From 7b023e533dec469ac98a84a04605797d323845aa Mon Sep 17 00:00:00 2001 From: Bartosz Taudul Date: Fri, 15 Feb 2019 01:58:23 +0100 Subject: [PATCH] Use big allocation mode for Vector's reserve_exact. --- server/TracyVector.hpp | 15 ++++++--------- server/TracyWorker.cpp | 38 +++++++++++++++++++------------------- 2 files changed, 25 insertions(+), 28 deletions(-) diff --git a/server/TracyVector.hpp b/server/TracyVector.hpp index a565259d..74bb8c91 100644 --- a/server/TracyVector.hpp +++ b/server/TracyVector.hpp @@ -9,6 +9,7 @@ #include "../common/TracyForceInline.hpp" #include "TracyMemory.hpp" #include "TracyPopcnt.hpp" +#include "TracySlab.hpp" namespace tracy { @@ -46,15 +47,11 @@ public: ~Vector() { - if( m_capacity == std::numeric_limits::max() ) - { - memUsage.fetch_sub( m_size * sizeof( T ) ); - } - else + if( m_capacity != std::numeric_limits::max() ) { memUsage.fetch_sub( Capacity() * sizeof( T ), std::memory_order_relaxed ); + delete[] m_ptr; } - delete[] m_ptr; } Vector& operator=( const Vector& ) = delete; @@ -234,13 +231,13 @@ public: m_size = sz; } - tracy_force_inline void reserve_exact( uint32_t sz ) + template + tracy_force_inline void reserve_exact( uint32_t sz, Slab& slab ) { assert( !m_ptr ); m_capacity = std::numeric_limits::max(); m_size = sz; - m_ptr = new T[sz]; - memUsage.fetch_add( sz * sizeof( T ) ); + m_ptr = (T*)slab.AllocBig( sizeof( T ) * sz ); } tracy_force_inline void clear() diff --git a/server/TracyWorker.cpp b/server/TracyWorker.cpp index 54336668..93912437 100644 --- a/server/TracyWorker.cpp +++ b/server/TracyWorker.cpp @@ -344,7 +344,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) if( fileVer >= FileVersion( 0, 3, 202 ) ) { f.Read( sz ); - m_data.frames.Data().reserve_exact( sz ); + m_data.frames.Data().reserve_exact( sz, m_slab ); for( uint64_t i=0; i(); @@ -352,7 +352,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) f.Read( &ptr->continuous, sizeof( ptr->continuous ) ); uint64_t fsz; f.Read( &fsz, sizeof( fsz ) ); - ptr->frames.reserve_exact( fsz ); + ptr->frames.reserve_exact( fsz, m_slab ); if( fileVer >= FileVersion( 0, 4, 2 ) ) { int64_t refTime = 0; @@ -400,7 +400,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) ptr->name = 0; ptr->continuous = 1; f.Read( sz ); - ptr->frames.reserve_exact( sz ); + ptr->frames.reserve_exact( sz, m_slab ); for( uint64_t i=0; iframes[i].start, sizeof( int64_t ) ); @@ -413,7 +413,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) flat_hash_map> pointerMap; f.Read( sz ); - m_data.stringData.reserve_exact( sz ); + m_data.stringData.reserve_exact( sz, m_slab ); for( uint64_t i=0; i(); @@ -566,7 +566,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) lockmap.threadList.emplace_back( t ); } f.Read( tsz ); - lockmap.timeline.reserve_exact( tsz ); + lockmap.timeline.reserve_exact( tsz, m_slab ); auto ptr = lockmap.timeline.data(); if( fileVer >= FileVersion( 0, 4, 2 ) ) { @@ -677,7 +677,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) f.Read( sz ); if( eventMask & EventType::Messages ) { - m_data.messages.reserve_exact( sz ); + m_data.messages.reserve_exact( sz, m_slab ); if( fileVer >= FileVersion( 0, 4, 2 ) ) { int64_t refTime = 0; @@ -721,7 +721,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) s_loadProgress.progress.store( LoadProgress::Zones, std::memory_order_relaxed ); f.Read( sz ); - m_data.threads.reserve_exact( sz ); + m_data.threads.reserve_exact( sz, m_slab ); for( uint64_t i=0; i(); @@ -749,7 +749,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) f.Read( msz ); if( eventMask & EventType::Messages ) { - td->messages.reserve_exact( msz ); + td->messages.reserve_exact( msz, m_slab ); for( uint64_t j=0; j(); @@ -825,7 +825,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) f.Read( pd->max ); uint64_t psz; f.Read( psz ); - pd->data.reserve_exact( psz ); + pd->data.reserve_exact( psz, m_slab ); if( fileVer >= FileVersion( 0, 4, 2 ) ) { int64_t refTime = 0; @@ -863,13 +863,13 @@ Worker::Worker( FileRead& f, EventType::Type eventMask ) f.Read( sz ); if( eventMask & EventType::Memory ) { - m_data.memory.data.reserve_exact( sz ); + m_data.memory.data.reserve_exact( sz, m_slab ); if( fileVer >= FileVersion( 0, 3, 201 ) ) { uint64_t activeSz, freesSz; f.Read2( activeSz, freesSz ); m_data.memory.active.reserve( activeSz ); - m_data.memory.frees.reserve_exact( freesSz ); + m_data.memory.frees.reserve_exact( freesSz, m_slab ); } auto mem = m_data.memory.data.data(); s_loadProgress.subTotal.store( sz, std::memory_order_relaxed ); @@ -3124,7 +3124,7 @@ void Worker::ReconstructMemAllocPlot() plot->name = 0; plot->type = PlotType::Memory; - plot->data.reserve_exact( psz ); + plot->data.reserve_exact( psz, m_slab ); auto aptr = mem.data.begin(); auto aend = mem.data.end(); @@ -3322,7 +3322,7 @@ void Worker::ReadTimelineUpdateStatistics( ZoneEvent* zone, uint16_t thread ) void Worker::ReadTimeline( FileRead& f, Vector& vec, uint16_t thread, uint64_t size, int64_t& refTime ) { assert( size != 0 ); - vec.reserve_exact( size ); + vec.reserve_exact( size, m_slab ); m_data.zonesCnt += size; for( uint64_t i=0; i& vec, uint16_t thread void Worker::ReadTimelinePre042( FileRead& f, Vector& vec, uint16_t thread, uint64_t size, int fileVer ) { assert( size != 0 ); - vec.reserve_exact( size ); + vec.reserve_exact( size, m_slab ); m_data.zonesCnt += size; for( uint64_t i=0; i& vec, uint16_t void Worker::ReadTimeline( FileRead& f, Vector& vec, uint64_t size, int64_t& refTime, int64_t& refGpuTime ) { assert( size != 0 ); - vec.reserve_exact( size ); + vec.reserve_exact( size, m_slab ); for( uint64_t i=0; i& vec, uint64_t size, i void Worker::ReadTimelinePre042( FileRead& f, Vector& vec, uint64_t size, int fileVer ) { assert( size != 0 ); - vec.reserve_exact( size ); + vec.reserve_exact( size, m_slab ); for( uint64_t i=0; i