Merge pull request #7339 from hvlad/PageCache/HashTable_PR

Hash table based on lock-free list for page cache.

.github/workflows/main.yml

@@ -15,7 +15,7 @@ jobs:
 
       - name: Prepare
         run: |
-          sudo apt-get install libtool-bin libtomcrypt1 libtomcrypt-dev libtommath1 libtommath-dev libicu-dev zlib1g-dev
+          sudo apt-get install libtool-bin libtomcrypt1 libtomcrypt-dev libtommath1 libtommath-dev libicu-dev zlib1g-dev cmake
 
       - name: Build
         run: |
@@ -135,7 +135,7 @@ jobs:
           fetch-depth: 10
 
       - name: Prepare
-        run: apk update && apk --no-cache --update add build-base libtool git autoconf automake zlib-dev icu-dev ncurses-dev libedit-dev linux-headers tar
+        run: apk update && apk --no-cache --update add build-base libtool git autoconf automake cmake zlib-dev icu-dev ncurses-dev libedit-dev linux-headers tar
 
       - name: Build
         run: |
@@ -323,7 +323,7 @@ jobs:
 
       - name: Prepare - Install tools
         run: |
-          brew install automake libtool ninja
+          brew install automake cmake libtool ninja
 
       - name: Cache - libc++ install
         id: cache-libcxx-install-macos

.gitignore

@@ -27,3 +27,4 @@ extern/ttmath/release/
 /src/include/gen/parse.h
 /src/include/gen/autoconfig.auto
 /src/include/gen/autoconfig.h
+extern/libcds/lib/

appveyor.yml

@@ -28,7 +28,7 @@ install:
   - cmd: cd builds\win32
   - cmd: run_all.bat JUSTBUILD
   - cmd: set ARTIFACTS_PATH=output_%FB_OUTPUT_SUFFIX%
-  - sh: export APT_PACKAGES="libtool-bin"
+  - sh: export APT_PACKAGES="libtool-bin cmake"
   - sh: if [ $PLATFORM = "x64" ]; then export APT_PACKAGES="$APT_PACKAGES libtommath1 libtommath-dev libicu-dev zlib1g-dev"; fi
   - sh: if [ $PLATFORM = "x86" ]; then export APT_PACKAGES="$APT_PACKAGES gcc-multilib g++-multilib libncurses5-dev:i386 libtommath-dev:i386 libicu-dev:i386 zlib1g-dev:i386"; fi
   - sh: if [ $PLATFORM = "x64" ]; then export CC="gcc" CXX="g++"; fi

builds/posix/Makefile.in

@@ -83,6 +83,13 @@ LTC_LDFLAGS='-L$(LIB) $(subst $,$$$$,$(call LIB_LINK_RPATH,lib))'
 endif
 endif
 
+# correct build type for cmake builds
+FB_CMAKE_BUILD_TYPE=$(TARGET)
+ifeq ($(FB_CMAKE_BUILD_TYPE),Native)
+  FB_CMAKE_BUILD_TYPE=Release
+endif
+
+
 .PHONY:	master_process cross_process firebird Debug Release external
 
 all:	firebird
@@ -190,6 +197,8 @@ ifeq ($(RE2_BUILD_FLG),Y)
 	ln -sf $(ROOT)/extern/re2/obj/libre2.a $(LIB)
 endif
 
+	$(MAKE) libcds
+
 ifeq ($(TOMMATH_BUILD_FLG),Y)
 	CC="$(CC)" CFLAGS="$(CFLAGS)" AR="$(AR)" $(MAKE) -C $(ROOT)/extern/libtommath -f makefile.shared GCC="$(CC)"
 
@@ -279,6 +288,23 @@ $(RE2_LIB):	$(RE2_Objs)
 	-$(RM) $@
 	$(STATICLIB_LINK) $@ $^
 
+#___________________________________________________________________________
+# libcds
+#
+
+.PHONY:	libcds
+libcds:
+	mkdir -p $(LIBCDS)/lib/$(TARGET)
+	cd $(LIBCDS)/lib/$(TARGET); \
+	cmake -DCMAKE_BUILD_TYPE=$(FB_CMAKE_BUILD_TYPE) -DCMAKE_CXX_FLAGS=-fPIC $(LIBCDS)
+	AR="$(AR)" $(MAKE) -C $(LIBCDS)/lib/$(TARGET)
+
+ifeq ($(TARGET),Debug)
+	ln -sf $(LIBCDS)/lib/$(TARGET)/bin/libcds-s_d.a $(LIB)/libcds.a
+else
+	ln -sf $(LIBCDS)/lib/$(TARGET)/bin/libcds-s.a $(LIB)/libcds.a
+endif
+
 #___________________________________________________________________________
 # main build target for both debug and release builds
 #
@@ -343,6 +369,7 @@ cross2:
 	ln -sf $(ROOT)/extern/decNumber/libdecFloat$(CROSS).a $(LIB)
 	CXX="$(CXX)" CXXFLAGS="$(CXXFLAGS)" $(MAKE) -C $(ROOT)/extern/int128/absl/numeric
 	ln -sf $(ROOT)/extern/int128/absl/numeric/libi128$(CROSS).a $(LIB)
+	$(MAKE) libcds
 	$(MAKE) yvalve
 	$(MAKE) engine
 	$(MAKE) fbintl

builds/posix/make.android.arm64

@@ -41,7 +41,7 @@ DEV_FLAGS=$(COMMON_FLAGS) $(WARN_FLAGS)
 CROSS_CONFIG=android.arm64
 
 LDFLAGS += -static-libstdc++
-DroidLibs := -lm -ldl $(DECLIB) $(RE2LIB) $(I128LIB)
+DroidLibs := -lm -ldl $(DECLIB) $(RE2LIB) $(I128LIB) $(LIBCDSLIB)
 UDR_SUPPORT_LIBS :=
 
 LINK_LIBS = $(DroidLibs)

builds/posix/make.android.arme

@@ -41,7 +41,7 @@ DEV_FLAGS=$(COMMON_FLAGS) $(WARN_FLAGS)
 CROSS_CONFIG=android.arme
 
 LDFLAGS += -static-libstdc++
-DroidLibs := -lm -ldl $(DECLIB) $(RE2LIB) $(I128LIB)
+DroidLibs := -lm -ldl $(DECLIB) $(RE2LIB) $(I128LIB) $(LIBCDSLIB)
 UDR_SUPPORT_LIBS :=
 
 LINK_LIBS = $(DroidLibs)

builds/posix/make.android.x86

@@ -41,7 +41,7 @@ DEV_FLAGS=$(COMMON_FLAGS) $(WARN_FLAGS)
 CROSS_CONFIG=android.x86
 
 LDFLAGS += -static-libstdc++
-DroidLibs := -lm -ldl $(DECLIB) $(RE2LIB) $(I128LIB)
+DroidLibs := -lm -ldl $(DECLIB) $(RE2LIB) $(I128LIB) $(LIBCDSLIB)
 UDR_SUPPORT_LIBS :=
 
 LINK_LIBS = $(DroidLibs)

builds/posix/make.android.x86_64

@@ -41,7 +41,7 @@ DEV_FLAGS=$(COMMON_FLAGS) $(WARN_FLAGS)
 CROSS_CONFIG=android.x86_64
 
 LDFLAGS += -static-libstdc++
-DroidLibs := -lm -ldl $(DECLIB) $(RE2LIB) $(I128LIB)
+DroidLibs := -lm -ldl $(DECLIB) $(RE2LIB) $(I128LIB) $(LIBCDSLIB)
 UDR_SUPPORT_LIBS :=
 
 LINK_LIBS = $(DroidLibs)

builds/posix/make.defaults

@@ -148,6 +148,8 @@ else
   I128LIB=
 endif
 
+LIBCDSLIB=-lcds
+
 # crypt library
 CRYPTLIB=@CRYPTLIB@
 
@@ -204,8 +206,8 @@ endif
 
 STATICLIB_LINK = $(AR) crus
 
-LINK_LIBS = @LIBS@ $(DECLIB) $(RE2LIB) $(I128LIB)
+LINK_LIBS = @LIBS@ $(DECLIB) $(RE2LIB) $(I128LIB) $(LIBCDSLIB)
-SO_LINK_LIBS = @LIBS@ $(DECLIB) $(RE2LIB) $(I128LIB)
+SO_LINK_LIBS = @LIBS@ $(DECLIB) $(RE2LIB) $(I128LIB) $(LIBCDSLIB)
 
 # Default extensions
 
@@ -294,6 +296,11 @@ TOMCRYPT_INC=$(TOMCRYPT)/src/headers
 TOMCRYPT_SO=$(TOMCRYPT)/.libs/libtomcrypt.so
 TOMCRYPT_VER=1
 
+# Own libcds support
+LIBCDS=$(ROOT)/extern/libcds
+LIBCDS_INC=$(LIBCDS)
+LIBCDS_DEF=CDS_BUILD_STATIC_LIB
+
 # LINKER OPTIONS
 #
 

builds/posix/make.rules

@@ -38,6 +38,8 @@ ifneq ($(SYSTEM_BOOST_FLG),Y)
   WFLAGS += -I$(ROOT)/extern/boost
 endif
 
+WFLAGS += -I$(LIBCDS_INC) -D$(LIBCDS_DEF)
+
 ifeq ($(TOMMATH_BUILD_FLG),Y)
   WFLAGS += -I$(TOMMATH_INC)
 endif

builds/win32/clean_all.bat

@@ -22,6 +22,16 @@ for %%v in ( %* )  do (
 @echo Cleaning icu...
 @rmdir /S /Q "%FB_ROOT_PATH%\extern\icu\%FB_TARGET_PLATFORM%\%FBBUILD_BUILDTYPE%" 2>nul
 
+@echo Cleaning cds...
+@for /D %%d in ("%FB_ROOT_PATH%\extern\libcds\obj\*") do (
+  rmdir /S /Q "%%d\%FB_TARGET_PLATFORM%\cds\%FB_CONFIG%-static" 2>nul
+)
+
+@for /D %%d in ("%FB_ROOT_PATH%\extern\libcds\bin\*") do (
+  rmdir /S /Q "%%d\%FB_TARGET_PLATFORM%-%FB_CONFIG%-static" 2>nul
+)
+
+
 @echo Cleaning decNumber...
 @rmdir /S /Q "%FB_ROOT_PATH%\extern\decNumber\lib\%FB_TARGET_PLATFORM%" 2>nul
 @rmdir /S /Q "%FB_ROOT_PATH%\extern\decNumber\temp\%FB_TARGET_PLATFORM%" 2>nul

builds/win32/compile.bat

@@ -18,6 +18,11 @@ set projects=
 	set config=debug
 )
 
+:: Special case for CDS, set in make_boot only
+@if "%FB_LIBCDS%"=="1" (
+	set config=%config%-static
+)
+
 shift
 shift
 

builds/win32/make_boot.bat

@@ -34,6 +34,9 @@ if "%ERRLEV%"=="1" goto :END
 call :btyacc
 if "%ERRLEV%"=="1" goto :END
 
+call :libcds
+if "%ERRLEV%"=="1" goto :END
+
 call :LibTom
 if "%ERRLEV%"=="1" goto :END
 
@@ -145,6 +148,17 @@ goto :EOF
 if errorlevel 1 call :boot2 decNumber_%FB_OBJ_DIR%
 goto :EOF
 
+::===================
+:: Build libcds
+:libcds
+@echo.
+set FB_LIBCDS=1
+@echo Building libcds (%FB_OBJ_DIR%)...
+@call compile.bat extern\libcds\projects\Win\vc141\cds libcds_%FB_CONFIG%_%FB_TARGET_PLATFORM%.log cds
+if errorlevel 1 call :boot2 libcds%FB_OBJ_DIR%
+set FB_LIBCDS=
+goto :EOF
+
 ::===================
 :: BUILD ttmath
 :ttmath

builds/win32/msvc15/engine.vcxproj

@@ -113,24 +113,28 @@
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdRelease.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdDebug.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdRelease.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdDebug.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <PropertyGroup Label="UserMacros" />
   <PropertyGroup>

builds/win32/msvc15/engine_static.vcxproj

@@ -89,6 +89,7 @@
     <ClCompile Include="..\..\..\src\jrd\GlobalRWLock.cpp" />
     <ClCompile Include="..\..\..\src\jrd\idx.cpp" />
     <ClCompile Include="..\..\..\src\jrd\inf.cpp" />
+    <ClCompile Include="..\..\..\src\jrd\InitCDSLib.cpp" />
     <ClCompile Include="..\..\..\src\jrd\intl.cpp" />
     <ClCompile Include="..\..\..\src\jrd\IntlManager.cpp" />
     <ClCompile Include="..\..\..\src\jrd\intl_builtin.cpp" />
@@ -276,6 +277,7 @@
     <ClInclude Include="..\..\..\src\jrd\inf_proto.h" />
     <ClInclude Include="..\..\..\src\jrd\inf_pub.h" />
     <ClInclude Include="..\..\..\src\jrd\ini.h" />
+    <ClInclude Include="..\..\..\src\jrd\InitCDSLib.h" />
     <ClInclude Include="..\..\..\src\jrd\ini_proto.h" />
     <ClInclude Include="..\..\..\src\jrd\intl.h" />
     <ClInclude Include="..\..\..\src\jrd\IntlManager.h" />
@@ -429,24 +431,28 @@
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdRelease.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdDebug.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdRelease.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdDebug.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <PropertyGroup Label="UserMacros" />
   <PropertyGroup>

builds/win32/msvc15/engine_static.vcxproj.filters

@@ -513,6 +513,9 @@
     <ClCompile Include="..\..\..\src\jrd\MetaName.cpp">
       <Filter>JRD files</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\..\src\jrd\InitCDSLib.cpp">
+      <Filter>JRD files</Filter>
+    </ClCompile>
     <ClCompile Include="..\..\..\src\jrd\optimizer\InnerJoin.cpp">
       <Filter>Optimizer</Filter>
     </ClCompile>
@@ -1070,6 +1073,9 @@
     <ClInclude Include="..\..\..\src\jrd\QualifiedName.h">
       <Filter>Header files</Filter>
     </ClInclude>
+    <ClInclude Include="..\..\..\src\jrd\InitCDSLib.h">
+      <Filter>Header files</Filter>
+    </ClInclude>
     <ClInclude Include="..\..\..\src\jrd\WorkerAttachment.h">
       <Filter>Header files</Filter>
     </ClInclude>

builds/win32/msvc15/engine_test.vcxproj

@@ -66,24 +66,28 @@
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdDebug.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdRelease.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdDebug.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="PropertySheets">
     <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
     <Import Project="$(VCTargetsPath)Microsoft.CPP.UpgradeFromVC71.props" />
     <Import Project="FirebirdCommon.props" />
     <Import Project="FirebirdRelease.props" />
+    <Import Project="libcds.props" />
   </ImportGroup>
   <PropertyGroup Label="UserMacros" />
   <PropertyGroup>

builds/win32/msvc15/fbserver.vcxproj

@@ -239,4 +239,4 @@
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets">
   </ImportGroup>
-</Project>
+</Project>

builds/win32/msvc15/libcds.props

@@ -0,0 +1,23 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ImportGroup Label="PropertySheets" />
+  <PropertyGroup Label="UserMacros">
+    <LIBCDS_ROOT>..\..\..\extern\libcds</LIBCDS_ROOT>
+  </PropertyGroup>
+  <PropertyGroup />
+  <ItemDefinitionGroup>
+    <ClCompile>
+      <AdditionalIncludeDirectories>$(LIBCDS_ROOT);%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <PreprocessorDefinitions>CDS_BUILD_STATIC_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+    </ClCompile>
+    <Link>
+      <AdditionalLibraryDirectories>$(LIBCDS_ROOT)\bin\vc.$(PlatformToolset)\$(Platform)-$(Configuration)-static;%(AdditionalLibraryDirectories)</AdditionalLibraryDirectories>
+      <AdditionalDependencies>libcds-$(PlatformTarget).lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemGroup>
+    <BuildMacro Include="LIBCDS_ROOT">
+      <Value>$(LIBCDS_ROOT)</Value>
+    </BuildMacro>
+  </ItemGroup>
+</Project>

extern/libcds/.gitignore

@@ -0,0 +1,21 @@
+/doc
+/sandbox
+*.o
+*.d
+/bin
+/obj
+/projects/Win/vc14/cds.sdf
+/projects/Win/vc14/cds.v14.suo
+/projects/Win/vc14/*.user
+/projects/Win/vc14/*.opensdf
+/projects/Win/vc14/.vs/
+/projects/Win/vc141/.vs/
+/projects/Win/vc141/*.user
+*.log
+/.project
+/projects/Win/vc14/*.opendb
+/test/stress/data/dictionary.txt
+/projects/Win/vc14/cds.VC.db
+/.cproject
+/.settings/
+/tools/change_license.pl

extern/libcds/.travis.yml

@@ -0,0 +1,265 @@
+language: cpp
+
+install: 
+  - chmod +x ./build/CI/travis-ci/install.sh
+  - ./build/CI/travis-ci/install.sh
+
+script:
+  - chmod +x ./build/CI/travis-ci/run.sh
+  - ./build/CI/travis-ci/run.sh
+
+linux: &linux_gcc
+  os: linux
+  addons:
+    apt:
+      sources:
+        - ubuntu-toolchain-r-test
+      packages:
+        - g++-6
+  compiler:
+    - g++-6
+  before_install:
+    - eval "CC=gcc-6 && CXX=g++-6"
+    
+
+linux: &linux_clang
+  os: linux
+  addons:
+    apt:
+      sources:
+        - ubuntu-toolchain-r-test
+        - llvm-toolchain-trusty-4.0
+      packages:
+        - clang-4.0
+  compiler:
+    - clang-4.0          
+  before_install:
+    - eval "CC=clang-4.0 && CXX=clang++-4.0"
+  
+      
+osx: &osx
+   os: osx
+   osx_image: xcode8.3
+   compiler:
+    - clang
+   before_install:
+    - eval "CC=clang && CXX=clang++"
+    
+
+matrix:
+  include:
+##   BUILD_TYPE=Release CXX_COMPILER=g++-6
+    - <<: *linux_gcc
+      env: TARGET=unit-deque BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-ilist BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-list BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-map BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-misc BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-pqueue BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-queue BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset-feldman BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset-michael-michael BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset-michael-lazy BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset-michael-iterable BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset-skip BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset-split-michael BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset-split-lazy BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset-split-iterable BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-set BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-striped-set BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-stack BUILD_TYPE=Release 
+    - <<: *linux_gcc
+      env: TARGET=unit-tree BUILD_TYPE=Release 
+
+##   BUILD_TYPE=Debug CXX_COMPILER=g++-6
+    - <<: *linux_gcc
+      env: TARGET=unit-deque BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-ilist BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-list BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-map BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-misc BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-pqueue BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-queue BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-iset BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-set BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-striped-set BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-stack BUILD_TYPE=Debug 
+    - <<: *linux_gcc
+      env: TARGET=unit-tree BUILD_TYPE=Debug 
+
+##   BUILD_TYPE=Release CXX_COMPILER=clang-4.0
+    - <<: *linux_clang
+      env: TARGET=unit-deque BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-ilist BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-list BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-misc BUILD_TYPE=Release  LINKER_FLAGS=-latomic
+    - <<: *linux_clang
+      env: TARGET=unit-pqueue BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-queue BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-set-feldman BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-set-michael-michael BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-set-michael-iterable BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-set-michael-lazy BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-set-skip BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-set-split-iterable BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-set-split-michael BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-set-split-lazy BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-striped-set BUILD_TYPE=Release 
+    - <<: *linux_clang
+      env: TARGET=unit-stack BUILD_TYPE=Release 
+# FIXME: building too long. Travis-ci will stop building.
+#  - BUILD_TYPE=Release  TARGET=unit-map
+#  - BUILD_TYPE=Release  TARGET=unit-iset
+#  - BUILD_TYPE=Release  TARGET=unit-tree
+
+##   BUILD_TYPE=Debug CXX_COMPILER=clang-4.0
+    - <<: *linux_clang
+      env: TARGET=unit-deque BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-ilist BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-list BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-map BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-misc BUILD_TYPE=Debug  LINKER_FLAGS=-latomic
+    - <<: *linux_clang
+      env: TARGET=unit-pqueue BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-queue BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-iset BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-set BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-striped-set BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-stack BUILD_TYPE=Debug 
+    - <<: *linux_clang
+      env: TARGET=unit-tree BUILD_TYPE=Debug 
+
+# RELEASE
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-deque 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-ilist 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-list 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-misc 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-pqueue 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-queue 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset-feldman 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset-michael-michael 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset-michael-lazy 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset-michael-iterable 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset-skip 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset-split-michael 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset-split-lazy 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-iset-split-iterable 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-striped-set 
+    - <<: *osx
+      env: BUILD_TYPE=Release TARGET=unit-stack 
+# FIXME: building too long. Travis-ci will stop building.
+#    - <<: *osx
+#      env: BUILD_TYPE=Release TARGET=unit-map 
+#    - <<: *osx
+#      env: BUILD_TYPE=Release TARGET=unit-set 
+#    - <<: *osx
+#      env: BUILD_TYPE=Release TARGET=unit-tree 
+
+# DEBUG
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-deque 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-ilist 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-list 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-map 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-misc 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-pqueue 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-queue 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset-feldman 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset-michael-michael 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset-michael-lazy 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset-michael-iterable 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset-skip 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset-split-michael 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset-split-lazy 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-iset-split-iterable 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-set 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-striped-set 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-stack 
+    - <<: *osx
+      env: BUILD_TYPE=Debug TARGET=unit-tree 
+      

extern/libcds/CMakeLists.txt

@@ -0,0 +1,250 @@
+cmake_minimum_required(VERSION 3.0.2 FATAL_ERROR)
+
+cmake_policy(SET CMP0016 NEW)
+if(POLICY CMP0042)
+    cmake_policy(SET CMP0042 NEW)
+endif()
+
+set(CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/build/cmake ${CMAKE_MODULE_PATH})
+include(TargetArch)
+include(CheckIncludeFileCXX)
+
+project(cds)
+
+set(PROJECT_VERSION 2.3.3)
+
+# Options
+option(WITH_TESTS "Build unit tests" OFF)
+option(WITH_TESTS_COVERAGE "Analyze test coverage using gcov (only for gcc)" OFF)
+option(WITH_BOOST_ATOMIC "Use boost atomics (only for boost >= 1.54)" OFF)
+option(WITH_ASAN "Build ASan+UBSan instrumented code" OFF)
+option(WITH_TSAN "Build TSan instrumented code" OFF)
+option(ENABLE_UNIT_TEST "Enable unit test" ON)
+option(ENABLE_STRESS_TEST "Enable stress test" ON)
+set(CMAKE_TARGET_ARCHITECTURE "" CACHE string "Target build architecture")
+
+find_package(Threads)
+
+if(NOT CMAKE_TARGET_ARCHITECTURE)
+    target_architecture(CMAKE_TARGET_ARCHITECTURE)
+endif()
+
+if(APPLE)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -D_DARWIN_C_SOURCE")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_DARWIN_C_SOURCE")
+endif()
+
+if(WITH_BOOST_ATOMIC)
+    if(TARGET boost::atomic)
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DCDS_USE_BOOST_ATOMIC")
+        link_libraries(boost::atomic)
+    else()
+        find_package(Boost 1.53 COMPONENTS atomic)
+        if(Boost_FOUND)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DCDS_USE_BOOST_ATOMIC")
+            message(STATUS "Boost version allows using of boost.atomic: activated")
+        endif()
+    endif()
+endif(WITH_BOOST_ATOMIC)
+
+if(WITH_ASAN)
+    if(CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+        set(CMAKE_CXX_FLAGS_DEBUG "-D_DEBUG")
+        set(CMAKE_CXX_FLAGS_RELEASE "-DNDEBUG")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O1 -fPIC -fsanitize=address,undefined -g -DCDS_ADDRESS_SANITIZER_ENABLED -fno-omit-frame-pointer -fno-optimize-sibling-calls")
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O1 -fsanitize=address,undefined -g -DCDS_ASAN_ENABLED -fno-omit-frame-pointer -fno-optimize-sibling-calls")
+        set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=address,undefined -pie")
+    else()
+        message(WARNING "Compiler does not support AddressSanitizer")
+    endif()
+endif(WITH_ASAN)
+
+if(WITH_TSAN)
+    if(CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+        set(CMAKE_CXX_FLAGS_DEBUG "-D_DEBUG")
+        set(CMAKE_CXX_FLAGS_RELEASE "-DNDEBUG")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O1 -fPIC -fsanitize=thread -g -DCDS_THREAD_SANITIZER_ENABLED -fno-omit-frame-pointer")
+        set(CMAKE_C_FLAGS "${CMAKE_CXX_FLAGS} -O1 -fPIC -fsanitize=thread -g -DCDS_THREAD_SANITIZER_ENABLED -fno-omit-frame-pointer")
+        set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fsanitize=thread -pie")
+    else()
+        message(WARNING "Compiler does not support ThreadSanitizer")
+    endif()
+endif(WITH_TSAN)
+
+if(WITH_TESTS_COVERAGE)
+    if(CMAKE_COMPILER_IS_GNUCXX)
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fprofile-arcs -ftest-coverage")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fprofile-arcs -ftest-coverage")
+        set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fprofile-arcs -ftest-coverage")
+        message(STATUS "Test coverage analysis: activated")
+    else()
+        message(WARNING "Compiler is not GNU gcc! Test coverage couldn't be analyzed")
+    endif()
+endif(WITH_TESTS_COVERAGE)
+
+set(CDS_SHARED_LIBRARY ${PROJECT_NAME})
+set(CDS_STATIC_LIBRARY ${PROJECT_NAME}-s)
+
+set(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE)
+set(CMAKE_INCLUDE_CURRENT_DIR ON)
+
+if(CDS_BIN_DIR)
+    set(EXECUTABLE_OUTPUT_PATH ${CDS_BIN_DIR})
+    set(LIBRARY_OUTPUT_PATH ${CDS_BIN_DIR})
+else()
+    set(EXECUTABLE_OUTPUT_PATH ${PROJECT_BINARY_DIR}/bin)
+    set(LIBRARY_OUTPUT_PATH ${PROJECT_BINARY_DIR}/bin)
+endif()
+message(STATUS "Binary output path: ${EXECUTABLE_OUTPUT_PATH}")
+
+if(NOT CMAKE_BUILD_TYPE)
+    set(CMAKE_BUILD_TYPE Debug CACHE STRING "Default build type to Debug" FORCE)
+endif()
+
+if(CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang")
+    string(REGEX MATCHALL "-std=[^ ]+" cxx_std_found ${CMAKE_CXX_FLAGS} " dummy@rg")
+    if(cxx_std_found)
+        message("C++ std: ${cxx_std_found}")
+    else()
+        list(APPEND LIBCDS_PUBLIC_CXX_FLAGS "-std=c++11")
+        message("C++ std: -std=c++11 (default)")
+    endif()
+
+    list(APPEND LIBCDS_PRIVATE_CXX_FLAGS "-Wall" "-Wextra" "-pedantic")
+
+    if(CMAKE_TARGET_ARCHITECTURE STREQUAL "x86_64")
+        list(APPEND LIBCDS_PUBLIC_CXX_FLAGS "-mcx16")
+        set(LIB_SUFFIX "64")
+
+        # GCC-7: 128-bit atomics support is implemented via libatomic on amd64
+        #        see https://gcc.gnu.org/ml/gcc/2017-01/msg00167.html
+        # Maybe, it will be changed in future
+        if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER "7.0.0" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS "8.0.0")
+            set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -latomic")
+        endif()
+    endif()
+endif()
+
+
+if(CMAKE_SYSTEM_NAME STREQUAL "AIX")
+    set(CMAKE_CXX_ARCHIVE_CREATE "<CMAKE_AR> -q -c ${CMAKE_STATIC_LINKER_FLAGS} -o <TARGET> <OBJECTS>")
+    list(APPEND LIBCDS_PRIVATE_CXX_FLAGS "-Wl,-G")
+    set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,-brtl")
+endif()
+
+set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -D_DEBUG")
+
+CHECK_INCLUDE_FILE_CXX(linux/membarrier.h CDS_HAVE_LINUX_MEMBARRIER_H CMAKE_CXX_FLAGS)
+
+message("Build type -- ${CMAKE_BUILD_TYPE}")
+message("Compiler version: ${CMAKE_CXX_COMPILER_ID} ${CMAKE_CXX_COMPILER_VERSION}")
+message("System: ${CMAKE_SYSTEM_NAME} version: ${CMAKE_SYSTEM_VERSION}")
+message("Target architecture: ${CMAKE_TARGET_ARCHITECTURE}")
+if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+    message("Compiler flags: ${CMAKE_CXX_FLAGS} ${CMAKE_CXX_FLAGS_DEBUG} ${LIBCDS_PUBLIC_CXX_FLAGS} ${LIBCDS_PRIVATE_CXX_FLAGS}")
+else()
+    message("Compiler flags: ${CMAKE_CXX_FLAGS} ${CMAKE_CXX_FLAGS_RELEASE} ${LIBCDS_PUBLIC_CXX_FLAGS} ${LIBCDS_PRIVATE_CXX_FLAGS}")
+endif()
+message("Exe flags: ${CMAKE_EXE_LINKER_FLAGS}")
+
+# Component names for separate distribution in rpms, debs etc.
+set(LIBRARIES_COMPONENT lib)
+set(HEADERS_COMPONENT devel)
+
+set(SOURCES src/init.cpp
+            src/hp.cpp
+            src/dhp.cpp
+            src/urcu_gp.cpp
+            src/urcu_sh.cpp
+            src/thread_data.cpp
+            src/topology_hpux.cpp
+            src/topology_linux.cpp
+            src/topology_osx.cpp
+            src/dllmain.cpp)
+
+add_library(${CDS_SHARED_LIBRARY} SHARED ${SOURCES})
+set_target_properties(${CDS_SHARED_LIBRARY} PROPERTIES VERSION ${PROJECT_VERSION}
+        DEBUG_POSTFIX "_d")
+if(MINGW)
+    set_target_properties(${CDS_SHARED_LIBRARY} PROPERTIES DEFINE_SYMBOL CDS_BUILD_LIB)
+endif()
+
+add_library(${CDS_STATIC_LIBRARY} STATIC ${SOURCES})
+set_target_properties(${CDS_STATIC_LIBRARY} PROPERTIES DEBUG_POSTFIX "_d")
+if(MINGW)
+    target_compile_definitions(${CDS_STATIC_LIBRARY} PRIVATE CDS_BUILD_STATIC_LIB)
+endif()
+
+target_link_libraries(${CDS_SHARED_LIBRARY} PRIVATE ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${CDS_STATIC_LIBRARY} PRIVATE ${CMAKE_THREAD_LIBS_INIT})
+target_include_directories(${CDS_SHARED_LIBRARY} INTERFACE "$<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}>"
+                                                            $<INSTALL_INTERFACE:include>)
+target_include_directories(${CDS_STATIC_LIBRARY} INTERFACE "$<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}>"
+                                                            $<INSTALL_INTERFACE:include>)
+target_compile_options(${CDS_SHARED_LIBRARY} PUBLIC "${LIBCDS_PUBLIC_CXX_FLAGS}")
+target_compile_options(${CDS_STATIC_LIBRARY} PUBLIC "${LIBCDS_PUBLIC_CXX_FLAGS}")
+target_compile_options(${CDS_SHARED_LIBRARY} PRIVATE "${LIBCDS_PRIVATE_CXX_FLAGS}")
+target_compile_options(${CDS_STATIC_LIBRARY} PRIVATE "${LIBCDS_PRIVATE_CXX_FLAGS}")
+
+install(TARGETS ${CDS_SHARED_LIBRARY} EXPORT LibCDSConfig LIBRARY DESTINATION lib${LIB_SUFFIX} COMPONENT ${LIBRARIES_COMPONENT} NAMELINK_SKIP RUNTIME DESTINATION lib${LIB_SUFFIX})
+install(TARGETS ${CDS_SHARED_LIBRARY} EXPORT LibCDSConfig LIBRARY DESTINATION lib${LIB_SUFFIX} COMPONENT ${HEADERS_COMPONENT} NAMELINK_ONLY)
+install(TARGETS ${CDS_STATIC_LIBRARY} EXPORT LibCDSConfig DESTINATION lib${LIB_SUFFIX} COMPONENT ${LIBRARIES_COMPONENT})
+install(EXPORT LibCDSConfig FILE LibCDSConfig.cmake NAMESPACE LibCDS:: DESTINATION lib/cmake/LibCDS)
+install(DIRECTORY ${PROJECT_SOURCE_DIR}/cds DESTINATION include COMPONENT ${HEADERS_COMPONENT})
+
+if(WITH_TESTS)
+    enable_testing()
+    add_subdirectory(${PROJECT_SOURCE_DIR}/test)
+    message(STATUS "Build tests: activated")
+endif(WITH_TESTS)
+
+### FOR PACKAGING in RPM, TGZ, DEB, NSYS...###############################################################################
+set(CPACK_PACKAGE_VERSION ${PROJECT_VERSION})
+set(CPACK_PACKAGE_NAME ${PROJECT_NAME})
+set(CPACK_PACKAGE_CONTACT "Max Khizhinsky <libcds-user@lists.sourceforge.net>")
+set(CPACK_PACKAGE_RELEASE 1)
+set(CPACK_PACKAGE_INSTALL_DIRECTORY "cds")
+set(CPACK_PACKAGE_DESCRIPTION_FILE "${PROJECT_SOURCE_DIR}/build/cmake/description.txt")
+set(CPACK_PACKAGE_DESCRIPTION_SUMMARY "Library of concurrent data structures")
+set(CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION}-${CPACK_PACKAGE_RELEASE}")
+set(DEPLOY_PACKAGE_FILE_NAME "${CPACK_PACKAGE_FILE_NAME}")
+
+# TGZ specific
+set(CPACK_ARCHIVE_COMPONENT_INSTALL ON)
+
+# RPM specific
+set(CPACK_RPM_COMPONENT_INSTALL ON)
+set(CPACK_RPM_PACKAGE_RELEASE ${CPACK_PACKAGE_RELEASE})
+set(CPACK_RPM_POST_INSTALL_SCRIPT_FILE "${PROJECT_SOURCE_DIR}/build/cmake/post_install_script.sh")
+set(CPACK_RPM_POST_UNINSTALL_SCRIPT_FILE "${PROJECT_SOURCE_DIR}/build/cmake/post_uninstall_script.sh")
+set(CPACK_RPM_PACKAGE_URL https://github.com/khizmax/libcds)
+set(CPACK_RPM_PACKAGE_LICENSE GPL)
+set(CPACK_RPM_PACKAGE_GROUP "System Environment/Base")
+set(CPACK_RPM_PACKAGE_REQUIRES "boost >= 1.50")
+set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION ${CPACK_PACKAGING_INSTALL_PREFIX})
+set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION /usr/local)
+set(CPACK_RPM_devel_PACKAGE_REQUIRES "boost >= 1.50, cds-lib = ${PROJECT_VERSION}")
+
+# DEB specific
+set(CPACK_DEB_COMPONENT_INSTALL ON)
+set(CPACK_DEBIAN_PACKAGE_DEPENDS "boost (>= 1.50)")
+set(CPACK_DEBIAN_PACKAGE_HOMEPAGE "https://github.com/khizmax/libcds")
+set(CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${PROJECT_SOURCE_DIR}/build/cmake/post_install_script.sh;;${PROJECT_SOURCE_DIR}/build/cmake/post_uninstall_script.sh;")
+
+# NSYS specific
+set(CPACK_NSIS_PACKAGE_NAME "${CPACK_PACKAGE_NAME}")
+set(CPACK_NSIS_DISPLAY_NAME "${CPACK_PACKAGE_NAME}")
+set(CPACK_NSIS_CONTACT ${CPACK_PACKAGE_CONTACT})
+set(CPACK_NSIS_ENABLE_UNINSTALL_BEFORE_INSTALL ON)
+set(CPACK_NSIS_MODIFY_PATH ON)
+
+# Components grouping for Mac OS X and Windows installers
+set(CPACK_COMPONENT_${LIBRARIES_COMPONENT}_GROUP "Runtime")
+set(CPACK_COMPONENT_${HEADERS_COMPONENT}_GROUP "Development")
+set(CPACK_COMPONENT_${LIBRARIES_COMPONENT}_DISPLAY_NAME "Libraries")
+set(CPACK_COMPONENT_${HEADERS_COMPONENT}_DISPLAY_NAME "C++ Headers")
+set(CPACK_COMPONENT_${HEADERS_COMPONENT}_DEPENDS ${LIBRARIES_COMPONENT})
+set(CPACK_COMPONENT_GROUP_DEVELOPMENT_DESCRIPTION "All of the tools you'll ever need to develop lock-free oriented software with libcds")
+set(CPACK_COMPONENT_GROUP_RUNTIME_DESCRIPTION "Only libcds library for runtime")
+
+include(CPack)

extern/libcds/LICENSE

@@ -0,0 +1,23 @@
+Boost Software License - Version 1.0 - August 17th, 2003
+
+Permission is hereby granted, free of charge, to any person or organization
+obtaining a copy of the software and accompanying documentation covered by
+this license (the "Software") to use, reproduce, display, distribute,
+execute, and transmit the Software, and to prepare derivative works of the
+Software, and to permit third-parties to whom the Software is furnished to
+do so, all subject to the following:
+
+The copyright notices in the Software and this entire statement, including
+the above license grant, this restriction and the following disclaimer,
+must be included in all copies of the Software, in whole or in part, and
+all derivative works of the Software, unless such copies or derivative
+works are solely in the form of machine-executable object code generated by
+a source language processor.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
+SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
+FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.

extern/libcds/appveyor.yml

@@ -0,0 +1,91 @@
+build: false
+
+shallow_clone: true              # (git clone --depth 1)
+
+image:
+  - Visual Studio 2017
+
+install:
+   - sed -i "/boost/d" conanfile.txt # delete boost from conanfile. Conan renamed name of boost libs so we'll use appveyour's boost
+   - cmd: echo "Downloading conan..."
+   - cmd: set PATH=%PATH%;%PYTHON%/Scripts/
+   - cmd: pip.exe install conan
+   - cmd: conan user # Create the conan data directory
+   - cmd: conan --version
+   - cmd: conan install --build=missing -s arch=x86 -s build_type=%configuration% .
+   #- cmd: conan install --build=missing -s arch=x86_64 -s build_type=%configuration% conanfileWin.txt
+   - cmd: echo =======================
+   - cmd: echo %configuration%
+   - cmd: echo %platform%
+   - cmd: echo =======================
+   - cmd: set GTEST_ROOT=C:/projects/libcds/deps
+   - cmd: set BOOST_PATH=C:\Libraries\boost_1_66_0
+   - cmd: set GTEST_LIB32=C:\projects\libcds\deps\lib;C:\Libraries\boost_1_66_0\lib32-msvc-14.1
+   - cmd: dir %GTEST_LIB32%
+   - cmd: echo =======================
+   # - cmd: set GTEST_LIB64=C:\projects\libcds\deps\lib
+   
+   
+environment:
+    matrix:
+    # require a library with name libboost_thread-vc141-mt-x32-1_66.lib,
+    - TARGET: gtest-stack
+    - TARGET: gtest-deque
+    - TARGET: gtest-pqueue
+    - TARGET: gtest-queue
+    
+    - TARGET: gtest-map-skip
+    - TARGET: gtest-map-split-michael
+    - TARGET: gtest-set-split-michael
+    - TARGET: gtest-misc
+    - TARGET: gtest-ilist-iterable
+    - TARGET: gtest-ilist-lazy
+    - TARGET: gtest-ilist-michael
+    - TARGET: gtest-iset-feldman
+    - TARGET: gtest-iset-michael
+    - TARGET: gtest-iset-michael-iterable
+    - TARGET: gtest-iset-michael-lazy
+    - TARGET: gtest-iset-skip
+    - TARGET: gtest-iset-split-iterable
+    - TARGET: gtest-iset-split-lazy
+    - TARGET: gtest-iset-split-michael
+    - TARGET: gtest-list-iterable
+    - TARGET: gtest-list-lazy
+    - TARGET: gtest-list-michael
+    - TARGET: gtest-map-feldman
+    - TARGET: gtest-map-michael
+    - TARGET: gtest-map-michael-iterable
+    - TARGET: gtest-map-michael-lazy
+    - TARGET: gtest-map-split-iterable
+    - TARGET: gtest-map-split-lazy
+    - TARGET: gtest-set-feldman
+    - TARGET: gtest-set-michael
+    - TARGET: gtest-set-michael-iterable
+    - TARGET: gtest-set-michael-lazy
+    - TARGET: gtest-set-skip
+    - TARGET: gtest-set-split-iterable
+    - TARGET: gtest-set-split-lazy
+    - TARGET: gtest-striped-map-boost
+    - TARGET: gtest-striped-map-cuckoo
+    - TARGET: gtest-striped-map-std
+    - TARGET: gtest-striped-set-boost
+    - TARGET: gtest-striped-set-cuckoo
+    - TARGET: gtest-striped-set-std
+    - TARGET: gtest-tree-bronson
+    - TARGET: gtest-tree-ellen
+configuration: 
+  - Release
+platform:
+  - Win32
+  
+build_script:
+     - msbuild projects/Win/vc141/cds.vcxproj /p:Configuration=%configuration% /p:PlatformTarget="Win32"
+     - msbuild projects/Win/vc141/%TARGET%.vcxproj /p:Configuration=%configuration% /p:PlatformTarget="Win32"
+     #- msbuild projects/Win/vc141/cds.vcxproj /p:Configuration=%configuration% /p:PlatformTarget=x64 
+     #- msbuild projects/Win/vc141/%TARGET%.vcxproj /p:Configuration=%configuration% /p:PlatformTarget=x64
+     
+test_script:
+     - cmd: set path=%path%;%GTEST_LIB32%;C:\projects\libcds\bin\vc.v141\%platform%-release\
+     - cmd: C:\projects\libcds\bin\vc.v141\%platform%-release\%TARGET%.exe
+      
+  

extern/libcds/build/CI/VASEx-CI-2/cds-libs

@@ -0,0 +1,239 @@
+#########################################
+# Generic parameters
+
+workspace:              $WORKSPACE
+libcds-source:          source
+make-job:               10
+gtest-include:          $GTEST_ROOT/googletest/include
+
+#########################################
+#GCC-4.8
+gcc-4.8-root:           $GCC48_ROOT/bin
+gcc-4.8-cxx:            g++-4.8
+gcc-4.8-cc:             gcc-4.8
+gcc-4.8-exe-ldflags:    -L$GCC48_ROOT/lib64 -Wl,-rpath=$GCC48_ROOT/lib64
+gcc-4.8-extlib:         rt
+gcc-4.8-boost:          $BOOST_ROOT
+gcc-4.8-64-boost-lib:   stage64-gcc4.8/lib
+gcc-4.8-gtest:          $GTEST_ROOT
+gcc-4.8-64-gtest-lib:   $GTEST_ROOT/lib-gcc4.8/libgtest.a
+
+
+########################################
+#GCC-4.9
+gcc-4.9-root:           $GCC49_ROOT/bin
+gcc-4.9-cxx:            g++-4.9
+gcc-4.9-cc:             gcc-4.9
+gcc-4.9-exe-ldflags:    -Wl,-rpath=$GCC49_ROOT/lib64
+gcc-4.9-extlib:         rt
+gcc-4.9-boost:          $BOOST_ROOT
+gcc-4.9-64-boost-lib:   stage64-gcc4.9/lib
+gcc-4.9-gtest:          $GTEST_ROOT
+gcc-4.9-64-gtest-lib:   $GTEST_ROOT/lib-gcc4.9/libgtest.a
+
+########################################
+#GCC-5
+gcc-5-root:           $GCC5_ROOT/bin
+gcc-5-cxx:            g++-5
+gcc-5-cc:             gcc-5
+gcc-5-boost:          $BOOST_ROOT
+gcc-5-exe-ldflags:    -Wl,-rpath=$GCC5_ROOT/lib64
+gcc-5-extlib:         rt
+gcc-5-64-boost-lib:   stage64-gcc5/lib
+gcc-5-64-asan-boost-lib:   stage64-gcc5-asan/lib
+gcc-5-64-tsan-boost-lib:   stage64-gcc5-tsan/lib
+gcc-5-gtest:          $GTEST_ROOT
+gcc-5-64-gtest-lib:   $GTEST_ROOT/lib-gcc5/libgtest.a
+
+########################################
+#GCC-6
+gcc-6-root:           $GCC6_ROOT/bin
+gcc-6-cxx:            g++-6
+gcc-6-cc:             gcc-6
+gcc-6-boost:          $BOOST_ROOT
+gcc-6-cxxflags:       -march=native  -std=c++14
+gcc-6-exe-ldflags:    -Wl,-rpath=$GCC6_ROOT/lib64
+gcc-6-extlib:         rt
+gcc-6-64-boost-lib:   stage64-gcc6/lib
+gcc-6-64-asan-boost-lib:   stage64-gcc6-asan/lib
+gcc-6-64-tsan-boost-lib:   stage64-gcc6-tsan/lib
+gcc-6-gtest:          $GTEST_ROOT
+gcc-6-64-gtest-lib:   $GTEST_ROOT/lib-gcc6/libgtest.a
+
+########################################
+#GCC-7
+gcc-7-root:           $GCC7_ROOT/bin
+gcc-7-cxx:            g++-7
+gcc-7-cc:             gcc-7
+gcc-7-boost:          $BOOST_ROOT
+gcc-7-cxxflags:       -march=native  -std=c++1z
+gcc-7-exe-ldflags:    -Wl,-rpath=$GCC7_ROOT/lib64
+gcc-7-extlib:         rt
+gcc-7-64-boost-lib:   stage64-gcc7/lib
+gcc-7-64-asan-boost-lib:   stage64-gcc7-asan/lib
+gcc-7-64-tsan-boost-lib:   stage64-gcc7-tsan/lib
+gcc-7-gtest:          $GTEST_ROOT
+gcc-7-64-gtest-lib:   $GTEST_ROOT/lib-gcc7/libgtest.a
+
+########################################
+#GCC-8
+gcc-8-root:           $GCC8_ROOT/bin
+gcc-8-cxx:            g++-8
+gcc-8-cc:             gcc-8
+gcc-8-boost:          $BOOST_ROOT
+gcc-8-cxxflags:       -march=native  -std=c++17 -Wmultistatement-macros
+gcc-8-exe-ldflags:    -Wl,-rpath=$GCC8_ROOT/lib64
+gcc-8-extlib:         rt
+gcc-8-path:           $DEVTOOLSET6_BIN
+gcc-8-64-boost-lib:   stage64-gcc7/lib
+gcc-8-64-asan-boost-lib:   stage64-gcc7-asan/lib
+gcc-8-64-tsan-boost-lib:   stage64-gcc7-tsan/lib
+gcc-8-gtest:          $GTEST_ROOT
+gcc-8-64-gtest-lib:   $GTEST_ROOT/lib-gcc7/libgtest.a
+
+########################################
+# clang-3.6
+clang-3.6-root:           $CLANG36_ROOT/bin
+clang-3.6-ld-lib-path:    $GCC5_ROOT/lib64
+clang-3.6-cxx:            clang++
+clang-3.6-cc:             clang
+clang-3.6-cxxflags:       -Wdocumentation 
+clang-3.6-exe-ldflags:    -L$GCC5_ROOT/lib64 -latomic -Wl,-rpath=$GCC5_ROOT/lib64
+clang-3.6-extlib:         rt
+clang-3.6-boost:          $BOOST_ROOT
+clang-3.6-64-boost-lib:   stage64-clang3.6/lib
+clang-3.6-gtest:          $GTEST_ROOT
+clang-3.6-64-gtest-lib:   $GTEST_ROOT/lib-clang3.6/libgtest.a
+
+########################################
+# clang-3.7
+clang-3.7-root:           $CLANG37_ROOT/bin
+clang-3.7-ld-lib-path:    $GCC6_ROOT/lib64
+clang-3.7-cxx:            clang++
+clang-3.7-cc:             clang
+clang-3.7-cxxflags:       -stdlib=libc++ -Wdocumentation
+clang-3.7-exe-ldflags:    -L$CLANG37_ROOT/lib -Wl,-rpath=$CLANG37_ROOT/lib -lc++abi
+clang-3.7-extlib:         rt
+clang-3.7-boost:          $BOOST_ROOT
+clang-3.7-64-boost-lib:   stage64-clang3.7/lib
+clang-3.7-gtest:          $GTEST_ROOT
+clang-3.7-64-gtest-lib:   $GTEST_ROOT/lib-clang3.7/libgtest.a
+clang-3.7-cmake-flags:    -DCMAKE_C_COMPILER_WORKS=1 -DCMAKE_CXX_COMPILER_WORKS=1
+
+
+########################################
+# clang-3.8
+clang-3.8-root:           $CLANG38_ROOT/bin
+clang-3.8-ld-lib-path:    $GCC6_ROOT/lib64
+clang-3.8-cxx:            clang++
+clang-3.8-cc:             clang
+clang-3.8-cxxflags:       -stdlib=libc++ -Wdocumentation
+clang-3.8-exe-ldflags:    -L$CLANG38_ROOT/lib -Wl,-rpath=$CLANG38_ROOT/lib
+clang-3.8-extlib:         rt
+clang-3.8-boost:          $BOOST_ROOT
+clang-3.8-64-boost-lib:   stage64-clang3.8/lib
+clang-3.8-gtest:          $GTEST_ROOT
+clang-3.8-64-gtest-lib:   $GTEST_ROOT/lib-clang3.8/libgtest.a
+
+
+########################################
+# clang-3.9
+clang-3.9-root:           $CLANG39_ROOT/bin
+clang-3.9-ld-lib-path:    $GCC6_ROOT/lib64
+clang-3.9-cxx:            clang++
+clang-3.9-cc:             clang
+clang-3.9-cxxflags:       -stdlib=libc++ -Wdocumentation
+clang-3.9-exe-ldflags:    -L$CLANG39_ROOT/lib -Wl,-rpath=$CLANG39_ROOT/lib
+clang-3.9-extlib:         rt
+clang-3.9-boost:          $BOOST_ROOT
+clang-3.9-64-boost-lib:   stage64-clang3.9/lib
+clang-3.9-64-asan-boost-lib:   stage64-clang3.9-asan/lib
+clang-3.9-64-tsan-boost-lib:   stage64-clang3.9-tsan/lib
+clang-3.9-gtest:          $GTEST_ROOT
+clang-3.9-64-gtest-lib:   $GTEST_ROOT/lib-clang3.9/libgtest.a
+
+
+########################################
+# clang-4
+clang-4-root:           $CLANG4_ROOT/bin
+clang-4-cxx:            clang++
+clang-4-cc:             clang
+clang-4-cxxflags:       -stdlib=libc++ -Wdocumentation -std=c++14 
+clang-4-exe-ldflags:    -L$CLANG4_ROOT/lib -Wl,-rpath=$CLANG4_ROOT/lib
+clang-4-extlib:         rt
+clang-4-boost:          $BOOST_ROOT
+clang-4-64-boost-lib:   stage64-clang4/lib
+clang-4-64-asan-boost-lib:   stage64-clang4-asan/lib
+clang-4-64-tsan-boost-lib:   stage64-clang4-tsan/lib
+clang-4-gtest:          $GTEST_ROOT
+clang-4-64-gtest-lib:   $GTEST_ROOT/lib-clang4/libgtest.a
+
+
+########################################
+# clang-5
+clang-5-root:           $CLANG5_ROOT/bin
+clang-5-cxx:            clang++
+clang-5-cc:             clang
+clang-5-cxxflags:       -stdlib=libc++ -Wdocumentation -std=c++1z
+clang-5-exe-ldflags:    -L$CLANG5_ROOT/lib -Wl,-rpath=$CLANG5_ROOT/lib
+clang-5-extlib:         rt
+clang-5-path:           $DEVTOOLSET6_BIN
+clang-5-boost:          $LIB_ROOT/boost_1_65_1
+clang-5-64-boost-lib:   stage64-clang5-std17/lib
+clang-5-64-asan-boost-lib:   stage64-clang5-asan/lib
+clang-5-64-tsan-boost-lib:   stage64-clang5-tsan/lib
+clang-5-gtest:          $GTEST_ROOT
+clang-5-64-gtest-lib:   $GTEST_ROOT/lib-clang5/libgtest.a
+
+
+########################################
+# clang-6
+clang-6-root:           $CLANG6_ROOT/bin
+clang-6-cxx:            clang++
+clang-6-cc:             clang
+clang-6-cxxflags:       -stdlib=libc++ -Wdocumentation -std=c++17
+clang-6-exe-ldflags:    -L$CLANG6_ROOT/lib -Wl,-rpath=$CLANG6_ROOT/lib
+clang-6-extlib:         rt
+clang-6-path:           $DEVTOOLSET6_BIN
+clang-6-boost:          $LIB_ROOT/boost_1_65_1
+clang-6-64-boost-lib:   stage64-clang6-std17/lib
+clang-6-64-asan-boost-lib:   stage64-clang6-asan/lib
+clang-6-64-tsan-boost-lib:   stage64-clang6-tsan/lib
+clang-6-gtest:          $GTEST_ROOT
+clang-6-64-gtest-lib:   $GTEST_ROOT/lib-clang6/libgtest.a
+
+
+########################################
+# clang-7
+clang-7-root:           $CLANG6_ROOT/bin
+clang-7-cxx:            clang++
+clang-7-cc:             clang
+clang-7-cxxflags:       -stdlib=libc++ -Wdocumentation -std=c++17
+clang-7-exe-ldflags:    -fuse-ld=lld -L$CLANG7_ROOT/lib -Wl,-rpath=$CLANG7_ROOT/lib
+clang-7-extlib:         rt
+clang-7-path:           $DEVTOOLSET6_BIN
+clang-7-boost:          $LIB_ROOT/boost_1_65_1
+clang-7-64-boost-lib:   stage64-clang7-std17/lib
+clang-7-64-asan-boost-lib:   stage64-clang7-asan/lib
+clang-7-64-tsan-boost-lib:   stage64-clang7-tsan/lib
+clang-7-gtest:          $GTEST_ROOT
+clang-7-64-gtest-lib:   $GTEST_ROOT/lib-clang6/libgtest.a
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

extern/libcds/build/CI/VASEx-CI-2/ci-build

@@ -0,0 +1,88 @@
+#! /bin/bash
+
+# Useful envvars:
+# CI_SCRIPT_PATH - path where to find scripts
+# TOOLSET - toolset: x64-gcc-5, x64-clang-3.9 and so on
+# BUILD_TYPE - build type: 'dbg', 'rel', 'asan', 'tsan'
+# WORKSPACE - path where to build
+
+env|sort
+
+case "$TOOLSET" in
+    "x64-gcc-4.8")
+	echo "GCC-4.8 '$BUILD_TYPE', toolset root: $GCC48_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-4.8-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-gcc-4.9")
+	echo "GCC-4.9 '$BUILD_TYPE', toolset root: $GCC49_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-4.9-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-gcc-5")
+	echo "GCC-5 '$BUILD_TYPE', toolset root: $GCC5_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-5-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-gcc-6")
+	echo "GCC-6 '$BUILD_TYPE', toolset root: $GCC6_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-6-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-gcc-7")
+	echo "GCC-7 '$BUILD_TYPE', toolset root: $GCC7_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-7-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-gcc-8")
+	echo "GCC-8 '$BUILD_TYPE', toolset root: $GCC8_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-8-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-clang-3.6")
+	echo "clang-3.6 '$BUILD_TYPE', toolset root: $CLANG36_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-3.6-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-clang-3.7")
+	echo "clang-3.7 '$BUILD_TYPE', toolset root: $CLANG37_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-3.7-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-clang-3.8")
+	echo "clang-3.8 '$BUILD_TYPE', toolset root: $CLANG38_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-3.8-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-clang-3.9")
+	echo "clang-3.9 '$BUILD_TYPE', toolset root: $CLANG39_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-3.9-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-clang-4")
+	echo "clang-4 '$BUILD_TYPE', toolset root: $CLANG4_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-4-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-clang-5")
+	echo "clang-5 '$BUILD_TYPE', toolset root: $CLANG5_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-5-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-clang-6")
+	echo "clang-6 '$BUILD_TYPE', toolset root: $CLANG6_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-6-64 $*
+	EXIT_CODE=$?
+	;;
+    "x64-clang-7")
+	echo "clang-7 '$BUILD_TYPE', toolset root: $CLANG7_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-7-64 $*
+	EXIT_CODE=$?
+	;;
+    * )
+	echo "Undefined toolset '$TOOLSET'"
+	exit 1
+	;;
+esac
+
+exit $EXIT_CODE

extern/libcds/build/CI/VASEx-CI-2/ci-env

@@ -0,0 +1,38 @@
+#CMAKE_2_8_12=/home/libcds-ci/bin/cmake/cmake-2.8.12/bin
+CMAKE_3_6_2=/home/libcds-ci/bin/cmake-3.6/bin
+CMAKE3=$CMAKE_3_6_2
+
+PATH=$CMAKE3:$PATH:$HOME/.local/bin:$HOME/bin
+
+DEVTOOLSET6_BIN=/opt/rh/devtoolset-6/root/usr/bin
+
+TOOLSET_ROOT=$HOME/bin
+
+GCC48_ROOT=$TOOLSET_ROOT/gcc-4.8
+GCC49_ROOT=$TOOLSET_ROOT/gcc-4.9
+GCC5_ROOT=$TOOLSET_ROOT/gcc-5
+GCC6_ROOT=$TOOLSET_ROOT/gcc-6
+GCC7_ROOT=$TOOLSET_ROOT/gcc-7
+GCC8_ROOT=$TOOLSET_ROOT/gcc-8
+
+CLANG36_ROOT=$TOOLSET_ROOT/clang-3.6
+CLANG37_ROOT=$TOOLSET_ROOT/clang-3.7
+CLANG38_ROOT=$TOOLSET_ROOT/clang-3.8
+CLANG39_ROOT=$TOOLSET_ROOT/clang-3.9
+CLANG4_ROOT=$TOOLSET_ROOT/clang-4
+CLANG5_ROOT=$TOOLSET_ROOT/clang-5
+CLANG6_ROOT=$TOOLSET_ROOT/clang-6
+CLANG7_ROOT=$TOOLSET_ROOT/clang-7
+
+CLANG_STDLIB="-stdlib=libc++"
+CLANG37_CXXFLAGS=$CLANG_STDLIB
+CLANG38_CXXFLAGS=$CLANG_STDLIB
+CLANG39_CXXFLAGS=$CLANG_STDLIB
+CLANG4_CXXFLAGS=$CLANG_STDLIB
+CLANG5_CXXFLAGS=$CLANG_STDLIB
+CLANG6_CXXFLAGS=$CLANG_STDLIB
+CLANG7_CXXFLAGS="$CLANG_STDLIB -fuse-ld=lld"
+
+LIB_ROOT=$HOME/lib
+BOOST_ROOT=$LIB_ROOT/boost
+GTEST_ROOT=$LIB_ROOT/gtest

