The |nsIFile*| one is only called by the |const nsACString&| one, so this patch
combines them.
--HG--
extra : rebase_source : d8338e88cef4799d95e590c056ab343d5a1c546a
profiler_get_gatherer() exposes ProfileGatherer to the outside world in a way
that makes future changes difficult.
This patch:
- Removes ProfileGatherer.h from the list of headers exported from the
profiler.
- Removes nsIProfiler.profileGatherer and nsProfiler::GetProfileGatherer().
- Replaces profiler_get_gatherer() with three new functions that provide
minimal but sufficient access to ProfileGatherer:
profiler_will_gather_OOP_profile(), profiler_gathered_OOP_profile(), and
profiler_OOP_exit_profile().
These functions provide access to the ProfileGatherer in a similar fashion to
the pre-existing functions profiler_get_profile_jsobject_async() and
profiler_save_profile_to_file_async()
This significantly reduces the size of the profiler's API surface.
--HG--
rename : tools/profiler/public/ProfileGatherer.h => tools/profiler/gecko/ProfileGatherer.h
extra : rebase_source : d8e06a1133d4098c3a214858d3ff2c4bdcd9f1f2
This removes the one use of gStartTime outside of platform*.cpp, which lets us
restrict its visibility to just that compilation unit.
--HG--
extra : rebase_source : bf7207572cba5c1a31b544ea73e783ecd559978a
PlatformStart() and PlatformStop() are currently responsible for setting and
clearing gIsActive, but it's better if we do it in profiler_{start,stop}().
The patch also does the following.
- Adds some missing emacs/vim modelines.
- Makes Platform{Start,Stop}() crash if they have failures. I'm not at all
confident that ignoring the errors as is currently done will result in
sensible behaviour, so brittleness is better.
--HG--
extra : rebase_source : b9ab8437f5b92f6a8993ba7677ecb74a321ce219
This patch mostly does formatting fixes.
It also removes some declarations from platform.h that are no longer necessary
now that platform-linux-android.cpp is in the same compilation unit as
platform.cpp (due to it being #include-d directly); this required reordering
some things.
--HG--
extra : rebase_source : d07ef71455885fe8f1414d87c261ca054989a6a8
This avoids the need for platform-linux-android.cpp to read gInterval off the
main thread in an awkward spot. It also makes platform-linux-android.cpp
more like platform-{win32,macos}.cpp.
--HG--
extra : rebase_source : c1c76a382d6373f9fd2e3f89a1e1f8fef9072257
There is another PlatformStop() call earlier in the function, and gIsActive is
always false by the time we reach the removed call, so it's dead code.
--HG--
extra : rebase_source : 3b358b6bef47d394d6d6bc76d1153ea38968919e
It's only being used in a boolean fashion, so this patch replaces it with a
boolean.
--HG--
extra : rebase_source : 91152dff81107070fa49b3984e1b6759e0cd6d20
This change means that all the relevant code is now within
platform-linux-android.cpp, which is nice.
--HG--
extra : rebase_source : 886a31005fdb67fae65e6f4209796973f1391244
- Don't bother checking gSampler in ProfilerSignalHandler. It is equivalent
to checking gIsActive and we do that at the top of the loop in
SignalSender(). There is no point repeatedly checking the same condition in
the middle of that loop; that just opens up the possibility of partially
complete samples where some threads are missing.
- Clear gCurrentThreadInfo in SignalSender() instead of in
ProfilerSignalHandler(). The effect is much the same, but this change means
gCurrentThreadInfo is both set and cleared in SignalSender(), i.e. on a
single thread, removing any need for Atomic<>.
--HG--
extra : rebase_source : 645d321de4cad6fdb32383b6f1d0c7cbe54308fc
We don't need OS now that the platform-*.cpp files are in the same compilation
unit as platform.cpp.
The patch removes the sleep functions because they are unnecessary indirection.
OS::Startup() is necessary, but the patch renames it PlatformInit() to match
Platform{Start,Stop}() and profiler_init(), from which it is called.
Specifically:
- platform-linux.cc -> platform-linux-android.cpp
- platform-macos.cc -> platform-macos.cpp
- platform-win32.cc -> platform-win32.cpp
Adding "android" to the first one is the most important part, because it makes
things clearer. The .cc to .cpp change is less important but I might as well do
it while I'm in here.
