Xref: news2.mv.net comp.os.msdos.djgpp:6533
From: Pål-Kristian Engstad <engstad AT funcom DOT com>
Newsgroups: comp.os.msdos.djgpp
Subject: Re: Inline assembly and DMA transfers.
Date: Mon, 29 Jul 1996 13:23:55 +0200
Organization: Funcom Oslo AS
Lines: 40
Message-ID: <31FC9F4B.35F5@funcom.com>
References: <199607230018 DOT KAA25409 AT gbrmpa DOT gov DOT au>
Reply-To: engstad AT funcom DOT com
NNTP-Posting-Host: khazad.funcom.com
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp

Leath Muller wrote:
> include <stdio.h>
> 
> volatile char *mem_block = NULL;
> volatile int value = 0;
> 
> void fill_block()
> {
[--snip--]
> }
> 

Try this:

static __inline__ void fill_block(char *mem_block, int value)
{
	asm volatile (	"1:	movl %1, (%0)\n"
			"	addl $4, %0\n"
			"	decl %2\n"
			"	cmpl $0, %2\n"
			"	jnz  1b\n"
			: /* Outputs. */
			: /* Inputs. */
			  "r" (mem_block), "r" (value), "r" (50) 
			: /* Modifies. */
			  "cc", 	/* Condition flags. */
			  "memory" 	/* Memory.          */
	);
}

Note that with this scheme, you'll get:
	1. The function inlined, avoiding a precious call.
	2. Register allocation automatic by the compiler.

PKE.
-- 
----------------------------------------------------------------------
| Pål-Kristian Engstad | engstad AT funcom DOT com  | Games programmer, PSX |
| Funcom Oslo AS       | Ph +47 22 42 01 02  | developer & Linux Fan |
----------------------------------------------------------------------