| commit | 89795fd9193d7375a8380c14d84c5f4418c20829 | [log] [tgz] |
|---|---|---|
| author | Primiano Tucci <primiano@google.com> | Mon Feb 18 23:08:06 2019 +0000 |
| committer | Primiano Tucci <primiano@google.com> | Thu Feb 21 09:05:48 2019 +0000 |
| tree | 1674d293ba03b07a7dad88291a4a52eeae8bdd58 | |
| parent | fd734301165726c46786e8c3895e07007c78f0d6 [diff] |
Introduce base::CircularQueue and improve trace processor sorter
This CL introduce a queue container backed by a circular array.
This vastly improves TraceSorter's performance, because it makes
delete-front O(1) (% invoking dtors), but still allows std::sort()
of the elements with the same cost of a vector.
This solution is also faster than using a std::deque for |events_|
instead of the std::vector, this is because CircularBuffer is way
simpler than a std::deque.
Benchmark of sorting-stage, obtained commenting out the parsing
stage and running on a 7GB trace [1] on a macbook pro 15 2018.
std::vector: 70 MB/s
std::deque: 260 MB/s
CircularBuffer: 285 MB/s
More CLs will further improve the sorter, but for now this CL alone
makes the trace processor cope better with large traces.
[1] SHA1: ad54e7de3a7,
trace-walleye-QP1A.190211.001-2019-02-15-20-27-27.perfetto-trace
from www/~primiano/perfetto/sample_traces
Test: perfetto_unittests --gtest_filter=CircularBufferTest.*
Change-Id: Ie804f54e08d7a9d374aefbe6141cbba65ddf25ecPerfetto is an open-source project for performance instrumentation and tracing of Linux/Android/Chrome platforms and user-space apps.
See www.perfetto.dev for docs.