Mail Archives: djgpp/1996/02/23/21:09:24
In article <Pine DOT SUN DOT 3 DOT 91 DOT 960220095423 DOT 24086A AT chris>,
Carrot <st2edea AT doc DOT ntu DOT ac DOT uk> wrote:
>
>Oh man. Somebody help me please....
>
>I'm new to DJGPP and have started using inline assembly within my code.
>At the moment, I'm trying to convert my old graphics routines to DJGPP v2 and
>am experiencing segment problems when accessing VGA memory at 0xA0000
>(mode 13h).
>
>Here's an example of the problem I'm getting...
>
[example code snipped]
I'd rewrite this to:
unsigned char *screen = __djgpp_conventional_base() + 0xa0000;
void Clear_VGA() {
__djgpp_nearptr_enable();
asm(
"movl _screen, %edi\n"
"xorl %eax, %eax\n" /* clear eax */
"movl $16000, %ecx\n"
"cld\n" /* GCC always uses this, you should too */
"rep\n"
"stosl\n"
: : eax,edi,ecx);
__djgpp_nearptr_disable();
}
Your problem is that GCC assumes you will take care of restoring clobbered
registers. It does this so it doesn't have to stop optimizing the code around
the asm statement. You weren't doing this, so you were just lucky it worked
as a one-off.
That last little list is the list of clobbered registers. Now GCC knows not
to count on them keeping their value after your asm, and will optimize
around it. It won't necessarily push them, it might decide not to use them
at all.
__djgpp_nearptr_enable lets your default es address DOS memory offset by
__djgpp_conventional_base.
Generally you actually want to do only one __djgpp_nearptr* pairing, but
it's good to leave memory protection off as little as possible while coding.
I don't guarantee this'll actually work verbatim, I'm not at my machine.
I recommend you take a look at
http://www.rt66.com/~brennan/djgpp/how_to_access_vga.html
By the way, on -O1 and up, memcpy with a constant length is compiled into
an inline rep movsl anyway.
--brennan
I think I'll write a "How to use the asm() statement tutorial for people
who hate fighting the Info program." Is there one already?
--
brennan AT rt66 DOT com | .sig this!
- Raw text -