Improve handling of post-load background jobs.

Background tasks (source location zones sorting, reconstruction of
memory plot) are now started only after trace loading is finished.
Multithreaded sorting was previously impacting trace load times.

Only one thread is used to perform both jobs, one after another.
This commit is contained in:
Bartosz Taudul 2019-02-14 01:17:37 +01:00
parent 080873003b
commit 92c1420c30
2 changed files with 26 additions and 24 deletions

View File

@ -766,22 +766,6 @@ Worker::Worker( FileRead& f, EventType::Type eventMask )
m_data.threads[i] = td; m_data.threads[i] = td;
} }
#ifndef TRACY_NO_STATISTICS
m_threadZones = std::thread( [this] {
for( auto& v : m_data.sourceLocationZones )
{
auto& zones = v.second.zones;
#ifdef MY_LIBCPP_SUCKS
pdqsort_branchless( zones.begin(), zones.end(), []( const auto& lhs, const auto& rhs ) { return lhs.zone->start < rhs.zone->start; } );
#else
std::sort( std::execution::par_unseq, zones.begin(), zones.end(), []( const auto& lhs, const auto& rhs ) { return lhs.zone->start < rhs.zone->start; } );
#endif
}
std::lock_guard<TracyMutex> lock( m_data.lock );
m_data.sourceLocationZonesReady = true;
} );
#endif
s_loadProgress.progress.store( LoadProgress::GpuZones, std::memory_order_relaxed ); s_loadProgress.progress.store( LoadProgress::GpuZones, std::memory_order_relaxed );
f.Read( sz ); f.Read( sz );
m_data.gpuData.reserve_exact( sz ); m_data.gpuData.reserve_exact( sz );
@ -1052,13 +1036,32 @@ Worker::Worker( FileRead& f, EventType::Type eventMask )
} }
finishLoading: finishLoading:
if( reconstructMemAllocPlot )
{
m_threadMemory = std::thread( [this] { ReconstructMemAllocPlot(); } );
}
s_loadProgress.total.store( 0, std::memory_order_relaxed ); s_loadProgress.total.store( 0, std::memory_order_relaxed );
m_loadTime = std::chrono::duration_cast<std::chrono::nanoseconds>( std::chrono::high_resolution_clock::now() - loadStart ).count(); m_loadTime = std::chrono::duration_cast<std::chrono::nanoseconds>( std::chrono::high_resolution_clock::now() - loadStart ).count();
#ifndef TRACY_NO_STATISTICS
m_threadBackground = std::thread( [this, reconstructMemAllocPlot] {
for( auto& v : m_data.sourceLocationZones )
{
auto& zones = v.second.zones;
#ifdef MY_LIBCPP_SUCKS
pdqsort_branchless( zones.begin(), zones.end(), []( const auto& lhs, const auto& rhs ) { return lhs.zone->start < rhs.zone->start; } );
#else
std::sort( std::execution::par_unseq, zones.begin(), zones.end(), []( const auto& lhs, const auto& rhs ) { return lhs.zone->start < rhs.zone->start; } );
#endif
}
{
std::lock_guard<TracyMutex> lock( m_data.lock );
m_data.sourceLocationZonesReady = true;
}
if( reconstructMemAllocPlot ) ReconstructMemAllocPlot();
} );
#else
if( reconstructMemAllocPlot )
{
m_threadBackground = std::thread( [this] { ReconstructMemAllocPlot(); } );
}
#endif
} }
Worker::~Worker() Worker::~Worker()
@ -1066,8 +1069,7 @@ Worker::~Worker()
Shutdown(); Shutdown();
if( m_thread.joinable() ) m_thread.join(); if( m_thread.joinable() ) m_thread.join();
if( m_threadMemory.joinable() ) m_threadMemory.join(); if( m_threadBackground.joinable() ) m_threadBackground.join();
if( m_threadZones.joinable() ) m_threadZones.join();
delete[] m_buffer; delete[] m_buffer;
LZ4_freeStreamDecode( m_stream ); LZ4_freeStreamDecode( m_stream );

View File

@ -416,7 +416,7 @@ private:
std::atomic<bool> m_hasData; std::atomic<bool> m_hasData;
std::atomic<bool> m_shutdown = { false }; std::atomic<bool> m_shutdown = { false };
std::thread m_threadMemory, m_threadZones; std::thread m_threadBackground;
int64_t m_delay; int64_t m_delay;
int64_t m_resolution; int64_t m_resolution;