On Tuesday 05 June 2007, Sam Ravnborg wrote:
> It's stuff like this that worries me..
> In the kernel we rely on a lot of things - more than in usual projects.
> And one thing we have established in the 2.6 kernel is a trust in the
> build system. If you change something kbuild will recompile
> eve
On Tue, Jun 05, 2007 at 11:20:22AM +0200, Arnd Bergmann wrote:
> On Tuesday 05 June 2007, Vegard Nossum wrote:
> > > You also need to take some care about the gcc arguments that you pass.
> > > E.g. passing -DMODULE when building the precompiled header means that
> > > you can't use that .pch when
On Tuesday 05 June 2007, Vegard Nossum wrote:
> > You also need to take some care about the gcc arguments that you pass.
> > E.g. passing -DMODULE when building the precompiled header means that
> > you can't use that .pch when you build a file where you don't pass
> > -DMODULE.
>
> So does it mak
On 5/31/07, Arnd Bergmann <[EMAIL PROTECTED]> wrote:
> One way that might be feasible is to have a single header file precompiled,
> and make that include all the common files (fs.h, mm.h, module.h, autoconf.h,
> sched.h, init.h). You then need to pass that header file to gcc as the
> first one tha
On Thursday 31 May 2007, Vegard Nossum wrote:
> By the way, according to the GCC manual [1], "Only one precompiled
> header can be used in a particular compilation", which, if I interpret
> it correctly, might make header precompilation useless for the kernel
> anyway. But here's my attempt (I hope
Hello,
I decided to try out the header precompilation feature of newer GCCs
to see if it would speed up kernel building. Following is a patch
which 1) adds "headers" target to top-level makefile, 2) allows and
compiles .h files into .h.gch using gcc. When gcc compiles the normal
C files, it looks
