tree a1a4f40005020256a587096ccbc87ebf69e47f08
parent c42c66b2dd053527068cab84c7af44df08b00475
author Isabelle Taylor <taylori@google.com> 1537353319 +0100
committer Isabelle Taylor <taylori@google.com> 1537353319 +0100

trace_processor: Flatten CPU freq storage and remove cycles

In order to join the cpu freq information quickly with the sched table
the cpu freq events need to be in timestamp order, rather than
stored per cpu. A side effect of that is that we can't quickly compute
cycles the way that we were doing before, but that's OK because we
can compute cycles when we span join with the sched table.

The counters table is not optimised. We let SQLite do the
filtering and ordering for now.

Since the counters are no longer stored by cpu, duration can not
be computed on the fly. Instead a stage is added to sched_tracker
(rename pending) similar to the sched slices that waits for the next
event on the same cpu before outputting the prev event with it's
duration.

Change-Id: I27da5366aae6bd4cf6a7d30b1e248b938f50b965