extern/libcds/build/CI/VASEx-CI/cds-libs

@@ -0,0 +1,195 @@
+#########################################
+# Generic parameters
+
+workspace:              $WORKSPACE
+libcds-source:          source
+make-job:               10
+gtest-include:          $GTEST_ROOT/googletest/include
+
+#########################################
+#GCC-4.8
+gcc-4.8-root:           $GCC48_ROOT/bin
+gcc-4.8-cxx:            g++-4.8
+gcc-4.8-cc:             gcc-4.8
+gcc-4.8-exe-ldflags:    -L$GCC48_ROOT/lib64 -Wl,-rpath=$GCC48_ROOT/lib64
+gcc-4.8-boost:          $BOOST_ROOT
+gcc-4.8-64-boost-lib:   stage64-gcc4.8/lib
+gcc-4.8-gtest:          $GTEST_ROOT
+gcc-4.8-64-gtest-lib:   $GTEST_ROOT/lib-gcc4.8/libgtest.a
+
+
+########################################
+#GCC-4.9
+gcc-4.9-root:           $GCC49_ROOT/bin
+gcc-4.9-cxx:            g++-4.9
+gcc-4.9-cc:             gcc-4.9
+gcc-4.9-exe-ldflags:    -Wl,-rpath=$GCC49_ROOT/lib64
+gcc-4.9-boost:          $BOOST_ROOT
+gcc-4.9-64-boost-lib:   stage64-gcc4.9/lib
+gcc-4.9-gtest:          $GTEST_ROOT
+gcc-4.9-64-gtest-lib:   $GTEST_ROOT/lib-gcc4.9/libgtest.a
+
+########################################
+#GCC-5
+gcc-5-root:           $GCC5_ROOT/bin
+gcc-5-cxx:            g++-5
+gcc-5-cc:             gcc-5
+gcc-5-boost:          $BOOST_ROOT
+gcc-5-exe-ldflags:    -Wl,-rpath=$GCC5_ROOT/lib64
+gcc-5-64-boost-lib:   stage64-gcc5/lib
+gcc-5-64-asan-boost-lib:   stage64-gcc5-asan/lib
+gcc-5-64-tsan-boost-lib:   stage64-gcc5-tsan/lib
+gcc-5-gtest:          $GTEST_ROOT
+gcc-5-64-gtest-lib:   $GTEST_ROOT/lib-gcc5/libgtest.a
+
+########################################
+#GCC-6
+gcc-6-root:           $GCC6_ROOT/bin
+gcc-6-cxx:            g++-6
+gcc-6-cc:             gcc-6
+gcc-6-boost:          $BOOST_ROOT
+gcc-6-cxxflags:       -march=native -std=c++14
+gcc-6-exe-ldflags:    -Wl,-rpath=$GCC6_ROOT/lib64
+gcc-6-64-boost-lib:   stage64-gcc6/lib
+gcc-6-64-asan-boost-lib:   stage64-gcc6-asan/lib
+gcc-6-64-tsan-boost-lib:   stage64-gcc6-tsan/lib
+gcc-6-gtest:          $GTEST_ROOT
+gcc-6-64-gtest-lib:   $GTEST_ROOT/lib-gcc6/libgtest.a
+
+
+########################################
+#GCC-7
+gcc-7-root:           $GCC7_ROOT/bin
+gcc-7-cxx:            g++-7
+gcc-7-cc:             gcc-7
+gcc-7-boost:          $BOOST_ROOT
+gcc-7-cxxflags:       -march=native -std=c++1z
+gcc-7-exe-ldflags:    -Wl,-rpath=$GCC7_ROOT/lib64
+gcc-7-64-boost-lib:   stage64-gcc7/lib
+gcc-7-64-asan-boost-lib:   stage64-gcc7-asan/lib
+gcc-7-64-tsan-boost-lib:   stage64-gcc7-tsan/lib
+gcc-7-gtest:          $GTEST_ROOT
+gcc-7-64-gtest-lib:   $GTEST_ROOT/lib-gcc7/libgtest.a
+
+########################################
+#GCC-8
+gcc-8-root:           $GCC8_ROOT/bin
+gcc-8-cxx:            g++-8
+gcc-8-cc:             gcc-8
+gcc-8-boost:          $BOOST_ROOT
+gcc-8-cxxflags:       -march=native  -std=c++17 -Wmultistatement-macros
+gcc-8-exe-ldflags:    -Wl,-rpath=$GCC8_ROOT/lib64
+gcc-8-extlib:         rt
+gcc-8-64-boost-lib:   stage64-gcc7/lib
+gcc-8-64-asan-boost-lib:   stage64-gcc7-asan/lib
+gcc-8-64-tsan-boost-lib:   stage64-gcc7-tsan/lib
+gcc-8-gtest:          $GTEST_ROOT
+gcc-8-64-gtest-lib:   $GTEST_ROOT/lib-gcc7/libgtest.a
+
+
+########################################
+# clang-3.6
+clang-3.6-root:           $CLANG36_ROOT/bin
+clang-3.6-ld-lib-path:    $GCC6_ROOT/lib64
+clang-3.6-cxx:            clang++
+clang-3.6-cc:             clang
+clang-3.6-cxxflags:       -Wdocumentation
+clang-3.6-exe-ldflags:    -L$GCC5_ROOT/lib64 -latomic -Wl,-rpath=$GCC5_ROOT/lib64
+clang-3.6-boost:          $BOOST_ROOT
+clang-3.6-64-boost-lib:   stage64-clang3.6/lib
+clang-3.6-gtest:          $GTEST_ROOT
+clang-3.6-64-gtest-lib:   $GTEST_ROOT/lib-clang3.6/libgtest.a
+
+########################################
+# clang-3.7
+clang-3.7-root:           $CLANG37_ROOT/bin
+clang-3.7-cxx:            clang++
+clang-3.7-cc:             clang
+clang-3.7-cxxflags:       -stdlib=libc++ -Wdocumentation
+clang-3.7-exe-ldflags:    -L$CLANG37_ROOT/lib -Wl,-rpath=$CLANG37_ROOT/lib -lc++abi
+clang-3.7-boost:          $BOOST_ROOT
+clang-3.7-64-boost-lib:   stage64-clang3.7/lib
+clang-3.7-gtest:          $GTEST_ROOT
+clang-3.7-64-gtest-lib:   $GTEST_ROOT/lib-clang3.7/libgtest.a
+clang-3.7-cmake-flags:    -DCMAKE_C_COMPILER_WORKS=1 -DCMAKE_CXX_COMPILER_WORKS=1
+
+########################################
+# clang-3.8
+clang-3.8-root:           $CLANG38_ROOT/bin
+clang-3.8-cxx:            clang++
+clang-3.8-cc:             clang
+clang-3.8-cxxflags:       -stdlib=libc++ -Wdocumentation
+clang-3.8-exe-ldflags:    -L$CLANG38_ROOT/lib -Wl,-rpath=$CLANG38_ROOT/lib
+clang-3.8-boost:          $BOOST_ROOT
+clang-3.8-64-boost-lib:   stage64-clang3.8/lib
+clang-3.8-gtest:          $GTEST_ROOT
+clang-3.8-64-gtest-lib:   $GTEST_ROOT/lib-clang3.8/libgtest.a
+
+
+########################################
+# clang-3.9
+clang-3.9-root:           $CLANG39_ROOT/bin
+clang-3.9-cxx:            clang++
+clang-3.9-cc:             clang
+clang-3.9-cxxflags:       -stdlib=libc++ -Wdocumentation
+clang-3.9-exe-ldflags:    -L$CLANG39_ROOT/lib -Wl,-rpath=$CLANG39_ROOT/lib
+clang-3.9-boost:          $BOOST_ROOT
+clang-3.9-64-boost-lib:   stage64-clang3.9/lib
+clang-3.9-64-asan-boost-lib:   stage64-clang3.9-asan/lib
+clang-3.9-64-tsan-boost-lib:   stage64-clang3.9-tsan/lib
+clang-3.9-gtest:          $GTEST_ROOT
+clang-3.9-64-gtest-lib:   $GTEST_ROOT/lib-clang3.9/libgtest.a
+
+
+########################################
+# clang-4
+clang-4-root:           $CLANG4_ROOT/bin
+clang-4-cxx:            clang++
+clang-4-cc:             clang
+clang-4-cxxflags:       -stdlib=libc++ -Wdocumentation -std=c++14
+clang-4-exe-ldflags:    -L$CLANG4_ROOT/lib -Wl,-rpath=$CLANG4_ROOT/lib
+clang-4-boost:          $BOOST_ROOT
+clang-4-64-boost-lib:   stage64-clang4/lib
+clang-4-64-asan-boost-lib:   stage64-clang4-asan/lib
+clang-4-64-tsan-boost-lib:   stage64-clang4-tsan/lib
+clang-4-gtest:          $GTEST_ROOT
+clang-4-64-gtest-lib:   $GTEST_ROOT/lib-clang4/libgtest.a
+
+########################################
+# clang-5
+clang-5-root:           $CLANG5_ROOT/bin
+clang-5-cxx:            clang++
+clang-5-cc:             clang
+clang-5-cxxflags:       -stdlib=libc++ -Wdocumentation -std=c++1z
+clang-5-exe-ldflags:    -L$CLANG5_ROOT/lib -Wl,-rpath=$CLANG5_ROOT/lib
+clang-5-boost:          $LIB_ROOT/boost_1_65_1
+clang-5-64-boost-lib:   stage64-clang5-std17/lib
+clang-5-64-asan-boost-lib:   stage64-clang5-asan/lib
+clang-5-64-tsan-boost-lib:   stage64-clang5-tsan/lib
+clang-5-gtest:          $GTEST_ROOT
+clang-5-64-gtest-lib:   $GTEST_ROOT/lib-clang5/libgtest.a
+
+########################################
+# clang-6
+clang-6-root:           $CLANG6_ROOT/bin
+clang-6-cxx:            clang++
+clang-6-cc:             clang
+clang-6-cxxflags:       -stdlib=libc++ -Wdocumentation -std=c++17
+clang-6-exe-ldflags:    -L$CLANG6_ROOT/lib -Wl,-rpath=$CLANG6_ROOT/lib
+clang-6-boost:          $LIB_ROOT/boost_1_65_1
+clang-6-64-boost-lib:   stage64-clang6-std17/lib
+clang-6-64-asan-boost-lib:   stage64-clang6-asan/lib
+clang-6-64-tsan-boost-lib:   stage64-clang6-tsan/lib
+clang-6-gtest:          $GTEST_ROOT
+clang-6-64-gtest-lib:   $GTEST_ROOT/lib-clang6/libgtest.a
+
+
+
+
+
+
+
+
+
+
+

extern/libcds/build/CI/VASEx-CI/ci-build

@@ -0,0 +1,81 @@
+#! /bin/bash
+
+# Useful envvars:
+# CI_SCRIPT_PATH - path where to find scripts
+# TOOLSET - toolset: x64-gcc-5, x64-clang-3.9 and so on
+# BUILD_TYPE - build type: 'dbg', 'rel', 'asan', 'tsan'
+# WORKSPACE - path where to build
+
+env|sort
+
+case "$TOOLSET" in
+    "x64-gcc-4.8")
+	echo "GCC-4.8 '$BUILD_TYPE', toolset root: $GCC48_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-4.8-64 $*
+	exit $?
+	;;
+    "x64-gcc-4.9")
+	echo "GCC-4.9 '$BUILD_TYPE', toolset root: $GCC49_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-4.9-64 $*
+	exit $?
+	;;
+    "x64-gcc-5")
+	echo "GCC-5 '$BUILD_TYPE', toolset root: $GCC5_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-5-64 $*
+	exit $?
+	;;
+    "x64-gcc-6")
+	echo "GCC-6 '$BUILD_TYPE', toolset root: $GCC6_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-6-64 $*
+	exit $?
+	;;
+    "x64-gcc-7")
+	echo "GCC-7 '$BUILD_TYPE', toolset root: $GCC7_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-7-64 $*
+	exit $?
+	;;
+    "x64-gcc-8")
+	echo "GCC-8 '$BUILD_TYPE', toolset root: $GCC8_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-gcc-8-64 $*
+	exit $?
+	;;
+    "x64-clang-3.6")
+	echo "clang-3.6 '$BUILD_TYPE', toolset root: $CLANG36_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-3.6-64 $*
+	exit $?
+	;;
+    "x64-clang-3.7")
+	echo "clang-3.7 '$BUILD_TYPE', toolset root: $CLANG37_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-3.7-64 $*
+	exit $?
+	;;
+    "x64-clang-3.8")
+	echo "clang-3.8 '$BUILD_TYPE', toolset root: $CLANG38_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-3.8-64 $*
+	exit $?
+	;;
+    "x64-clang-3.9")
+	echo "clang-3.9 '$BUILD_TYPE', toolset root: $CLANG39_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-3.9-64 $*
+	exit $?
+	;;
+    "x64-clang-4")
+	echo "clang-4 '$BUILD_TYPE', toolset root: $CLANG4_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-4-64 $*
+	exit $?
+	;;
+    "x64-clang-5")
+	echo "clang-5 '$BUILD_TYPE', toolset root: $CLANG5_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-5-64 $*
+	exit $?
+	;;
+    "x64-clang-6")
+	echo "clang-6 '$BUILD_TYPE', toolset root: $CLANG6_ROOT"
+	$CI_SCRIPT_PATH/ci-script/cds-$BUILD_TYPE-clang-6-64 $*
+	exit $?
+	;;
+    * )
+	echo "Undefined toolset '$TOOLSET'"
+	exit 1
+	;;
+esac

extern/libcds/build/CI/VASEx-CI/ci-env

@@ -0,0 +1,35 @@
+CMAKE_2_8_12=/home/libcds-ci/bin/cmake/cmake-2.8.12/bin
+CMAKE_3_6_2=/home/libcds-ci/bin/cmake/cmake-3.6.2/bin
+CMAKE3=$CMAKE_3_6_2
+
+PATH=$CMAKE3:$PATH:$HOME/.local/bin:$HOME/bin
+
+TOOLSET_ROOT=$HOME/bin
+
+GCC48_ROOT=$TOOLSET_ROOT/gcc-4.8
+GCC49_ROOT=$TOOLSET_ROOT/gcc-4.9
+GCC5_ROOT=$TOOLSET_ROOT/gcc-5
+GCC6_ROOT=$TOOLSET_ROOT/gcc-6
+GCC7_ROOT=$TOOLSET_ROOT/gcc-7
+GCC8_ROOT=$TOOLSET_ROOT/gcc-8
+
+CLANG35_ROOT=$TOOLSET_ROOT/clang-3.5
+CLANG36_ROOT=$TOOLSET_ROOT/clang-3.6
+CLANG37_ROOT=$TOOLSET_ROOT/clang-3.7
+CLANG38_ROOT=$TOOLSET_ROOT/clang-3.8
+CLANG39_ROOT=$TOOLSET_ROOT/clang-3.9
+CLANG4_ROOT=$TOOLSET_ROOT/clang-4
+CLANG5_ROOT=$TOOLSET_ROOT/clang-5
+CLANG6_ROOT=$TOOLSET_ROOT/clang-6
+
+CLANG_STDLIB="-stdlib=libc++"
+CLANG37_CXXFLAGS=$CLANG_STDLIB
+CLANG38_CXXFLAGS=$CLANG_STDLIB
+CLANG39_CXXFLAGS=$CLANG_STDLIB
+CLANG4_CXXFLAGS=$CLANG_STDLIB
+CLANG5_CXXFLAGS=$CLANG_STDLIB
+CLANG6_CXXFLAGS=$CLANG_STDLIB
+
+LIB_ROOT=$HOME/lib
+BOOST_ROOT=$LIB_ROOT/boost
+GTEST_ROOT=$LIB_ROOT/gtest

extern/libcds/build/CI/cmake-gen

@@ -0,0 +1,113 @@
+#! /usr/bin/perl
+
+my $compiler=shift;
+my $bitness =shift;
+my $build   =shift;
+$build="rel" unless $build;
+
+my $cmake_build="RELEASE";
+$cmake_build="DEBUG" if $build eq 'dbg';
+
+my $cds_libs="cds-libs";
+
+# get generic props
+my $workspace=get_gen_prop("workspace") || "$HOME";
+my $cds_source=get_gen_prop("libcds-source") || "../libcds";
+my $make_jobs=get_gen_prop("make-job") || 2;
+
+# get compiler-specific props
+my $comp_root=get_prop("root");
+my $boost=get_prop( "boost" );
+my $boost_libs=get_prop( "boost-lib" );
+my $gtest=get_prop("gtest");
+my $gtest_lib=get_prop( "gtest-lib");
+my $gtest_inc=get_prop("gtest-include") || get_gen_prop("gtest-include");
+my $cxx=get_prop("cxx") or $compiler;
+my $cc=get_prop("cc") or $compiler;
+my $cxxflags=get_prop("cxxflags");
+my $ldflags=get_prop("ldflags");
+my $cmake_exe_ldflags=get_prop("exe-ldflags");
+my $ext_lib=get_prop("extlib");
+my $ld_lib_path=get_prop("ld-lib-path");
+my $sys_path=get_prop("path");
+my $cmake_flags=get_prop("cmake-flags");
+
+my $filename="cds-$build-$compiler-$bitness";
+open( my $out, ">", $filename )  or die "Cannot open cds-$build-$compiler-$bitness";
+
+print $out "#! /bin/sh\n\n";
+print $out "root=$workspace\n";
+print $out "CDS_SOURCE=\$root/$cds_source\n";
+print $out "OBJ_ROOT=\$root/obj\n";
+print $out "BIN_ROOT=\$root/bin\n";
+print $out "GTEST_ROOT=$gtest\n" if $gtest;
+print $out "export PATH=$sys_path:\$PATH\n" if $sys_path;
+print $out "\n";
+print $out "rm -fr \$OBJ_ROOT\n";
+print $out "rm -fr \$BIN_ROOT\n";
+print $out "mkdir -p \$OBJ_ROOT\n";
+print $out "#cp -f run-ctest-rel \$OBJ_ROOT/run-ctest\n" if $build eq 'rel';
+print $out "#cp -f run-ctest-dbg \$OBJ_ROOT/run-ctest\n" unless $build eq 'rel';
+print $out "cd \$OBJ_ROOT\n";
+print $out "\n";
+print $out "LD_LIBRARY_PATH=$ld_lib_path:\$LD_LIBRARY_PATH \\\n" if $ld_lib_path;
+print $out "LDFLAGS=\"$ldflags\" \\\n" if $ldflags;
+print $out "cmake -G \"Unix Makefiles\" \\\n";
+print $out " -DCMAKE_BUILD_TYPE=$cmake_build \\\n";
+print $out " -DCMAKE_C_COMPILER=$comp_root/$cc \\\n";
+print $out " -DCMAKE_CXX_COMPILER=$comp_root/$cxx \\\n";
+print $out " -DCMAKE_CXX_FLAGS=\"$cxxflags\" \\\n" if $cxxflags;
+print $out " -DCMAKE_EXE_LINKER_FLAGS=\"$cmake_exe_ldflags\" \\\n" if $cmake_exe_ldflags;
+print $out " -DCDS_BIN_DIR=\$BIN_ROOT \\\n";
+print $out " -DWITH_TESTS=ON \\\n";
+print $out " -DWITH_ASAN=ON \\\n" if $build eq 'asan';
+print $out " -DWITH_TSAN=ON \\\n" if $build eq 'tsan';
+print $out " -DBOOST_ROOT=$boost \\\n";
+print $out " -DBOOST_LIBRARYDIR=$boost/$boost_libs \\\n" if $boost_libs;
+print $out " -DGTEST_INCLUDE_DIRS=$gtest_inc \\\n" if $gtest_inc;
+print $out " -DGTEST_LIBRARIES=$gtest_lib \\\n" if $gtest_lib;
+print $out " -DEXTERNAL_SYSTEM_LIBS=\"$ext_lib\" \\\n" if $ext_lib;
+print $out " $cmake_flags \\\n" if $cmake_flags;
+print $out " \$CDS_SOURCE && \\\n";
+print $out "make -j $make_jobs \$* \n";
+
+close $out;
+chmod 0755, $filename;
+
+sub get_prop($@)
+{
+   my $what=shift;
+   my $key="$compiler-$bitness-$build-$what:";
+
+   my $grep = `grep -P $key $cds_libs`;
+   if ( $grep ) {
+      my @ret = $grep =~ /^$key\s+(\S.*\S*)\s+/;
+      return $ret[0] if @ret;
+   }
+
+   $key = "$compiler-$bitness-$what:";
+   my $grep = `grep -P $key $cds_libs`;
+   if ( $grep ) {
+      my @ret = $grep =~ /^$key\s+(\S.*\S*)\s+/;
+      return $ret[0] if @ret;
+   }
+
+   $key = "$compiler-$what:";
+   my $grep = `grep -P $key $cds_libs`;
+   if ( $grep ) {
+      my @ret = $grep =~ /^$key\s+(\S.*\S*)\s+/;
+      return $ret[0] if @ret;
+   }
+}
+
+sub get_gen_prop($@)
+{
+   my $key=shift;
+   $key = "$key:";
+
+   my $grep = `grep -P $key $cds_libs`;
+   if ( $grep ) {
+      my @ret = $grep =~ /^$key\s+(\S.*\S*)\s+/;
+      return $ret[0] if @ret;
+   }
+}

extern/libcds/build/CI/gen-all

@@ -0,0 +1,48 @@
+#! /bin/sh
+
+./cmake-gen gcc-4.8 64 dbg
+./cmake-gen gcc-4.8 64 rel
+./cmake-gen gcc-4.9 64 dbg
+./cmake-gen gcc-4.9 64 rel
+./cmake-gen gcc-5 64 dbg
+./cmake-gen gcc-5 64 rel
+./cmake-gen gcc-5 64 tsan
+./cmake-gen gcc-5 64 asan
+./cmake-gen gcc-6 64 dbg
+./cmake-gen gcc-6 64 rel
+./cmake-gen gcc-6 64 tsan
+./cmake-gen gcc-6 64 asan
+./cmake-gen gcc-7 64 dbg
+./cmake-gen gcc-7 64 rel
+./cmake-gen gcc-7 64 tsan
+./cmake-gen gcc-7 64 asan
+./cmake-gen gcc-8 64 dbg
+./cmake-gen gcc-8 64 rel
+./cmake-gen gcc-8 64 tsan
+./cmake-gen gcc-8 64 asan
+./cmake-gen clang-3.6 64 dbg
+./cmake-gen clang-3.6 64 rel
+./cmake-gen clang-3.7 64 dbg
+./cmake-gen clang-3.7 64 rel
+./cmake-gen clang-3.8 64 dbg
+./cmake-gen clang-3.8 64 rel
+./cmake-gen clang-3.9 64 dbg
+./cmake-gen clang-3.9 64 rel
+./cmake-gen clang-3.9 64 asan
+./cmake-gen clang-3.9 64 tsan
+./cmake-gen clang-4 64 dbg
+./cmake-gen clang-4 64 rel
+./cmake-gen clang-4 64 asan
+./cmake-gen clang-4 64 tsan
+./cmake-gen clang-5 64 dbg
+./cmake-gen clang-5 64 rel
+./cmake-gen clang-5 64 asan
+./cmake-gen clang-5 64 tsan
+./cmake-gen clang-6 64 dbg
+./cmake-gen clang-6 64 rel
+./cmake-gen clang-6 64 asan
+./cmake-gen clang-6 64 tsan
+./cmake-gen clang-7 64 dbg
+./cmake-gen clang-7 64 rel
+./cmake-gen clang-7 64 asan
+./cmake-gen clang-7 64 tsan

extern/libcds/build/CI/travis-ci/install.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+
+set -e
+set -x
+
+if [[ "$(uname -s)" == 'Darwin' ]]; then
+    brew update || brew update
+    brew outdated pyenv || brew upgrade pyenv
+    brew install pyenv-virtualenv
+    brew install cmake || true
+
+    if which pyenv > /dev/null; then
+        eval "$(pyenv init -)"
+    fi
+
+    pyenv install 2.7.10
+    pyenv virtualenv 2.7.10 conan
+    pyenv rehash
+    pyenv activate conan
+    
+    pip install conan --upgrade
+    pip install conan_package_tools
+
+    conan user
+    exit 0
+fi
+
+pip install --user conan --upgrade
+pip install --user conan_package_tools
+
+conan user

extern/libcds/build/CI/travis-ci/run.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+
+set -e
+set -x
+
+CONAN_INSTALL_FLAGS="-s compiler.libcxx=libstdc++11"
+
+if [[ "$(uname -s)" == 'Darwin' ]]; then
+    if which pyenv > /dev/null; then
+        eval "$(pyenv init -)"
+    fi
+    pyenv activate conan
+    CONAN_INSTALL_FLAGS=""
+fi
+
+#export CXX=$CXX_COMPILER
+#export CC=$C_COMPILER
+mkdir build-test && cd build-test
+conan install --build $CONAN_INSTALL_FLAGS -s build_type=$BUILD_TYPE ..
+cmake -DCMAKE_PREFIX_PATH="$TRAVIS_BUILD_DIR/build-test/deps" -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_EXE_LINKER_FLAGS=$LINKER_FLAGS -DWITH_TESTS=ON ..
+cmake --build . -- -j2 $TARGET
+
+if [[ "$(uname -s)" == 'Darwin' ]]; then
+	export DYLD_LIBRARY_PATH=$TRAVIS_BUILD_DIR/build-test/deps/lib
+fi
+ctest -VV -R $TARGET

extern/libcds/build/cmake/TargetArch.cmake

@@ -0,0 +1,141 @@
+# Source: https://github.com/axr/solar-cmake
+# Based on the Qt 5 processor detection code, so should be very accurate
+# https://qt.gitorious.org/qt/qtbase/blobs/master/src/corelib/global/qprocessordetection.h
+# Currently handles arm (v5, v6, v7), x86 (32/64), ia64, and ppc (32/64)
+
+# Regarding POWER/PowerPC, just as is noted in the Qt source,
+# "There are many more known variants/revisions that we do not handle/detect."
+
+set(archdetect_c_code "
+#if defined(__arm__) || defined(__TARGET_ARCH_ARM)
+    #if defined(__ARM_ARCH_7__) \\
+        || defined(__ARM_ARCH_7A__) \\
+        || defined(__ARM_ARCH_7R__) \\
+        || defined(__ARM_ARCH_7M__) \\
+        || (defined(__TARGET_ARCH_ARM) && __TARGET_ARCH_ARM-0 >= 7)
+        #error cmake_ARCH armv7
+    #elif defined(__ARM_ARCH_6__) \\
+        || defined(__ARM_ARCH_6J__) \\
+        || defined(__ARM_ARCH_6T2__) \\
+        || defined(__ARM_ARCH_6Z__) \\
+        || defined(__ARM_ARCH_6K__) \\
+        || defined(__ARM_ARCH_6ZK__) \\
+        || defined(__ARM_ARCH_6M__) \\
+        || (defined(__TARGET_ARCH_ARM) && __TARGET_ARCH_ARM-0 >= 6)
+        #error cmake_ARCH armv6
+    #elif defined(__ARM_ARCH_5TEJ__) \\
+        || (defined(__TARGET_ARCH_ARM) && __TARGET_ARCH_ARM-0 >= 5)
+        #error cmake_ARCH armv5
+    #else
+        #error cmake_ARCH arm
+    #endif
+#elif defined(__aarch64__)
+    #if defined(__ARM_ARCH) && __ARM_ARCH == 8
+        #error cmake_ARCH armv8
+    #else
+        #error cmake_ARCH arm64
+    #endif
+#elif defined(__i386) || defined(__i386__) || defined(_M_IX86)
+    #error cmake_ARCH i386
+#elif defined(__x86_64) || defined(__x86_64__) || defined(__amd64) || defined(_M_X64)
+    #error cmake_ARCH x86_64
+#elif defined(__ia64) || defined(__ia64__) || defined(_M_IA64)
+    #error cmake_ARCH ia64
+#elif defined(__ppc__) || defined(__ppc) || defined(__powerpc__) \\
+      || defined(_ARCH_COM) || defined(_ARCH_PWR) || defined(_ARCH_PPC)  \\
+      || defined(_M_MPPC) || defined(_M_PPC)
+    #if defined(__ppc64__) || defined(__powerpc64__) || defined(__64BIT__)
+        #error cmake_ARCH ppc64
+    #else
+        #error cmake_ARCH ppc
+    #endif
+#endif
+
+#error cmake_ARCH unknown
+")
+
+# Set ppc_support to TRUE before including this file or ppc and ppc64
+# will be treated as invalid architectures since they are no longer supported by Apple
+
+function(target_architecture output_var)
+    if(APPLE AND CMAKE_OSX_ARCHITECTURES)
+        # On OS X we use CMAKE_OSX_ARCHITECTURES *if* it was set
+        # First let's normalize the order of the values
+
+        # Note that it's not possible to compile PowerPC applications if you are using
+        # the OS X SDK version 10.6 or later - you'll need 10.4/10.5 for that, so we
+        # disable it by default
+        # See this page for more information:
+        # http://stackoverflow.com/questions/5333490/how-can-we-restore-ppc-ppc64-as-well-as-full-10-4-10-5-sdk-support-to-xcode-4
+
+        # Architecture defaults to i386 or ppc on OS X 10.5 and earlier, depending on the CPU type detected at runtime.
+        # On OS X 10.6+ the default is x86_64 if the CPU supports it, i386 otherwise.
+
+        foreach(osx_arch ${CMAKE_OSX_ARCHITECTURES})
+            if("${osx_arch}" STREQUAL "ppc" AND ppc_support)
+                set(osx_arch_ppc TRUE)
+            elseif("${osx_arch}" STREQUAL "i386")
+                set(osx_arch_i386 TRUE)
+            elseif("${osx_arch}" STREQUAL "x86_64")
+                set(osx_arch_x86_64 TRUE)
+            elseif("${osx_arch}" STREQUAL "ppc64" AND ppc_support)
+                set(osx_arch_ppc64 TRUE)
+            else()
+                message(FATAL_ERROR "Invalid OS X arch name: ${osx_arch}")
+            endif()
+        endforeach()
+
+        # Now add all the architectures in our normalized order
+        if(osx_arch_ppc)
+            list(APPEND ARCH ppc)
+        endif()
+
+        if(osx_arch_i386)
+            list(APPEND ARCH i386)
+        endif()
+
+        if(osx_arch_x86_64)
+            list(APPEND ARCH x86_64)
+        endif()
+
+        if(osx_arch_ppc64)
+            list(APPEND ARCH ppc64)
+        endif()
+    else()
+        file(WRITE "${CMAKE_BINARY_DIR}/arch.c" "${archdetect_c_code}")
+
+        enable_language(C)
+
+        # Detect the architecture in a rather creative way...
+        # This compiles a small C program which is a series of ifdefs that selects a
+        # particular #error preprocessor directive whose message string contains the
+        # target architecture. The program will always fail to compile (both because
+        # file is not a valid C program, and obviously because of the presence of the
+        # #error preprocessor directives... but by exploiting the preprocessor in this
+        # way, we can detect the correct target architecture even when cross-compiling,
+        # since the program itself never needs to be run (only the compiler/preprocessor)
+        try_run(
+            run_result_unused
+            compile_result_unused
+            "${CMAKE_BINARY_DIR}"
+            "${CMAKE_BINARY_DIR}/arch.c"
+            COMPILE_OUTPUT_VARIABLE ARCH
+            CMAKE_FLAGS CMAKE_OSX_ARCHITECTURES=${CMAKE_OSX_ARCHITECTURES}
+        )
+
+        # Parse the architecture name from the compiler output
+        string(REGEX MATCH "cmake_ARCH ([a-zA-Z0-9_]+)" ARCH "${ARCH}")
+
+        # Get rid of the value marker leaving just the architecture name
+        string(REPLACE "cmake_ARCH " "" ARCH "${ARCH}")
+
+        # If we are compiling with an unknown architecture this variable should
+        # already be set to "unknown" but in the case that it's empty (i.e. due
+        # to a typo in the code), then set it to unknown
+        if (NOT ARCH)
+            set(ARCH unknown)
+        endif()
+    endif()
+
+    set(${output_var} "${ARCH}" PARENT_SCOPE)
+endfunction()

extern/libcds/build/cmake/description.txt

@@ -0,0 +1 @@
+libcds - Concurrent Data Structure C++ library

extern/libcds/build/cmake/post_install_script.sh

@@ -0,0 +1 @@
+ldconfig

extern/libcds/build/cmake/post_uninstall_script.sh

@@ -0,0 +1 @@
+ldconfig

extern/libcds/build/cmake/readme.md

@@ -0,0 +1,104 @@
+Building library with CMake
+===============
+
+CDS suports both in-source and out-of-source cmake build types. Now project uses:
+
+- CMake: general cross-platform building
+- CTest: all unit tests can be run in a standard way by *ctest* command
+- CPack: for making rpm/deb/nsys etc. packages
+
+Compiling and testing
+----------
+**Building out-of-source in "RELEASE" mode ("DEBUG" is default)**
+
+- Wherever create empty directory for building, for instance `libcds-debug`
+- Prepare: `cmake -DCMAKE_BUILD_TYPE=RELEASE <path to the project's root directory with CMakeLists.txt>`
+- Compile: `make -j4`
+- As a result you'll see shared and static cds libraries in the build directory
+
+**Warning**: We strongly recommend not to use static cds library. Static library is not tested and not maintained. You can use it on own risk.
+
+After using command cmake -L <path to the project's root directory with CMakeLists.txt> one can see some additional variables, that can activate additional features:
+
+- `WITH_TESTS:BOOL=OFF`: if you want to build library with unit testing support use *-DWITH_TESTS=ON* on prepare step. Be careful with this flag, because compile time will dramatically increase
+- `WITH_TESTS_COVERAGE:BOOL=OFF`: Analyze test coverage using gcov (only for gcc)
+- `WITH_BOOST_ATOMIC:BOOL=OFF`: Use boost atomics (only for boost >= 1.54)
+- `WITH_ASAN:BOOL=OFF`: compile libcds with AddressSanitizer instrumentation
+- `WITH_TSAN:BOOL=OFF`: compile libcds with ThreadSanitizer instrumentation
+
+Additional gtest hints (for unit and stress tests only):
+- `GTEST_INCLUDE_DIRS=path`: gives full `path` to gtest include dir. 
+- `GTEST_LIBRARY=path`: gives full `path` to `libgtest.a`.
+
+
+Packaging
+----------
+
+In order to package library *CPack* is used, command *cpack -G <Generator>* should create correspondent packages for particular operating system. Now the project supports building the following package types:
+
+- *RPM*: redhat-based linux distribs        
+- *DEB*: debian-based linux distribs
+- *TGZ*: simple "*tgz*" archive with library and headers
+- *NSYS*: windows installer package (NSYS should be installed)   
+  
+"Live" building and packaging example
+----------
+- `git clone https://github.com/khizmax/libcds.git`
+- `mkdir libcds-release`
+- `cd libcds-release`
+- `cmake -DWITH\_TESTS=ON -DCMAKE\_BUILD_TYPE=RELEASE ../libcds`
+```
+    -- The C compiler identification is GNU 4.8.3
+    -- The CXX compiler identification is GNU 4.8.3
+    ...
+    -- Found Threads: TRUE
+    -- Boost version: 1.54.0
+    -- Found the following Boost libraries:
+    --   system
+    --   thread
+    Build type -- RELEASE
+    -- Configuring done
+    -- Generating done
+    -- Build files have been written to: <...>/libcds-release
+``` 
+- `make -j4`
+```
+    Scanning dependencies of target cds
+    Scanning dependencies of target test-common
+    Scanning dependencies of target cds-s
+    Scanning dependencies of target test-hdr-offsetof
+    [  1%] Building CXX object CMakeFiles/cds-s.dir/src/hp_gc.cpp.o
+    ...
+    [100%] Built target test-hdr
+```
+
+- `ctest`
+```
+    Test project /home/kel/projects_cds/libcds-debug
+        Start 1: test-hdr
+    1/7 Test #1: test-hdr .........................   Passed  1352.24 sec
+        Start 2: cdsu-misc
+    2/7 Test #2: cdsu-misc ........................   Passed    0.00 sec
+        Start 3: cdsu-map
+    ...
+```
+
+- `cpack -G RPM`
+```
+    CPack: Create package using RPM
+    CPack: Install projects
+    CPack: - Run preinstall target for: cds
+    CPack: - Install project: cds
+    CPack: -   Install component: devel
+    CPack: -   Install component: lib
+    CPack: Create package
+    CPackRPM:Debug: Adding /usr/local to builtin omit list.
+    CPackRPM: Will use GENERATED spec file: /home/kel/projects_cds/libcds-debug/_CPack_Packages/Linux/RPM/SPECS/cds-devel.spec
+    CPackRPM: Will use GENERATED spec file: /home/kel/projects_cds/libcds-debug/_CPack_Packages/Linux/RPM/SPECS/cds-lib.spec
+    CPack: - package: /home/kel/projects_cds/libcds-debug/cds-2.1.0-1-devel.rpm generated.
+    CPack: - package: /home/kel/projects_cds/libcds-debug/cds-2.1.0-1-lib.rpm generated.
+```
+
+Future development
+----------
+- CDash: use CI system

extern/libcds/cds/algo/atomic.h

@@ -0,0 +1,496 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CXX11_ATOMIC_H
+#define CDSLIB_CXX11_ATOMIC_H
+
+#include <cds/details/defs.h>
+#include <cds/user_setup/cache_line.h>
+
+namespace cds {
+
+/// C++11 Atomic library support
+/** @anchor cds_cxx11_atomic
+    \p libcds can use the following implementations of the atomics:
+    - STL \p &lt;atomic&gt;. This is used by default
+    - \p boost.atomic for boost 1.54 and above. To use it you should define \p CDS_USE_BOOST_ATOMIC for
+      your compiler invocation, for example, for gcc specify \p -DCDS_USE_BOOST_ATOMIC
+      in command line
+    - \p libcds implementation of atomic operation according to C++11 standard as
+      specified in <a href="http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2011/n3242.pdf">N3242, p.29</a>.
+      \p libcds implementation is not the full standard compliant, it provides only C++ part of standard,
+      for example, \p libcds has no static initialization of the atomic variables and some other C features.
+      However, that imlementation is enough for the library purposes. Supported architecture: x86, amd64,
+      ia64 (Itanium) 64bit, 64bit Sparc. To use \p libcds atomic you should define \p CDS_USE_LIBCDS_ATOMIC
+      in the compiler command line (\p -DCDS_USE_LIBCDS_ATOMIC for gcc/clang).
+
+      @note For Clang compiler \p libcds doesn't use native \p libc++ \p &lt;atomic&gt; due some problems.
+      Instead, \p libcds atomic is used by default, or you can try to use \p boost.atomic.
+
+      The library defines \p atomics alias for atomic namespace:
+      - <tt>namespace atomics = std</tt> for STL
+      - <tt>namespace atomics = boost</tt> for \p boost.atomic
+      - <tt>namespace atomics = cds::cxx11_atomic</tt> for library-provided atomic implementation
+*/
+namespace cxx11_atomic {
+}} // namespace cds::cxx11_atomic
+
+//@cond
+#if defined(CDS_USE_BOOST_ATOMIC)
+    // boost atomic
+#   include <boost/version.hpp>
+#   if BOOST_VERSION >= 105400
+#       include <boost/atomic.hpp>
+        namespace atomics = boost;
+#       define CDS_CXX11_ATOMIC_BEGIN_NAMESPACE namespace boost {
+#       define CDS_CXX11_ATOMIC_END_NAMESPACE }
+#   else
+#       error "Boost version 1.54 or above is needed for boost.atomic"
+#   endif
+#elif defined(CDS_USE_LIBCDS_ATOMIC)
+    // libcds atomic
+#   include <cds/compiler/cxx11_atomic.h>
+    namespace atomics = cds::cxx11_atomic;
+#   define CDS_CXX11_ATOMIC_BEGIN_NAMESPACE namespace cds { namespace cxx11_atomic {
+#   define CDS_CXX11_ATOMIC_END_NAMESPACE }}
+#else
+    // Compiler provided C++11 atomic
+#   include <atomic>
+    namespace atomics = std;
+#   define CDS_CXX11_ATOMIC_BEGIN_NAMESPACE namespace std {
+#   define CDS_CXX11_ATOMIC_END_NAMESPACE }
+#endif
+//@endcond
+
+namespace cds {
+
+    /// Atomic primitives
+    /**
+        This namespace contains useful primitives derived from <tt>std::atomic</tt>.
+    */
+    namespace atomicity {
+
+        /// Atomic event counter.
+        /**
+            This class is based on <tt>std::atomic_size_t</tt>.
+            It uses relaxed memory ordering \p memory_order_relaxed and may be used as a statistic counter.
+        */
+        class event_counter
+        {
+            //@cond
+            atomics::atomic_size_t   m_counter;
+            //@endcond
+
+        public:
+            typedef size_t      value_type  ;       ///< Type of counter
+
+        public:
+            // Initializes event counter with zero
+            event_counter() noexcept
+                : m_counter(size_t(0))
+            {}
+
+            /// Assign operator
+            /**
+                Returns \p n.
+            */
+            value_type operator =(
+                value_type n    ///< new value of the counter
+            ) noexcept
+            {
+                m_counter.exchange( n, atomics::memory_order_relaxed );
+                return n;
+            }
+
+            /// Addition
+            /**
+                Returns new value of the atomic counter.
+            */
+            size_t operator +=(
+                size_t n    ///< addendum
+            ) noexcept
+            {
+                return m_counter.fetch_add( n, atomics::memory_order_relaxed ) + n;
+            }
+
+            /// Substraction
+            /**
+                Returns new value of the atomic counter.
+            */
+            size_t operator -=(
+                size_t n    ///< subtrahend
+            ) noexcept
+            {
+                return m_counter.fetch_sub( n, atomics::memory_order_relaxed ) - n;
+            }
+
+            /// Get current value of the counter
+            operator size_t () const noexcept
+            {
+                return m_counter.load( atomics::memory_order_relaxed );
+            }
+
+            /// Preincrement
+            size_t operator ++() noexcept
+            {
+                return m_counter.fetch_add( 1, atomics::memory_order_relaxed ) + 1;
+            }
+            /// Postincrement
+            size_t operator ++(int) noexcept
+            {
+                return m_counter.fetch_add( 1, atomics::memory_order_relaxed );
+            }
+
+            /// Predecrement
+            size_t operator --() noexcept
+            {
+                return m_counter.fetch_sub( 1, atomics::memory_order_relaxed ) - 1;
+            }
+            /// Postdecrement
+            size_t operator --(int) noexcept
+            {
+                return m_counter.fetch_sub( 1, atomics::memory_order_relaxed );
+            }
+
+            /// Get current value of the counter
+            size_t get() const noexcept
+            {
+                return m_counter.load( atomics::memory_order_relaxed );
+            }
+
+            /// Resets the counter to 0
+            void reset() noexcept
+            {
+                m_counter.store( 0, atomics::memory_order_release );
+            }
+        };
+
+        /// Atomic item counter
+        /**
+            This class is simplified interface around \p std::atomic_size_t.
+            The class supports getting current value of the counter and increment/decrement its value.
+
+            See also: improved version that eliminates false sharing - \p cache_friendly_item_counter.
+        */
+        class item_counter
+        {
+        public:
+            typedef atomics::atomic_size_t   atomic_type;   ///< atomic type used
+            typedef size_t counter_type;                    ///< Integral item counter type (size_t)
+
+        private:
+            //@cond
+            atomic_type     m_Counter;   ///< Atomic item counter
+            //@endcond
+
+        public:
+            /// Default ctor initializes the counter to zero.
+            item_counter()
+                : m_Counter(counter_type(0))
+            {}
+
+            /// Returns current value of the counter
+            counter_type value(atomics::memory_order order = atomics::memory_order_relaxed) const
+            {
+                return m_Counter.load( order );
+            }
+
+            /// Same as \ref value() with relaxed memory ordering
+            operator counter_type() const
+            {
+                return value();
+            }
+
+            /// Returns underlying atomic interface
+            atomic_type& getAtomic()
+            {
+                return m_Counter;
+            }
+
+            /// Returns underlying atomic interface (const)
+            const atomic_type& getAtomic() const
+            {
+                return m_Counter;
+            }
+
+            /// Increments the counter. Semantics: postincrement
+            counter_type inc(atomics::memory_order order = atomics::memory_order_relaxed )
+            {
+                return m_Counter.fetch_add( 1, order );
+            }
+
+            /// Increments the counter. Semantics: postincrement
+            counter_type inc( counter_type count, atomics::memory_order order = atomics::memory_order_relaxed )
+            {
+                return m_Counter.fetch_add( count, order );
+            }
+
+            /// Decrements the counter. Semantics: postdecrement
+            counter_type dec(atomics::memory_order order = atomics::memory_order_relaxed)
+            {
+                return m_Counter.fetch_sub( 1, order );
+            }
+
+            /// Decrements the counter. Semantics: postdecrement
+            counter_type dec( counter_type count, atomics::memory_order order = atomics::memory_order_relaxed )
+            {
+                return m_Counter.fetch_sub( count, order );
+            }
+
+            /// Preincrement
+            counter_type operator ++()
+            {
+                return inc() + 1;
+            }
+            /// Postincrement
+            counter_type operator ++(int)
+            {
+                return inc();
+            }
+
+            /// Predecrement
+            counter_type operator --()
+            {
+                return dec() - 1;
+            }
+            /// Postdecrement
+            counter_type operator --(int)
+            {
+                return dec();
+            }
+
+            /// Increment by \p count
+            counter_type operator +=( counter_type count )
+            {
+                return inc( count ) + count;
+            }
+
+            /// Decrement by \p count
+            counter_type operator -=( counter_type count )
+            {
+                return dec( count ) - count;
+            }
+
+            /// Resets count to 0
+            void reset(atomics::memory_order order = atomics::memory_order_relaxed)
+            {
+                m_Counter.store( 0, order );
+            }
+        };
+
+#if CDS_COMPILER == CDS_COMPILER_CLANG
+    // CLang unhappy: pad1_ and pad2_ - unused private field warning
+#   pragma GCC diagnostic push
+#   pragma GCC diagnostic ignored "-Wunused-private-field"
+#endif
+        /// Atomic cache-friendly item counter
+        /**
+            Atomic item counter with cache-line padding to avoid false sharing.
+            Adding cache-line padding before and after atomic counter eliminates the contention
+            in read path of many containers and can notably improve search operations in sets/maps.
+        */
+        class cache_friendly_item_counter
+        {
+        public:
+            typedef atomics::atomic_size_t   atomic_type;   ///< atomic type used
+            typedef size_t counter_type;                    ///< Integral item counter type (size_t)
+
+        private:
+            //@cond
+            char            pad1_[cds::c_nCacheLineSize];
+            atomic_type     m_Counter;   ///< Atomic item counter
+            char            pad2_[cds::c_nCacheLineSize - sizeof( atomic_type )];
+            //@endcond
+
+        public:
+            /// Default ctor initializes the counter to zero.
+            cache_friendly_item_counter()
+                : m_Counter(counter_type(0))
+            {}
+
+            /// Returns current value of the counter
+            counter_type value(atomics::memory_order order = atomics::memory_order_relaxed) const
+            {
+                return m_Counter.load( order );
+            }
+
+            /// Same as \ref value() with relaxed memory ordering
+            operator counter_type() const
+            {
+                return value();
+            }
+
+            /// Returns underlying atomic interface
+            atomic_type& getAtomic()
+            {
+                return m_Counter;
+            }
+
+            /// Returns underlying atomic interface (const)
+            const atomic_type& getAtomic() const
+            {
+                return m_Counter;
+            }
+
+            /// Increments the counter. Semantics: postincrement
+            counter_type inc(atomics::memory_order order = atomics::memory_order_relaxed )
+            {
+                return m_Counter.fetch_add( 1, order );
+            }
+
+            /// Increments the counter. Semantics: postincrement
+            counter_type inc( counter_type count, atomics::memory_order order = atomics::memory_order_relaxed )
+            {
+                return m_Counter.fetch_add( count, order );
+            }
+
+            /// Decrements the counter. Semantics: postdecrement
+            counter_type dec(atomics::memory_order order = atomics::memory_order_relaxed)
+            {
+                return m_Counter.fetch_sub( 1, order );
+            }
+
+            /// Decrements the counter. Semantics: postdecrement
+            counter_type dec( counter_type count, atomics::memory_order order = atomics::memory_order_relaxed )
+            {
+                return m_Counter.fetch_sub( count, order );
+            }
+
+            /// Preincrement
+            counter_type operator ++()
+            {
+                return inc() + 1;
+            }
+            /// Postincrement
+            counter_type operator ++(int)
+            {
+                return inc();
+            }
+
+            /// Predecrement
+            counter_type operator --()
+            {
+                return dec() - 1;
+            }
+            /// Postdecrement
+            counter_type operator --(int)
+            {
+                return dec();
+            }
+
+            /// Increment by \p count
+            counter_type operator +=( counter_type count )
+            {
+                return inc( count ) + count;
+            }
+
+            /// Decrement by \p count
+            counter_type operator -=( counter_type count )
+            {
+                return dec( count ) - count;
+            }
+
+            /// Resets count to 0
+            void reset(atomics::memory_order order = atomics::memory_order_relaxed)
+            {
+                m_Counter.store( 0, order );
+            }
+        };
+#if CDS_COMPILER == CDS_COMPILER_CLANG
+#   pragma GCC diagnostic pop
+#endif
+
+        /// Empty item counter
+        /**
+            This class may be used instead of \ref item_counter when you do not need full \ref item_counter interface.
+            All methods of the class is empty and returns 0.
+
+            The object of this class should not be used in data structure that behavior significantly depends on item counting
+            (for example, in many hash map implementation).
+        */
+        class empty_item_counter {
+        public:
+            typedef size_t counter_type    ;  ///< Counter type
+        public:
+            /// Returns 0
+            static counter_type value(atomics::memory_order /*order*/ = atomics::memory_order_relaxed)
+            {
+                return 0;
+            }
+
+            /// Same as \ref value(), always returns 0.
+            operator counter_type() const
+            {
+                return value();
+            }
+
+            /// Dummy increment. Always returns 0
+            static counter_type inc(atomics::memory_order /*order*/ = atomics::memory_order_relaxed)
+            {
+                return 0;
+            }
+
+            /// Dummy increment. Always returns 0
+            static counter_type inc( counter_type /*count*/, atomics::memory_order /*order*/ = atomics::memory_order_relaxed )
+            {
+                return 0;
+            }
+
+            /// Dummy increment. Always returns 0
+            static counter_type dec(atomics::memory_order /*order*/ = atomics::memory_order_relaxed)
+            {
+                return 0;
+            }
+
+            /// Dummy increment. Always returns 0
+            static counter_type dec( counter_type /*count*/, atomics::memory_order /*order*/ = atomics::memory_order_relaxed )
+            {
+                return 0;
+            }
+
+            /// Dummy pre-increment. Always returns 0
+            counter_type operator ++() const
+            {
+                return 0;
+            }
+            /// Dummy post-increment. Always returns 0
+            counter_type operator ++(int) const
+            {
+                return 0;
+            }
+
+            /// Dummy pre-decrement. Always returns 0
+            counter_type operator --() const
+            {
+                return 0;
+            }
+            /// Dummy post-decrement. Always returns 0
+            counter_type operator --(int) const
+            {
+                return 0;
+            }
+
+            /// Dummy increment by \p count, always returns 0
+            counter_type operator +=( counter_type count )
+            {
+                CDS_UNUSED( count );
+                return 0;
+            }
+
+            /// Dummy decrement by \p count, always returns 0
+            counter_type operator -=( counter_type count )
+            {
+                CDS_UNUSED( count );
+                return 0;
+            }
+
+            /// Dummy function
+            static void reset(atomics::memory_order /*order*/ = atomics::memory_order_relaxed)
+            {}
+        };
+    }   // namespace atomicity
+}   // namespace cds
+
+#endif // #ifndef CDSLIB_CXX11_ATOMIC_H

