Pack StringIdx in 24 bits.

This reduces ZoneEvent size from 32 to 30 bytes.

Memory usage reduction on selected traces (sizes in MB):

big             9902 -> 9527  (96%)
chicken         2172 -> 2107  (97%)
ctx-big          311 ->  309  (99%)
drl-l-b         1570 -> 1479  (94%)
long            5496 -> 5412  (98%)
mem             6468 -> 6468  (100%)
q3bsp-mt        5784 -> 5592  (96%)
selfprofile     1486 -> 1443  (97%)
This commit is contained in:
Bartosz Taudul 2019-09-29 20:32:42 +02:00
parent 781ebeb835
commit d228bcb622
5 changed files with 159 additions and 55 deletions

View File

@ -41,27 +41,45 @@ struct StringRef
};
};
struct StringIdx
class StringIdx
{
StringIdx() : __data( 0 ) {}
StringIdx( uint32_t _idx )
: __data( 0 )
public:
StringIdx() { memset( m_idx, 0, sizeof( m_idx ) ); }
StringIdx( uint32_t idx )
{
idx = _idx;
active = 1;
SetIdx( idx );
}
union
void SetIdx( uint32_t idx )
{
struct
{
uint32_t idx : 31;
uint32_t active : 1;
};
uint32_t __data;
};
idx++;
memcpy( m_idx, &idx, 3 );
}
uint32_t Idx() const
{
uint32_t idx = 0;
memcpy( &idx, m_idx, 3 );
return idx - 1;
}
bool Active() const
{
uint32_t zero = 0;
return memcmp( m_idx, &zero, 3 ) != 0;
}
private:
uint8_t m_idx[3];
};
struct __StringIdxOld
{
uint32_t idx : 31;
uint32_t active : 1;
};
struct SourceLocation
{
StringRef name;

View File

@ -7,7 +7,7 @@ namespace Version
{
enum { Major = 0 };
enum { Minor = 5 };
enum { Patch = 7 };
enum { Patch = 8 };
}
}

View File

@ -2996,7 +2996,7 @@ int View::DrawZoneLevel( const Vector<ZoneEvent*>& vec, bool hover, double pxns,
{
const auto color = GetZoneColor( ev, tid, depth );
const char* zoneName = m_worker.GetZoneName( ev );
int dmul = ev.text.active ? 2 : 1;
int dmul = ev.text.Active() ? 2 : 1;
if( ev.child >= 0 )
{
@ -5336,7 +5336,7 @@ void View::DrawZoneInfoWindow()
auto threadData = GetZoneThreadData( ev );
assert( threadData );
const auto tid = threadData->id;
if( ev.name.active )
if( ev.name.Active() )
{
if( m_bigFont ) ImGui::PushFont( m_bigFont );
TextFocused( "Zone name:", m_worker.GetString( ev.name ) );
@ -5371,7 +5371,7 @@ void View::DrawZoneInfoWindow()
ImGui::TextDisabled( "(%s)", RealToString( tid, true ) );
ImGui::SameLine();
SmallColorBox( GetThreadColor( tid, 0 ) );
if( ev.text.active )
if( ev.text.Active() )
{
TextFocused( "User text:", m_worker.GetString( ev.text ) );
dmul++;
@ -7237,7 +7237,7 @@ uint64_t View::GetSelectionTarget( const Worker::ZoneThreadData& ev, FindZone::G
case FindZone::GroupBy::Thread:
return ev.Thread();
case FindZone::GroupBy::UserText:
return ev.Zone()->text.active ? ev.Zone()->text.idx : std::numeric_limits<uint64_t>::max();
return ev.Zone()->text.Active() ? ev.Zone()->text.Idx() : std::numeric_limits<uint64_t>::max();
case FindZone::GroupBy::Callstack:
return ev.Zone()->callstack;
default:
@ -8265,7 +8265,7 @@ void View::DrawFindZone()
group = &m_findZone.groups[ev.Thread()];
break;
case FindZone::GroupBy::UserText:
group = &m_findZone.groups[ev.Zone()->text.active ? ev.Zone()->text.idx : std::numeric_limits<uint64_t>::max()];
group = &m_findZone.groups[ev.Zone()->text.Active() ? ev.Zone()->text.Idx() : std::numeric_limits<uint64_t>::max()];
break;
case FindZone::GroupBy::Callstack:
group = &m_findZone.groups[ev.Zone()->callstack];
@ -8548,7 +8548,7 @@ void View::DrawZoneList( const Vector<ZoneEvent*>& zones )
break;
case FindZone::TableSortBy::Name:
pdqsort_branchless( sortedZones.begin(), sortedZones.end(), [this]( const auto& lhs, const auto& rhs ) {
if( lhs->name.active != rhs->name.active ) return lhs->name.active > rhs->name.active;
if( lhs->name.Active() != rhs->name.Active() ) return lhs->name.Active() > rhs->name.Active();
return strcmp( m_worker.GetString( lhs->name ), m_worker.GetString( rhs->name ) ) < 0;
} );
break;
@ -8592,7 +8592,7 @@ void View::DrawZoneList( const Vector<ZoneEvent*>& zones )
ImGui::NextColumn();
ImGui::TextUnformatted( TimeToString( timespan ) );
ImGui::NextColumn();
if( ev->name.active )
if( ev->name.Active() )
{
ImGui::TextUnformatted( m_worker.GetString( ev->name ) );
}
@ -11508,7 +11508,7 @@ static tracy_force_inline CallstackFrameTree* GetFrameTreeItemGroup( flat_hash_m
{
auto& frameData = *worker.GetCallstackFrame( idx );
auto& frame = frameData.data[frameData.size-1];
auto fidx = frame.name.idx;
auto fidx = frame.name.Idx();
auto it = tree.find( fidx );
if( it == tree.end() )
@ -12562,7 +12562,7 @@ void View::ZoneTooltip( const ZoneEvent& ev )
const auto selftime = GetZoneSelfTime( ev );
ImGui::BeginTooltip();
if( ev.name.active )
if( ev.name.Active() )
{
ImGui::TextUnformatted( m_worker.GetString( ev.name ) );
}
@ -12610,7 +12610,7 @@ void View::ZoneTooltip( const ZoneEvent& ev )
TextFocused( "Running state regions:", RealToString( cnt, true ) );
}
}
if( ev.text.active )
if( ev.text.Active() )
{
ImGui::NewLine();
TextColoredUnformatted( ImVec4( 0xCC / 255.f, 0xCC / 255.f, 0x22 / 255.f, 1.f ), m_worker.GetString( ev.text ) );

