shogo82148 / qrcode / 37125429002

34%

main: 34%

Pull #131

github

improve performance of the mask penalty score calculation

Point() is called for each of the 8 mask candidates when Mask is MaskAuto,
and it was the bottleneck of encoding.

- unpack the bitmap into a []uint8 grid once instead of calling BinaryAt for each pixel
- detect finder patterns with a sliding window bit mask
- count long runs and 2x2 blocks without branches

The results are exactly the same as the previous implementation.
The previous implementation is kept in the test to verify it.

EncodeToBitmap (Version 40, MaskAuto): 16.9ms -> 2.47ms

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

123 of 125 new or added lines in 1 file covered. (98.4%)

2901 of 8543 relevant lines covered (33.96%)

2.94 hits per line