commit | ca06498ef3233fc78e7790f4d8d7190e1ea1c870 | [log] [tgz] |
---|---|---|
author | Ryan Savitski <rsavitski@google.com> | Tue Sep 24 16:51:13 2019 +0100 |
committer | Ryan Savitski <rsavitski@google.com> | Fri Sep 27 11:13:30 2019 +0000 |
tree | 80ecfc294672841692b2addc87bff12e62b84648 | |
parent | f9291073cc2b24f1131578b51f33715b54f51ccc [diff] |
android builds (incl. standalone): switch from Oz to O2 This speeds up ftrace parsing by about ~25% (12% little core -> 9% little core) on a crosshatch under medium load. In terms of binary size (android tree builds on master): HEAD: 12K bin/traced_probes 12K bin/traced 600K bin/perfetto 316K bin/trigger_perfetto 940K lib64/libperfetto.so 32K lib64/libperfetto_android_internal.so This topic: 12K bin/traced_probes 12K bin/traced 708K bin/perfetto 364K bin/trigger_perfetto 1.1M lib64/libperfetto.so 28K lib64/libperfetto_android_internal.so On GetStaticEventInfo in particular: * without the posted change, we're about +75K on libperfetto.so when switching to O2. I've kept the return type as a vector since the proto translation table ends up operating on a dynamic vector in the end. I've also kept it separate from trace_processor's ftrace descriptor array, since they are different enough that a shared definition ends up being more convoluted. "bloaty" diff on libperfetto.so (this topic (O2) vs head (Oz)): file size +165K vm size +203K https://paste.googleplex.com/5528773631934464 Change-Id: I442d42914dcc7ab56a09f068cb86b2842c2c3297
Perfetto 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.