Detect context switch callstack samples.

Context switch callstack samples are not included in the sampling data
statistics (as these don't represent random sampling), but are rather put into
a separate dedicated data structure.

For this to work, a complete context switch data has to be available for the
callstack timestamp. There is no guarantee it will be present at the time it
is needed, so a second structure is added to allow postponing qualification of
callstacks.

server/TracyEvent.hpp

@@ -666,9 +666,11 @@ struct ThreadData
     Vector<int64_t> childTimeStack;
     Vector<GhostZone> ghostZones;
     uint64_t ghostIdx;
+    Vector<SampleData> postponedSamples;
 #endif
     Vector<SampleData> samples;
     SampleData pendingSample;
+    Vector<SampleData> ctxSwitchSamples;
     uint64_t kernelSampleCnt;
     uint8_t isFiber;
     ThreadData* fiber;

server/TracyWorker.cpp

@@ -6162,7 +6162,43 @@ void Worker::ProcessCallstackSampleImpl( const SampleData& sd, ThreadData& td )
     }
 
 #ifndef TRACY_NO_STATISTICS
+    bool postpone = false;
+    auto ctx = GetContextSwitchData( td.id );
+    if( !ctx )
     {
+        postpone = true;
+    }
+    else
+    {
+        auto it = std::lower_bound( ctx->v.begin(), ctx->v.end(), sd.time.Val(), [] ( const auto& l, const auto& r ) { return (uint64_t)l.End() < (uint64_t)r; } );
+        if( it == ctx->v.end() )
+        {
+            postpone = true;
+        }
+        else if( sd.time.Val() == it->Start() )
+        {
+            td.ctxSwitchSamples.push_back( sd );
+        }
+        else
+        {
+            ProcessCallstackSampleImplStats( sd, td );
+        }
+    }
+    if( postpone )
+    {
+        td.postponedSamples.push_back( sd );
+    }
+#endif
+}
+
+#ifndef TRACY_NO_STATISTICS
+void Worker::ProcessCallstackSampleImplStats( const SampleData& sd, ThreadData& td )
+{
+    const auto t = sd.time.Val();
+    const auto callstack = sd.callstack.Val();
+    const auto& cs = GetCallstack( callstack );
+    const auto& ip = cs[0];
+
     uint16_t tid = CompressThread( td.id );
 
     auto frame = GetCallstackFrame( ip );
@@ -6225,7 +6261,7 @@ void Worker::ProcessCallstackSampleImpl( const SampleData& sd, ThreadData& td )
             sit->second.push_back_non_empty( SampleDataRange { sd.time, tid, ip } );
         }
     }
-    }
+
     for( uint16_t i=1; i<cs.size(); i++ )
     {
         auto addr = GetCanonicalPointer( cs[i] );
@@ -6251,8 +6287,8 @@ void Worker::ProcessCallstackSampleImpl( const SampleData& sd, ThreadData& td )
     {
         m_data.ghostZonesPostponed = true;
     }
-#endif
 }
+#endif
 
 void Worker::ProcessCallstackSample( const QueueCallstackSample& ev )
 {

server/TracyWorker.hpp

@@ -720,6 +720,9 @@ private:
     tracy_force_inline MemEvent* ProcessMemAllocImpl( uint64_t memname, MemData& memdata, const QueueMemAlloc& ev );
     tracy_force_inline MemEvent* ProcessMemFreeImpl( uint64_t memname, MemData& memdata, const QueueMemFree& ev );
     tracy_force_inline void ProcessCallstackSampleImpl( const SampleData& sd, ThreadData& td );
+#ifndef TRACY_NO_STATISTICS
+    tracy_force_inline void ProcessCallstackSampleImplStats( const SampleData& sd, ThreadData& td );
+#endif
 
     void ZoneStackFailure( uint64_t thread, const ZoneEvent* ev );
     void ZoneDoubleEndFailure( uint64_t thread, const ZoneEvent* ev );