Thread: Signed Long comparison in IF Statement

Originally Posted by iMalc

What happens if you replace:

Code:

else if(output < -255)

with this?:

Code:

else if ((unsigned long)output >= 0x80000000UL && (unsigned long)output < 0xFFFFFF01UL)

Cause if that works, I'd say you've run into a compiler bug.

Note for others: I'm not interested in hearing about undefined behaviour when casting a negative to unsigned here.
I'm interested in diagnosing a real-world problem that we can be pretty sure relates to a two's-complement machine.

I'll try that on the hardware to see what happens

Originally Posted by rcgldr

How are KP, KI, and KD defined? If these are unsigned, then the calculation for output will get "promoted" to unsigned (although with 2's complement math, I'm not sure why this would cause a problem).

If that's not the problem, then taking a guess here, on the compiler you are using, what is the default variable type for a decimal constant such as 255 or -255 as used in your if statements? Have you tried suffixing those constants with the letter L: 255L or -255L to see if that makes any difference?

Can you try something simple like output = -255; sprintf(string,"%ld\n\r",output); ?

I tried the 255L and -255L but that didn't work.
I also just printed out the signed long output = -255, and it prints out -255. So that looks fine

As far as compiler version, I think I have AVRGCC 3.4.1.95 that has GCC 4.6.2 for 8-bit AVR microcontrollers.

I am using Atmel Studio 6 for the IDE.

the max range that I need is from -20,000 to 20,000. So obviously I don't need signed long. Do you think switching to a smaller type like int16_t would be better and have less problems? I modified a code that I downloaded from the internet, but left the datatypes as it was since the code was suppose to be for a microcontroller board that was a PID controller for a motor.

Originally Posted by laserlight

Plain int is guaranteed to have at least the range [-32767, 32767], so that would cover the max range that you have in mind, though it wouldn't hurt to #include <limits.h> and check INT_MIN and INT_MAX just in case your compiler is non-standard in that respect.

yeah so I checked this

Code:

	int setValue = 0;
	int feedbackValue = 0;
	static int errorValue = 0;
	static int prevError = 0;
	static int integral = 0;
	int derivative = 0;
	int output = 0;

It works with changing the variables to "INT" from "SIGNED LONG", so I'm not quite sure what the difference is causing the problem or if it's a compiler thing.