Deployment

To execute the benchmark, you need to place the generated VMEM binaries in the appropriate scratchpad memories.

• vector_loader/build/ram.vmem -> Worker Cores D-SPM
• vector_loader/build/rom.vmem -> Worker Cores I-SPM
• build/ram.vmem -> Management Core D-SPM
• build/rom.vmem -> Management Core I-SPM

License

Unless otherwise noted, the code in this repository is licensed under the MIT License. See LICENSE for details. Some components come from third-party libraries and are licensed under their respective licenses. These are listed in the NOTICE file.