On Fri, 18 Oct 2013 13:45:22 Trent W. Buck wrote:
Russell Coker <russell@coker.com.au> writes:
Some people have compiled a 32bit address system with 64bit operations and all AMD64 registers. For everything that doesn't require more than 3G of address space per process it should beat i386 or standard AMD64 builds.
https://en.wikipedia.org/wiki/X32_ABI
Those people are strange and should be avoided.
The Wikipedia page you cited notes that some operations are 32% faster with X32, that's a significant difference. http://etbe.coker.com.au/2008/09/11/execmod-and-se-linux-i386-must-die/ http://etbe.coker.com.au/2008/09/12/fixing-execmod-textrel-problems-in-lenny... http://etbe.coker.com.au/2011/02/16/mplayer-squeeze-selinux/ It also notes that the average is 5% to 8% improvement. One thing that's been done on i386 for similar performance gains is to use libraries that have self- modifying code to avoid using a register for offsets. Above are the URLs for some blog posts I've written about this. I have done a fair bit of work maintaining an alternate repository for mplayer and other video related packages to deal with the fact that the Debian maintainers and upstream want the extra 5% or so performance while I want systems that are difficult to exploit. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/