From: terje.mathisen@nospicedham.tmsw.no   
      
   Robert Prins wrote:   
   > My strings currently have a length of 47 bytes or less, and as you   
   > know, I'm coding in Pascal, so not zero-terminated, but with a   
   > one-byte length preceding the (short)string. I'm using code I copied   
   > from one of the above URL's, pretty much unaltered,   
   >   
   > xor   eax, eax   
   >   
   > mov   esi, utf8 movzx edx, byte ptr [esi] test  edx, edx jmp   @04   
   >   
   > @01: mov   cl, [edx + esi]   
      
   > shl   cl, 1   
      
   > js    @02   
      
   > jc    @03   
   >   
   > @02: inc   eax   
   >   
      
   That's sneaky, you are skipping the increment only when the top bit is   
   set and not the second, the only problem is the number of branches   
      
   > @03: dec   edx   
   >   
   > @04: jnz   @01   
   >   
   > It's probably fast enough, given that the program takes 0.28 seconds   
   > to do what it needs to do, slurp in an 11,430 line 1.2 Mb CSV file,   
   > and spit out 97,501 lines in seven file with results, totalling   
   > around 7.3Mb.   
      
   The rest of your code is going to dominate, that's pretty clear!   
      
   What about code like this, since you already have the count in EDX when   
   you start:   
      
      mov eax,edx   
      
   ...   
      
      
      mov cl,[esi+edx]   
      sub cl,80h   
      cmp cl,40h   
      sbb eax,0   
      
   I.e. only if the input char is in the 80-bf range will you get a carry   
   from the CMP and that carry is subtracted from the original count.   
      
   No branches to predict and the same number of instructions.   
      
   Terje   
      
   --   
   -    
   "almost all programming can be viewed as an exercise in caching"   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)