Skip to content

Latest commit

 

History

History
146 lines (121 loc) · 7.02 KB

README.md

File metadata and controls

146 lines (121 loc) · 7.02 KB
Raw

About GenMap

GenMap is an application mapping framework for spatially mapping CGRAs implemented with Python. It uses a multi-objective genetic algorithm. Therefore, it is easy to add your own objectives to be optimized. It also contains a leakage power optimization method, a dynamic power estimation model, and RoMultiC configuration generation.

Publications

  1. Takuya Kojima, Hayate Okuhara, Masaaki Kondo, Hideharu Amano, "Body Bias Control on a CGRA based on Convex Optimization", COOLCHIPS25, Japan, April, 2022. [IEEE Xplore]
  2. Takuya Kojima, Nguyen Anh Vu Doan, Hideharu Amano, "GenMap: A Genetic Algorithmic Approach for Optimizing Spatial Mapping of Coarse Grained Reconfigurable Architectures", IEEE Transactions on Very Large Scale Integration Systems (VLSI), Vol. 28, no. 11, pp.2383-2396, Nov 2020. [IEEE Xplore]
  3. Takeharu Ikezoe, Takuya Kojima and Hideharu Amano, "A Coarse-Grained Reconfigurable Architecture with a Fault Tolerant Non-Volatile Configurable Memory," 2019 International Conference on Field-Programmable Technology (ICFPT), Tianjin, China, 2019, pp. 81-89. [IEEE Xplore]
  4. Takuya Kojima and Hideharu Amano, "A Configuration Data Multicasting Method for Coarse-Grained Reconfigurable Architectures", 28th International Conference on Field Programmable Logic and Applications (FPL), Dublin, Ireland, August, 2018. [IEEE Xplore]
  5. Takuya Kojima, Naoki Ando, Hayate Okuhara, Hideharu Amano, "Glitch-aware Variable Pipeline Optimization for CGRAs", ReConFig 2017, Mexico, December 2017. [IEEE Xplore]
  6. Takuya Kojima, Naoki Ando, Hayate Okuhara, Ng. Anh Vu Doan, Hideharu Amano, "Body Bias Optimization for Variable Pipelined CGRA", 27th International Conference on Field-Programmable Logic and Applications(FPL), Belgium, September 2017. [IEEE Xplore]

Installation

Please see Installation guide

Included architecture definitions

GenMap supports the following CGRA architectures by default:

  1. CMA-SOTB2 [K.Masuyama, et al] (a.k.a. NVCMA [T.Ikezoe, et al] )
  2. CC-SOTB [Y.Matsushita, et al]
  3. CC-SOTB2 (VPCMA) [N.Ando, et al]
  4. VPCMA2 [T.Kojima, et al]
  5. RHP-CGRA [A.Podobas, et al]

Architecture information and parameters for some simulations are in the chip_files

Quick Tutorial

This tutorial introduces you to GenMap overview.

Run GenMap

First, make a working directory. In this tutorial, the working directory is in the same directory as the GenMap.

(GenMap_env) # mkdir work
(GenMap_env) # cd work

Please copy a sample application DFG file to a working directory (gray in this example).

(GenMap_env) # cp ../GenMap/app_samples/gray.dot ./

Please copy the architecture definition file and simulation parameter file to the working directory (CCSOTB2 in this example).

(GenMap_env) # cp ../GenMap/chip_files/CCSOTB2/arch.xml ./
(GenMap_env) # cp ../GenMap/chip_files/CCSOTB2/simdata.xml ./

Please copy the optimization setting to the working directory.

(GenMap_env) # cp ../GenMap/OptimizationParameters.xml ./

To run GenMap, please execute:

(GenMap_env) # python3 ../GenMap/GenMap.py gray.dot 10 [--nproc num]

At least, you have to specify a path to an application DFG file as a positional argument. For other optional arguments, please see help (python GenMap.py -h) You can specify the process count for multiprocessing (default 1).

After GenMap starts, the optimization status will appear like below.

Generation 121:  40%|█▌  | 120/300 [1:20:02<2:07:56, 42.65s/it, hof_len=97, stall=6]
Wire_Length: , max=29, min=27                                                       
Mapping_Width: , max=3, min=2                                                       
Op_Mapping_Width: , max=3, min=2                                                    
Power_Consumption: , max=0.589, min=0.347                                           
Time_Slack: , max=61.2, min=0.462

Generate Configuration

If the above optimization finishes successfully, it will save a result gray.dump (in default). By loading this result, you can generate configuration data for the architecture.

Please execute:

(GenMap_env) # python3 ../GenMap/CCSOTB2_ConfGen.py gray.dump

Then a shell will start.

=== GenMap solution selection utility ===
GenMap shell>

To show all solutions, execute:

GenMap shell> show
+-----+-------------+---------------+------------------+--------------------+---------------------+
|  ID | Wire_Length | Mapping_Width | Op_Mapping_Width | Power_Consumption  |      Time_Slack     |
+-----+-------------+---------------+------------------+--------------------+---------------------+
|  0  |     26.0    |      3.0      |       3.0        | 2.375386336518864  | 0.25644671033332855 |
|  1  |     26.0    |      3.0      |       3.0        | 2.4910062914105766 |  2.4540015123333276 |
...
...
| 157 |     37.0    |      7.0      |       7.0        | 1.029479529863405  |  1.0105863153333274 |
| 158 |     37.0    |      7.0      |       7.0        | 1.0658122223530275 |  1.5835502433333275 |
+-----+-------------+---------------+------------------+--------------------+---------------------+

To select a solution, execute:

GenMap shell> select 0 # to specify the ID number

To save the configuration, execute:

GenMap shell>  save

In the GenMap Shell, the following commands are available.

  1. show: show the result
  2. sort: sort the solutions by specified objective
  3. filter: filter the solutions with some conditions
  4. select: select a solution
  5. reset: reset the filtering and selection
  6. view: view mapping of the selected solution
  7. save: save configurations of the selected solution
  8. quit: quit the shell

You can see the usage of these commands in GenMap Shell with --help option as follows.

GenMap shell> save --help
usage: save [options...]
It generates configuration files of selected solutions

optional arguments:
  -h, --help            show this help message and exit
  -o OUTPUT_DIR, --output_dir OUTPUT_DIR
                        specify output directory name
  -f, --force           overwrite without prompt
  -p PREFIX, --prefix PREFIX
                        specify prefix of output file names (default:
                        app_name)
  -s STYLE [STYLE ...], --style STYLE [STYLE ...]
                        Pass the space separated arguments to configuration
                        generator

Detailed documentation

Please refer to [docs]