Faster division of a float?

[gwdeveloper 275]

The FR5739 has the MPY32 hardware multiplier. I would like to use this to speed up the division of a floating point number, float/int. From what I assume, in place of X/Y, I can use X*(1/Y) and use the hardware multiplier. From http://www.ti.com/lit/an/slaa329/slaa329.pdf it would seem that I lose the remainder when doing the division. Is there a faster way to do floating point division with integers but retain accuracy?

 

Currently, I am using the DMA to speed up ADC conversions but then losing cycles during the division. I know I can use the DMA to directly enter the data into the MPY32 registers but then I'll still lose accuracy.

[SirZusa 34]

x / y

 

multiply x by 10000 and truncate it - so you get an integer out of the float with 4 decimals, then do a normal integer - integer - division ... and in the last step divide it by 10000 to get the result - i dont know if it is faster - but you can give it a try

 

something like this (untested)

float x = 1.234567;
int y = 7;
int z = x * 10000;
float result = z / (y * 10000);

[gwdeveloper 275]

I'll give it a shot. Gotta reset CCS after figuring out how to make it count CPU cycles. I never though to multiply it by 10000 to make it a whole number first. That may actually make it easier to use the hardware multiplier and still speed things up a bit.