Code

Speedup C encoder up to 100x#256

Code

Open

homm wants to merge 14 commits intowoltapp:masterwoltapp/blurhash:masterfrom homm:optimizationhomm/blurhash:optimizationCopy head branch name to clipboard

+158-60

Commits

Commits on Sep 25, 2024

• Define math consts

  homm

  committed
  9c138a9

  Copy full SHA for 9c138a9

• Fix number of arguments in blurhash_encoder

  homm

  committed
  3152f57

  Copy full SHA for 3152f57

• Show main warnings like unused variables

  homm

  committed
  b027b16

  Copy full SHA for b027b16

• Build object files separate for compilation speedup

  homm

  committed
  7fe900e

  Copy full SHA for 7fe900e

• Use sRGBToLinear_cache (4.5x speedup)

  homm

  committed
  6af77c9

  Copy full SHA for 6af77c9

• cosX cache (5.6x speedup)

  homm

  committed
  d936afb

  Copy full SHA for d936afb

• Prepare cosX && cosY once for all passes

  homm

  committed
  2e19ea7

  Copy full SHA for 2e19ea7

• Calculate factors in one call (up to 1.6x speedup)

  homm

  committed
  d3d26c1

  Copy full SHA for d3d26c1

Commits on Oct 11, 2024

• unroll multiplyBasisFunction loop (2.5x speedup)

  homm

  committed
  52d4a62

  Copy full SHA for 52d4a62

Commits on Oct 18, 2024

• Assign sRGBToLinear_cache after population to avoid races

  homm

  committed
  89a2524

  Copy full SHA for 89a2524

Commits on Oct 23, 2024

• decoder: cosf is about 17% faster

  homm

  committed
  3ab38fe

  Copy full SHA for 3ab38fe

• decoder: cache cos (3.2x faster)

  homm

  committed
  6f02d6e

  Copy full SHA for 6f02d6e

• decode: unroll inner loop (20% faster)

  homm

  committed
  0f2d8c9

  Copy full SHA for 0f2d8c9

• decode: Cache linearTosRGB (2.75x speedup)

  homm

  committed
  d2a09cd

  Copy full SHA for d2a09cd