GR712RC LEON3 Power Model Data

Citation Author(s):
Kris
Nikov
University of Bristol
Marcos
Martinez
Thales Alenia Space
Pedro
Vallejo
Thales Alenia Space
Abel
Balbis
Thales Alenia Space
Jose
Nunez-Yanez
University of Bristol
Kerstin
Eder
University of Bristol
Submitted by:
Kris Nikov
Last updated:
Tue, 05/17/2022 - 22:18
DOI:
10.21227/1y7r-am78
Research Article Link:
Robust and accurate fine-grain power models for embedded systems with no on-chip PMU
License:
Creative Commons Attribution
288 Views
Categories:
Power and Energy
0
0 ratings - Please login to submit your rating.
ACCESS DATASET CITE SHARE/EMBED

Abstract 

This dataset contains sensor and PMU data for the LEON3 CPU on the Gaisler GR712RC development platform running a subset of the BEEBS benchmark suite as well as four implementations of an industry-representative image vision algorithm. It is meant to be used by the Robust Energy and Power Predictor Selection(REPPS) methodology [https://github.com/TSL-UOB/TP-REPPS] in order to generate accurate fine-grain models.

Instructions: 

# Generating the models from the LEON3 sample data

The data for this paper is generated using a custom open-source methodology called **__REPPS__**. In order to replicate the results, first you must follow all the steps in [https://github.com/TSL-UOB/TP-REPPS](https://github.com/TSL-UOB/TP-REPPS) in order to install and configure all the scripts and supporting programs. Afterwards you can proceed with executing the following commands to generate the various models from the LEON3 data.

**DISCLAIMER - If you have any issues please don't hesitate to contact via [email](mailto:kris.nikov@bris.ac.uk).**

## Generate models trained on BEEBS and validated on the use_case_core application

### ASIC Only Model
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_use_case_finegrain.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_use_case_split.data -p 6 -e 4 -d 2 -o 2 -s 20210421_leon3_beebs_ucc_pwr_fngr_nocyc_nocth_asicdata_avgrelerr_nfolds_ools.data
```

### Bottom-Up Search
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_use_case_finegrain.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_use_case_split.data -p 6 -l 9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24 -m 1 -n 16 -c 1 -g -i 50 -d 2 -o 2 -s 20210425_leon3_beebs_ucc_pwr_fngr_allev_nocyc_nocth_botup_avgrelerr_nfolds_ools.data
```

### Top-Down Search
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_use_case_finegrain.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_use_case_split.data -p 6 -l 9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24 -m 2 -n 1 -c 1 -g -i 50 -d 2 -o 2 -s 20210425_leon3_beebs_ucc_pwr_fngr_allev_nocyc_nocth_topdown_avgrelerr_nfolds_ools
```

## Validate the previous models on BEEBS as well (no need to redo all the event selection, just use same events)

### ASIC Only Model
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_BEEBS_split.data -p 6 -e 4 -d 2 -o 2 -s 20210421_leon3_beebs_beebs_pwr_fngr_nocyc_nocth_asicdata_avgrelerr_nfolds_ools.data
```

### Bottom-Up Search
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_BEEBS_split.data -p 6 -e 24 -d 2 -o 2 -s 20210421_leon3_beebs_beebs_pwr_fngr_allev_nocyc_nocth_botup_avgrelerr_nfolds_ools.data
```

### Top-Down Search
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_BEEBS_split.data -p 6 -e 9,10,12,13,14,15,16,18,19,20,22,23 -d 2 -o 2 -s 20210421_leon3_beebs_beebs_pwr_fngr_allev_nocyc_nocth_topdown_avgrelerr_nfolds_ools.data
```

# Visualise the data

## Generate model per-sample breakdown files for the 1st run of the use_case_opt application

### ASIC Only Model
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_use_case_finegrain_1run.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_onlyusecaseopt_split.data -p 6 -e 4 -d 2 -o 6 -s /PATH/TO/ESL_paper_data/20210421_leon3_beebs_uco_pwr_fngr_nocyc_nocth_asicdata_avgrelerr_nfolds_ools_1r.data
```

### Bottom-Up Search
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_use_case_finegrain_1run.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_onlyusecaseopt_split.data -p 6 -e 24 -d 2 -o 6 -s /PATH/TO/ESL_paper_data/20210427_leon3_beebs_uco_pwr_fngr_allev_nocyc_nocth_botup_avgrelerr_nfolds_ools_1r.data
```

