Dan,

Thanks for the document.

I also tried the paper&pencil method on the your routine above - it apparently works OK, but I am still puzzled why that number of moves between the three memory locations (that confuses me a lot I admit) - I guess this is the result of optimisation for PICs, is it?

I could not resist, and (before I took the fist glimpse into that document; and also without an attempt to "tune" it specifically to SDCC or Keil, I swerve :-) ) I wrote down the following:

void CRC1(unsigned char b) {
  b ^= CRCLo;
  CRCLo = CRCHi;
  CRCHi = b ^ (b << 4);
  CRCLo ^= (CRCHi >> 4) ^ (CRCHi << 3);
  CRCHi ^= (CRCHi >> 5);
}

This gave me 27 cycles on SDCC and 26 on Keil 7.5 eval I have here.

Again, I admit this might be less optimal for PICs due to the different architecture.

JW