Extract thread compression into a separate class.

capture/build/win32/capture.vcxproj

@@ -135,6 +135,7 @@
     <ClCompile Include="..\..\..\common\tracy_lz4hc.cpp" />
     <ClCompile Include="..\..\..\server\TracyMemory.cpp" />
     <ClCompile Include="..\..\..\server\TracyPrint.cpp" />
+    <ClCompile Include="..\..\..\server\TracyThreadCompress.cpp" />
     <ClCompile Include="..\..\..\server\TracyWorker.cpp" />
     <ClCompile Include="..\..\src\capture.cpp" />
     <ClCompile Include="..\..\src\getopt.c" />
@@ -159,6 +160,7 @@
     <ClInclude Include="..\..\..\server\TracyPopcnt.hpp" />
     <ClInclude Include="..\..\..\server\TracyPrint.hpp" />
     <ClInclude Include="..\..\..\server\TracySlab.hpp" />
+    <ClInclude Include="..\..\..\server\TracyThreadCompress.hpp" />
     <ClInclude Include="..\..\..\server\TracyVector.hpp" />
     <ClInclude Include="..\..\..\server\TracyWorker.hpp" />
     <ClInclude Include="..\..\..\server\tracy_flat_hash_map.hpp" />

capture/build/win32/capture.vcxproj.filters

@@ -39,6 +39,9 @@
     <ClCompile Include="..\..\..\server\TracyPrint.cpp">
       <Filter>server</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\..\server\TracyThreadCompress.cpp">
+      <Filter>server</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\..\..\common\tracy_lz4.hpp">
@@ -110,5 +113,8 @@
     <ClInclude Include="..\..\..\server\TracyPrint.hpp">
       <Filter>server</Filter>
     </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyThreadCompress.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
   </ItemGroup>
 </Project>

profiler/build/win32/Tracy.vcxproj

@@ -117,6 +117,7 @@
     <ClCompile Include="..\..\..\server\TracyPrint.cpp" />
     <ClCompile Include="..\..\..\server\TracyStorage.cpp" />
     <ClCompile Include="..\..\..\server\TracyTexture.cpp" />
+    <ClCompile Include="..\..\..\server\TracyThreadCompress.cpp" />
     <ClCompile Include="..\..\..\server\TracyUserData.cpp" />
     <ClCompile Include="..\..\..\server\TracyView.cpp" />
     <ClCompile Include="..\..\..\server\TracyWorker.cpp" />
@@ -172,6 +173,7 @@
     <ClInclude Include="..\..\..\server\TracyStorage.hpp" />
     <ClInclude Include="..\..\..\server\TracyStringDiscovery.hpp" />
     <ClInclude Include="..\..\..\server\TracyTexture.hpp" />
+    <ClInclude Include="..\..\..\server\TracyThreadCompress.hpp" />
     <ClInclude Include="..\..\..\server\TracyUserData.hpp" />
     <ClInclude Include="..\..\..\server\TracyVarArray.hpp" />
     <ClInclude Include="..\..\..\server\TracyVector.hpp" />

profiler/build/win32/Tracy.vcxproj.filters

@@ -105,6 +105,9 @@
     <ClCompile Include="..\..\..\server\TracyUserData.cpp">
       <Filter>server</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\..\server\TracyThreadCompress.cpp">
+      <Filter>server</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\..\..\common\tracy_lz4.hpp">
@@ -281,6 +284,9 @@
     <ClInclude Include="..\..\..\server\TracyUserData.hpp">
       <Filter>server</Filter>
     </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyThreadCompress.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <Natvis Include="DebugVis.natvis" />

