Re: [Sdcc-user] movx gets "optimized" out of loop!

[Bo-Yi L. <bl...@ik...>]

Ok,

Thanks to all for the replies.  I was using the correct forward declaration.
Sandeep thinks this needs fixing.  Maybe he forgot.  I'll wait for the next
release because I cannot get CVS to run on my workstation.

Bo

----- Original Message -----
From: "Johan Knol" <joh...@id...>
To: <sdc...@li...>
Sent: Saturday, March 17, 2001 8:10 AM
Subject: Re: [Sdcc-user] movx gets "optimized" out of loop!


> > Hi,
>  >
>  > I'm trying to write a polling routine that polls for a low status bit
at
> a
>  > specific device register address.  Here's a code snippet:
>  > -------------------------------------------------------------------
>  > #typedef unsigned char BYTE
>  > void PollLo(volatile xdata BYTE * addr, BYTE bitmask) {
>  > BYTE in;
>  >
>  > do {
>  > in = *addr;
>  > } while (in & bitmask);
>  > }
>  .
>  .
>  > -------------------------------------------------------------------
>  > The declaration says *addr is a pointer to a volatile BYTE.  But if I
do
>  > this, the movx (in=*addr) gets placed outside the do loop.  The only
way
> I
>  > can get the movx inside the do loop is to use
>  > (xdata BYTE * volatile addr) instead.  But in effect I'm saying the
>  pointer
>  > itself is volatile, which is not what really I want because I do not
need
>  > the pointer to be dereferenced inside the loop.  Does anybody know how
to
>  > get the compiler to do this right?
>
>  It was a bug in the compiler that is now fixed.
>
>  > Any why does it keep saving dptr to
>  > r2/r3 and back again?  dptr is already set by the caller routine so the
>  > exercise seems pointless.
>
>  Because addr is assigned to registers r2/r3. Dp registers are used as
>  temporaries. But you are right, this should be optimized.
>
>  Johan
>
>
>
> _______________________________________________
> Sdcc-user mailing list
> Sdc...@li...
> http://lists.sourceforge.net/lists/listinfo/sdcc-user
>