"The original post used only integers, and a lot of brackets: that meant the integer result in the denominator was zero - and, hence, the divide-by-zero error."

The original post was:

#define Xtal 22.1184
#define tic12 (12/Xtal)
#define _10msec (65536 - ((1000 * 10) / tic12))

.
.
TL1 = _10msec;
TH1 = _10msec >> 8 ;

which does have a floating point constant in it.

The expansion is:

TL1 = (65536 - ((1000 * 10) / (12/22.1184)));
TH1 = (65536 - ((1000 * 10) / (12/22.1184))) >> 8 ;

which, contrary to what the OP reports, correctly generates a value of zero for TL1. The real error is in the second line where an attempt is being made to right shift a floating point value.