From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-help-return-18824-listarch-gcc-help=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 983 invoked by alias); 8 Jan 2005 05:59:46 -0000
Mailing-List: contact gcc-help-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Archive: <http://gcc.gnu.org/ml/gcc-help/>
List-Post: <mailto:gcc-help@gcc.gnu.org>
List-Help: <mailto:gcc-help-help@gcc.gnu.org>
Sender: gcc-help-owner@gcc.gnu.org
Received: (qmail 940 invoked from network); 8 Jan 2005 05:59:38 -0000
Received: from unknown (HELO eastnet.febras.ru) (212.91.197.11)
  by sourceware.org with SMTP; 8 Jan 2005 05:59:38 -0000
Received: from marine.febras.ru ([212.91.197.10])
	by eastnet.febras.ru with esmtp (Exim 4.43)
	id 1Cn9dW-0006nn-P9
	for gcc-help@gcc.gnu.org; Sat, 08 Jan 2005 15:59:42 +1000
Received: from [192.168.11.218] (helo=192.168.11.218)
	by marine.febras.ru with esmtp (Exim 4.12)
	id 1Cn9dS-0001pY-00
	for gcc-help@gcc.gnu.org; Sat, 08 Jan 2005 15:59:38 +1000
From: "Dmitry K." <dmixm@marine.febras.ru>
To: gcc-help@gcc.gnu.org
Subject: How to force GNU CC use "inc" for `++' operator?
Date: Sat, 08 Jan 2005 05:59:00 -0000
User-Agent: KMail/1.5
MIME-Version: 1.0
Content-Type: text/plain;
  charset="us-ascii"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
Message-Id: <200501081601.12229.dmixm@marine.febras.ru>
X-SW-Source: 2005-01/txt/msg00039.txt.bz2

Hi.

In the next example GCC differently processes two alike lines, containing
increment of `c':

   extern int foo1 (int,int);
   int foo (int x, int y)
   {
       unsigned char c= 0;
       if (foo1(x,y)) c++;
       if (foo1(x,y)) c++;
       return c;
   }

`inc r15' is used at the second increment.  But at first -- pair operations:
`ldi' and `mov'. This occupies two words instead of one and is executed twice
longer:

   foo:
   /* prologue: frame size=0 */
        push r15
        push r16
        push r17
        push r28
        push r29
   /* prologue end (size=5) */
        movw r16,r24
        movw r28,r22
        clr r15
        rcall foo1
        or r24,r25
        breq .L2
        ldi r24,lo8(1)		/**************/
        mov r15,r24		/**************/
   .L2:
        movw r22,r28
        movw r24,r16
        rcall foo1
        or r24,r25
        breq .L3
        inc r15			/**************/
   .L3:
        mov r24,r15
        clr r25
   /* epilogue: frame size=0 */
        pop r29
        pop r28
        pop r17
        pop r16
        pop r15
        ret

Certainly, 4% is rubbish. But this example is an extract from real program,
where lack of the free register gave considerable decline.

Was used: gcc 3.4.3, --target=avr, -Os, -mmcu=atmega8

Thanks in advance.