On Mon, Sep 23, 2019 at 02:03:35PM +0200, Andrzej Ostruszka wrote: > On 9/23/19 9:23 AM, Thomas Monjalon wrote: > [...] > > Please can we get some numbers to understand how longer it is? > > Below numbers are for make based (make -j8) clean build on my system: > > non-LTO > real: 144.56s, user:451.81s, sys:48.46s, CPU:346% > > LTO > real: 607.20s, user:2141.71s, sys:88.36s, CPU:367% > > So it is similar ~5x increase as Mattias has reported. Have not > measured it, but the lion share of that increase is due to linking of > 'test' apps. >
Interesting. Do we want to explicitly not use lto for the test app? > I would vote for leaving LTO as an option - although I must admit I did > not get what Ray meant by saying: > > 20/09/2019 09:38, Ray Kinsella: > [...] > > I would just like to see it in or out, not another build time option. > > If "in or out" means "either accept the patches with LTO on and no > config option or reject them" then I disagree. Even if run time > improvements are questionable I find the additional link time warnings > beneficial and would like to have an easy way to turn them on when doing > final touches before pushing out. Looking at it from the meson build view-point, support for lto is built into the tool itself, so the support can't be removed as such - it will either result in a working build or not. Therefore, I'm for taking a patch to support lto in meson, whatever about supporting it through make etc. /Bruce
