On Fri, Jan 12, 2018 at 11:10 PM, Segher Boessenkool
<segher@xxxxxxxxxxxxxxxxxxx> wrote:
> On Fri, Jan 12, 2018 at 10:45:31PM +0100, Arnd Bergmann wrote:
>> > I guess you could enable the _x routines whenever you use ubsan? Ubsan
>> > will cause much bigger code growth than the handful of insns in those
>> > routines?
>>
>> Right, that could work, too. My patch that Herbert merged intentionally
>> used -Os also for non-UBSAN builds because it turned out to
>> be much faster (see gcc PR83651),
>
> "Much"?
>
> -Os is *slower* with 8.0, 5% faster with 7.2, 4% faster with 7.1,
> slower with 7.0 and 6.3. Your numbers, #c1.
>
> Anf this is the generic code of course, which is slow anyway (not to
> mention insecure).
Right. I've done some more investigation anyway, starting over with the
analysis of the gcc options that change it. I've found now that turning
off '-fcode-hoisting' but leaving on the other options I had suspected
earlier (-O2 instead of -Os, -ftree-sra, -ftree-pre) also fixes the
stack problem, and appears to result in the best performance so
far. I need to rerun the whole test matrix, but that seems rather
promising, and the result may also help debug what's really happening.
Arnd