extern/libcds/cds/algo/backoff_strategy.h

@@ -0,0 +1,439 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_BACKOFF_STRATEGY_H
+#define CDSLIB_BACKOFF_STRATEGY_H
+
+/*
+    Filename: backoff_strategy.h
+    Created 2007.03.01 by Maxim Khiszinsky
+
+    Description:
+         Generic back-off strategies
+
+    Editions:
+    2007.03.01  Maxim Khiszinsky    Created
+    2008.10.02  Maxim Khiszinsky    Backoff action transfers from contructor to operator() for all backoff schemas
+    2009.09.10  Maxim Khiszinsky    reset() function added
+*/
+
+#include <utility>      // declval
+#include <thread>
+#include <chrono>
+#include <cds/compiler/backoff.h>
+
+namespace cds {
+    /// Different backoff schemes
+    /**
+        Back-off schema may be used in lock-free algorithms when the algorithm cannot perform some action because a conflict
+        with the other concurrent operation is encountered. In this case current thread can do another work or can call
+        processor's performance hint.
+
+        The interface of back-off strategy is following:
+        \code
+            struct backoff_strategy {
+                void operator()();
+                template <typename Predicate> bool operator()( Predicate pr );
+                void reset();
+            };
+        \endcode
+
+        \p operator() operator calls back-off strategy's action. It is main part of back-off strategy.
+
+        Interruptible back-off <tt>template < typename Predicate > bool operator()( Predicate pr )</tt>
+        allows to interrupt back-off spinning if \p pr predicate returns \p true.
+        \p Predicate is a functor with the following interface:
+        \code
+        struct predicate {
+            bool operator()();
+        };
+        \endcode
+
+        \p reset() function resets internal state of back-off strategy to initial state. It is required for some
+        back-off strategies, for example, exponential back-off.
+    */
+    namespace backoff {
+
+        /// Empty backoff strategy. Do nothing
+        struct empty {
+            //@cond
+            void operator ()() const noexcept
+            {}
+
+            template <typename Predicate>
+            bool operator()(Predicate pr) const noexcept( noexcept(std::declval<Predicate>()()))
+            {
+                return pr();
+            }
+
+            static void reset() noexcept
+            {}
+            //@endcond
+        };
+
+        /// Switch to another thread (yield). Good for thread preemption architecture.
+        struct yield {
+            //@cond
+            void operator ()() const noexcept
+            {
+                std::this_thread::yield();
+            }
+
+            template <typename Predicate>
+            bool operator()(Predicate pr) const noexcept( noexcept(std::declval<Predicate>()()))
+            {
+                if ( pr())
+                    return true;
+                operator()();
+                return false;
+            }
+
+            static void reset() noexcept
+            {}
+            //@endcond
+        };
+
+        /// Random pause
+        /**
+            This back-off strategy calls processor-specific pause hint instruction
+            if one is available for the processor architecture.
+        */
+        struct pause {
+            //@cond
+            void operator ()() const noexcept
+            {
+#            ifdef CDS_backoff_hint_defined
+                platform::backoff_hint();
+#            endif
+            }
+
+            template <typename Predicate>
+            bool operator()(Predicate pr) const noexcept( noexcept(std::declval<Predicate>()()))
+            {
+                if ( pr())
+                    return true;
+                operator()();
+                return false;
+            }
+
+            static void reset() noexcept
+            {}
+            //@endcond
+        };
+
+        /// Processor hint back-off
+        /**
+            This back-off schema calls performance hint instruction if it is available for current processor.
+            Otherwise, it calls \p nop.
+        */
+        struct hint
+        {
+        //@cond
+            void operator ()() const noexcept
+            {
+#           if defined(CDS_backoff_hint_defined)
+                platform::backoff_hint();
+#           elif defined(CDS_backoff_nop_defined)
+                platform::backoff_nop();
+#           endif
+            }
+
+            template <typename Predicate>
+            bool operator()(Predicate pr) const noexcept(noexcept(std::declval<Predicate>()()))
+            {
+                if ( pr())
+                    return true;
+                operator()();
+                return false;
+            }
+
+            static void reset() noexcept
+            {}
+        //@endcond
+        };
+
+        /// \p backoff::exponential const traits
+        struct exponential_const_traits
+        {
+            typedef hint    fast_path_backoff;  ///< Fast-path back-off strategy
+            typedef yield   slow_path_backoff;  ///< Slow-path back-off strategy
+
+            enum: size_t {
+                lower_bound = 16,         ///< Minimum spinning limit
+                upper_bound = 16 * 1024   ///< Maximum spinning limit
+            };
+        };
+
+        /// \p nackoff::exponential runtime traits
+        struct exponential_runtime_traits
+        {
+            typedef hint    fast_path_backoff;  ///< Fast-path back-off strategy
+            typedef yield   slow_path_backoff;  ///< Slow-path back-off strategy
+
+            static size_t lower_bound;    ///< Minimum spinning limit, default is 16
+            static size_t upper_bound;    ///< Maximum spinning limit, default is 16*1024
+        };
+
+        /// Exponential back-off
+        /**
+            This back-off strategy is composite. It consists of \p SpinBkoff and \p YieldBkoff
+            back-off strategy. In first, the strategy tries to apply repeatedly \p SpinBkoff
+            (spinning phase) until internal counter of failed attempts reaches its maximum
+            spinning value. Then, the strategy transits to high-contention phase
+            where it applies \p YieldBkoff until \p reset() is called.
+            On each spinning iteration the internal spinning counter is doubled.
+
+            Selecting the best value for maximum spinning limit is platform and application specific task.
+            The limits are described by \p Traits template parameter.
+            There are two types of \p Traits:
+            - constant traits \p exponential_const_traits - specifies the lower and upper limits
+              as a compile-time constants; to change the limits you should recompile your application
+            - runtime traits \p exponential_runtime_traits - specifies the limits as \p s_nExpMin
+              and \p s_nExpMax variables which can be changed at runtime to tune back-off strategy.
+
+            The traits class must declare two data member:
+            - \p lower_bound - the lower bound of spinning loop
+            - \p upper_bound - the upper boudn of spinning loop
+
+            You may use \p Traits template parameter to separate back-off implementations.
+            For example, you may define two \p exponential back-offs that is the best for your task A and B:
+            \code
+
+            #include <cds/algo/backoff_strategy.h>
+            namespace bkoff = cds::backoff;
+
+            // the best bounds for task A
+            struct traits_A: public bkoff::exponential_const_traits
+            {
+                static size_t lower_bound;
+                static size_t upper_bound;
+            };
+            size_t traits_A::lower_bound = 1024;
+            size_t traits_A::upper_bound = 8 * 1024;
+
+            // the best bounds for task B
+            struct traits_B: public bkoff::exponential_const_traits
+            {
+                static size_t lower_bound;
+                static size_t upper_bound;
+            };
+            size_t traits_A::lower_bound = 16;
+            size_t traits_A::upper_bound = 1024;
+
+            // // define your back-off specialization
+            typedef bkoff::exponential<traits_A> expBackOffA;
+            typedef bkoff::exponential<traits_B> expBackOffB;
+            \endcode
+        */
+        template <typename Traits = exponential_const_traits >
+        class exponential
+        {
+        public:
+            typedef Traits     traits;   ///< Traits
+
+            typedef typename traits::fast_path_backoff  spin_backoff    ;   ///< spin (fast-path) back-off strategy
+            typedef typename traits::slow_path_backoff  yield_backoff   ;   ///< yield (slow-path) back-off strategy
+
+        protected:
+            size_t  m_nExpCur   ;           ///< Current spin counter in range [traits::s_nExpMin, traits::s_nExpMax]
+
+            spin_backoff    m_bkSpin    ;   ///< Spinning (fast-path) phase back-off strategy
+            yield_backoff   m_bkYield   ;   ///< Yield phase back-off strategy
+
+        public:
+            /// Default ctor
+            exponential() noexcept
+                : m_nExpCur( traits::lower_bound )
+            {}
+
+            //@cond
+            void operator ()() noexcept(noexcept(std::declval<spin_backoff>()()) && noexcept(std::declval<yield_backoff>()()))
+            {
+                if ( m_nExpCur <= traits::upper_bound ) {
+                    for ( size_t n = 0; n < m_nExpCur; ++n )
+                        m_bkSpin();
+                    m_nExpCur *= 2;
+                }
+                else
+                    m_bkYield();
+            }
+
+            template <typename Predicate>
+            bool operator()( Predicate pr ) noexcept( noexcept(std::declval<Predicate>()()) && noexcept(std::declval<spin_backoff>()()) && noexcept(std::declval<yield_backoff>()()))
+            {
+                if ( m_nExpCur <= traits::upper_bound ) {
+                    for ( size_t n = 0; n < m_nExpCur; ++n ) {
+                        if ( m_bkSpin(pr))
+                            return true;
+                    }
+                    m_nExpCur *= 2;
+                }
+                else
+                    return m_bkYield(pr);
+                return false;
+            }
+
+            void reset() noexcept( noexcept( std::declval<spin_backoff>().reset()) && noexcept( std::declval<yield_backoff>().reset()))
+            {
+                m_nExpCur = traits::lower_bound;
+                m_bkSpin.reset();
+                m_bkYield.reset();
+            }
+            //@endcond
+        };
+
+        //@cond
+        template <typename FastPathBkOff, typename SlowPathBkOff>
+        struct make_exponential
+        {
+            struct traits: public exponential_const_traits
+            {
+                typedef FastPathBkOff   fast_path_backoff;
+                typedef SlowPathBkOff   slow_path_backoff;
+            };
+
+            typedef exponential<traits> type;
+        };
+
+        template <typename FastPathBkOff, typename SlowPathBkOff>
+        using make_exponential_t = typename make_exponential<FastPathBkOff, SlowPathBkOff>::type;
+        //@endcond
+
+        /// Constant traits for \ref delay back-off strategy
+        struct delay_const_traits
+        {
+            typedef std::chrono::milliseconds duration_type;    ///< Timeout type
+            enum: unsigned {
+                timeout = 5                                     ///< Delay timeout
+            };
+        };
+
+        /// Runtime traits for \ref delay back-off strategy
+        struct delay_runtime_traits
+        {
+            typedef std::chrono::milliseconds duration_type;    ///< Timeout type
+            static unsigned timeout;                            ///< Delay timeout, default 5
+        };
+
+        /// Delay back-off strategy
+        /**
+            Template arguments:
+            - \p Duration - duration type, default is \p std::chrono::milliseconds
+            - \p Traits - a class that defines default timeout.
+
+            Choosing the best value for th timeout is platform and application specific task.
+            The default values for timeout is provided by \p Traits class that should
+            \p timeout data member. There are two predefined \p Traits implementation:
+            - \p delay_const_traits - defines \p timeout as a constant (enum).
+              To change timeout you should recompile your application.
+            - \p delay_runtime_traits - specifies timeout as static data member that can be changed
+              at runtime to tune the back-off strategy.
+
+            You may use \p Traits template parameter to separate back-off implementations.
+            For example, you may define two \p delay back-offs for 5 and 10 ms timeout:
+            \code
+
+            #include <cds/algo/backoff_strategy.h>
+            namespace bkoff = cds::backoff;
+
+            // 5ms delay
+            struct ms5
+            {
+                typedef std::chrono::milliseconds duration_type;
+                enum: unsigned { timeout = 5 };
+            };
+
+            // 10ms delay, runtime support
+            struct ms10
+            {
+                typedef std::chrono::milliseconds duration_type;
+                static unsigned timeout;
+            };
+            unsigned ms10::timeout = 10;
+
+            // define your back-off specialization
+            typedef bkoff::delay<std::chrono::milliseconds, ms5>  delay5;
+            typedef bkoff::delay<std::chrono::milliseconds, ms10> delay10;
+
+            \endcode
+        */
+        template <typename Traits = delay_const_traits>
+        class delay
+        {
+        public:
+            typedef Traits   traits;        ///< Traits
+            typedef typename Traits::duration_type duration_type; ///< Duration type (default \p std::chrono::milliseconds)
+
+        protected:
+            ///@cond
+            duration_type const timeout;
+            ///@endcond
+
+        public:
+            /// Default ctor takes the timeout from \p traits::timeout
+            delay() noexcept
+                : timeout( traits::timeout )
+            {}
+
+            /// Initializes timeout from \p nTimeout
+            constexpr explicit delay( unsigned int nTimeout ) noexcept
+                : timeout( nTimeout )
+            {}
+
+            //@cond
+            void operator()() const
+            {
+                std::this_thread::sleep_for( timeout );
+            }
+
+            template <typename Predicate>
+            bool operator()(Predicate pr) const
+            {
+                for ( unsigned int i = 0; i < traits::timeout; i += 2 ) {
+                    if ( pr())
+                        return true;
+                    std::this_thread::sleep_for( duration_type( 2 ));
+                }
+                return false;
+            }
+
+            static void reset() noexcept
+            {}
+            //@endcond
+        };
+
+        //@cond
+        template <unsigned int Timeout, class Duration = std::chrono::milliseconds >
+        struct make_delay_of
+        {
+            struct traits {
+                typedef Duration duration_type;
+                enum: unsigned { timeout = Timeout };
+            };
+
+            typedef delay<traits> type;
+        };
+        //@endcond
+
+        /// Delay back-off strategy, template version
+        /**
+            This is a simplified version of \p backoff::delay class.
+            Template parameter \p Timeout sets a delay timeout of \p Duration unit.
+        */
+        template <unsigned int Timeout, class Duration = std::chrono::milliseconds >
+        using delay_of = typename make_delay_of< Timeout, Duration >::type;
+
+
+        /// Default backoff strategy
+        typedef exponential<exponential_const_traits>    Default;
+
+        /// Default back-off strategy for lock primitives
+        typedef exponential<exponential_const_traits>    LockDefault;
+
+    } // namespace backoff
+} // namespace cds
+
+
+#endif // #ifndef CDSLIB_BACKOFF_STRATEGY_H

extern/libcds/cds/algo/base.h

@@ -0,0 +1,18 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_BASE_H
+#define CDSLIB_ALGO_BASE_H
+
+#include <cds/details/defs.h>
+
+namespace cds {
+
+    /// Different approaches and techniques for supporting high-concurrent data structure
+    namespace algo {}
+
+} // namespace cds
+
+#endif // #ifndef CDSLIB_ALGO_BASE_H

extern/libcds/cds/algo/bit_reversal.h

@@ -0,0 +1,159 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_BIT_REVERSAL_H
+#define CDSLIB_ALGO_BIT_REVERSAL_H
+
+#include <cds/algo/base.h>
+
+    // Source: http://stackoverflow.com/questions/746171/best-algorithm-for-bit-reversal-from-msb-lsb-to-lsb-msb-in-c
+namespace cds { namespace algo {
+
+    /// Bit reversal algorithms
+    namespace bit_reversal {
+
+        /// SWAR algorithm (source: http://aggregate.org/MAGIC/#Bit%20Reversal)
+        struct swar {
+            /// 32bit
+            uint32_t operator()( uint32_t x ) const
+            {
+                x = ( ( ( x & 0xaaaaaaaa ) >> 1 ) | ( ( x & 0x55555555 ) << 1 ));
+                x = ( ( ( x & 0xcccccccc ) >> 2 ) | ( ( x & 0x33333333 ) << 2 ));
+                x = ( ( ( x & 0xf0f0f0f0 ) >> 4 ) | ( ( x & 0x0f0f0f0f ) << 4 ));
+                x = ( ( ( x & 0xff00ff00 ) >> 8 ) | ( ( x & 0x00ff00ff ) << 8 ));
+                return( ( x >> 16 ) | ( x << 16 ));
+            }
+
+            /// 64bit
+            uint64_t operator()( uint64_t x ) const
+            {
+                return ( static_cast<uint64_t>( operator()( static_cast<uint32_t>( x ))) << 32 )  // low 32bit
+                    | ( static_cast<uint64_t>( operator()( static_cast<uint32_t>( x >> 32 ))));  // high 32bit
+            }
+        };
+
+        /// Lookup table algorithm
+        struct lookup {
+            /// 32bit
+            uint32_t operator()( uint32_t x ) const
+            {
+                static uint8_t const table[] = {
+                    0x00, 0x80, 0x40, 0xC0, 0x20, 0xA0, 0x60, 0xE0, 0x10, 0x90, 0x50, 0xD0, 0x30, 0xB0, 0x70, 0xF0,
+                    0x08, 0x88, 0x48, 0xC8, 0x28, 0xA8, 0x68, 0xE8, 0x18, 0x98, 0x58, 0xD8, 0x38, 0xB8, 0x78, 0xF8,
+                    0x04, 0x84, 0x44, 0xC4, 0x24, 0xA4, 0x64, 0xE4, 0x14, 0x94, 0x54, 0xD4, 0x34, 0xB4, 0x74, 0xF4,
+                    0x0C, 0x8C, 0x4C, 0xCC, 0x2C, 0xAC, 0x6C, 0xEC, 0x1C, 0x9C, 0x5C, 0xDC, 0x3C, 0xBC, 0x7C, 0xFC,
+                    0x02, 0x82, 0x42, 0xC2, 0x22, 0xA2, 0x62, 0xE2, 0x12, 0x92, 0x52, 0xD2, 0x32, 0xB2, 0x72, 0xF2,
+                    0x0A, 0x8A, 0x4A, 0xCA, 0x2A, 0xAA, 0x6A, 0xEA, 0x1A, 0x9A, 0x5A, 0xDA, 0x3A, 0xBA, 0x7A, 0xFA,
+                    0x06, 0x86, 0x46, 0xC6, 0x26, 0xA6, 0x66, 0xE6, 0x16, 0x96, 0x56, 0xD6, 0x36, 0xB6, 0x76, 0xF6,
+                    0x0E, 0x8E, 0x4E, 0xCE, 0x2E, 0xAE, 0x6E, 0xEE, 0x1E, 0x9E, 0x5E, 0xDE, 0x3E, 0xBE, 0x7E, 0xFE,
+                    0x01, 0x81, 0x41, 0xC1, 0x21, 0xA1, 0x61, 0xE1, 0x11, 0x91, 0x51, 0xD1, 0x31, 0xB1, 0x71, 0xF1,
+                    0x09, 0x89, 0x49, 0xC9, 0x29, 0xA9, 0x69, 0xE9, 0x19, 0x99, 0x59, 0xD9, 0x39, 0xB9, 0x79, 0xF9,
+                    0x05, 0x85, 0x45, 0xC5, 0x25, 0xA5, 0x65, 0xE5, 0x15, 0x95, 0x55, 0xD5, 0x35, 0xB5, 0x75, 0xF5,
+                    0x0D, 0x8D, 0x4D, 0xCD, 0x2D, 0xAD, 0x6D, 0xED, 0x1D, 0x9D, 0x5D, 0xDD, 0x3D, 0xBD, 0x7D, 0xFD,
+                    0x03, 0x83, 0x43, 0xC3, 0x23, 0xA3, 0x63, 0xE3, 0x13, 0x93, 0x53, 0xD3, 0x33, 0xB3, 0x73, 0xF3,
+                    0x0B, 0x8B, 0x4B, 0xCB, 0x2B, 0xAB, 0x6B, 0xEB, 0x1B, 0x9B, 0x5B, 0xDB, 0x3B, 0xBB, 0x7B, 0xFB,
+                    0x07, 0x87, 0x47, 0xC7, 0x27, 0xA7, 0x67, 0xE7, 0x17, 0x97, 0x57, 0xD7, 0x37, 0xB7, 0x77, 0xF7,
+                    0x0F, 0x8F, 0x4F, 0xCF, 0x2F, 0xAF, 0x6F, 0xEF, 0x1F, 0x9F, 0x5F, 0xDF, 0x3F, 0xBF, 0x7F, 0xFF
+                };
+                static_assert( sizeof( table ) / sizeof( table[0] ) == 256, "Table size mismatch" );
+
+                return ( static_cast<uint32_t>( table[x & 0xff] ) << 24 ) |
+                    ( static_cast<uint32_t>( table[( x >> 8 ) & 0xff] ) << 16 ) |
+                    ( static_cast<uint32_t>( table[( x >> 16 ) & 0xff] ) << 8 ) |
+                    ( static_cast<uint32_t>( table[( x >> 24 ) & 0xff] ));
+            }
+
+            /// 64bit
+            uint64_t operator()( uint64_t x ) const
+            {
+                return ( static_cast<uint64_t>( operator()( static_cast<uint32_t>( x ))) << 32 ) |
+                    static_cast<uint64_t>( operator()( static_cast<uint32_t>( x >> 32 )));
+            }
+        };
+
+
+        /// Mul-Div algorithm for 32bit architectire
+
+        /// Mul-Div algorithm
+        struct muldiv {
+            //@cond
+            static uint8_t muldiv32_byte( uint8_t b )
+            {
+                return static_cast<uint8_t>( ( ( b * 0x0802LU & 0x22110LU ) | ( b * 0x8020LU & 0x88440LU )) * 0x10101LU >> 16 );
+            }
+
+            static uint8_t muldiv64_byte( uint8_t b )
+            {
+                return static_cast<uint8_t>( ( b * 0x0202020202ULL & 0x010884422010ULL ) % 1023 );
+            }
+
+            // for 32bit architecture
+            static uint32_t muldiv32( uint32_t x )
+            {
+                return static_cast<uint32_t>( muldiv32_byte( static_cast<uint8_t>( x >> 24 )))
+                    | ( static_cast<uint32_t>( muldiv32_byte( static_cast<uint8_t>( x >> 16 ))) << 8 )
+                    | ( static_cast<uint32_t>( muldiv32_byte( static_cast<uint8_t>( x >> 8 ))) << 16 )
+                    | ( static_cast<uint32_t>( muldiv32_byte( static_cast<uint8_t>( x ))) << 24 );
+            }
+
+            static uint64_t muldiv32( uint64_t x )
+            {
+                return static_cast<uint64_t>( muldiv32_byte( static_cast<uint8_t>( x >> 56 )))
+                    | ( static_cast<uint64_t>( muldiv32_byte( static_cast<uint8_t>( x >> 48 ))) << 8 )
+                    | ( static_cast<uint64_t>( muldiv32_byte( static_cast<uint8_t>( x >> 40 ))) << 16 )
+                    | ( static_cast<uint64_t>( muldiv32_byte( static_cast<uint8_t>( x >> 32 ))) << 24 )
+                    | ( static_cast<uint64_t>( muldiv32_byte( static_cast<uint8_t>( x >> 24 ))) << 32 )
+                    | ( static_cast<uint64_t>( muldiv32_byte( static_cast<uint8_t>( x >> 16 ))) << 40 )
+                    | ( static_cast<uint64_t>( muldiv32_byte( static_cast<uint8_t>( x >> 8 ))) << 48 )
+                    | ( static_cast<uint64_t>( muldiv32_byte( static_cast<uint8_t>( x ))) << 56 );
+            }
+
+            /// for 64bit architectire
+            static uint32_t muldiv64( uint32_t x )
+            {
+                return static_cast<uint32_t>( muldiv64_byte( static_cast<uint8_t>( x >> 24 )))
+                    | ( static_cast<uint32_t>( muldiv64_byte( static_cast<uint8_t>( x >> 16 ))) << 8 )
+                    | ( static_cast<uint32_t>( muldiv64_byte( static_cast<uint8_t>( x >> 8 ))) << 16 )
+                    | ( static_cast<uint32_t>( muldiv64_byte( static_cast<uint8_t>( x ))) << 24 );
+            }
+
+            static uint64_t muldiv64( uint64_t x )
+            {
+                return static_cast<uint64_t>( muldiv64_byte( static_cast<uint8_t>( x >> 56 )))
+                    | ( static_cast<uint64_t>( muldiv64_byte( static_cast<uint8_t>( x >> 48 ))) << 8 )
+                    | ( static_cast<uint64_t>( muldiv64_byte( static_cast<uint8_t>( x >> 40 ))) << 16 )
+                    | ( static_cast<uint64_t>( muldiv64_byte( static_cast<uint8_t>( x >> 32 ))) << 24 )
+                    | ( static_cast<uint64_t>( muldiv64_byte( static_cast<uint8_t>( x >> 24 ))) << 32 )
+                    | ( static_cast<uint64_t>( muldiv64_byte( static_cast<uint8_t>( x >> 16 ))) << 40 )
+                    | ( static_cast<uint64_t>( muldiv64_byte( static_cast<uint8_t>( x >> 8 ))) << 48 )
+                    | ( static_cast<uint64_t>( muldiv64_byte( static_cast<uint8_t>( x ))) << 56 );
+            }
+            //@endcond
+
+            /// 32bit
+            uint32_t operator()( uint32_t x ) const
+            {
+#           if CDS_BUILD_BITS == 32
+                return muldiv32( x );
+#           else
+                return muldiv64( x );
+#           endif
+            }
+
+            /// 64bit
+            uint64_t operator()( uint64_t x ) const
+            {
+#           if CDS_BUILD_BITS == 32
+                return muldiv32( x );
+#           else
+                return muldiv64( x );
+#           endif
+            }
+        };
+
+    } // namespace bit_reversal
+}} // namespace cds::algo
+
+#endif // #ifndef CDSLIB_ALGO_BIT_REVERSAL_H

extern/libcds/cds/algo/bitop.h

@@ -0,0 +1,143 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_BITOP_H
+#define CDSLIB_BITOP_H
+
+/*
+    Different bit algorithms:
+        LSB        get least significant bit number
+        MSB        get most significant bit number
+        bswap    swap byte order of word
+        RBO        reverse bit order of word
+
+    Editions:
+        2007.10.08    Maxim.Khiszinsky    Created
+*/
+
+#include <cds/details/defs.h>
+#include <cds/compiler/bitop.h>
+
+namespace cds {
+    /// Bit operations
+    namespace bitop {
+
+        ///@cond none
+        namespace details {
+            template <int> struct BitOps;
+
+            // 32-bit bit ops
+            template <> struct BitOps<4> {
+                typedef uint32_t        TUInt;
+
+                static int      MSB( TUInt x )      { return bitop::platform::msb32( x );   }
+                static int      LSB( TUInt x )      { return bitop::platform::lsb32( x );   }
+                static int      MSBnz( TUInt x )    { return bitop::platform::msb32nz( x ); }
+                static int      LSBnz( TUInt x )    { return bitop::platform::lsb32nz( x ); }
+                static int      SBC( TUInt x )      { return bitop::platform::sbc32( x ) ;  }
+                static int      ZBC( TUInt x )      { return bitop::platform::zbc32( x ) ;  }
+
+                static TUInt    RBO( TUInt x )      { return bitop::platform::rbo32( x );   }
+                static bool     complement( TUInt& x, int nBit ) { return bitop::platform::complement32( &x, nBit ); }
+
+                static TUInt    RandXorShift(TUInt x) { return bitop::platform::RandXorShift32(x); }
+            };
+
+            // 64-bit bit ops
+            template <> struct BitOps<8> {
+                typedef uint64_t TUInt;
+
+                static int      MSB( TUInt x )        { return bitop::platform::msb64( x );     }
+                static int      LSB( TUInt x )        { return bitop::platform::lsb64( x );     }
+                static int      MSBnz( TUInt x )      { return bitop::platform::msb64nz( x );   }
+                static int      LSBnz( TUInt x )      { return bitop::platform::lsb64nz( x );   }
+                static  int     SBC( TUInt x )        { return bitop::platform::sbc64( x ) ;    }
+                static  int     ZBC( TUInt x )        { return bitop::platform::zbc64( x ) ;    }
+
+                static TUInt    RBO( TUInt x )        { return bitop::platform::rbo64( x );     }
+                static bool     complement( TUInt& x, int nBit ) { return bitop::platform::complement64( &x, nBit ); }
+
+                static TUInt    RandXorShift(TUInt x) { return bitop::platform::RandXorShift64(x); }
+            };
+        }    // namespace details
+        //@endcond
+
+
+        /// Get least significant bit (LSB) number (1..32/64), 0 if nArg == 0
+        template <typename T>
+        static inline int LSB( T nArg )
+        {
+            return details::BitOps< sizeof(T) >::LSB( (typename details::BitOps<sizeof(T)>::TUInt) nArg );
+        }
+
+        /// Get least significant bit (LSB) number (0..31/63)
+        /**
+            Precondition: nArg != 0
+        */
+        template <typename T>
+        static inline int LSBnz( T nArg )
+        {
+            assert( nArg != 0 );
+            return details::BitOps< sizeof(T) >::LSBnz( (typename details::BitOps<sizeof(T)>::TUInt) nArg );
+        }
+
+        /// Get most significant bit (MSB) number (1..32/64), 0 if nArg == 0
+        template <typename T>
+        static inline int MSB( T nArg )
+        {
+            return details::BitOps< sizeof(T) >::MSB( (typename details::BitOps<sizeof(T)>::TUInt) nArg );
+        }
+
+        /// Get most significant bit (MSB) number (0..31/63)
+        /**
+            Precondition: nArg != 0
+        */
+        template <typename T>
+        static inline int MSBnz( T nArg )
+        {
+            assert( nArg != 0 );
+            return details::BitOps< sizeof(T) >::MSBnz( (typename details::BitOps<sizeof(T)>::TUInt) nArg );
+        }
+
+        /// Get non-zero bit count of a word
+        template <typename T>
+        static inline int SBC( T nArg )
+        {
+            return details::BitOps< sizeof(T) >::SBC( (typename details::BitOps<sizeof(T)>::TUInt) nArg );
+        }
+
+        /// Get zero bit count of a word
+        template <typename T>
+        static inline int ZBC( T nArg )
+        {
+            return details::BitOps< sizeof(T) >::ZBC( (typename details::BitOps<sizeof(T)>::TUInt) nArg );
+        }
+
+        /// Reverse bit order of \p nArg
+        template <typename T>
+        static inline T RBO( T nArg )
+        {
+            return (T) details::BitOps< sizeof(T) >::RBO( (typename details::BitOps<sizeof(T)>::TUInt) nArg );
+        }
+
+        /// Complement bit \p nBit in \p nArg
+        template <typename T>
+        static inline bool complement( T& nArg, int nBit )
+        {
+            return details::BitOps< sizeof(T) >::complement( reinterpret_cast< typename details::BitOps<sizeof(T)>::TUInt& >( nArg ), nBit );
+        }
+
+        /// Simple random number generator
+        template <typename T>
+        static inline T RandXorShift( T x)
+        {
+            return (T) details::BitOps< sizeof(T) >::RandXorShift(x);
+        }
+
+    } // namespace bitop
+} //namespace cds
+
+#endif    // #ifndef CDSLIB_BITOP_H
+

extern/libcds/cds/algo/elimination.h

@@ -0,0 +1,61 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_ELIMINATION_H
+#define CDSLIB_ALGO_ELIMINATION_H
+
+#include <cds/algo/elimination_tls.h>
+#include <cds/algo/elimination_opt.h>
+#include <cds/algo/atomic.h>
+#include <cds/threading/model.h>
+
+namespace cds { namespace algo {
+
+    /// Elimination technique
+    /** @anchor cds_elimination_description
+        Elimination technique allows highly distributed coupling and execution of operations with reverse
+        semantics like the pushes and pops on a stack. If a push followed by a pop are performed
+        on a stack, the data structure's state does not change (similarly for a pop followed by a push).
+        This means that if one can cause pairs of pushes and pops to meet and pair up in
+        separate locations, the threads can exchange values without having to touch a centralized structure
+        since they have anyhow "eliminated" each other's effect on it. Elimination can be implemented
+        by using a collision array in which threads pick random locations in order to try and collide.
+        Pairs of threads that "collide" in some location run through a synchronization protocol,
+        and all such disjoint collisions can be performed in parallel. If a thread has not met another
+        in the selected location or if it met a thread with an operation that cannot be eliminated
+        (such as two push operations), an alternative scheme must be used.
+    */
+    namespace elimination {
+
+        /// Base class describing an operation for eliminating
+        /**
+            This class contains some debugng info.
+            Actual operation descriptor depends on real container and its interface.
+        */
+        struct operation_desc
+        {
+            record * pOwner;    ///< Owner of the descriptor
+        };
+
+        /// Acquires elimination record for the current thread
+        template <typename OperationDesc>
+        static inline record * init_record( OperationDesc& op )
+        {
+            record& rec = cds::threading::elimination_record();
+            assert( rec.is_free());
+            op.pOwner = &rec;
+            rec.pOp = static_cast<operation_desc *>( &op );
+            return &rec;
+        }
+
+        /// Releases elimination record for the current thread
+        static inline void clear_record()
+        {
+            cds::threading::elimination_record().pOp = nullptr;
+        }
+    } // namespace elimination
+}} // namespace cds::algo
+
+#endif // CDSLIB_ALGO_ELIMINATION_H

extern/libcds/cds/algo/elimination_opt.h

@@ -0,0 +1,40 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_ELIMINATION_OPT_H
+#define CDSLIB_ALGO_ELIMINATION_OPT_H
+
+#include <cds/details/defs.h>
+
+namespace cds { namespace opt {
+
+    /// Enable \ref cds_elimination_description "elimination back-off" for the container
+    template <bool Enable>
+    struct enable_elimination {
+        //@cond
+        template <class Base> struct pack: public Base
+        {
+            static constexpr const bool enable_elimination = Enable;
+        };
+        //@endcond
+    };
+
+    /// \ref cds_elimination_description "Elimination back-off strategy" option setter
+    /**
+        Back-off strategy for elimination.
+        Usually, elimination back-off strategy is \p cds::backoff::delay.
+    */
+    template <typename Type>
+    struct elimination_backoff {
+        //@cond
+        template <class Base> struct pack: public Base
+        {
+            typedef Type elimination_backoff;
+        };
+        //@endcond
+    };
+}} // namespace cds::opt
+
+#endif // #ifndef CDSLIB_ALGO_ELIMINATION_OPT_H

extern/libcds/cds/algo/elimination_tls.h

@@ -0,0 +1,37 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_ELIMINATION_TLS_H
+#define CDSLIB_ALGO_ELIMINATION_TLS_H
+
+#include <cds/algo/base.h>
+
+namespace cds { namespace algo { namespace elimination {
+
+    // Forwards
+    struct operation_desc;
+
+    /// Per-thread elimination record
+    /** @headerfile cds/algo/elimination.h
+    */
+    struct record
+    {
+        operation_desc *    pOp ;   ///< Operation descriptor
+
+        /// Initialization
+        record()
+            : pOp( nullptr )
+        {}
+
+        /// Checks if the record is free
+        bool is_free() const
+        {
+            return pOp == nullptr;
+        }
+    };
+
+}}} // cds::algo::elimination
+
+#endif // #ifndef CDSLIB_ALGO_ELIMINATION_TLS_H

extern/libcds/cds/algo/flat_combining.h

@@ -0,0 +1,11 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_FLAT_COMBINING_H
+#define CDSLIB_ALGO_FLAT_COMBINING_H
+
+#include <cds/algo/flat_combining/kernel.h>
+
+#endif // #ifndef CDSLIB_ALGO_FLAT_COMBINING_H

extern/libcds/cds/algo/flat_combining/defs.h

@@ -0,0 +1,67 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_FLAT_COMBINING_DEFS_H
+#define CDSLIB_ALGO_FLAT_COMBINING_DEFS_H
+
+#include <cds/algo/atomic.h>
+
+
+namespace cds { namespace algo { namespace flat_combining {
+
+    /// Special values of \p publication_record::nRequest
+    enum request_value
+    {
+        req_EmptyRecord,    ///< Publication record is empty
+        req_Response,       ///< Operation is done
+
+        req_Operation       ///< First operation id for derived classes
+    };
+
+    /// \p publication_record state
+    enum record_state {
+        inactive,       ///< Record is inactive
+        active,         ///< Record is active
+        removed         ///< Record should be removed
+    };
+
+    /// Record of publication list
+    /**
+        Each data structure based on flat combining contains a class derived from \p %publication_record
+    */
+    struct publication_record {
+        atomics::atomic<unsigned int>           nRequest;   ///< Request field (depends on data structure)
+        atomics::atomic<unsigned int>           nState;     ///< Record state: inactive, active, removed
+        atomics::atomic<unsigned int>           nAge;       ///< Age of the record
+        atomics::atomic<publication_record *>   pNext;      ///< Next record in active publication list
+        atomics::atomic<publication_record *>   pNextAllocated; ///< Next record in allocated publication list
+
+        /// Initializes publication record
+        publication_record()
+            : nRequest( req_EmptyRecord )
+            , nAge( 0 )
+            , pNext( nullptr )
+            , pNextAllocated( nullptr )
+        {
+            nState.store( inactive, atomics::memory_order_release );
+        }
+
+        /// Returns the value of \p nRequest field
+        unsigned int op( atomics::memory_order mo = atomics::memory_order_relaxed ) const
+        {
+            return nRequest.load( mo );
+        }
+
+        /// Checks if the operation is done
+        bool is_done() const
+        {
+            return nRequest.load( atomics::memory_order_relaxed ) == req_Response;
+        }
+    };
+
+
+}}} // namespace cds::algo::flat_combining
+
+#endif // CDSLIB_ALGO_FLAT_COMBINING_DEFS_H

extern/libcds/cds/algo/flat_combining/kernel.h

@@ -0,0 +1,875 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_FLAT_COMBINING_KERNEL_H
+#define CDSLIB_ALGO_FLAT_COMBINING_KERNEL_H
+
+#include <cds/algo/flat_combining/defs.h>
+#include <cds/algo/flat_combining/wait_strategy.h>
+
+#include <cds/sync/spinlock.h>
+#include <cds/details/allocator.h>
+#include <cds/opt/options.h>
+#include <cds/algo/int_algo.h>
+
+namespace cds { namespace algo {
+
+    /// @defgroup cds_flat_combining_intrusive Intrusive flat combining containers
+    /// @defgroup cds_flat_combining_container Non-intrusive flat combining containers
+
+    /// Flat combining
+    /**
+        @anchor cds_flat_combining_description
+        Flat combining (FC) technique is invented by Hendler, Incze, Shavit and Tzafrir in their paper
+        [2010] <i>"Flat Combining and the Synchronization-Parallelism Tradeoff"</i>.
+        The technique converts a sequential data structure to its concurrent implementation.
+        A few structures are added to the sequential implementation: a <i>global lock</i>,
+        a <i>count</i> of the number of combining passes, and a pointer to the <i>head</i>
+        of a <i>publication list</i>. The publication list is a list of thread-local records
+        of a size proportional to the number of threads that are concurrently accessing the shared object.
+
+        Each thread \p t accessing the structure to perform an invocation of some method \p f()
+        on the shared object executes the following sequence of steps:
+        <ol>
+        <li>Write the invocation opcode and parameters (if any) of the method \p f() to be applied
+        sequentially to the shared object in the <i>request</i> field of your thread local publication
+        record (there is no need to use a load-store memory barrier). The <i>request</i> field will later
+        be used to receive the response. If your thread local publication record is marked as active
+        continue to step 2, otherwise continue to step 5.</li>
+        <li>Check if the global lock is taken. If so (another thread is an active combiner), spin on the <i>request</i>
+        field waiting for a response to the invocation (one can add a yield at this point to allow other threads
+        on the same core to run). Once in a while while spinning check if the lock is still taken and that your
+        record is active (you may use any of \p wait_strategy instead of spinning). If your record is inactive proceed to step 5.
+        Once the response is available, reset the request field to null and return the response.</li>
+        <li>If the lock is not taken, attempt to acquire it and become a combiner. If you fail,
+        return to spinning in step 2.</li>
+        <li>Otherwise, you hold the lock and are a combiner.
+        <ul>
+            <li>Increment the combining pass count by one.</li>
+            <li>Execute a \p fc_apply() by traversing the publication list from the head,
+            combining all non-null method call invocations, setting the <i>age</i> of each of these records
+            to the current <i>count</i>, applying the combined method calls to the structure D, and returning
+            responses to all the invocations. This traversal is guaranteed to be wait-free.</li>
+            <li>If the <i>count</i> is such that a cleanup needs to be performed, traverse the publication
+            list from the <i>head</i>. Starting from the second item (we always leave the item pointed to
+            by the head in the list), remove from the publication list all records whose <i>age</i> is
+            much smaller than the current <i>count</i>. This is done by removing the node and marking it
+            as inactive.</li>
+            <li>Release the lock.</li>
+        </ul>
+        <li>If you have no thread local publication record allocate one, marked as active. If you already
+        have one marked as inactive, mark it as active. Execute a store-load memory barrier. Proceed to insert
+        the record into the list with a successful CAS to the <i>head</i>. Then proceed to step 1.</li>
+        </ol>
+
+        As the test results show, the flat combining technique is suitable for non-intrusive containers
+        like stack, queue, deque. For intrusive concurrent containers the flat combining demonstrates
+        less impressive results.
+
+        \ref cds_flat_combining_container "List of FC-based containers" in libcds.
+
+        \ref cds_flat_combining_intrusive "List of intrusive FC-based containers" in libcds.
+    */
+    namespace flat_combining {
+
+        /// Flat combining internal statistics
+        template <typename Counter = cds::atomicity::event_counter >
+        struct stat
+        {
+            typedef Counter counter_type;   ///< Event counter type
+
+            counter_type    m_nOperationCount   ;   ///< How many operations have been performed
+            counter_type    m_nCombiningCount   ;   ///< Combining call count
+            counter_type    m_nCompactPublicationList; ///< Count of publication list compacting
+            counter_type    m_nDeactivatePubRecord; ///< How many publication records were deactivated during compacting
+            counter_type    m_nActivatePubRecord;   ///< Count of publication record activating
+            counter_type    m_nPubRecordCreated ;   ///< Count of created publication records
+            counter_type    m_nPubRecordDeleted ;   ///< Count of deleted publication records
+            counter_type    m_nPassiveWaitCall;     ///< Count of passive waiting call (\p kernel::wait_for_combining())
+            counter_type    m_nPassiveWaitIteration;///< Count of iteration inside passive waiting
+            counter_type    m_nPassiveWaitWakeup;   ///< Count of forcing wake-up of passive wait cycle
+            counter_type    m_nInvokeExclusive;     ///< Count of call \p kernel::invoke_exclusive()
+            counter_type    m_nWakeupByNotifying;   ///< How many times the passive thread be waked up by a notification
+            counter_type    m_nPassiveToCombiner;   ///< How many times the passive thread becomes the combiner
+
+            /// Returns current combining factor
+            /**
+                Combining factor is how many operations perform in one combine pass:
+                <tt>combining_factor := m_nOperationCount / m_nCombiningCount</tt>
+            */
+            double combining_factor() const
+            {
+                return m_nCombiningCount.get() ? double( m_nOperationCount.get()) / m_nCombiningCount.get() : 0.0;
+            }
+
+            //@cond
+            void    onOperation()               { ++m_nOperationCount;          }
+            void    onCombining()               { ++m_nCombiningCount;          }
+            void    onCompactPublicationList()  { ++m_nCompactPublicationList;  }
+            void    onDeactivatePubRecord()     { ++m_nDeactivatePubRecord;     }
+            void    onActivatePubRecord()       { ++m_nActivatePubRecord;       }
+            void    onCreatePubRecord()         { ++m_nPubRecordCreated;        }
+            void    onDeletePubRecord()         { ++m_nPubRecordDeleted;        }
+            void    onPassiveWait()             { ++m_nPassiveWaitCall;         }
+            void    onPassiveWaitIteration()    { ++m_nPassiveWaitIteration;    }
+            void    onPassiveWaitWakeup()       { ++m_nPassiveWaitWakeup;       }
+            void    onInvokeExclusive()         { ++m_nInvokeExclusive;         }
+            void    onWakeupByNotifying()       { ++m_nWakeupByNotifying;       }
+            void    onPassiveToCombiner()       { ++m_nPassiveToCombiner;       }
+
+            //@endcond
+        };
+
+        /// Flat combining dummy internal statistics
+        struct empty_stat
+        {
+            //@cond
+            void    onOperation()               const {}
+            void    onCombining()               const {}
+            void    onCompactPublicationList()  const {}
+            void    onDeactivatePubRecord()     const {}
+            void    onActivatePubRecord()       const {}
+            void    onCreatePubRecord()         const {}
+            void    onDeletePubRecord()         const {}
+            void    onPassiveWait()             const {}
+            void    onPassiveWaitIteration()    const {}
+            void    onPassiveWaitWakeup()       const {}
+            void    onInvokeExclusive()         const {}
+            void    onWakeupByNotifying()       const {}
+            void    onPassiveToCombiner()       const {}
+            //@endcond
+        };
+
+        /// Type traits of \ref kernel class
+        /**
+            You can define different type traits for \ref kernel
+            by specifying your struct based on \p %traits
+            or by using \ref make_traits metafunction.
+        */
+        struct traits
+        {
+            typedef cds::sync::spin             lock_type;  ///< Lock type
+            typedef cds::algo::flat_combining::wait_strategy::backoff< cds::backoff::delay_of<2>> wait_strategy; ///< Wait strategy
+            typedef CDS_DEFAULT_ALLOCATOR       allocator;  ///< Allocator used for TLS data (allocating \p publication_record derivatives)
+            typedef empty_stat                  stat;       ///< Internal statistics
+            typedef opt::v::relaxed_ordering  memory_model; ///< /// C++ memory ordering model
+        };
+
+        /// Metafunction converting option list to traits
+        /**
+            \p Options are:
+            - \p opt::lock_type - mutex type, default is \p cds::sync::spin
+            - \p opt::wait_strategy - wait strategy, see \p wait_strategy namespace, default is \p wait_strategy::backoff.
+            - \p opt::allocator - allocator type, default is \ref CDS_DEFAULT_ALLOCATOR
+            - \p opt::stat - internal statistics, possible type: \ref stat, \ref empty_stat (the default)
+            - \p opt::memory_model - C++ memory ordering model.
+                List of all available memory ordering see \p opt::memory_model.
+                Default is \p cds::opt::v::relaxed_ordering
+        */
+        template <typename... Options>
+        struct make_traits {
+#   ifdef CDS_DOXYGEN_INVOKED
+            typedef implementation_defined type ;   ///< Metafunction result
+#   else
+            typedef typename cds::opt::make_options<
+                typename cds::opt::find_type_traits< traits, Options... >::type
+                ,Options...
+            >::type   type;
+#   endif
+        };
+
+        /// The kernel of flat combining
+        /**
+            Template parameters:
+            - \p PublicationRecord - a type derived from \ref publication_record
+            - \p Traits - a type traits of flat combining, default is \p flat_combining::traits.
+                \ref make_traits metafunction can be used to create type traits
+
+            The kernel object should be a member of a container class. The container cooperates with flat combining
+            kernel object. There are two ways to interact with the kernel:
+            - One-by-one processing the active records of the publication list. This mode provides by \p combine() function:
+              the container acquires its publication record by \p acquire_record(), fills its fields and calls
+              \p combine() function of its kernel object. If the current thread becomes a combiner, the kernel
+              calls \p fc_apply() function of the container for each active non-empty record. Then, the container
+              should release its publication record by \p release_record(). Only one pass through the publication
+              list is possible.
+            - Batch processing - \p batch_combine() function. It this mode the container obtains access
+              to entire publication list. This mode allows the container to perform an elimination, for example,
+              the stack can collide \p push() and \p pop() requests. The sequence of invocations is the following:
+              the container acquires its publication record by \p acquire_record(), fills its field and call
+              \p batch_combine() function of its kernel object. If the current thread becomes a combiner,
+              the kernel calls \p fc_process() function of the container passing two iterators pointing to
+              the begin and the end of publication list (see \ref iterator class). The iterators allow
+              multiple pass through active records of publication list. For each processed record the container
+              should call \p operation_done() function. On the end, the container should release
+              its record by \p release_record().
+        */
+        template <
+            typename PublicationRecord
+            ,typename Traits = traits
+        >
+        class kernel
+        {
+        public:
+            typedef Traits   traits;                               ///< Type traits
+            typedef typename traits::lock_type global_lock_type;   ///< Global lock type
+            typedef typename traits::wait_strategy wait_strategy;  ///< Wait strategy type
+            typedef typename traits::allocator allocator;          ///< Allocator type (used for allocating publication_record_type data)
+            typedef typename traits::stat      stat;               ///< Internal statistics
+            typedef typename traits::memory_model memory_model;    ///< C++ memory model
+
+            typedef typename wait_strategy::template make_publication_record<PublicationRecord>::type publication_record_type; ///< Publication record type
+
+        protected:
+            //@cond
+            typedef cds::details::Allocator< publication_record_type, allocator >   cxx11_allocator; ///< internal helper cds::details::Allocator
+            typedef std::lock_guard<global_lock_type> lock_guard;
+            //@endcond
+
+        protected:
+            atomics::atomic<unsigned int>  m_nCount;    ///< Total count of combining passes. Used as an age.
+            publication_record_type*    m_pHead;        ///< Head of active publication list
+            publication_record_type*    m_pAllocatedHead; ///< Head of allocated publication list
+            boost::thread_specific_ptr< publication_record_type > m_pThreadRec;   ///< Thread-local publication record
+            mutable global_lock_type    m_Mutex;        ///< Global mutex
+            mutable stat                m_Stat;         ///< Internal statistics
+            unsigned int const          m_nCompactFactor;    ///< Publication list compacting factor (the list will be compacted through \p %m_nCompactFactor combining passes)
+            unsigned int const          m_nCombinePassCount; ///< Number of combining passes
+            wait_strategy               m_waitStrategy;      ///< Wait strategy
+
+        public:
+            /// Initializes the object
+            /**
+                Compact factor = 1024
+
+                Combiner pass count = 8
+            */
+            kernel()
+                : kernel( 1024, 8 )
+            {}
+
+            /// Initializes the object
+            kernel(
+                unsigned int nCompactFactor  ///< Publication list compacting factor (the list will be compacted through \p nCompactFactor combining passes)
+                ,unsigned int nCombinePassCount ///< Number of combining passes for combiner thread
+                )
+                : m_nCount(0)
+                , m_pHead( nullptr )
+                , m_pAllocatedHead( nullptr )
+                , m_pThreadRec( tls_cleanup )
+                , m_nCompactFactor( static_cast<unsigned>( cds::beans::ceil2( static_cast<size_t>( nCompactFactor )) - 1 ))   // binary mask
+                , m_nCombinePassCount( nCombinePassCount )
+            {
+                assert( m_pThreadRec.get() == nullptr );
+                publication_record_type* pRec = cxx11_allocator().New();
+                m_pAllocatedHead =
+                    m_pHead = pRec;
+                m_pThreadRec.reset( pRec );
+                m_Stat.onCreatePubRecord();
+            }
+
+            /// Destroys the object and all publication records
+            ~kernel()
+            {
+                m_pThreadRec.reset();   // calls tls_cleanup()
+
+                // delete all publication records
+                for ( publication_record* p = m_pAllocatedHead; p; ) {
+                    publication_record * pRec = p;
+                    p = p->pNextAllocated.load( memory_model::memory_order_relaxed );
+                    free_publication_record( static_cast<publication_record_type *>( pRec ));
+                }
+            }
+
+            /// Gets publication list record for the current thread
+            /**
+                If there is no publication record for the current thread
+                the function allocates it.
+            */
+            publication_record_type * acquire_record()
+            {
+                publication_record_type * pRec = m_pThreadRec.get();
+                if ( !pRec ) {
+                    // Allocate new publication record
+                    pRec = cxx11_allocator().New();
+                    m_pThreadRec.reset( pRec );
+                    m_Stat.onCreatePubRecord();
+
+                    // Insert in allocated list
+                    assert( m_pAllocatedHead != nullptr );
+                    publication_record* p = m_pAllocatedHead->pNextAllocated.load( memory_model::memory_order_relaxed );
+                    do {
+                        pRec->pNextAllocated.store( p, memory_model::memory_order_release );
+                    } while ( !m_pAllocatedHead->pNextAllocated.compare_exchange_weak( p, pRec, memory_model::memory_order_release, atomics::memory_order_acquire ));
+
+                    publish( pRec );
+                }
+                else if ( pRec->nState.load( memory_model::memory_order_acquire ) != active )
+                    publish( pRec );
+
+                assert( pRec->op() == req_EmptyRecord );
+
+                return pRec;
+            }
+
+            /// Marks publication record for the current thread as empty
+            void release_record( publication_record_type * pRec )
+            {
+                assert( pRec->is_done());
+                pRec->nRequest.store( req_EmptyRecord, memory_model::memory_order_release );
+            }
+
+            /// Trying to execute operation \p nOpId
+            /**
+                \p pRec is the publication record acquiring by \ref acquire_record earlier.
+                \p owner is a container that is owner of flat combining kernel object.
+                As a result the current thread can become a combiner or can wait for
+                another combiner performs \p pRec operation.
+
+                If the thread becomes a combiner, the kernel calls \p owner.fc_apply
+                for each active non-empty publication record.
+            */
+            template <class Container>
+            void combine( unsigned int nOpId, publication_record_type * pRec, Container& owner )
+            {
+                assert( nOpId >= req_Operation );
+                assert( pRec );
+
+                pRec->nRequest.store( nOpId, memory_model::memory_order_release );
+                m_Stat.onOperation();
+
+                try_combining( owner, pRec );
+            }
+
+            /// Trying to execute operation \p nOpId in batch-combine mode
+            /**
+                \p pRec is the publication record acquiring by \p acquire_record() earlier.
+                \p owner is a container that owns flat combining kernel object.
+                As a result the current thread can become a combiner or can wait for
+                another combiner performs \p pRec operation.
+
+                If the thread becomes a combiner, the kernel calls \p owner.fc_process()
+                giving the container the full access over publication list. This function
+                is useful for an elimination technique if the container supports any kind of
+                that. The container can perform multiple pass through publication list.
+
+                \p owner.fc_process() has two arguments - forward iterators on begin and end of
+                publication list, see \ref iterator class. For each processed record the container
+                should call \p operation_done() function to mark the record as processed.
+
+                On the end of \p %batch_combine the \p combine() function is called
+                to process rest of publication records.
+            */
+            template <class Container>
+            void batch_combine( unsigned int nOpId, publication_record_type* pRec, Container& owner )
+            {
+                assert( nOpId >= req_Operation );
+                assert( pRec );
+
+                pRec->nRequest.store( nOpId, memory_model::memory_order_release );
+                m_Stat.onOperation();
+
+                try_batch_combining( owner, pRec );
+            }
+
+            /// Invokes \p Func in exclusive mode
+            /**
+                Some operation in flat combining containers should be called in exclusive mode
+                i.e the current thread should become the combiner to process the operation.
+                The typical example is \p empty() function.
+
+                \p %invoke_exclusive() allows do that: the current thread becomes the combiner,
+                invokes \p f exclusively but unlike a typical usage the thread does not process any pending request.
+                Instead, after end of \p f call the current thread wakes up a pending thread if any.
+            */
+            template <typename Func>
+            void invoke_exclusive( Func f )
+            {
+                {
+                    lock_guard l( m_Mutex );
+                    f();
+                }
+                m_waitStrategy.wakeup( *this );
+                m_Stat.onInvokeExclusive();
+            }
+
+            /// Marks \p rec as executed
+            /**
+                This function should be called by container if \p batch_combine() mode is used.
+                For usual combining (see \p combine()) this function is excess.
+            */
+            void operation_done( publication_record& rec )
+            {
+                rec.nRequest.store( req_Response, memory_model::memory_order_release );
+                m_waitStrategy.notify( *this, static_cast<publication_record_type&>( rec ));
+            }
+
+            /// Internal statistics
+            stat const& statistics() const
+            {
+                return m_Stat;
+            }
+
+            //@cond
+            // For container classes based on flat combining
+            stat& internal_statistics() const
+            {
+                return m_Stat;
+            }
+            //@endcond
+
+            /// Returns the compact factor
+            unsigned int compact_factor() const
+            {
+                return m_nCompactFactor + 1;
+            }
+
+            /// Returns number of combining passes for combiner thread
+            unsigned int combine_pass_count() const
+            {
+                return m_nCombinePassCount;
+            }
+
+        public:
+            /// Publication list iterator
+            /**
+                Iterators are intended for batch processing by container's
+                \p fc_process function.
+                The iterator allows iterate through active publication list.
+            */
+            class iterator
+            {
+                //@cond
+                friend class kernel;
+                publication_record_type * m_pRec;
+                //@endcond
+
+            protected:
+                //@cond
+                iterator( publication_record_type * pRec )
+                    : m_pRec( pRec )
+                {
+                    skip_inactive();
+                }
+
+                void skip_inactive()
+                {
+                    while ( m_pRec && (m_pRec->nState.load( memory_model::memory_order_acquire ) != active
+                                    || m_pRec->op( memory_model::memory_order_relaxed) < req_Operation ))
+                    {
+                        m_pRec = static_cast<publication_record_type*>(m_pRec->pNext.load( memory_model::memory_order_acquire ));
+                    }
+                }
+                //@endcond
+
+            public:
+                /// Initializes an empty iterator object
+                iterator()
+                    : m_pRec( nullptr )
+                {}
+
+                /// Copy ctor
+                iterator( iterator const& src )
+                    : m_pRec( src.m_pRec )
+                {}
+
+                /// Pre-increment
+                iterator& operator++()
+                {
+                    assert( m_pRec );
+                    m_pRec = static_cast<publication_record_type *>( m_pRec->pNext.load( memory_model::memory_order_acquire ));
+                    skip_inactive();
+                    return *this;
+                }
+
+                /// Post-increment
+                iterator operator++(int)
+                {
+                    assert( m_pRec );
+                    iterator it(*this);
+                    ++(*this);
+                    return it;
+                }
+
+                /// Dereference operator, can return \p nullptr
+                publication_record_type* operator ->()
+                {
+                    return m_pRec;
+                }
+
+                /// Dereference operator, the iterator should not be an end iterator
+                publication_record_type& operator*()
+                {
+                    assert( m_pRec );
+                    return *m_pRec;
+                }
+
+                /// Iterator equality
+                friend bool operator==( iterator it1, iterator it2 )
+                {
+                    return it1.m_pRec == it2.m_pRec;
+                }
+
+                /// Iterator inequality
+                friend bool operator!=( iterator it1, iterator it2 )
+                {
+                    return !( it1 == it2 );
+                }
+            };
+
+            /// Returns an iterator to the first active publication record
+            iterator begin()    { return iterator(m_pHead); }
+
+            /// Returns an iterator to the end of publication list. Should not be dereferenced.
+            iterator end()      { return iterator(); }
+
+        public:
+            /// Gets current value of \p rec.nRequest
+            /**
+                This function is intended for invoking from a wait strategy
+            */
+            int get_operation( publication_record& rec )
+            {
+                return rec.op( memory_model::memory_order_acquire );
+            }
+
+            /// Wakes up any waiting thread
+            /**
+                This function is intended for invoking from a wait strategy
+            */
+            void wakeup_any()
+            {
+                publication_record* pRec = m_pHead;
+                while ( pRec ) {
+                    if ( pRec->nState.load( memory_model::memory_order_acquire ) == active
+                      && pRec->op( memory_model::memory_order_acquire ) >= req_Operation )
+                    {
+                        m_waitStrategy.notify( *this, static_cast<publication_record_type&>( *pRec ));
+                        break;
+                    }
+                    pRec = pRec->pNext.load( memory_model::memory_order_acquire );
+                }
+            }
+
+        private:
+            //@cond
+            static void tls_cleanup( publication_record_type* pRec )
+            {
+                // Thread done
+                // pRec that is TLS data should be excluded from publication list
+                pRec->nState.store( removed, memory_model::memory_order_release );
+            }
+
+            void free_publication_record( publication_record_type* pRec )
+            {
+                cxx11_allocator().Delete( pRec );
+                m_Stat.onDeletePubRecord();
+            }
+
+            void publish( publication_record_type* pRec )
+            {
+                assert( pRec->nState.load( memory_model::memory_order_relaxed ) == inactive );
+
+                pRec->nAge.store( m_nCount.load(memory_model::memory_order_relaxed), memory_model::memory_order_relaxed );
+                pRec->nState.store( active, memory_model::memory_order_relaxed );
+
+                // Insert record to publication list
+                if ( m_pHead != static_cast<publication_record *>(pRec)) {
+                    publication_record * p = m_pHead->pNext.load( memory_model::memory_order_relaxed );
+                    if ( p != static_cast<publication_record *>( pRec )) {
+                        do {
+                            pRec->pNext.store( p, memory_model::memory_order_release );
+                            // Failed CAS changes p
+                        } while ( !m_pHead->pNext.compare_exchange_weak( p, static_cast<publication_record *>(pRec),
+                            memory_model::memory_order_release, atomics::memory_order_acquire ));
+                        m_Stat.onActivatePubRecord();
+                    }
+                }
+            }
+
+            void republish( publication_record_type* pRec )
+            {
+                if ( pRec->nState.load( memory_model::memory_order_relaxed ) != active ) {
+                    // The record has been excluded from publication list. Reinsert it
+                    publish( pRec );
+                }
+            }
+
+            template <class Container>
+            void try_combining( Container& owner, publication_record_type* pRec )
+            {
+                if ( m_Mutex.try_lock()) {
+                    // The thread becomes a combiner
+                    lock_guard l( m_Mutex, std::adopt_lock_t());
+
+                    // The record pRec can be excluded from publication list. Re-publish it
+                    republish( pRec );
+
+                    combining( owner );
+                    assert( pRec->op( memory_model::memory_order_relaxed ) == req_Response );
+                }
+                else {
+                    // There is another combiner, wait while it executes our request
+                    if ( !wait_for_combining( pRec )) {
+                        // The thread becomes a combiner
+                        lock_guard l( m_Mutex, std::adopt_lock_t());
+
+                        // The record pRec can be excluded from publication list. Re-publish it
+                        republish( pRec );
+
+                        combining( owner );
+                        assert( pRec->op( memory_model::memory_order_relaxed ) == req_Response );
+                    }
+                }
+            }
+
+            template <class Container>
+            void try_batch_combining( Container& owner, publication_record_type * pRec )
+            {
+                if ( m_Mutex.try_lock()) {
+                    // The thread becomes a combiner
+                    lock_guard l( m_Mutex, std::adopt_lock_t());
+
+                    // The record pRec can be excluded from publication list. Re-publish it
+                    republish( pRec );
+
+                    batch_combining( owner );
+                    assert( pRec->op( memory_model::memory_order_relaxed ) == req_Response );
+                }
+                else {
+                    // There is another combiner, wait while it executes our request
+                    if ( !wait_for_combining( pRec )) {
+                        // The thread becomes a combiner
+                        lock_guard l( m_Mutex, std::adopt_lock_t());
+
+                        // The record pRec can be excluded from publication list. Re-publish it
+                        republish( pRec );
+
+                        batch_combining( owner );
+                        assert( pRec->op( memory_model::memory_order_relaxed ) == req_Response );
+                    }
+                }
+            }
+
+            template <class Container>
+            void combining( Container& owner )
+            {
+                // The thread is a combiner
+                assert( !m_Mutex.try_lock());
+
+                unsigned int const nCurAge = m_nCount.fetch_add( 1, memory_model::memory_order_relaxed ) + 1;
+
+                unsigned int nEmptyPassCount = 0;
+                unsigned int nUsefulPassCount = 0;
+                for ( unsigned int nPass = 0; nPass < m_nCombinePassCount; ++nPass ) {
+                    if ( combining_pass( owner, nCurAge ))
+                        ++nUsefulPassCount;
+                    else if ( ++nEmptyPassCount > nUsefulPassCount )
+                        break;
+                }
+
+                m_Stat.onCombining();
+                if ( ( nCurAge & m_nCompactFactor ) == 0 )
+                    compact_list( nCurAge );
+            }
+
+            template <class Container>
+            bool combining_pass( Container& owner, unsigned int nCurAge )
+            {
+                publication_record* p = m_pHead;
+                bool bOpDone = false;
+                while ( p ) {
+                    switch ( p->nState.load( memory_model::memory_order_acquire )) {
+                    case active:
+                        if ( p->op( memory_model::memory_order_acquire ) >= req_Operation ) {
+                            p->nAge.store( nCurAge, memory_model::memory_order_relaxed );
+                            owner.fc_apply( static_cast<publication_record_type*>( p ));
+                            operation_done( *p );
+                            bOpDone = true;
+                        }
+                        break;
+                    case inactive:
+                        // Only m_pHead can be inactive in the publication list
+                        assert( p == m_pHead );
+                        break;
+                    case removed:
+                        // Such record will be removed on compacting phase
+                        break;
+                    default:
+                        /// ??? That is impossible
+                        assert( false );
+                    }
+                    p = p->pNext.load( memory_model::memory_order_acquire );
+                }
+                return bOpDone;
+            }
+
+            template <class Container>
+            void batch_combining( Container& owner )
+            {
+                // The thread is a combiner
+                assert( !m_Mutex.try_lock());
+
+                unsigned int const nCurAge = m_nCount.fetch_add( 1, memory_model::memory_order_relaxed ) + 1;
+
+                for ( unsigned int nPass = 0; nPass < m_nCombinePassCount; ++nPass )
+                    owner.fc_process( begin(), end());
+
+                combining_pass( owner, nCurAge );
+                m_Stat.onCombining();
+                if ( ( nCurAge & m_nCompactFactor ) == 0 )
+                    compact_list( nCurAge );
+            }
+
+            bool wait_for_combining( publication_record_type* pRec )
+            {
+                m_waitStrategy.prepare( *pRec );
+                m_Stat.onPassiveWait();
+
+                while ( pRec->op( memory_model::memory_order_acquire ) != req_Response ) {
+                    // The record can be excluded from publication list. Reinsert it
+                    republish( pRec );
+
+                    m_Stat.onPassiveWaitIteration();
+
+                    // Wait while operation processing
+                    if ( m_waitStrategy.wait( *this, *pRec ))
+                        m_Stat.onWakeupByNotifying();
+
+                    if ( m_Mutex.try_lock()) {
+                        if ( pRec->op( memory_model::memory_order_acquire ) == req_Response ) {
+                            // Operation is done
+                            m_Mutex.unlock();
+
+                            // Wake up a pending threads
+                            m_waitStrategy.wakeup( *this );
+                            m_Stat.onPassiveWaitWakeup();
+
+                            break;
+                        }
+                        // The thread becomes a combiner
+                        m_Stat.onPassiveToCombiner();
+                        return false;
+                    }
+                }
+                return true;
+            }
+
+            void compact_list( unsigned int nCurAge )
+            {
+                // Compacts publication list
+                // This function is called only by combiner thread
+
+            try_again:
+                publication_record * pPrev = m_pHead;
+                for ( publication_record * p = pPrev->pNext.load( memory_model::memory_order_acquire ); p; ) {
+                    switch ( p->nState.load( memory_model::memory_order_relaxed )) {
+                    case active:
+                        if ( p->nAge.load( memory_model::memory_order_relaxed ) + m_nCompactFactor < nCurAge )
+                        {
+                            publication_record * pNext = p->pNext.load( memory_model::memory_order_relaxed );
+                            if ( pPrev->pNext.compare_exchange_strong( p, pNext,
+                                memory_model::memory_order_acquire, atomics::memory_order_relaxed ))
+                            {
+                                p->nState.store( inactive, memory_model::memory_order_release );
+                                p = pNext;
+                                m_Stat.onDeactivatePubRecord();
+                                continue;
+                            }
+                        }
+                        break;
+
+                    case removed:
+                        publication_record * pNext = p->pNext.load( memory_model::memory_order_acquire );
+                        if ( cds_likely( pPrev->pNext.compare_exchange_strong( p, pNext, memory_model::memory_order_acquire, atomics::memory_order_relaxed ))) {
+                            p = pNext;
+                            continue;
+                        }
+                        else {
+                            // CAS can be failed only in beginning of list
+                            assert( pPrev == m_pHead );
+                            goto try_again;
+                        }
+                    }
+                    pPrev = p;
+                    p = p->pNext.load( memory_model::memory_order_acquire );
+                }
+
+                // Iterate over allocated list to find removed records
+                pPrev = m_pAllocatedHead;
+                for ( publication_record * p = pPrev->pNextAllocated.load( memory_model::memory_order_acquire ); p; ) {
+                    if ( p->nState.load( memory_model::memory_order_relaxed ) == removed ) {
+                        publication_record * pNext = p->pNextAllocated.load( memory_model::memory_order_relaxed );
+                        if ( pPrev->pNextAllocated.compare_exchange_strong( p, pNext, memory_model::memory_order_acquire, atomics::memory_order_relaxed )) {
+                            free_publication_record( static_cast<publication_record_type *>( p ));
+                            p = pNext;
+                            continue;
+                        }
+                    }
+
+                    pPrev = p;
+                    p = p->pNextAllocated.load( memory_model::memory_order_relaxed );
+                }
+
+                m_Stat.onCompactPublicationList();
+            }
+            //@endcond
+        };
+
+        //@cond
+        class container
+        {
+        public:
+            template <typename PubRecord>
+            void fc_apply( PubRecord * )
+            {
+                assert( false );
+            }
+
+            template <typename Iterator>
+            void fc_process( Iterator, Iterator )
+            {
+                assert( false );
+            }
+        };
+        //@endcond
+
+    } // namespace flat_combining
+}} // namespace cds::algo
+
+/*
+  CppMem model  (http://svr-pes20-cppmem.cl.cam.ac.uk/cppmem/)
+
+  // Combiner thread - slave (waiting) thread
+int main() {
+      atomic_int y = 0;  // pRec->op
+      int x = 0;         // pRec->data
+   {{{
+      { // slave thread (not combiner)
+        // Op data
+        x = 1;
+        // Annotate request (op)
+        y.store(1, release);
+        // Wait while request done
+        y.load(acquire).readsvalue(2);
+        // Read result
+        r2=x;
+      }
+   |||
+      { // Combiner thread
+        // Read request (op)
+        r1=y.load(acquire).readsvalue(1);
+        // Execute request - change request data
+        x = 2;
+        // store "request processed" flag (pRec->op := req_Response)
+        y.store(2, release);
+      }
+   }}};
+   return 0;
+}
+
+*/
+
+#endif // #ifndef CDSLIB_ALGO_FLAT_COMBINING_KERNEL_H