### Top-Down Search
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_use_case_finegrain_1run.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_onlyusecaseopt_split.data -p 6 -e 9,10,12,13,14,15,16,18,19,20,22,23 -d 2 -o 6 -s /PATH/TO/ESL_paper_data/20210427_leon3_beebs_uco_pwr_fngr_allev_nocyc_nocth_topdown_avgrelerr_nfolds_ools_1r.data
```

## Generate model per-sample breakdown files for the 1st run of the BEEBS benchmarks

### ASIC Only Model
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain_1run.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_BEEBS_split.data -p 6 -e 4 -d 2 -o 6 -s /PATH/TO/ESL_paper_data/20210423_leon3_beebs_beebs_pwr_fngr_nocyc_nocth_asicdata_avgrelerr_nfolds_ools_1r.data
```

### Bottom-Up Search
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain_1run.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_BEEBS_split.data -p 6 -e 24 -d 2 -o 6 -s /PATH/TO/ESL_paper_data/20210427_leon3_beebs_beebs_pwr_fngr_allev_nocyc_nocth_botup_avgrelerr_nfolds_ools_1r.data
```

### Top-Down Search
```
./octave_makemodel.sh -r /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain.data -t /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain_1run.data -b /PATH/TO/ESL_paper_data/split/LEON3_BEEBS_BEEBS_split.data -p 6 -e 9,10,12,13,14,15,16,18,19,20,22,23 -d 2 -o 6 -s /PATH/TO/ESL_paper_data/20210427_leon3_beebs_beebs_pwr_fngr_allev_nocyc_nocth_topdown_avgrelerr_nfolds_ools_1r.data
```

## Plot the model per-sample breakdwon data using `MODELDATA_plot.py`

### Plot the use_case_opt 1st run per-sample physical measurements and model errors
```
./MODELDATA_plot.py -p 1 -x "Samples[#]" -t 10 -y "Power[W]" -b /PATH/TO/ESL_paper_data/data/LEON3_use_case_opt_finegrain_1run.data -l "Sensor Data" -i /PATH/TO/ESL_paper_data/20210421_leon3_beebs_uco_pwr_fngr_nocyc_nocth_asicdata_avgrelerr_nfolds_ools_1r.data -a 'ASIC Data Only' -i /PATH/TO/ESL_paper_data/20210427_leon3_beebs_uco_pwr_fngr_allev_nocyc_nocth_botup_avgrelerr_nfolds_ools_1r.data -a "Bottom-Up Search" -i /PATH/TO/ESL_paper_data/20210427_leon3_beebs_uco_pwr_fngr_allev_nocyc_nocth_topdown_avgrelerr_nfolds_ools_1r.data -a "Top-Down Search"
```

### Plot the BEEBS 1st run per-sample physical measurements and model errors
```
./MODELDATA_plot.py -p 1 -x "Samples[#]" -t 10 -y "Power[W]" -b /PATH/TO/ESL_paper_data/data/LEON3_BEEBS_finegrain_1run_physicaldata.data -l "Sensor Data" -i /PATH/TO/ESL_paper_data/20210423_leon3_beebs_beebs_pwr_fngr_nocyc_nocth_asicdata_avgrelerr_nfolds_ools_1r.data -a 'ASIC Data Only' -i /PATH/TO/ESL_paper_data/20210427_leon3_beebs_beebs_pwr_fngr_allev_nocyc_nocth_botup_avgrelerr_nfolds_ools_1r.data -a "Bottom-Up Search" -i /PATH/TO/ESL_paper_data/20210427_leon3_beebs_beebs_pwr_fngr_allev_nocyc_nocth_topdown_avgrelerr_nfolds_ools_1r.data -a "Top-Down Search"
```

Dataset Files

LOGIN TO ACCESS DATASET FILES

 

QUESTIONS?