Ok. I have just looked at Murray's code.

1. He has already stated that you need to calibrate the function.

2. Yes. It would need re-calibration for a change of memory model

3. Yes. It may need re-calibration for a change of compiler version. It may need re-calibration for a change in optimisation setting.


You can solve the memory model by specifying the data type of the local variable. Then the basic loop will remain the same. The preamble will obviously have a different but insignificant execution time. e.g:

void delay_ms(volatile unsigned int xcount)
{
    data volatile unsigned int count = xcount;
    for(count *= DELAY_MULTPLR; count > 0; count--) continue;

    return;
}

As I said in an earlier post, other compilers and cpus permit inline assembly. So you can use a macro that will produce the exact number of machine cycles.

Personally I would seek a delay_us(us) macro and a delay_ms(int ms) function. And this is exactly what you do with sdcc. But given the limitations of Keil, you could write a single delay.asm with conditional assembly for each memory model.

But for the average Joe, he can just check the given calibration constant and go.

As an exercise for Mr Neil and Mr Malund, I would be interested in a compiler version or setting that creates a different timing loop:

?C0001:
	SETB 	C
	MOV  	A,count?041+01H
	SUBB 	A,#00H
	MOV  	A,count?041
	SUBB 	A,#00H
	JC   	?C0004
?C0003:
	MOV  	A,count?041+01H
	DEC  	count?041+01H
	JNZ  	?C0001
	DEC  	count?041
?C0005:
	SJMP 	?C0001

Yes. I am well aware that the compiler is quite free to create different code. I doubt that it will alter the generation unless there is a version change. In which case the user would re-calibrate.

David.