Replace parallel STL with PPQSort.

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.

.github/workflows/linux.yml

@@ -12,7 +12,7 @@ jobs:
     container: archlinux:base-devel
     steps:
     - name: Install dependencies
-      run: pacman -Syu --noconfirm && pacman -S --noconfirm --needed freetype2 tbb debuginfod wayland dbus libxkbcommon libglvnd meson cmake git wayland-protocols nodejs
+      run: pacman -Syu --noconfirm && pacman -S --noconfirm --needed freetype2 debuginfod wayland dbus libxkbcommon libglvnd meson cmake git wayland-protocols nodejs
     - name: Trust git repo
       run: git config --global --add safe.directory '*'
     - uses: actions/checkout@v4

capture/CMakeLists.txt

@@ -2,7 +2,6 @@ cmake_minimum_required(VERSION 3.16)
 
 option(NO_ISA_EXTENSIONS "Disable ISA extensions (don't pass -march=native or -mcpu=native to the compiler)" OFF)
 option(NO_STATISTICS "Disable calculation of statistics" ON)
-option(NO_PARALLEL_STL "Disable parallel STL" OFF)
 
 include(${CMAKE_CURRENT_LIST_DIR}/../cmake/version.cmake)
 

cmake/server.cmake

@@ -29,11 +29,7 @@ list(TRANSFORM TRACY_SERVER_SOURCES PREPEND "${TRACY_SERVER_DIR}/")
 
 add_library(TracyServer STATIC EXCLUDE_FROM_ALL ${TRACY_COMMON_SOURCES} ${TRACY_SERVER_SOURCES})
 target_include_directories(TracyServer PUBLIC ${TRACY_COMMON_DIR} ${TRACY_SERVER_DIR})
-target_link_libraries(TracyServer PUBLIC TracyCapstone TracyZstd)
+target_link_libraries(TracyServer PUBLIC TracyCapstone TracyZstd PPQSort::PPQSort)
 if(NO_STATISTICS)
     target_compile_definitions(TracyServer PUBLIC TRACY_NO_STATISTICS)
 endif()
-
-if(NOT NO_PARALLEL_STL AND UNIX AND NOT APPLE AND NOT EMSCRIPTEN)
-    target_link_libraries(TracyServer PRIVATE TracyTbb)
-endif()

cmake/vendor.cmake

@@ -214,32 +214,11 @@ if (NOT NO_FILESELECTOR AND NOT EMSCRIPTEN)
     endif()
 endif()
 
-# TBB
+# PPQSort
-if (NO_PARALLEL_STL)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DNO_PARALLEL_SORT")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DNO_PARALLEL_SORT")
-else()
-    if (UNIX AND NOT APPLE AND NOT EMSCRIPTEN)
-        # Tracy does not use TBB directly, but the implementation of parallel algorithms
-        # in some versions of libstdc++ depends on TBB. When it does, you must
-        # explicitly link against -ltbb.
-        #
-        # Some distributions have pgk-config files for TBB, others don't.
 
-        pkg_check_modules(TBB tbb)
-        if (TBB_FOUND)
-            add_library(TracyTbb INTERFACE)
-            target_include_directories(TracyTbb INTERFACE ${TBB_INCLUDE_DIRS})
-            target_link_libraries(TracyTbb INTERFACE ${TBB_LINK_LIBRARIES})
-        else()
 CPMAddPackage(
-                NAME tbb
+    NAME PPQSort
-                GITHUB_REPOSITORY oneapi-src/oneTBB
+    GITHUB_REPOSITORY GabTux/PPQSort
-                GIT_TAG v2021.12.0-rc2
+    VERSION 1.0.3
-                OPTIONS "TBB_TEST OFF"
+    EXCLUDE_FROM_ALL TRUE
 )
-            add_library(TracyTbb INTERFACE)
-            target_link_libraries(TracyTbb INTERFACE tbb)
-        endif()
-    endif()
-endif()

csvexport/CMakeLists.txt

@@ -1,7 +1,6 @@
 cmake_minimum_required(VERSION 3.16)
 