View File

@ -911,10 +911,10 @@ Worker::Worker( FileRead& f, EventType::Type eventMask )
{
ReadTimelinePre042( f, td->timeline, CompressThread( tid ), tsz, fileVer );
}
else if( fileVer <= FileVersion( 0, 5, 1 ) )
else if( fileVer <= FileVersion( 0, 5, 7 ) )
{
int64_t refTime = 0;
ReadTimelinePre052( f, td->timeline, CompressThread( tid ), tsz, refTime, fileVer );
ReadTimelinePre058( f, td->timeline, CompressThread( tid ), tsz, refTime, fileVer );
}
else
{
@ -1298,7 +1298,7 @@ Worker::Worker( FileRead& f, EventType::Type eventMask )
}
}
if( fileVer >= FileVersion( 0, 4, 6 ) )
if( fileVer >= FileVersion( 0, 5, 8 ) )
{
f.Read( sz );
m_data.callstackFrameMap.reserve( sz );
@ -1316,20 +1316,54 @@ Worker::Worker( FileRead& f, EventType::Type eventMask )
m_data.callstackFrameMap.emplace( id, frameData );
}
}
else if( fileVer >= FileVersion( 0, 4, 6 ) )
{
f.Read( sz );
m_data.callstackFrameMap.reserve( sz );
for( uint64_t i=0; i<sz; i++ )
{
__StringIdxOld str;
CallstackFrameId id;
f.Read( id );
auto frameData = m_slab.Alloc<CallstackFrameData>();
f.Read( frameData->size );
frameData->data = m_slab.AllocInit<CallstackFrame>( frameData->size );
for( uint8_t j=0; j<frameData->size; j++ )
{
f.Read( str );
if( str.active ) frameData->data[j].name.SetIdx( str.idx );
f.Read( str );
if( str.active ) frameData->data[j].file.SetIdx( str.idx );
f.Read( frameData->data[j].line );
}
m_data.callstackFrameMap.emplace( id, frameData );
}
}
else if( fileVer >= FileVersion( 0, 4, 3 ) )
{
f.Read( sz );
m_data.callstackFrameMap.reserve( sz );
for( uint64_t i=0; i<sz; i++ )
{
__StringIdxOld str;
uint64_t ptr;
f.Read( ptr );
auto frameData = m_slab.Alloc<CallstackFrameData>();
f.Read( frameData->size );
frameData->data = m_slab.Alloc<CallstackFrame>( frameData->size );
f.Read( frameData->data, sizeof( CallstackFrame ) * frameData->size );
frameData->data = m_slab.AllocInit<CallstackFrame>( frameData->size );
for( uint8_t j=0; j<frameData->size; j++ )
{
f.Read( str );
if( str.active ) frameData->data[j].name.SetIdx( str.idx );
f.Read( str );
if( str.active ) frameData->data[j].file.SetIdx( str.idx );
f.Read( frameData->data[j].line );
}
m_data.callstackFrameMap.emplace( PackPointer( ptr ), frameData );
}
@ -1340,14 +1374,19 @@ Worker::Worker( FileRead& f, EventType::Type eventMask )
m_data.callstackFrameMap.reserve( sz );
for( uint64_t i=0; i<sz; i++ )
{
__StringIdxOld str;
uint64_t ptr;
f.Read( ptr );
auto frameData = m_slab.Alloc<CallstackFrameData>();
frameData->size = 1;
frameData->data = m_slab.Alloc<CallstackFrame>();
f.Read( frameData->data, sizeof( CallstackFrame ) );
frameData->data = m_slab.AllocInit<CallstackFrame>();
f.Read( str );
if( str.active ) frameData->data->name.SetIdx( str.idx );
f.Read( str );
if( str.active ) frameData->data->file.SetIdx( str.idx );
f.Read( frameData->data->line );
m_data.callstackFrameMap.emplace( PackPointer( ptr ), frameData );
}
@ -1931,8 +1970,8 @@ const char* Worker::GetString( const StringRef& ref ) const
const char* Worker::GetString( const StringIdx& idx ) const
{
assert( idx.active );
return m_data.stringData[idx.idx];
assert( idx.Active() );
return m_data.stringData[idx.Idx()];
}
static const char* BadExternalThreadNames[] = {
@ -2030,7 +2069,7 @@ const char* Worker::GetZoneName( const ZoneEvent& ev ) const
const char* Worker::GetZoneName( const ZoneEvent& ev, const SourceLocation& srcloc ) const
{
if( ev.name.active )
if( ev.name.Active() )
{
return GetString( ev.name );
}
@ -4537,7 +4576,7 @@ void Worker::ReadTimelinePre042( FileRead& f, ZoneEvent* zone, uint16_t thread,
}
}