extern/libcds/cds/algo/flat_combining/wait_strategy.h

@@ -0,0 +1,417 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_FLAT_COMBINING_WAIT_STRATEGY_H
+#define CDSLIB_ALGO_FLAT_COMBINING_WAIT_STRATEGY_H
+
+#include <cds/algo/flat_combining/defs.h>
+#include <cds/algo/backoff_strategy.h>
+#include <mutex>
+#include <condition_variable>
+#include <boost/thread/tss.hpp>  // thread_specific_ptr
+
+
+namespace cds { namespace opt {
+
+    /// Wait strategy option for \p flat_combining::kernel
+    template <typename Strategy>
+    struct wait_strategy {
+        //@cond
+        template <typename Base> struct pack: public Base
+        {
+            typedef Strategy wait_strategy;
+        };
+        //@endcond
+    };
+
+}} // namespace cds::opt
+
+namespace cds { namespace algo { namespace flat_combining {
+
+    /// Wait strategies for \p flat_combining technique
+    /**
+        Wait strategy specifies how a thread waits until its request is performed by the combiner.
+        See \p wait_strategy::empty wait strategy to explain the interface.
+    */
+    namespace wait_strategy {
+
+        /// Empty wait strategy
+        /**
+            Empty wait strategy is just spinning on request field.
+            All functions are empty.
+        */
+        struct empty
+        {
+            /// Metafunction for defining a publication record for flat combining technique
+            /**
+                Any wait strategy may expand the publication record for storing
+                its own private data.
+                \p PublicationRecord is the type specified by \p flat_combining::kernel.
+                - If the strategy has no thread-private data, it should typedef \p PublicationRecord
+                  as a return \p type of metafunction.
+                - Otherwise, if the strategy wants to store anything in thread-local data,
+                  it should expand \p PublicationRecord, for example:
+                  \code
+                  template <typename PublicationRecord>
+                  struct make_publication_record {
+                    struct type: public PublicationRecord
+                    {
+                        int strategy_data;
+                    };
+                  };
+                  \endcode
+            */
+            template <typename PublicationRecord>
+            struct make_publication_record {
+                typedef PublicationRecord type; ///< Metafunction result
+            };
+
+            /// Prepares the strategy
+            /**
+                This function is called before enter to waiting cycle.
+                Some strategies need to prepare its thread-local data in \p rec.
+
+                \p PublicationRecord is thread's publication record of type \p make_publication_record::type
+            */
+            template <typename PublicationRecord>
+            void prepare( PublicationRecord& rec )
+            {
+                CDS_UNUSED( rec );
+            }
+
+            /// Waits for the combiner
+            /**
+                The thread calls this function to wait for the combiner process
+                the request.
+                The function returns \p true if the thread was waked up by the combiner,
+                otherwise it should return \p false.
+
+                \p FCKernel is a \p flat_combining::kernel object,
+                \p PublicationRecord is thread's publication record of type \p make_publication_record::type
+            */
+            template <typename FCKernel, typename PublicationRecord>
+            bool wait( FCKernel& fc, PublicationRecord& rec )
+            {
+                CDS_UNUSED( fc );
+                CDS_UNUSED( rec );
+                return false;
+            }
+
+            /// Wakes up the thread
+            /**
+                The combiner calls \p %notify() when it has been processed the request.
+
+                \p FCKernel is a \p flat_combining::kernel object,
+                \p PublicationRecord is thread's publication record of type \p make_publication_record::type
+            */
+            template <typename FCKernel, typename PublicationRecord>
+            void notify( FCKernel& fc, PublicationRecord& rec )
+            {
+                CDS_UNUSED( fc );
+                CDS_UNUSED( rec );
+            }
+
+            /// Moves control to other thread
+            /**
+                This function is called when the thread becomes the combiner
+                but the request of the thread is already processed.
+                The strategy may call \p fc.wakeup_any() instructs the kernel
+                to wake up any pending thread.
+
+                \p FCKernel is a \p flat_combining::kernel object,
+            */
+            template <typename FCKernel>
+            void wakeup( FCKernel& fc )
+            {
+                CDS_UNUSED( fc );
+            }
+        };
+
+        /// Back-off wait strategy
+        /**
+            Template argument \p Backoff specifies back-off strategy, default is cds::backoff::delay_of<2>
+        */
+        template <typename BackOff = cds::backoff::delay_of<2>>
+        struct backoff
+        {
+            typedef BackOff back_off;   ///< Back-off strategy
+
+            /// Incorporates back-off strategy into publication record
+            template <typename PublicationRecord>
+            struct make_publication_record
+            {
+                //@cond
+                struct type: public PublicationRecord
+                {
+                    back_off bkoff;
+                };
+                //@endcond
+            };
+
+            /// Resets back-off strategy in \p rec
+            template <typename PublicationRecord>
+            void prepare( PublicationRecord& rec )
+            {
+                rec.bkoff.reset();
+            }
+
+            /// Calls back-off strategy
+            template <typename FCKernel, typename PublicationRecord>
+            bool wait( FCKernel& /*fc*/, PublicationRecord& rec )
+            {
+                rec.bkoff();
+                return false;
+            }
+
+            /// Does nothing
+            template <typename FCKernel, typename PublicationRecord>
+            void notify( FCKernel& /*fc*/, PublicationRecord& /*rec*/ )
+            {}
+
+            /// Does nothing
+            template <typename FCKernel>
+            void wakeup( FCKernel& )
+            {}
+        };
+
+        /// Wait strategy based on the single mutex and the condition variable
+        /**
+            The strategy shares the mutex and conditional variable for all thread.
+
+            Template parameter \p Milliseconds specifies waiting duration;
+            the minimal value is 1.
+        */
+        template <int Milliseconds = 2>
+        class single_mutex_single_condvar
+        {
+        //@cond
+            std::mutex  m_mutex;
+            std::condition_variable m_condvar;
+            bool        m_wakeup;
+
+            typedef std::unique_lock< std::mutex > unique_lock;
+        //@endcond
+
+        public:
+            enum {
+                c_nWaitMilliseconds = Milliseconds < 1 ? 1 : Milliseconds ///< Waiting duration
+            };
+
+            /// Empty metafunction
+            template <typename PublicationRecord>
+            struct make_publication_record {
+                typedef PublicationRecord type; ///< publication record type
+            };
+
+            /// Default ctor
+            single_mutex_single_condvar()
+                : m_wakeup( false )
+            {}
+
+            /// Does nothing
+            template <typename PublicationRecord>
+            void prepare( PublicationRecord& /*rec*/ )
+            {}
+
+            /// Sleeps on condition variable waiting for notification from combiner
+            template <typename FCKernel, typename PublicationRecord>
+            bool wait( FCKernel& fc, PublicationRecord& rec )
+            {
+                if ( fc.get_operation( rec ) >= req_Operation ) {
+                    unique_lock lock( m_mutex );
+                    if ( fc.get_operation( rec ) >= req_Operation ) {
+                        if ( m_wakeup ) {
+                            m_wakeup = false;
+                            return true;
+                        }
+
+                        bool ret = m_condvar.wait_for( lock, std::chrono::milliseconds( c_nWaitMilliseconds )) == std::cv_status::no_timeout;
+                        m_wakeup = false;
+                        return ret;
+                    }
+                }
+                return false;
+            }
+
+            /// Calls condition variable function \p notify_all()
+            template <typename FCKernel, typename PublicationRecord>
+            void notify( FCKernel& fc, PublicationRecord& /*rec*/ )
+            {
+                wakeup( fc );
+            }
+
+            /// Calls condition variable function \p notify_all()
+            template <typename FCKernel>
+            void wakeup( FCKernel& /*fc*/ )
+            {
+                unique_lock lock( m_mutex );
+                m_wakeup = true;
+                m_condvar.notify_all();
+            }
+        };
+
+        /// Wait strategy based on the single mutex and thread-local condition variables
+        /**
+            The strategy shares the mutex, but each thread has its own conditional variable
+
+            Template parameter \p Milliseconds specifies waiting duration;
+            the minimal value is 1.
+        */
+        template <int Milliseconds = 2>
+        class single_mutex_multi_condvar
+        {
+        //@cond
+            std::mutex  m_mutex;
+            bool        m_wakeup;
+
+            typedef std::unique_lock< std::mutex > unique_lock;
+        //@endcond
+
+        public:
+            enum {
+                c_nWaitMilliseconds = Milliseconds < 1 ? 1 : Milliseconds  ///< Waiting duration
+            };
+
+            /// Incorporates a condition variable into \p PublicationRecord
+            template <typename PublicationRecord>
+            struct make_publication_record {
+                /// Metafunction result
+                struct type: public PublicationRecord
+                {
+                    //@cond
+                    std::condition_variable m_condvar;
+                    //@endcond
+                };
+            };
+
+            /// Default ctor
+            single_mutex_multi_condvar()
+                : m_wakeup( false )
+            {}
+
+            /// Does nothing
+            template <typename PublicationRecord>
+            void prepare( PublicationRecord& /*rec*/ )
+            {}
+
+            /// Sleeps on condition variable waiting for notification from combiner
+            template <typename FCKernel, typename PublicationRecord>
+            bool wait( FCKernel& fc, PublicationRecord& rec )
+            {
+                if ( fc.get_operation( rec ) >= req_Operation ) {
+                    unique_lock lock( m_mutex );
+
+                    if ( fc.get_operation( rec ) >= req_Operation ) {
+                        if ( m_wakeup ) {
+                            m_wakeup = false;
+                            return true;
+                        }
+
+                        bool ret = rec.m_condvar.wait_for( lock, std::chrono::milliseconds( c_nWaitMilliseconds )) == std::cv_status::no_timeout;
+                        m_wakeup = false;
+                        return ret;
+                    }
+                }
+                return false;
+            }
+
+            /// Calls condition variable function \p notify_one()
+            template <typename FCKernel, typename PublicationRecord>
+            void notify( FCKernel& /*fc*/, PublicationRecord& rec )
+            {
+                unique_lock lock( m_mutex );
+                m_wakeup = true;
+                rec.m_condvar.notify_one();
+            }
+
+            /// Calls \p fc.wakeup_any() to wake up any pending thread
+            template <typename FCKernel>
+            void wakeup( FCKernel& fc )
+            {
+                fc.wakeup_any();
+            }
+        };
+
+        /// Wait strategy where each thread has a mutex and a condition variable
+        /**
+            Template parameter \p Milliseconds specifies waiting duration;
+            the minimal value is 1.
+        */
+        template <int Milliseconds = 2>
+        class multi_mutex_multi_condvar
+        {
+        //@cond
+            typedef std::unique_lock< std::mutex > unique_lock;
+        //@endcond
+        public:
+            enum {
+                c_nWaitMilliseconds = Milliseconds < 1 ? 1 : Milliseconds   ///< Waiting duration
+            };
+
+            /// Incorporates a condition variable and a mutex into \p PublicationRecord
+            template <typename PublicationRecord>
+            struct make_publication_record {
+                /// Metafunction result
+                struct type: public PublicationRecord
+                {
+                    //@cond
+                    std::mutex              m_mutex;
+                    std::condition_variable m_condvar;
+                    bool                    m_wakeup;
+
+                    type()
+                        : m_wakeup( false )
+                    {}
+                    //@endcond
+                };
+            };
+
+            /// Does nothing
+            template <typename PublicationRecord>
+            void prepare( PublicationRecord& /*rec*/ )
+            {}
+
+            /// Sleeps on condition variable waiting for notification from combiner
+            template <typename FCKernel, typename PublicationRecord>
+            bool wait( FCKernel& fc, PublicationRecord& rec )
+            {
+                if ( fc.get_operation( rec ) >= req_Operation ) {
+                    unique_lock lock( rec.m_mutex );
+
+                    if ( fc.get_operation( rec ) >= req_Operation ) {
+                        if ( rec.m_wakeup ) {
+                            rec.m_wakeup = false;
+                            return true;
+                        }
+
+                        bool ret = rec.m_condvar.wait_for( lock, std::chrono::milliseconds( c_nWaitMilliseconds )) == std::cv_status::no_timeout;
+                        rec.m_wakeup = false;
+                        return ret;
+                    }
+                }
+                return false;
+            }
+
+            /// Calls condition variable function \p notify_one()
+            template <typename FCKernel, typename PublicationRecord>
+            void notify( FCKernel& /*fc*/, PublicationRecord& rec )
+            {
+                unique_lock lock( rec.m_mutex );
+                rec.m_wakeup = true;
+                rec.m_condvar.notify_one();
+            }
+
+            /// Calls \p fc.wakeup_any() to wake up any pending thread
+            template <typename FCKernel>
+            void wakeup( FCKernel& fc )
+            {
+                fc.wakeup_any();
+            }
+        };
+
+    } // namespace wait_strategy
+}}} // namespace cds::algo::flat_combining
+
+#endif //CDSLIB_ALGO_FLAT_COMBINING_WAIT_STRATEGY_H

extern/libcds/cds/algo/int_algo.h

@@ -0,0 +1,147 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_INT_ALGO_H
+#define CDSLIB_INT_ALGO_H
+
+#include <cds/algo/bitop.h>
+
+namespace cds { namespace beans {
+    /// Returns largest previous integer for <tt>log2( n )</tt>
+    static inline size_t log2floor( size_t n )
+    {
+        return n ? cds::bitop::MSBnz( n ) : 0;
+    }
+
+    /// Returns smallest following integer for <tt>log2( n )</tt>
+    static inline size_t log2ceil( size_t n )
+    {
+        size_t i = log2floor( n );
+        return ( size_t( 1 ) << i ) < n ? i + 1 : i;
+    }
+
+    /// Returns largest previous power of 2 for \p n
+    /**
+        Examples:
+        \code
+        floor2(0) == 1   // !!!
+        floor2(1) == 1
+        floor2(2) == 2
+        floor2(3) == 2
+        floor2(4) == 4
+        floor2(15) == 8
+        floor2(16) == 16
+        floor2(17) == 16
+        \endcode
+    */
+    static inline size_t floor2( size_t n )
+    {
+        return size_t(1) << log2floor( n );
+    }
+
+    /// Returns smallest following power of 2 for \p n
+    /**
+        Examples:
+        \code
+        ceil2(0) == 1   // !!!
+        ceil2(1) == 1
+        ceil2(2) == 2
+        ceil2(3) == 4
+        ceil2(4) == 4
+        ceil2(15) == 16
+        ceil2(16) == 16
+        ceil2(17) == 32
+        \endcode
+    */
+    static inline size_t ceil2( size_t n )
+    {
+        return size_t(1) << log2ceil( n );
+    }
+
+    /// Checks if \p n is power of 2
+    constexpr static inline bool is_power2( size_t n ) noexcept
+    {
+        return (n & (n - 1)) == 0 && n;
+    }
+
+    /// Returns binary logarithm of \p n if \p n is power of two, otherwise returns 0
+    static inline size_t log2( size_t n )
+    {
+        return is_power2(n) ? log2floor(n) : 0;
+    }
+
+#if CDS_BUILD_BITS == 32
+    //@cond
+    // 64bit specializations
+
+/// Returns largest previous integer for <tt>log2( n )</tt>
+    static inline uint64_t log2floor( uint64_t n )
+    {
+        return n ? cds::bitop::MSBnz( n ) : 0;
+    }
+
+/// Returns smallest following integer for <tt>log2( n )</tt>
+    static inline uint64_t log2ceil( uint64_t n )
+    {
+        uint64_t i = log2floor( n );
+        return (uint64_t( 1 ) << i) < n ? i + 1 : i;
+    }
+
+/// Returns largest previous power of 2 for \p n
+    /**
+        Examples:
+        \code
+        floor2(0) == 1   // !!!
+        floor2(1) == 1
+        floor2(2) == 2
+        floor2(3) == 2
+        floor2(4) == 4
+        floor2(15) == 8
+        floor2(16) == 16
+        floor2(17) == 16
+        \endcode
+    */
+    static inline uint64_t floor2( uint64_t n )
+    {
+        return uint64_t( 1 ) << log2floor( n );
+    }
+
+/// Returns smallest following power of 2 for \p n
+    /**
+        Examples:
+        \code
+        ceil2(0) == 1   // !!!
+        ceil2(1) == 1
+        ceil2(2) == 2
+        ceil2(3) == 4
+        ceil2(4) == 4
+        ceil2(15) == 16
+        ceil2(16) == 16
+        ceil2(17) == 32
+        \endcode
+    */
+    static inline uint64_t ceil2( uint64_t n )
+    {
+        return uint64_t( 1 ) << log2ceil( n );
+    }
+
+/// Checks if \p n is power of 2
+    constexpr static inline bool is_power2( uint64_t n ) noexcept
+    {
+        return (n & (n - 1)) == 0 && n;
+    }
+
+/// Returns binary logarithm of \p n if \p n is power of two, otherwise returns 0
+    static inline uint64_t log2( uint64_t n )
+    {
+        return is_power2( n ) ? log2floor( n ) : 0;
+    }
+
+    //@endcond
+#endif //#if CDS_BUILD_BITS == 32
+
+}}   // namespace cds::beans
+
+#endif  // #ifndef CDSLIB_INT_ALGO_H

extern/libcds/cds/algo/split_bitstring.h

@@ -0,0 +1,445 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_ALGO_SPLIT_BITSTRING_H
+#define CDSLIB_ALGO_SPLIT_BITSTRING_H
+
+#include <cds/algo/base.h>
+
+namespace cds { namespace algo {
+
+    /// Cuts a bit sequence from fixed-size bit-string
+    /**
+        The splitter can be used as an iterator over bit-string.
+        Each call of \p cut() or \p safe_cut() cuts the bit count specified
+        and keeps the position inside bit-string for the next call.
+
+        The splitter stores a const reference to bit-string, not a copy.
+        The maximum count of bits that can be cut in a single call is <tt> sizeof(UInt) * 8 </tt>
+
+        The splitter keeps byte order.
+
+        Template parameters:
+        - \p BitString - a fixed-sized type that interprets as bit string
+        - \p BitStringSize - the size of \p BitString in bytes, default is <tt>sizeof( BitString )</tt>.
+             You can specify 0 for default.
+        - \p UInt - an unsigned integer, return type for \p cut(), default is \p unsigned
+
+        There are specialized splitters:
+        - a simplified \p byte_splitter algorithm that is suitable when count is multiple of 8.
+        - \p number_splitter algorithm is suitable for a number
+    */
+    template <typename BitString, size_t BitStringSize = sizeof( BitString ), typename UInt = unsigned >
+    class split_bitstring
+    {
+    public:
+        typedef BitString bitstring;    ///< Bit-string type
+        typedef UInt      uint_type;    ///< Result type of \p cut() function
+        static constexpr size_t const c_bitstring_size = BitStringSize ? BitStringSize : sizeof( BitString ); ///< size of \p BitString in bytes
+
+        //@cond
+        static constexpr unsigned const c_nBitPerByte = 8;
+        //@endcond
+
+    public:
+        /// Initializises the splitter with reference to \p h and zero start bit offset
+        explicit split_bitstring( bitstring const& h )
+            : cur_( reinterpret_cast<uint8_t const*>( &h ))
+            , offset_( 0 )
+            , first_( cur_ )
+            , last_( cur_ + c_bitstring_size )
+        {}
+
+        /// Initializises the splitter with reference to \p h and start bit offset \p nBitOffset
+        split_bitstring( bitstring const& h, size_t nBitOffset )
+            : cur_( reinterpret_cast<uint8_t const*>( &h ) + nBitOffset / c_nBitPerByte )
+            , offset_( nBitOffset % c_nBitPerByte  )
+            , first_( reinterpret_cast<uint8_t const*>( &h ))
+            , last_( first_ + c_bitstring_size )
+        {}
+
+        /// Returns \p true if end-of-string is not reached yet
+        explicit operator bool() const
+        {
+            return !eos();
+        }
+
+        /// Returns \p true if end-of-stream encountered
+        bool eos() const
+        {
+            return cur_ >= last_;
+        }
+
+        /// Cuts next \p count bits from bit-string
+        /**
+            For performance reason, the function does not manage out-of-bound condition.
+            To control that use \p safe_cut().
+        */
+        uint_type cut( unsigned count )
+        {
+            assert( !eos());
+
+            uint_type result = 0;
+#       if defined( CDS_ARCH_LITTLE_ENDIAN )
+            for ( unsigned done = 0; done < count; ) {
+                assert( cur_ < last_ );
+                unsigned bits = count - done;
+                if ( bits > c_nBitPerByte - offset_ )
+                    bits = c_nBitPerByte - offset_;
+
+                result |= static_cast<uint_type>(( *cur_ >> offset_ ) & (( 1 << bits ) - 1 )) << done;
+
+                offset_ += bits;
+                assert( offset_ <= c_nBitPerByte );
+                if ( offset_ == c_nBitPerByte ) {
+                    offset_ = 0;
+                    ++cur_;
+                }
+                done += bits;
+            }
+#       else
+            while ( count ) {
+                assert( cur_ < last_ );
+
+                unsigned bits = count <= ( c_nBitPerByte - offset_ ) ? count : c_nBitPerByte - offset_;
+
+                result = ( result << bits ) | (( *cur_ >> offset_ ) & ( ( 1 << bits ) - 1 ));
+
+                offset_ += bits;
+                assert( offset_ <= c_nBitPerByte );
+                if ( offset_ == c_nBitPerByte ) {
+                    offset_ = 0;
+                    ++cur_;
+                }
+                count -= bits;
+            }
+#       endif
+
+            return result;
+        }
+
+        /// Cuts up to \p count from the bit-string
+        /**
+            Safe analog of \p cut() but if \p count is more than the rest of bit-string,
+            only the rest is returned.
+            When \p eos() condition is met the function returns 0.
+        */
+        uint_type safe_cut( unsigned count )
+        {
+            if ( eos())
+                return 0;
+
+            unsigned const rest = static_cast<unsigned>( last_ - cur_ - 1 ) * c_nBitPerByte + ( c_nBitPerByte - offset_ );
+            if ( rest < count )
+                count = rest;
+            return count ? cut( count ) : 0;
+        }
+
+        /// Resets the splitter
+        void reset() noexcept
+        {
+            cur_ = first_;
+            offset_ = 0;
+        }
+
+        /// Returns pointer to source bitstring
+        bitstring const * source() const
+        {
+            return reinterpret_cast<bitstring const *>( first_ );
+        }
+
+        /// Returns current bit offset from beginning of bit-string
+        size_t bit_offset() const
+        {
+            return offset_ + (cur_ - first_) * c_nBitPerByte;
+        }
+
+        /// Returns how many bits remain
+        size_t rest_count() const
+        {
+            return c_bitstring_size * c_nBitPerByte - bit_offset();
+        }
+
+        /// Returns \p true for any argument
+        static constexpr bool is_correct( unsigned /*count*/ )
+        {
+            return true;
+        }
+
+    private:
+        //@cond
+        uint8_t const*  cur_;
+        unsigned        offset_;
+        uint8_t const* const    first_;
+        uint8_t const* const    last_;
+        //@endcond
+    };
+
+    /// Simplified \p split_bitstring algorithm when \p count is multiple of 8
+    template <typename BitString, size_t BitStringSize = sizeof( BitString ), typename UInt = unsigned >
+    class byte_splitter
+    {
+    public:
+        typedef BitString bitstring;    ///< Bit-string type
+        typedef UInt      uint_type;    ///< Result type of \p cut() function
+        static constexpr size_t const c_bitstring_size = BitStringSize ? BitStringSize : sizeof( BitString ); ///< size of \p BitString in bytes
+
+        //@cond
+        static constexpr unsigned const c_nBitPerByte = 8;
+        //@endcond
+
+    public:
+        /// Initializises the splitter with reference to \p h and zero start bit offset
+        explicit byte_splitter( bitstring const& h )
+            : cur_( reinterpret_cast<uint8_t const*>( &h ))
+            , first_( cur_ )
+            , last_( cur_ + c_bitstring_size )
+        {}
+
+        /// Initializises the splitter with reference to \p h and start bit offset \p nBitOffset
+        byte_splitter( bitstring const& h, size_t nBitOffset )
+            : cur_( reinterpret_cast<uint8_t const*>( &h ) + nBitOffset / c_nBitPerByte )
+            , first_( reinterpret_cast<uint8_t const*>( &h ))
+            , last_( first_ + c_bitstring_size )
+        {
+            assert( is_correct( static_cast<unsigned>( nBitOffset )));
+            assert( !eos());
+        }
+
+        /// Returns \p true if end-of-string is not reached yet
+        explicit operator bool() const
+        {
+            return !eos();
+        }
+
+        /// Returns \p true if end-of-stream encountered
+        bool eos() const
+        {
+            return cur_ >= last_;
+        }
+
+        /// Cuts next \p count bits (must be multiplier of 8) from bit-string
+        /**
+            For performance reason, the function does not manage out-of-bound condition.
+            To control that use \p safe_cut().
+        */
+        uint_type cut( unsigned count )
+        {
+            assert( !eos());
+            assert( is_correct( count ));
+
+            uint_type result = 0;
+
+#       if defined( CDS_ARCH_LITTLE_ENDIAN )
+            for ( unsigned i = 0; i < count; i += c_nBitPerByte ) {
+                result |= static_cast<uint_type>( *cur_ ) << i;
+                ++cur_;
+            }
+#       else
+            for ( ; count; count -= c_nBitPerByte ) {
+                result = ( result << c_nBitPerByte ) | *cur_;
+                ++cur_;
+            }
+#       endif
+
+            return result;
+        }
+
+        /// Cuts up to \p count from the bit-string
+        /**
+            Safe analog of \p cut(): if \p count is more than the rest of bit-string,
+            only the rest is returned.
+            When \p eos() condition is met the function returns 0.
+        */
+        uint_type safe_cut( unsigned count )
+        {
+            if ( eos())
+                return 0;
+
+            unsigned const rest = static_cast<unsigned>( last_ - cur_ - 1 ) * c_nBitPerByte;
+            if ( rest < count )
+                count = rest;
+            return count ? cut( count ) : 0;
+        }
+
+        /// Resets the splitter
+        void reset() noexcept
+        {
+            cur_ = first_;
+        }
+
+        /// Returns pointer to source bitstring
+        bitstring const* source() const
+        {
+            return reinterpret_cast<bitstring const *>( first_ );
+        }
+
+        /// Returns current bit offset from beginning of bit-string
+        size_t bit_offset() const
+        {
+            return (cur_ - first_) * c_nBitPerByte;
+        }
+
+        /// Returns how many bits remain
+        size_t rest_count() const
+        {
+            return c_bitstring_size * c_nBitPerByte - bit_offset();
+        }
+
+        /// Checks if \p count is multiple of 8
+        static constexpr bool is_correct( unsigned count )
+        {
+            return count % 8 == 0;
+        }
+
+    private:
+        //@cond
+        uint8_t const*  cur_;
+        uint8_t const* const    first_;
+        uint8_t const* const    last_;
+        //@endcond
+    };
+
+
+    /// Cuts a bit sequence from a number
+    /**
+        The splitter can be used as an iterator over bit representation of the number of type \p Int.
+        Each call of \p cut() or \p safe_cut() cuts the bit count specified
+        and keeps the position inside the number for the next call.
+    */
+    template <typename Int>
+    class number_splitter
+    {
+    public:
+        typedef Int       int_type;     ///< Number type
+        typedef Int       uint_type;    ///< Result type of \p cut() function
+
+        //@cond
+        static constexpr unsigned const c_nBitPerByte = 8;
+        //@endcond
+
+    public:
+        /// Initalizes the splitter with nymber \p n and initial bit offset 0
+        explicit number_splitter( int_type n )
+            : number_( n )
+            , shift_( 0 )
+        {}
+
+        /// Initalizes the splitter with nymber \p n and initial bit offset \p initial_offset
+        number_splitter( int_type n, size_t initial_offset )
+            : number_( n )
+            , shift_( static_cast<unsigned>( initial_offset ))
+        {
+            assert( initial_offset < sizeof( int_type ) * c_nBitPerByte );
+        }
+
+        /// Returns \p true if end-of-string is not reached yet
+        explicit operator bool() const
+        {
+            return !eos();
+        }
+
+        /// Returns \p true if end-of-stream encountered
+        bool eos() const
+        {
+            return shift_ >= sizeof( int_type ) * c_nBitPerByte;
+        }
+
+        /// Cuts next \p count bits (must be multiplier of 8) from the number
+        /**
+            For performance reason, the function does not manage out-of-bound condition.
+            To control that use \p safe_cut().
+        */
+        int_type cut( unsigned count )
+        {
+            assert( !eos());
+            assert( is_correct( count ));
+
+            int_type result = ( number_ >> shift_ ) & (( 1 << count ) - 1 );
+            shift_ += count;
+
+            return result;
+        }
+
+        /// Cuts up to \p count from the bit-string
+        /**
+            Safe analog of \p cut(): if \p count is more than the rest of \p int_type,
+            only the rest is returned.
+            When \p eos() condition is met the function returns 0.
+        */
+        int_type safe_cut( unsigned count )
+        {
+            if ( eos())
+                return 0;
+
+            unsigned rest = static_cast<unsigned>( rest_count());
+            if ( rest < count )
+                count = rest;
+            return count ? cut( count ) : 0;
+        }
+
+        /// Resets the splitter
+        void reset() noexcept
+        {
+            shift_ = 0;
+        }
+
+        /// Returns initial number
+        int_type source() const
+        {
+            return number_;
+        }
+
+        /// Returns current bit offset from beginning of the number
+        size_t bit_offset() const
+        {
+            return shift_;
+        }
+
+        /// Returns how many bits remain
+        size_t rest_count() const
+        {
+            return sizeof( int_type ) * c_nBitPerByte - shift_;
+        }
+
+        /// Checks if \p count is multiple of 8
+        static constexpr bool is_correct( unsigned count )
+        {
+            return count < sizeof( int_type ) * c_nBitPerByte;
+        }
+
+    private:
+        //@cond
+        int_type const  number_;
+        unsigned        shift_;
+        //@endcond
+    };
+
+    /// Metafunctin to select a most suitable splitter for type \p BitString of size \p BitStringSize
+    template <typename BitString, size_t BitStringSize >
+    struct select_splitter
+    {
+        typedef split_bitstring< BitString, BitStringSize > type; ///< metafunction result
+    };
+
+    //@cond
+#   define CDS_SELECT_NUMBER_SPLITTER( num_type ) \
+        template <> struct select_splitter<num_type, sizeof(num_type)> { typedef number_splitter<num_type> type; }
+
+    CDS_SELECT_NUMBER_SPLITTER( int );
+    CDS_SELECT_NUMBER_SPLITTER( unsigned );
+    CDS_SELECT_NUMBER_SPLITTER( short );
+    CDS_SELECT_NUMBER_SPLITTER( unsigned short );
+    CDS_SELECT_NUMBER_SPLITTER( long );
+    CDS_SELECT_NUMBER_SPLITTER( unsigned long );
+    CDS_SELECT_NUMBER_SPLITTER( long long );
+    CDS_SELECT_NUMBER_SPLITTER( unsigned long long );
+
+#   undef CDS_SELECT_NUMBER_SPLITTER
+    //@endcond
+
+}} // namespace cds::algo
+
+#endif // #ifndef CDSLIB_ALGO_SPLIT_BITSTRING_H

extern/libcds/cds/compiler/backoff.h

@@ -0,0 +1,39 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_BACKOFF_IMPL_H
+#define CDSLIB_COMPILER_BACKOFF_IMPL_H
+
+#include <cds/details/defs.h>
+
+#if CDS_COMPILER == CDS_COMPILER_MSVC || (CDS_COMPILER == CDS_COMPILER_INTEL && CDS_OS_INTERFACE == CDS_OSI_WINDOWS)
+#   if CDS_PROCESSOR_ARCH == CDS_PROCESSOR_X86
+#       include <cds/compiler/vc/x86/backoff.h>
+#   elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_AMD64
+#       include <cds/compiler/vc/amd64/backoff.h>
+#   else
+#       error "MS VC++ compiler: unsupported processor architecture"
+#   endif
+#elif CDS_COMPILER == CDS_COMPILER_GCC || CDS_COMPILER == CDS_COMPILER_CLANG || CDS_COMPILER == CDS_COMPILER_INTEL
+#   if CDS_PROCESSOR_ARCH == CDS_PROCESSOR_X86
+#       include <cds/compiler/gcc/x86/backoff.h>
+#   elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_AMD64
+#       include <cds/compiler/gcc/amd64/backoff.h>
+#   elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_IA64
+#       include <cds/compiler/gcc/ia64/backoff.h>
+#   elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_SPARC
+#       include <cds/compiler/gcc/sparc/backoff.h>
+#   elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_PPC64
+#       include <cds/compiler/gcc/ppc64/backoff.h>
+#   elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_ARM7
+#       include <cds/compiler/gcc/arm7/backoff.h>
+#   elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_ARM8
+#       include <cds/compiler/gcc/arm8/backoff.h>
+#   endif
+#else
+#   error "Undefined compiler"
+#endif
+
+#endif  // #ifndef CDSLIB_COMPILER_BACKOFF_IMPL_H

extern/libcds/cds/compiler/bitop.h

@@ -0,0 +1,43 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_BITOP_H
+#define CDSLIB_COMPILER_BITOP_H
+
+// Choose appropriate header for current architecture and compiler
+
+#if CDS_COMPILER == CDS_COMPILER_MSVC || (CDS_COMPILER == CDS_COMPILER_INTEL && CDS_OS_INTERFACE == CDS_OSI_WINDOWS)
+/************************************************************************/
+/* MS Visual C++                                                        */
+/************************************************************************/
+
+#    if CDS_PROCESSOR_ARCH == CDS_PROCESSOR_X86
+#        include <cds/compiler/vc/x86/bitop.h>
+#   elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_AMD64
+#        include <cds/compiler/vc/amd64/bitop.h>
+#    endif
+
+#elif CDS_COMPILER == CDS_COMPILER_GCC || CDS_COMPILER == CDS_COMPILER_CLANG || CDS_COMPILER == CDS_COMPILER_INTEL
+/************************************************************************/
+/* GCC                                                                  */
+/************************************************************************/
+
+#    if CDS_PROCESSOR_ARCH == CDS_PROCESSOR_X86
+#       include <cds/compiler/gcc/x86/bitop.h>
+#    elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_AMD64
+#       include <cds/compiler/gcc/amd64/bitop.h>
+#    elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_SPARC
+#       include <cds/compiler/gcc/sparc/bitop.h>
+#    elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_IA64
+#       include <cds/compiler/gcc/ia64/bitop.h>
+#    elif CDS_PROCESSOR_ARCH == CDS_PROCESSOR_PPC64
+#       include <cds/compiler/gcc/ppc64/bitop.h>
+#   endif
+#endif  // Compiler choice
+
+// Generic (C) implementation
+#include <cds/details/bitop_generic.h>
+
+#endif    // #ifndef CDSLIB_COMPILER_BITOP_H

extern/libcds/cds/compiler/clang/defs.h

@@ -0,0 +1,127 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_CLANG_DEFS_H
+#define CDSLIB_COMPILER_CLANG_DEFS_H
+
+// Compiler version
+#define CDS_COMPILER_VERSION (__clang_major__ * 10000 + __clang_minor__ * 100 + __clang_patchlevel__)
+
+// Compiler name
+#define  CDS_COMPILER__NAME    ("clang " __clang_version__)
+#define  CDS_COMPILER__NICK    "clang"
+
+#if CDS_COMPILER_VERSION < 30600
+#   error "Compiler version error. Clang version 3.6.0 and above is supported"
+#endif
+
+#if __cplusplus < CDS_CPLUSPLUS_11
+#   error C++11 and above is required
+#endif
+
+#if defined(_LIBCPP_VERSION) && !defined(CDS_USE_BOOST_ATOMIC) && CDS_COMPILER_VERSION < 30700
+    // Note: Clang libc++ atomic leads to program crash.
+    // So, we use libcds atomic implementation
+#   define CDS_USE_LIBCDS_ATOMIC
+#endif
+
+// clang for Windows
+#if defined( _MSC_VER )
+#   define CDS_OS_INTERFACE     CDS_OSI_WINDOWS
+#   if defined(_WIN64)
+#       define CDS_OS_TYPE      CDS_OS_WIN64
+#       define CDS_OS__NAME     "Win64"
+#       define CDS_OS__NICK     "Win64"
+#   elif defined(_WIN32)
+#       define CDS_OS_TYPE      CDS_OS_WIN32
+#       define CDS_OS__NAME     "Win32"
+#       define CDS_OS__NICK     "Win32"
+#   endif
+#endif
+
+#include <cds/compiler/gcc/compiler_macro.h>
+
+#define alignof __alignof__
+
+// C++11 thread_local keyword
+#if !(CDS_OS_TYPE == CDS_OS_OSX && CDS_COMPILER_VERSION < 30600)
+    // OS X error?
+    // See http://stackoverflow.com/questions/23791060/c-thread-local-storage-clang-503-0-40-mac-osx
+    // http://stackoverflow.com/questions/28094794/why-does-apple-clang-disallow-c11-thread-local-when-official-clang-supports
+    // clang 3.6 ok?..
+#   define CDS_CXX11_THREAD_LOCAL_SUPPORT
+#endif
+
+// Attributes
+#if CDS_COMPILER_VERSION >= 30600
+#   if __cplusplus == CDS_CPLUSPLUS_11 // C++11
+#       define CDS_DEPRECATED( reason ) [[gnu::deprecated(reason)]]
+#   else  // C++14
+#       define CDS_DEPRECATED( reason ) [[deprecated(reason)]]
+#   endif
+#endif
+
+#define CDS_NORETURN __attribute__((__noreturn__))
+
+// *************************************************
+// Features
+#if defined(__has_feature) && __has_feature(thread_sanitizer)
+#   ifndef CDS_THREAD_SANITIZER_ENABLED
+#       define CDS_THREAD_SANITIZER_ENABLED
+#   endif
+#endif
+
+#if defined(__has_feature) && __has_feature(address_sanitizer)
+#   ifndef CDS_ADDRESS_SANITIZER_ENABLED
+#       define CDS_ADDRESS_SANITIZER_ENABLED
+#   endif
+#endif
+
+
+// *************************************************
+// Alignment macro
+
+#define CDS_TYPE_ALIGNMENT(n)   __attribute__ ((aligned (n)))
+#define CDS_CLASS_ALIGNMENT(n)  __attribute__ ((aligned (n)))
+#define CDS_DATA_ALIGNMENT(n)   __attribute__ ((aligned (n)))
+
+
+// likely/unlikely
+
+#define cds_likely( expr )   __builtin_expect( !!( expr ), 1 )
+#define cds_unlikely( expr ) __builtin_expect( !!( expr ), 0 )
+
+// Exceptions
+#if defined( __EXCEPTIONS ) && __EXCEPTIONS == 1
+#   define CDS_EXCEPTION_ENABLED
+#endif
+
+
+// double-width CAS support - only for libc++
+// You can manually suppress wide-atomic support by defining in compiler command line:
+//  for 64bit platform: -DCDS_DISABLE_128BIT_ATOMIC
+//  for 32bit platform: -DCDS_DISABLE_64BIT_ATOMIC
+#ifdef _LIBCPP_VERSION
+#   if CDS_BUILD_BITS == 64
+#       if !defined( CDS_DISABLE_128BIT_ATOMIC ) && defined( __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16 )
+#           define CDS_DCAS_SUPPORT
+#       endif
+#   else
+#       if !defined( CDS_DISABLE_64BIT_ATOMIC ) && defined( __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8 )
+#           define CDS_DCAS_SUPPORT
+#       endif
+#   endif
+#endif
+
+//if constexpr support (C++17)
+#ifndef constexpr_if
+#   if defined( __cpp_if_constexpr ) && __cpp_if_constexpr >= 201606
+#       define constexpr_if if constexpr
+#   endif
+#endif
+
+#include <cds/compiler/gcc/compiler_barriers.h>
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_DEFS_H

extern/libcds/cds/compiler/defs.h

@@ -0,0 +1,51 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_DEFS_H
+#define CDSLIB_COMPILER_DEFS_H
+
+// __cplusplus values
+#define CDS_CPLUSPLUS_11     201103L
+#define CDS_CPLUSPLUS_14     201402L
+#define CDS_CPLUSPLUS_17     201703L
+
+// VC 2017 is not full C++11-compatible yet
+//#if __cplusplus < CDS_CPLUSPLUS_11
+//#   error C++11 and above is required
+//#endif
+
+
+#if CDS_COMPILER == CDS_COMPILER_MSVC
+#   include <cds/compiler/vc/defs.h>
+#elif CDS_COMPILER == CDS_COMPILER_GCC
+#   include <cds/compiler/gcc/defs.h>
+#elif CDS_COMPILER == CDS_COMPILER_INTEL
+#   include <cds/compiler/icl/defs.h>
+#elif CDS_COMPILER == CDS_COMPILER_CLANG
+#   include <cds/compiler/clang/defs.h>
+#elif CDS_COMPILER == CDS_COMPILER_UNKNOWN
+#   error Unknown compiler. Compilation aborted
+#else
+#   error Unknown value of CDS_COMPILER macro
+#endif
+
+#ifndef CDS_EXPORT_API
+#   define CDS_EXPORT_API
+#endif
+
+#ifndef cds_likely
+#   define cds_likely( expr )   expr
+#   define cds_unlikely( expr ) expr
+#endif
+
+//if constexpr support (C++17)
+#ifndef constexpr_if
+#   define constexpr_if if
+#endif
+
+// Features
+#include <cds/compiler/feature_tsan.h>
+
+#endif  // #ifndef CDSLIB_COMPILER_DEFS_H

extern/libcds/cds/compiler/feature_tsan.h

@@ -0,0 +1,91 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_FEATURE_TSAN_H
+#define CDSLIB_COMPILER_FEATURE_TSAN_H
+
+// Thread Sanitizer annotations.
+// From http://llvm.org/viewvc/llvm-project/compiler-rt/trunk/test/tsan/annotate_happens_before.cc?view=markup
+
+//@cond
+
+#ifdef CDS_THREAD_SANITIZER_ENABLED
+#   define CDS_TSAN_ANNOTATE_HAPPENS_BEFORE(addr)   AnnotateHappensBefore(__FILE__, __LINE__, reinterpret_cast<void*>(addr))
+#   define CDS_TSAN_ANNOTATE_HAPPENS_AFTER(addr)    AnnotateHappensAfter(__FILE__, __LINE__, reinterpret_cast<void*>(addr))
+
+#   define CDS_TSAN_ANNOTATE_IGNORE_READS_BEGIN     AnnotateIgnoreReadsBegin(__FILE__, __LINE__)
+#   define CDS_TSAN_ANNOTATE_IGNORE_READS_END       AnnotateIgnoreReadsEnd(__FILE__, __LINE__)
+#   define CDS_TSAN_ANNOTATE_IGNORE_WRITES_BEGIN    AnnotateIgnoreWritesBegin(__FILE__, __LINE__)
+#   define CDS_TSAN_ANNOTATE_IGNORE_WRITES_END      AnnotateIgnoreWritesEnd(__FILE__, __LINE__)
+#   define CDS_TSAN_ANNOTATE_IGNORE_RW_BEGIN        \
+                                                    CDS_TSAN_ANNOTATE_IGNORE_READS_BEGIN; \
+                                                    CDS_TSAN_ANNOTATE_IGNORE_WRITES_BEGIN
+#   define CDS_TSAN_ANNOTATE_IGNORE_RW_END          \
+                                                    CDS_TSAN_ANNOTATE_IGNORE_WRITES_END;\
+                                                    CDS_TSAN_ANNOTATE_IGNORE_READS_END
+#   define CDS_TSAN_ANNOTATE_NEW_MEMORY( addr, sz ) AnnotateNewMemory( __FILE__, __LINE__, reinterpret_cast<void *>(addr), sz )
+
+// Publish/unpublish - DEPRECATED
+#if 0
+#   define CDS_TSAN_ANNOTATE_PUBLISH_MEMORY_RANGE( addr, sz ) AnnotatePublishMemoryRange( __FILE__, __LINE__, reinterpret_cast<void *>(addr), sz )
+#   define CDS_TSAN_ANNOTATE_UNPUBLISH_MEMORY_RANGE( addr, sz ) AnnotateUnpublishMemoryRange( __FILE__, __LINE__, reinterpret_cast<void *>(addr), sz )
+#endif
+
+#   define CDS_TSAN_ANNOTATE_MUTEX_CREATE( addr )    AnnotateRWLockCreate( __FILE__, __LINE__, reinterpret_cast<void *>(addr))
+#   define CDS_TSAN_ANNOTATE_MUTEX_DESTROY( addr )   AnnotateRWLockDestroy( __FILE__, __LINE__, reinterpret_cast<void *>(addr))
+        // must be called after actual acquire
+#   define CDS_TSAN_ANNOTATE_MUTEX_ACQUIRED( addr )  AnnotateRWLockAcquired( __FILE__, __LINE__, reinterpret_cast<void *>(addr), 1 )
+        // must be called before actual release
+#   define CDS_TSAN_ANNOTATE_MUTEX_RELEASED( addr )  AnnotateRWLockReleased( __FILE__, __LINE__, reinterpret_cast<void *>(addr), 1 )
+
+    // provided by TSan
+    extern "C" {
+        void AnnotateHappensBefore(const char *f, int l, void *addr);
+        void AnnotateHappensAfter(const char *f, int l, void *addr);
+
+        void AnnotateIgnoreReadsBegin(const char *f, int l);
+        void AnnotateIgnoreReadsEnd(const char *f, int l);
+        void AnnotateIgnoreWritesBegin(const char *f, int l);
+        void AnnotateIgnoreWritesEnd(const char *f, int l);
+
+#if 0
+        void AnnotatePublishMemoryRange( const char *f, int l, void * mem, size_t size );
+        void AnnotateUnpublishMemoryRange( const char *f, int l, void * addr, size_t size );
+#endif
+        void AnnotateNewMemory( const char *f, int l, void * mem, size_t size );
+
+        void AnnotateRWLockCreate( const char *f, int l, void* m );
+        void AnnotateRWLockDestroy( const char *f, int l, void* m );
+        void AnnotateRWLockAcquired( const char *f, int l, void *m, long is_w );
+        void AnnotateRWLockReleased( const char *f, int l, void *m, long is_w );
+    }
+
+#else // CDS_THREAD_SANITIZER_ENABLED
+
+#   define CDS_TSAN_ANNOTATE_HAPPENS_BEFORE(addr)
+#   define CDS_TSAN_ANNOTATE_HAPPENS_AFTER(addr)
+
+#   define CDS_TSAN_ANNOTATE_IGNORE_READS_BEGIN
+#   define CDS_TSAN_ANNOTATE_IGNORE_READS_END
+#   define CDS_TSAN_ANNOTATE_IGNORE_WRITES_BEGIN
+#   define CDS_TSAN_ANNOTATE_IGNORE_WRITES_END
+#   define CDS_TSAN_ANNOTATE_IGNORE_RW_BEGIN
+#   define CDS_TSAN_ANNOTATE_IGNORE_RW_END
+
+#if 0
+#   define CDS_TSAN_ANNOTATE_PUBLISH_MEMORY_RANGE( addr, sz )
+#   define CDS_TSAN_ANNOTATE_UNPUBLISH_MEMORY_RANGE( addr, sz )
+#endif
+#   define CDS_TSAN_ANNOTATE_NEW_MEMORY( addr, sz )
+
+#   define CDS_TSAN_ANNOTATE_MUTEX_CREATE( addr )
+#   define CDS_TSAN_ANNOTATE_MUTEX_DESTROY( addr )
+#   define CDS_TSAN_ANNOTATE_MUTEX_ACQUIRED( addr )
+#   define CDS_TSAN_ANNOTATE_MUTEX_RELEASED( addr )
+
+#endif
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_FEATURE_TSAN_H

extern/libcds/cds/compiler/gcc/amd64/backoff.h

@@ -0,0 +1,35 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_AMD64_BACKOFF_H
+#define CDSLIB_COMPILER_GCC_AMD64_BACKOFF_H
+
+//@cond none
+
+namespace cds { namespace backoff {
+    namespace gcc { namespace amd64 {
+
+#       define CDS_backoff_nop_defined
+        static inline void backoff_nop()
+        {
+            asm volatile ( "nop;" );
+        }
+
+#       define CDS_backoff_hint_defined
+        static inline void backoff_hint()
+        {
+            asm volatile ( "pause;" );
+        }
+
+
+    }} // namespace gcc::amd64
+
+    namespace platform {
+        using namespace gcc::amd64;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_GCC_AMD64_BACKOFF_H

extern/libcds/cds/compiler/gcc/amd64/bitop.h

@@ -0,0 +1,159 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_AMD64_BITOP_H
+#define CDSLIB_COMPILER_GCC_AMD64_BITOP_H
+
+//@cond none
+namespace cds {
+    namespace bitop { namespace platform { namespace gcc { namespace amd64 {
+        // MSB - return index (1..32) of most significant bit in nArg. If nArg == 0 return 0
+#        define cds_bitop_msb32_DEFINED
+        static inline int msb32( uint32_t nArg )
+        {
+            int        nRet;
+            __asm__ __volatile__ (
+                "bsrl        %[nArg], %[nRet]     ;\n\t"
+                "jnz        1f                    ;\n\t"
+                "xorl        %[nRet], %[nRet]    ;\n\t"
+                "subl        $1, %[nRet]         ;\n\t"
+                "1:"
+                "addl        $1, %[nRet]         ;\n\t"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return nRet;
+        }
+
+#        define cds_bitop_msb32nz_DEFINED
+        static inline int msb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            int        nRet;
+            __asm__ __volatile__ (
+                "bsrl        %[nArg], %[nRet]    ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return nRet;
+        }
+
+        // LSB - return index (0..31) of least significant bit in nArg. If nArg == 0 return -1U
+#        define cds_bitop_lsb32_DEFINED
+        static inline int lsb32( uint32_t nArg )
+        {
+
+            int        nRet;
+            __asm__ __volatile__ (
+                "bsfl        %[nArg], %[nRet]     ;"
+                "jnz        1f        ;"
+                "xorl        %[nRet], %[nRet]    ;"
+                "subl        $1, %[nRet]         ;"
+                "1:"
+                "addl        $1, %[nRet]         ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return nRet;
+
+        }
+
+        // LSB - return index (0..31) of least significant bit in nArg.
+        // Condition: nArg != 0
+#        define cds_bitop_lsb32nz_DEFINED
+        static inline int lsb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            int        nRet;
+            __asm__ __volatile__ (
+                "bsfl        %[nArg], %[nRet]    ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return nRet;
+        }
+
+#        define cds_bitop_msb64_DEFINED
+        static inline int msb64( uint64_t nArg )
+        {
+            uint64_t  nRet;
+            asm volatile (
+                "bsrq        %[nArg], %[nRet]     ;\n\t"
+                "jnz        1f                    ;\n\t"
+                "xorq        %[nRet], %[nRet]    ;\n\t"
+                "subq        $1, %[nRet]         ;\n\t"
+                "1:"
+                "addq        $1, %[nRet]         ;\n\t"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return (int) nRet;
+        }
+
+#        define cds_bitop_msb64nz_DEFINED
+        static inline int msb64nz( uint64_t nArg )
+        {
+            assert( nArg != 0 );
+            uint64_t nRet;
+            __asm__ __volatile__ (
+                "bsrq        %[nArg], %[nRet]    ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return (int) nRet;
+        }
+
+        // LSB - return index (0..31) of least significant bit in nArg. If nArg == 0 return -1U
+#        define cds_bitop_lsb64_DEFINED
+        static inline int lsb64( uint64_t nArg )
+        {
+            uint64_t nRet;
+            __asm__ __volatile__ (
+                "bsfq        %[nArg], %[nRet]     ;"
+                "jnz        1f        ;"
+                "xorq        %[nRet], %[nRet]    ;"
+                "subq        $1, %[nRet]         ;"
+                "1:"
+                "addq        $1, %[nRet]         ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return (int) nRet;
+
+        }
+
+        // LSB - return index (0..31) of least significant bit in nArg.
+        // Condition: nArg != 0
+#        define cds_bitop_lsb64nz_DEFINED
+        static inline int lsb64nz( uint64_t nArg )
+        {
+            assert( nArg != 0 );
+            uint64_t nRet;
+            __asm__ __volatile__ (
+                "bsfq        %[nArg], %[nRet]    ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return (int) nRet;
+        }
+
+
+    }} // namespace gcc::amd64
+
+    using namespace gcc::amd64;
+
+    }}}    // namespace cds::bitop::platform
+
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_AMD64_BITOP_H

