Code

Optimize bokeh blur convolution#1475

Code

Merged

JimBobSquarePants merged 22 commits intomasterSixLabors/ImageSharp:masterfrom sp/bokeh-blur-speedupSixLabors/ImageSharp:sp/bokeh-blur-speedupCopy head branch name to clipboardDec 15, 2020

+465-177

Commits

Commits on Dec 12, 2020

• Add BokehBlur benchmark

  Sergio0694

  committed
  efeb1c4

• Minor code refactoring to improve flexibility

  Sergio0694

  committed
  d8453bc

• Switched bokeh blur to optimized pipeline

  Sergio0694

  committed
  faa1ad1

• Specialize bokeh blur operations for 1D kernels

  Sergio0694

  committed
  6bb300a

• Minor code tweaks

  Sergio0694

  committed
  c72a3bb

• Restore temporary changes

  Sergio0694

  committed
  04d66a9

• Remove unnecessary code

  Sergio0694

  committed
  b2f397d

• Fix gamma processing out of image bounds

  Sergio0694

  committed
  3180ec4

• Fix blur processing when constrained to region

  Sergio0694

  committed
  7db1225

• Fix NullReferenceException in KernelSamplingMap.Dispose

  Sergio0694

  committed
  50f9716

Commits on Dec 13, 2020

• Remove allocation constrained test for bokeh blur

  Sergio0694

  committed
  9fca028

• Remove unnecessary offset indirections

  Sergio0694

  committed
  2e53a44

Commits on Dec 14, 2020

• Add optimized paths for default gamma exposure

  Sergio0694

  committed
  e4ba017

• Switch to vectorized clamping

  Sergio0694

  committed
  cb18c58

• Initial vectorized cube root implementation

  Sergio0694

  committed
  5c89d0a

• Fix vectorized cube root on x86-64 with no SSE41

  Sergio0694

  committed
  442e467

• Minor codegen tweaks

  Sergio0694

  committed
  c38fc81

• Add discontigous buffers and intrinsics tests

  JimBobSquarePants

  committed
  0d27e04

• Fix feature test runner

  JimBobSquarePants

  committed
  8ab5e6f

• Switch to explicit SSE Newton approximations

  Sergio0694

  committed
  c52112c

• Add FMA support, more SSE optimizations

  Show description for 9142e72
  Sergio0694

  committed
  9142e72

• Add more codegen improvements

  Sergio0694

  committed
  a8cae3f