zhangengru/tracy
1
0
Fork 0
mirror of https://github.com/wolfpld/tracy.git synced 2025-03-20 07:40:02 +08:00
Code Issues Projects Wiki Activity

Replace parallel STL with PPQSort.

Browse Source 
PPQSort is supposedly quite fast: https://github.com/GabTux/PPQSort

More importantly, it does not depend on TBB fuckery, so there's no longer
a need to link with an external library that people may or may not have.

The NO_PARALLEL_STL option is out, as it was provided solely to deal with
TBB being not available. Sequential sorting is still used on emscripten.
This commit is contained in:
Bartosz Taudul
2024-09-26 14:38:11 +02:00
parent d400183483
commit 1c1faeff2d
22 changed files with 42 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files
server/TracyWorker.cpp
+13 -11
View File
@@ -37,6 +37,7 @@
#include "TracySort.hpp"
#include "TracyTaskDispatch.hpp"
#include "TracyWorker.hpp"
#include "tracy_pdqsort.h"
namespace tracy
{
@@ -1552,12 +1553,13 @@ Worker::Worker( FileRead& f, EventType::Type eventMask, bool bgTasks, bool allow
m_data.symbolLoc[symIdx++] = SymbolLocation { symAddr, size.Val() };
}
}
#ifdef NO_PARALLEL_SORT
#ifdef __EMSCRIPTEN__
pdqsort_branchless( m_data.symbolLoc.begin(), m_data.symbolLoc.end(), [] ( const auto& l, const auto& r ) { return l.addr < r.addr; } );
pdqsort_branchless( m_data.symbolLocInline.begin(), m_data.symbolLocInline.end() );
#else
std::sort( std::execution::par_unseq, m_data.symbolLoc.begin(), m_data.symbolLoc.end(), [] ( const auto& l, const auto& r ) { return l.addr < r.addr; } );
std::sort( std::execution::par_unseq, m_data.symbolLocInline.begin(), m_data.symbolLocInline.end() );
ppqsort::sort( ppqsort::execution::par, m_data.symbolLoc.begin(), m_data.symbolLoc.end(), [] ( const auto& l, const auto& r ) { return l.addr < r.addr; } );
ppqsort::sort( ppqsort::execution::par, m_data.symbolLocInline.begin(), m_data.symbolLocInline.end() );
#endif
f.Read( sz );
@@ -4155,10 +4157,10 @@ void Worker::DoPostponedSymbols()
{
if( m_data.newSymbolsIndex >= 0 )
{
#ifdef NO_PARALLEL_SORT
#ifdef __EMSCRIPTEN__
pdqsort_branchless( m_data.symbolLoc.begin() + m_data.newSymbolsIndex, m_data.symbolLoc.end(), [] ( const auto& l, const auto& r ) { return l.addr < r.addr; } );
#else
std::sort( std::execution::par_unseq, m_data.symbolLoc.begin() + m_data.newSymbolsIndex, m_data.symbolLoc.end(), [] ( const auto& l, const auto& r ) { return l.addr < r.addr; } );
ppqsort::sort( ppqsort::execution::par, m_data.symbolLoc.begin() + m_data.newSymbolsIndex, m_data.symbolLoc.end(), [] ( const auto& l, const auto& r ) { return l.addr < r.addr; } );
#endif
const auto ms = std::lower_bound( m_data.symbolLoc.begin(), m_data.symbolLoc.begin() + m_data.newSymbolsIndex, m_data.symbolLoc[m_data.newSymbolsIndex], [] ( const auto& l, const auto& r ) { return l.addr < r.addr; } );
std::inplace_merge( ms, m_data.symbolLoc.begin() + m_data.newSymbolsIndex, m_data.symbolLoc.end(), [] ( const auto& l, const auto& r ) { return l.addr < r.addr; } );
@@ -4170,10 +4172,10 @@ void Worker::DoPostponedInlineSymbols()
{
if( m_data.newInlineSymbolsIndex >= 0 )
{
#ifdef NO_PARALLEL_SORT
#ifdef __EMSCRIPTEN__
pdqsort_branchless( m_data.symbolLocInline.begin() + m_data.newInlineSymbolsIndex, m_data.symbolLocInline.end() );
#else
std::sort( std::execution::par_unseq, m_data.symbolLocInline.begin() + m_data.newInlineSymbolsIndex, m_data.symbolLocInline.end() );
ppqsort::sort( ppqsort::execution::par, m_data.symbolLocInline.begin() + m_data.newInlineSymbolsIndex, m_data.symbolLocInline.end() );
#endif
const auto ms = std::lower_bound( m_data.symbolLocInline.begin(), m_data.symbolLocInline.begin() + m_data.newInlineSymbolsIndex, m_data.symbolLocInline[m_data.newInlineSymbolsIndex] );
std::inplace_merge( ms, m_data.symbolLocInline.begin() + m_data.newInlineSymbolsIndex, m_data.symbolLocInline.end() );
@@ -7017,10 +7019,10 @@ void Worker::CreateMemAllocPlot( MemData& memdata )
void Worker::ReconstructMemAllocPlot( MemData& mem )
{
#ifdef NO_PARALLEL_SORT
#ifdef __EMSCRIPTEN__
pdqsort_branchless( mem.frees.begin(), mem.frees.end(), [&mem] ( const auto& lhs, const auto& rhs ) { return mem.data[lhs].TimeFree() < mem.data[rhs].TimeFree(); } );
#else
std::sort( std::execution::par_unseq, mem.frees.begin(), mem.frees.end(), [&mem] ( const auto& lhs, const auto& rhs ) { return mem.data[lhs].TimeFree() < mem.data[rhs].TimeFree(); } );
ppqsort::sort( ppqsort::execution::par, mem.frees.begin(), mem.frees.end(), [&mem] ( const auto& lhs, const auto& rhs ) { return mem.data[lhs].TimeFree() < mem.data[rhs].TimeFree(); } );
#endif
const auto psz = mem.data.size() + mem.frees.size() + 1;
@@ -7962,10 +7964,10 @@ void Worker::Write( FileWrite& f, bool fiDict )
}
if( m_inconsistentSamples )
{
#ifdef NO_PARALLEL_SORT
#ifdef __EMSCRIPTEN__
pdqsort_branchless( thread->samples.begin(), thread->samples.end(), [] ( const auto& lhs, const auto& rhs ) { return lhs.time.Val() < rhs.time.Val(); } );
#else
std::sort( std::execution::par_unseq, thread->samples.begin(), thread->samples.end(), [] ( const auto& lhs, const auto& rhs ) { return lhs.time.Val() < rhs.time.Val(); } );
ppqsort::sort( ppqsort::execution::par, thread->samples.begin(), thread->samples.end(), [] ( const auto& lhs, const auto& rhs ) { return lhs.time.Val() < rhs.time.Val(); } );
#endif
}
sz = thread->samples.size();