Lpc2000

On Tuesday 13 December 2005 11:36, uedogan wrote:
> Hi everybody,
>
> i'm currently working with the LPC2136 and the latest KEIL C-compiler.
> Does anybody know some functions that provide a (much) faster 16-bit
> unsigned integer division than KEIL currently has? I have ~58MHz and
> the division currently needs about 2us.
> I would even prefer some 32-Bit unsigned integer divisions but these
> run totaly out of my available timing (KEIL needs ~3.9us for that).

OK, I have not tried this out, but I'm quite surprised  it takes soooooo long. 
Keil have a good reputation and I would have expected a better result than 
this.

I looked at the gcc code for divi3 and it is approx 256 bytes long, but not 
all of that is getting executed. I would expect this to run faster than the 
numbers you give.

Curious how are you doing the measurement?

You probably already know this.... ARM does not have a built-in divide 
instruction so it uses a funky multiply instead. Most of execution goes into 
calculating the multipliers.

Shifting is very cheap on ARM, so if you're doing a lot of divisions by a 
shiftable number (eg. power of 2), then use shifts instead. The compiler 
should be able to figure this out itself for constants (eg. x = y / 16; ==> x 
= y >> 4;).

Since remainder (x = y % z) also uses a division, you should also consider 
using and masking  instead (x = y % 8; ==> x = y & 7;)