Keith,
Thanks for confirming what I suspected. I tried your technique on my code
and was able to cut the stack allocation from (link a6,#-584) down to (link
a6,#-224).
The value of 584 would lead to a Palm OS 2 crash as expressions were
executed sequentially in the code... 224 works just fine. There were 6
floor() calls, 30 sin() calls, and 8 or so cos() calls, with a few sqrt()
and atan() calls. The original old function executed the 6 floor
expressions then caused a stack overflow on the first expression using the
sin().
Perhaps this is a problem that can be corrected in a future release of the
compiler for CodeWarrior?
Another work around that I was using while waiting for a potential answer
was to break up the main function that required the larger stack allocation
into multiple smaller functions.
Thanks again for your help!
Ed
----- Original Message -----
From: <[EMAIL PROTECTED]>
To: "Palm Developer Forum" <[EMAIL PROTECTED]>
Sent: Saturday, September 09, 2000 7:38 PM
Subject: Re: CodeWarrior Compiler Glitch with temporary stack space?
>
>
> Ed,
>
> FWIW, I think you're right. I think that CW *is* using up too much stack
space
> for temporaries. It could easily use one temporary for all of the results
from
> floor(). Instead, it's creating a new temporary for each result.
>
> I played around with a lot of different ways of jiggering the compiler to
be
> more space efficient. I tried putting braces around all of the
expressions, I
> tried inline functions, I tried all the different optimization levels I
could
> think of, etc. None of those worked, but I finally found something that
did --
> create your own temporary and use that. Here's my result. It uses 24 bytes
of
> stack space instead of 64. I think it could get away with 16, but I'm
happy with
> what we've got.
>
> #pragma stack_cleanup on
>
> #define PI2 (3.14159265 * 2)
>
> typedef struct MOONPOSDATA
> {
> double l;
> double m;
> double f;
> double d;
> double n;
> double g;
> } MOONPOSDATA;
>
> extern double floor (double);
>
> void MoonPosTest(double t, double *a5, double *d5, double *r5)
> {
> MOONPOSDATA* mp;
> double temp;
>
> mp = MemPtrNew(sizeof(MOONPOSDATA));
>
> mp->l = (mp->l - (temp = floor(mp->l))) * PI2;
> mp->m = (mp->m - (temp = floor(mp->m))) * PI2;
> mp->f = (mp->f - (temp = floor(mp->f))) * PI2;
> mp->d = (mp->d - (temp = floor(mp->d))) * PI2;
> mp->n = (mp->n - (temp = floor(mp->n))) * PI2;
> mp->g = (mp->g - (temp = floor(mp->g))) * PI2;
>
> MemPtrFree(mp);
> }
>
> -- Keith Rollin
> -- Palm OS Emulator engineer
>
>
>
>
>
>
> "M. Edward Wilborne III" <[EMAIL PROTECTED]> on 09/09/2000 09:00:03 AM
>
> Please respond to "Palm Developer Forum" <[EMAIL PROTECTED]>
>
> Sent by: "M. Edward Wilborne III" <[EMAIL PROTECTED]>
>
>
> To: "Palm Developer Forum" <[EMAIL PROTECTED]>
> cc: (Keith Rollin/US/PALM)
> Subject: Re: CodeWarrior Compiler Glitch with temporary stack space?
>
>
>
> Additional information...
>
> With:
>
> #pragma stack_cleanup on
>
> in the code, there appears to be 1 additional assemply instruction added
to
> each function:
>
> 00000xxx: 4FEF 0014 lea 20(a7),a7
>
> after the sysTrapFlpEmDispatch instruction.
>
> However, the use of stack relative to A6 appears to be the same as it was
> with the stack clean up set off. Which still leads me to believe that the
> expressions are using more stack space than necessary.
>
> Ed
>
> ----- Original Message -----
> From: "M. Edward Wilborne III" <[EMAIL PROTECTED]>
> To: "Palm Developer Forum" <[EMAIL PROTECTED]>
> Sent: Saturday, September 09, 2000 11:26 AM
> Subject: CodeWarrior Compiler Glitch with temporary stack space?
>
>
> > I apologize for the long nature of this message (with the assembly code
in
> > it, well it gets pretty large).
> >
> > This is just a snippet of working code, please disregard the fact that
it
> > returns no useful result. It is intended as an example only.
> >
> > I was compiling a much larger version of this function ("useful results
> > version") and kept having problems with stack overflows on Palm OS 3.0,
> and
> > 2.0, but the function would work fine on Palm OS 3.5.
> >
> > After taking a look at the disassembled version of the code, and trying
to
> > make heads of tails of what is going on, the conclusion that I have come
> up
> > with is that CodeWarrior is setting aside space on the stack for
temporary
> > or intermediate results, but doesn't appear to resuse that space once an
> > expression is done with the intermediate result. If you pile a bunch of
> > expressions into the function, then codewarrior allocates a larger
amount
> of
> > stack space, and eventually you can overflow the stack.
>
>
>
> --
> For information on using the Palm Developer Forums, or to unsubscribe,
please
> see http://www.palmos.com/dev/tech/support/forums/
>
>
>
>
>
> --
> For information on using the Palm Developer Forums, or to unsubscribe,
please see http://www.palmos.com/dev/tech/support/forums/
>
--
For information on using the Palm Developer Forums, or to unsubscribe, please see
http://www.palmos.com/dev/tech/support/forums/
