Speeding up Floyd-Steinberg dithering: an optimization exercise

Hi I shaved off 40% more just by replacing multiplication with bitwise operations and ported the code to a standard Python C Extension to avoid relying on JIT.<p><a href="https:&#x2F;&#x2F;github.com&#x2F;szabolcsdombi&#x2F;optimized-floyd-steinberg-dithering">https:&#x2F;&#x2F;github.com&#x2F;szabolcsdombi&#x2F;optimized-floyd-steinberg-d...</a><p>X * 7 is equal to (X &lt;&lt; 3) - X<p>X * 3 is equal to (X &lt;&lt; 1) + X<p>X * 5 is equal to (X &lt;&lt; 2) + X