extern/libcds/cds/compiler/gcc/amd64/cxx11_atomic.h

@@ -0,0 +1,203 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_AMD64_CXX11_ATOMIC_H
+#define CDSLIB_COMPILER_GCC_AMD64_CXX11_ATOMIC_H
+
+#include <cstdint>
+#include <cds/compiler/gcc/x86/cxx11_atomic32.h>
+
+//@cond
+namespace cds { namespace cxx11_atomic {
+    namespace platform { inline namespace gcc { inline namespace amd64 {
+
+        //-----------------------------------------------------------------------------
+        // 64bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline bool cas64_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            T prev = expected;
+            fence_before(mo_success);
+            __asm__ __volatile__ (
+                "lock ; cmpxchgq %[desired], %[pDest]"
+                : [prev] "+a" (prev), [pDest] "+m" (*pDest)
+                : [desired] "r" (desired)
+                );
+            bool success = (prev == expected);
+            expected = prev;
+            if (success)
+                fence_after(mo_success);
+            else
+                fence_after(mo_fail);
+            return success;
+        }
+
+        template <typename T>
+        static inline bool cas64_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas64_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T load64( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+            assert( cds::details::is_aligned( pSrc, 8 ));
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+
+        template <typename T>
+        static inline T exchange64( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            fence_before(order);
+            __asm__ __volatile__ (
+                "xchgq %[v], %[pDest]"
+                : [v] "+r" (v), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return v;
+        }
+
+        template <typename T>
+        static inline void store64( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            if (order != memory_order_seq_cst) {
+                fence_before(order);
+                *pDest = val;
+            }
+            else {
+                exchange64( pDest, val, order);
+            }
+        }
+
+#       define CDS_ATOMIC_fetch64_add_defined
+        template <typename T>
+        static inline T fetch64_add( T volatile * pDest, T v, memory_order order) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            fence_before(order);
+            __asm__ __volatile__ (
+                "lock ; xaddq %[v], %[pDest]"
+                : [v] "+r" (v), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return v;
+        }
+
+#       define CDS_ATOMIC_fetch64_sub_defined
+        template <typename T>
+        static inline T fetch64_sub( T volatile * pDest, T v, memory_order order) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            fence_before(order);
+            __asm__ __volatile__ (
+                "negq   %[v] ; \n"
+                "lock ; xaddq %[v], %[pDest]"
+                : [v] "+r" (v), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return v;
+        }
+
+
+        //-----------------------------------------------------------------------------
+        // pointer primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T * exchange_ptr( T * volatile * pDest, T * v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+
+            return (T *) exchange64( (uint64_t volatile *) pDest, (uint64_t) v, order );
+        }
+
+        template <typename T>
+        static inline void store_ptr( T * volatile * pDest, T * src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange_ptr( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T * load_ptr( T * volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            T * v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_strong( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+
+            return cas64_strong( (uint64_t volatile *) pDest, *reinterpret_cast<uint64_t *>( &expected ), (uint64_t) desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_weak( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas_ptr_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+    }} // namespace gcc::amd64
+
+    }   // namespace platform
+
+}}  // namespace cds::cxx11_atomic
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_AMD64_CXX11_ATOMIC_H

extern/libcds/cds/compiler/gcc/arm7/backoff.h

@@ -0,0 +1,27 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_ARM7_BACKOFF_H
+#define CDSLIB_COMPILER_GCC_ARM7_BACKOFF_H
+
+//@cond none
+
+namespace cds { namespace backoff {
+    namespace gcc { namespace arm7 {
+
+#       define CDS_backoff_hint_defined
+        static inline void backoff_hint()
+        {
+            asm volatile( "yield" ::: "memory" );
+        }
+    }} // namespace gcc::arm7
+
+    namespace platform {
+        using namespace gcc::arm7;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_GCC_ARM7_BACKOFF_H

extern/libcds/cds/compiler/gcc/arm8/backoff.h

@@ -0,0 +1,27 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_ARM8_BACKOFF_H
+#define CDSLIB_COMPILER_GCC_ARM8_BACKOFF_H
+
+//@cond none
+
+namespace cds { namespace backoff {
+    namespace gcc { namespace arm8 {
+
+#       define CDS_backoff_hint_defined
+        static inline void backoff_hint()
+        {
+            asm volatile( "yield" ::: "memory" );
+        }
+    }} // namespace gcc::arm8
+
+    namespace platform {
+        using namespace gcc::arm8;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_GCC_ARM8_BACKOFF_H

extern/libcds/cds/compiler/gcc/compiler_barriers.h

@@ -0,0 +1,13 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_COMPILER_BARRIERS_H
+#define CDSLIB_COMPILER_GCC_COMPILER_BARRIERS_H
+
+#define CDS_COMPILER_RW_BARRIER  __asm__ __volatile__ ( "" ::: "memory" )
+#define CDS_COMPILER_R_BARRIER   CDS_COMPILER_RW_BARRIER
+#define CDS_COMPILER_W_BARRIER   CDS_COMPILER_RW_BARRIER
+
+#endif  // #ifndef CDSLIB_COMPILER_GCC_COMPILER_BARRIERS_H

extern/libcds/cds/compiler/gcc/compiler_macro.h

@@ -0,0 +1,170 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_COMPILER_MACRO_H
+#define CDSLIB_COMPILER_GCC_COMPILER_MACRO_H
+
+// OS interface && OS name
+#ifndef CDS_OS_TYPE
+#   if defined( __linux__ )
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_LINUX
+#       define CDS_OS__NAME         "linux"
+#       define CDS_OS__NICK         "linux"
+#   elif defined( __sun__ )
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_SUN_SOLARIS
+#       define CDS_OS__NAME         "Sun Solaris"
+#       define CDS_OS__NICK         "sun"
+#   elif defined( __hpux__ )
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_HPUX
+#       define CDS_OS__NAME         "HP-UX"
+#       define CDS_OS__NICK         "hpux"
+#   elif defined( _AIX )
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_AIX
+#       define CDS_OS__NAME         "AIX"
+#       define CDS_OS__NICK         "aix"
+#   elif defined( __FreeBSD__ )
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_FREE_BSD
+#       define CDS_OS__NAME         "FreeBSD"
+#       define CDS_OS__NICK         "freebsd"
+#   elif defined( __OpenBSD__ )
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_OPEN_BSD
+#       define CDS_OS__NAME         "OpenBSD"
+#       define CDS_OS__NICK         "openbsd"
+#   elif defined( __NetBSD__ )
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_NET_BSD
+#       define CDS_OS__NAME         "NetBSD"
+#       define CDS_OS__NICK         "netbsd"
+#   elif defined(__MINGW32__) || defined( __MINGW64__)
+#       define CDS_OS_INTERFACE     CDS_OSI_WINDOWS
+#       define CDS_OS_TYPE          CDS_OS_MINGW
+#       define CDS_OS__NAME         "MinGW"
+#       define CDS_OS__NICK         "mingw"
+#   elif defined(__MACH__)
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_OSX
+#       define CDS_OS__NAME         "OS X"
+#       define CDS_OS__NICK         "osx"
+#   else
+#       define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#       define CDS_OS_TYPE          CDS_OS_PTHREAD
+#       define CDS_OS__NAME         "pthread"
+#       define CDS_OS__NICK         "pthread"
+#   endif
+#endif // #ifndef CDS_OS_TYPE
+
+// Processor architecture
+
+#if defined(__arm__) && !defined(__ARM_ARCH)
+// GCC 4.6 does not defined __ARM_ARCH
+#   if defined(__ARM_ARCH_8A__) || defined(__ARM_ARCH_8S__) || defined(__aarch64__) || defined(__ARM_ARCH_ISA_A64)
+#       define __ARM_ARCH   8
+#   elif defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7S__)
+#       define __ARM_ARCH   7
+#   else
+#       define __ARM_ARCH   5
+#   endif
+#endif
+
+#if defined(__x86_64__) || defined(__amd64__) || defined(__amd64)
+#   define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_AMD64
+#   define CDS_BUILD_BITS        64
+#   define CDS_PROCESSOR__NAME   "Intel x86-64"
+#   define CDS_PROCESSOR__NICK   "amd64"
+#elif defined(__i386__)
+#   define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_X86
+#   define CDS_BUILD_BITS        32
+#   define CDS_PROCESSOR__NAME   "Intel x86"
+#   define CDS_PROCESSOR__NICK   "x86"
+#elif defined(sparc) || defined (__sparc__)
+#   define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_SPARC
+#   define CDS_PROCESSOR__NAME   "Sparc"
+#   define CDS_PROCESSOR__NICK   "sparc"
+#   ifdef __arch64__
+#       define CDS_BUILD_BITS   64
+#    else
+#       error Sparc 32bit is not supported
+#    endif
+#elif defined( __ia64__)
+#    define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_IA64
+#    define CDS_BUILD_BITS        64
+#    define CDS_PROCESSOR__NAME   "Intel IA64"
+#    define CDS_PROCESSOR__NICK   "ia64"
+#elif defined(_ARCH_PPC64)
+#    define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_PPC64
+#    define CDS_BUILD_BITS        64
+#    define CDS_PROCESSOR__NAME   "IBM PowerPC64"
+#    define CDS_PROCESSOR__NICK   "ppc64"
+#elif defined(__arm__) && __SIZEOF_POINTER__ == 4 && __ARM_ARCH >= 7 && __ARM_ARCH < 8
+#    define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_ARM7
+#    define CDS_BUILD_BITS        32
+#    define CDS_PROCESSOR__NAME   "ARM v7"
+#    define CDS_PROCESSOR__NICK   "arm7"
+#elif ( defined(__arm__)  || defined(__aarch64__)) && __ARM_ARCH >= 8
+#    define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_ARM8
+#    define CDS_BUILD_BITS        64
+#    define CDS_PROCESSOR__NAME   "ARM v8"
+#    define CDS_PROCESSOR__NICK   "arm8"
+#elif defined(__arm__)  || defined(__aarch64__)
+#   define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_ARM8
+#   define CDS_PROCESSOR__NAME   "ARM"
+#   define CDS_PROCESSOR__NICK   "arm"
+#   if __SIZEOF_POINTER__ == 8
+#       define CDS_BUILD_BITS        64
+#   else
+#       define CDS_BUILD_BITS        32
+#   endif
+#else
+#   if defined(CDS_USE_LIBCDS_ATOMIC)
+#       error "Libcds does not support atomic implementation for the processor architecture. Try to use C++11-compatible compiler and remove CDS_USE_LIBCDS_ATOMIC flag from compiler command line"
+#   else
+#       define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_UNKNOWN
+#       define CDS_BUILD_BITS        32
+#       define CDS_PROCESSOR__NAME   "unknown"
+#       define CDS_PROCESSOR__NICK   "unknown"
+#   endif
+#endif
+
+#if CDS_OS_TYPE == CDS_OS_MINGW
+#   ifdef CDS_BUILD_LIB
+#       define CDS_EXPORT_API          __declspec(dllexport)
+#   elif !defined(CDS_BUILD_STATIC_LIB)
+#       define CDS_EXPORT_API          __declspec(dllimport)
+#   endif
+#else
+#   ifndef __declspec
+#       define __declspec(_x)
+#   endif
+#endif
+
+// Byte order
+#if !defined(CDS_ARCH_LITTLE_ENDIAN) && !defined(CDS_ARCH_BIG_ENDIAN)
+#   ifdef __BYTE_ORDER__
+#       if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+#           define CDS_ARCH_LITTLE_ENDIAN
+#       elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+#           define CDS_ARCH_BIG_ENDIAN
+#       endif
+#   else
+#       warning "Undefined byte order for current architecture (no __BYTE_ORDER__ preprocessor definition)"
+#   endif
+#endif
+
+// Sanitizer attributes
+// Example: CDS_DISABLE_SANITIZE( "function" )
+#ifdef CDS_ADDRESS_SANITIZER_ENABLED
+#   define CDS_SUPPRESS_SANITIZE( ... ) __attribute__(( no_sanitize( __VA_ARGS__ )))
+#else
+#   define CDS_SUPPRESS_SANITIZE( ... )
+#endif
+
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_COMPILER_MACRO_H

extern/libcds/cds/compiler/gcc/defs.h

@@ -0,0 +1,107 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_DEFS_H
+#define CDSLIB_COMPILER_GCC_DEFS_H
+
+// Compiler version
+#define CDS_COMPILER_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__)
+
+#if CDS_COMPILER_VERSION < 40800
+#   error "Compiler version error. GCC version 4.8.0 and above is supported"
+#endif
+
+// Compiler name
+#ifdef __VERSION__
+#   define  CDS_COMPILER__NAME    ("GNU C++ " __VERSION__)
+#else
+#   define  CDS_COMPILER__NAME    "GNU C++"
+#endif
+#define  CDS_COMPILER__NICK        "gcc"
+
+#if __cplusplus < CDS_CPLUSPLUS_11
+#   error C++11 and above is required
+#endif
+
+
+#include <cds/compiler/gcc/compiler_macro.h>
+
+#define alignof __alignof__
+
+// ***************************************
+// C++11 features
+
+// C++11 thread_local keyword
+#define CDS_CXX11_THREAD_LOCAL_SUPPORT
+
+// *************************************************
+// Features
+// If you run under Thread Sanitizer, pass -DCDS_THREAD_SANITIZER_ENABLED in compiler command line
+// UPD: Seems, GCC 5+ has predefined macro __SANITIZE_THREAD__, see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64354
+#if defined(__SANITIZE_THREAD__) && !defined(CDS_THREAD_SANITIZER_ENABLED)
+#   define CDS_THREAD_SANITIZER_ENABLED
+#endif
+
+// *************************************************
+// Alignment macro
+
+#define CDS_TYPE_ALIGNMENT(n)   __attribute__ ((aligned (n)))
+#define CDS_CLASS_ALIGNMENT(n)  __attribute__ ((aligned (n)))
+#define CDS_DATA_ALIGNMENT(n)   __attribute__ ((aligned (n)))
+
+// Attributes
+#if CDS_COMPILER_VERSION >= 40900
+#   if __cplusplus == CDS_CPLUSPLUS_11 // C++11
+#       define CDS_DEPRECATED( reason ) [[gnu::deprecated(reason)]]
+#   else  // C++14
+#       define CDS_DEPRECATED( reason ) [[deprecated(reason)]]
+#   endif
+#else
+    // GCC 4.8
+#   define CDS_DEPRECATED( reason ) __attribute__((deprecated( reason )))
+#endif
+
+#define CDS_NORETURN __attribute__((__noreturn__))
+
+// likely/unlikely
+
+#define cds_likely( expr )   __builtin_expect( !!( expr ), 1 )
+#define cds_unlikely( expr ) __builtin_expect( !!( expr ), 0 )
+
+// Exceptions
+#if defined( __EXCEPTIONS ) && __EXCEPTIONS == 1
+#   define CDS_EXCEPTION_ENABLED
+#endif
+
+// double-width CAS support
+// note: gcc-4.8 does not support double-word atomics
+//       gcc-4.9: a lot of crashes when use DCAS
+//       gcc-7: 128-bit atomic is not lock-free, see https://gcc.gnu.org/ml/gcc/2017-01/msg00167.html
+// You can manually suppress wide-atomic support by defining in compiler command line:
+//  for 64bit platform: -DCDS_DISABLE_128BIT_ATOMIC
+//  for 32bit platform: -DCDS_DISABLE_64BIT_ATOMIC
+#if CDS_COMPILER_VERSION >= 50000
+#   if CDS_BUILD_BITS == 64
+#       if !defined( CDS_DISABLE_128BIT_ATOMIC ) && defined( __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16 ) && CDS_COMPILER_VERSION < 70000
+#           define CDS_DCAS_SUPPORT
+#       endif
+#   else
+#       if !defined( CDS_DISABLE_64BIT_ATOMIC ) && defined( __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8 )
+#           define CDS_DCAS_SUPPORT
+#       endif
+#   endif
+#endif
+
+//if constexpr support (C++17)
+#ifndef constexpr_if
+#   if defined( __cpp_if_constexpr ) && __cpp_if_constexpr >= 201606
+#       define constexpr_if if constexpr
+#   endif
+#endif
+
+
+#include <cds/compiler/gcc/compiler_barriers.h>
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_DEFS_H

extern/libcds/cds/compiler/gcc/ia64/backoff.h

@@ -0,0 +1,34 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_IA64_BACKOFF_H
+#define CDSLIB_COMPILER_GCC_IA64_BACKOFF_H
+
+//@cond none
+
+namespace cds { namespace backoff {
+    namespace gcc { namespace ia64 {
+
+#       define CDS_backoff_hint_defined
+        static inline void backoff_hint()
+        {
+            asm volatile ( "hint @pause;;" );
+        }
+
+#       define CDS_backoff_nop_defined
+        static inline void backoff_nop()
+        {
+            asm volatile ( "nop;;" );
+        }
+
+    }} // namespace gcc::ia64
+
+    namespace platform {
+        using namespace gcc::ia64;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_GCC_IA64_BACKOFF_H

extern/libcds/cds/compiler/gcc/ia64/bitop.h

@@ -0,0 +1,65 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_IA64_BITOP_H
+#define CDSLIB_COMPILER_GCC_IA64_BITOP_H
+
+//@cond none
+namespace cds {
+    namespace bitop { namespace platform { namespace gcc { namespace ia64 {
+
+        // MSB - return index (1..32) of most significant bit in x. If x == 0 return 0
+#        define cds_bitop_msb32_DEFINED
+        static inline int msb32( uint32_t nArg )
+        {
+            if ( !nArg )
+                return 0;
+            uint64_t x = nArg;
+            x |= x >> 1;
+            x |= x >> 2;
+            x |= x >> 4;
+            x |= x >> 8;
+            x |= x >> 16;
+
+            uint64_t    nRes;
+            asm __volatile__( "popcnt %0=%1\n\t" : "=r" (nRes) : "r" (x));
+            return (int) nRes;
+        }
+
+        // It is not compiled on HP-UX. Why?..
+#if CDS_OS_TYPE != CDS_OS_HPUX
+        // MSB - return index (0..31) of most significant bit in nArg.
+        // !!! nArg != 0
+#        define cds_bitop_msb32nz_DEFINED
+        static inline int msb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            long double d = nArg;
+            long nExp;
+            asm __volatile__("getf.exp %0=%1\n\t" : "=r"(nExp) : "f"(d));
+            return (int) (nExp - 0xffff);
+        }
+
+        // MSB - return index (0..63) of most significant bit in nArg.
+        // !!! nArg != 0
+#        define cds_bitop_msb64nz_DEFINED
+        static inline int msb64nz( uint64_t nArg )
+        {
+            assert( nArg != 0 );
+            long double d = nArg;
+            long nExp;
+            asm __volatile__("getf.exp %0=%1\n\t" : "=r" (nExp) : "f" (d));
+            return (int) (nExp - 0xffff);
+        }
+#endif    // #if CDS_OS_TYPE != CDS_OS_HPUX
+
+    }} // namespace gcc::ia64
+
+    using namespace gcc::ia64;
+
+}}}    // namespace cds::bitop::platform
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_IA64_BITOP_H

extern/libcds/cds/compiler/gcc/ia64/cxx11_atomic.h

@@ -0,0 +1,653 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_IA64_CXX11_ATOMIC_H
+#define CDSLIB_COMPILER_GCC_IA64_CXX11_ATOMIC_H
+
+/*
+    Source:
+        1. load/store: http://www.decadent.org.uk/pipermail/cpp-threads/2008-December/001932.html
+        2. Mapping to C++ Memory Model: http://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
+*/
+
+#include <cstdint>
+
+//@cond
+namespace cds { namespace cxx11_atomic {
+    namespace platform { inline namespace gcc { inline namespace ia64 {
+
+        static inline void itanium_full_fence() noexcept
+        {
+            __asm__ __volatile__ ( "mf \n\t" ::: "memory" );
+        }
+
+        static inline void fence_before( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_consume:
+            case memory_order_acquire:
+                break;
+            case memory_order_release:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_seq_cst:
+                itanium_full_fence();
+                break;
+            }
+        }
+
+        static inline void fence_after( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_acquire:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_relaxed:
+            case memory_order_consume:
+            case memory_order_release:
+                break;
+            case memory_order_seq_cst:
+                itanium_full_fence();
+                break;
+            }
+        }
+
+
+        //-----------------------------------------------------------------------------
+        // fences
+        //-----------------------------------------------------------------------------
+        static inline void thread_fence(memory_order order) noexcept
+        {
+            switch(order)
+            {
+                case memory_order_relaxed:
+                case memory_order_consume:
+                    break;
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                case memory_order_seq_cst:
+                    itanium_full_fence();
+                    break;
+                default:;
+            }
+        }
+
+        static inline void signal_fence(memory_order order) noexcept
+        {
+            // C++11: 29.8.8: only compiler optimization, no hardware instructions
+            switch(order)
+            {
+                case memory_order_relaxed:
+                    break;
+                case memory_order_consume:
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                case memory_order_seq_cst:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                default:;
+            }
+        }
+
+#define CDS_ITANIUM_ATOMIC_LOAD( n_bytes, n_bits )   \
+        template <typename T>   \
+        static inline T load##n_bits( T volatile const * pSrc, memory_order order ) noexcept \
+        { \
+            static_assert( sizeof(T) == n_bytes, "Illegal size of operand" )   ; \
+            assert( order ==  memory_order_relaxed \
+                || order ==  memory_order_consume  \
+                || order ==  memory_order_acquire  \
+                || order == memory_order_seq_cst   \
+                ) ; \
+            assert( pSrc )  ; \
+            T val    ; \
+            __asm__ __volatile__ ( \
+                "ld" #n_bytes ".acq %[val] = [%[pSrc]]  \n\t" \
+                : [val] "=r" (val) \
+                : [pSrc] "r" (pSrc) \
+                : "memory" \
+                ) ; \
+            return val ; \
+        }
+
+#define CDS_ITANIUM_ATOMIC_STORE( n_bytes, n_bits ) \
+        template <typename T> \
+        static inline void store##n_bits( T volatile * pDest, T val, memory_order order ) noexcept \
+        { \
+            static_assert( sizeof(T) == n_bytes, "Illegal size of operand" )   ; \
+            assert( order ==  memory_order_relaxed \
+                || order ==  memory_order_release  \
+                || order == memory_order_seq_cst   \
+                ) ; \
+            assert( pDest )  ; \
+            if ( order == memory_order_seq_cst ) { \
+                __asm__ __volatile__ ( \
+                    "st" #n_bytes ".rel [%[pDest]] = %[val] \n\t" \
+                    "mf     \n\t" \
+                    :: [pDest] "r" (pDest), [val] "r" (val) \
+                    : "memory" \
+                    ) ; \
+            } \
+            else { \
+                __asm__ __volatile__ ( \
+                    "st" #n_bytes ".rel [%[pDest]] = %[val] \n\t" \
+                    :: [pDest] "r" (pDest), [val] "r" (val) \
+                    : "memory" \
+                    ) ; \
+                fence_after(order) ; \
+            } \
+        }
+
+#define CDS_ITANIUM_ATOMIC_CAS( n_bytes, n_bits ) \
+        template <typename T> \
+        static inline bool cas##n_bits##_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order /*mo_fail*/ ) noexcept \
+        { \
+            static_assert( sizeof(T) == n_bytes, "Illegal size of operand" )   ; \
+            T current ; \
+            switch(mo_success) { \
+            case memory_order_relaxed: \
+            case memory_order_consume: \
+            case memory_order_acquire: \
+                __asm__ __volatile__ ( \
+                    "mov ar.ccv = %[expected] ;;\n\t" \
+                    "cmpxchg" #n_bytes ".acq %[current] = [%[pDest]], %[desired], ar.ccv\n\t" \
+                    : [current] "=r" (current) \
+                    : [pDest] "r" (pDest), [expected] "r" (expected), [desired] "r" (desired) \
+                    : "ar.ccv", "memory" \
+                    ); \
+                break ; \
+            case memory_order_release: \
+                __asm__ __volatile__ ( \
+                    "mov ar.ccv = %[expected] ;;\n\t" \
+                    "cmpxchg" #n_bytes ".rel %[current] = [%[pDest]], %[desired], ar.ccv\n\t" \
+                    : [current] "=r" (current) \
+                    : [pDest] "r" (pDest), [expected] "r" (expected), [desired] "r" (desired) \
+                    : "ar.ccv", "memory" \
+                    ); \
+                break ; \
+            case memory_order_acq_rel: \
+            case memory_order_seq_cst: \
+                __asm__ __volatile__ ( \
+                    "mov ar.ccv = %[expected] ;;\n\t" \
+                    "cmpxchg" #n_bytes ".rel %[current] = [%[pDest]], %[desired], ar.ccv\n\t" \
+                    "mf \n\t" \
+                    : [current] "=r" (current) \
+                    : [pDest] "r" (pDest), [expected] "r" (expected), [desired] "r" (desired) \
+                    : "ar.ccv", "memory" \
+                    ); \
+                break; \
+            default: \
+                assert(false); \
+            } \
+            bool bSuccess = expected == current ; \
+            expected = current ; \
+            return bSuccess ; \
+        } \
+        template <typename T> \
+        static inline bool cas##n_bits##_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept \
+        { return cas##n_bits##_strong( pDest, expected, desired, mo_success, mo_fail ); }
+
+        // xchg is performed with acquire semantics
+#define CDS_ITANIUM_ATOMIC_EXCHANGE( n_bytes, n_bits ) \
+        template <typename T> \
+        static inline T exchange##n_bits( T volatile * pDest, T val, memory_order order ) noexcept \
+        { \
+            static_assert( sizeof(T) == n_bytes, "Illegal size of operand" )   ; \
+            assert( pDest ) ; \
+            T current ; \
+            switch(order) \
+            { \
+            case memory_order_relaxed: \
+            case memory_order_consume: \
+            case memory_order_acquire: \
+                __asm__ __volatile__ ( \
+                    "xchg" #n_bytes " %[current] = [%[pDest]], %[val]\n\t" \
+                    : [current] "=r" (current) \
+                    : [pDest] "r" (pDest), [val] "r" (val) \
+                    : "memory" \
+                    ); \
+                break; \
+            case memory_order_acq_rel: \
+            case memory_order_release: \
+            case memory_order_seq_cst: \
+                __asm__ __volatile__ ( \
+                    "mf \n\t" \
+                    "xchg" #n_bytes " %[current] = [%[pDest]], %[val]\n\t" \
+                    : [current] "=r" (current) \
+                    : [pDest] "r" (pDest), [val] "r" (val) \
+                    : "memory" \
+                    ); \
+                break; \
+            default: assert(false); \
+            } \
+            return current ; \
+        }
+
+#define CDS_ITANIUM_ATOMIC_FETCH_ADD( n_bytes, n_add )  \
+        switch (order) { \
+            case memory_order_relaxed: \
+            case memory_order_consume: \
+            case memory_order_acquire: \
+                __asm__ __volatile__ ( \
+                    "fetchadd" #n_bytes ".acq %[cur] = [%[pDest]], " #n_add " \n\t" \
+                    : [cur] "=r" (cur) \
+                    : [pDest] "r" (pDest) \
+                    : "memory" \
+                    ); \
+                break ; \
+            case memory_order_release: \
+                __asm__ __volatile__ ( \
+                    "fetchadd" #n_bytes ".rel %[cur] = [%[pDest]], " #n_add " \n\t" \
+                    : [cur] "=r" (cur) \
+                    : [pDest] "r" (pDest) \
+                    : "memory" \
+                    ); \
+                break ; \
+            case memory_order_acq_rel: \
+            case memory_order_seq_cst: \
+                __asm__ __volatile__ ( \
+                    "fetchadd" #n_bytes ".rel %[cur] = [%[pDest]], " #n_add " \n\t" \
+                    "mf \n\t" \
+                    : [cur] "=r" (cur) \
+                    : [pDest] "r" (pDest) \
+                    : "memory" \
+                    ); \
+                break ; \
+            default: \
+                assert(false); \
+        }
+
+        //-----------------------------------------------------------------------------
+        // 8bit primitives
+        //-----------------------------------------------------------------------------
+
+        CDS_ITANIUM_ATOMIC_LOAD( 1, 8 )
+        CDS_ITANIUM_ATOMIC_STORE( 1, 8 )
+        CDS_ITANIUM_ATOMIC_CAS( 1, 8 )
+        CDS_ITANIUM_ATOMIC_EXCHANGE( 1, 8 )
+
+        //-----------------------------------------------------------------------------
+        // 16bit primitives
+        //-----------------------------------------------------------------------------
+
+        CDS_ITANIUM_ATOMIC_LOAD( 2, 16 )
+        CDS_ITANIUM_ATOMIC_STORE( 2, 16 )
+        CDS_ITANIUM_ATOMIC_CAS( 2, 16 )
+        CDS_ITANIUM_ATOMIC_EXCHANGE( 2, 16 )
+
+        //-----------------------------------------------------------------------------
+        // 32bit primitives
+        //-----------------------------------------------------------------------------
+
+        CDS_ITANIUM_ATOMIC_LOAD( 4, 32 )
+        CDS_ITANIUM_ATOMIC_STORE( 4, 32 )
+        CDS_ITANIUM_ATOMIC_CAS( 4, 32 )
+        CDS_ITANIUM_ATOMIC_EXCHANGE( 4, 32 )
+
+#       define CDS_ATOMIC_fetch32_add_defined
+        template <typename T>
+        static inline T fetch32_add( T volatile * pDest, T val, memory_order order) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( pDest );
+
+            T cur;
+            switch ( val ) {
+                case 1:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 4, 1 );
+                    break;
+                case 4:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 4, 4 );
+                    break;
+                case 8:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 4, 8 );
+                    break;
+                case 16:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 4, 16 );
+                    break;
+                default:
+                    cur = load32( pDest, memory_order_relaxed );
+                    do {} while ( !cas32_strong( pDest, cur, cur + val, order, memory_order_relaxed ));
+                    break;
+            }
+            return cur;
+        }
+
+#       define CDS_ATOMIC_fetch32_sub_defined
+        template <typename T>
+        static inline T fetch32_sub( T volatile * pDest, T val, memory_order order) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( pDest );
+            T cur;
+            switch ( val ) {
+                case 1:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 4, -1 );
+                    break;
+                case 4:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 4, -4 );
+                    break;
+                case 8:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 4, -8 );
+                    break;
+                case 16:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 4, -16 );
+                    break;
+                default:
+                    cur = load32( pDest, memory_order_relaxed );
+                    do {} while ( !cas32_strong( pDest, cur, cur - val, order, memory_order_relaxed ));
+                    break;
+            }
+            return cur;
+        }
+
+        //-----------------------------------------------------------------------------
+        // 64bit primitives
+        //-----------------------------------------------------------------------------
+
+        CDS_ITANIUM_ATOMIC_LOAD( 8, 64 )
+        CDS_ITANIUM_ATOMIC_STORE( 8, 64 )
+        CDS_ITANIUM_ATOMIC_CAS( 8, 64 )
+        CDS_ITANIUM_ATOMIC_EXCHANGE( 8, 64 )
+
+#       define CDS_ATOMIC_fetch64_add_defined
+        template <typename T>
+        static inline T fetch64_add( T volatile * pDest, T val, memory_order order) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( pDest );
+
+            T cur;
+            switch ( val ) {
+                case 1:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, 1 );
+                    break;
+                case 4:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, 4 );
+                    break;
+                case 8:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, 8 );
+                    break;
+                case 16:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, 16 );
+                    break;
+                default:
+                    cur = load64( pDest, memory_order_relaxed );
+                    do {} while ( !cas64_strong( pDest, cur, cur + val, order, memory_order_relaxed ));
+                    break;
+            }
+            return cur;
+        }
+
+#       define CDS_ATOMIC_fetch64_sub_defined
+        template <typename T>
+        static inline T fetch64_sub( T volatile * pDest, T val, memory_order order) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( pDest );
+            T cur;
+            switch ( val ) {
+                case 1:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, -1 );
+                    break;
+                case 4:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, -4 );
+                    break;
+                case 8:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, -8 );
+                    break;
+                case 16:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, -16 );
+                    break;
+                default:
+                    cur = load64( pDest, memory_order_relaxed );
+                    do {} while ( !cas64_strong( pDest, cur, cur - val, order, memory_order_relaxed ));
+                    break;
+            }
+            return cur;
+        }
+
+        //-----------------------------------------------------------------------------
+        // pointer primitives
+        //-----------------------------------------------------------------------------
+        template <typename T>
+        static inline T * load_ptr( T * volatile const * pSrc, memory_order order ) noexcept
+        {
+            assert( order ==  memory_order_relaxed
+                 || order ==  memory_order_consume
+                 || order ==  memory_order_acquire
+                 || order == memory_order_seq_cst
+            );
+            assert( pSrc );
+            T * val;
+            __asm__ __volatile__ (
+                "ld8.acq %[val] = [%[pSrc]]  \n\t"
+                : [val] "=r" (val)
+                : [pSrc] "r" (pSrc)
+                : "memory"
+            );
+            return val;
+        }
+
+        template <typename T>
+        static inline void store_ptr( T * volatile * pDest, T * val, memory_order order ) noexcept
+        {
+            assert( order ==  memory_order_relaxed
+                 || order ==  memory_order_release
+                 || order == memory_order_seq_cst
+            );
+            assert( pDest );
+
+            if ( order == memory_order_seq_cst ) {
+                __asm__ __volatile__ (
+                    "st8.rel [%[pDest]] = %[val] \n\t"
+                    "mf     \n\t"
+                    :: [pDest] "r" (pDest), [val] "r" (val)
+                    : "memory"
+                );
+            }
+            else {
+                __asm__ __volatile__ (
+                    "st8.rel [%[pDest]] = %[val] \n\t"
+                    :: [pDest] "r" (pDest), [val] "r" (val)
+                    : "memory"
+                );
+                fence_after(order);
+            }
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_strong( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T *) == 8, "Illegal size of operand" );
+            assert( pDest );
+
+            T * current;
+
+            switch(mo_success) {
+            case memory_order_relaxed:
+            case memory_order_consume:
+            case memory_order_acquire:
+                __asm__ __volatile__ (
+                    "mov ar.ccv = %[expected] ;;\n\t"
+                    "cmpxchg8.acq %[current] = [%[pDest]], %[desired], ar.ccv\n\t"
+                    : [current] "=r" (current)
+                    : [pDest] "r" (pDest), [expected] "r" (expected), [desired] "r" (desired)
+                    : "ar.ccv", "memory"
+                );
+                break;
+            case memory_order_release:
+               __asm__ __volatile__ (
+                   "mov ar.ccv = %[expected] ;;\n\t"
+                   "cmpxchg8.rel %[current] = [%[pDest]], %[desired], ar.ccv\n\t"
+                   : [current] "=r" (current)
+                   : [pDest] "r" (pDest), [expected] "r" (expected), [desired] "r" (desired)
+                   : "ar.ccv", "memory"
+               );
+               break;
+            case memory_order_acq_rel:
+            case memory_order_seq_cst:
+               __asm__ __volatile__ (
+                   "mov ar.ccv = %[expected] ;;\n\t"
+                   "cmpxchg8.rel %[current] = [%[pDest]], %[desired], ar.ccv\n\t"
+                   "mf \n\t"
+                   : [current] "=r" (current)
+                   : [pDest] "r" (pDest), [expected] "r" (expected), [desired] "r" (desired)
+                   : "ar.ccv", "memory"
+               );
+               break;
+            default:
+                assert(false);
+            }
+
+            bool bSuccess = expected == current;
+            expected = current;
+            if ( !bSuccess )
+                fence_after( mo_fail );
+            return bSuccess;
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_weak( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas_ptr_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T * exchange_ptr( T * volatile * pDest, T * val, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == 8, "Illegal size of operand" );
+            assert( pDest );
+
+            T * current;
+            switch(order) {
+                case memory_order_relaxed:
+                case memory_order_consume:
+                case memory_order_acquire:
+                    __asm__ __volatile__ (
+                        "xchg8 %[current] = [%[pDest]], %[val]\n\t"
+                        : [current] "=r" (current)
+                        : [pDest] "r" (pDest), [val] "r" (val)
+                        : "memory"
+                    );
+                    break;
+                case memory_order_acq_rel:
+                case memory_order_release:
+                case memory_order_seq_cst:
+                    __asm__ __volatile__ (
+                        "mf \n\t"
+                        "xchg8 %[current] = [%[pDest]], %[val]\n\t"
+                        : [current] "=r" (current)
+                        : [pDest] "r" (pDest), [val] "r" (val)
+                        : "memory"
+                    );
+                    break;
+                default: assert(false);
+            }
+            return current;
+        }
+
+
+        template <typename T> struct atomic_pointer_sizeof { enum { value = sizeof(T) }; };
+        template <> struct atomic_pointer_sizeof<void> { enum { value = 1 }; };
+
+        // It does not work properly
+        // atomic.fetch_add( ... ) returns nullptr, why?..
+//#       define CDS_ATOMIC_fetch_ptr_add_defined
+        template <typename T>
+        static inline T * fetch_ptr_add( T * volatile * pDest, ptrdiff_t val, memory_order order) noexcept
+        {
+            static_assert( sizeof(T *) == 8, "Illegal size of operand" );
+            assert( pDest );
+
+            T * cur;
+            val *= atomic_pointer_sizeof<T>::value;
+            switch ( val ) {
+                case 1:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, 1 );
+                    break;
+                case 4:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, 4 );
+                    break;
+                case 8:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, 8 );
+                    break;
+                case 16:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, 16 );
+                    break;
+                default:
+                    cur = load_ptr( pDest, memory_order_relaxed );
+                    do {} while ( !cas_ptr_strong( pDest, cur, reinterpret_cast<T *>(reinterpret_cast<uint8_t *>(cur) + val), order, memory_order_relaxed ));
+                    break;
+            }
+            return cur;
+        }
+
+        // It does not work properly
+        // atomic.fetch_sub( ... ) returns nullptr, why?..
+//#       define CDS_ATOMIC_fetch_ptr_sub_defined
+        template <typename T>
+        static inline T * fetch_ptr_sub( T * volatile * pDest, ptrdiff_t val, memory_order order) noexcept
+        {
+            static_assert( sizeof(T *) == 8, "Illegal size of operand" );
+            assert( pDest );
+            T * cur;
+            val *= atomic_pointer_sizeof<T>::value;
+            switch ( val ) {
+                case 1:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, -1 );
+                    break;
+                case 4:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, -4 );
+                    break;
+                case 8:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, -8 );
+                    break;
+                case 16:
+                    CDS_ITANIUM_ATOMIC_FETCH_ADD( 8, -16 );
+                    break;
+                default:
+                    cur = load_ptr( pDest, memory_order_relaxed );
+                    do {} while ( !cas_ptr_strong( pDest, cur, reinterpret_cast<T *>(reinterpret_cast<uint8_t *>(cur) - val), order, memory_order_relaxed ));
+                    break;
+            }
+            return cur;
+        }
+
+        //-----------------------------------------------------------------------------
+        // atomic flag primitives
+        //-----------------------------------------------------------------------------
+
+        typedef bool atomic_flag_type;
+        static inline bool atomic_flag_tas( atomic_flag_type volatile * pFlag, memory_order order ) noexcept
+        {
+            return exchange8( pFlag, true, order );
+        }
+
+        static inline void atomic_flag_clear( atomic_flag_type volatile * pFlag, memory_order order ) noexcept
+        {
+            store8( pFlag, false, order );
+        }
+
+#undef CDS_ITANIUM_ATOMIC_LOAD
+#undef CDS_ITANIUM_ATOMIC_STORE
+#undef CDS_ITANIUM_ATOMIC_CAS
+#undef CDS_ITANIUM_ATOMIC_EXCHANGE
+#undef CDS_ITANIUM_ATOMIC_FETCH_ADD
+
+    }} // namespace gcc::ia64
+
+    }   // namespace platform
+}}  // namespace cds::cxx11_atomic
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_IA64_CXX11_ATOMIC_H

extern/libcds/cds/compiler/gcc/ppc64/backoff.h

@@ -0,0 +1,29 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_PPC64_BACKOFF_H
+#define CDSLIB_COMPILER_GCC_PPC64_BACKOFF_H
+
+//@cond none
+
+namespace cds { namespace backoff {
+    namespace gcc { namespace ppc64 {
+
+#       define CDS_backoff_hint_defined
+        static inline void backoff_hint()
+        {
+            // Provide a hint that performance will probably be improved
+            // if shared resources dedicated to the executing processor are released for use by other processors
+            asm volatile( "or 27,27,27   # yield" );
+        }
+    }} // namespace gcc::ppc64
+
+    namespace platform {
+        using namespace gcc::ppc64;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_GCC_PPC64_BACKOFF_H

extern/libcds/cds/compiler/gcc/ppc64/bitop.h

@@ -0,0 +1,20 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_PPC64_BITOP_H
+#define CDSLIB_COMPILER_GCC_PPC64_BITOP_H
+
+//@cond none
+namespace cds {
+    namespace bitop { namespace platform { namespace gcc { namespace ppc64 {
+
+    }} // namespace gcc::ppc64
+
+    using namespace gcc::ppc64;
+
+}}}    // namespace cds::bitop::platform
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_PPC64_BITOP_H

extern/libcds/cds/compiler/gcc/sparc/backoff.h

@@ -0,0 +1,29 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_SPARC_BACKOFF_H
+#define CDSLIB_COMPILER_GCC_SPARC_BACKOFF_H
+
+//@cond none
+
+namespace cds { namespace backoff {
+    namespace gcc { namespace Sparc {
+
+#       define CDS_backoff_nop_defined
+        static inline void backoff_nop()
+        {
+            asm volatile ( "nop;" );
+        }
+
+
+    }} // namespace gcc::Sparc
+
+    namespace platform {
+        using namespace gcc::Sparc;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_GCC_SPARC_BACKOFF_H

extern/libcds/cds/compiler/gcc/sparc/bitop.h

@@ -0,0 +1,45 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_SPARC_BITOP_H
+#define CDSLIB_COMPILER_GCC_SPARC_BITOP_H
+
+//@cond none
+namespace cds {
+    namespace bitop { namespace platform { namespace gcc { namespace Sparc {
+
+        // MSB - return index (1..64) of most significant bit in nArg. If nArg == 0 return 0
+        // Source: UltraSPARC Architecture 2007
+        //
+        // Test result: this variant and its variation about 100 times slower then generic implementation :-(
+        static inline int sparc_msb64( uint64_t nArg )
+        {
+            uint64_t result;
+            asm volatile (
+                "neg %[nArg], %[result] \n\t"
+                "xnor %[nArg], %[result], %%g5 \n\t"
+                "popc %%g5, %[result] \n\t"
+                "movrz %[nArg], %%g0, %[result] \n\t"
+                : [result] "=r" (result)
+                : [nArg] "r" (nArg)
+                : "g5"
+            );
+            return result;
+        }
+
+        // MSB - return index (1..32) of most significant bit in nArg. If nArg == 0 return 0
+        static inline int sparc_msb32( uint32_t nArg )
+        {
+            return sparc_msb64( (uint64_t) nArg );
+        }
+
+    }} // namespace gcc::Sparc
+
+    using namespace gcc::Sparc;
+
+}}}    // namespace cds::bitop::platform
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_SPARC_BITOP_H

extern/libcds/cds/compiler/gcc/sparc/cxx11_atomic.h

@@ -0,0 +1,610 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_SPARC_CXX11_ATOMIC_H
+#define CDSLIB_COMPILER_GCC_SPARC_CXX11_ATOMIC_H
+
+#include <cstdint>
+
+/*
+    Source:
+
+    1. [Doug Lea "JSR-133 Cookbook for Compiler Writers]:
+
+        Acquire semantics: load; LoadLoad+LoadStore
+        Release semantics: LoadStore+StoreStore; store
+
+    2. boost::atomic library by Helge Bahman
+    3. OpenSparc source code
+*/
+
+#if CDS_OS_TYPE == CDS_OS_LINUX
+#   define CDS_SPARC_RMO_MEMORY_MODEL
+#endif
+
+#define CDS_SPARC_MB_FULL    "membar #Sync \n\t"
+#ifdef CDS_SPARC_RMO_MEMORY_MODEL
+        // RMO memory model (Linux only?..) Untested
+#   define CDS_SPARC_MB_LL_LS       "membar #LoadLoad|#LoadStore \n\t"
+#   define CDS_SPARC_MB_LS_SS       "membar #LoadStore|#StoreStore \n\t"
+#   define CDS_SPARC_MB_LL_LS_SS    "membar #LoadLoad|#LoadStore|#StoreStore \n\t"
+#else
+        // TSO memory model (default; Solaris uses this model)
+#   define CDS_SPARC_MB_LL_LS
+#   define CDS_SPARC_MB_LS_SS
+#   define CDS_SPARC_MB_LL_LS_SS
+#endif
+
+#define CDS_SPARC_MB_ACQ        CDS_SPARC_MB_LL_LS
+#define CDS_SPARC_MB_REL        CDS_SPARC_MB_LS_SS
+#define CDS_SPARC_MB_ACQ_REL    CDS_SPARC_MB_LL_LS_SS
+#define CDS_SPARC_MB_SEQ_CST    CDS_SPARC_MB_FULL
+
+//@cond
+namespace cds { namespace cxx11_atomic {
+    namespace platform { inline namespace gcc { inline namespace Sparc {
+
+        static inline void fence_before( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_acquire:
+            case memory_order_consume:
+                break;
+            case memory_order_release:
+            case memory_order_acq_rel:
+                __asm__ __volatile__ ( "" CDS_SPARC_MB_REL ::: "memory" );
+                break;
+            case memory_order_seq_cst:
+                __asm__ __volatile__ ( "" CDS_SPARC_MB_FULL ::: "memory" );
+                break;
+            }
+        }
+
+        static inline void fence_after( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_consume:
+            case memory_order_release:
+                break;
+            case memory_order_acquire:
+            case memory_order_acq_rel:
+                __asm__ __volatile__ ( "" CDS_SPARC_MB_ACQ ::: "memory" );
+                break;
+            case memory_order_seq_cst:
+                __asm__ __volatile__ ( "" CDS_SPARC_MB_FULL ::: "memory" );
+                break;
+            }
+        }
+
+
+        //-----------------------------------------------------------------------------
+        // fences
+        //-----------------------------------------------------------------------------
+        static inline void thread_fence(memory_order order) noexcept
+        {
+            switch(order)
+            {
+                case memory_order_relaxed:
+                case memory_order_consume:
+                    break;
+                case memory_order_acquire:
+                    __asm__ __volatile__ ( "" CDS_SPARC_MB_ACQ ::: "memory" );
+                    break;
+                case memory_order_release:
+                    __asm__ __volatile__ ( "" CDS_SPARC_MB_REL ::: "memory" );
+                    break;
+                case memory_order_acq_rel:
+                    __asm__ __volatile__ ( "" CDS_SPARC_MB_ACQ_REL ::: "memory" );
+                    break;
+                case memory_order_seq_cst:
+                    __asm__ __volatile__ ( "" CDS_SPARC_MB_SEQ_CST ::: "memory"  );
+                    break;
+                default:;
+            }
+        }
+
+        static inline void signal_fence(memory_order order) noexcept
+        {
+            // C++11: 29.8.8: only compiler optimization, no hardware instructions
+            switch(order)
+            {
+                case memory_order_relaxed:
+                    break;
+                case memory_order_consume:
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                case memory_order_seq_cst:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                default:;
+            }
+        }
+
+        //-----------------------------------------------------------------------------
+        // atomic flag primitives
+        //-----------------------------------------------------------------------------
+
+        typedef unsigned char atomic_flag_type;
+        static inline bool atomic_flag_tas( atomic_flag_type volatile * pFlag, memory_order order ) noexcept
+        {
+            atomic_flag_type fCur;
+            fence_before( order );
+            __asm__ __volatile__(
+                "ldstub    [%[pFlag]], %[fCur] \n\t"
+                : [fCur] "=r"(fCur)
+                : [pFlag] "r"(pFlag)
+                : "memory", "cc"
+                );
+            fence_after( order );
+            return fCur != 0;
+        }
+
+        static inline void atomic_flag_clear( atomic_flag_type volatile * pFlag, memory_order order ) noexcept
+        {
+            fence_before( order );
+            __asm__ __volatile__(
+                CDS_SPARC_MB_REL
+                "stub    %%g0, [%[pFlag]] \n\t"
+                :: [pFlag] "r"(pFlag)
+                : "memory"
+            );
+            fence_after( order );
+        }
+
+        //-----------------------------------------------------------------------------
+        // 32bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline void store32( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            fence_before(order);
+            *pDest = src;
+            fence_after(order);
+        }
+
+        template <typename T>
+        static inline T load32( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            fence_before(order);
+            T v = *pSrc;
+            fence_after(order);
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas32_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( pDest );
+
+            fence_before( mo_success );
+            __asm__ __volatile__(
+                "cas [%[pDest]], %[expected], %[desired]"
+                : [desired] "+r" (desired)
+                : [pDest] "r" (pDest), [expected] "r" (expected)
+                : "memory"
+                );
+
+            // desired contains current value
+
+            bool bSuccess = desired == expected;
+            if ( bSuccess )
+                fence_after( mo_success );
+            else {
+                fence_after(mo_fail);
+                expected = desired;
+            }
+
+            return bSuccess;
+        }
+
+        template <typename T>
+        static inline bool cas32_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas32_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T exchange32( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( pDest );
+
+            // This primitive could be implemented via "swap" instruction but "swap" is deprecated in UltraSparc
+
+            T cur = load32( pDest, memory_order_relaxed );
+            do {} while ( !cas32_strong( pDest, cur, v, order, memory_order_relaxed ));
+            return cur;
+        }
+
+        //-----------------------------------------------------------------------------
+        // 64bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T load64( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            fence_before(order);
+            T v = *pSrc;
+            fence_after(order);
+            return v;
+        }
+
+        template <typename T>
+        static inline void store64( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            fence_before(order);
+            *pDest = val;
+            fence_after(order);
+
+        }
+
+        template <typename T>
+        static inline bool cas64_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( pDest );
+
+            fence_before( mo_success );
+            __asm__ __volatile__(
+                "casx [%[pDest]], %[expected], %[desired]"
+                : [desired] "+r" (desired)
+                : [pDest] "r" (pDest), [expected] "r" (expected)
+                : "memory"
+                );
+
+            // desired contains current value
+
+            bool bSuccess = desired == expected;
+            if ( bSuccess ) {
+                fence_after( mo_success );
+            }
+            else {
+                fence_after(mo_fail);
+                expected = desired;
+            }
+
+            return bSuccess;
+        }
+
+        template <typename T>
+        static inline bool cas64_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas64_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T exchange64( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( pDest );
+
+            T cur = load64( pDest, memory_order_relaxed );
+            do {} while ( !cas64_strong( pDest, cur, v, order, memory_order_relaxed ));
+            return cur;
+        }
+
+        //-----------------------------------------------------------------------------
+        // 8bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline void store8( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            fence_before( order );
+            *pDest = src;
+            fence_after( order );
+        }
+
+        template <typename T>
+        static inline T load8( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order == memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            fence_before( order );
+            T v = *pSrc;
+            fence_after( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas8_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( pDest );
+
+            union u32 {
+                uint32_t    w;
+                T           c[4];
+            };
+            static_assert( sizeof(u32) == sizeof(uint32_t), "Argument size error" );
+
+            u32 volatile *  pDest32 = (u32 *)( uintptr_t( pDest ) & ~0x03 );
+            size_t const  nCharIdx = (size_t)( uintptr_t( pDest ) & 0x03 );
+            u32 uExpected;
+            u32 uDesired;
+
+            bool bSuccess;
+            for (;;) {
+                uExpected.w =
+                    uDesired.w = pDest32->w;
+                uExpected.c[nCharIdx] = expected;
+                uDesired.c[nCharIdx] = desired;
+
+                bSuccess = cas32_weak( reinterpret_cast<uint32_t volatile *>(pDest32), uExpected.w, uDesired.w, mo_success, mo_fail );
+                if ( bSuccess || uExpected.c[nCharIdx] != expected )
+                    break;
+            }
+
+            expected = uExpected.c[nCharIdx];
+            return bSuccess;
+        }
+
+        template <typename T>
+        static inline bool cas8_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( pDest );
+
+            union u32 {
+                uint32_t    w;
+                T           c[4];
+            };
+            static_assert( sizeof(u32) == sizeof(uint32_t), "Argument size error" );
+
+            u32 volatile * pDest32 = (u32 *)( uintptr_t( pDest ) & ~0x03 );
+            size_t const  nCharIdx = (size_t)( uintptr_t( pDest ) & 0x03 );
+            u32 uExpected;
+            u32 uDesired;
+
+            uExpected.w =
+                uDesired.w = pDest32->w;
+            uExpected.c[nCharIdx] = expected;
+            uDesired.c[nCharIdx] = desired;
+
+            bool bSuccess = cas32_weak( reinterpret_cast<uint32_t volatile *>(pDest32), uExpected.w, uDesired.w, mo_success, mo_fail );
+
+            expected = uExpected.c[nCharIdx];
+            return bSuccess;
+        }
+
+        template <typename T>
+        static inline T exchange8( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( pDest );
+
+            T cur = load8( pDest, memory_order_relaxed );
+            do {} while ( !cas8_strong( pDest, cur, v, order, memory_order_relaxed ));
+            return cur;
+        }
+
+        //-----------------------------------------------------------------------------
+        // 16bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T load16( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            fence_before( order );
+            T v = *pSrc;
+            fence_after( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline void store16( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            fence_before(order);
+            *pDest = src;
+            fence_after(order);
+        }
+
+        template <typename T>
+        static inline bool cas16_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( pDest );
+
+            union u32 {
+                uint32_t    w;
+                T           c[2];
+            };
+            static_assert( sizeof(u32) == sizeof(uint32_t), "Argument size error" );
+
+            u32 volatile *  pDest32 = (u32 *)( uintptr_t( pDest ) & ~0x03 );
+            size_t const  nIdx = (size_t)( (uintptr_t( pDest ) >> 1) & 0x01 );
+            u32 uExpected;
+            u32 uDesired;
+
+            bool bSuccess;
+            for (;;) {
+                uExpected.w =
+                    uDesired.w = pDest32->w;
+                uExpected.c[nIdx] = expected;
+                uDesired.c[nIdx] = desired;
+
+                bSuccess = cas32_weak( reinterpret_cast<uint32_t volatile *>(pDest32), uExpected.w, uDesired.w, mo_success, mo_fail );
+                if ( bSuccess || uExpected.c[nIdx] != expected )
+                    break;
+            }
+
+            expected = uExpected.c[nIdx];
+            return bSuccess;
+        }
+
+        template <typename T>
+        static inline bool cas16_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( pDest );
+
+            union u32 {
+                uint32_t    w;
+                T           c[2];
+            };
+            static_assert( sizeof(u32) == sizeof(uint32_t), "Argument size error" );
+
+            u32 volatile * pDest32 = (u32 *)( uintptr_t( pDest ) & ~0x03 );
+            size_t const  nIdx = (size_t)( (uintptr_t( pDest ) >> 1) & 0x01 );
+            u32 uExpected;
+            u32 uDesired;
+
+            uExpected.w =
+                uDesired.w = pDest32->w;
+            uExpected.c[nIdx] = expected;
+            uDesired.c[nIdx] = desired;
+
+            bool bSuccess = cas32_weak( reinterpret_cast<uint32_t volatile *>(pDest32), uExpected.w, uDesired.w, mo_success, mo_fail );
+
+            expected = uExpected.c[nIdx];
+            return bSuccess;
+        }
+
+        template <typename T>
+        static inline T exchange16( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( pDest );
+
+            T cur = load16( pDest, memory_order_relaxed );
+            do {} while ( !cas16_strong( pDest, cur, v, order, memory_order_relaxed ));
+            return cur;
+        }
+
+        //-----------------------------------------------------------------------------
+        // pointer primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline void store_ptr( T * volatile * pDest, T * src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            fence_before(order);
+            *pDest = src;
+            fence_after(order);
+        }
+
+        template <typename T>
+        static inline T * load_ptr( T * volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            fence_before( order );
+            T * v = *pSrc;
+            fence_after( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_strong( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+
+            return cas64_strong( (uint64_t volatile *) pDest, *reinterpret_cast<uint64_t *>( &expected ), (uint64_t) desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_weak( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas_ptr_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T * exchange_ptr( T * volatile * pDest, T * v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            return (T *) exchange64( (uint64_t volatile *) pDest, (uint64_t) v, order );
+        }
+
+    }} // namespace gcc::Sparc
+
+    }   // namespace platform
+}}  // namespace cds::cxx11_atomic
+//@endcond
+
+#undef CDS_SPARC_MB_ACQ
+#undef CDS_SPARC_MB_REL
+#undef CDS_SPARC_MB_SEQ_CST
+#undef CDS_SPARC_MB_FULL
+#undef CDS_SPARC_MB_LL_LS
+#undef CDS_SPARC_MB_LS_SS
+#undef CDS_SPARC_MB_LL_LS_SS
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_AMD64_CXX11_ATOMIC_H

