README.md 3.77 KB
Newer Older
Long Chung Chan's avatar
Long Chung Chan committed
1
2
# Partitioning FPGA-Optimized Systolic Arrays

LongChan's avatar
LongChan committed
3
We provide a fast optimization algorithm and a step-to-step guide on how to generate the dataset for a specific board and topologies to be used by our optimization tool.
Long Chung Chan's avatar
Long Chung Chan committed
4
5

>    *Long Chung Chan, Gurshaant Singh Malik and Nachiket Kapre*
Harry Chan's avatar
Harry Chan committed
6
>    [**"Partitioning FPGA-Optimized Systolic Arrays for Fun and Profit"**](https://git.uwaterloo.ca/watcag-public/fpga-syspart/blob/master/optimization_algo/paper/PID6211513.pdf),
Long Chung Chan's avatar
Long Chung Chan committed
7
8
>    2019 International Conference on Field-Programmable Technology 

Harry Chan's avatar
Harry Chan committed
9
## TODO
Harry Chan's avatar
Harry Chan committed
10
11
- [x] Add a public link to the paper
- [x] Basic Demo Flow
Long Chung Chan's avatar
Long Chung Chan committed
12
13
14
15
16
17
18
- [] Step-by-step guide on
  - [] their own data set
  - [] running the optimzation algo on the generated dataset
- [] What each files/directories are responsible for
- [] Changes done to ScaleSim
- [] Explanation of the testing result

Harry Chan's avatar
Harry Chan committed
19
## Demo
LongChan's avatar
LongChan committed
20
The following demos use pre-generated datasets and topologies that can be found in:
Long Chung Chan's avatar
Long Chung Chan committed
21

Harry Chan's avatar
Harry Chan committed
22
23
- [topologies](https://git.uwaterloo.ca/watcag-public/fpga-syspart/blob/master/optimization_algo/topologies/) contains all the topologies descriping their respective CNN structures
- [data_source](https://git.uwaterloo.ca/watcag-public/fpga-syspart/blob/master/optimization_algo/data_source/) contains all the cycle-accurate data generated using [SCALE sim](https://github.com/ARM-software/SCALE-Sim)
LongChan's avatar
LongChan committed
24
25
26
27
28
29
30
31
32
33
34

The instruction below will do a sweep run on each of the following networks:
   - FasterRCNN
   - Mobilenet
   - Yolo tiny
   - Googlenet
   - Alexnet
   - AlphaGoZero
   - NCF_rec
   - Resnet_50_v1

LongChan's avatar
LongChan committed
35
To obtain individual optimization result for a specific network and a specfic number of partition, please refer to the section below.
LongChan's avatar
LongChan committed
36
37

To get optimization result with 
LongChan's avatar
LongChan committed
38
1. Covariance Matrix Adaptation Evolution Strategy (CMA-es)
LongChan's avatar
LongChan committed
39
40
41
42
43
    ```bash
    # cd optimization_algo/scripts
    # ./sweep_nets_cma.sh
    ```

LongChan's avatar
LongChan committed
44
2. Genetic Algorithm (GA)
LongChan's avatar
LongChan committed
45
46
    ```bash
    # cd optimization_algo/scripts
LongChan's avatar
LongChan committed
47
    # ./sweep_nets_ga.sh
LongChan's avatar
LongChan committed
48
49
    ```

LongChan's avatar
LongChan committed
50
3. Hyperparameter Optimiztion
LongChan's avatar
LongChan committed
51
52
    ```bash
    # cd optimization_algo/scripts
LongChan's avatar
LongChan committed
53
    # ./sweep_nets_ho.sh
LongChan's avatar
LongChan committed
54
55
    ```

LongChan's avatar
LongChan committed
56
4. Brute Force
LongChan's avatar
LongChan committed
57
58
    ```bash
    # cd optimization_algo/scripts
LongChan's avatar
LongChan committed
59
    # ./sweep_nets_brute.sh
LongChan's avatar
LongChan committed
60
    ```
Long Chung Chan's avatar
Long Chung Chan committed
61

Harry Chan's avatar
Harry Chan committed
62
63
Result of the optimization will be added to the corresponding csv file under this [folder](https://git.uwaterloo.ca/watcag-public/fpga-syspart/blob/master/optimization_algo/resulting_csv)

Harry Chan's avatar
Harry Chan committed
64
## Step-by-step detail guide
LongChan's avatar
LongChan committed
65

Harry Chan's avatar
Harry Chan committed
66
### Custom topologies
Harry Chan's avatar
Harry Chan committed
67

Harry Chan's avatar
Harry Chan committed
68
### Custom target board
Harry Chan's avatar
Harry Chan committed
69

Harry Chan's avatar
Harry Chan committed
70
### Generate data source using SCALE-sim
Long Chung Chan's avatar
Long Chung Chan committed
71

Harry Chan's avatar
Harry Chan committed
72
### Running script targeting specific approach
Long Chung Chan's avatar
Long Chung Chan committed
73

Harry Chan's avatar
Harry Chan committed
74
## Repo Breakdown
Long Chung Chan's avatar
Long Chung Chan committed
75

Harry Chan's avatar
Harry Chan committed
76
## Scale-sim change
Long Chung Chan's avatar
Long Chung Chan committed
77

LongChan's avatar
LongChan committed
78
## License
Long Chung Chan's avatar
Long Chung Chan committed
79
This tool is distributed under MIT license.
LongChan's avatar
LongChan committed
80
Copyright (c) 2019 Long Chung Chan, Gurshaant Singh Malik, Nachiket Kapre
Long Chung Chan's avatar
Long Chung Chan committed
81
82
83
84
85
86
87
88
89
90
91
92
93
94

<div style="text-align: justify;"> 
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
<br><br>
</div>

<div style="text-align: justify;"> 
<b>The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.</b>
<br><br>
</div>

<div style="text-align: justify;"> 
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 </div>