--HG--
rename : tools/profiler/core/platform-linux.cc => tools/profiler/core/platform-linux-android.cpp
rename : tools/profiler/core/platform-macos.cc => tools/profiler/core/platform-macos.cpp
rename : tools/profiler/core/platform-win32.cc => tools/profiler/core/platform-win32.cpp
extra : rebase_source : 371f91c4cd95e88e1723e192e68f16ba66965c8f
Currently we use the SPS_* macros in some places, but also use other ones like
__arm__ and ANDROID and XP_{WIN,MAC,LINUX}. This patch makes the profiler
consistently use the SPS_* macros and removes the V8_HOST_ARCH_* macros.
The patch also does the following.
- Cleans up some header inclusions, e.g. including pthread.h directly in the
files that use it, and removing some unneeded android/log.h inclusions.
- Removes an unused branch in SetSampleContext() -- we don't support ARM on
anything other than Android, and glibc 2.3 is ancient.
- Doesn't use SPS_* in PseudoStack.h because that would require exporting
PlatformMacros.h, which doesn't seem worthwhile.
Some things that aid the understanding of this patch.
- XP_LINUX and LINUX are both defined for Linux *and* Android.
- x86/Android is the only supported platform that doesn't define
HAVE_NATIVE_UNWIND.
- Every platform that defines USE_LUL_STACKWALK also defines
HAVE_NATIVE_UNWIND.
--HG--
extra : rebase_source : 561b708f9434cabd9c0e00d4f4bfdd53f7008670
It's defined if any of XP_{WIN,MAC,LINUX} are defined and the latter includes
Android as well. So it's defined on all the OSes the profiler supports.
--HG--
extra : rebase_source : 1fa9c1fb573a99375b477a048c0b4575ac1eeca0
They must be set before the call to MaybeSetProfiler(), which checks
gProfileThreads, and before the gTaskTracer check.
This patch fixes the failure of some threads (e.g. Compositor) to be profiled,
as well as allowing TaskTracer to start up again!
This change increases consistency:
- Each OS is dealt with one at a time (no more interleaving).
- For each OS, x86 is now the first listed architecture.
The patch also adds the missing "#undef SPS_PLAT_x86_android".
--HG--
extra : rebase_source : 6ab278ff2523269424aa6dc872ec35ca6dc735a4
They duplicate the equivalent SPS_* macros. (The SPS_* macros have already
crept into use in some places within LUL.)
--HG--
extra : rebase_source : 65ed6e6e147189814511b0ca38342fec118478b9
Instead of calling MaybeShowStats() every 16th time around the inner loop we
now check every time around the outer loop. Because there are typically 20--50
threads running at once, this results in a slightly lower freqency of printing,
but that seems fine because this is debug-only code.
The new names make it clearer that these actions apply to just one thread.
- profiler_sleep_start() --> profiler_thread_sleep()
- profiler_sleep_end() --> profiler_thread_wake()
- profiler_is_sleeping() --> profiler_thread_is_sleeping()
- GeckoProfilerSleepRAII --> GeckoProfilerThreadSleepRAII
- GeckoProfilerWakeRAII --> GeckoProfilerThreadWakeRAII
This patch:
- Improves some comments.
- Adds some assertions.
- Removes unnecessary differences between some Mac and Windows code.
- Removes the unnecessary SamplerThread::mStackSize on Windows.
"unwinder" and "jank" are never checked for via hasFeature(), and are not
mentioned anywhere else in the profiler. This patch removes them from
profiler_get_features().
"restyle" *does* have a hasFeature() call and is handled by
profiler_feature_active(). This patch adds it to profiler_get_features().
Sleep handling is currently done with two counters (mSleepId, mSleepIdObserved)
and a boolean (mSleeping). This patch replaces it with a single tri-state
variable (mSleep), which is enough to replicate the functionality.
The patch also clearly documents how this variable is accessed from multiple
threads and the potential inaccuracies that can result from races.
This patch:
- Reformats PseudoStack.h to more closely follow Mozilla style.
- Rewrites some comments to make them more readable, e.g. by properly
delimiting sentences with upper-case letters and full stops.
- Replaces sMin() with std::min(), because <algorithm> no longer causes
problems.
- Reorders PseudoStack so that all the data members are at the end, which makes
them easier to see.
Sampler's constructor and destructor are inlined into profiler_{start,stop}(),
respectively. A trivial Sampler class is left behind (now in platform.cpp)
because there are still numerous gSampler null checks, which I will remove in a
follow-up bug.
PseudoStack::flushSamplerOnShutdown() moves into platform.cpp. MAXPATHLEN is
unused and is deleted.
Nicholas Nethercote
Marco Castelluccio
Wes Kocher
Jan de Mooij
Jeff Walden
Tom Tromey