From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 21711 invoked by alias); 26 Nov 2004 10:27:42 -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 21670 invoked from network); 26 Nov 2004 10:27:35 -0000 Received: from unknown (HELO ibm-4.MPA-Garching.MPG.DE) (130.183.83.34) by sourceware.org with SMTP; 26 Nov 2004 10:27:35 -0000 Received: from mpa-garching.mpg.de (nck-4.MPA-Garching.MPG.DE [130.183.84.214]) by ibm-4.MPA-Garching.MPG.DE (AIX4.3/8.9.3p2/8.9.1) with ESMTP id LAA86002; Fri, 26 Nov 2004 11:27:34 +0100 Message-ID: <41A70516.6080805@mpa-garching.mpg.de> Date: Fri, 26 Nov 2004 13:02:00 -0000 From: Martin Reinecke User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.6) Gecko/20040113 MIME-Version: 1.0 To: gcc@gcc.gnu.org CC: roger@eyesopen.com Subject: Re: [BENCHMARK]-mfpmath=sse should disable x387 intrinsics Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit X-SW-Source: 2004-11/txt/msg01020.txt.bz2 Roger Sayle wrote: > Could you present the performance results for your testcase with > "-mfpmath=387", "-mfpmath=sse" and "-mfpmath=sse,387"? It's relatively > rare for "-mfpmath=sse" to be a win on a Pentium4 benchmark, and to quote > Robert Scott Ladd from his Coyote Gulch benchmarking: > >>From http://www.coyotegulch.com/products/acovea/acovea_4.html >>> Much to my surprise, I have yet to find any consistent evidence that >>> options like -mfpmath=sse improve program performance. Thus Acovea >>> bears out my personal experience, though it does not explain why so >>> many people continue to suggest that I should use -mfpmath=sse to >>> generate floating-point code. If someone could suggest a good >>> "-mfpmath=sse", I'd appreciate seeing it. I have a real-world C++ code which gets a 10% performance gain on a P4 when compiled with -mfpmath=sse (using the mainline compiler of 20041125). Its critical inner loop consists of multiplications, additions and subtractions only. Hopefully this code will be open-sourced soon; as soon as that happens, I could contribute it as a floating-point benchmark. Cheers, Martin