Xref: news2.mv.net comp.os.msdos.djgpp:4467
From: Shawn Hargreaves <slh100 AT york DOT ac DOT uk>
Newsgroups: comp.os.msdos.djgpp
Subject: Re: Need help with ASM sprite routine
Date: Sat, 1 Jun 1996 21:10:57 +0100
Organization: The University of York, UK
Lines: 51
Message-ID: <Pine.SGI.3.91.960601205941.18048C-100000@tower.york.ac.uk>
NNTP-Posting-Host: tower.york.ac.uk
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
In-Reply-To: <833618156.15278.0@gememail.demon.co.uk>
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp

On Sat, 1 Jun 1996, Owen wrote:

> I have written a small routine to draw sprites on to bitmap images and
> I have got it to work in C. I have tried rewritting the inner loop
> using inline assembly but I can't get it to work. When I complie the

[stuff deleted]

> __asm__ __volatile__ ("pushl %%ecx \n"
>                       "pushl %%ebx \n"
>                       "pushl %%edi \n"
>                       "pushl %%esi \n"
>                       "movl %0,%%edi \n"
>                       "movl %1,%%esi \n"
>                       "movl %2,%%ecx \n"

[stuff deleted]

>                       : "r" (dest_data), "r" (src_data),"r" (width));

The way you are passing the parameters to the asm code is not correct. 
Think about it: you are telling gcc to put dest_data, src_data, and width 
in registers (any registers it likes), and then moving them to other 
registers. What would happen if it put src_data in edi? When you move %0 
(the dest_data value) to %%edi it will clobber your src_data value.

You can fix this by telling gcc exactly which registers you want the 
parameters in. Get rid of the push and move instructions at the top of 
the function, and change the constraints to:

	: "D" (dest_data), "S" (src_data), "c" (width));

"D" means put it in edi, "S" means esi, and "c" means ecx. Alternatively 
you could leave gcc with the choice of register, but then you should use 
the registers it picks throughout your function, not move them to 
hardcoded registers of your own, ie. your function should work entirely 
with the %0 and %1 values.

Finally, if you need to save and restore specific registers, rather than 
pushing and popping the values it is easier to just add them to the 
clobbered list at the end of the asm statement. This way gcc will take 
care of them for you, but it is more efficient because they will only be 
saved/restored if they contain live data.


/*
 *	Shawn Hargreaves.        Why is 'phonetic' spelt with a ph?
 *	Check out Allegro and FED on http://www.york.ac.uk/~slh100/
 */