extern/libcds/cds/compiler/gcc/x86/backoff.h

@@ -0,0 +1,35 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_X86_BACKOFF_H
+#define CDSLIB_COMPILER_GCC_X86_BACKOFF_H
+
+//@cond none
+
+namespace cds { namespace backoff {
+    namespace gcc { namespace x86 {
+
+#       define CDS_backoff_nop_defined
+        static inline void backoff_nop()
+        {
+            asm volatile ( "nop;" );
+        }
+
+#       define CDS_backoff_hint_defined
+        static inline void backoff_hint()
+        {
+            asm volatile ( "pause;" );
+        }
+
+
+    }} // namespace gcc::x86
+
+    namespace platform {
+        using namespace gcc::x86;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_GCC_X86_BACKOFF_H

extern/libcds/cds/compiler/gcc/x86/bitop.h

@@ -0,0 +1,89 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_X86_BITOP_H
+#define CDSLIB_COMPILER_GCC_X86_BITOP_H
+
+//@cond none
+namespace cds {
+    namespace bitop { namespace platform { namespace gcc { namespace x86 {
+        // MSB - return index (1..32) of most significant bit in nArg. If nArg == 0 return 0
+#        define cds_bitop_msb32_DEFINED
+        static inline int msb32( uint32_t nArg )
+        {
+            int        nRet;
+            __asm__ __volatile__ (
+                "bsrl        %[nArg], %[nRet]     ;\n\t"
+                "jnz        1f                    ;\n\t"
+                "xorl        %[nRet], %[nRet]    ;\n\t"
+                "subl        $1, %[nRet]         ;\n\t"
+            "1:"
+                "addl        $1, %[nRet]         ;\n\t"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+            );
+            return nRet;
+        }
+
+#        define cds_bitop_msb32nz_DEFINED
+        static inline int msb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            int        nRet;
+            __asm__ __volatile__ (
+                "bsrl        %[nArg], %[nRet]    ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+            );
+            return nRet;
+        }
+
+        // LSB - return index (0..31) of least significant bit in nArg. If nArg == 0 return -1U
+#        define cds_bitop_lsb32_DEFINED
+        static inline int lsb32( uint32_t nArg )
+        {
+
+            int        nRet;
+            __asm__ __volatile__ (
+                "bsfl        %[nArg], %[nRet]     ;"
+                "jnz        1f        ;"
+                "xorl        %[nRet], %[nRet]    ;"
+                "subl        $1, %[nRet]         ;"
+                "1:"
+                "addl        $1, %[nRet]         ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return nRet;
+
+        }
+
+        // LSB - return index (0..31) of least significant bit in nArg.
+        // Condition: nArg != 0
+#        define cds_bitop_lsb32nz_DEFINED
+        static inline int lsb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            int        nRet;
+            __asm__ __volatile__ (
+                "bsfl        %[nArg], %[nRet]    ;"
+                : [nRet] "=a" (nRet)
+                : [nArg] "r" (nArg)
+                : "cc"
+                );
+            return nRet;
+        }
+
+    }} // namespace gcc::x86
+
+    using namespace gcc::x86;
+
+}}}    // namespace cds::bitop::platform
+//@endcond
+
+#endif    // #ifndef CDSLIB_ARH_X86_GCC_BITOP_H

extern/libcds/cds/compiler/gcc/x86/cxx11_atomic.h

@@ -0,0 +1,185 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_X86_CXX11_ATOMIC_H
+#define CDSLIB_COMPILER_GCC_X86_CXX11_ATOMIC_H
+
+#include <cstdint>
+#include <cds/compiler/gcc/x86/cxx11_atomic32.h>
+
+//@cond
+namespace cds { namespace cxx11_atomic {
+    namespace platform { inline namespace gcc { inline namespace x86 {
+
+        //-----------------------------------------------------------------------------
+        // 64bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline bool cas64_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            uint32_t ebxStore;
+            T prev = expected;
+
+            fence_before(mo_success);
+
+            // We must save EBX in PIC mode
+            __asm__ __volatile__ (
+                "movl %%ebx, %[ebxStore]\n"
+                "movl %[desiredLo], %%ebx\n"
+                "lock; cmpxchg8b 0(%[pDest])\n"
+                "movl %[ebxStore], %%ebx\n"
+                : [prev] "=A" (prev), [ebxStore] "=m" (ebxStore)
+                : [desiredLo] "D" ((int)desired), [desiredHi] "c" ((int)(desired >> 32)), [pDest] "S" (pDest), "0" (prev)
+                : "memory");
+            bool success = (prev == expected);
+            if (success)
+                fence_after(mo_success);
+            else {
+                fence_after(mo_fail);
+                expected = prev;
+            }
+            return success;
+        }
+
+        template <typename T>
+        static inline bool cas64_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas64_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T load64( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+            assert( cds::details::is_aligned( pSrc, 8 ));
+            CDS_UNUSED( order );
+
+            T CDS_DATA_ALIGNMENT(8) v;
+            __asm__ __volatile__(
+                "movq   (%[pSrc]), %[v]   ;   \n\t"
+                : [v] "=x" (v)
+                : [pSrc] "r" (pSrc)
+                :
+            );
+            return v;
+        }
+
+
+        template <typename T>
+        static inline T exchange64( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            T cur = load64( pDest, memory_order_relaxed );
+            do {
+            } while (!cas64_weak( pDest, cur, v, order, memory_order_relaxed ));
+            return cur;
+        }
+
+        template <typename T>
+        static inline void store64( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                // Atomically stores 64bit value by SSE instruction
+                __asm__ __volatile__(
+                    "movq       %[val], (%[pDest])   ;   \n\t"
+                    :
+                    : [val] "x" (val), [pDest] "r" (pDest)
+                    : "memory"
+                    );
+            }
+            else {
+                exchange64( pDest, val, order );
+            }
+        }
+
+
+        //-----------------------------------------------------------------------------
+        // pointer primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T * exchange_ptr( T * volatile * pDest, T * v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+
+            return (T *) exchange32( (uint32_t volatile *) pDest, (uint32_t) v, order );
+        }
+
+        template <typename T>
+        static inline void store_ptr( T * volatile * pDest, T * src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange_ptr( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T * load_ptr( T * volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            T * v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_strong( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+
+            return cas32_strong( (uint32_t volatile *) pDest, *reinterpret_cast<uint32_t *>( &expected ), (uint32_t) desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_weak( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas_ptr_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+    }} // namespace gcc::x86
+
+    }   // namespace platform
+}}  // namespace cds::cxx11_atomic
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_X86_CXX11_ATOMIC_H

extern/libcds/cds/compiler/gcc/x86/cxx11_atomic32.h

@@ -0,0 +1,477 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_GCC_X86_CXX11_ATOMIC32_H
+#define CDSLIB_COMPILER_GCC_X86_CXX11_ATOMIC32_H
+
+#include <cstdint>
+#include <cds/details/is_aligned.h>
+
+//@cond
+namespace cds { namespace cxx11_atomic {
+    namespace platform { inline namespace gcc { inline namespace x86 {
+
+        static inline void fence_before( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_acquire:
+            case memory_order_consume:
+                break;
+            case memory_order_release:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_seq_cst:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            }
+        }
+
+        static inline void fence_after( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_acquire:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_relaxed:
+            case memory_order_consume:
+            case memory_order_release:
+                break;
+            case memory_order_seq_cst:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            }
+        }
+
+
+        static inline void fence_after_load(memory_order order) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_release:
+                break;
+            case memory_order_acquire:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_consume:
+                break;
+            case memory_order_seq_cst:
+                __asm__ __volatile__ ( "mfence" ::: "memory"  );
+                break;
+            default:;
+            }
+        }
+
+        //-----------------------------------------------------------------------------
+        // fences
+        //-----------------------------------------------------------------------------
+        static inline void thread_fence(memory_order order) noexcept
+        {
+            switch(order)
+            {
+                case memory_order_relaxed:
+                case memory_order_consume:
+                    break;
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                case memory_order_seq_cst:
+                    __asm__ __volatile__ ( "mfence" ::: "memory"  );
+                    break;
+                default:;
+            }
+        }
+
+        static inline void signal_fence(memory_order order) noexcept
+        {
+            // C++11: 29.8.8: only compiler optimization, no hardware instructions
+            switch(order)
+            {
+                case memory_order_relaxed:
+                    break;
+                case memory_order_consume:
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                case memory_order_seq_cst:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                default:;
+            }
+        }
+
+        //-----------------------------------------------------------------------------
+        // 8bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline bool cas8_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+
+            T prev = expected;
+            fence_before(mo_success);
+            __asm__ __volatile__  (
+                "lock ; cmpxchgb %[desired], %[pDest]"
+                : [prev] "+a" (prev), [pDest] "+m" (*pDest)
+                : [desired] "q" (desired)
+                );
+            bool success = (prev == expected);
+            expected = prev;
+            if (success)
+                fence_after(mo_success);
+            else
+                fence_after(mo_fail);
+            return success;
+        }
+
+        template <typename T>
+        static inline bool cas8_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas8_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T exchange8( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+
+            fence_before(order);
+            __asm__ __volatile__  (
+                "xchgb %[v], %[pDest]"
+                : [v] "+q" (v), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return v;
+        }
+
+        template <typename T>
+        static inline void store8( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest != NULL );
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange8( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load8( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order == memory_order_seq_cst
+                );
+            assert( pSrc != NULL );
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+#       define CDS_ATOMIC_fetch8_add_defined
+        template <typename T>
+        static inline T fetch8_add( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            fence_before(order);
+            __asm__ __volatile__  (
+                "lock ; xaddb %[val], %[pDest]"
+                : [val] "+q" (val), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return val;
+        }
+
+#       define CDS_ATOMIC_fetch8_sub_defined
+        template <typename T>
+        static inline T fetch8_sub( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            fence_before(order);
+            __asm__ __volatile__  (
+                "negb %[val] ; \n"
+                "lock ; xaddb %[val], %[pDest]"
+                : [val] "+q" (val), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return val;
+        }
+
+        //-----------------------------------------------------------------------------
+        // atomic flag primitives
+        //-----------------------------------------------------------------------------
+
+        typedef bool atomic_flag_type;
+        static inline bool atomic_flag_tas( atomic_flag_type volatile * pFlag, memory_order order ) noexcept
+        {
+            return exchange8( pFlag, true, order );
+        }
+
+        static inline void atomic_flag_clear( atomic_flag_type volatile * pFlag, memory_order order ) noexcept
+        {
+            store8( pFlag, false, order );
+        }
+
+        //-----------------------------------------------------------------------------
+        // 16bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T exchange16( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            fence_before(order);
+            __asm__ __volatile__  (
+                "xchgw %[v], %[pDest]"
+                : [v] "+q" (v), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return v;
+        }
+
+        template <typename T>
+        static inline void store16( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest != NULL );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange16( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load16( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc != NULL );
+            assert( cds::details::is_aligned( pSrc, 2 ));
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas16_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            T prev = expected;
+            fence_before(mo_success);
+            __asm__ __volatile__  (
+                "lock ; cmpxchgw %[desired], %[pDest]"
+                : [prev] "+a" (prev), [pDest] "+m" (*pDest)
+                : [desired] "q" (desired)
+                );
+            bool success = prev == expected;
+            if (success)
+                fence_after(mo_success);
+            else {
+                fence_after(mo_fail);
+                expected = prev;
+            }
+
+            return success;
+        }
+
+        template <typename T>
+        static inline bool cas16_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas16_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+#       define CDS_ATOMIC_fetch16_add_defined
+        template <typename T>
+        static inline T fetch16_add( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            fence_before(order);
+            __asm__ __volatile__  (
+                "lock ; xaddw %[val], %[pDest]"
+                : [val] "+q" (val), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return val;
+        }
+
+#       define CDS_ATOMIC_fetch16_sub_defined
+        template <typename T>
+        static inline T fetch16_sub( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            fence_before(order);
+            __asm__ __volatile__  (
+                "negw %[val] ; \n"
+                "lock ; xaddw %[val], %[pDest]"
+                : [val] "+q" (val), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return val;
+        }
+
+        //-----------------------------------------------------------------------------
+        // 32bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T exchange32( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            fence_before(order);
+            __asm__ __volatile__  (
+                "xchgl %[v], %[pDest]"
+                : [v] "+r" (v), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return v;
+        }
+
+        template <typename T>
+        static inline void store32( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest != NULL );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange32( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load32( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc != NULL );
+            assert( cds::details::is_aligned( pSrc, 4 ));
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas32_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            T prev = expected;
+            fence_before(mo_success);
+            __asm__ __volatile__  (
+                "lock ; cmpxchgl %[desired], %[pDest]"
+                : [prev] "+a" (prev), [pDest] "+m" (*pDest)
+                : [desired] "r" (desired)
+                );
+            bool success = prev == expected;
+            if (success)
+                fence_after(mo_success);
+            else {
+                fence_after(mo_fail);
+                expected = prev;
+            }
+            return success;
+        }
+
+        template <typename T>
+        static inline bool cas32_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas32_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        // fetch_xxx may be emulated via cas32
+        // If the platform has special fetch_xxx instruction
+        // then it should define CDS_ATOMIC_fetch32_xxx_defined macro
+
+#       define CDS_ATOMIC_fetch32_add_defined
+        template <typename T>
+        static inline T fetch32_add( T volatile * pDest, T v, memory_order order) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            fence_before(order);
+            __asm__ __volatile__  (
+                "lock ; xaddl %[v], %[pDest]"
+                : [v] "+r" (v), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return v;
+        }
+
+#       define CDS_ATOMIC_fetch32_sub_defined
+        template <typename T>
+        static inline T fetch32_sub( T volatile * pDest, T v, memory_order order) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            fence_before(order);
+            __asm__ __volatile__  (
+                "negl   %[v] ; \n"
+                "lock ; xaddl %[v], %[pDest]"
+                : [v] "+r" (v), [pDest] "+m" (*pDest)
+                );
+            fence_after(order);
+            return v;
+        }
+
+    }}} // namespace platform::gcc::x86
+}}  // namespace cds::cxx11_atomic
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_GCC_X86_CXX11_ATOMIC32_H

extern/libcds/cds/compiler/icl/compiler_barriers.h

@@ -0,0 +1,30 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_ICL_COMPILER_BARRIERS_H
+#define CDSLIB_COMPILER_ICL_COMPILER_BARRIERS_H
+
+#if defined(_MSC_VER) && _MSC_VER < 1700
+    // VC++ up to vc10
+#   include <intrin.h>
+
+#   pragma intrinsic(_ReadWriteBarrier)
+#   pragma intrinsic(_ReadBarrier)
+#   pragma intrinsic(_WriteBarrier)
+
+#   define CDS_COMPILER_RW_BARRIER  _ReadWriteBarrier()
+#   define CDS_COMPILER_R_BARRIER   _ReadBarrier()
+#   define CDS_COMPILER_W_BARRIER   _WriteBarrier()
+
+#else
+    // MS VC11+, linux
+#   include <atomic>
+
+#   define CDS_COMPILER_RW_BARRIER  std::atomic_thread_fence( std::memory_order_acq_rel )
+#   define CDS_COMPILER_R_BARRIER   CDS_COMPILER_RW_BARRIER
+#   define CDS_COMPILER_W_BARRIER   CDS_COMPILER_RW_BARRIER
+#endif
+
+#endif  // #ifndef CDSLIB_COMPILER_ICL_COMPILER_BARRIERS_H

extern/libcds/cds/compiler/icl/defs.h

@@ -0,0 +1,133 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_ICL_DEFS_H
+#define CDSLIB_COMPILER_ICL_DEFS_H
+//@cond
+
+// Compiler version
+#ifdef __ICL
+#   define CDS_COMPILER_VERSION    __ICL
+#else
+#   define CDS_COMPILER_VERSION    __INTEL_COMPILER
+#endif
+
+// Compiler name
+// Supported compilers: MS VC 2008, 2010, 2012
+//
+#   define  CDS_COMPILER__NAME  "Intel C++"
+#   define  CDS_COMPILER__NICK  "icl"
+
+// OS name
+#if defined(_WIN64)
+#   define CDS_OS_INTERFACE CDS_OSI_WINDOWS
+#   define CDS_OS_TYPE      CDS_OS_WIN64
+#   define CDS_OS__NAME     "Win64"
+#   define CDS_OS__NICK     "Win64"
+#elif defined(_WIN32)
+#   define CDS_OS_INTERFACE CDS_OSI_WINDOWS
+#   define CDS_OS_TYPE      CDS_OS_WIN32
+#   define CDS_OS__NAME     "Win32"
+#   define CDS_OS__NICK     "Win32"
+#elif defined( __linux__ )
+#   define CDS_OS_INTERFACE     CDS_OSI_UNIX
+#   define CDS_OS_TYPE          CDS_OS_LINUX
+#   define CDS_OS__NAME         "linux"
+#   define CDS_OS__NICK         "linux"
+#endif
+
+// Processor architecture
+#if defined(_M_X64) || defined(_M_AMD64) || defined(__amd64__) || defined(__amd64)
+#   define CDS_BUILD_BITS       64
+#   define CDS_PROCESSOR_ARCH   CDS_PROCESSOR_AMD64
+#   define CDS_PROCESSOR__NAME  "AMD64"
+#   define CDS_PROCESSOR__NICK  "amd64"
+#elif defined(_M_IX86) || defined(__i386__) || defined(__i386)
+#   define CDS_BUILD_BITS       32
+#   define CDS_PROCESSOR_ARCH   CDS_PROCESSOR_X86
+#   define CDS_PROCESSOR__NAME  "Intel x86"
+#   define CDS_PROCESSOR__NICK  "x86"
+#else
+#   define CDS_BUILD_BITS        -1
+#   define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_UNKNOWN
+#   define CDS_PROCESSOR__NAME    "<<Undefined>>"
+#   error Intel C++ compiler is supported for x86 only
+#endif
+
+#if CDS_OS_INTERFACE == CDS_OSI_WINDOWS
+#   define  __attribute__( _x )
+#endif
+
+#if CDS_OS_INTERFACE == CDS_OSI_WINDOWS
+#   ifdef CDS_BUILD_LIB
+#       define CDS_EXPORT_API          __declspec(dllexport)
+#   else
+#       define CDS_EXPORT_API          __declspec(dllimport)
+#   endif
+#endif
+
+#if CDS_OS_INTERFACE == CDS_OSI_WINDOWS
+#   define alignof     __alignof
+#else
+#   define alignof __alignof__
+#endif
+
+// *************************************************
+// Alignment macro
+
+#if CDS_OS_INTERFACE == CDS_OSI_WINDOWS
+#   define CDS_TYPE_ALIGNMENT(n)     __declspec( align(n))
+#   define CDS_DATA_ALIGNMENT(n)     __declspec( align(n))
+#   define CDS_CLASS_ALIGNMENT(n)    __declspec( align(n))
+#else
+#   define CDS_TYPE_ALIGNMENT(n)   __attribute__ ((aligned (n)))
+#   define CDS_CLASS_ALIGNMENT(n)  __attribute__ ((aligned (n)))
+#   define CDS_DATA_ALIGNMENT(n)   __attribute__ ((aligned (n)))
+#endif
+
+// Attributes
+#if CDS_OS_INTERFACE == CDS_OSI_WINDOWS
+#   define CDS_DEPRECATED( reason ) __declspec(deprecated( reason ))
+#   define CDS_NORETURN __declspec(noreturn)
+#else
+#   define CDS_DEPRECATED( reason ) __attribute__((deprecated( reason )))
+#   define CDS_NORETURN __attribute__((__noreturn__))
+#endif
+
+// Exceptions
+#if CDS_OS_INTERFACE == CDS_OSI_WINDOWS
+#   if defined( _CPPUNWIND )
+#       define CDS_EXCEPTION_ENABLED
+#   endif
+#else
+#   if defined( __EXCEPTIONS ) && __EXCEPTIONS == 1
+#       define CDS_EXCEPTION_ENABLED
+#   endif
+#endif
+
+// Byte order
+#if !defined(CDS_ARCH_LITTLE_ENDIAN) && !defined(CDS_ARCH_BIG_ENDIAN)
+#   if CDS_OS_INTERFACE == CDS_OSI_WINDOWS
+#       define CDS_ARCH_LITTLE_ENDIAN
+#   else
+#       ifdef __BYTE_ORDER__
+#           if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+#               define CDS_ARCH_LITTLE_ENDIAN
+#           elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+#               define CDS_ARCH_BIG_ENDIAN
+#           endif
+#       else
+#           warning "Undefined byte order for current architecture (no __BYTE_ORDER__ preprocessor definition)"
+#       endif
+#   endif
+#endif
+
+// Sanitizer attributes (not supported)
+#define CDS_SUPPRESS_SANITIZE( ... )
+
+#include <cds/compiler/icl/compiler_barriers.h>
+
+//@endcond
+#endif // #ifndef CDSLIB_COMPILER_VC_DEFS_H

extern/libcds/cds/compiler/vc/amd64/backoff.h

@@ -0,0 +1,35 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_VC_AMD64_BACKOFF_H
+#define CDSLIB_COMPILER_VC_AMD64_BACKOFF_H
+
+//@cond none
+#include <intrin.h>
+
+namespace cds { namespace backoff {
+    namespace vc { namespace amd64 {
+
+#       define CDS_backoff_hint_defined
+        static inline void backoff_hint()
+        {
+            _mm_pause();
+        }
+
+#       define CDS_backoff_nop_defined
+        static inline void backoff_nop()
+        {
+            __nop();
+        }
+
+    }} // namespace vc::amd64
+
+    namespace platform {
+        using namespace vc::amd64;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_VC_AMD64_BACKOFF_H

extern/libcds/cds/compiler/vc/amd64/bitop.h

@@ -0,0 +1,129 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_VC_AMD64_BITOP_H
+#define CDSLIB_COMPILER_VC_AMD64_BITOP_H
+
+#if _MSC_VER == 1500
+    /*
+        VC 2008 bug:
+            math.h(136) : warning C4985: 'ceil': attributes not present on previous declaration.
+            intrin.h(142) : see declaration of 'ceil'
+
+        See http://connect.microsoft.com/VisualStudio/feedback/details/381422/warning-of-attributes-not-present-on-previous-declaration-on-ceil-using-both-math-h-and-intrin-h
+    */
+#   pragma warning(push)
+#   pragma warning(disable: 4985)
+#   include <intrin.h>
+#   pragma warning(pop)
+#else
+#   include <intrin.h>
+#endif
+
+#pragma intrinsic(_BitScanReverse)
+#pragma intrinsic(_BitScanForward)
+#pragma intrinsic(_BitScanReverse64)
+#pragma intrinsic(_BitScanForward64)
+
+//@cond none
+namespace cds {
+    namespace bitop { namespace platform { namespace vc { namespace amd64 {
+
+        // MSB - return index (1..32) of most significant bit in nArg. If nArg == 0 return 0
+#        define cds_bitop_msb32_DEFINED
+        static inline int msb32( uint32_t nArg )
+        {
+            unsigned long nIndex;
+            if ( _BitScanReverse( &nIndex, nArg ))
+                return (int) nIndex + 1;
+            return 0;
+        }
+
+#        define cds_bitop_msb32nz_DEFINED
+        static inline int msb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            unsigned long nIndex;
+            _BitScanReverse( &nIndex, nArg );
+            return (int) nIndex;
+        }
+
+        // LSB - return index (1..32) of least significant bit in nArg. If nArg == 0 return -1U
+#        define cds_bitop_lsb32_DEFINED
+        static inline int lsb32( uint32_t nArg )
+        {
+            unsigned long nIndex;
+            if ( _BitScanForward( &nIndex, nArg ))
+                return (int) nIndex + 1;
+            return 0;
+        }
+
+#        define cds_bitop_lsb32nz_DEFINED
+        static inline int lsb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            unsigned long nIndex;
+            _BitScanForward( &nIndex, nArg );
+            return (int) nIndex;
+        }
+
+
+#        define cds_bitop_msb64_DEFINED
+        static inline int msb64( uint64_t nArg )
+        {
+            unsigned long nIndex;
+            if ( _BitScanReverse64( &nIndex, nArg ))
+                return (int) nIndex + 1;
+            return 0;
+        }
+
+#        define cds_bitop_msb64nz_DEFINED
+        static inline int msb64nz( uint64_t nArg )
+        {
+            assert( nArg != 0 );
+            unsigned long nIndex;
+            _BitScanReverse64( &nIndex, nArg );
+            return (int) nIndex;
+        }
+
+#        define cds_bitop_lsb64_DEFINED
+        static inline int lsb64( uint64_t nArg )
+        {
+            unsigned long nIndex;
+            if ( _BitScanForward64( &nIndex, nArg ))
+                return (int) nIndex + 1;
+            return 0;
+        }
+
+#        define cds_bitop_lsb64nz_DEFINED
+        static inline int lsb64nz( uint64_t nArg )
+        {
+            assert( nArg != 0 );
+            unsigned long nIndex;
+            _BitScanForward64( &nIndex, nArg );
+            return (int) nIndex;
+        }
+
+#       define cds_bitop_complement32_DEFINED
+        static inline bool complement32( uint32_t * pArg, unsigned int nBit )
+        {
+            return _bittestandcomplement( reinterpret_cast<long *>( pArg ), nBit ) != 0;
+        }
+
+#       define cds_bitop_complement64_DEFINED
+        static inline bool complement64( uint64_t * pArg, unsigned int nBit )
+        {
+            return _bittestandcomplement64( reinterpret_cast<__int64 *>( pArg ), nBit ) != 0;
+        }
+
+
+    }} // namespace vc::amd64
+
+    using namespace vc::amd64;
+
+}}}    // namespace cds::bitop::platform
+//@endcond
+
+#endif    // #ifndef CDSLIB_COMPILER_VC_AMD64_BITOP_H

extern/libcds/cds/compiler/vc/amd64/cxx11_atomic.h

@@ -0,0 +1,584 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_VC_AMD64_CXX11_ATOMIC_H
+#define CDSLIB_COMPILER_VC_AMD64_CXX11_ATOMIC_H
+
+#include <intrin.h>
+#include <emmintrin.h>  // for 128bit atomic load/store
+#include <cds/details/is_aligned.h>
+
+#pragma intrinsic( _InterlockedIncrement )
+#pragma intrinsic( _InterlockedDecrement )
+#pragma intrinsic( _InterlockedCompareExchange )
+#pragma intrinsic( _InterlockedCompareExchangePointer )
+#pragma intrinsic( _InterlockedCompareExchange16 )
+#pragma intrinsic( _InterlockedCompareExchange64 )
+#pragma intrinsic( _InterlockedExchange )
+#pragma intrinsic( _InterlockedExchange64 )
+#pragma intrinsic( _InterlockedExchangePointer )
+#pragma intrinsic( _InterlockedExchangeAdd )
+#pragma intrinsic( _InterlockedExchangeAdd64 )
+//#pragma intrinsic( _InterlockedAnd )
+//#pragma intrinsic( _InterlockedOr )
+//#pragma intrinsic( _InterlockedXor )
+//#pragma intrinsic( _InterlockedAnd64 )
+//#pragma intrinsic( _InterlockedOr64 )
+//#pragma intrinsic( _InterlockedXor64 )
+#pragma intrinsic( _interlockedbittestandset )
+#if _MSC_VER >= 1600
+#   pragma intrinsic( _InterlockedCompareExchange8 )
+#   pragma intrinsic( _InterlockedExchange8 )
+#   pragma intrinsic( _InterlockedExchange16 )
+#endif
+
+//@cond
+namespace cds { namespace cxx11_atomic {
+    namespace platform { inline namespace vc { inline namespace amd64 {
+
+        static inline void fence_before( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_acquire:
+            case memory_order_consume:
+                break;
+            case memory_order_release:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_seq_cst:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            }
+        }
+
+        static inline void fence_after( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_acquire:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_relaxed:
+            case memory_order_consume:
+            case memory_order_release:
+                break;
+            case memory_order_seq_cst:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            }
+        }
+
+        static inline void full_fence()
+        {
+            // MS VC does not support inline assembler in C code.
+            // So, we use InterlockedExchange for full fence instead of mfence inst
+            long t;
+            _InterlockedExchange( &t, 0 );
+        }
+
+        static inline void fence_after_load(memory_order order) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_release:
+                break;
+            case memory_order_acquire:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_consume:
+                break;
+            case memory_order_seq_cst:
+                full_fence();
+                break;
+            default:;
+            }
+        }
+
+        //-----------------------------------------------------------------------------
+        // fences
+        //-----------------------------------------------------------------------------
+        static inline void thread_fence(memory_order order) noexcept
+        {
+            switch(order)
+            {
+                case memory_order_relaxed:
+                case memory_order_consume:
+                    break;
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                case memory_order_seq_cst:
+                    full_fence();
+                    break;
+                default:;
+            }
+        }
+
+        static inline void signal_fence(memory_order order) noexcept
+        {
+            // C++11: 29.8.8: only compiler optimization, no hardware instructions
+            switch(order)
+            {
+                case memory_order_relaxed:
+                    break;
+                case memory_order_consume:
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                case memory_order_seq_cst:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                default:;
+            }
+        }
+
+        //-----------------------------------------------------------------------------
+        // atomic flag primitives
+        //-----------------------------------------------------------------------------
+
+        typedef unsigned char atomic_flag_type;
+        static inline bool atomic_flag_tas( atomic_flag_type volatile * pFlag, memory_order /*order*/ ) noexcept
+        {
+            return _interlockedbittestandset( (long volatile *) pFlag, 0 ) != 0;
+        }
+
+        static inline void atomic_flag_clear( atomic_flag_type volatile * pFlag, memory_order order ) noexcept
+        {
+            assert( order != memory_order_acquire
+                && order != memory_order_acq_rel
+                );
+
+            fence_before( order );
+            *pFlag = 0;
+            fence_after( order );
+        }
+
+        //-----------------------------------------------------------------------------
+        // 8bit primitives
+        //-----------------------------------------------------------------------------
+
+#if _MSC_VER >= 1600
+#   pragma warning(push)
+        // Disable warning C4800: 'char' : forcing value to bool 'true' or 'false' (performance warning)
+#   pragma warning( disable: 4800 )
+#endif
+        template <typename T>
+        static inline bool cas8_strong( T volatile * pDest, T& expected, T desired, memory_order /*mo_success*/, memory_order /*mo_fail*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+
+#       if _MSC_VER >= 1600
+            // VC 2010 +
+            T prev = expected;
+            expected = (T) _InterlockedCompareExchange8( (char volatile*) pDest, (char) desired, (char) expected );
+            return expected == prev;
+#       else
+            // VC 2008
+            unsigned int *  pnDest = (unsigned int *)( ((unsigned __int64) pDest) & ~(unsigned __int64(3)));
+            unsigned int    nOffset = ((unsigned __int64) pDest) & 3;
+            unsigned int    nExpected;
+            unsigned int    nDesired;
+
+            for (;;) {
+                nExpected =
+                    nDesired = *pnDest;
+                memcpy( reinterpret_cast<T *>(&nExpected) + nOffset, &expected, sizeof(T));
+                memcpy( reinterpret_cast<T *>(&nDesired) + nOffset, &desired, sizeof(T));
+
+                unsigned int nPrev = (unsigned int) _InterlockedCompareExchange( (long *) pnDest, (long) nDesired, (long) nExpected );
+                if ( nPrev == nExpected )
+                    return true;
+                T nByte;
+                memcpy( &nByte, reinterpret_cast<T *>(&nPrev) + nOffset, sizeof(T));
+                if ( nByte != expected ) {
+                    expected = nByte;
+                    return false;
+                }
+            }
+#       endif
+        }
+#if _MSC_VER >= 1600
+#   pragma warning(pop)
+#endif
+
+        template <typename T>
+        static inline bool cas8_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas8_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+#if _MSC_VER >= 1600
+#   pragma warning(push)
+        // Disable warning C4800: 'char' : forcing value to bool 'true' or 'false' (performance warning)
+#   pragma warning( disable: 4800 )
+#endif
+        template <typename T>
+        static inline T exchange8( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+
+#       if _MSC_VER >= 1600
+            CDS_UNUSED(order);
+            return (T) _InterlockedExchange8( (char volatile *) pDest, (char) v );
+#       else
+            T expected = *pDest;
+            do {} while ( !cas8_strong( pDest, expected, v, order, memory_order_relaxed ));
+            return expected;
+#       endif
+        }
+#if _MSC_VER >= 1600
+#   pragma warning(pop)
+#endif
+
+        template <typename T>
+        static inline void store8( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange8( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load8( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order == memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        //-----------------------------------------------------------------------------
+        // 16bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline bool cas16_strong( T volatile * pDest, T& expected, T desired, memory_order /*mo_success*/, memory_order /*mo_fail*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            // _InterlockedCompareExchange behave as read-write memory barriers
+            T prev = expected;
+            expected = (T) _InterlockedCompareExchange16( (short *) pDest, (short) desired, (short) expected );
+            return expected == prev;
+        }
+
+        template <typename T>
+        static inline bool cas16_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas16_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T exchange16( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+#       if _MSC_VER >= 1600
+            order;
+            return (T) _InterlockedExchange16( (short volatile *) pDest, (short) v );
+#       else
+            T expected = *pDest;
+            do {} while ( !cas16_strong( pDest, expected, v, order, memory_order_relaxed ));
+            return expected;
+#       endif
+        }
+
+        template <typename T>
+        static inline void store16( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange16( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load16( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+            assert( cds::details::is_aligned( pSrc, 2 ));
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        //-----------------------------------------------------------------------------
+        // 32bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T exchange32( T volatile * pDest, T v, memory_order /*order*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            return (T) _InterlockedExchange( (long *) pDest, (long) v );
+        }
+
+        template <typename T>
+        static inline void store32( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange32( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load32( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+            assert( cds::details::is_aligned( pSrc, 4 ));
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas32_strong( T volatile * pDest, T& expected, T desired, memory_order /*mo_success*/, memory_order /*mo_fail*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            // _InterlockedCompareExchange behave as read-write memory barriers
+            T prev = expected;
+            expected = (T) _InterlockedCompareExchange( (long *) pDest, (long) desired, (long) expected );
+            return expected == prev;
+        }
+
+        template <typename T>
+        static inline bool cas32_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas32_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        // fetch_xxx may be emulated via cas32
+        // If the platform has special fetch_xxx instruction
+        // then it should define CDS_ATOMIC_fetch32_xxx_defined macro
+
+#       define CDS_ATOMIC_fetch32_add_defined
+        template <typename T>
+        static inline T fetch32_add( T volatile * pDest, T v, memory_order /*order*/) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            // _InterlockedExchangeAdd behave as read-write memory barriers
+            return (T) _InterlockedExchangeAdd( (long *) pDest, (long) v );
+        }
+
+        //-----------------------------------------------------------------------------
+        // 64bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline bool cas64_strong( T volatile * pDest, T& expected, T desired, memory_order /*mo_success*/, memory_order /*mo_fail*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            // _InterlockedCompareExchange behave as read-write memory barriers
+            T prev = expected;
+            expected = (T) _InterlockedCompareExchange64( (__int64 *) pDest, (__int64) desired, (__int64) expected );
+            return expected == prev;
+        }
+
+        template <typename T>
+        static inline bool cas64_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas64_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T load64( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+            assert( cds::details::is_aligned( pSrc, 8 ));
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+
+        template <typename T>
+        static inline T exchange64( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+
+            T cur = load64( pDest, memory_order_relaxed );
+            do {
+            } while (!cas64_weak( pDest, cur, v, order, memory_order_relaxed ));
+            return cur;
+        }
+
+        template <typename T>
+        static inline void store64( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = val;
+            }
+            else {
+                exchange64( pDest, val, order );
+            }
+        }
+
+#       define CDS_ATOMIC_fetch64_add_defined
+        template <typename T>
+        static inline T fetch64_add( T volatile * pDest, T v, memory_order /*order*/) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal size of operand" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            // _InterlockedExchangeAdd64 behave as read-write memory barriers
+            return (T) _InterlockedExchangeAdd64( (__int64 *) pDest, (__int64) v );
+        }
+
+        //-----------------------------------------------------------------------------
+        // pointer primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T * exchange_ptr( T * volatile * pDest, T * v, memory_order /*order*/ ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            return (T *) _InterlockedExchangePointer( (void * volatile *) pDest, reinterpret_cast<void *>(v));
+        }
+
+        template <typename T>
+        static inline void store_ptr( T * volatile * pDest, T * src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange_ptr( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T * load_ptr( T * volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            T * v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_strong( T * volatile * pDest, T *& expected, T * desired, memory_order /*mo_success*/, memory_order /*mo_fail*/ ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal size of operand" );
+
+            // _InterlockedCompareExchangePointer behave as read-write memory barriers
+            T * prev = expected;
+            expected = (T *) _InterlockedCompareExchangePointer( (void * volatile *) pDest, (void *) desired, (void *) expected );
+            return expected == prev;
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_weak( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas_ptr_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+    }} // namespace vc::amd64
+
+    } // namespace platform
+}}  // namespace cds::cxx11_atomic
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_VC_AMD64_CXX11_ATOMIC_H

extern/libcds/cds/compiler/vc/compiler_barriers.h

@@ -0,0 +1,19 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_VC_COMPILER_BARRIERS_H
+#define CDSLIB_COMPILER_VC_COMPILER_BARRIERS_H
+
+#include <intrin.h>
+
+#pragma intrinsic(_ReadWriteBarrier)
+#pragma intrinsic(_ReadBarrier)
+#pragma intrinsic(_WriteBarrier)
+
+#define CDS_COMPILER_RW_BARRIER  _ReadWriteBarrier()
+#define CDS_COMPILER_R_BARRIER   _ReadBarrier()
+#define CDS_COMPILER_W_BARRIER   _WriteBarrier()
+
+#endif  // #ifndef CDSLIB_COMPILER_VC_COMPILER_BARRIERS_H

extern/libcds/cds/compiler/vc/defs.h

@@ -0,0 +1,143 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_VC_DEFS_H
+#define CDSLIB_COMPILER_VC_DEFS_H
+//@cond
+
+// Compiler version
+#define CDS_COMPILER_VERSION    _MSC_VER
+
+// Compiler name
+// Supported compilers: MS VC 2015 +
+// C++ compiler versions:
+#define CDS_COMPILER_MSVC14         1900    // 2015 vc14
+#define CDS_COMPILER_MSVC14_1       1910    // 2017 vc14.1
+#define CDS_COMPILER_MSVC14_1_3     1911    // 2017 vc14.1 (VS 15.3)
+#define CDS_COMPILER_MSVC14_1_5     1912    // 2017 vc14.1 (VS 15.5)
+#define CDS_COMPILER_MSVC15         2000    // next Visual Studio
+
+#if CDS_COMPILER_VERSION < CDS_COMPILER_MSVC14
+#   error "Only MS Visual C++ 14 (2015) and above is supported"
+#endif
+
+#if _MSC_VER == 1900
+#   define  CDS_COMPILER__NAME  "MS Visual C++ 2015"
+#   define  CDS_COMPILER__NICK  "vc14"
+#   define  CDS_COMPILER_LIBCDS_SUFFIX "vcv140"
+#elif _MSC_VER < 2000
+#   define  CDS_COMPILER__NAME  "MS Visual C++ 2017"
+#   define  CDS_COMPILER__NICK  "vc141"
+#   define  CDS_COMPILER_LIBCDS_SUFFIX "vcv141"
+#else
+#   define  CDS_COMPILER__NAME  "MS Visual C++"
+#   define  CDS_COMPILER__NICK  "msvc"
+#   define  CDS_COMPILER_LIBCDS_SUFFIX "vc"
+#endif
+
+// OS interface
+#define CDS_OS_INTERFACE CDS_OSI_WINDOWS
+
+// OS name
+#if defined(_WIN64)
+#   define CDS_OS_TYPE      CDS_OS_WIN64
+#   define CDS_OS__NAME     "Win64"
+#   define CDS_OS__NICK     "Win64"
+#elif defined(_WIN32)
+#   define CDS_OS_TYPE      CDS_OS_WIN32
+#   define CDS_OS__NAME     "Win32"
+#   define CDS_OS__NICK     "Win32"
+#endif
+
+// Processor architecture
+#ifdef _M_IX86
+#   define CDS_BUILD_BITS       32
+#   define CDS_PROCESSOR_ARCH   CDS_PROCESSOR_X86
+#   define CDS_PROCESSOR__NAME  "Intel x86"
+#   define CDS_PROCESSOR__NICK  "x86"
+#elif _M_X64
+#   define CDS_BUILD_BITS       64
+#   define CDS_PROCESSOR_ARCH   CDS_PROCESSOR_AMD64
+#   define CDS_PROCESSOR__NAME  "AMD64"
+#   define CDS_PROCESSOR__NICK  "amd64"
+#else
+#   define CDS_BUILD_BITS        -1
+#   define CDS_PROCESSOR_ARCH    CDS_PROCESSOR_UNKNOWN
+#   define CDS_PROCESSOR__NAME    "<<Undefined>>"
+#   error Microsoft Visual C++ compiler is supported for x86 only
+#endif
+
+#define  __attribute__( _x )
+
+#ifndef CDS_BUILD_STATIC_LIB
+#   ifdef CDS_BUILD_LIB
+#       define CDS_EXPORT_API          __declspec(dllexport)
+#   else
+#       define CDS_EXPORT_API          __declspec(dllimport)
+#   endif
+#else
+#   define CDS_EXPORT_API
+#endif
+
+#define alignof     __alignof
+
+// Memory leaks detection (debug build only)
+#ifdef _DEBUG
+#   define _CRTDBG_MAP_ALLOC
+#   define _CRTDBG_MAPALLOC
+#   include <stdlib.h>
+#   include <crtdbg.h>
+#   define CDS_MSVC_MEMORY_LEAKS_DETECTING_ENABLED
+#endif
+
+// *************************************************
+// Alignment macro
+
+#define CDS_TYPE_ALIGNMENT(n)     __declspec( align(n))
+#define CDS_DATA_ALIGNMENT(n)     __declspec( align(n))
+#define CDS_CLASS_ALIGNMENT(n)    __declspec( align(n))
+
+// Attributes
+#define CDS_DEPRECATED( reason ) [[deprecated( reason )]]
+
+#define CDS_NORETURN __declspec(noreturn)
+
+// Exceptions
+#if defined( _CPPUNWIND )
+#   define CDS_EXCEPTION_ENABLED
+#endif
+
+
+// double-width CAS support
+//#define CDS_DCAS_SUPPORT
+
+// Byte order
+//  It seems, MSVC works only on little-endian architecture?..
+#if !defined(CDS_ARCH_LITTLE_ENDIAN) && !defined(CDS_ARCH_BIG_ENDIAN)
+#   define CDS_ARCH_LITTLE_ENDIAN
+#endif
+
+//if constexpr support (C++17)
+#ifndef constexpr_if
+    // Standard way to check if the compiler supports "if constexpr"
+    // Of course, MS VC doesn't support any standard way
+#   if defined __cpp_if_constexpr
+#       if __cpp_if_constexpr >= 201606
+#           define constexpr_if if constexpr
+#       endif
+#   elif CDS_COMPILER_VERSION >= CDS_COMPILER_MSVC14_1_3 && _MSVC_LANG > CDS_CPLUSPLUS_14
+        // MS-specific WTF.
+        // Don't work in /std:c++17 because /std:c++17 key defines _MSVC_LANG=201402 (c++14) in VC 15.3
+#       define constexpr_if if constexpr
+#   endif
+#endif
+
+// Sanitizer attributes (not supported)
+#define CDS_SUPPRESS_SANITIZE( ... )
+
+#include <cds/compiler/vc/compiler_barriers.h>
+
+//@endcond
+#endif // #ifndef CDSLIB_COMPILER_VC_DEFS_H

extern/libcds/cds/compiler/vc/x86/backoff.h

@@ -0,0 +1,35 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_VC_X86_BACKOFF_H
+#define CDSLIB_COMPILER_VC_X86_BACKOFF_H
+
+//@cond none
+#include <intrin.h>
+
+namespace cds { namespace backoff {
+    namespace vc { namespace x86 {
+
+#       define CDS_backoff_hint_defined
+        static inline void backoff_hint()
+        {
+            _mm_pause();
+        }
+
+#       define CDS_backoff_nop_defined
+        static inline void backoff_nop()
+        {
+            __nop();
+        }
+
+    }} // namespace vc::x86
+
+    namespace platform {
+        using namespace vc::x86;
+    }
+}}  // namespace cds::backoff
+
+//@endcond
+#endif  // #ifndef CDSLIB_COMPILER_VC_X86_BACKOFF_H

extern/libcds/cds/compiler/vc/x86/bitop.h

@@ -0,0 +1,86 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_VC_X86_BITOP_H
+#define CDSLIB_COMPILER_VC_X86_BITOP_H
+
+#include <intrin.h>
+#pragma intrinsic(_BitScanReverse)
+#pragma intrinsic(_BitScanForward)
+
+//@cond none
+namespace cds {
+    namespace bitop { namespace platform { namespace vc { namespace x86 {
+        // MSB - return index (1..32) of most significant bit in nArg. If nArg == 0 return 0
+#        define cds_bitop_msb32_DEFINED
+        static inline int msb32( uint32_t nArg )
+        {
+            unsigned long nIndex;
+            if ( _BitScanReverse( &nIndex, nArg ))
+                return (int) nIndex + 1;
+            return 0;
+        }
+
+#        define cds_bitop_msb32nz_DEFINED
+        static inline int msb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            unsigned long nIndex;
+            _BitScanReverse( &nIndex, nArg );
+            return (int) nIndex;
+        }
+
+        // LSB - return index (1..32) of least significant bit in nArg. If nArg == 0 return -1U
+#        define cds_bitop_lsb32_DEFINED
+        static inline int lsb32( uint32_t nArg )
+        {
+            unsigned long nIndex;
+            if ( _BitScanForward( &nIndex, nArg ))
+                return (int) nIndex + 1;
+            return 0;
+        }
+
+#        define cds_bitop_lsb32nz_DEFINED
+        static inline int lsb32nz( uint32_t nArg )
+        {
+            assert( nArg != 0 );
+            unsigned long nIndex;
+            _BitScanForward( &nIndex, nArg );
+            return (int) nIndex;
+        }
+
+        // bswap - Reverses the byte order of a 32-bit word
+#        define cds_bitop_bswap32_DEFINED
+        static inline uint32_t bswap32( uint32_t nArg )
+        {
+            __asm {
+                mov    eax, nArg;
+                bswap    eax;
+            }
+        }
+
+#       define cds_bitop_complement32_DEFINED
+        static inline bool complement32( uint32_t * pArg, unsigned int nBit )
+        {
+            return _bittestandcomplement( reinterpret_cast<long *>( pArg ), nBit ) != 0;
+        }
+
+#       define cds_bitop_complement64_DEFINED
+        static inline bool complement64( uint64_t * pArg, unsigned int nBit )
+        {
+            if ( nBit < 32 )
+                return _bittestandcomplement( reinterpret_cast<long *>( pArg ), nBit ) != 0;
+            else
+                return _bittestandcomplement( reinterpret_cast<long *>( pArg ) + 1, nBit - 32 ) != 0;
+        }
+
+    }} // namespace vc::x86
+
+    using namespace vc::x86;
+
+}}}    // namespace cds::bitop::platform
+//@endcond
+
+#endif    // #ifndef CDSLIB_COMPILER_VC_X86_BITOP_H

extern/libcds/cds/compiler/vc/x86/cxx11_atomic.h

@@ -0,0 +1,556 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_COMPILER_VC_X86_CXX11_ATOMIC_H
+#define CDSLIB_COMPILER_VC_X86_CXX11_ATOMIC_H
+
+#include <intrin.h>
+#include <emmintrin.h>  // for 64bit atomic load/store
+#include <cds/details/is_aligned.h>
+
+#pragma intrinsic( _InterlockedIncrement )
+#pragma intrinsic( _InterlockedDecrement )
+#pragma intrinsic( _InterlockedCompareExchange )
+//#pragma intrinsic( _InterlockedCompareExchangePointer )   // On the x86 architecture, _InterlockedCompareExchangePointer is a macro that calls _InterlockedCompareExchange
+#pragma intrinsic( _InterlockedCompareExchange16 )
+#pragma intrinsic( _InterlockedCompareExchange64 )
+#pragma intrinsic( _InterlockedExchange )
+//#pragma intrinsic( _InterlockedExchangePointer )  // On the x86 architecture, _InterlockedExchangePointer is a macro that calls _InterlockedExchange
+#pragma intrinsic( _InterlockedExchangeAdd )
+#pragma intrinsic( _InterlockedXor )
+#pragma intrinsic( _InterlockedOr )
+#pragma intrinsic( _InterlockedAnd )
+#pragma intrinsic( _interlockedbittestandset )
+#if _MSC_VER >= 1600
+#   pragma intrinsic( _InterlockedCompareExchange8 )
+#   pragma intrinsic( _InterlockedExchange8 )
+#   pragma intrinsic( _InterlockedExchange16 )
+#endif
+
+//@cond
+namespace cds { namespace cxx11_atomic {
+    namespace platform { inline namespace vc { inline namespace x86 {
+
+        static inline void fence_before( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_acquire:
+            case memory_order_consume:
+                break;
+            case memory_order_release:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_seq_cst:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            }
+        }
+
+        static inline void fence_after( memory_order order ) noexcept
+        {
+            switch(order) {
+            case memory_order_acquire:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_relaxed:
+            case memory_order_consume:
+            case memory_order_release:
+                break;
+            case memory_order_seq_cst:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            }
+        }
+
+
+        static inline void fence_after_load(memory_order order) noexcept
+        {
+            switch(order) {
+            case memory_order_relaxed:
+            case memory_order_release:
+                break;
+            case memory_order_acquire:
+            case memory_order_acq_rel:
+                CDS_COMPILER_RW_BARRIER;
+                break;
+            case memory_order_consume:
+                break;
+            case memory_order_seq_cst:
+                __asm { mfence };
+                break;
+            default:;
+            }
+        }
+
+        //-----------------------------------------------------------------------------
+        // fences
+        //-----------------------------------------------------------------------------
+        static inline void thread_fence(memory_order order) noexcept
+        {
+            switch(order)
+            {
+                case memory_order_relaxed:
+                case memory_order_consume:
+                    break;
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                case memory_order_seq_cst:
+                    __asm { mfence };
+                    break;
+                default:;
+            }
+        }
+
+        static inline void signal_fence(memory_order order) noexcept
+        {
+            // C++11: 29.8.8: only compiler optimization, no hardware instructions
+            switch(order)
+            {
+                case memory_order_relaxed:
+                    break;
+                case memory_order_consume:
+                case memory_order_release:
+                case memory_order_acquire:
+                case memory_order_acq_rel:
+                case memory_order_seq_cst:
+                    CDS_COMPILER_RW_BARRIER;
+                    break;
+                default:;
+            }
+        }
+
+        //-----------------------------------------------------------------------------
+        // atomic flag primitives
+        //-----------------------------------------------------------------------------
+
+        typedef unsigned char atomic_flag_type;
+        static inline bool atomic_flag_tas( atomic_flag_type volatile * pFlag, memory_order /*order*/ ) noexcept
+        {
+            return _interlockedbittestandset( (long volatile *) pFlag, 0 ) != 0;
+        }
+
+        static inline void atomic_flag_clear( atomic_flag_type volatile * pFlag, memory_order order ) noexcept
+        {
+            assert( order != memory_order_acquire
+                && order != memory_order_acq_rel
+                );
+
+            fence_before( order );
+            *pFlag = 0;
+            fence_after( order );
+        }
+
+
+        //-----------------------------------------------------------------------------
+        // 8bit primitives
+        //-----------------------------------------------------------------------------
+
+#if _MSC_VER >= 1600
+#   pragma warning(push)
+        // Disable warning C4800: 'char' : forcing value to bool 'true' or 'false' (performance warning)
+#   pragma warning( disable: 4800 )
+#endif
+        template <typename T>
+        static inline bool cas8_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal operand size"  );
+
+#       if _MSC_VER >= 1600
+            T prev = expected;
+            expected = (T) _InterlockedCompareExchange8( reinterpret_cast<char volatile*>(pDest), (char) desired, (char) expected );
+            return expected == prev;
+#       else
+            bool bRet = false;
+            __asm {
+                mov ecx, pDest;
+                mov edx, expected;
+                mov al, byte ptr [edx];
+                mov ah, desired;
+                lock cmpxchg byte ptr [ecx], ah;
+                mov byte ptr [edx], al;
+                setz bRet;
+            }
+            return bRet;
+#       endif
+        }
+#if _MSC_VER >= 1600
+#   pragma warning(pop)
+#endif
+
+        template <typename T>
+        static inline bool cas8_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas8_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+#if _MSC_VER >= 1600
+#   pragma warning(push)
+    // Disable warning C4800: 'char' : forcing value to bool 'true' or 'false' (performance warning)
+#   pragma warning( disable: 4800 )
+#endif
+        template <typename T>
+        static inline T exchange8( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal operand size" );
+
+#       if _MSC_VER >= 1600
+            return (T) _InterlockedExchange8( reinterpret_cast<char volatile *>(pDest), (char) v );
+#       else
+            __asm {
+                mov al, v;
+                mov ecx, pDest;
+                lock xchg byte ptr [ecx], al;
+            }
+#       endif
+        }
+#if _MSC_VER >= 1600
+#   pragma warning(pop)
+#endif
+
+        template <typename T>
+        static inline void store8( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange8( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load8( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 1, "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order == memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        //-----------------------------------------------------------------------------
+        // 16bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T exchange16( T volatile * pDest, T v, memory_order /*order*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal operand size" );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+#       if _MSC_VER >= 1600
+            return (T) _InterlockedExchange16( (short volatile *) pDest, (short) v );
+#       else
+            __asm {
+                mov ax, v;
+                mov ecx, pDest;
+                lock xchg word ptr [ecx], ax;
+            }
+#       endif
+        }
+
+        template <typename T>
+        static inline void store16( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange16( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load16( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+            assert( cds::details::is_aligned( pSrc, 2 ));
+
+            T v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas16_strong( T volatile * pDest, T& expected, T desired, memory_order /*mo_success*/, memory_order /*mo_fail*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 2, "Illegal operand size" );
+            assert( cds::details::is_aligned( pDest, 2 ));
+
+            // _InterlockedCompareExchange behave as read-write memory barriers
+            T prev = expected;
+            expected = (T) _InterlockedCompareExchange16( (short *) pDest, (short) desired, (short) expected );
+            return expected == prev;
+        }
+
+        template <typename T>
+        static inline bool cas16_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas16_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        //-----------------------------------------------------------------------------
+        // 32bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T exchange32( T volatile * pDest, T v, memory_order /*order*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal operand size" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            return (T) _InterlockedExchange( (long *) pDest, (long) v );
+        }
+
+        template <typename T>
+        static inline void store32( T volatile * pDest, T src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange32( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T load32( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+            assert( cds::details::is_aligned( pSrc, 4 ));
+
+            T v( *pSrc );
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas32_strong( T volatile * pDest, T& expected, T desired, memory_order /*mo_success*/, memory_order /*mo_fail*/ ) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal operand size" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            // _InterlockedCompareExchange behave as read-write memory barriers
+            T prev = expected;
+            expected = (T) _InterlockedCompareExchange( (long *) pDest, (long) desired, (long) expected );
+            return expected == prev;
+        }
+
+        template <typename T>
+        static inline bool cas32_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas32_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        // fetch_xxx may be emulated via cas32
+        // If the platform has special fetch_xxx instruction
+        // then it should define CDS_ATOMIC_fetch32_xxx_defined macro
+
+#       define CDS_ATOMIC_fetch32_add_defined
+        template <typename T>
+        static inline T fetch32_add( T volatile * pDest, T v, memory_order /*order*/) noexcept
+        {
+            static_assert( sizeof(T) == 4, "Illegal operand size" );
+            assert( cds::details::is_aligned( pDest, 4 ));
+
+            // _InterlockedExchangeAdd behave as read-write memory barriers
+            return (T) _InterlockedExchangeAdd( (long *) pDest, (long) v );
+        }
+
+        //-----------------------------------------------------------------------------
+        // 64bit primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline bool cas64_strong( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal operand size" );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            // _InterlockedCompareExchange behave as read-write memory barriers
+            T prev = expected;
+            expected = (T) _InterlockedCompareExchange64( (__int64 *) pDest, (__int64) desired, (__int64) expected );
+            return expected == prev;
+        }
+
+        template <typename T>
+        static inline bool cas64_weak( T volatile * pDest, T& expected, T desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas64_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+
+        template <typename T>
+        static inline T load64( T volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+            assert( cds::details::is_aligned( pSrc, 8 ));
+
+            // Atomically loads 64bit value by SSE intrinsics
+            __m128i volatile v = _mm_loadl_epi64( (__m128i const *) pSrc );
+            fence_after_load( order );
+            return (T) v.m128i_i64[0];
+        }
+
+
+        template <typename T>
+        static inline T exchange64( T volatile * pDest, T v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal operand size" );
+
+            T cur = load64( pDest, memory_order_relaxed );
+            do {
+            } while (!cas64_weak( pDest, cur, v, order, memory_order_relaxed ));
+            return cur;
+        }
+
+        template <typename T>
+        static inline void store64( T volatile * pDest, T val, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T) == 8, "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+            assert( cds::details::is_aligned( pDest, 8 ));
+
+            if ( order != memory_order_seq_cst ) {
+                __m128i v;
+                v.m128i_i64[0] = val;
+                fence_before( order );
+                _mm_storel_epi64( (__m128i *) pDest, v );
+            }
+            else {
+                exchange64( pDest, val, order );
+            }
+        }
+
+
+        //-----------------------------------------------------------------------------
+        // pointer primitives
+        //-----------------------------------------------------------------------------
+
+        template <typename T>
+        static inline T * exchange_ptr( T * volatile * pDest, T * v, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal operand size" );
+            return (T *) _InterlockedExchange( (long volatile *) pDest, (uintptr_t) v );
+            //return (T *) _InterlockedExchangePointer( (void * volatile *) pDest, reinterpret_cast<void *>(v));
+        }
+
+        template <typename T>
+        static inline void store_ptr( T * volatile * pDest, T * src, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_release
+                || order == memory_order_seq_cst
+                );
+            assert( pDest );
+
+            if ( order != memory_order_seq_cst ) {
+                fence_before( order );
+                *pDest = src;
+            }
+            else {
+                exchange_ptr( pDest, src, order );
+            }
+        }
+
+        template <typename T>
+        static inline T * load_ptr( T * volatile const * pSrc, memory_order order ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal operand size" );
+            assert( order ==  memory_order_relaxed
+                || order ==  memory_order_consume
+                || order ==  memory_order_acquire
+                || order ==  memory_order_seq_cst
+                );
+            assert( pSrc );
+
+            T * v = *pSrc;
+            fence_after_load( order );
+            return v;
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_strong( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            static_assert( sizeof(T *) == sizeof(void *), "Illegal operand size" );
+
+            // _InterlockedCompareExchangePointer behave as read-write memory barriers
+            T * prev = expected;
+            expected = (T *) _InterlockedCompareExchange( (long volatile *) pDest, (uintptr_t) desired, (uintptr_t) prev );
+            return expected == prev;
+        }
+
+        template <typename T>
+        static inline bool cas_ptr_weak( T * volatile * pDest, T *& expected, T * desired, memory_order mo_success, memory_order mo_fail ) noexcept
+        {
+            return cas_ptr_strong( pDest, expected, desired, mo_success, mo_fail );
+        }
+    }} // namespace vc::x86
+
+    } // namespace platform
+}}  // namespace cds::cxx11_atomic
+//@endcond
+
+#endif // #ifndef CDSLIB_COMPILER_VC_X86_CXX11_ATOMIC_H

