From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 11930 invoked by alias); 3 Apr 2003 11:30:38 -0000 Mailing-List: contact gcc-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Archive: List-Post: List-Help: Sender: gcc-owner@gcc.gnu.org Received: (qmail 11920 invoked from network); 3 Apr 2003 11:30:38 -0000 Received: from unknown (HELO anchor-post-39.mail.demon.net) (194.217.242.80) by sources.redhat.com with SMTP; 3 Apr 2003 11:30:38 -0000 Received: from mailgate.cyclicode.net ([62.49.221.218] helo=humbug.cyclicode.net) by anchor-post-39.mail.demon.net with esmtp (Exim 3.36 #2) id 1912vQ-0006Tr-0d; Thu, 03 Apr 2003 12:30:32 +0100 Received: from cyclicode.net (humbug.cyclicode.net [62.49.221.218]) by humbug.cyclicode.net (8.11.6/8.11.2) with ESMTP id h33BWR810984; Thu, 3 Apr 2003 12:32:27 +0100 Message-ID: <3E8C1BCA.5090502@cyclicode.net> Date: Thu, 03 Apr 2003 12:59:00 -0000 From: Dave Hudson User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.2.1) Gecko/20021130 X-Accept-Language: en-gb, en-us MIME-Version: 1.0 To: Gabriele Caracausi CC: GCC ML Subject: Re: help about new porting (.h file) References: <20030403095715.19015.qmail@mx7.aruba.it> In-Reply-To: <20030403095715.19015.qmail@mx7.aruba.it> Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit X-SW-Source: 2003-04/txt/msg00135.txt.bz2 Hi Gabriele, Gabriele Caracausi wrote: > Hi all, > * POINTER_SIZE is setted to 16 bit > * Pmode is setted to HImode > * INT_TYPE_SIZE is setted to 16 bit so HImode should be 8 bit. No, HImode is 16 bits for the IP2k. > In MODES_TIEBLE_P is written that QImode and HImode are readable as the > same type and this should mean that they have both the same size, but if > HImode is 16 bit wide QImode should be 8 bit wide and so they should not When the modes are tieable it means that we can avoid copying things around. I just reviewed this code though and the current definition for the IP2k is just plain wrong (as was the predecessor to it). The code suffers some minor size regressions if I just define it to be "1" but also wins in a number of other places. In this particular port the key thing is to avoid gcc thinking that it can tie certain HImode or QImode regs to SImode or DImode because the 16-bit IP and DP pointer regs are not contiguous with the 32 8-bit GPRs. Regards, Dave