 option(NO_ISA_EXTENSIONS "Disable ISA extensions (don't pass -march=native or -mcpu=native to the compiler)" OFF)
-option(NO_PARALLEL_STL "Disable parallel STL" OFF)
 
 set(NO_STATISTICS OFF)
 

import/CMakeLists.txt

@@ -2,7 +2,6 @@ cmake_minimum_required(VERSION 3.16)
 
 option(NO_ISA_EXTENSIONS "Disable ISA extensions (don't pass -march=native or -mcpu=native to the compiler)" OFF)
 option(NO_STATISTICS "Disable calculation of statistics" ON)
-option(NO_PARALLEL_STL "Disable parallel STL" OFF)
 
 include(${CMAKE_CURRENT_LIST_DIR}/../cmake/version.cmake)
 

manual/tracy.tex

@@ -837,8 +837,6 @@ There is no need to install external libraries (e.g. with vcpkg). All libraries
 
 On Unix systems (including Linux), you will need to install the \texttt{pkg-config} utility to provide information about libraries.
 
-Due to some questionable design decisions by the compiler developers, you will most likely also need the \texttt{tbb} library\footnote{Technically, this is not a Tracy dependency, but rather a \texttt{libstdc++} dependency, but it may still not be installed by default.}. If not found, this library is downloaded automatically.
-
 Installation of the libraries on OSX can be facilitated using the \texttt{brew} package manager.
 
 \paragraph{Linux}

profiler/CMakeLists.txt

@@ -6,7 +6,6 @@ option(LEGACY "Instead of Wayland, use the legacy X11 backend on Linux" OFF)
 option(NO_ISA_EXTENSIONS "Disable ISA extensions (don't pass -march=native or -mcpu=native to the compiler)" OFF)
 option(NO_STATISTICS "Disable calculation of statistics" OFF)
 option(SELF_PROFILE "Enable self-profiling" OFF)
-option(NO_PARALLEL_STL "Disable parallel STL" OFF)
 
 include(${CMAKE_CURRENT_LIST_DIR}/../cmake/version.cmake)
 

profiler/src/profiler/TracySourceView.cpp

@@ -13,11 +13,11 @@
 #include "TracyImGui.hpp"
 #include "TracyMicroArchitecture.hpp"
 #include "TracyPrint.hpp"
-#include "TracySort.hpp"
 #include "TracySourceView.hpp"
 #include "TracyUtility.hpp"
 #include "TracyView.hpp"
 #include "TracyWorker.hpp"
+#include "tracy_pdqsort.h"
 
 #include "IconsFontAwesome6.h"
 

profiler/src/profiler/TracyTimelineItemPlot.cpp

@@ -5,6 +5,7 @@
 #include "TracyUtility.hpp"
 #include "TracyView.hpp"
 #include "TracyWorker.hpp"
+#include "tracy_pdqsort.h"
 
 namespace tracy
 {

profiler/src/profiler/TracyView_Annotations.cpp

@@ -1,6 +1,7 @@
 #include "TracyImGui.hpp"
 #include "TracyPrint.hpp"
 #include "TracyView.hpp"
+#include "tracy_pdqsort.h"
 
 namespace tracy
 {

profiler/src/profiler/TracyView_ContextSwitch.cpp

@@ -6,6 +6,7 @@
 #include "TracyTimelineContext.hpp"
 #include "TracyTimelineDraw.hpp"
 #include "TracyView.hpp"
+#include "tracy_pdqsort.h"
 
 namespace tracy
 {

profiler/src/profiler/TracyView_CpuData.cpp

@@ -8,6 +8,7 @@
 #include "TracyTimelineItem.hpp"
 #include "TracyTimelineContext.hpp"
 #include "TracyView.hpp"
+#include "tracy_pdqsort.h"
 
 constexpr float MinVisSize = 3;
 

profiler/src/profiler/TracyView_FindZone.cpp

@@ -7,7 +7,9 @@
 #include "TracyImGui.hpp"
 #include "TracyMouse.hpp"
 #include "TracyPrint.hpp"
+#include "TracySort.hpp"
 #include "TracyView.hpp"
+#include "tracy_pdqsort.h"
 
