PIC24FJ64GA306 startup problem

[slavka012]

I have an application that works fine when I used compiler V5.075. Recently I have switched to V5.093, and the program no longer works. As far as I can tell it never reaches the main(). I mean MY main. In the beginning of the main() I see a compiler generated code that initializes global variables. On the V5.075 it is a linear piece of code that simple loads a value into W and then stores into the corresponding memory area. On the V.5.093 it is more complex. I see the data area is defined with values and addresses of the variables and there is a code that iterates through the data. That piece of code is getting stuck and the actual main() never executes.

[Ttelmah]

Have just built a minimum project, with you published 'main', using
5.075, and 5.093, and the initialisation is identical.

Now I suspect two things. First you are using restart_wdt, which suggests
you have the watchdog setup to enable in hardware, possibly with quite a
short period?.
Possibly you use #ZERO_RAM?
The only 'loop' initialisation, is if you have zero_ram enabled. Even with this
though the initialisation is identical:

[slavka012]

[Ttelmah]

I've just built one with a 256 element array. The initialisation is identical between 5.075, and 5.093. Same data statements and loop to load it.
It is something specific you are doing, that is making the initialisation
differ. You need to work out what it is.
Are you sure you are comparing with 5.075?. there were changes to the data
initialisation a while before this.

[slavka012]

[Ttelmah]

If you mean you are loading it with values using rand, then this is not done
during initialization. It is run time code. You can't load using a function
as part of the initialization.
The initialisation code you show just loads the data values from the data
statements into an array in RAM.

[slavka012]

[Ttelmah]

You are misunderstanding what I am saying.
I am not saying 'you are doing something stupid'. I am saying you are doing
something _different_ to what you describe.

The point is that if I compile exactly what you describe, there is no difference
at all between the initialisation code of the two compilers. So you have to be
doing something else beyond what you are describing to make the difference
occur. I set my test array up as a 256 element array, and the initialisation
is identical.
Now looking at your code, you are initialising I think perhaps seven arrays.
Post your real setup code.
Unless we can reproduce the behaviour, we can't know what is actually
happening.

[slavka012]

I did a bit more digging. Turned out it is WDT after all. On the 5.075 initialization code finishes up in 107mS (at 31KHz clock) and on 5.093 it takes over 120mS. WDT should nominally be 124mS, but somehow expires sooner.

[Ttelmah]

Some of the watchdog timers are _very_ inaccurate. On some chips they can
vary by over 50%!...