extern/libcds/cds/container/basket_queue.h

@@ -0,0 +1,456 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_BASKET_QUEUE_H
+#define CDSLIB_CONTAINER_BASKET_QUEUE_H
+
+#include <cds/intrusive/basket_queue.h>
+#include <cds/container/details/base.h>
+#include <memory>
+
+namespace cds { namespace container {
+
+    /// BasketQueue related definitions
+    /** @ingroup cds_nonintrusive_helper
+    */
+    namespace basket_queue {
+
+        /// Internal statistics
+        template <typename Counter = cds::intrusive::basket_queue::stat<>::counter_type >
+        using stat = cds::intrusive::basket_queue::stat< Counter >;
+
+        /// Dummy internal statistics
+        typedef cds::intrusive::basket_queue::empty_stat empty_stat;
+
+        /// BasketQueue default type traits
+        struct traits
+        {
+            /// Node allocator
+            typedef CDS_DEFAULT_ALLOCATOR       allocator;
+
+            /// Back-off strategy
+            typedef cds::backoff::empty         back_off;
+
+            /// Item counting feature; by default, disabled. Use \p cds::atomicity::item_counter to enable item counting
+            typedef atomicity::empty_item_counter   item_counter;
+
+            /// Internal statistics (by default, disabled)
+            /**
+                Possible option value are: \p basket_queue::stat, \p basket_queue::empty_stat (the default),
+                user-provided class that supports \p %basket_queue::stat interface.
+            */
+            typedef basket_queue::empty_stat         stat;
+
+            /// C++ memory ordering model
+            /**
+                Can be \p opt::v::relaxed_ordering (relaxed memory model, the default)
+                or \p opt::v::sequential_consistent (sequentially consisnent memory model).
+            */
+            typedef opt::v::relaxed_ordering    memory_model;
+
+            /// Padding for internal critical atomic data. Default is \p opt::cache_line_padding
+            enum { padding = opt::cache_line_padding };
+        };
+
+        /// Metafunction converting option list to \p basket_queue::traits
+        /**
+            Supported \p Options are:
+            - \p opt::allocator - allocator (like \p std::allocator) used for allocating queue nodes. Default is \ref CDS_DEFAULT_ALLOCATOR
+            - \p opt::back_off - back-off strategy used, default is \p cds::backoff::empty.
+            - \p opt::item_counter - the type of item counting feature. Default is \p cds::atomicity::empty_item_counter (item counting disabled)
+                To enable item counting use \p cds::atomicity::item_counter
+            - \ opt::stat - the type to gather internal statistics.
+                Possible statistics types are: \p basket_queue::stat, \p basket_queue::empty_stat, user-provided class that supports \p %basket_queue::stat interface.
+                Default is \p %basket_queue::empty_stat.
+            - \p opt::padding - padding for internal critical atomic data. Default is \p opt::cache_line_padding
+            - \p opt::memory_model - C++ memory ordering model. Can be \p opt::v::relaxed_ordering (relaxed memory model, the default)
+                or \p opt::v::sequential_consistent (sequentially consisnent memory model).
+
+            Example: declare \p %BasketQueue with item counting and internal statistics
+            \code
+            typedef cds::container::BasketQueue< cds::gc::HP, Foo,
+                typename cds::container::basket_queue::make_traits<
+                    cds::opt::item_counte< cds::atomicity::item_counter >,
+                    cds::opt::stat< cds::intrusive::basket_queue::stat<> >
+                >::type
+            > myQueue;
+            \endcode
+        */
+        template <typename... Options>
+        struct make_traits {
+#   ifdef CDS_DOXYGEN_INVOKED
+            typedef implementation_defined type;   ///< Metafunction result
+#   else
+            typedef typename cds::opt::make_options<
+                typename cds::opt::find_type_traits< traits, Options... >::type
+                , Options...
+            >::type type;
+#   endif
+        };
+    } // namespace basket_queue
+
+    //@cond
+    namespace details {
+        template <typename GC, typename T, typename Traits>
+        struct make_basket_queue
+        {
+            typedef GC gc;
+            typedef T value_type;
+            typedef Traits traits;
+
+            struct node_type: public intrusive::basket_queue::node< gc >
+            {
+                value_type  m_value;
+
+                node_type( const value_type& val )
+                    : m_value( val )
+                {}
+                template <typename... Args>
+                node_type( Args&&... args )
+                    : m_value( std::forward<Args>(args)...)
+                {}
+            };
+
+            typedef typename std::allocator_traits< typename traits::allocator >::template rebind_alloc< node_type > allocator_type;
+            //typedef typename traits::allocator::template rebind<node_type>::other allocator_type;
+            typedef cds::details::Allocator< node_type, allocator_type >           cxx_allocator;
+
+            struct node_deallocator
+            {
+                void operator ()( node_type * pNode )
+                {
+                    cxx_allocator().Delete( pNode );
+                }
+            };
+
+            struct intrusive_traits : public traits
+            {
+                typedef cds::intrusive::basket_queue::base_hook< opt::gc<gc> > hook;
+                typedef node_deallocator disposer;
+                static constexpr const cds::intrusive::opt::link_check_type link_checker = cds::intrusive::basket_queue::traits::link_checker;
+            };
+
+            typedef cds::intrusive::BasketQueue< gc, node_type, intrusive_traits > type;
+        };
+    }
+    //@endcond
+
+    /// Basket lock-free queue (non-intrusive variant)
+    /** @ingroup cds_nonintrusive_queue
+        It is non-intrusive version of basket queue algorithm based on intrusive::BasketQueue counterpart.
+
+        \par Source:
+            [2007] Moshe Hoffman, Ori Shalev, Nir Shavit "The Baskets Queue"
+
+        <b>Key idea</b>
+
+        In the 'basket' approach, instead of
+        the traditional ordered list of nodes, the queue consists of an ordered list of groups
+        of nodes (logical baskets). The order of nodes in each basket need not be specified, and in
+        fact, it is easiest to maintain them in LIFO order. The baskets fulfill the following basic
+        rules:
+        - Each basket has a time interval in which all its nodes' enqueue operations overlap.
+        - The baskets are ordered by the order of their respective time intervals.
+        - For each basket, its nodes' dequeue operations occur after its time interval.
+        - The dequeue operations are performed according to the order of baskets.
+
+        Two properties define the FIFO order of nodes:
+        - The order of nodes in a basket is not specified.
+        - The order of nodes in different baskets is the FIFO-order of their respective baskets.
+
+        In algorithms such as the MS-queue or optimistic
+        queue, threads enqueue items by applying a Compare-and-swap (CAS) operation to the
+        queue's tail pointer, and all the threads that fail on a particular CAS operation (and also
+        the winner of that CAS) overlap in time. In particular, they share the time interval of
+        the CAS operation itself. Hence, all the threads that fail to CAS on the tail-node of
+        the queue may be inserted into the same basket. By integrating the basket-mechanism
+        as the back-off mechanism, the time usually spent on backing-off before trying to link
+        onto the new tail, can now be utilized to insert the failed operations into the basket,
+        allowing enqueues to complete sooner. In the meantime, the next successful CAS operations
+        by enqueues allow new baskets to be formed down the list, and these can be
+        filled concurrently. Moreover, the failed operations don't retry their link attempt on the
+        new tail, lowering the overall contention on it. This leads to a queue
+        algorithm that unlike all former concurrent queue algorithms requires virtually no tuning
+        of the backoff mechanisms to reduce contention, making the algorithm an attractive
+        out-of-the-box queue.
+
+        In order to enqueue, just as in MSQueue, a thread first tries to link the new node to
+        the last node. If it failed to do so, then another thread has already succeeded. Thus it
+        tries to insert the new node into the new basket that was created by the winner thread.
+        To dequeue a node, a thread first reads the head of the queue to obtain the
+        oldest basket. It may then dequeue any node in the oldest basket.
+
+
+        Template arguments:
+        - \p GC - garbage collector type: \p gc::HP, \p gc::DHP
+        - \p T - type of value to be stored in the queue
+        - \p Traits - queue traits, default is \p basket_queue::traits. You can use \p basket_queue::make_traits
+            metafunction to make your traits or just derive your traits from \p %basket_queue::traits:
+            \code
+            struct myTraits: public cds::container::basket_queue::traits {
+                typedef cds::intrusive::basket_queue::stat<> stat;
+                typedef cds::atomicity::item_counter    item_counter;
+            };
+            typedef cds::container::BasketQueue< cds::gc::HP, Foo, myTraits > myQueue;
+
+            // Equivalent make_traits example:
+            typedef cds::container::BasketQueue< cds::gc::HP, Foo,
+                typename cds::container::basket_queue::make_traits<
+                    cds::opt::stat< cds::container::basket_queue::stat<> >,
+                    cds::opt::item_counter< cds::atomicity::item_counter >
+                >::type
+            > myQueue;
+            \endcode
+    */
+    template <typename GC, typename T, typename Traits = basket_queue::traits >
+    class BasketQueue:
+#ifdef CDS_DOXYGEN_INVOKED
+        private intrusive::BasketQueue< GC, intrusive::basket_queue::node< T >, Traits >
+#else
+        protected details::make_basket_queue< GC, T, Traits >::type
+#endif
+    {
+        //@cond
+        typedef details::make_basket_queue< GC, T, Traits > maker;
+        typedef typename maker::type base_class;
+        //@endcond
+
+    public:
+        /// Rebind template arguments
+        template <typename GC2, typename T2, typename Traits2>
+        struct rebind {
+            typedef BasketQueue< GC2, T2, Traits2> other   ;   ///< Rebinding result
+        };
+
+    public:
+        typedef GC gc;          ///< Garbage collector
+        typedef T  value_type;  ///< Type of value to be stored in the queue
+        typedef Traits traits;  ///< Queue's traits
+
+        typedef typename base_class::back_off       back_off;       ///< Back-off strategy used
+        typedef typename maker::allocator_type      allocator_type; ///< Allocator type used for allocate/deallocate the nodes
+        typedef typename base_class::item_counter   item_counter;   ///< Item counting policy used
+        typedef typename base_class::stat           stat;           ///< Internal statistics policy used
+        typedef typename base_class::memory_model   memory_model;   ///< Memory ordering. See cds::opt::memory_model option
+
+        static constexpr const size_t c_nHazardPtrCount = base_class::c_nHazardPtrCount; ///< Count of hazard pointer required for the algorithm
+
+    protected:
+        typedef typename maker::node_type node_type; ///< queue node type (derived from intrusive::basket_queue::node)
+
+        //@cond
+        typedef typename maker::cxx_allocator     cxx_allocator;
+        typedef typename maker::node_deallocator  node_deallocator;   // deallocate node
+        typedef typename base_class::node_traits  node_traits;
+        //@endcond
+
+    protected:
+        ///@cond
+        static node_type * alloc_node()
+        {
+            return cxx_allocator().New();
+        }
+        static node_type * alloc_node( const value_type& val )
+        {
+            return cxx_allocator().New( val );
+        }
+        template <typename... Args>
+        static node_type * alloc_node_move( Args&&... args )
+        {
+            return cxx_allocator().MoveNew( std::forward<Args>( args )... );
+        }
+        static void free_node( node_type * p )
+        {
+            node_deallocator()( p );
+        }
+
+        struct node_disposer {
+            void operator()( node_type * pNode )
+            {
+                free_node( pNode );
+            }
+        };
+        typedef std::unique_ptr< node_type, node_disposer > scoped_node_ptr;
+        //@endcond
+
+    public:
+        /// Initializes empty queue
+        BasketQueue()
+        {}
+
+        /// Destructor clears the queue
+        ~BasketQueue()
+        {}
+
+        /// Enqueues \p val value into the queue.
+        /**
+            The function makes queue node in dynamic memory calling copy constructor for \p val
+            and then it calls \p intrusive::BasketQueue::enqueue().
+            Returns \p true if success, \p false otherwise.
+        */
+        bool enqueue( value_type const& val )
+        {
+            scoped_node_ptr p( alloc_node(val));
+            if ( base_class::enqueue( *p )) {
+                p.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// Enqueues \p val value into the queue, move semantics
+        bool enqueue( value_type&& val )
+        {
+            scoped_node_ptr p( alloc_node_move( std::move( val )));
+            if ( base_class::enqueue( *p )) {
+                p.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// Enqueues \p data to queue using a functor
+        /**
+            \p Func is a functor called to create node.
+            The functor \p f takes one argument - a reference to a new node of type \ref value_type :
+            \code
+            cds::container::BasketQueue< cds::gc::HP, Foo > myQueue;
+            Bar bar;
+            myQueue.enqueue_with( [&bar]( Foo& dest ) { dest = bar; } );
+            \endcode
+        */
+        template <typename Func>
+        bool enqueue_with( Func f )
+        {
+            scoped_node_ptr p( alloc_node());
+            f( p->m_value );
+            if ( base_class::enqueue( *p )) {
+                p.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// Synonym for \p enqueue() function
+        bool push( value_type const& val )
+        {
+            return enqueue( val );
+        }
+
+        /// Synonym for \p enqueue() function, move semantics
+        bool push( value_type&& val )
+        {
+            return enqueue( std::move( val ));
+        }
+
+        /// Synonym for \p enqueue_with() function
+        template <typename Func>
+        bool push_with( Func f )
+        {
+            return enqueue_with( f );
+        }
+
+        /// Enqueues data of type \ref value_type constructed with <tt>std::forward<Args>(args)...</tt>
+        template <typename... Args>
+        bool emplace( Args&&... args )
+        {
+            scoped_node_ptr p( alloc_node_move( std::forward<Args>(args)...));
+            if ( base_class::enqueue( *p )) {
+                p.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// Dequeues a value from the queue
+        /**
+            If queue is not empty, the function returns \p true, \p dest contains copy of
+            dequeued value. The assignment operator for \p value_type is invoked.
+            If queue is empty, the function returns \p false, \p dest is unchanged.
+        */
+        bool dequeue( value_type& dest )
+        {
+            return dequeue_with( [&dest]( value_type& src ) {
+                // TSan finds a race between this read of \p src and node_type constructor
+                // I think, it is wrong
+                CDS_TSAN_ANNOTATE_IGNORE_READS_BEGIN;
+                dest = std::move( src );
+                CDS_TSAN_ANNOTATE_IGNORE_READS_END;
+            });
+        }
+
+        /// Dequeues a value using a functor
+        /**
+            \p Func is a functor called to copy dequeued value.
+            The functor takes one argument - a reference to removed node:
+            \code
+            cds:container::BasketQueue< cds::gc::HP, Foo > myQueue;
+            Bar bar;
+            myQueue.dequeue_with( [&bar]( Foo& src ) { bar = std::move( src );});
+            \endcode
+            The functor is called only if the queue is not empty.
+        */
+        template <typename Func>
+        bool dequeue_with( Func f )
+        {
+            typename base_class::dequeue_result res;
+            if ( base_class::do_dequeue( res, true )) {
+                f( node_traits::to_value_ptr( *res.pNext )->m_value );
+                return true;
+            }
+            return false;
+        }
+
+        /// Synonym for \p dequeue() function
+        bool pop( value_type& dest )
+        {
+            return dequeue( dest );
+        }
+
+        /// Synonym for \p dequeue_with() function
+        template <typename Func>
+        bool pop_with( Func f )
+        {
+            return dequeue_with( f );
+        }
+
+        /// Checks if the queue is empty
+        /**
+            Note that this function is not \p const.
+            The function is based on \p dequeue() algorithm.
+        */
+        bool empty()
+        {
+            return base_class::empty();
+        }
+
+        /// Clear the queue
+        /**
+            The function repeatedly calls \ref dequeue until it returns \p nullptr.
+        */
+        void clear()
+        {
+            base_class::clear();
+        }
+
+        /// Returns queue's item count
+        /** \copydetails cds::intrusive::BasketQueue::size()
+        */
+        size_t size() const
+        {
+            return base_class::size();
+        }
+
+        /// Returns reference to internal statistics
+        const stat& statistics() const
+        {
+            return base_class::statistics();
+        }
+
+    };
+
+}}  // namespace cds::container
+
+#endif  // #ifndef CDSLIB_CONTAINER_BASKET_QUEUE_H

extern/libcds/cds/container/bronson_avltree_map_rcu.h

@@ -0,0 +1,687 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_BRONSON_AVLTREE_MAP_RCU_H
+#define CDSLIB_CONTAINER_BRONSON_AVLTREE_MAP_RCU_H
+
+#include <functional>
+#include <cds/container/impl/bronson_avltree_map_rcu.h>
+
+namespace cds { namespace container {
+
+    namespace bronson_avltree {
+        //@cond
+        namespace details {
+            template < class RCU, typename Key, typename T, typename Traits>
+            struct make_map
+            {
+                typedef Key key_type;
+                typedef T mapped_type;
+                typedef Traits original_traits;
+
+                typedef cds::details::Allocator< mapped_type, typename original_traits::allocator > cxx_allocator;
+
+                struct traits : public original_traits
+                {
+                    struct disposer {
+                        void operator()( mapped_type * p ) const
+                        {
+                            cxx_allocator().Delete( p );
+                        }
+                    };
+                };
+
+                // Metafunction result
+                typedef BronsonAVLTreeMap< RCU, Key, mapped_type *, traits > type;
+            };
+        } // namespace details
+        //@endcond
+    } // namespace bronson_avltree
+
+    /// Bronson et al AVL-tree (RCU specialization)
+    /** @ingroup cds_nonintrusive_map
+        @ingroup cds_nonintrusive_tree
+        @anchor cds_container_BronsonAVLTreeMap_rcu
+
+        Source:
+            - [2010] N.Bronson, J.Casper, H.Chafi, K.Olukotun "A Practical Concurrent Binary Search Tree"
+            - <a href="http://github.com/nbronson/snaptree">Java implementation</a>
+
+        This is a concurrent AVL tree algorithm that uses hand-over-hand optimistic validation,
+        a concurrency control mechanism for searching and navigating a binary search tree.
+        This mechanism minimizes spurious retries when concurrent structural changes cannot
+        affect the correctness of the search or navigation result.
+        The algorithm is based on partially external trees, a simple scheme that simplifies deletions
+        by leaving a routing node in the tree when deleting a node that has two children,
+        then opportunistically unlinking routing nodes during rebalancing. As in external trees,
+        which store values only in leaf nodes, deletions can be performed locally while holding
+        a fixed number of locks. Partially external trees, however, require far fewer routing nodes
+        than an external tree for most sequences of insertions and deletions.
+        The algorithm uses optimistic concurrency control, but carefully manage the
+        tree in such a way that all atomic regions have fixed read and write sets
+        that are known ahead of time. This allows to reduce practical overheads by embedding
+        the concurrency control directly. To perform tree operations using only fixed sized
+        atomic regions the algo uses the following mechanisms: search operations overlap atomic blocks as
+        in the hand-over-hand locking technique; mutations perform rebalancing separately;
+        and deletions occasionally leave a routing node in the tree.
+
+        <b>Template arguments</b>:
+        - \p RCU - one of \ref cds_urcu_gc "RCU type"
+        - \p Key - key type
+        - \p T - value type to be stored in tree's nodes.
+        - \p Traits - tree traits, default is \p bronson_avltree::traits
+            It is possible to declare option-based tree with \p bronson_avltree::make_traits metafunction
+            instead of \p Traits template argument.
+
+        There is \ref cds_container_BronsonAVLTreeMap_rcu_ptr "a specialization" for "key -> value pointer" map.
+
+        @note Before including <tt><cds/container/bronson_avltree_map_rcu.h></tt> you should include appropriate RCU header file,
+        see \ref cds_urcu_gc "RCU type" for list of existing RCU class and corresponding header files.
+    */
+    template <
+        typename RCU,
+        typename Key,
+        typename T,
+#   ifdef CDS_DOXYGEN_INVOKED
+        typename Traits = bronson_avltree::traits
+#else
+        typename Traits
+#endif
+    >
+    class BronsonAVLTreeMap< cds::urcu::gc<RCU>, Key, T, Traits >
+#ifdef CDS_DOXYGEN_INVOKED
+        : private BronsonAVLTreeMap< cds::urcu::gc<RCU>, Key, T*, Traits >
+#else
+        : private bronson_avltree::details::make_map< cds::urcu::gc<RCU>, Key, T, Traits >::type
+#endif
+    {
+        //@cond
+        typedef bronson_avltree::details::make_map< cds::urcu::gc<RCU>, Key, T, Traits > maker;
+        typedef typename maker::type base_class;
+        //@endcond
+
+    public:
+        typedef cds::urcu::gc<RCU>  gc;   ///< RCU Garbage collector
+        typedef Key     key_type;    ///< type of a key stored in the map
+        typedef T       mapped_type; ///< type of value stored in the map
+        typedef Traits  traits;      ///< Traits template parameter
+
+        typedef typename base_class::key_comparator     key_comparator;     ///< key compare functor based on \p Traits::compare and \p Traits::less
+        typedef typename traits::item_counter           item_counter;       ///< Item counting policy
+        typedef typename traits::memory_model           memory_model;       ///< Memory ordering, see \p cds::opt::memory_model option
+        typedef typename traits::allocator              allocator_type;     ///< allocator for value
+        typedef typename traits::node_allocator         node_allocator_type;///< allocator for maintaining internal nodes
+        typedef typename traits::stat                   stat;               ///< internal statistics
+        typedef typename traits::rcu_check_deadlock     rcu_check_deadlock; ///< Deadlock checking policy
+        typedef typename traits::back_off               back_off;           ///< Back-off strategy
+        typedef typename traits::sync_monitor           sync_monitor;       ///< @ref cds_sync_monitor "Synchronization monitor" type for node-level locking
+
+        /// Enabled or disabled @ref bronson_avltree::relaxed_insert "relaxed insertion"
+        static bool const c_bRelaxedInsert = traits::relaxed_insert;
+
+        /// Group of \p extract_xxx functions does not require external locking
+        static constexpr const bool c_bExtractLockExternal = base_class::c_bExtractLockExternal;
+
+        typedef typename base_class::rcu_lock   rcu_lock;  ///< RCU scoped lock
+
+        /// Returned pointer to \p mapped_type of extracted node
+        typedef typename base_class::exempt_ptr exempt_ptr;
+
+    protected:
+        //@cond
+        typedef typename base_class::node_type        node_type;
+        typedef typename base_class::node_scoped_lock node_scoped_lock;
+        typedef typename maker::cxx_allocator         cxx_allocator;
+
+        typedef typename base_class::update_flags update_flags;
+        //@endcond
+
+    public:
+        /// Creates empty map
+        BronsonAVLTreeMap()
+        {}
+
+        /// Destroys the map
+        ~BronsonAVLTreeMap()
+        {}
+
+        /// Inserts new node with \p key and default value
+        /**
+            The function creates a node with \p key and default value, and then inserts the node created into the map.
+
+            Preconditions:
+            - The \p key_type should be constructible from a value of type \p K.
+            - The \p mapped_type should be default-constructible.
+
+            RCU \p synchronize() can be called. RCU should not be locked.
+
+            Returns \p true if inserting successful, \p false otherwise.
+        */
+        template <typename K>
+        bool insert( K const& key )
+        {
+            return base_class::do_update(key, key_comparator(),
+                []( node_type * pNode ) -> mapped_type*
+                {
+                    assert( pNode->m_pValue.load( memory_model::memory_order_relaxed ) == nullptr );
+                    CDS_UNUSED( pNode );
+                    return cxx_allocator().New();
+                },
+                update_flags::allow_insert
+            ) == update_flags::result_inserted;
+        }
+
+        /// Inserts new node
+        /**
+            The function creates a node with copy of \p val value
+            and then inserts the node created into the map.
+
+            Preconditions:
+            - The \p key_type should be constructible from \p key of type \p K.
+            - The \p mapped_type should be constructible from \p val of type \p V.
+
+            RCU \p synchronize() method can be called. RCU should not be locked.
+
+            Returns \p true if \p val is inserted into the map, \p false otherwise.
+        */
+        template <typename K, typename V>
+        bool insert( K const& key, V const& val )
+        {
+            return base_class::do_update( key, key_comparator(),
+                [&val]( node_type * pNode ) -> mapped_type*
+                {
+                    assert( pNode->m_pValue.load( memory_model::memory_order_relaxed ) == nullptr );
+                    CDS_UNUSED( pNode );
+                    return cxx_allocator().New( val );
+                },
+                update_flags::allow_insert
+            ) == update_flags::result_inserted;
+        }
+
+        /// Inserts new node and initialize it by a functor
+        /**
+            This function inserts new node with key \p key and if inserting is successful then it calls
+            \p func functor with signature
+            \code
+                struct functor {
+                    void operator()( key_type const& key, mapped_type& item );
+                };
+            \endcode
+
+            The key_type should be constructible from value of type \p K.
+
+            The function allows to split creating of new item into two part:
+            - create item from \p key;
+            - insert new item into the map;
+            - if inserting is successful, initialize the value of item by calling \p func functor
+
+            This can be useful if complete initialization of object of \p value_type is heavyweight and
+            it is preferable that the initialization should be completed only if inserting is successful.
+            The functor is called under the node lock.
+
+            RCU \p synchronize() method can be called. RCU should not be locked.
+        */
+        template <typename K, typename Func>
+        bool insert_with( K const& key, Func func )
+        {
+            return base_class::do_update( key, key_comparator(),
+                [&func]( node_type * pNode ) -> mapped_type*
+                {
+                    assert( pNode->m_pValue.load( memory_model::memory_order_relaxed ) == nullptr );
+                    mapped_type * pVal = cxx_allocator().New();
+                    func( pNode->m_key, *pVal );
+                    return pVal;
+                },
+                update_flags::allow_insert
+            ) == update_flags::result_inserted;
+        }
+
+        /// For \p key inserts data of type \p mapped_type created in-place from \p args
+        /**
+            Returns \p true if inserting successful, \p false otherwise.
+
+            RCU \p synchronize() method can be called. RCU should not be locked.
+        */
+        template <typename K, typename... Args>
+        bool emplace( K&& key, Args&&... args )
+        {
+            struct scoped_ptr
+            {
+                mapped_type * pVal;
+                scoped_ptr( mapped_type * p ): pVal( p ) {}
+                ~scoped_ptr() { if ( pVal ) cxx_allocator().Delete( pVal ); }
+                void release() { pVal = nullptr; }
+            };
+
+            scoped_ptr p( cxx_allocator().MoveNew( std::forward<Args>( args )... ));
+            if ( base_class::insert( std::forward<K>( key ), p.pVal )) {
+                p.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// Updates the value for \p key
+        /**
+            The operation performs inserting or changing data with lock-free manner.
+
+            If the \p key not found in the map, then the new item created from \p key
+            will be inserted into the map iff \p bAllowInsert is \p true
+            (note that in this case the \ref key_type should be constructible from type \p K).
+            Otherwise, the functor \p func is called with item found.
+            The functor \p Func signature is:
+            \code
+                struct my_functor {
+                    void operator()( bool bNew, key_type const& key, mapped_type& item );
+                };
+            \endcode
+
+            with arguments:
+            - \p bNew - \p true if the item has been inserted, \p false otherwise
+            - \p item - value
+
+            The functor may change any fields of the \p item. The functor is called under the node lock,
+            the caller can change any field of \p item.
+
+            RCU \p synchronize() method can be called. RCU should not be locked.
+
+            Returns <tt> std::pair<bool, bool> </tt> where \p first is \p true if operation is successful,
+            \p second is \p true if new item has been added or \p false if the item with \p key
+            already exists.
+        */
+        template <typename K, typename Func>
+        std::pair<bool, bool> update( K const& key, Func func, bool bAllowInsert = true )
+        {
+            int result = base_class::do_update( key, key_comparator(),
+                [&func]( node_type * pNode ) -> mapped_type*
+                {
+                    mapped_type * pVal = pNode->m_pValue.load( memory_model::memory_order_relaxed );
+                    if ( !pVal ) {
+                        pVal = cxx_allocator().New();
+                        func( true, pNode->m_key, *pVal );
+                    }
+                    else
+                        func( false, pNode->m_key, *pVal );
+                    return pVal;
+                },
+                (bAllowInsert ? update_flags::allow_insert : 0) | update_flags::allow_update
+            );
+            return std::make_pair( result != 0, (result & update_flags::result_inserted) != 0 );
+        }
+
+
+        /// Delete \p key from the map
+        /**
+            RCU \p synchronize() method can be called. RCU should not be locked.
+
+            Return \p true if \p key is found and deleted, \p false otherwise
+        */
+        template <typename K>
+        bool erase( K const& key )
+        {
+            return base_class::erase( key );
+        }
+
+        /// Deletes the item from the map using \p pred predicate for searching
+        /**
+            The function is an analog of \p erase(K const&)
+            but \p pred is used for key comparing.
+            \p Less functor has the interface like \p std::less.
+            \p Less must imply the same element order as the comparator used for building the map.
+        */
+        template <typename K, typename Less>
+        bool erase_with( K const& key, Less pred )
+        {
+            return base_class::erase_with( key, pred );
+        }
+
+        /// Delete \p key from the map
+        /** \anchor cds_nonintrusive_BronsonAVLTreeMap_rcu_erase_func
+
+            The function searches an item with key \p key, calls \p f functor
+            and deletes the item. If \p key is not found, the functor is not called.
+
+            The functor \p Func interface:
+            \code
+            struct extractor {
+                void operator()(key_type const& key, mapped_type& item) { ... }
+            };
+            \endcode
+
+            RCU \p synchronize method can be called. RCU should not be locked.
+
+            Return \p true if key is found and deleted, \p false otherwise
+        */
+        template <typename K, typename Func>
+        bool erase( K const& key, Func f )
+        {
+            return base_class::erase( key, f );
+        }
+
+        /// Deletes the item from the map using \p pred predicate for searching
+        /**
+            The function is an analog of \ref cds_nonintrusive_BronsonAVLTreeMap_rcu_erase_func "erase(K const&, Func)"
+            but \p pred is used for key comparing.
+            \p Less functor has the interface like \p std::less.
+            \p Less must imply the same element order as the comparator used for building the map.
+        */
+        template <typename K, typename Less, typename Func>
+        bool erase_with( K const& key, Less pred, Func f )
+        {
+            return base_class::erase_with( key, pred, f );
+        }
+
+        /// Extracts a value with minimal key from the map
+        /**
+            Returns \p exempt_ptr pointer to the leftmost item.
+            If the set is empty, returns empty \p exempt_ptr.
+
+            Note that the function returns only the value for minimal key.
+            To retrieve its key use \p extract_min( Func ) member function.
+
+            @note Due the concurrent nature of the map, the function extracts <i>nearly</i> minimum key.
+            It means that the function gets leftmost leaf of the tree and tries to unlink it.
+            During unlinking, a concurrent thread may insert an item with key less than leftmost item's key.
+            So, the function returns the item with minimum key at the moment of tree traversing.
+
+            RCU \p synchronize method can be called. RCU should NOT be locked.
+            The function does not free the item.
+            The deallocator will be implicitly invoked when the returned object is destroyed or when
+            its \p release() member function is called.
+        */
+        exempt_ptr extract_min()
+        {
+            return base_class::extract_min();
+        }
+
+        /// Extracts minimal key and corresponding value
+        /**
+            Returns \p exempt_ptr to the leftmost item.
+            If the tree is empty, returns empty \p exempt_ptr.
+
+            \p Func functor is used to store minimal key.
+            \p Func has the following signature:
+            \code
+                struct functor {
+                    void operator()( key_type const& key );
+                };
+            \endcode
+            If the tree is empty, \p f is not called.
+            Otherwise, is it called with minimal key, the pointer to corresponding value is returned
+            as \p exempt_ptr.
+
+            @note Due the concurrent nature of the map, the function extracts <i>nearly</i> minimum key.
+            It means that the function gets leftmost leaf of the tree and tries to unlink it.
+            During unlinking, a concurrent thread may insert an item with key less than leftmost item's key.
+            So, the function returns the item with minimum key at the moment of tree traversing.
+
+            RCU \p synchronize method can be called. RCU should NOT be locked.
+            The function does not free the item.
+            The deallocator will be implicitly invoked when the returned object is destroyed or when
+            its \p release() member function is called.
+        */
+        template <typename Func>
+        exempt_ptr extract_min( Func f )
+        {
+            return base_class::extract_min( f );
+        }
+
+        /// Extracts minimal key and corresponding value
+        /**
+            This function is a shortcut for the following call:
+            \code
+                key_type key;
+                exempt_ptr xp = theTree.extract_min( [&key]( key_type const& k ) { key = k; } );
+            \endcode
+            \p key_type should be copy-assignable. The copy of minimal key
+            is returned in \p min_key argument.
+        */
+        typename std::enable_if< std::is_copy_assignable<key_type>::value, exempt_ptr >::type
+        extract_min_key( key_type& min_key )
+        {
+            return base_class::extract_min_key( min_key );
+        }
+
+        /// Extracts an item with maximal key from the map
+        /**
+            Returns \p exempt_ptr pointer to the rightmost item.
+            If the set is empty, returns empty \p exempt_ptr.
+
+            Note that the function returns only the value for maximal key.
+            To retrieve its key use \p extract_max( Func ) or \p extract_max_key(key_type&) member function.
+
+            @note Due the concurrent nature of the map, the function extracts <i>nearly</i> maximal key.
+            It means that the function gets rightmost leaf of the tree and tries to unlink it.
+            During unlinking, a concurrent thread may insert an item with key greater than rightmost item's key.
+            So, the function returns the item with maximum key at the moment of tree traversing.
+
+            RCU \p synchronize method can be called. RCU should NOT be locked.
+            The function does not free the item.
+            The deallocator will be implicitly invoked when the returned object is destroyed or when
+            its \p release() is called.
+        */
+        exempt_ptr extract_max()
+        {
+            return base_class::extract_max();
+        }
+
+        /// Extracts the maximal key and corresponding value
+        /**
+            Returns \p exempt_ptr pointer to the rightmost item.
+            If the set is empty, returns empty \p exempt_ptr.
+
+            \p Func functor is used to store maximal key.
+            \p Func has the following signature:
+            \code
+                struct functor {
+                    void operator()( key_type const& key );
+                };
+            \endcode
+            If the tree is empty, \p f is not called.
+            Otherwise, is it called with maximal key, the pointer to corresponding value is returned
+            as \p exempt_ptr.
+
+            @note Due the concurrent nature of the map, the function extracts <i>nearly</i> maximal key.
+            It means that the function gets rightmost leaf of the tree and tries to unlink it.
+            During unlinking, a concurrent thread may insert an item with key greater than rightmost item's key.
+            So, the function returns the item with maximum key at the moment of tree traversing.
+
+            RCU \p synchronize method can be called. RCU should NOT be locked.
+            The function does not free the item.
+            The deallocator will be implicitly invoked when the returned object is destroyed or when
+            its \p release() is called.
+        */
+        template <typename Func>
+        exempt_ptr extract_max( Func f )
+        {
+            return base_class::extract_max( f );
+        }
+
+        /// Extracts the maximal key and corresponding value
+        /**
+            This function is a shortcut for the following call:
+            \code
+                key_type key;
+                exempt_ptr xp = theTree.extract_max( [&key]( key_type const& k ) { key = k; } );
+            \endcode
+            \p key_type should be copy-assignable. The copy of maximal key
+            is returned in \p max_key argument.
+        */
+        typename std::enable_if< std::is_copy_assignable<key_type>::value, exempt_ptr >::type
+        extract_max_key( key_type& max_key )
+        {
+            return base_class::extract_max_key( max_key );
+        }
+
+        /// Extracts an item from the map
+        /**
+            The function searches an item with key equal to \p key in the tree,
+            unlinks it, and returns \p exempt_ptr pointer to a value found.
+            If \p key is not found the function returns an empty \p exempt_ptr.
+
+            RCU \p synchronize method can be called. RCU should NOT be locked.
+            The function does not destroy the value found.
+            The dealloctor will be implicitly invoked when the returned object is destroyed or when
+            its \p release() member function is called.
+        */
+        template <typename Q>
+        exempt_ptr extract( Q const& key )
+        {
+            return base_class::extract( key );
+        }
+
+        /// Extracts an item from the map using \p pred for searching
+        /**
+            The function is an analog of \p extract(Q const&)
+            but \p pred is used for key compare.
+            \p Less has the interface like \p std::less.
+            \p pred must imply the same element order as the comparator used for building the map.
+        */
+        template <typename Q, typename Less>
+        exempt_ptr extract_with( Q const& key, Less pred )
+        {
+            return base_class::extract_with( key, pred );
+        }
+
+        /// Find the key \p key
+        /**
+            The function searches the item with key equal to \p key and calls the functor \p f for item found.
+            The interface of \p Func functor is:
+            \code
+            struct functor {
+                void operator()( key_type const& key, mapped_type& val );
+            };
+            \endcode
+            where \p val is the item found for \p key
+            The functor is called under node-level lock.
+
+            The function applies RCU lock internally.
+
+            The function returns \p true if \p key is found, \p false otherwise.
+        */
+        template <typename K, typename Func>
+        bool find( K const& key, Func f )
+        {
+            return base_class::find( key, f );
+        }
+
+        /// Finds the key \p val using \p pred predicate for searching
+        /**
+            The function is an analog of \p find(K const&, Func)
+            but \p pred is used for key comparing.
+            \p Less functor has the interface like \p std::less.
+            \p Less must imply the same element order as the comparator used for building the map.
+        */
+        template <typename K, typename Less, typename Func>
+        bool find_with( K const& key, Less pred, Func f )
+        {
+            return base_class::find_with( key, pred, f );
+        }
+
+        /// Checks whether the map contains \p key
+        /**
+            The function searches the item with key equal to \p key
+            and returns \p true if it is found, and \p false otherwise.
+
+            The function applies RCU lock internally.
+        */
+        template <typename K>
+        bool contains( K const& key )
+        {
+            return base_class::contains( key );
+        }
+
+        /// Checks whether the map contains \p key using \p pred predicate for searching
+        /**
+            The function is similar to <tt>contains( key )</tt> but \p pred is used for key comparing.
+            \p Less functor has the interface like \p std::less.
+            \p Less must imply the same element order as the comparator used for building the set.
+        */
+        template <typename K, typename Less>
+        bool contains( K const& key, Less pred )
+        {
+            return base_class::contains( key, pred );
+        }
+
+        /// Clears the map
+        void clear()
+        {
+            base_class::clear();
+        }
+
+        /// Checks if the map is empty
+        bool empty() const
+        {
+            return base_class::empty();
+        }
+
+        /// Returns item count in the map
+        /**
+            Only leaf nodes containing user data are counted.
+
+            The value returned depends on item counter type provided by \p Traits template parameter.
+            If it is \p atomicity::empty_item_counter this function always returns 0.
+
+            The function is not suitable for checking the tree emptiness, use \p empty()
+            member function for this purpose.
+        */
+        size_t size() const
+        {
+            return base_class::size();
+        }
+
+        /// Returns const reference to internal statistics
+        stat const& statistics() const
+        {
+            return base_class::statistics();
+        }
+
+        /// Returns reference to \p sync_monitor object
+        sync_monitor& monitor()
+        {
+            return base_class::monitor();
+        }
+        //@cond
+        sync_monitor const& monitor() const
+        {
+            return base_class::monitor();
+        }
+        //@endcond
+
+        /// Checks internal consistency (not atomic, not thread-safe)
+        /**
+            The debugging function to check internal consistency of the tree.
+        */
+        bool check_consistency() const
+        {
+            return base_class::check_consistency();
+        }
+
+        /// Checks internal consistency (not atomic, not thread-safe)
+        /**
+            The debugging function to check internal consistency of the tree.
+            The functor \p Func is called if a violation of internal tree structure
+            is found:
+            \code
+            struct functor {
+                void operator()( size_t nLevel, size_t hLeft, size_t hRight );
+            };
+            \endcode
+            where
+            - \p nLevel - the level where the violation is found
+            - \p hLeft - the height of left subtree
+            - \p hRight - the height of right subtree
+
+            The functor is called for each violation found.
+        */
+        template <typename Func>
+        bool check_consistency( Func f ) const
+        {
+            return base_class::check_consistency( f );
+        }
+    };
+}} // namespace cds::container
+
+#endif // #ifndef CDSLIB_CONTAINER_IMPL_BRONSON_AVLTREE_MAP_RCU_H

extern/libcds/cds/container/cuckoo_map.h

@@ -0,0 +1,747 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_CUCKOO_MAP_H
+#define CDSLIB_CONTAINER_CUCKOO_MAP_H
+
+#include <cds/container/details/cuckoo_base.h>
+#include <cds/details/binary_functor_wrapper.h>
+
+namespace cds { namespace container {
+
+    //@cond
+    namespace details {
+        template <typename Key, typename T, typename Traits>
+        struct make_cuckoo_map
+        {
+            typedef Key key_type;    ///< key type
+            typedef T   mapped_type; ///< type of value stored in the map
+            typedef std::pair<key_type const, mapped_type>   value_type;   ///< Pair type
+
+            typedef Traits original_traits;
+            typedef typename original_traits::probeset_type probeset_type;
+            static bool const store_hash = original_traits::store_hash;
+            static unsigned int const store_hash_count = store_hash ? ((unsigned int) std::tuple_size< typename original_traits::hash::hash_tuple_type >::value) : 0;
+
+            struct node_type: public intrusive::cuckoo::node<probeset_type, store_hash_count>
+            {
+                value_type  m_val;
+
+                template <typename K>
+                node_type( K const& key )
+                    : m_val( std::make_pair( key_type(key), mapped_type()))
+                {}
+
+                template <typename K, typename Q>
+                node_type( K const& key, Q const& v )
+                    : m_val( std::make_pair( key_type(key), mapped_type(v)))
+                {}
+
+                template <typename K, typename... Args>
+                node_type( K&& key, Args&&... args )
+                    : m_val( std::forward<K>(key), std::move( mapped_type(std::forward<Args>(args)...)))
+                {}
+            };
+
+            struct key_accessor {
+                key_type const& operator()( node_type const& node ) const
+                {
+                    return node.m_val.first;
+                }
+            };
+
+            struct intrusive_traits: public original_traits
+            {
+                typedef intrusive::cuckoo::base_hook<
+                    cds::intrusive::cuckoo::probeset_type< probeset_type >
+                    ,cds::intrusive::cuckoo::store_hash< store_hash_count >
+                >  hook;
+
+                typedef cds::intrusive::cuckoo::traits::disposer   disposer;
+
+                typedef typename std::conditional<
+                    std::is_same< typename original_traits::equal_to, opt::none >::value
+                    , opt::none
+                    , cds::details::predicate_wrapper< node_type, typename original_traits::equal_to, key_accessor >
+                >::type equal_to;
+
+                typedef typename std::conditional<
+                    std::is_same< typename original_traits::compare, opt::none >::value
+                    , opt::none
+                    , cds::details::compare_wrapper< node_type, typename original_traits::compare, key_accessor >
+                >::type compare;
+
+                typedef typename std::conditional<
+                    std::is_same< typename original_traits::less, opt::none >::value
+                    ,opt::none
+                    ,cds::details::predicate_wrapper< node_type, typename original_traits::less, key_accessor >
+                >::type less;
+
+                typedef opt::details::hash_list_wrapper< typename original_traits::hash, node_type, key_accessor >    hash;
+            };
+
+            typedef intrusive::CuckooSet< node_type, intrusive_traits > type;
+        };
+    }   // namespace details
+    //@endcond
+
+    /// Cuckoo hash map
+    /** @ingroup cds_nonintrusive_map
+
+        Source
+            - [2007] M.Herlihy, N.Shavit, M.Tzafrir "Concurrent Cuckoo Hashing. Technical report"
+            - [2008] Maurice Herlihy, Nir Shavit "The Art of Multiprocessor Programming"
+
+        <b>About Cuckoo hashing</b>
+
+            [From "The Art of Multiprocessor Programming"]
+            <a href="https://en.wikipedia.org/wiki/Cuckoo_hashing">Cuckoo hashing</a> is a hashing algorithm in which a newly added item displaces any earlier item
+            occupying the same slot. For brevity, a table is a k-entry array of items. For a hash set f size
+            N = 2k we use a two-entry array of tables, and two independent hash functions,
+            <tt> h0, h1: KeyRange -> 0,...,k-1</tt>
+            mapping the set of possible keys to entries in he array. To test whether a value \p x is in the set,
+            <tt>find(x)</tt> tests whether either <tt>table[0][h0(x)]</tt> or <tt>table[1][h1(x)]</tt> is
+            equal to \p x. Similarly, <tt>erase(x)</tt>checks whether \p x is in either <tt>table[0][h0(x)]</tt>
+            or <tt>table[1][h1(x)]</tt>, ad removes it if found.
+
+            The <tt>insert(x)</tt> successively "kicks out" conflicting items until every key has a slot.
+            To add \p x, the method swaps \p x with \p y, the current occupant of <tt>table[0][h0(x)]</tt>.
+            If the prior value was \p nullptr, it is done. Otherwise, it swaps the newly nest-less value \p y
+            for the current occupant of <tt>table[1][h1(y)]</tt> in the same way. As before, if the prior value
+            was \p nullptr, it is done. Otherwise, the method continues swapping entries (alternating tables)
+            until it finds an empty slot. We might not find an empty slot, either because the table is full,
+            or because the sequence of displacement forms a cycle. We therefore need an upper limit on the
+            number of successive displacements we are willing to undertake. When this limit is exceeded,
+            we resize the hash table, choose new hash functions and start over.
+
+            For concurrent cuckoo hashing, rather than organizing the set as a two-dimensional table of
+            items, we use two-dimensional table of probe sets, where a probe set is a constant-sized set
+            of items with the same hash code. Each probe set holds at most \p PROBE_SIZE items, but the algorithm
+            tries to ensure that when the set is quiescent (i.e no method call in progress) each probe set
+            holds no more than <tt>THRESHOLD < PROBE_SET</tt> items. While method calls are in-flight, a probe
+            set may temporarily hold more than \p THRESHOLD but never more than \p PROBE_SET items.
+
+            In current implementation, a probe set can be defined either as a (single-linked) list
+            or as a fixed-sized vector, optionally ordered.
+
+            In description above two-table cuckoo hashing (<tt>k = 2</tt>) has been considered.
+            We can generalize this approach for <tt>k >= 2</tt> when we have \p k hash functions
+            <tt>h[0], ... h[k-1]</tt> and \p k tables <tt>table[0], ... table[k-1]</tt>.
+
+            The search in probe set is linear, the complexity is <tt> O(PROBE_SET) </tt>.
+            The probe set may be ordered or not. Ordered probe set can be a little better since
+            the average search complexity is <tt>O(PROBE_SET/2)</tt>.
+            However, the overhead of sorting can eliminate a gain of ordered search.
+
+            The probe set is ordered if \p compare or \p less is specified in \p Traits
+            template parameter. Otherwise, the probe set is unordered and \p Traits must contain
+            \p equal_to predicate.
+
+        Template arguments:
+        - \p Key - key type
+        - \p T - the type stored in the map.
+        - \p Traits - map traits, default is \p cuckoo::traits.
+            It is possible to declare option-based set with \p cuckoo::make_traits metafunction
+            result as \p Traits template argument.
+
+       <b>Examples</b>
+
+       Declares cuckoo mapping from \p std::string to struct \p foo.
+       For cuckoo hashing we should provide at least two hash functions:
+       \code
+        struct hash1 {
+            size_t operator()(std::string const& s) const
+            {
+                return cds::opt::v::hash<std::string>( s );
+            }
+        };
+
+        struct hash2: private hash1 {
+            size_t operator()(std::string const& s) const
+            {
+                size_t h = ~( hash1::operator()(s));
+                return ~h + 0x9e3779b9 + (h << 6) + (h >> 2);
+            }
+        };
+       \endcode
+
+        Cuckoo-map with list-based unordered probe set and storing hash values
+        \code
+        #include <cds/container/cuckoo_map.h>
+
+        // Declare type traits
+        struct my_traits: public cds::container::cuckoo::traits
+        {
+            typedef std::equal_to< std::string > equal_to;
+            typedef std::tuple< hash1, hash2 >  hash;
+
+            static bool const store_hash = true;
+        };
+
+        // Declare CuckooMap type
+        typedef cds::container::CuckooMap< std::string, foo, my_traits > my_cuckoo_map;
+
+        // Equal option-based declaration
+        typedef cds::container::CuckooMap< std::string, foo,
+            cds::container::cuckoo::make_traits<
+                cds::opt::hash< std::tuple< hash1, hash2 > >
+                ,cds::opt::equal_to< std::equal_to< std::string > >
+                ,cds::container::cuckoo::store_hash< true >
+            >::type
+        > opt_cuckoo_map;
+        \endcode
+
+        If we provide \p less functor instead of \p equal_to
+        we get as a result a cuckoo map with ordered probe set that may improve
+        performance.
+        Example for ordered vector-based probe-set:
+
+        \code
+        #include <cds/container/cuckoo_map.h>
+
+        // Declare type traits
+        // We use a vector of capacity 4 as probe-set container and store hash values in the node
+        struct my_traits: public cds::container::cuckoo::traits
+        {
+            typedef std::less< std::string > less;
+            typedef std::tuple< hash1, hash2 >  hash;
+            typedef cds::container::cuckoo::vector<4> probeset_type;
+
+            static bool const store_hash = true;
+        };
+
+        // Declare CuckooMap type
+        typedef cds::container::CuckooMap< std::string, foo, my_traits > my_cuckoo_map;
+
+        // Equal option-based declaration
+        typedef cds::container::CuckooMap< std::string, foo,
+            cds::container::cuckoo::make_traits<
+                cds::opt::hash< std::tuple< hash1, hash2 > >
+                ,cds::opt::less< std::less< std::string > >
+                ,cds::container::cuckoo::probeset_type< cds::container::cuckoo::vector<4> >
+                ,cds::container::cuckoo::store_hash< true >
+            >::type
+        > opt_cuckoo_map;
+        \endcode
+
+    */
+    template <typename Key, typename T, typename Traits = cuckoo::traits>
+    class CuckooMap:
+#ifdef CDS_DOXYGEN_INVOKED
+        protected intrusive::CuckooSet< std::pair< Key const, T>, Traits>
+#else
+        protected details::make_cuckoo_map<Key, T, Traits>::type
+#endif
+    {
+        //@cond
+        typedef details::make_cuckoo_map<Key, T, Traits>    maker;
+        typedef typename maker::type  base_class;
+        //@endcond
+    public:
+        typedef Key     key_type;    ///< key type
+        typedef T       mapped_type; ///< value type stored in the container
+        typedef std::pair<key_type const, mapped_type> value_type;   ///< Key-value pair type stored in the map
+        typedef Traits  traits;     ///< Map traits
+
+        typedef typename traits::hash                 hash;            ///< hash functor tuple wrapped for internal use
+        typedef typename base_class::hash_tuple_type  hash_tuple_type; ///< hash tuple type
+
+        typedef typename base_class::mutex_policy mutex_policy; ///< Concurrent access policy, see \p cuckoo::traits::mutex_policy
+        typedef typename base_class::stat         stat;         ///< internal statistics type
+
+        static bool const c_isSorted = base_class::c_isSorted; ///< whether the probe set should be ordered
+        static size_t const c_nArity = base_class::c_nArity;   ///< the arity of cuckoo hashing: the number of hash functors provided; minimum 2.
+
+        typedef typename base_class::key_equal_to key_equal_to; ///< Key equality functor; used only for unordered probe-set
+
+        typedef typename base_class::key_comparator  key_comparator; ///< key comparing functor based on opt::compare and opt::less option setter. Used only for ordered probe set
+
+        typedef typename base_class::allocator     allocator; ///< allocator type used for internal bucket table allocations
+
+        /// Node allocator type
+        typedef typename std::conditional<
+            std::is_same< typename traits::node_allocator, opt::none >::value,
+            allocator,
+            typename traits::node_allocator
+        >::type node_allocator;
+
+        /// item counter type
+        typedef typename traits::item_counter  item_counter;
+
+    protected:
+        //@cond
+        typedef typename base_class::scoped_cell_lock   scoped_cell_lock;
+        typedef typename base_class::scoped_full_lock   scoped_full_lock;
+        typedef typename base_class::scoped_resize_lock scoped_resize_lock;
+        typedef typename maker::key_accessor            key_accessor;
+
+        typedef typename base_class::value_type         node_type;
+        typedef cds::details::Allocator< node_type, node_allocator >    cxx_node_allocator;
+        //@endcond
+
+    public:
+        static unsigned int const   c_nDefaultProbesetSize = base_class::c_nDefaultProbesetSize; ///< default probeset size
+        static size_t const         c_nDefaultInitialSize = base_class::c_nDefaultInitialSize;   ///< default initial size
+        static unsigned int const   c_nRelocateLimit = base_class::c_nRelocateLimit;             ///< Count of attempts to relocate before giving up
+
+    protected:
+        //@cond
+        template <typename K>
+        static node_type * alloc_node( K const& key )
+        {
+            return cxx_node_allocator().New( key );
+        }
+        template <typename K, typename... Args>
+        static node_type * alloc_node( K&& key, Args&&... args )
+        {
+            return cxx_node_allocator().MoveNew( std::forward<K>( key ), std::forward<Args>(args)... );
+        }
+
+        static void free_node( node_type * pNode )
+        {
+            cxx_node_allocator().Delete( pNode );
+        }
+        //@endcond
+
+    protected:
+        //@cond
+        struct node_disposer {
+            void operator()( node_type * pNode )
+            {
+                free_node( pNode );
+            }
+        };
+
+        typedef std::unique_ptr< node_type, node_disposer >     scoped_node_ptr;
+
+        //@endcond
+
+    public:
+        /// Default constructor
+        /**
+            Initial size = \ref c_nDefaultInitialSize
+
+            Probe set size:
+            - \ref c_nDefaultProbesetSize if \p probeset_type is \p cuckoo::list
+            - \p Capacity if \p probeset_type is <tt> cuckoo::vector<Capacity> </tt>
+
+            Probe set threshold = probe set size - 1
+        */
+        CuckooMap()
+        {}
+
+        /// Constructs an object with given probe set size and threshold
+        /**
+            If probe set type is <tt> cuckoo::vector<Capacity> </tt> vector
+            then \p nProbesetSize should be equal to vector's \p Capacity.
+        */
+        CuckooMap(
+            size_t nInitialSize                     ///< Initial map size; if 0 - use default initial size \ref c_nDefaultInitialSize
+            , unsigned int nProbesetSize            ///< probe set size
+            , unsigned int nProbesetThreshold = 0   ///< probe set threshold, <tt>nProbesetThreshold < nProbesetSize</tt>. If 0, nProbesetThreshold = nProbesetSize - 1
+        )
+        : base_class( nInitialSize, nProbesetSize, nProbesetThreshold )
+        {}
+
+        /// Constructs an object with given hash functor tuple
+        /**
+            The probe set size and threshold are set as default, see CuckooSet()
+        */
+        CuckooMap(
+            hash_tuple_type const& h    ///< hash functor tuple of type <tt>std::tuple<H1, H2, ... Hn></tt> where <tt> n == \ref c_nArity </tt>
+        )
+        : base_class( h )
+        {}
+
+        /// Constructs a map with given probe set properties and hash functor tuple
+        /**
+            If probe set type is <tt> cuckoo::vector<Capacity> </tt> vector
+            then \p nProbesetSize should be equal to vector's \p Capacity.
+        */
+        CuckooMap(
+            size_t nInitialSize                 ///< Initial map size; if 0 - use default initial size \ref c_nDefaultInitialSize
+            , unsigned int nProbesetSize        ///< probe set size
+            , unsigned int nProbesetThreshold   ///< probe set threshold, <tt>nProbesetThreshold < nProbesetSize</tt>. If 0, nProbesetThreshold = nProbesetSize - 1
+            , hash_tuple_type const& h    ///< hash functor tuple of type <tt>std::tuple<H1, H2, ... Hn></tt> where <tt> n == \ref c_nArity </tt>
+        )
+        : base_class( nInitialSize, nProbesetSize, nProbesetThreshold, h )
+        {}
+
+        /// Constructs a map with given hash functor tuple (move semantics)
+        /**
+            The probe set size and threshold are set as default, see CuckooSet()
+        */
+        CuckooMap(
+            hash_tuple_type&& h     ///< hash functor tuple of type <tt>std::tuple<H1, H2, ... Hn></tt> where <tt> n == \ref c_nArity </tt>
+        )
+        : base_class( std::forward<hash_tuple_type>(h))
+        {}
+
+        /// Constructs a map with given probe set properties and hash functor tuple (move semantics)
+        /**
+            If probe set type is <tt> cuckoo::vector<Capacity> </tt> vector
+            then \p nProbesetSize should be equal to vector's \p Capacity.
+        */
+        CuckooMap(
+            size_t nInitialSize                 ///< Initial map size; if 0 - use default initial size \ref c_nDefaultInitialSize
+            , unsigned int nProbesetSize        ///< probe set size
+            , unsigned int nProbesetThreshold   ///< probe set threshold, <tt>nProbesetThreshold < nProbesetSize</tt>. If 0, nProbesetThreshold = nProbesetSize - 1
+            , hash_tuple_type&& h    ///< hash functor tuple of type <tt>std::tuple<H1, H2, ... Hn></tt> where <tt> n == \ref c_nArity </tt>
+        )
+        : base_class( nInitialSize, nProbesetSize, nProbesetThreshold, std::forward<hash_tuple_type>(h))
+        {}
+
+        /// Destructor clears the map
+        ~CuckooMap()
+        {
+            clear();
+        }
+
+    public:
+        /// Inserts new node with key and default value
+        /**
+            The function creates a node with \p key and default value, and then inserts the node created into the map.
+
+            Preconditions:
+            - The \ref key_type should be constructible from a value of type \p K.
+                In trivial case, \p K is equal to \ref key_type.
+            - The \ref mapped_type should be default-constructible.
+
+            Returns \p true if inserting successful, \p false otherwise.
+        */
+        template <typename K>
+        bool insert( K const& key )
+        {
+            return insert_with( key, [](value_type&){} );
+        }
+
+        /// Inserts new node
+        /**
+            The function creates a node with copy of \p val value
+            and then inserts the node created into the map.
+
+            Preconditions:
+            - The \ref key_type should be constructible from \p key of type \p K.
+            - The \ref value_type should be constructible from \p val of type \p V.
+
+            Returns \p true if \p val is inserted into the set, \p false otherwise.
+        */
+        template <typename K, typename V>
+        bool insert( K const& key, V const& val )
+        {
+            return insert_with( key, [&val](value_type& item) { item.second = val ; } );
+        }
+
+        /// Inserts new node and initialize it by a functor
+        /**
+            This function inserts new node with key \p key and if inserting is successful then it calls
+            \p func functor with signature
+            \code
+                struct functor {
+                    void operator()( value_type& item );
+                };
+            \endcode
+
+            The argument \p item of user-defined functor \p func is the reference
+            to the map's item inserted:
+                - <tt>item.first</tt> is a const reference to item's key that cannot be changed.
+                - <tt>item.second</tt> is a reference to item's value that may be changed.
+
+            The key_type should be constructible from value of type \p K.
+
+            The function allows to split creating of new item into two part:
+            - create item from \p key;
+            - insert new item into the map;
+            - if inserting is successful, initialize the value of item by calling \p func functor
+
+            This can be useful if complete initialization of object of \p value_type is heavyweight and
+            it is preferable that the initialization should be completed only if inserting is successful.
+        */
+        template <typename K, typename Func>
+        bool insert_with( const K& key, Func func )
+        {
+            scoped_node_ptr pNode( alloc_node( key ));
+            if ( base_class::insert( *pNode, [&func]( node_type& item ) { func( item.m_val ); } )) {
+                pNode.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// For key \p key inserts data of type \ref value_type constructed with <tt>std::forward<Args>(args)...</tt>
+        /**
+            Returns \p true if inserting successful, \p false otherwise.
+        */
+        template <typename K, typename... Args>
+        bool emplace( K&& key, Args&&... args )
+        {
+            scoped_node_ptr pNode( alloc_node( std::forward<K>(key), std::forward<Args>(args)... ));
+            if ( base_class::insert( *pNode )) {
+                pNode.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// Updates the node
+        /**
+            The operation performs inserting or changing data with lock-free manner.
+
+            If \p key is not found in the map, then \p key is inserted iff \p bAllowInsert is \p true.
+            Otherwise, the functor \p func is called with item found.
+            The functor \p func signature is:
+            \code
+                struct my_functor {
+                    void operator()( bool bNew, value_type& item );
+                };
+            \endcode
+            with arguments:
+            - \p bNew - \p true if the item has been inserted, \p false otherwise
+            - \p item - an item of the map for \p key
+
+            Returns std::pair<bool, bool> where \p first is \p true if operation is successful,
+            i.e. the node has been inserted or updated,
+            \p second is \p true if new item has been added or \p false if the item with \p key
+            already exists.
+        */
+        template <typename K, typename Func>
+        std::pair<bool, bool> update( K const& key, Func func, bool bAllowInsert = true )
+        {
+            scoped_node_ptr pNode( alloc_node( key ));
+            std::pair<bool, bool> res = base_class::update( *pNode,
+                [&func](bool bNew, node_type& item, node_type const& ){ func( bNew, item.m_val ); },
+                bAllowInsert
+            );
+            if ( res.first && res.second )
+                pNode.release();
+            return res;
+        }
+        //@cond
+        template <typename K, typename Func>
+        CDS_DEPRECATED("ensure() is deprecated, use update()")
+        std::pair<bool, bool> ensure( K const& key, Func func )
+        {
+            return update( key, func, true );
+        }
+        //@endcond
+
+        /// Delete \p key from the map
+        /** \anchor cds_nonintrusive_CuckooMap_erase_val
+
+            Return \p true if \p key is found and deleted, \p false otherwise
+        */
+        template <typename K>
+        bool erase( K const& key )
+        {
+            node_type * pNode = base_class::erase(key);
+            if ( pNode ) {
+                free_node( pNode );
+                return true;
+            }
+            return false;
+        }
+
+        /// Deletes the item from the list using \p pred predicate for searching
+        /**
+            The function is an analog of \ref cds_nonintrusive_CuckooMap_erase_val "erase(Q const&)"
+            but \p pred is used for key comparing.
+            If cuckoo map is ordered, then \p Predicate should have the interface and semantics like \p std::less.
+            If cuckoo map is unordered, then \p Predicate should have the interface and semantics like \p std::equal_to.
+            \p Predicate must imply the same element order as the comparator used for building the map.
+        */
+        template <typename K, typename Predicate>
+        bool erase_with( K const& key, Predicate pred )
+        {
+            CDS_UNUSED( pred );
+            node_type * pNode = base_class::erase_with(key, cds::details::predicate_wrapper<node_type, Predicate, key_accessor>());
+            if ( pNode ) {
+                free_node( pNode );
+                return true;
+            }
+            return false;
+        }
+
+        /// Delete \p key from the map
+        /** \anchor cds_nonintrusive_CuckooMap_erase_func
+
+            The function searches an item with key \p key, calls \p f functor
+            and deletes the item. If \p key is not found, the functor is not called.
+
+            The functor \p Func interface:
+            \code
+            struct extractor {
+                void operator()(value_type& item) { ... }
+            };
+            \endcode
+
+            Return \p true if key is found and deleted, \p false otherwise
+
+            See also: \ref erase
+        */
+        template <typename K, typename Func>
+        bool erase( K const& key, Func f )
+        {
+            node_type * pNode = base_class::erase( key );
+            if ( pNode ) {
+                f( pNode->m_val );
+                free_node( pNode );
+                return true;
+            }
+            return false;
+        }
+
+        /// Deletes the item from the list using \p pred predicate for searching
+        /**
+            The function is an analog of \ref cds_nonintrusive_CuckooMap_erase_func "erase(Q const&, Func)"
+            but \p pred is used for key comparing.
+            If cuckoo map is ordered, then \p Predicate should have the interface and semantics like \p std::less.
+            If cuckoo map is unordered, then \p Predicate should have the interface and semantics like \p std::equal_to.
+            \p Predicate must imply the same element order as the comparator used for building the map.
+        */
+        template <typename K, typename Predicate, typename Func>
+        bool erase_with( K const& key, Predicate pred, Func f )
+        {
+            CDS_UNUSED( pred );
+            node_type * pNode = base_class::erase_with( key, cds::details::predicate_wrapper<node_type, Predicate, key_accessor>());
+            if ( pNode ) {
+                f( pNode->m_val );
+                free_node( pNode );
+                return true;
+            }
+            return false;
+        }
+
+        /// Find the key \p key
+        /** \anchor cds_nonintrusive_CuckooMap_find_func
+
+            The function searches the item with key equal to \p key and calls the functor \p f for item found.
+            The interface of \p Func functor is:
+            \code
+            struct functor {
+                void operator()( value_type& item );
+            };
+            \endcode
+            where \p item is the item found.
+
+            The functor may change \p item.second.
+
+            The function returns \p true if \p key is found, \p false otherwise.
+        */
+        template <typename K, typename Func>
+        bool find( K const& key, Func f )
+        {
+            return base_class::find( key, [&f](node_type& item, K const& ) { f( item.m_val );});
+        }
+
+        /// Find the key \p val using \p pred predicate for comparing
+        /**
+            The function is an analog of \ref cds_nonintrusive_CuckooMap_find_func "find(K const&, Func)"
+            but \p pred is used for key comparison.
+            If you use ordered cuckoo map, then \p Predicate should have the interface and semantics like \p std::less.
+            If you use unordered cuckoo map, then \p Predicate should have the interface and semantics like \p std::equal_to.
+            \p pred must imply the same element order as the comparator used for building the map.
+        */
+        template <typename K, typename Predicate, typename Func>
+        bool find_with( K const& key, Predicate pred, Func f )
+        {
+            CDS_UNUSED( pred );
+            return base_class::find_with( key, cds::details::predicate_wrapper<node_type, Predicate, key_accessor>(),
+                [&f](node_type& item, K const& ) { f( item.m_val );});
+        }
+
+        /// Checks whether the map contains \p key
+        /**
+            The function searches the item with key equal to \p key
+            and returns \p true if it is found, and \p false otherwise.
+        */
+        template <typename K>
+        bool contains( K const& key )
+        {
+            return base_class::contains( key );
+        }
+        //@cond
+        template <typename K>
+        CDS_DEPRECATED("the function is deprecated, use contains()")
+        bool find( K const& key )
+        {
+            return contains( key );
+        }
+        //@endcond
+
+        /// Checks whether the map contains \p key using \p pred predicate for searching
+        /**
+            The function is similar to <tt>contains( key )</tt> but \p pred is used for key comparing.
+            \p Less functor has the interface like \p std::less.
+            \p Less must imply the same element order as the comparator used for building the map.
+        */
+        template <typename K, typename Predicate>
+        bool contains( K const& key, Predicate pred )
+        {
+            CDS_UNUSED( pred );
+            return base_class::contains( key, cds::details::predicate_wrapper<node_type, Predicate, key_accessor>());
+        }
+        //@cond
+        template <typename K, typename Predicate>
+        CDS_DEPRECATED("the function is deprecated, use contains()")
+        bool find_with( K const& key, Predicate pred )
+        {
+            return contains( key, pred );
+        }
+        //@endcond
+
+        /// Clears the map
+        void clear()
+        {
+            base_class::clear_and_dispose( node_disposer());
+        }
+
+        /// Checks if the map is empty
+        /**
+            Emptiness is checked by item counting: if item count is zero then the map is empty.
+        */
+        bool empty() const
+        {
+            return base_class::empty();
+        }
+
+        /// Returns item count in the map
+        size_t size() const
+        {
+            return base_class::size();
+        }
+
+        /// Returns the size of hash table
+        /**
+            The hash table size is non-constant and can be increased via resizing.
+        */
+        size_t bucket_count() const
+        {
+            return base_class::bucket_count();
+        }
+
+        /// Returns lock array size
+        /**
+            The lock array size is constant.
+        */
+        size_t lock_count() const
+        {
+            return base_class::lock_count();
+        }
+
+        /// Returns const reference to internal statistics
+        stat const& statistics() const
+        {
+            return base_class::statistics();
+        }
+
+        /// Returns const reference to mutex policy internal statistics
+        typename mutex_policy::statistics_type const& mutex_policy_statistics() const
+        {
+            return base_class::mutex_policy_statistics();
+        }
+    };
+}}  // namespace cds::container
+
+#endif //#ifndef CDSLIB_CONTAINER_CUCKOO_MAP_H

