As reported in #138173, enabling `-ftime-report` adds pass timing info
to the stats file if `-stats-file` is specified. This was determined to
be WAI. However, if one intentionally wants to put timer information in
the stats file, using `-ftime-report` may lead to a lot of logspam (that
can't be removed by directing stderr to `/dev/null` as that would also
redirect compiler errors). To address this, this PR adds a flag
`-stats-file-timers` that adds timer data to the stats file without
outputting to stderr.
This patch adds a new flag, -ftime-report-json, which outputs the same
information as -ftime-report but as JSON instead of -ftime-report's
pretty printed format.
This reverts commit 31ebe6647b7f1fc7f6778a5438175b12f82357ae.
The reason for the test failure is likely due to
`Name2PairMap::getTimerGroup(...)` not holding a lock.
Additionally, remove the behavior for both pass manager's timer manager
classes (`PassTimingInfo` for the old pass manager and
`TimePassesHandler` for the new pass manager) where these classes would
print the values of their timers upon destruction.
Currently, each pass manager manages their own `TimerGroup`s. This is
problematic because of duplicate `TimerGroup`s (both pass managers have
a `TimerGroup` for pass times with identical names and descriptions).
The result is that in Clang, `-ftime-report` has two "Pass execution
timing report" sections (one for the new pass manager which manages
optimization passes, and one for the old pass manager which manages the
backend). The result of this change is that Clang's `-ftime-report` now
prints both optimization and backend pass timing info in a unified "Pass
execution timing report" section.
Moving the ownership of the `TimerGroups` to globals also makes it
easier to implement JSON-formatted `-ftime-report`. This was not
possible with the old structure because the two pass managers were
created and destroyed in far parts of the codebase and outputting JSON
requires the printing logic to be at the same place because of
formatting.
Previous discourse discussion:
https://discourse.llvm.org/t/difficulties-with-implementing-json-formatted-ftime-report/84353
Currently, -ftime-report + new pass manager emits one line of report for each
pass run. This potentially causes huge output text especially with regular LTO
or large single file (Obeserved in private tests and was reported in D51276).
The behaviour of -ftime-report + legacy pass manager is
emitting one line of report for each pass object which has relatively reasonable
text output size. This patch adds a flag `-ftime-report=` to control time report
aggregation for new pass manager.
The flag is for new pass manager only. Using it with legacy pass manager gives
an error. It is a driver and cc1 flag. `per-pass` is the new default so
`-ftime-report` is aliased to `-ftime-report=per-pass`. Before this patch,
functionality-wise `-ftime-report` is aliased to `-ftime-report=per-pass-run`.
* Adds an boolean variable TimePassesHandler::PerRun to control per-pass vs per-pass-run.
* Adds a new clang CodeGen flag CodeGenOptions::TimePassesPerRun to work with the existing CodeGenOptions::TimePasses.
* Remove FrontendOptions::ShowTimers, its uses are replaced by the existing CodeGenOptions::TimePasses.
* Remove FrontendTimesIsEnabled (It was introduced in D45619 which was largely reverted.)
Differential Revision: https://reviews.llvm.org/D92436
