A kernel library targeting Neural Network Accelerators developed in the PULP group.
This library is considered unstable and might go through major changes until a stable release of v1.0.0.
The interface to each accelerator consists of these functions:
void <accelerator>_nnx_init(<accelerator>_dev_t *dev, <accelerator>_pulp_conf_t *conf);
void <accelerator>_nnx_term(<accelerator>_dev_t *dev);
int <accelerator>_nnx_dispatch(<accelerator>_dev_t *dev, <accelerator>_task_t *task);
int <accelerator>_nnx_dispatch_check(<accelerator>_dev_t *dev);
void <accelerator>_nnx_dispatch_wait(<accelerator>_dev_t *dev);
int <accelerator>_nnx_resolve_check(<accelerator>_dev_t *dev, <accelerator>_task_t *task);
void <accelerator>_nnx_resolve_wait(<accelerator>_dev_t *dev, <accelerator>_task_t *task);
Each accelerator has their own named function in case there exist multiple types of accelerators on a same board.
Each function accepts a pointer to a <accelerator>_dev_t
type which to discern between each accelerator.
Note: The accelerator can provide additional helper functions if needed.
- inc: nnx interface for each accelerator
- src: implementation for each accelerator
- util: utilities used by all the accelerators
- <accelerator>:
- hal: hardware abstraction layer
- gvsoc: gvsoc-specific functions
- bsp: board support package for each board that has the accelerator
- test: testing folder (more info)
You can find information about testing in the dedicated README.
The library was tested with following pairs of SDKs and compilers:
SDK | SDK Commit Hash | Compiler | Compiler Commit Hash |
---|---|---|---|
gap_sdk (obtainable from GreenWaves Technologies) | 90df4ce219 | gap_gnu_toolchain | 360fd4f9d6 |
pulp-sdk | c216298881 | pulp-riscv-gnu-toolchain | 9938bd8fcf (release v1.0.16) |
Bug reports and feature requests should be reported through issues.
All the development should be done through forks and merged onto the dev
branch with pull requests.
The library will follow the Semantic Versioning.
If you use PULP-NNX in your work, you can cite us:
@inproceedings{10.1145/3607889.3609092,
author = {Macan, Luka and Burrello, Alessio and Benini, Luca and Conti, Francesco},
title = {WIP: Automatic DNN Deployment on Heterogeneous Platforms: the GAP9 Case Study},
year = {2024},
isbn = {9798400702907},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3607889.3609092},
doi = {10.1145/3607889.3609092},
abstract = {Emerging Artificial-Intelligence-enabled System-on-Chips (AI-SoCs) combine a flexible microcontroller with parallel Digital Signal Processors (DSP) and heterogeneous acceleration capabilities. In this Work-in-Progress paper, we focus on the GAP9 RISC-V SoC as a case study to show how the open-source DORY Deep Neural Network (DNN) tool flow can be extended for heterogeneous acceleration by fine grained interleaving of a dedicated Neural Engine and a cluster of RISC-V cores. Our results show that up to 91\% of the peak accelerator throughput can be extracted in end-to-end execution of benchmarks based on MobileNet-V1 and V2.},
booktitle = {Proceedings of the International Conference on Compilers, Architecture, and Synthesis for Embedded Systems},
pages = {9–10},
numpages = {2},
keywords = {TinyML, MCUs, deep learning, HW accelerators},
location = {<conf-loc>, <city>Hamburg</city>, <country>Germany</country>, </conf-loc>},
series = {CASES '23 Companion}
}
- Luka Macan <luka.macan@unibo.it>
- Francesco Conti <fconti@unibo.it>
- Arpan Suravi Prasad <prasadar@iis.ee.ethz.ch>
Licensed under Apache-2.0; the whole text of the license can be found in the LICENSE file.