Efficient C Programming: C Loop Structure (Part 1)

Introduction: Loop bodies are the focus of program design and optimization. This section will focus on the most effective way to handle for and while loops on ARM. (EEFOCUS)

14.5C Loop Structure

The loop body is the focus of program design and optimization. This section will focus on the most effective way to handle for and while loops on ARM.

14.5.1 Loop Termination

First, let's look at the following example, which shows two different loop exit conditions and the different assembly codes they generate.

The C source program is shown below.

intfact1(intn)

{

inti,fact=1;

for(i=1;i<=n;i++)

fact*=i;

return(fact);

}

intfact2(intn)

{

inti,fact=1;

for(i=n;i!=0;i--)

fact*=i;

return(fact);

}

The generated assembly code is shown below.

fact1

MOVa3,#1

MOVa2,#1

CMPa1,#1

BLT|L000020.J5.fact1|

|L000010.J4.fact1|

MULa3,a2,a3

ADDa2,a2,#1

CMPa2,a1

BLE|L000010.J4.fact1|

|L000020.J5.fact1|

MOVa1,a3

MOVpc,lr

fact2

MOVSa2,a1

MOVa1,#1

MOVEQpc,lr

|L000034.J4.fact2|

MULa1,a2,a1

SUBSa2,a2,#1

BNE|L000034.J4.fact2|

MOVpc,lr

From the generated assembly code, we can see that although the two functions implement the same function, the generated code efficiency is not exactly the same. The key here is that the loop termination condition should be countdown to zero, rather than countup to a certain value. Since the countdown result is stored in the conditional flag, the instruction to compare with zero can be omitted. At the same time, one less register can be used to store the loop termination value.

Notice

The above example uses the -O2–Otime compilation options. If the -Ospace option is used, the compilation results will be different.

For loop count value i, if i is unsigned, the loop continuation condition can be either i!=0 or i>0. Since i cannot be a negative number, these two conditions are equivalent. For a signed loop count value, it is best not to use the condition i>0 as the loop continuation condition. If i>0 is used as the loop continuation condition, the compiler will generate the following code.

SUBa2,a2,#1

CMPr1,#0

BGT|L000034.J4.fact2|

At this time, the compiler adds an extra CMP instruction, mainly to prevent the signed number i=−0x8000000. In short, no matter for signed or unsigned loop count values, i!=0 should be used as the loop end condition. For the signed number i, this is one less instruction than using i>0.