From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 14831 invoked by alias); 15 Apr 2002 02:36:00 -0000 Mailing-List: contact gcc-prs-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Archive: List-Post: List-Help: Sender: gcc-prs-owner@gcc.gnu.org Received: (qmail 14817 invoked by uid 71); 15 Apr 2002 02:36:00 -0000 Date: Sun, 14 Apr 2002 19:36:00 -0000 Message-ID: <20020415023600.14816.qmail@sources.redhat.com> To: nobody@gcc.gnu.org Cc: gcc-prs@gcc.gnu.org, From: Jason Merrill Subject: Re: libstdc++/4150: catastrophic performance decrease in C++ code Reply-To: Jason Merrill X-SW-Source: 2002-04/txt/msg00760.txt.bz2 List-Id: The following reply was made to PR libstdc++/4150; it has been noted by GNATS. From: Jason Merrill To: libstdc++@gcc.gnu.org Cc: gcc-gnats@gcc.gnu.org Subject: Re: libstdc++/4150: catastrophic performance decrease in C++ code Date: Mon, 15 Apr 2002 03:25:59 +0100 --=-=-= Loren suggested defining _GLIBCPP_AVOID_FSEEK as a fix for 4150; I've tried it. It avoids the useless seeks for the stdin case, but it's a rather Pyrrhic improvement; it also slows down execution from 0:42 to 1:50 (vs 0:20 in 2.95). The comment in ios.cc says that this "hurts" performance; I would suggest that verb isn't strong enough. Perhaps "eviscerates". But going back to the problem without that define, the basic bug seems to be that after basic_filebuf::underflow() reads in a chunk of data, it then always rewinds the file to the beginning of that chunk. The next time it gets called, it seeks ahead to the end of that chunk to read the next one. And so on. So we get two useless seeks between each read. There may be a point to these seeks, but there are no comments in the code to justify them. Simply tearing them out brings us up to 2.95 performance levels. At least seekoff would need to be adjusted accordingly, but is there any reason why we can't leave the file pointer at the end of the last read block, rather than at the beginning as we do now? Jason --=-=-= Content-Type: text/x-patch Content-Disposition: inline *** include/bits/fstream.tcc.~1~ Fri Apr 12 11:28:30 2002 --- include/bits/fstream.tcc Mon Apr 15 02:49:14 2002 *************** namespace std *** 268,280 **** { if (__testout) _M_really_overflow(); ! #if _GLIBCPP_AVOID_FSEEK ! else if ((_M_in_cur - _M_in_beg) == 1) ! _M_file->sys_getc(); ! #endif ! else ! _M_file->seekoff(_M_in_cur - _M_in_beg, ! ios_base::cur, ios_base::in); } if (__testinit || __testget) --- 268,275 ---- { if (__testout) _M_really_overflow(); ! if (_M_in_cur < _M_in_end) ! abort (); } if (__testinit || __testget) *************** namespace std *** 316,331 **** if (__testout) _M_out_cur = _M_in_cur; __ret = traits_type::to_int_type(*_M_in_cur); - #if _GLIBCPP_AVOID_FSEEK - if (__elen == 1) - _M_file->sys_ungetc(*_M_in_cur); - else - { - #endif - _M_file->seekoff(-__elen, ios_base::cur, ios_base::in); - #if _GLIBCPP_AVOID_FSEEK - } - #endif } } } --- 311,316 ---- --=-=-=--