Function calling - which is better, and why? ...

OK ... which of these two is better, and why?

Example 1

Code:

---

void CHighResolutionTimer::SecondsElapsed(double *ElapsedTime)
{
        QueryPerformanceCounter((LARGE_INTEGER*) &m_i64_ElapsedTime);
        *ElapsedTime = double(m_i64_ElapsedTime - m_i64_StartTime) / double(m_i64_Frequency);
}

---

Example 2

Code:

---

double CHighResolutionTimer::SecondsElapsed(void)
{
        QueryPerformanceCounter((LARGE_INTEGER*) &m_i64_ElapsedTime);
        return double(m_i64_ElapsedTime - m_i64_StartTime) / double(m_i64_Frequency);
}

---

I could then call then like ...


CBlah.SecondsElapsed(&var);
or
var = CBlah.SecondsElapsed();


I'm interested in the performance rather than the readability.

Regards

In example 2 a temporary variable is created to return, while in example 1 there is no temporary variable created. Example 1 should therefore have a slight performance benefit.

But you don't have the syntactic benefit of being able to nest calls.

double movement = velocity * timer.getSecondsElapsed();

is impossible with the pass a pointer and fill it notation.

I'm pretty sure that 2 should run faster. 2 has to do an extra copy, but 1 has to access memory through a pointer, which can be relatively expensive.

I did a little test (I used sqrt() versus QueryPerformanceCounter()), and Example 2 was indeed just a tad bit faster.

Just thought about it a little more... #2 should definately be faster with optimizations since the single double could probably be handled at register level while pass through reference... well, that's memory no matter how you cut it.

I finally got a machine to acturately test the two returning methods:

VC++6 ... No optomisation, no inlining

Worst case - the two were identical
Best case - version two about 60% faster.

Thanks for all the input :)

Cheers all

The Second is faster
because:

the varriable is returned in the Proccessor register eax

while the first uses the stack memory with push and pop
therefore is slower...