From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 31816 invoked by alias); 13 May 2004 11:15:52 -0000 Mailing-List: contact libc-hacker-help@sources.redhat.com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: libc-hacker-owner@sources.redhat.com Received: (qmail 31768 invoked from network); 13 May 2004 11:15:50 -0000 Received: from unknown (HELO sunsite.ms.mff.cuni.cz) (195.113.15.26) by sourceware.org with SMTP; 13 May 2004 11:15:50 -0000 Received: from sunsite.ms.mff.cuni.cz (sunsite.mff.cuni.cz [127.0.0.1]) by sunsite.ms.mff.cuni.cz (8.12.8/8.12.8) with ESMTP id i4D9363j012379; Thu, 13 May 2004 11:03:06 +0200 Received: (from jakub@localhost) by sunsite.ms.mff.cuni.cz (8.12.8/8.12.8/Submit) id i4D936EC012369; Thu, 13 May 2004 11:03:06 +0200 Date: Thu, 13 May 2004 11:15:00 -0000 From: Jakub Jelinek To: Ulrich Drepper Cc: Glibc hackers Subject: [PATCH] Speed up sprintf Message-ID: <20040513090306.GA5191@sunsite.ms.mff.cuni.cz> Reply-To: Jakub Jelinek Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4i X-SW-Source: 2004-05/txt/msg00026.txt.bz2 Hi! My http://sources.redhat.com/ml/libc-hacker/2003-05/msg00032.html change apparently slowed down sprintf quite a lot on longer strings, if buffer passed to sprintf is located in the first half of address space. The problem is that the fix from last year for sprintf now sets write_end to ~0L and if write_ptr is small enough, (ssize_t) (write_end - write_ptr) is negative. This results in overflow being called for every single character instead of doing a mempcpy. The following patch seems to fix it, now sprintf at least on the sample testcase I have on IA-64 is 30% faster than sprintf in glibc 2.2.4 (as opposed to being almost 4 times slower). 2004-05-13 Jakub Jelinek * libio/genops.c (_IO_default_xsputn): Avoid one overflow per char if count is negative, yet write_ptr < write_end. (_IO_default_xsgetn): Avoid one underflow per char if count is negative, yet read_ptr < read_end. --- libc/libio/genops.c.jj 2004-02-12 11:52:09.000000000 +0100 +++ libc/libio/genops.c 2004-05-13 13:02:15.715861960 +0200 @@ -447,10 +447,10 @@ _IO_default_xsputn (f, data, n) for (;;) { /* Space available. */ - _IO_ssize_t count = f->_IO_write_end - f->_IO_write_ptr; - if (count > 0) + if (f->_IO_write_ptr < f->_IO_write_end) { - if ((_IO_size_t) count > more) + _IO_size_t count = f->_IO_write_end - f->_IO_write_ptr; + if (count > more) count = more; if (count > 20) { @@ -462,9 +462,7 @@ _IO_default_xsputn (f, data, n) #endif s += count; } - else if (count <= 0) - count = 0; - else + else if (count) { char *p = f->_IO_write_ptr; _IO_ssize_t i; @@ -504,10 +502,10 @@ _IO_default_xsgetn (fp, data, n) for (;;) { /* Data available. */ - _IO_ssize_t count = fp->_IO_read_end - fp->_IO_read_ptr; - if (count > 0) + if (fp->_IO_read_ptr < fp->_IO_read_end) { - if ((_IO_size_t) count > more) + _IO_size_t count = fp->_IO_read_end - fp->_IO_read_ptr; + if (count > more) count = more; if (count > 20) { @@ -519,9 +517,7 @@ _IO_default_xsgetn (fp, data, n) #endif fp->_IO_read_ptr += count; } - else if (count <= 0) - count = 0; - else + else if (count) { char *p = fp->_IO_read_ptr; int i = (int) count; Jakub