Commits on May 30, 2016

1. True colour to xterm256 conversion: significant speed up …

   Speed:
   In a hot loop, where the old code has its ~2K pre-calculated data
   already cached, this new code is x40-x50 faster (gcc/clang) with
   arbitrary RGB values, and ~x3 faster with RGB values which have an
   exact xterm256 colour representation (e.g. #5f87ff).
   
   When used sparsely, or otherwise when the old code doesn't have the
   data cached, the new code is likely to be at least another order of
   magnitude faster, as it does not use pre-calculated data (technically,
   it uses less than 20 constants).
   
   Further 50% speed-up is possible with 256 bytes of pre-calculated date,
   or 100% (x2 speed up) with 1K of pre-calculated data.
   
   Compatibility:
   This code returns identical values as the old code for the same inputs.
   
   However, the old code is implicitly rounding down from the middle. For
   instance, for RGB of (13, 13, 13) which is exactly between the
   potential return values 232 (8, 8, 8) and 233 (18, 18, 18), it returns
   232. It does the same with colour values at 16..231 .
   
   While there's no right or wrong in this case between rounding up/down,
   the new code also supports the more conventional paradigm of rounding
   up from the middle. To enable it, set `LC` (legacy compatibility) to 0.

   

   avih committed May 30, 2016
   Copy the full SHA

   05e82d6 View commit details

   Browse the repository at this point in the history