search

unzvfu / cuda-fixnum

Extended-precision modular arithmetic library that targets CUDA.
MIT License
34 stars 8 forks source link

Overhaul benchmarking system #30

Open unzvfu opened 4 years ago

unzvfu commented 4 years ago

From https://github.com/data61/cuda-fixnum/issues/50:

Including, in no particular order

  • [ ] Generate data and graph it, rather than generating large difficult-to-interpret tables.
  • [ ] Generate data for all functions.
  • [ ] Allow generating data for specific functions.
  • [ ] Generate graphs of relative performance for related functions (e.g. How much faster is squaring to multiplication for various sizes?)
  • [ ] For each operation, work out the appropriate amount of data to use (currently we use a tiny amount for small fixnums and a largish amount for bigger fixnums -- this is inconsistent).
  • [ ] Understand what the optimal performance could be in important cases (e.g. mulmod, modexp).
  • [ ] Normalise the data gathered relative to some appropriately chosen device parameters (e.g. #SMs, #cores, core frequency, ...)
unzvfu commented 4 years ago

Consider using the Google Benchmark library.