From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from joooj.vinc17.net (joooj.vinc17.net [155.133.131.76]) by sourceware.org (Postfix) with ESMTPS id 7787E3858C5F for ; Fri, 4 Aug 2023 14:51:02 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 7787E3858C5F Authentication-Results: sourceware.org; dmarc=none (p=none dis=none) header.from=vinc17.net Authentication-Results: sourceware.org; spf=none smtp.mailfrom=vinc17.net Received: from smtp-zira.vinc17.net (atoulouse-657-1-1002-191.w90-55.abo.wanadoo.fr [90.55.204.191]) by joooj.vinc17.net (Postfix) with ESMTPSA id 3FB8629D; Fri, 4 Aug 2023 16:51:01 +0200 (CEST) Received: by zira.vinc17.org (Postfix, from userid 1000) id AEAA228002B8; Fri, 4 Aug 2023 16:51:00 +0200 (CEST) Date: Fri, 4 Aug 2023 16:51:00 +0200 From: Vincent Lefevre To: Paul Zimmermann Cc: Szabolcs Nagy , Wilco.Dijkstra@arm.com, libc-alpha@sourceware.org, Joe.Ramsay@arm.com Subject: Re: [PATCH] aarch64: Improve SVE sin polynomial Message-ID: <20230804145100.GB2741@zira.vinc17.org> Mail-Followup-To: Vincent Lefevre , Paul Zimmermann , Szabolcs Nagy , Wilco.Dijkstra@arm.com, libc-alpha@sourceware.org, Joe.Ramsay@arm.com References: MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: X-Mailer-Info: https://www.vinc17.net/mutt/ User-Agent: Mutt/2.2.10+64 (b470a9ad) vl-149028 (2023-06-05) X-Spam-Status: No, score=-1.1 required=5.0 tests=BAYES_00,KAM_DMARC_STATUS,KAM_LAZY_DOMAIN_SECURITY,SPF_HELO_NONE,SPF_NONE,TXREP autolearn=no autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: On 2023-08-04 09:28:53 +0200, Paul Zimmermann via Libc-alpha wrote: > Hi Szabolcs, > > > i'd say we want to minimize the maximum of > > > > (poly(x) - sin(x))/ulp(sin(x)) > > > > instead of > > > > (poly(x) - sin(x))/sin(x) > > interesting question! I have asked the Sollya developers if this is > possible. I would say that a large change of the weight function (e.g. a discontinuity point) is likely to generate an extrema for the error function, which could affect the quality of the minimax approximation. That's from the theory. In practice, it is still possible that this works by chance, but I suppose that this depends on the considered function and domain. Concerning Sollya, doesn't it uses the derivative of the error function (making the direct use of ulp impossible)? -- Vincent Lefèvre - Web: 100% accessible validated (X)HTML - Blog: Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)