From: sam AT campbellsci DOT co DOT uk (Samuel James Liddicott) Subject: Re: Time and motion studies of gcc and egcs and LCC 3 Feb 1998 18:16:53 -0800 Message-ID: <004701bd30c0$77196910$2a0110ac.cygnus.gnu-win32@sam.ethernet> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit To: marcus AT bighorn DOT dr DOT lucent DOT com, gnu-win32 AT cygnus DOT com -----Original Message----- From: marcus AT bighorn DOT dr DOT lucent DOT com To: gnu-win32 AT cygnus DOT com Date: 03 February 1998 04:02 Subject: Re: Time and motion studies of gcc and egcs and LCC >I don't think that the overhead of getting into a DLL is all that great, so >I think that the performance hit is somewhere else, like inside cygwin.dll >and NT itself. The cost to enter a DLL is an additional jump instruction >(indirect through the linkage pointer). To return is no cost. Granted, this >does bring an additional page into the working set (or two, actually, because >the linkage pointer is probably on a different page), but if the calls are >frequent, it should not be flushed frequently, so the page fault cost is >not very great either. I read that in win95 the pages are flushed in order of first-in-first-out, and if a flushed page is a commonly used page, it will just be first-in again a bit quicker, and all this to save on the overheads of working out which pages are least used. I was shocked when I read it too, but imagine the overheads of doing page usage sorting every time you need to load a new page? MS seemed to think little was gained. Sam Liddicott | Nothing I say is to be attributed as | Campbell Scientific Ltd. | a company statement or representation. | Campbell Park, 80 Hathern Road, *----------------------------------------+ Shepshed, Leic. United Kingdom. LE12 9AL Phone: +44 (0) 1509 601141 Email: sam AT campbellsci DOT co DOT uk Fax: +44 (0) 1509 601091 - For help on using this list (especially unsubscribing), send a message to "gnu-win32-request AT cygnus DOT com" with one line of text: "help".