Construct parent call stacks for sampled stack traces.

This commit is contained in:
Bartosz Taudul 2020-02-29 16:24:15 +01:00
parent 935eb4cc61
commit 6cc4de8d28
3 changed files with 89 additions and 10 deletions

View File

@ -354,8 +354,9 @@ union CallstackFrameId
{
struct
{
uint64_t idx : 63;
uint64_t idx : 62;
uint64_t sel : 1;
uint64_t custom : 1;
};
uint64_t data;
};

View File

@ -30,10 +30,11 @@ namespace tracy
static inline CallstackFrameId PackPointer( uint64_t ptr )
{
assert( ( ( ptr & 0x4000000000000000 ) << 1 ) == ( ptr & 0x8000000000000000 ) );
assert( ( ( ptr & 0x3000000000000000 ) << 2 ) == ( ptr & 0xC000000000000000 ) );
CallstackFrameId id;
id.idx = ptr;
id.sel = 0;
id.custom = 0;
return id;
}
@ -3123,7 +3124,7 @@ void Worker::AddFrameImageData( uint64_t ptr, const char* data, size_t sz )
uint64_t Worker::GetCanonicalPointer( const CallstackFrameId& id ) const
{
assert( id.sel == 0 );
return ( id.idx & 0x7FFFFFFFFFFFFFFF ) | ( ( id.idx & 0x4000000000000000 ) << 1 );
return ( id.idx & 0x3FFFFFFFFFFFFFFF ) | ( ( id.idx & 0xC000000000000000 ) << 2 );
}
void Worker::AddCallstackPayload( uint64_t ptr, const char* _data, size_t _sz )
@ -3201,6 +3202,7 @@ void Worker::AddCallstackAllocPayload( uint64_t ptr, const char* data, size_t _s
frameData->size = 1;
id.idx = m_callstackAllocNextIdx++;
id.sel = 1;
id.custom = 0;
m_data.callstackFrameMap.emplace( id, frameData );
m_data.revFrameMap.emplace( frameData, id );
}
@ -5304,7 +5306,7 @@ void Worker::UpdateSampleStatistics( uint32_t callstack, uint32_t count, bool ca
}
}
UpdateSampleStatisticsImpl( frames, cssz, count );
UpdateSampleStatisticsImpl( frames, cssz, count, cs );
}
void Worker::UpdateSampleStatisticsPostponed( decltype(Worker::DataBlock::postponedSamples.begin())& it )
@ -5324,20 +5326,20 @@ void Worker::UpdateSampleStatisticsPostponed( decltype(Worker::DataBlock::postpo
frames[i] = frame;
}
UpdateSampleStatisticsImpl( frames, cssz, it->second );
UpdateSampleStatisticsImpl( frames, cssz, it->second, cs );
it = m_data.postponedSamples.erase( it );
}
void Worker::UpdateSampleStatisticsImpl( const CallstackFrameData** frames, uint8_t framesCount, uint32_t count )
void Worker::UpdateSampleStatisticsImpl( const CallstackFrameData** frames, uint8_t framesCount, uint32_t count, const VarArray<CallstackFrameId>& cs )
{
const auto fexcl = frames[0];
const auto fxsz = fexcl->size;
{
const auto fexcl = frames[0];
const auto fsz = fexcl->size;
const auto& frame0 = fexcl->data[0];
auto sym = m_data.symbolStats.find( frame0.symAddr );
if( sym == m_data.symbolStats.end() ) sym = m_data.symbolStats.emplace( frame0.symAddr, SymbolStats {} ).first;
sym->second.excl += count;
for( uint8_t f=1; f<fsz; f++ )
for( uint8_t f=1; f<fxsz; f++ )
{
const auto& frame = fexcl->data[f];
sym = m_data.symbolStats.find( frame.symAddr );
@ -5357,6 +5359,78 @@ void Worker::UpdateSampleStatisticsImpl( const CallstackFrameData** frames, uint
sym->second.incl += count;
}
}
CallstackFrameId parentFrameId;
if( fxsz != 1 )
{
auto cfdata = (CallstackFrame*)alloca( ( fxsz-1 ) * sizeof( CallstackFrame ) );
for( int i=0; i<fxsz-1; i++ )
{
cfdata[i] = fexcl->data[i+1];
}
CallstackFrameData cfd;
cfd.data = cfdata;
cfd.size = fxsz-1;
cfd.imageName = fexcl->imageName;
auto it = m_data.revParentFrameMap.find( &cfd );
if( it == m_data.revParentFrameMap.end() )
{
auto frame = m_slab.Alloc<CallstackFrame>( fxsz-1 );
memcpy( frame, cfdata, fxsz * sizeof( CallstackFrame ) );
auto frameData = m_slab.AllocInit<CallstackFrameData>();
frameData->data = frame;
frameData->size = fxsz - 1;
parentFrameId.idx = m_callstackParentNextIdx++;
parentFrameId.sel = 0;
parentFrameId.custom = 1;
m_data.callstackFrameMap.emplace( parentFrameId, frameData );
m_data.revParentFrameMap.emplace( frameData, parentFrameId );
}
else
{
parentFrameId = it->second;
}
}
const auto sz = framesCount - ( fxsz == 1 );
const auto memsize = sizeof( VarArray<CallstackFrameId> ) + sz * sizeof( CallstackFrameId );
auto mem = (char*)m_slab.AllocRaw( memsize );
auto data = (CallstackFrameId*)mem;
auto dst = data;
if( fxsz == 1 )
{
for( int i=0; i<sz; i++ )
{
*dst++ = cs[i+1];
}
}
else
{
*dst++ = parentFrameId;
for( int i=1; i<sz; i++ )
{
*dst++ = cs[i];
}
}
auto arr = (VarArray<CallstackFrameId>*)( mem + sz * sizeof( CallstackFrameId ) );
new(arr) VarArray<CallstackFrameId>( sz, data );
uint32_t idx;
auto it = m_data.parentCallstackMap.find( arr );
if( it == m_data.parentCallstackMap.end() )
{
idx = m_data.parentCallstackPayload.size();
m_data.parentCallstackMap.emplace( arr, idx );
m_data.parentCallstackPayload.push_back( arr );
}
else
{
idx = it->second;
m_slab.Unalloc( memsize );
}
}
#endif

