Which is faster: floating point division or casting to 64-bit int?

[aberger]

Consider the simple calculation:

result = (A/B)*scaleFactor

where all quantities are unsigned, 32-bit integers.

A/B is strictly less than 1, but 

A*scaleFactor could be larger than 2^32.

Therefore, it seems I have two options:

result = ((float)(A)/B)*scaleFactor

or

result = ((uint64_t)(A)*scaleFactor)/B

Which sequence of operations is faster on the K64?

[aberger]

I was able to find this partial response by digging into the web page source:

My guess was that your first version with the single precision float operation will be faster, as the ARM Cortex-M4F can use the FPU for this, while your 64bit version requires a library runtime routine which very likely is slower. All depends to some extend to compiler optimizations and value of da...

[ErichStyger]

Not sure what could have been the original solution, but a while back I wrote about this:

https://mcuoneclipse.com/2019/03/29/be-aware-floating-point-operations-on-arm-cortex-m4f/

[ErichStyger]

I have no idea. I even cannot display the content of this thread, all what I get is

"This widget could not be displayed."

[myke_predko]

@aberger @BlackNight 

What happened to this thread?  I'd be interested in the comments.

[aberger]

Why did the solution to this discussion vanish?

[aberger]

Thanks Erich, this is great. At this time, I'm using the P&E debugger in MCUXpresso, so it unfortunately looks like I can't do the DWT cycle counting.

[BlackNight]

Yes, that's currently only supported with the LinkServer connection. For P&E you could directly read the DWT cycle counter in the memory view or store the values in your code, see Cycle Counting on ARM Cortex-M with DWT | MCU on Eclipse 

I hope this helps,

Erich

[aberger]

Hi @BlackNight - You previously provided the accepted solution to this post. For some reason NXP has removed that solution. Is there any chance you still have the your original post to this question? Would you mind re-sharing it? Thank you.