On Mon, Jul 15, 2019 at 04:41:28PM +0200, Martin Liška wrote:
> On 7/15/19 4:31 PM, Segher Boessenkool wrote:
> > On Mon, Jul 15, 2019 at 04:23:18PM +0200, Martin Liška wrote:
> >> On 7/15/19 4:17 PM, Segher Boessenkool wrote:
> >>> Is that enough? Can there be two entries with the same count as well
> >>> as value?
> >>
> >> Yes, it can happen very unlikely in a multi-threaded instrumentation,
> >> but it will not make a problem.
> >
> > It will result in different sort order on different hosts, then. Is
> > that okay here? Does a different order here not lead to different
> > compiler output?
>
> The memory layout of the {value, count} tuple can be different. But the
> function will return you K-th most common value. And that will return the
> same.
Ah okay, we only look at the value later. This should be documented I
think (just in the comparison routine is fine).
Segher
