From: Sam S. <sd...@gn...> - 2009-08-16 20:44:35
|
> * Vladimir Tzankov <igmnaxbi@tznvy.pbz> [2009-08-16 20:11:26 +0300]: > > On 8/14/09, Sam Steingold <sd...@gn...> wrote: >> Bruno Haible wrote: >>> It's possible. Last time I measured it, it caused a performance decrease >>> of 17%, >>> on x86. You have to know that position-independent code means that every >>> access >>> to a global variable needs special code so that the access goes through >>> %eip. >> >> this pretty much kills the idea. >> however, these 17% looks like a lot - if it were true, people should have >> screamed about it it should have been fixed. > > It's somehow relevant. > I observe that MT builds are 10-20% slower than single thread ones - > mainly due to additional indirection when accessing globals. is this cumulative with -fPIC? i.e., is there any penalty from -fPIC in -DMULTITHREAD? -- Sam Steingold (http://sds.podval.org/) on Ubuntu 9.04 (jaunty) http://dhimmi.com http://jihadwatch.org http://honestreporting.com http://pmw.org.il http://palestinefacts.org http://memri.org http://ffii.org PI seconds is a nanocentury |