Code

Performance improvements#62

Code

Open

Agade09 wants to merge 8 commits intora1nty:mainra1nty/DXcam:mainfrom Agade09:mainAgade09/DXcam:mainCopy head branch name to clipboard

+22-14

Commits

Commits on Apr 15, 2023

• Use numpy directly for BGRA->RGB and BGRA->BGR conversion.

  Show description for a70ab3a
  Agade09

  committed
  a70ab3a

• Reduce overhead in ctypes.string_at() when grabbing from a region smaller than the whole screen.

  Show description for b7ac4c2
  Agade09

  committed
  b7ac4c2

Commits on Apr 16, 2023

• Fixed regression where numpy_processor was no longer correct for rotations of (90,180,270)

  Agade09

  committed
  2833caf

• Fixed not applying performance optimization if a region is defined where its width region matches the screen's width.

  Agade09

  committed
  94bec3f

Commits on Apr 17, 2023

• Almost completely removed overhead of ctypes.string_at by using the from_address API instead. In profiling a 1440x2560 grab, total time spent went from 20% in string_at() to almost 0% in from_addre…

  Show description for c9859c2
  Agade09

  committed
  c9859c2

• Don't call _validate_region in grab if the region is None because in that case self.region is used, and self.region was already validated when it was defined.

  Show description for 87a8117
  Agade09

  committed
  87a8117

• Translate 'BGRA' color_mode to None in NumpyProcessor so the existing if statement bypassed the call to self.process_cvtcolor(). Simplify code in process_cvtcolor since it no longer needs to handle…

  Show description for bf02b69
  Agade09

  committed
  bf02b69

• Revert using numpy for conversion of BGRA->BGR/RGB. Numpy was faster but it was producing a non-contiguous array, which changes the behavior of the library

  Agade09

  committed
  204a558