MK22 Floating Conversions Takes Really Long Time

[ec123]

My processor is running at 40MHz bus.

I'm working with an array of uint32_t and converting it to floats to another array.   The maximum size is 128 elements in each array.

I verified my application takes 6ms+ to loop and narrowed down the problem to the code below.  This code takes 6ms to run:

for ( i = 0; i < currentBufferSize && i < nSamples; i++ )

{

FsamplesA[i] = (float)samplesA[i]*3.3/4096.0 - fZeroCurrent;

FsamplesB[i] = (float)samplesB[i]*3.3/4096.0 - fZeroCurrent;

FsamplesC[i] = (float)samplesC[i]*3.3/4096.0 - fZeroCurrent;

}

So there is an assignment after a cast, multiplication, division and a simple offset subtraction.

I calculated it takes 624 instruction cycles to do the calcs and populate just a single float buffer element!

Does this seem right for a Cortex-M4F?

Using MCUXpresso V10.2.0.

Thanks!

[ec123]

Thanks for the input Erich.  I tired using the f-suffix, but I'm getting about the same results as before.  Anyway, I've since changed the code so that I do the conversions after each ADC sample is ready and just populate the float buffer with that value -  no intermediate buffers/conversions.  With that, I have the main loop at about 400us (from 6ms).

[BlackNight]

Hi Ed,

see as well the discussion in Be aware: Floating Point Operations on ARM Cortex-M4F | MCU on Eclipse  on that subject. You can use a gcc compiler option to force single precision constants too.

I hope this helps,

Erich