I got around to incorporating your implementations into my benchmark so that I had a set of comparable figures.

For the most part there are no surprises, with your implementations coming in more or less similar to my implementations of the same algorithms. Just a few percent either way as you'd expect. There are however, two stand outs.

The first is your uuencode() version--well done for getting it to work which I never did--which proves to be some 20% faster than the next nearest pure perl solution.

The second is your C implementation of the basic mask and shift algorithm which comes out 7% to 8% faster than the other C routines. Looking at the basic mechanics I do not see any great differences that the compiler shouldn't be able to optimise away. The only difference I can see that might account for that 8% is that you aren't mortalising your return values.

I've noticed that sv_2mortal() seems more costly that you'd think it should be in the past and tried to sidestep it, but every time I have, it has led to memory leaks. When I add that to your code, the difference over the other C routines drops to 2%. Still an advantage though, so maybe your implementation is more compiler/optimiser friendly.

Many thanks for your responses, it is always instructive to see how other people code the same algorithm.

(Also, congrats that all your routines produced the same, correct output!)

               Rate upuu upuu0  buk jethro salva mlut  asu jmcn   uu C
+_.._2 Crob C_24to32  uic
upuu        25622/s   --   -4% -13%   -48%  -53% -54% -54% -56% -64% -
+87%  -87%     -87% -88%
upuu0       26597/s   4%    -- -10%   -46%  -51% -52% -53% -54% -62% -
+86%  -86%     -86% -87%
buk         29407/s  15%   11%   --   -40%  -46% -47% -48% -49% -58% -
+85%  -85%     -85% -86%
jethro      49099/s  92%   85%  67%     --  -10% -12% -13% -15% -30% -
+74%  -74%     -75% -76%
salva       54710/s 114%  106%  86%    11%    --  -2%  -3%  -6% -22% -
+72%  -72%     -72% -74%
mlut        55864/s 118%  110%  90%    14%    2%   --  -1%  -4% -20% -
+71%  -71%     -71% -73%
asu         56264/s 120%  112%  91%    15%    3%   1%   --  -3% -20% -
+71%  -71%     -71% -73%
jmcnamara   57962/s 126%  118%  97%    18%    6%   4%   3%   -- -17% -
+70%  -70%     -70% -72%
uu          70207/s 174%  164% 139%    43%   28%  26%  25%  21%   -- -
+64%  -64%     -64% -66%
C_24to32_2 192359/s 651%  623% 554%   292%  252% 244% 242% 232% 174%  
+ --   -0%      -1%  -7%
Croboticus 192543/s 651%  624% 555%   292%  252% 245% 242% 232% 174%  
+ 0%    --      -1%  -7%
C_24to32   194211/s 658%  630% 560%   296%  255% 248% 245% 235% 177%  
+ 1%    1%       --  -6%
uic        207580/s 710%  680% 606%   323%  279% 272% 269% 258% 196%  
+ 8%    8%       7%   --
[download]

(Gah! I wish I could defeat that stupid f*** wrap algorithm!)