Path: csiph.com!usenet.blueworldhosting.com!feeder01.blueworldhosting.com!peer01.iad.highwinds-media.com!news.highwinds-media.com!feed-me.highwinds-media.com!post01.iad.highwinds-media.com!fx23.iad.POSTED!not-for-mail From: Robert Wessel Newsgroups: comp.arch.arithmetic Subject: Re: SSE "denormals are zeros" Message-ID: References: User-Agent: ForteAgent/7.20.32.1218 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Lines: 16 X-Complaints-To: abuse@easynews.com Organization: Forte - www.forteinc.com X-Complaints-Info: Please be sure to forward a copy of ALL headers otherwise we will be unable to process your complaint properly. Date: Fri, 17 Jun 2016 12:33:32 -0500 X-Received-Bytes: 1477 X-Received-Body-CRC: 2815026940 Xref: csiph.com comp.arch.arithmetic:84 On Fri, 17 Jun 2016 18:11:28 +0200, Bonita Montero wrote: >Does anyone know what the "denormals are zeros" flag of the >x86 MXCSR is good for? >Or more precisely: I know what it does, but I don't know why >it should make sense to consider denormal values as zeros. Mainly performance - denormals tend to be slow (although less so on recent x86s). Some codes do things like converge to zero, but end up passing through the denormal range first - just skipping that can sometimes be a considerable performance improvement. There are some downsize to disabling gradual underflow, but in practice many cases where you get them you're on your way to zero anyway, and in most cases the advantages of gradual underflow are very small.