[solved]PIC18f series Hardware multiply

edmundopt · Joined: 08 Dec 2011 Posts: 15 Location: Portugal

First of all, personaly I do not use CCS, but I'm translating my Microchip C18 to CCS v4.124, my question is, when the IC is included, example:

#include <18F4620.h>

the following code :

unsigned int8 num_high, num_low;
unsigned int16 result;

result = num_high * 256;
result = result + num_low;

or the following code:

result = make16(num_high,num_low);

should it not the assembly code use the ASM instruction MULWF??
is there a way to change the resulting assembly code?

I'm refering pic18f series hardware multiply very usefull capability
sorry for any mistakes.

Battery David · Joined: 01 Feb 2010 Posts: 25

I haven't tried it but the manual lists a _mul() function to use the hardware multiplier.

Please let us know how it works.

David

edmundopt · Joined: 08 Dec 2011 Posts: 15 Location: Portugal

Thank you, I found out that :

unsigned int8 num_high, num_low;
unsigned int8 factor = 256;
unsigned int16 result;

result = num_high * factor;
result = result + num_low;

uses the MULWF..

strange.....

ckielstra · Joined: 18 Mar 2004 Posts: 3680 Location: The Netherlands

edmundopt · Joined: 08 Dec 2011 Posts: 15 Location: Portugal

you are right, really bad example about multiplication, because 256 is out of range for int8, anyway thanks because that was the problem in my code, out of range values!

and yes, something like :

result = result + high_int8;
result = (result<<8);
result = result + low_int8;

thanks

RF_Developer · Joined: 07 Feb 2011 Posts: 839

CCS will generally often optimise multiplication by constant powers of two into shifts. So

b = a * 2;
c = a * 8;
d = a * 64;

will be implemented as:

b = a << 1;
c = a << 3;
d = a << 5;

this is generally true even with 8 bit values , which can be multiplied in a single instruction by hardware and for which the speed and code size gains are relatively small, and is certainly true for 16 and more bit variables which gain substantially by this optimisation.

Whatever the compiler does however, always remember the calculation must be done in a form that is big enough to contain the expected result, and that C selects its arithmetic based on the input variables, NOT the result. If you want a 16 bit result, you must cast if needed to ensure the arithmetic is done as 16 bit.

RF Developer