On 07/03/14 14:26, Justin Cormack wrote: > On Fri, Mar 7, 2014 at 2:20 PM, Antti Kantee <[email protected]> wrote: >> I already knew that curlwp needs to be optimized -- does anyone have any >> really good ideas on how to do that on x86? -- but at least now there's >> confirmation. > > I seem to remember that using __thread rather than pthread_key_* was > noticeably faster, as a quick improvement.
We should definitely do that as a first step. On archs with plentiful registers it's possible to use -ffixed to make curlwp essentially free (with some minor pain in wrapping hypercalls since they may call non-ffixed code), but I don't know of a similar scheme on x86. The kernel uses segmentation (at least last I checked), so I don't think that'll translate to userspace. ------------------------------------------------------------------------------ Subversion Kills Productivity. Get off Subversion & Make the Move to Perforce. With Perforce, you get hassle-free workflows. Merge that actually works. Faster operations. Version large binaries. Built-in WAN optimization and the freedom to use Git, Perforce or both. Make the move to Perforce. http://pubads.g.doubleclick.net/gampad/clk?id=122218951&iu=/4140/ostg.clktrk _______________________________________________ rumpkernel-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/rumpkernel-users
