Skip to content
Tool for generating efficient setting of the cuFFT library
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.gitignore
LICENSE
Makefile
README.md
advisor.cpp
advisor.h
benchmarkResult.cpp
benchmarkResult.h
benchmarker.cpp
benchmarker.h
cudaAsserts.cpp
cudaAsserts.h
cudaUtils.h
generalTransform.cpp
generalTransform.h
inputParser.cpp
inputParser.h
main.cpp
sizeOptimizer.cpp
sizeOptimizer.h
transform.cpp
transform.h
transformGenerator.cpp
transformGenerator.h
utils.h
validator.cpp
validator.h

README.md

cuFFTAdvisor

Tool for generating optimal setting of the cuFFT library.

This tool is able to:

  • benchmark some transformation.
  • run fast heuristics which will recommend faster settings, using either padding or cropping.
  • autotune the heuristics proposals and thus obtainign the best setting possible

Publication

pending

Build

Simply run make in the root directory. Make sure that you have NVCC and Cuda libraries in the PATH

Example

./build/cuFFTAdvisor -help

./build/cuFFTAdvisor -benchmark -device 0 -x 85 -y 1

./build/cuFFTAdvisor -recommend 30 -x 2273 -y 2273 -n 6 --outOfPlaceOnly --realOnly --floatOnly --batchOnly --forwardOnly --maxSignalInc 10

./build/cuFFTAdvisor -find 20 -x 445105 -n 5 --inPlaceOnly --complexOnly --doubleOnly --forwardOnly

Known limitations

Internal heuristics is based on Cuda v.8, so it might not be very accurate for other versions.

You can’t perform that action at this time.