RFR: 8311976: Inconsistency in usage of CITimeVerbose to generate compilation logs

[Ashutosh Mehra]

On Thu, 13 Jul 2023 23:07:35 GMT, Vladimir Kozlov <kvn at openjdk.org> wrote:

>> Please review this PR for controlling timing information of C1 compilation phases using CITimeVerbose option, same as for C2 compilations.
>> I also took this opportunity to fix some other minor issues with logging:
>> 1. The PhaseTraceTime object should be constructed after setting the compiler data as PhaseTraceTime constructor calls `Compilation::current()`. For this reason I moved the statement `PhaseTraceTime timeit(_t_compile)` after the call to `_env->set_compiler_data(this);`.
>> 2. Previous step also allowed to remove the nullptr check for `Compilation::current()` in PhaseTraceTime constructor.
>> 3. I noticed the call to ComileLog->done() only prints `phase_done` tag and ignores all other parameters passed to it. This was due to a bug in `xmlStream::va_done` which is also fixed in here.
>> 4. Remove unnecessary statements in TracePhase destructor as the object already has the fields computed in the constructor.
>> 5. Some bikeshedding like TimerName -> TimerId and TracePhase::C -> TracePhase::_compile
>> 
>> I felt these are all minor fixes so I clubbed them together. here If it feel inappropriate I can pull them in their own PRs.
>> 
>> Testing: GHA testing passed
>
> @ashu-mehra, can you show output before and after this changes? It would be nice to see what you are fixing.

@vnkozlov thanks for reviewing it.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/14880#issuecomment-1635826005