6 Trigger Benchmark Results

nyanpasu64 edytuje tę stronę 2018-10-29 04:51:22 -07:00

Spis treści

• Edge Postprocessing
• Make stereo waves mmaped, only flatten when indexing

• OS: linux kubuntu
• lscpu: Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz
• CPU max MHz: 2800.0000
• file: sd3-winter

Edge Postprocessing

https://github.com/nyanpasu64/ovgenpy/issues/54

CPU MHz: 2700.170

end_time: 60 unless otherwise specified

• master=3e2577c3908fca07767bf3938e00f46de9def388
  • FPS = 361.39417374896743 (end_time: ~)
  • FPS = 346.00137877976107
  • FPS = 37x to 381.9118519249378
• Add PerFrameCache class to cache period over multiple triggers
  • FPS = 306.3840671116295 (I will only keep top data)
  • FPS = 373.1213967657881
• [performance] Make CorrelationTrigger assume cache is empty
  • FPS = 345.4890131431575
  • FPS = 378.8953182312417
  • CACHE? WHO NEEDS TO WRITE TO CACHE?
    • FPS = 351.3518958113948 (end_time: ???)
• edge-post-trigger
  • FPS = 380.2575750516541
• edge-post-trigger + LocalPostTrigger
  • FPS = 326.25395052206613 to 331.948028419742
• Revert "[performance] Make CorrelationTrigger assume cache is empty"
  • FPS = 375.58309454038647

benchmark results are statistically bullshit. the same code runs 300..375fps on a 60-second segment. I think we should only keep the peak scores.

Make stereo waves mmaped, only flatten when indexing

https://github.com/nyanpasu64/ovgenpy/pull/21

precalculate division:

• ???

before:

• 388fps trigger
• 216fps render
• 133MB

after:

• 219fps trigger
• 183fps render
• 55-59MB

np.sum / stereo_chan:

• 258fps trigger

[...,0] + [...,1] ,/ 2

• 305fps trigger

cython

• ???