void Worker::ReadTimelinePre052( FileRead& f, ZoneEvent* zone, uint16_t thread, int64_t& refTime, int fileVer )
void Worker::ReadTimelinePre058( FileRead& f, ZoneEvent* zone, uint16_t thread, int64_t& refTime, int fileVer )
{
uint64_t sz;
f.Read( sz );
@ -4550,7 +4589,7 @@ void Worker::ReadTimelinePre052( FileRead& f, ZoneEvent* zone, uint16_t thread,
zone->child = m_data.zoneChildren.size();
m_data.zoneChildren.push_back( Vector<ZoneEvent*>() );
Vector<ZoneEvent*> tmp;
ReadTimelinePre052( f, tmp, thread, sz, refTime, fileVer );
ReadTimelinePre058( f, tmp, thread, sz, refTime, fileVer );
m_data.zoneChildren[zone->child] = std::move( tmp );
}
}
@ -4645,7 +4684,7 @@ void Worker::ReadTimeline( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t thread
do
{
s_loadProgress.subProgress.fetch_add( 1, std::memory_order_relaxed );
uint16_t srcloc;
int16_t srcloc;
f.Read( srcloc );
zone->SetSrcLoc( srcloc );
// Use zone->end as scratch buffer for zone start time offset.
@ -4682,7 +4721,26 @@ void Worker::ReadTimelinePre042( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t
f.Read( srcloc );
zone->SetSrcLoc( srcloc );
f.Skip( 4 );
f.Read( &zone->text, sizeof( zone->text ) + sizeof( zone->callstack ) + sizeof( zone->name ) );
__StringIdxOld str;
f.Read( str );
if( str.active )
{
zone->text.SetIdx( str.idx );
}
else
{
new ( &zone->text ) StringIdx();
}
f.Read( zone->callstack );
f.Read( str );
if( str.active )
{
zone->name.SetIdx( str.idx );
}
else
{
new ( &zone->name ) StringIdx();
}
ReadTimelinePre042( f, zone, thread, fileVer );
#ifdef TRACY_NO_STATISTICS
ReadTimelineUpdateStatistics( zone, thread );
@ -4690,9 +4748,9 @@ void Worker::ReadTimelinePre042( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t
}
}
void Worker::ReadTimelinePre052( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t thread, uint64_t size, int64_t& refTime, int fileVer )
void Worker::ReadTimelinePre058( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t thread, uint64_t size, int64_t& refTime, int fileVer )
{
assert( fileVer <= FileVersion( 0, 5, 1 ) );
assert( fileVer <= FileVersion( 0, 5, 7 ) );
assert( size != 0 );
vec.reserve_exact( size, m_slab );
m_data.zonesCnt += size;
@ -4706,23 +4764,51 @@ void Worker::ReadTimelinePre052( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t
do
{
s_loadProgress.subProgress.fetch_add( 1, std::memory_order_relaxed );
// Use zone->end as scratch buffer for zone start time offset.
f.Read( &zone->end, sizeof( zone->end ) );
int16_t srcloc;
f.Read( srcloc );
zone->SetSrcLoc( srcloc );
if( fileVer <= FileVersion( 0, 5, 0 ) )
if( fileVer >= FileVersion( 0, 5, 2 ) )
{
f.Skip( 4 );
int16_t srcloc;
f.Read( srcloc );
zone->SetSrcLoc( srcloc );
f.Read( &zone->end, sizeof( zone->end ) );
}
else
{
f.Skip( 2 );
f.Read( &zone->end, sizeof( zone->end ) );
int16_t srcloc;
f.Read( srcloc );
zone->SetSrcLoc( srcloc );
if( fileVer <= FileVersion( 0, 5, 0 ) )
{
f.Skip( 4 );
}
else
{
f.Skip( 2 );
}
}
__StringIdxOld str;
f.Read( str );
if( str.active )
{
zone->text.SetIdx( str.idx );
}
else
{
new ( &zone->text ) StringIdx();
}
f.Read( zone->callstack );
f.Read( str );
if( str.active )
{
zone->name.SetIdx( str.idx );
}
else
{
new ( &zone->name ) StringIdx();
}
f.Read( &zone->text, sizeof( zone->text ) + sizeof( zone->callstack ) + sizeof( zone->name ) );
refTime += zone->end;
zone->SetStart( refTime - m_data.baseTime );
ReadTimelinePre052( f, zone, thread, refTime, fileVer );
ReadTimelinePre058( f, zone, thread, refTime, fileVer );
zone->end = ReadTimeOffset( f, refTime );
if( zone->end >= 0 ) zone->end -= m_data.baseTime;
#ifdef TRACY_NO_STATISTICS

