Re: [PATCH 1/2] perf tools: Add reference timestamp to perf header

[Peter Zijlstra]

On Mon, 2010-12-13 at 19:05 +0100, Frederic Weisbecker wrote:
> 
> There is one writer per cpu and a (stupid) multiplexing at report time.
> It was working. Just the sorting on report time was (stupidly) slow.
> 
> I should perhaps rebase that branch and try to make something with it. 

Right, so we need to have a small ~16 events sort window per file [*],
and then a merge-sort over all files, simply consumer the most recent
event across all files.

We can have out-of-order events on a single cpu due to nesting and
taking the time-stamp _before_ we reserve the buffer. We could cure some
of this by moving the perf_clock() call from
__perf_event_header__init_id()/perf_prepare_sample() to
__perf_event_output_id_sample()/perf_output_sample().

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/