Question

Suppose I want to change the value of a variable when a flag is set. An obvious method is the following.

int a = 1, b = 2;
if(Flag)
    {
    a=b;
    Flag = false;
    }

This, however, is quite inefficient WRT the code generated to perform the operation. I have found that the following generates 40% less code to perform the operation.

a = ((!Flag)*a)+((Flag)*b);
Flag = false;

My question: I would rather not use the multiply “*” operator as multiplicative/division operators are slower to execute on my target. What else could I use to speed this up without increasing code space?

EDIT: The target device is an MSP430 running in the kHz range. code space and execution time are critical. the compiler is IAR C

Answer 1

The following achieves your desired logic without using multiply or any conditionals:

unsigned int mask = Flag-1;
a = (mask & a) | (~mask & b);

So when Flag==1 you get

a = (0x0 & a) | (0xffff & b);

and when Flag==0 you get

a = (0xffff & a) | (0x0 & b);

(You don’t really need a separate variable for mask either.. you can just do --Flag and use Flag instead of mask, I just thought it’d make the solution a little clearer.)