View File

@ -130,8 +130,8 @@ private:
{
const auto& v = data->data[i];
hash = ( ( hash << 5 ) + hash ) ^ size_t( v.line );
hash = ( ( hash << 5 ) + hash ) ^ size_t( v.file.__data );
hash = ( ( hash << 5 ) + hash ) ^ size_t( v.name.__data );
hash = ( ( hash << 5 ) + hash ) ^ size_t( v.file.Idx() );
hash = ( ( hash << 5 ) + hash ) ^ size_t( v.name.Idx() );
}
return hash;
}
@ -517,7 +517,7 @@ private:
tracy_force_inline void ReadTimeline( FileRead& f, ZoneEvent* zone, uint16_t thread, int64_t& refTime );
tracy_force_inline void ReadTimelinePre042( FileRead& f, ZoneEvent* zone, uint16_t thread, int fileVer );
tracy_force_inline void ReadTimelinePre052( FileRead& f, ZoneEvent* zone, uint16_t thread, int64_t& refTime, int fileVer );
tracy_force_inline void ReadTimelinePre058( FileRead& f, ZoneEvent* zone, uint16_t thread, int64_t& refTime, int fileVer );
tracy_force_inline void ReadTimeline( FileRead& f, GpuEvent* zone, int64_t& refTime, int64_t& refGpuTime );
tracy_force_inline void ReadTimelinePre052( FileRead& f, GpuEvent* zone, int64_t& refTime, int64_t& refGpuTime, int fileVer );
@ -525,7 +525,7 @@ private:
void ReadTimeline( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t thread, uint64_t size, int64_t& refTime );
void ReadTimelinePre042( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t thread, uint64_t size, int fileVer );
void ReadTimelinePre052( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t thread, uint64_t size, int64_t& refTime, int fileVer );
void ReadTimelinePre058( FileRead& f, Vector<ZoneEvent*>& vec, uint16_t thread, uint64_t size, int64_t& refTime, int fileVer );
void ReadTimeline( FileRead& f, Vector<GpuEvent*>& vec, uint64_t size, int64_t& refTime, int64_t& refGpuTime );
void ReadTimelinePre052( FileRead& f, Vector<GpuEvent*>& vec, uint64_t size, int64_t& refTime, int64_t& refGpuTime, int fileVer );