Reduce number of allocations in AutoStopwatch

This patch fixes two related issues.
1. The AutoStopwatch uses a stack-allocated `mozilla::Vector` to
communicate with its callback during each compartment switch.
This vector was designed to allow its contents to be stack-allocated
but they turned out to be accidentally heap-allocated.

2. During each tick, the stopwatch fills a vector `recentGroups_`.
This vector always started with minimal capacity and had to grow
repeatedly as groups were added, causing repeated reallocations.
This patch preallocates `recentGroups_` to have the same capacity as the
previous tick. We expect that this should eventually reach a stable size
that closely matches the actual needs of the process.

js/src/jsapi.h

@@ -6566,7 +6566,7 @@ struct JS_PUBLIC_API(PerformanceGroup) {
     uint64_t refCount_;
 };
 
-using PerformanceGroupVector = mozilla::Vector<RefPtr<js::PerformanceGroup>, 0, SystemAllocPolicy>;
+using PerformanceGroupVector = mozilla::Vector<RefPtr<js::PerformanceGroup>, 8, SystemAllocPolicy>;
 
 /**
  * Commit any Performance Monitoring data.

js/src/vm/Stopwatch.cpp

@@ -20,6 +20,7 @@
 #include "gc/Zone.h"
 #include "vm/Runtime.h"
 
+
 namespace js {
 
 bool
@@ -136,6 +137,9 @@ PerformanceMonitoring::start()
 bool
 PerformanceMonitoring::commit()
 {
+    // Maximal initialization size, in elements for the vector of groups.
+    static const size_t MAX_GROUPS_INIT_CAPACITY = 1024;
+
 #if !defined(MOZ_HAVE_RDTSC)
     // The AutoStopwatch is only executed if `MOZ_HAVE_RDTSC`.
     return false;
@@ -152,13 +156,24 @@ PerformanceMonitoring::commit()
         return true;
     }
 
-    PerformanceGroupVector recentGroups;
+    // The move operation is generally constant time, unless
-    recentGroups_.swap(recentGroups);
+    // `recentGroups_.length()` is very small, in which case
+    // it's fast just because it's small.
+    PerformanceGroupVector recentGroups(Move(recentGroups_));
+    recentGroups_ = PerformanceGroupVector(); // Reconstruct after `Move`.
 
     bool success = true;
     if (stopwatchCommitCallback)
         success = stopwatchCommitCallback(iteration_, recentGroups, stopwatchCommitClosure);
 
+    // Heuristic: we expect to have roughly the same number of groups as in
+    // the previous iteration.
+    const size_t capacity = recentGroups.capacity() < MAX_GROUPS_INIT_CAPACITY ?
+                            recentGroups.capacity() :
+                            MAX_GROUPS_INIT_CAPACITY;
+    success = recentGroups_.reserve(capacity)
+            && success;
+
     // Reset immediately, to make sure that we're not hit by the end
     // of a nested event loop (which would cause `commit` to be called
     // twice in succession).
@@ -227,7 +242,7 @@ AutoStopwatch::AutoStopwatch(JSContext* cx MOZ_GUARD_OBJECT_NOTIFIER_PARAM_IN_IM
     MOZ_GUARD_OBJECT_NOTIFIER_INIT;
 
     JSCompartment* compartment = cx_->compartment();
-    if (compartment->scheduledForDestruction)
+    if (MOZ_UNLIKELY(compartment->scheduledForDestruction))
         return;
 
     JSRuntime* runtime = cx_->runtime();
@@ -266,11 +281,11 @@ AutoStopwatch::~AutoStopwatch()
     }
 
     JSCompartment* compartment = cx_->compartment();
-    if (compartment->scheduledForDestruction)
+    if (MOZ_UNLIKELY(compartment->scheduledForDestruction))
         return;
 
     JSRuntime* runtime = cx_->runtime();
-    if (iteration_ != runtime->performanceMonitoring.iteration()) {
+    if (MOZ_UNLIKELY(iteration_ != runtime->performanceMonitoring.iteration())) {
         // We have entered a nested event loop at some point.
         // Any information we may have is obsolete.
         return;

toolkit/components/perfmonitoring/nsPerformanceStats.cpp

@@ -1082,6 +1082,9 @@ nsPerformanceStatsService::GetPerformanceGroups(JSContext* cx,
     return false;
   }
 
+  // Returning a vector that is too large would cause allocations all over the
+  // place in the JS engine. We want to be sure that all data is stored inline.
+  MOZ_ASSERT(out.length() <= out.sMaxInlineStorage);
   return true;
 }
 
@@ -1310,8 +1313,12 @@ nsPerformanceStatsService::GetResources(uint64_t* userTime,
 
 void
 nsPerformanceStatsService::NotifyJankObservers(const mozilla::Vector<uint64_t>& aPreviousJankLevels) {
-  GroupVector alerts;
+
-  mPendingAlerts.swap(alerts);
+  // The move operation is generally constant time, unless
+  // `mPendingAlerts.length()` is very small, in which case it's fast anyway.
+  GroupVector alerts(Move(mPendingAlerts));
+  mPendingAlerts = GroupVector(); // Reconstruct after `Move`.
+
   if (!mPendingAlertsCollector) {
     // We are shutting down.
     return;

toolkit/components/perfmonitoring/nsPerformanceStats.h

@@ -19,7 +19,7 @@
 class nsPerformanceGroup;
 class nsPerformanceGroupDetails;
 
-typedef mozilla::Vector<RefPtr<nsPerformanceGroup>> GroupVector;
+typedef mozilla::Vector<RefPtr<nsPerformanceGroup>, 8> GroupVector;
 
 /**
  * A data structure for registering observers interested in