Same line of code takes drastically different times to execute (Due)

Official Hardware Mega Due
1 
double answer = int2sinescale(input, minADC, maxADC) * int2sinescale(scaledsine[i], 0, 4095);

OK so when I wrap some timing code around it in the actual loop where it's being used, I get

28 microsecs (lockamp)
28 microsecs (lockamp)
27 microsecs (lockamp)
28 microsecs (lockamp)
28 microsecs (lockamp)
19 microsecs (lockamp)
28 microsecs (lockamp)

But when I just copy that line, put it into a shell command called "benchmark" (it looks like this):

    starttime = micros();
    double voltage = int2voltage(800, 0.55, 2.75);
    elapsedtime = micros() - starttime;
    Serial.print(elapsedtime); Serial.println(" microsecs (int2voltage)");
    starttime = micros();
    int sinescale = int2sinescale(2200, 688, 3424);
    elapsedtime = micros() - starttime;
    Serial.print(elapsedtime); Serial.println(" microsecs (int2sinescale)");
    starttime = micros();
    double answer = 0.55 * 2.7555324;
    elapsedtime = micros() - starttime;
    Serial.print(elapsedtime); Serial.println(" microsecs (multiply two doubles)");
    starttime = micros();
    voltage = int2sinescale(700, minADC, maxADC) * int2sinescale(scaledsine[3], 0, 4095);
    elapsedtime = micros() - starttime;
    Serial.print(elapsedtime); Serial.println(" microsecs (lockamp int2sinescale line)");

auteur > benchmark
1 microsecs (int2voltage)
1 microsecs (int2sinescale)
1 microsecs (multiply two doubles)
2 microsecs (lockamp int2sinescale line)

What the hell? None of these report such low execution times when I put timing code around them where they're actually being used in the program. What's actually going on?

2

You're timing the serial output?

3

randomshinichi:
OK so when I wrap some timing code around it in the actual loop where it's being used, I get

But when I just copy that line, put it into a shell command called "benchmark" (it looks like this):

What's actually going on?

How would anyone know? You haven't posted the code that gives the long times.

...R

4

As I said, it's just that same line of code.

It turned out that since I wasn't using the value, the compiler optimized it away. The asm("") trick that works on AVR Arduinos doesn't work, so I had to Serial.print the answers out. Thanks anyway!