PERFECT Suite

Home:

• https://hpc.pnnl.gov/PERFECT/
• https://github.com/pnnl/perfect

Abstract: Kernels and applications for evaluating performance-power-architecture tradeoffs in radar and image processing.

About: The PERFECT Suite is comprised of kernels and applications representing domains of interest in radar and image processing. The suite was formed through a survey of applications used by all teams as well as consultations with numerous application domain experts. Each kernel in its own right is computationally important and provides reasonable coverage of a domain of interest. Furthermore, each kernel is algorithmically interesting.

The Suite has been jointly developed by Pacific Northwest National Laboratory (PNNL) and Georgia Tech Research Institute (GTRI) with support from the DARPA PERFECT (Power Efficiency Revolution for Embedded Computing) program.

Citation:

Kevin Barker, Thomas Benson, Dan Campbell, David Ediger, Roberto Gioiosa, Adolfy Hoisie, Darren Kerbyson, Joseph Manzano, Andres Marquez, Leon Song, Nathan R. Tallent, and Antonino Tumeo. "PERFECT (Power Efficiency Revolution For Embedded Computing Technologies) Benchmark Suite Manual." Pacific Northwest National Laboratory and Georgia Tech Research Institute, December 2013. https://hpc.pnnl.gov/PERFECT/

Contacts: (firstname.lastname@pnnl.gov)

• Joseph Manzano
• Nathan R. Tallent (www), (www)

Contributors: (alphabetical)

• Kevin Barker (PNNL)
• Thomas Benson (GTRI)
• Dan Campbell (GTRI)
• David Ediger (GTRI)
• Nitin Gawande (PNNL)
• Roberto Gioiosa (PNNL)
• Adolfy Hoisie (PNNL)
• Darren Kerbyson (PNNL)
• Joseph Manzano (PNNL)
• Andres Marquez (PNNL)
• Leon Song (PNNL)
• Nathan Tallent (PNNL)
• Antonino Tumeo (PNNL)

Details

Manual:

• doc/PERFECT-Suite-manual.pdf (https://gitlab.pnnl.gov/perf-lab/perfect/perfect-suite/-/blob/master/doc/PERFECT-Suite-manual.pdf)
• See above for citation.

Contents: The PERFECT Suite contains the following kernels and applications. There are serial C and OpenMP versions for all; in select cases there are CUDA variants.

• suite/pa1 : "PERFECT Application 1" kernels and application.

  • Discrete Wavelet Transform
  • 2D Convolution
  • Histogram Equalization

• suite/stap : "Space-Time Adaptive Processing" kernels and application.

  • System Solver
  • Inner Product
  • Outer Product

• suite/sar : "Synthetic Aperture Radar" kernels.

  • Interpolation 1
  • Interpolation 2
  • Back Projection (Non-Fourier SAR)

• suite/wami : "Wide Area Motion Imaging" kernels and application.

  • Debayer
  • Image Registration
  • Change Detection

• suite/required : Required (common) kernels

  • Sort
  • FFT 1D
  • FFT 2D

Preparing large files

Note: For migration to GitHub, the sample input/output files over 100 MB have been split.

To recover them:

  for fnm_first in $(find . -name "*.split-aa") ; do
    fnm_base=${fnm_first%.split-aa}
    echo "*** Building ${fnm_base}"
    cat ${fnm_base}.split-* > ${fnm_base}
  done

To clean them:

find . \( -name "*.bin" -o -name "*.m" \) -size +100M  -exec rm '{}' \;

To generate them:

find . \( -name "*.bin" -o -name "*.m" \) -size +100M  -exec split -b 99MiB  '{}' '{}'.split- \;

Using

The make system is based on GNU make. Type "make help" to see a list of supported targets.

The make system defaults to GCC, but that can be changed by overriding CC and CFLAGS, e.g.,

# environment override
CC=clang make

or

# make override
make CC=clang

To build and test the suite's kernels, issue the command

make check

To manually select input sizes, override INPUT_SIZE with the values SMALL, MEDIUM, or LARGE, e.g.,

make INPUT_SIZE=SMALL check