This can be so much complicated part of code for compiler:

tmp = 65536 - (10 * AxisAngle + 460) * 221184 / 120000; //that is, by removing the two zeros
TH = tmp / 256;
TL = tmp - TH * 256;

I have some experience with getting better result and eliminating anomally with divide formula to smaller part and also expliciting conversion like this:

// temporarily variables
float tmp_float = 0;
unsigned int tmp = 0;

// your following formula:
// tmp = 65536 - (10 * AxisAngle + 460) * 221184 / 120000
// can be rewrite by this:
tmp_float = (float)AxisAngle * (float)10;
tmp_float += (float)460;
// this is equal for tmp_float = tmp_float + (float)460;
tmp_float *= (float)3456;
// this is equal for tmp_float = tmp_float * (float)3456;
tmp_float /= (float)1875;
// this is equal for tmp_float = tmp_float / (float)1875;
// 3456/1875 is equal to 221184/120000 but smaller numbers
tmp_float = (float)65536 - tmp_float;
// finally you can use here some of round functions to get...
// ...better result or simplify conversion like this:
tmp = (unsigned int)tmp_float;

// after this it is better using bit rotation rutines instead...
// ...of dividing and substracting
TH = (unsigned char)(tmp >> 8);
TL = (unsigned char)tmp;
// or TL = (unsigned char)(tmp & 0xFF);

...

This can be helpfull in other cases not just in this one. Keep this on your mind and I think you will have less problem with compiler anomally :) Good luck.

Michal