View File

@ -227,6 +227,9 @@ private:
unordered_flat_map<uint64_t, SymbolStats> symbolStats;
#ifndef TRACY_NO_STATISTICS
unordered_flat_map<VarArray<CallstackFrameId>*, uint32_t, VarArrayHasher<CallstackFrameId>, VarArrayComparator<CallstackFrameId>> parentCallstackMap;
Vector<short_ptr<VarArray<CallstackFrameId>>> parentCallstackPayload;
unordered_flat_map<CallstackFrameData*, CallstackFrameId, RevFrameHash, RevFrameComp> revParentFrameMap;
unordered_flat_map<uint32_t, uint32_t> postponedSamples;
bool newFramesWereReceived = false;
bool callstackSamplesReady = false;
@ -643,7 +646,7 @@ private:
void ReconstructContextSwitchUsage();
void UpdateSampleStatistics( uint32_t callstack, uint32_t count, bool canPostpone );
void UpdateSampleStatisticsPostponed( decltype(Worker::DataBlock::postponedSamples.begin())& it );
void UpdateSampleStatisticsImpl( const CallstackFrameData** frames, uint8_t framesCount, uint32_t count );
void UpdateSampleStatisticsImpl( const CallstackFrameData** frames, uint8_t framesCount, uint32_t count, const VarArray<CallstackFrameId>& cs );
#endif
tracy_force_inline int64_t ReadTimeline( FileRead& f, ZoneEvent* zone, int64_t refTime, int32_t& childIdx );
@ -731,6 +734,7 @@ private:
CallstackFrameData* m_callstackFrameStaging;
uint64_t m_callstackFrameStagingPtr;
uint64_t m_callstackAllocNextIdx = 0;
uint64_t m_callstackParentNextIdx = 0;
uint64_t m_lastMemActionCallstack;
bool m_lastMemActionWasAlloc;