From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from list by monty-python.gnu.org with tmda-scanned (Exim 4.34)
	id 1BPpOC-00022d-GF
	for qemu-devel@nongnu.org; Mon, 17 May 2004 17:11:12 -0400
Received: from mail by monty-python.gnu.org with spam-scanned (Exim 4.34)
	id 1BPpNc-0001re-Kd
	for qemu-devel@nongnu.org; Mon, 17 May 2004 17:11:08 -0400
Received: from [193.252.22.30] (helo=mwinf0102.wanadoo.fr)
	by monty-python.gnu.org with esmtp (Exim 4.34) id 1BPpNc-0001r1-5W
	for qemu-devel@nongnu.org; Mon, 17 May 2004 17:10:36 -0400
Received: from bellard.org (ATuileries-112-1-3-148.w81-48.abo.wanadoo.fr
	[81.48.134.148])
	by mwinf0102.wanadoo.fr (SMTP Server) with ESMTP id 84D821BFD38F
	for <qemu-devel@nongnu.org>; Mon, 17 May 2004 23:10:34 +0200 (CEST)
Message-ID: <40A92AD5.3090504@bellard.org>
Date: Mon, 17 May 2004 23:12:53 +0200
From: Fabrice Bellard <fabrice@bellard.org>
MIME-Version: 1.0
Subject: Re: [Qemu-devel] Speed up hack
References: <200405170918.i4H9I4OR001702@treas.simtreas.ru>
	<40A8E6C4.6050105@simtreas.ru>
In-Reply-To: <40A8E6C4.6050105@simtreas.ru>
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit
Reply-To: qemu-devel@nongnu.org
List-Id: qemu-devel.nongnu.org
List-Unsubscribe: <http://mail.nongnu.org/mailman/listinfo/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=unsubscribe>
List-Archive: <http://mail.gnu.org/pipermail/qemu-devel>
List-Post: <mailto:qemu-devel@nongnu.org>
List-Help: <mailto:qemu-devel-request@nongnu.org?subject=help>
List-Subscribe: <http://mail.nongnu.org/mailman/listinfo/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=subscribe>
To: qemu-devel@nongnu.org

Can you estimate the speed gains ? I am especially interested by the 
poll() overhead. For the rest, optimizing DMA_run() would be simpler and 
more efficient.

Fabrice.

Vladimir N. Oleynik wrote:
> Hi.
> 
> I generate little speed up hack:
> 
> 1) new TICKS_BEFORE_POLL constant from make cpu loop before
> call poll().
> 2) call DMA_run() if sb16 or floppy present
> 3) also, for extreme overclocker: if change MAX_FD from 2 to 1,
> you can make very little speed up also ;-)
> 
> 
> --- vl.c~       2004-05-17 20:15:26.000000000 +0400
> +++ vl.c        2004-05-17 19:52:44.000000000 +0400
> @@ -1634,6 +1634,8 @@
>      }
>  }
> 
> +#define TICKS_BEFORE_POLL 100
> +
>  int main_loop(void)
>  {
>  #ifndef _WIN32
> @@ -1641,6 +1643,7 @@
>      IOHandlerRecord *ioh, *ioh_next;
>      uint8_t buf[4096];
>      int n, max_size;
> +    int tick_before_poll=TICKS_BEFORE_POLL;
>  #endif
>      int ret, timeout;
>      CPUState *env = global_env;
> @@ -1670,6 +1673,11 @@
>             Sleep(timeout);
>  #else
> 
> +       if(tick_before_poll!=0 && timeout==0) {
> +               tick_before_poll--;
> +               goto ltick_before_poll;
> +       }
> +       tick_before_poll=TICKS_BEFORE_POLL;
>         /* poll any events */
>         /* XXX: separate device handlers from system ones */
>         pf = ufds;
> @@ -1741,6 +1749,7 @@
>         }
>  #endif
> 
> +       ltick_before_poll:
>  #endif
> 
>         if (vm_running) {
> @@ -1750,10 +1759,18 @@
>             if (audio_enabled) {
>                 /* XXX: add explicit timer */
>                 SB16_run();
> -           }
> +               DMA_run();
> +           } else {
> +               int i;
> 
> -           /* run dma transfers, if any */
> -           DMA_run();
> +               /* run dma transfers, if any */
> +               for(i = 0; i < MAX_FD; i++) {
> +                       if(fd_table[i]) {
> +                               DMA_run();
> +                               break;
> +                       }
> +               }
> +           }
>         }
> 
>         /* real time timers */
> 
> 
> --w
> vodz
> 
> 
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://mail.nongnu.org/mailman/listinfo/qemu-devel
>