server/TracyThreadCompress.cpp

@@ -0,0 +1,78 @@
+#include <limits>
+
+#include "TracyFileRead.hpp"
+#include "TracyFileWrite.hpp"
+#include "TracyThreadCompress.hpp"
+
+namespace tracy
+{
+
+ThreadCompress::ThreadCompress()
+    : m_threadLast( std::numeric_limits<uint64_t>::max(), 0 )
+{
+}
+
+void ThreadCompress::InitZero()
+{
+    assert( m_threadExpand.empty() );
+    m_threadExpand.push_back( 0 );
+}
+
+void ThreadCompress::Load( FileRead& f, int fileVer )
+{
+    assert( m_threadExpand.empty() );
+    assert( m_threadMap.empty() );
+
+    uint64_t sz;
+    if( fileVer >= FileVersion( 0, 4, 4 ) )
+    {
+        f.Read( sz );
+        m_threadExpand.reserve_and_use( sz );
+        f.Read( m_threadExpand.data(), sizeof( uint64_t ) * sz );
+        m_threadMap.reserve( sz );
+        for( size_t i=0; i<sz; i++ )
+        {
+            m_threadMap.emplace( m_threadExpand[i], i );
+        }
+    }
+    else
+    {
+        f.Read( sz );
+        m_threadExpand.reserve( sz );
+        m_threadExpand.push_back( 0 );
+    }
+}
+
+void ThreadCompress::Save( FileWrite& f ) const
+{
+    uint64_t sz = m_threadExpand.size();
+    f.Write( &sz, sizeof( sz ) );
+    f.Write( m_threadExpand.data(), sz * sizeof( uint64_t ) );
+}
+
+uint16_t ThreadCompress::CompressThreadReal( uint64_t thread )
+{
+    auto it = m_threadMap.find( thread );
+    if( it != m_threadMap.end() )
+    {
+        m_threadLast.first = thread;
+        m_threadLast.second = it->second;
+        return it->second;
+    }
+    else
+    {
+        return CompressThreadNew( thread );
+    }
+}
+
+uint16_t ThreadCompress::CompressThreadNew( uint64_t thread )
+{
+    auto sz = m_threadExpand.size();
+    m_threadExpand.push_back( thread );
+    m_threadMap.emplace( thread, sz );
+    m_threadLast.first = thread;
+    m_threadLast.second = sz;
+    return sz;
+}
+
+}

server/TracyThreadCompress.hpp

@@ -0,0 +1,61 @@
+#ifndef __TRACY__THREADCOMPRESS_HPP__
+#define __TRACY__THREADCOMPRESS_HPP__
+
+#include <assert.h>
+#include <stdint.h>
+
+#include "../common/TracyForceInline.hpp"
+#include "tracy_flat_hash_map.hpp"
+#include "TracyVector.hpp"
+
+namespace tracy
+{
+
+class FileRead;
+class FileWrite;
+
+class ThreadCompress
+{
+public:
+    ThreadCompress();
+
+    void InitZero();
+    void Load( FileRead& f, int fileVer );
+    void Save( FileWrite& f ) const;
+
+    tracy_force_inline uint16_t CompressThread( uint64_t thread )
+    {
+        if( m_threadLast.first == thread ) return m_threadLast.second;
+        return CompressThreadReal( thread );
+    }
+
+    tracy_force_inline uint64_t DecompressThread( uint16_t thread ) const
+    {
+        assert( thread < m_threadExpand.size() );
+        return m_threadExpand[thread];
+    }
+
+    tracy_force_inline uint16_t DecompressMustRaw( uint64_t thread ) const
+    {
+        auto it = m_threadMap.find( thread );
+        assert( it != m_threadMap.end() );
+        return it->second;
+    }
+
+    tracy_force_inline bool Exists( uint64_t thread ) const
+    {
+        return m_threadMap.find( thread ) != m_threadMap.end();
+    }
+
+private:
+    uint16_t CompressThreadReal( uint64_t thread );
+    uint16_t CompressThreadNew( uint64_t thread );
+
+    flat_hash_map<uint64_t, uint16_t, nohash<uint64_t>> m_threadMap;
+    Vector<uint64_t> m_threadExpand;
+    std::pair<uint64_t, uint16_t> m_threadLast;
+};
+
+}
+
+#endif

server/TracyWorker.cpp

@@ -250,7 +250,7 @@ Worker::Worker( const char* addr )
     , m_loadTime( 0 )
 {
     m_data.sourceLocationExpand.push_back( 0 );
-    m_data.threadExpand.push_back( 0 );
+    m_data.localThreadCompress.InitZero();
     m_data.callstackPayload.push_back( nullptr );
 
     memset( m_gpuCtxMap, 0, sizeof( m_gpuCtxMap ) );
@@ -515,23 +515,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask )
         }
     }
 