 namespace tracy
 {
@@ -507,10 +509,10 @@ void View::DrawFindZone()
                     }
                 }
                 auto mid = vec.begin() + vszorig;
-#ifdef NO_PARALLEL_SORT
+#ifdef __EMSCRIPTEN__
                 pdqsort_branchless( mid, vec.end() );
 #else
-                std::sort( std::execution::par_unseq, mid, vec.end() );
+                ppqsort::sort( ppqsort::execution::par, mid, vec.end() );
 #endif
                 std::inplace_merge( vec.begin(), mid, vec.end() );
 

profiler/src/profiler/TracyView_FrameTree.cpp

@@ -1,6 +1,7 @@
 #include "TracyImGui.hpp"
 #include "TracyPrint.hpp"
 #include "TracyView.hpp"
+#include "tracy_pdqsort.h"
 
 namespace tracy
 {

profiler/src/profiler/TracyView_Memory.cpp

@@ -4,6 +4,7 @@
 #include "TracyMouse.hpp"
 #include "TracyPrint.hpp"
 #include "TracyView.hpp"
+#include "tracy_pdqsort.h"
 
 namespace tracy
 {

profiler/src/profiler/TracyView_Samples.cpp

@@ -9,6 +9,7 @@
 #include "TracyTimelineContext.hpp"
 #include "TracyTimelineDraw.hpp"
 #include "TracyView.hpp"
+#include "tracy_pdqsort.h"
 
 namespace tracy
 {

profiler/src/profiler/TracyView_ZoneInfo.cpp

@@ -5,6 +5,7 @@
 #include "TracyPrint.hpp"
 #include "TracyMouse.hpp"
 #include "TracyView.hpp"
+#include "tracy_pdqsort.h"
 
 namespace tracy
 {

server/TracySort.hpp

@@ -1,19 +1,10 @@
 #ifndef __TRACYSORT_HPP__
 #define __TRACYSORT_HPP__
 
-#ifndef NO_PARALLEL_SORT
+#ifdef __EMSCRIPTEN__
-#  if !defined __APPLE__ && !defined __EMSCRIPTEN__ && ( ( defined _MSC_VER && _MSVC_LANG >= 201703L ) || __cplusplus >= 201703L )
-#    if __has_include(<execution>)
-#      include <algorithm>
-#      include <execution>
-#    else
-#      define NO_PARALLEL_SORT
-#    endif
-#  else
-#    define NO_PARALLEL_SORT
-#  endif
-#endif
-
 #  include "tracy_pdqsort.h"
+#else
+#  include <ppqsort.h>
+#endif
 
 #endif

server/TracySortedVector.hpp

@@ -102,10 +102,10 @@ public:
         const auto se = sb + sortedEnd;
         const auto sl = se - 1;
         const auto ue = v.end();
-#ifdef NO_PARALLEL_SORT
+#ifdef __EMSCRIPTEN__
-        pdqsort_branchless( se, ue, comp );
+        pdqsort_branchless( sb, se, comp );
 #else
-        std::sort( std::execution::par_unseq, se, ue, comp );
+        ppqsort::sort( ppqsort::execution::par, sb, se, comp );
 #endif
         const auto ss = std::lower_bound( sb, se, *se, comp );
         const auto uu = std::lower_bound( se, ue, *sl, comp );

server/TracyWorker.cpp

@@ -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; } );
+    ppqsort::sort( ppqsort::execution::par, 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.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();

update/CMakeLists.txt

@@ -2,7 +2,6 @@ cmake_minimum_required(VERSION 3.16)
 
 option(NO_ISA_EXTENSIONS "Disable ISA extensions (don't pass -march=native or -mcpu=native to the compiler)" OFF)
 option(NO_STATISTICS "Disable calculation of statistics" ON)
-option(NO_PARALLEL_STL "Disable parallel STL" OFF)
 
 include(${CMAKE_CURRENT_LIST_DIR}/../cmake/version.cmake)