On 06/27/2013 10:58 AM, Ralf Baechle wrote:
On Fri, Dec 07, 2012 at 04:09:46PM +0100, Ralf Baechle wrote: It's a while already - but I haven't forgotten about this little optimization.
[...]
Using a quick hack (not in below patch) I reduced CPU support to just a single type. Then not only code for the other 2 variants is dropped by the compiler but also dead code elimination can discard ifs and switches leaving a highly customized, smaller and faster kernel. The best of it is that it doesn't even rely on LTO, just a decent optimizer of a per-file compiler. Patch for your enjoyment below. I've done my testing with plain old boring gcc 4.7 btw.
Much appreciated. I will test it out here and get back to you. Thanks. Steve