-    if( fileVer >= FileVersion( 0, 4, 4 ) )
+    m_data.localThreadCompress.Load( f, fileVer );
-    {
-        f.Read( sz );
-        m_data.threadExpand.reserve_and_use( sz );
-        f.Read( m_data.threadExpand.data(), sizeof( uint64_t ) * sz );
-        m_data.threadMap.reserve( sz );
-        for( size_t i=0; i<sz; i++ )
-        {
-            m_data.threadMap.emplace( m_data.threadExpand[i], i );
-        }
-    }
-    else
-    {
-        f.Read( sz );
-        m_data.threadExpand.reserve( sz );
-        m_data.threadExpand.push_back( 0 );
-    }
 
     f.Read( sz );
     for( uint64_t i=0; i<sz; i++ )
@@ -1548,9 +1532,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask )
             if( !t->timeline.empty() )
             {
                 // Don't touch thread compression cache in a thread.
-                auto it = m_data.threadMap.find( t->id );
+                ProcessTimeline( t->timeline, m_data.localThreadCompress.DecompressMustRaw( t->id ) );
-                assert( it != m_data.threadMap.end() );
-                ProcessTimeline( t->timeline, it->second );
             }
         }
         for( auto& v : m_data.sourceLocationZones )
@@ -1881,7 +1863,7 @@ const char* Worker::GetThreadString( uint64_t id ) const
 
 bool Worker::IsThreadLocal( uint64_t id ) const
 {
-    return m_data.threadMap.find( id ) != m_data.threadMap.end();
+    return m_data.localThreadCompress.Exists( id );
 }
 
 const SourceLocation& Worker::GetSourceLocation( int16_t srcloc ) const
@@ -2040,31 +2022,6 @@ const Worker::SourceLocationZones& Worker::GetZonesForSourceLocation( int16_t sr
 }
 #endif
 
-uint16_t Worker::CompressThreadReal( uint64_t thread )
-{
-    auto it = m_data.threadMap.find( thread );
-    if( it != m_data.threadMap.end() )
-    {
-        m_data.threadLast.first = thread;
-        m_data.threadLast.second = it->second;
-        return it->second;
-    }
-    else
-    {
-        return CompressThreadNew( thread );
-    }
-}
-
-uint16_t Worker::CompressThreadNew( uint64_t thread )
-{
-    auto sz = m_data.threadExpand.size();
-    m_data.threadExpand.push_back( thread );
-    m_data.threadMap.emplace( thread, sz );
-    m_data.threadLast.first = thread;
-    m_data.threadLast.second = sz;
-    return sz;
-}
-
 void Worker::Exec()
 {
     auto ShouldExit = [this]
@@ -4828,9 +4785,7 @@ void Worker::Write( FileWrite& f )
         f.Write( &ptr, sizeof( ptr ) );
     }
 
-    sz = m_data.threadExpand.size();
+    m_data.localThreadCompress.Save( f );
-    f.Write( &sz, sizeof( sz ) );
-    f.Write( m_data.threadExpand.data(), sz * sizeof( uint64_t ) );
 
     sz = m_data.sourceLocation.size();
     f.Write( &sz, sizeof( sz ) );
@@ -5044,7 +4999,7 @@ void Worker::Write( FileWrite& f )
     ctxValid.reserve( m_data.ctxSwitch.size() );
     for( auto it = m_data.ctxSwitch.begin(); it != m_data.ctxSwitch.end(); ++it )
     {
-        if( m_data.threadMap.find( it->first ) != m_data.threadMap.end() )
+        if( m_data.localThreadCompress.Exists( it->first ) )
         {
             ctxValid.emplace_back( it );
         }

server/TracyWorker.hpp

@@ -18,6 +18,7 @@
 #include "TracyEvent.hpp"
 #include "TracySlab.hpp"
 #include "TracyStringDiscovery.hpp"
+#include "TracyThreadCompress.hpp"
 #include "TracyVarArray.hpp"
 
 namespace tracy
@@ -151,7 +152,6 @@ private:
             , baseTime( 0 )
             , lastTime( 0 )
             , frameOffset( 0 )
-            , threadLast( std::numeric_limits<uint64_t>::max(), 0 )
             , threadDataLast( std::numeric_limits<uint64_t>::max(), nullptr )
             , ctxSwitchLast( std::numeric_limits<uint64_t>::max(), nullptr )
         {}
@@ -193,9 +193,7 @@ private:
 
         flat_hash_map<uint32_t, LockMap*, nohash<uint32_t>> lockMap;
 
-        flat_hash_map<uint64_t, uint16_t, nohash<uint64_t>> threadMap;
+        ThreadCompress localThreadCompress;
-        Vector<uint64_t> threadExpand;
-        std::pair<uint64_t, uint16_t> threadLast;
         std::pair<uint64_t, ThreadData*> threadDataLast;
 
         Vector<Vector<ZoneEvent*>> zoneChildren;
@@ -359,12 +357,8 @@ public:
     bool AreSourceLocationZonesReady() const { return m_data.sourceLocationZonesReady; }
 #endif
 
-    tracy_force_inline uint16_t CompressThread( uint64_t thread )
+    tracy_force_inline uint16_t CompressThread( uint64_t thread ) { return m_data.localThreadCompress.CompressThread( thread ); }
-    {
+    tracy_force_inline uint64_t DecompressThread( uint16_t thread ) const { return m_data.localThreadCompress.DecompressThread( thread ); }
-        if( m_data.threadLast.first == thread ) return m_data.threadLast.second;
-        return CompressThreadReal( thread );
-    }
-    tracy_force_inline uint64_t DecompressThread( uint16_t thread ) const { assert( thread < m_data.threadExpand.size() ); return m_data.threadExpand[thread]; }
 
     std::shared_mutex& GetMbpsDataLock() { return m_mbpsData.lock; }
     const std::vector<float>& GetMbpsData() const { return m_mbpsData.mbps; }
@@ -508,8 +502,6 @@ private:
     void HandlePostponedPlots();
 
     StringLocation StoreString( char* str, size_t sz );
-    uint16_t CompressThreadReal( uint64_t thread );
-    uint16_t CompressThreadNew( uint64_t thread );
     const ContextSwitch* const GetContextSwitchDataImpl( uint64_t thread );
 
     tracy_force_inline void ReadTimeline( FileRead& f, ZoneEvent* zone, uint16_t thread, int64_t& refTime );

update/build/win32/update.vcxproj

@@ -134,6 +134,7 @@
     <ClCompile Include="..\..\..\common\tracy_lz4.cpp" />
     <ClCompile Include="..\..\..\common\tracy_lz4hc.cpp" />
     <ClCompile Include="..\..\..\server\TracyMemory.cpp" />
+    <ClCompile Include="..\..\..\server\TracyThreadCompress.cpp" />
     <ClCompile Include="..\..\..\server\TracyWorker.cpp" />
     <ClCompile Include="..\..\src\update.cpp" />
   </ItemGroup>
@@ -156,6 +157,7 @@
     <ClInclude Include="..\..\..\server\TracyMemory.hpp" />
     <ClInclude Include="..\..\..\server\TracyPopcnt.hpp" />
     <ClInclude Include="..\..\..\server\TracySlab.hpp" />
+    <ClInclude Include="..\..\..\server\TracyThreadCompress.hpp" />
     <ClInclude Include="..\..\..\server\TracyVector.hpp" />
     <ClInclude Include="..\..\..\server\TracyWorker.hpp" />
     <ClInclude Include="..\..\..\server\tracy_flat_hash_map.hpp" />

update/build/win32/update.vcxproj.filters

@@ -33,6 +33,9 @@
     <ClCompile Include="..\..\..\common\tracy_lz4hc.cpp">
       <Filter>common</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\..\server\TracyThreadCompress.cpp">
+      <Filter>server</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\..\..\common\tracy_lz4.hpp">
@@ -98,5 +101,8 @@
     <ClInclude Include="..\..\..\common\tracy_lz4hc.hpp">
       <Filter>common</Filter>
     </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyThreadCompress.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
   </ItemGroup>
 </Project>