extern/libcds/cds/container/cuckoo_set.h

@@ -0,0 +1,825 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_CUCKOO_SET_H
+#define CDSLIB_CONTAINER_CUCKOO_SET_H
+
+#include <cds/container/details/cuckoo_base.h>
+#include <cds/details/binary_functor_wrapper.h>
+
+namespace cds { namespace container {
+
+    //@cond
+    namespace details {
+        template <typename T, typename Traits>
+        struct make_cuckoo_set
+        {
+            typedef T value_type;
+            typedef Traits original_traits;
+            typedef typename original_traits::probeset_type probeset_type;
+            static bool const store_hash = original_traits::store_hash;
+            static unsigned int const store_hash_count = store_hash ? ((unsigned int) std::tuple_size< typename original_traits::hash::hash_tuple_type >::value) : 0;
+
+            struct node_type: public intrusive::cuckoo::node<probeset_type, store_hash_count>
+            {
+                value_type  m_val;
+
+                template <typename Q>
+                node_type( Q const& v )
+                    : m_val(v)
+                {}
+
+                template <typename... Args>
+                node_type( Args&&... args )
+                    : m_val( std::forward<Args>(args)...)
+                {}
+            };
+
+            struct value_accessor {
+                value_type const& operator()( node_type const& node ) const
+                {
+                    return node.m_val;
+                }
+            };
+
+            template <typename Pred, typename ReturnValue>
+            using predicate_wrapper = cds::details::binary_functor_wrapper< ReturnValue, Pred, node_type, value_accessor >;
+
+            struct intrusive_traits: public original_traits
+            {
+                typedef intrusive::cuckoo::base_hook<
+                    cds::intrusive::cuckoo::probeset_type< probeset_type >
+                    ,cds::intrusive::cuckoo::store_hash< store_hash_count >
+                >  hook;
+
+                typedef cds::intrusive::cuckoo::traits::disposer   disposer;
+
+                typedef typename std::conditional<
+                    std::is_same< typename original_traits::equal_to, opt::none >::value
+                    , opt::none
+                    , predicate_wrapper< typename original_traits::equal_to, bool >
+                >::type equal_to;
+
+                typedef typename std::conditional<
+                    std::is_same< typename original_traits::compare, opt::none >::value
+                    , opt::none
+                    , predicate_wrapper< typename original_traits::compare, int >
+                >::type compare;
+
+                typedef typename std::conditional<
+                    std::is_same< typename original_traits::less, opt::none >::value
+                    ,opt::none
+                    ,predicate_wrapper< typename original_traits::less, bool >
+                >::type less;
+
+                typedef opt::details::hash_list_wrapper< typename original_traits::hash, node_type, value_accessor >    hash;
+            };
+
+            typedef intrusive::CuckooSet< node_type, intrusive_traits > type;
+        };
+    } // namespace details
+    //@endcond
+
+    /// Cuckoo hash set
+    /** @ingroup cds_nonintrusive_set
+
+        Source
+            - [2007] M.Herlihy, N.Shavit, M.Tzafrir "Concurrent Cuckoo Hashing. Technical report"
+            - [2008] Maurice Herlihy, Nir Shavit "The Art of Multiprocessor Programming"
+
+        <b>About Cuckoo hashing</b>
+
+            [From "The Art of Multiprocessor Programming"]
+            <a href="https://en.wikipedia.org/wiki/Cuckoo_hashing">Cuckoo hashing</a> is a hashing algorithm in which a newly added item displaces any earlier item
+            occupying the same slot. For brevity, a table is a k-entry array of items. For a hash set f size
+            N = 2k we use a two-entry array of tables, and two independent hash functions,
+            <tt> h0, h1: KeyRange -> 0,...,k-1</tt>
+            mapping the set of possible keys to entries in he array. To test whether a value \p x is in the set,
+            <tt>find(x)</tt> tests whether either <tt>table[0][h0(x)]</tt> or <tt>table[1][h1(x)]</tt> is
+            equal to \p x. Similarly, <tt>erase(x)</tt>checks whether \p x is in either <tt>table[0][h0(x)]</tt>
+            or <tt>table[1][h1(x)]</tt>, ad removes it if found.
+
+            The <tt>insert(x)</tt> successively "kicks out" conflicting items until every key has a slot.
+            To add \p x, the method swaps \p x with \p y, the current occupant of <tt>table[0][h0(x)]</tt>.
+            If the prior value was \p nullptr, it is done. Otherwise, it swaps the newly nest-less value \p y
+            for the current occupant of <tt>table[1][h1(y)]</tt> in the same way. As before, if the prior value
+            was \p nullptr, it is done. Otherwise, the method continues swapping entries (alternating tables)
+            until it finds an empty slot. We might not find an empty slot, either because the table is full,
+            or because the sequence of displacement forms a cycle. We therefore need an upper limit on the
+            number of successive displacements we are willing to undertake. When this limit is exceeded,
+            we resize the hash table, choose new hash functions and start over.
+
+            For concurrent cuckoo hashing, rather than organizing the set as a two-dimensional table of
+            items, we use two-dimensional table of probe sets, where a probe set is a constant-sized set
+            of items with the same hash code. Each probe set holds at most \p PROBE_SIZE items, but the algorithm
+            tries to ensure that when the set is quiescent (i.e no method call in progress) each probe set
+            holds no more than <tt>THRESHOLD < PROBE_SET</tt> items. While method calls are in-flight, a probe
+            set may temporarily hold more than \p THRESHOLD but never more than \p PROBE_SET items.
+
+            In current implementation, a probe set can be defined either as a (single-linked) list
+            or as a fixed-sized vector, optionally ordered.
+
+            In description above two-table cuckoo hashing (<tt>k = 2</tt>) has been considered.
+            We can generalize this approach for <tt>k >= 2</tt> when we have \p k hash functions
+            <tt>h[0], ... h[k-1]</tt> and \p k tables <tt>table[0], ... table[k-1]</tt>.
+
+            The search in probe set is linear, the complexity is <tt> O(PROBE_SET) </tt>.
+            The probe set may be ordered or not. Ordered probe set can be a little better since
+            the average search complexity is <tt>O(PROBE_SET/2)</tt>.
+            However, the overhead of sorting can eliminate a gain of ordered search.
+
+            The probe set is ordered if \p compare or \p less is specified in \p Traits
+            template parameter. Otherwise, the probe set is unordered and \p Traits must contain
+            \p equal_to predicate.
+
+        Template arguments:
+        - \p T - the type stored in the set.
+        - \p Traits - type traits. See cuckoo::traits for explanation.
+            It is possible to declare option-based set with cuckoo::make_traits metafunction result as \p Traits template argument.
+
+        <b>Examples</b>
+
+        Cuckoo-set with list-based unordered probe set and storing hash values
+        \code
+        #include <cds/container/cuckoo_set.h>
+
+        // Data stored in cuckoo set
+        struct my_data
+        {
+            // key field
+            std::string     strKey;
+
+            // other data
+            // ...
+        };
+
+        // Provide equal_to functor for my_data since we will use unordered probe-set
+        struct my_data_equal_to {
+            bool operator()( const my_data& d1, const my_data& d2 ) const
+            {
+                return d1.strKey.compare( d2.strKey ) == 0;
+            }
+
+            bool operator()( const my_data& d, const std::string& s ) const
+            {
+                return d.strKey.compare(s) == 0;
+            }
+
+            bool operator()( const std::string& s, const my_data& d ) const
+            {
+                return s.compare( d.strKey ) == 0;
+            }
+        };
+
+        // Provide two hash functor for my_data
+        struct hash1 {
+            size_t operator()(std::string const& s) const
+            {
+                return cds::opt::v::hash<std::string>( s );
+            }
+            size_t operator()( my_data const& d ) const
+            {
+                return (*this)( d.strKey );
+            }
+        };
+
+        struct hash2: private hash1 {
+            size_t operator()(std::string const& s) const
+            {
+                size_t h = ~( hash1::operator()(s));
+                return ~h + 0x9e3779b9 + (h << 6) + (h >> 2);
+            }
+            size_t operator()( my_data const& d ) const
+            {
+                return (*this)( d.strKey );
+            }
+        };
+
+        // Declare type traits
+        struct my_traits: public cds::container::cuckoo::traits
+        {
+            typedef my_data_equa_to equal_to;
+            typedef std::tuple< hash1, hash2 >  hash;
+
+            static bool const store_hash = true;
+        };
+
+        // Declare CuckooSet type
+        typedef cds::container::CuckooSet< my_data, my_traits > my_cuckoo_set;
+
+        // Equal option-based declaration
+        typedef cds::container::CuckooSet< my_data,
+            cds::container::cuckoo::make_traits<
+                cds::opt::hash< std::tuple< hash1, hash2 > >
+                ,cds::opt::equal_to< my_data_equal_to >
+                ,cds::container::cuckoo::store_hash< true >
+            >::type
+        > opt_cuckoo_set;
+        \endcode
+
+        If we provide \p compare function instead of \p equal_to for \p my_data
+        we get as a result a cuckoo set with ordered probe set that may improve
+        performance.
+        Example for ordered vector-based probe-set:
+
+        \code
+        #include <cds/container/cuckoo_set.h>
+
+        // Data stored in cuckoo set
+        struct my_data
+        {
+            // key field
+            std::string     strKey;
+
+            // other data
+            // ...
+        };
+
+        // Provide compare functor for my_data since we want to use ordered probe-set
+        struct my_data_compare {
+            int operator()( const my_data& d1, const my_data& d2 ) const
+            {
+                return d1.strKey.compare( d2.strKey );
+            }
+
+            int operator()( const my_data& d, const std::string& s ) const
+            {
+                return d.strKey.compare(s);
+            }
+
+            int operator()( const std::string& s, const my_data& d ) const
+            {
+                return s.compare( d.strKey );
+            }
+        };
+
+        // Provide two hash functor for my_data
+        struct hash1 {
+            size_t operator()(std::string const& s) const
+            {
+                return cds::opt::v::hash<std::string>( s );
+            }
+            size_t operator()( my_data const& d ) const
+            {
+                return (*this)( d.strKey );
+            }
+        };
+
+        struct hash2: private hash1 {
+            size_t operator()(std::string const& s) const
+            {
+                size_t h = ~( hash1::operator()(s));
+                return ~h + 0x9e3779b9 + (h << 6) + (h >> 2);
+            }
+            size_t operator()( my_data const& d ) const
+            {
+                return (*this)( d.strKey );
+            }
+        };
+
+        // Declare type traits
+        // We use a vector of capacity 4 as probe-set container and store hash values in the node
+        struct my_traits: public cds::container::cuckoo::traits
+        {
+            typedef my_data_compare compare;
+            typedef std::tuple< hash1, hash2 >  hash;
+            typedef cds::container::cuckoo::vector<4> probeset_type;
+
+            static bool const store_hash = true;
+        };
+
+        // Declare CuckooSet type
+        typedef cds::container::CuckooSet< my_data, my_traits > my_cuckoo_set;
+
+        // Equal option-based declaration
+        typedef cds::container::CuckooSet< my_data,
+            cds::container::cuckoo::make_traits<
+                cds::opt::hash< std::tuple< hash1, hash2 > >
+                ,cds::opt::compare< my_data_compare >
+                ,cds::container::cuckoo::probeset_type< cds::container::cuckoo::vector<4> >
+                ,cds::container::cuckoo::store_hash< true >
+            >::type
+        > opt_cuckoo_set;
+        \endcode
+
+    */
+    template <typename T, typename Traits = cuckoo::traits>
+    class CuckooSet:
+#ifdef CDS_DOXYGEN_INVOKED
+        protected intrusive::CuckooSet<T, Traits>
+#else
+        protected details::make_cuckoo_set<T, Traits>::type
+#endif
+    {
+        //@cond
+        typedef details::make_cuckoo_set<T, Traits> maker;
+        typedef typename maker::type  base_class;
+        //@endcond
+
+    public:
+        typedef T       value_type  ;   ///< value type stored in the container
+        typedef Traits  traits     ;   ///< traits
+
+        typedef typename traits::hash                 hash;   ///< hash functor tuple wrapped for internal use
+        typedef typename base_class::hash_tuple_type  hash_tuple_type;   ///< Type of hash tuple
+
+        typedef typename base_class::mutex_policy mutex_policy; ///< Concurrent access policy, see cuckoo::traits::mutex_policy
+        typedef typename base_class::stat         stat;         ///< internal statistics type
+
+
+        static bool const c_isSorted = base_class::c_isSorted; ///< whether the probe set should be ordered
+        static size_t const c_nArity = base_class::c_nArity;   ///< the arity of cuckoo hashing: the number of hash functors provided; minimum 2.
+
+        typedef typename base_class::key_equal_to key_equal_to; ///< Key equality functor; used only for unordered probe-set
+
+        typedef typename base_class::key_comparator  key_comparator; ///< key comparing functor based on \p Traits::compare and \p Traits::less option setter. Used only for ordered probe set
+
+        typedef typename base_class::allocator     allocator; ///< allocator type used for internal bucket table allocations
+
+        /// Node allocator type
+        typedef typename std::conditional<
+            std::is_same< typename traits::node_allocator, opt::none >::value,
+            allocator,
+            typename traits::node_allocator
+        >::type node_allocator;
+
+        /// item counter type
+        typedef typename traits::item_counter  item_counter;
+
+    protected:
+        //@cond
+        typedef typename base_class::value_type         node_type;
+        typedef cds::details::Allocator< node_type, node_allocator >    cxx_node_allocator;
+        //@endcond
+
+    public:
+        static unsigned int const   c_nDefaultProbesetSize = base_class::c_nDefaultProbesetSize; ///< default probeset size
+        static size_t const         c_nDefaultInitialSize = base_class::c_nDefaultInitialSize;   ///< default initial size
+        static unsigned int const   c_nRelocateLimit = base_class::c_nRelocateLimit;             ///< Count of attempts to relocate before giving up
+
+    protected:
+        //@cond
+        template <typename Q>
+        static node_type * alloc_node( Q const& v )
+        {
+            return cxx_node_allocator().New( v );
+        }
+        template <typename... Args>
+        static node_type * alloc_node( Args&&... args )
+        {
+            return cxx_node_allocator().MoveNew( std::forward<Args>(args)... );
+        }
+
+        static void free_node( node_type * pNode )
+        {
+            cxx_node_allocator().Delete( pNode );
+        }
+        //@endcond
+
+    protected:
+        //@cond
+        struct node_disposer {
+            void operator()( node_type * pNode )
+            {
+                free_node( pNode );
+            }
+        };
+
+        typedef std::unique_ptr< node_type, node_disposer >     scoped_node_ptr;
+
+        //@endcond
+
+    public:
+        /// Default constructor
+        /**
+            Initial size = \ref c_nDefaultInitialSize
+
+            Probe set size:
+            - \ref c_nDefaultProbesetSize if \p probeset_type is \p cuckoo::list
+            - \p Capacity if \p probeset_type is <tt> cuckoo::vector<Capacity> </tt>
+
+            Probe set threshold = probe set size - 1
+        */
+        CuckooSet()
+        {}
+
+        /// Constructs the set object with given probe set size and threshold
+        /**
+            If probe set type is <tt> cuckoo::vector<Capacity> </tt> vector
+            then \p nProbesetSize should be equal to vector's \p Capacity.
+        */
+        CuckooSet(
+            size_t nInitialSize                 ///< Initial set size; if 0 - use default initial size \ref c_nDefaultInitialSize
+            , unsigned int nProbesetSize        ///< probe set size
+            , unsigned int nProbesetThreshold = 0  ///< probe set threshold, <tt>nProbesetThreshold < nProbesetSize</tt>. If 0, nProbesetThreshold = nProbesetSize - 1
+        )
+        : base_class( nInitialSize, nProbesetSize, nProbesetThreshold )
+        {}
+
+        /// Constructs the set object with given hash functor tuple
+        /**
+            The probe set size and threshold are set as default, see CuckooSet()
+        */
+        CuckooSet(
+            hash_tuple_type const& h    ///< hash functor tuple of type <tt>std::tuple<H1, H2, ... Hn></tt> where <tt> n == \ref c_nArity </tt>
+        )
+        : base_class( h )
+        {}
+
+        /// Constructs the set object with given probe set properties and hash functor tuple
+        /**
+            If probe set type is <tt> cuckoo::vector<Capacity> </tt> vector
+            then \p nProbesetSize should be equal to vector's \p Capacity.
+        */
+        CuckooSet(
+            size_t nInitialSize                 ///< Initial set size; if 0 - use default initial size \ref c_nDefaultInitialSize
+            , unsigned int nProbesetSize        ///< probe set size
+            , unsigned int nProbesetThreshold   ///< probe set threshold, <tt>nProbesetThreshold < nProbesetSize</tt>. If 0, nProbesetThreshold = nProbesetSize - 1
+            , hash_tuple_type const& h    ///< hash functor tuple of type <tt>std::tuple<H1, H2, ... Hn></tt> where <tt> n == \ref c_nArity </tt>
+        )
+        : base_class( nInitialSize, nProbesetSize, nProbesetThreshold, h )
+        {}
+
+        /// Constructs the set object with given hash functor tuple (move semantics)
+        /**
+            The probe set size and threshold are set as default, see CuckooSet()
+        */
+        CuckooSet(
+            hash_tuple_type&& h     ///< hash functor tuple of type <tt>std::tuple<H1, H2, ... Hn></tt> where <tt> n == \ref c_nArity </tt>
+        )
+        : base_class( std::forward<hash_tuple_type>(h))
+        {}
+
+        /// Constructs the set object with given probe set properties and hash functor tuple (move semantics)
+        /**
+            If probe set type is <tt> cuckoo::vector<Capacity> </tt> vector
+            then \p nProbesetSize should be equal to vector's \p Capacity.
+        */
+        CuckooSet(
+            size_t nInitialSize                 ///< Initial set size; if 0 - use default initial size \ref c_nDefaultInitialSize
+            , unsigned int nProbesetSize        ///< probe set size
+            , unsigned int nProbesetThreshold   ///< probe set threshold, <tt>nProbesetThreshold < nProbesetSize</tt>. If 0, nProbesetThreshold = nProbesetSize - 1
+            , hash_tuple_type&& h    ///< hash functor tuple of type <tt>std::tuple<H1, H2, ... Hn></tt> where <tt> n == \ref c_nArity </tt>
+        )
+        : base_class( nInitialSize, nProbesetSize, nProbesetThreshold, std::forward<hash_tuple_type>(h))
+        {}
+
+        /// Destructor clears the set
+        ~CuckooSet()
+        {
+            clear();
+        }
+
+    public:
+        /// Inserts new node
+        /**
+            The function creates a node with copy of \p val value
+            and then inserts the node created into the set.
+
+            The type \p Q should contain as minimum the complete key for the node.
+            The object of \ref value_type should be constructible from a value of type \p Q.
+            In trivial case, \p Q is equal to \ref value_type.
+
+            Returns \p true if \p val is inserted into the set, \p false otherwise.
+        */
+        template <typename Q>
+        bool insert( Q const& val )
+        {
+            return insert( val, []( value_type& ) {} );
+        }
+
+        /// Inserts new node
+        /**
+            The function allows to split creating of new item into two part:
+            - create item with key only
+            - insert new item into the set
+            - if inserting is success, calls \p f functor to initialize value-field of new item .
+
+            The functor signature is:
+            \code
+                void func( value_type& item );
+            \endcode
+            where \p item is the item inserted.
+
+            The type \p Q can differ from \ref value_type of items storing in the set.
+            Therefore, the \p value_type should be constructible from type \p Q.
+
+            The user-defined functor is called only if the inserting is success.
+        */
+        template <typename Q, typename Func>
+        bool insert( Q const& val, Func f )
+        {
+            scoped_node_ptr pNode( alloc_node( val ));
+            if ( base_class::insert( *pNode, [&f]( node_type& node ) { f( node.m_val ); } )) {
+                pNode.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// Inserts data of type \ref value_type constructed with <tt>std::forward<Args>(args)...</tt>
+        /**
+            Returns \p true if inserting successful, \p false otherwise.
+        */
+        template <typename... Args>
+        bool emplace( Args&&... args )
+        {
+            scoped_node_ptr pNode( alloc_node( std::forward<Args>(args)... ));
+            if ( base_class::insert( *pNode )) {
+                pNode.release();
+                return true;
+            }
+            return false;
+        }
+
+        /// Updates the node
+        /**
+            The operation performs inserting or changing data with lock-free manner.
+
+            If the item \p val is not found in the set, then \p val is inserted into the set
+            iff \p bAllowInsert is \p true.
+            Otherwise, the functor \p func is called with item found.
+            The functor \p func signature is:
+            \code
+                struct my_functor {
+                    void operator()( bool bNew, value_type& item, const Q& val );
+                };
+            \endcode
+            with arguments:
+            - \p bNew - \p true if the item has been inserted, \p false otherwise
+            - \p item - item of the set
+            - \p val - argument \p val passed into the \p %update() function
+            If new item has been inserted (i.e. \p bNew is \p true) then \p item and \p val arguments
+            refer to the same thing.
+
+            Returns std::pair<bool, bool> where \p first is \p true if operation is successful,
+            i.e. the node has been inserted or updated,
+            \p second is \p true if new item has been added or \p false if the item with \p key
+            already exists.
+        */
+        template <typename Q, typename Func>
+        std::pair<bool, bool> update( Q const& val, Func func, bool bAllowInsert = true )
+        {
+            scoped_node_ptr pNode( alloc_node( val ));
+            std::pair<bool, bool> res = base_class::update( *pNode,
+                [&val,&func](bool bNew, node_type& item, node_type const& ){ func( bNew, item.m_val, val ); },
+                bAllowInsert
+            );
+            if ( res.first && res.second )
+                pNode.release();
+            return res;
+        }
+        //@cond
+        template <typename Q, typename Func>
+        CDS_DEPRECATED("ensure() is deprecated, use update()")
+        std::pair<bool, bool> ensure( Q const& val, Func func )
+        {
+            return update( val, func, true );
+        }
+        //@endcond
+
+        /// Delete \p key from the set
+        /** \anchor cds_nonintrusive_CuckooSet_erase
+
+            Since the key of set's item type \ref value_type is not explicitly specified,
+            template parameter \p Q defines the key type searching in the list.
+            The set item comparator should be able to compare the type \p value_type
+            and the type \p Q.
+
+            Return \p true if key is found and deleted, \p false otherwise
+        */
+        template <typename Q>
+        bool erase( Q const& key )
+        {
+            node_type * pNode = base_class::erase( key );
+            if ( pNode ) {
+                free_node( pNode );
+                return true;
+            }
+            return false;
+        }
+
+        /// Deletes the item from the list using \p pred predicate for searching
+        /**
+            The function is an analog of \ref cds_nonintrusive_CuckooSet_erase "erase(Q const&)"
+            but \p pred is used for key comparing.
+            If cuckoo set is ordered, then \p Predicate should have the interface and semantics like \p std::less.
+            If cuckoo set is unordered, then \p Predicate should have the interface and semantics like \p std::equal_to.
+            \p Predicate must imply the same element order as the comparator used for building the set.
+        */
+        template <typename Q, typename Predicate>
+        bool erase_with( Q const& key, Predicate pred )
+        {
+            CDS_UNUSED( pred );
+            node_type * pNode = base_class::erase_with( key, typename maker::template predicate_wrapper<Predicate, bool>());
+            if ( pNode ) {
+                free_node( pNode );
+                return true;
+            }
+            return false;
+        }
+
+        /// Delete \p key from the set
+        /** \anchor cds_nonintrusive_CuckooSet_erase_func
+
+            The function searches an item with key \p key, calls \p f functor
+            and deletes the item. If \p key is not found, the functor is not called.
+
+            The functor \p Func interface is:
+            \code
+            struct functor {
+                void operator()(value_type const& val);
+            };
+            \endcode
+
+            Return \p true if key is found and deleted, \p false otherwise
+        */
+        template <typename Q, typename Func>
+        bool erase( Q const& key, Func f )
+        {
+            node_type * pNode = base_class::erase( key );
+            if ( pNode ) {
+                f( pNode->m_val );
+                free_node( pNode );
+                return true;
+            }
+            return false;
+        }
+
+        /// Deletes the item from the list using \p pred predicate for searching
+        /**
+            The function is an analog of \ref cds_nonintrusive_CuckooSet_erase_func "erase(Q const&, Func)"
+            but \p pred is used for key comparing.
+            If you use ordered cuckoo set, then \p Predicate should have the interface and semantics like \p std::less.
+            If you use unordered cuckoo set, then \p Predicate should have the interface and semantics like \p std::equal_to.
+            \p Predicate must imply the same element order as the comparator used for building the set.
+        */
+        template <typename Q, typename Predicate, typename Func>
+        bool erase_with( Q const& key, Predicate pred, Func f )
+        {
+            CDS_UNUSED( pred );
+            node_type * pNode = base_class::erase_with( key, typename maker::template predicate_wrapper<Predicate, bool>());
+            if ( pNode ) {
+                f( pNode->m_val );
+                free_node( pNode );
+                return true;
+            }
+            return false;
+        }
+
+        /// Find the key \p val
+        /** \anchor cds_nonintrusive_CuckooSet_find_func
+
+            The function searches the item with key equal to \p val and calls the functor \p f for item found.
+            The interface of \p Func functor is:
+            \code
+            struct functor {
+                void operator()( value_type& item, Q& val );
+            };
+            \endcode
+            where \p item is the item found, \p val is the <tt>find</tt> function argument.
+
+            The functor can change non-key fields of \p item.
+            The \p val argument is non-const since it can be used as \p f functor destination i.e., the functor
+            can modify both arguments.
+
+            The type \p Q can differ from \ref value_type of items storing in the container.
+            Therefore, the \p value_type should be comparable with type \p Q.
+
+            The function returns \p true if \p val is found, \p false otherwise.
+        */
+        template <typename Q, typename Func>
+        bool find( Q& val, Func f )
+        {
+            return base_class::find( val, [&f](node_type& item, Q& v) { f( item.m_val, v );});
+        }
+        //@cond
+        template <typename Q, typename Func>
+        bool find( Q const& val, Func f )
+        {
+            return base_class::find( val, [&f](node_type& item, Q const& v) { f( item.m_val, v );});
+        }
+        //@endcond
+
+        /// Find the key \p val using \p pred predicate for comparing
+        /**
+            The function is an analog of \ref cds_nonintrusive_CuckooSet_find_func "find(Q&, Func)"
+            but \p pred is used for key comparison.
+            If you use ordered cuckoo set, then \p Predicate should have the interface and semantics like \p std::less.
+            If you use unordered cuckoo set, then \p Predicate should have the interface and semantics like \p std::equal_to.
+            \p pred must imply the same element order as the comparator used for building the set.
+        */
+        template <typename Q, typename Predicate, typename Func>
+        bool find_with( Q& val, Predicate pred, Func f )
+        {
+            CDS_UNUSED( pred );
+            return base_class::find_with( val, typename maker::template predicate_wrapper<Predicate, bool>(),
+                [&f](node_type& item, Q& v) { f( item.m_val, v );});
+        }
+        //@cond
+        template <typename Q, typename Predicate, typename Func>
+        bool find_with( Q const& val, Predicate pred, Func f )
+        {
+            CDS_UNUSED( pred );
+            return base_class::find_with( val, typename maker::template predicate_wrapper<Predicate, bool>(),
+                [&f](node_type& item, Q const& v) { f( item.m_val, v );});
+        }
+        //@endcond
+
+        /// Checks whether the set contains \p key
+        /**
+            The function searches the item with key equal to \p key
+            and returns \p true if it is found, and \p false otherwise.
+        */
+        template <typename Q>
+        bool contains( Q const& key )
+        {
+            return base_class::find( key, [](node_type&, Q const&) {});
+        }
+        //@cond
+        template <typename Q>
+        CDS_DEPRECATED("the function is deprecated, use contains()")
+        bool find( Q const& key )
+        {
+            return contains( key );
+        }
+        //@endcond
+
+        /// Checks whether the set contains \p key using \p pred predicate for searching
+        /**
+            The function is similar to <tt>contains( key )</tt> but \p pred is used for key comparing.
+            \p Less functor has the interface like \p std::less.
+            \p Less must imply the same element order as the comparator used for building the set.
+        */
+        template <typename Q, typename Predicate>
+        bool contains( Q const& key, Predicate pred )
+        {
+            CDS_UNUSED( pred );
+            return base_class::find_with( key, typename maker::template predicate_wrapper<Predicate, bool>(), [](node_type&, Q const&) {});
+        }
+        //@cond
+        template <typename Q, typename Predicate>
+        CDS_DEPRECATED("the function is deprecated, use contains()")
+        bool find_with( Q const& key, Predicate pred )
+        {
+            return contains( key, pred );
+        }
+        //@endcond
+
+        /// Clears the set
+        /**
+            The function erases all items from the set.
+        */
+        void clear()
+        {
+            return base_class::clear_and_dispose( node_disposer());
+        }
+
+        /// Checks if the set is empty
+        /**
+            Emptiness is checked by item counting: if item count is zero then the set is empty.
+        */
+        bool empty() const
+        {
+            return base_class::empty();
+        }
+
+        /// Returns item count in the set
+        size_t size() const
+        {
+            return base_class::size();
+        }
+
+        /// Returns the size of hash table
+        /**
+            The hash table size is non-constant and can be increased via resizing.
+        */
+        size_t bucket_count() const
+        {
+            return base_class::bucket_count();
+        }
+
+        /// Returns lock array size
+        size_t lock_count() const
+        {
+            return base_class::lock_count();
+        }
+
+        /// Returns const reference to internal statistics
+        stat const& statistics() const
+        {
+            return base_class::statistics();
+        }
+
+        /// Returns const reference to mutex policy internal statistics
+        typename mutex_policy::statistics_type const& mutex_policy_statistics() const
+        {
+            return base_class::mutex_policy_statistics();
+        }
+    };
+
+}} // namespace cds::container
+
+#endif //#ifndef CDSLIB_CONTAINER_CUCKOO_SET_H

extern/libcds/cds/container/details/base.h

@@ -0,0 +1,78 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_DETAILS_BASE_H
+#define CDSLIB_CONTAINER_DETAILS_BASE_H
+
+#include <cds/intrusive/details/base.h>
+
+namespace cds {
+
+/// Standard (non-intrusive) containers
+/**
+    @ingroup cds_nonintrusive_containers
+    This namespace contains implementations of non-intrusive (std-like) lock-free containers.
+*/
+namespace container {
+
+    /// Common options for non-intrusive containers
+    /** @ingroup cds_nonintrusive_helper
+        This namespace contains options for non-intrusive containers that is, in general, the same as for the intrusive containers.
+        It imports all definitions from cds::opt and cds::intrusive::opt namespaces
+    */
+    namespace opt {
+        using namespace cds::intrusive::opt;
+    }   // namespace opt
+
+    /// @defgroup cds_nonintrusive_containers Non-intrusive containers
+    /** @defgroup cds_nonintrusive_helper Helper structs for non-intrusive containers
+        @ingroup cds_nonintrusive_containers
+    */
+
+    /** @defgroup cds_nonintrusive_stack Stack
+        @ingroup cds_nonintrusive_containers
+    */
+    /** @defgroup cds_nonintrusive_queue Queue
+        @ingroup cds_nonintrusive_containers
+    */
+    /** @defgroup cds_nonintrusive_deque Deque
+        @ingroup cds_nonintrusive_containers
+    */
+    /** @defgroup cds_nonintrusive_priority_queue Priority queue
+        @ingroup cds_nonintrusive_containers
+    */
+    /** @defgroup cds_nonintrusive_map Map
+        @ingroup cds_nonintrusive_containers
+    */
+    /** @defgroup cds_nonintrusive_set Set
+        @ingroup cds_nonintrusive_containers
+    */
+    /** @defgroup cds_nonintrusive_list List
+        @ingroup cds_nonintrusive_containers
+    */
+    /** @defgroup cds_nonintrusive_tree Tree
+        @ingroup cds_nonintrusive_containers
+    */
+
+
+    // Tag for selecting iterable list implementation
+    /** @ingroup cds_nonintrusive_helper
+        This struct is empty and it is used only as a tag for selecting \p IterableList
+        as ordered list implementation in declaration of some classes.
+
+        See \p split_list::traits::ordered_list as an example.
+    */
+    typedef intrusive::iterable_list_tag iterable_list_tag;
+
+    //@cond
+    template <typename List>
+    struct is_iterable_list: public cds::intrusive::is_iterable_list< List >
+    {};
+    //@endcond
+
+}   // namespace container
+}   // namespace cds
+
+#endif // #ifndef CDSLIB_CONTAINER_DETAILS_BASE_H

extern/libcds/cds/container/details/bronson_avltree_base.h

@@ -0,0 +1,503 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_DETAILS_BRONSON_AVLTREE_BASE_H
+#define CDSLIB_CONTAINER_DETAILS_BRONSON_AVLTREE_BASE_H
+
+#include <cds/container/details/base.h>
+#include <cds/opt/compare.h>
+#include <cds/urcu/options.h>
+#include <cds/sync/spinlock.h>
+#include <cds/sync/injecting_monitor.h>
+
+namespace cds { namespace container {
+
+    /// BronsonAVLTree related declarations
+    namespace bronson_avltree {
+
+        template <typename Key, typename T, typename SyncMonitor >
+        struct node;
+
+        //@cond
+        template <typename Node, typename T, typename SyncMonitor>
+        struct link_node
+        {
+            typedef Node     node_type;
+            typedef T        mapped_type;
+            typedef uint32_t version_type;  ///< version type (internal)
+
+            enum
+            {
+                shrinking = 1,
+                unlinked = 2,
+                version_flags = shrinking | unlinked
+                // the rest is version counter
+            };
+
+            atomics::atomic< int >          m_nHeight;  ///< Node height
+            atomics::atomic<version_type>   m_nVersion; ///< Version bits
+            atomics::atomic<node_type *>    m_pParent;  ///< Parent node
+            atomics::atomic<node_type *>    m_pLeft;    ///< Left child
+            atomics::atomic<node_type *>    m_pRight;   ///< Right child
+            typename SyncMonitor::node_injection m_SyncMonitorInjection;    ///< @ref cds_sync_monitor "synchronization monitor" injected data
+            atomics::atomic<mapped_type *>  m_pValue;   ///< Value
+
+        public:
+            link_node()
+                : m_nHeight( 0 )
+                , m_nVersion( 0 )
+                , m_pParent( nullptr )
+                , m_pLeft( nullptr )
+                , m_pRight( nullptr )
+            {
+                m_pValue.store( nullptr, atomics::memory_order_release );
+            }
+
+            link_node( int nHeight, version_type version, node_type * pParent, node_type * pLeft, node_type * pRight )
+                : m_nHeight( nHeight )
+                , m_nVersion( version )
+                , m_pParent( pParent )
+                , m_pLeft( pLeft )
+                , m_pRight( pRight )
+            {
+                m_pValue.store( nullptr, atomics::memory_order_release );
+            }
+
+            node_type * parent( atomics::memory_order order ) const
+            {
+                return m_pParent.load( order );
+            }
+
+            void parent( node_type * p, atomics::memory_order order )
+            {
+                m_pParent.store( p, order );
+            }
+
+            node_type * child( int nDirection, atomics::memory_order order ) const
+            {
+                assert( nDirection != 0 );
+                return nDirection < 0 ? m_pLeft.load( order ) : m_pRight.load( order );
+            }
+
+            void child( node_type * pChild, int nDirection, atomics::memory_order order )
+            {
+                assert( nDirection != 0 );
+                if ( nDirection < 0 )
+                    m_pLeft.store( pChild, order );
+                else
+                    m_pRight.store( pChild, order );
+            }
+
+            version_type version( atomics::memory_order order ) const
+            {
+                return m_nVersion.load( order );
+            }
+
+            void version( version_type ver, atomics::memory_order order )
+            {
+                m_nVersion.store( ver, order );
+            }
+
+            void exchange_version( version_type ver, atomics::memory_order order )
+            {
+                m_nVersion.exchange( ver, order );
+            }
+
+            int height( atomics::memory_order order ) const
+            {
+                return m_nHeight.load( order );
+            }
+
+            void height( int h, atomics::memory_order order )
+            {
+                m_nHeight.store( h, order );
+            }
+
+            template <typename BackOff>
+            void wait_until_shrink_completed( atomics::memory_order order ) const
+            {
+                BackOff bkoff;
+                while ( is_shrinking( order ))
+                    bkoff();
+            }
+
+            bool is_unlinked( atomics::memory_order order ) const
+            {
+                return m_nVersion.load( order ) == unlinked;
+            }
+
+            bool is_shrinking( atomics::memory_order order ) const
+            {
+                return (m_nVersion.load( order ) & shrinking) != 0;
+            }
+
+            mapped_type * value( atomics::memory_order order ) const
+            {
+                return m_pValue.load( order );
+            }
+
+            bool is_valued( atomics::memory_order order ) const
+            {
+                return value( order ) != nullptr;
+            }
+        };
+        //@endcond
+
+        /// BronsonAVLTree internal node
+        template <typename Key, typename T, typename SyncMonitor >
+        struct node<Key, T*, SyncMonitor>: public link_node< node<Key, T*, SyncMonitor>, T, SyncMonitor >
+        {
+            //@cond
+            typedef link_node< node<Key, T*, SyncMonitor>, T, SyncMonitor > base_class;
+            //@endcond
+
+            typedef Key key_type;       ///< key type
+            typedef T   mapped_type;    ///< value type
+            //@cond
+            typedef typename base_class::version_type version_type;
+            //@endcond
+
+            key_type const                  m_key;      ///< Key
+            node *                          m_pNextRemoved; ///< thread-local list of removed node
+
+        public:
+            //@cond
+            template <typename Q>
+            node( Q&& key )
+                : base_class()
+                , m_key( std::forward<Q>( key ))
+                , m_pNextRemoved( nullptr )
+            {}
+
+            template <typename Q>
+            node( Q&& key, int nHeight, version_type version, node * pParent, node * pLeft, node * pRight )
+                : base_class( nHeight, version, pParent, pLeft, pRight )
+                , m_key( std::forward<Q>( key ))
+                , m_pNextRemoved( nullptr )
+            {}
+            //@endcond
+        };
+
+        /// BronsonAVLTreeMap internal statistics
+        template <typename Counter = cds::atomicity::event_counter>
+        struct stat {
+            typedef Counter   event_counter; ///< Event counter type
+
+            event_counter   m_nFindSuccess; ///< Count of success \p find() call
+            event_counter   m_nFindFailed;  ///< Count of failed \p find() call
+            event_counter   m_nFindRetry;   ///< Count of retries during \p find()
+            event_counter   m_nFindWaitShrinking;   ///< Count of waiting until shrinking completed duting \p find() call
+
+            event_counter   m_nInsertSuccess;       ///< Count of inserting data node
+            event_counter   m_nInsertFailed;        ///< Count of insert failures
+            event_counter   m_nRelaxedInsertFailed; ///< Count of false creating of data nodes (only if @ref bronson_avltree::relaxed_insert "relaxed insertion" is enabled)
+            event_counter   m_nInsertRetry;         ///< Count of insert retries via concurrent operations
+            event_counter   m_nUpdateWaitShrinking; ///< Count of waiting until shrinking completed during \p update() call
+            event_counter   m_nUpdateRetry;         ///< Count of update retries via concurrent operations
+            event_counter   m_nUpdateRootWaitShrinking;  ///< Count of waiting until root shrinking completed duting \p update() call
+            event_counter   m_nUpdateSuccess;       ///< Count of updating data node
+            event_counter   m_nUpdateUnlinked;      ///< Count of attempts to update unlinked node
+            event_counter   m_nDisposedNode;        ///< Count of disposed node
+            event_counter   m_nDisposedValue;       ///< Count of disposed value
+            event_counter   m_nExtractedValue;      ///< Count of extracted value
+            event_counter   m_nRemoveSuccess;       ///< Count of successfully \p erase() call
+            event_counter   m_nRemoveFailed;        ///< Count of failed \p erase() call
+            event_counter   m_nRemoveRetry;         ///< Count o erase/extract retries
+            event_counter   m_nExtractSuccess;      ///< Count of successfully \p extract() call
+            event_counter   m_nExtractFailed;       ///< Count of failed \p extract() call
+            event_counter   m_nRemoveWaitShrinking; ///< ount of waiting until shrinking completed during \p erase() or \p extract() call
+            event_counter   m_nRemoveRootWaitShrinking;  ///< Count of waiting until root shrinking completed duting \p erase() or \p extract() call
+            event_counter   m_nMakeRoutingNode;     ///< How many nodes were converted to routing (valueless) nodes
+
+            event_counter   m_nRightRotation;       ///< Count of single right rotation
+            event_counter   m_nLeftRotation;        ///< Count of single left rotation
+            event_counter   m_nLeftRightRotation;   ///< Count of double left-over-right rotation
+            event_counter   m_nRightLeftRotation;   ///< Count of double right-over-left rotation
+
+            event_counter   m_nRotateAfterRightRotation; ///< Count of rotation required after single right rotation
+            event_counter   m_nRemoveAfterRightRotation; ///< Count of removal required after single right rotation
+            event_counter   m_nDamageAfterRightRotation; ///< Count of damaged node after single right rotation
+
+            event_counter   m_nRotateAfterLeftRotation;  ///< Count of rotation required after signle left rotation
+            event_counter   m_nRemoveAfterLeftRotation;  ///< Count of removal required after single left rotation
+            event_counter   m_nDamageAfterLeftRotation;  ///< Count of damaged node after single left rotation
+
+            event_counter   m_nRotateAfterRLRotation;    ///< Count of rotation required after right-over-left rotation
+            event_counter   m_nRemoveAfterRLRotation;    ///< Count of removal required after right-over-left rotation
+            event_counter   m_nRotateAfterLRRotation;    ///< Count of rotation required after left-over-right rotation
+            event_counter   m_nRemoveAfterLRRotation;    ///< Count of removal required after left-over-right rotation
+
+            event_counter   m_nInsertRebalanceReq;  ///< Count of rebalance required after inserting
+            event_counter   m_nRemoveRebalanceReq;  ///< Count of rebalance required after removing
+
+            //@cond
+            void onFindSuccess()        { ++m_nFindSuccess      ; }
+            void onFindFailed()         { ++m_nFindFailed       ; }
+            void onFindRetry()          { ++m_nFindRetry        ; }
+            void onFindWaitShrinking()  { ++m_nFindWaitShrinking; }
+
+            void onInsertSuccess()          { ++m_nInsertSuccess; }
+            void onInsertFailed()           { ++m_nInsertFailed; }
+            void onRelaxedInsertFailed()    { ++m_nRelaxedInsertFailed; }
+            void onInsertRetry()            { ++m_nInsertRetry      ; }
+            void onUpdateWaitShrinking()    { ++m_nUpdateWaitShrinking; }
+            void onUpdateRetry()            { ++m_nUpdateRetry; }
+            void onUpdateRootWaitShrinking() { ++m_nUpdateRootWaitShrinking; }
+            void onUpdateSuccess()          { ++m_nUpdateSuccess;  }
+            void onUpdateUnlinked()         { ++m_nUpdateUnlinked; }
+            void onDisposeNode()            { ++m_nDisposedNode; }
+            void onDisposeValue()           { ++m_nDisposedValue; }
+            void onExtractValue()           { ++m_nExtractedValue; }
+            void onRemove(bool bSuccess)
+            {
+                if ( bSuccess )
+                    ++m_nRemoveSuccess;
+                else
+                    ++m_nRemoveFailed;
+            }
+            void onExtract( bool bSuccess )
+            {
+                if ( bSuccess )
+                    ++m_nExtractSuccess;
+                else
+                    ++m_nExtractFailed;
+            }
+            void onRemoveRetry()            { ++m_nRemoveRetry; }
+            void onRemoveWaitShrinking()    { ++m_nRemoveWaitShrinking; }
+            void onRemoveRootWaitShrinking() { ++m_nRemoveRootWaitShrinking; }
+            void onMakeRoutingNode()        { ++m_nMakeRoutingNode; }
+
+            void onRotateRight()            { ++m_nRightRotation; }
+            void onRotateLeft()             { ++m_nLeftRotation; }
+            void onRotateRightOverLeft()    { ++m_nRightLeftRotation; }
+            void onRotateLeftOverRight()    { ++m_nLeftRightRotation; }
+
+            void onRotateAfterRightRotation() { ++m_nRotateAfterRightRotation; }
+            void onRemoveAfterRightRotation() { ++m_nRemoveAfterRightRotation; }
+            void onDamageAfterRightRotation() { ++m_nDamageAfterRightRotation; }
+
+            void onRotateAfterLeftRotation()  { ++m_nRotateAfterLeftRotation; }
+            void onRemoveAfterLeftRotation()  { ++m_nRemoveAfterLeftRotation; }
+            void onDamageAfterLeftRotation()  { ++m_nDamageAfterLeftRotation; }
+
+            void onRotateAfterRLRotation()    { ++m_nRotateAfterRLRotation; }
+            void onRemoveAfterRLRotation()    { ++m_nRemoveAfterRLRotation; }
+            void onRotateAfterLRRotation()    { ++m_nRotateAfterLRRotation; }
+            void onRemoveAfterLRRotation()    { ++m_nRemoveAfterLRRotation; }
+
+            void onInsertRebalanceRequired() { ++m_nInsertRebalanceReq; }
+            void onRemoveRebalanceRequired() { ++m_nRemoveRebalanceReq; }
+            //@endcond
+        };
+
+        /// BronsonAVLTreeMap empty statistics
+        struct empty_stat {
+            //@cond
+            void onFindSuccess()        const {}
+            void onFindFailed()         const {}
+            void onFindRetry()          const {}
+            void onFindWaitShrinking()  const {}
+
+            void onInsertSuccess()          const {}
+            void onInsertFailed()           const {}
+            void onRelaxedInsertFailed()    const {}
+            void onInsertRetry()            const {}
+            void onUpdateWaitShrinking()    const {}
+            void onUpdateRetry()            const {}
+            void onUpdateRootWaitShrinking() const {}
+            void onUpdateSuccess()          const {}
+            void onUpdateUnlinked()         const {}
+            void onDisposeNode()            const {}
+            void onDisposeValue()           const {}
+            void onExtractValue()           const {}
+            void onRemove(bool /*bSuccess*/) const {}
+            void onExtract(bool /*bSuccess*/) const {}
+            void onRemoveRetry()            const {}
+            void onRemoveWaitShrinking()    const {}
+            void onRemoveRootWaitShrinking() const {}
+            void onMakeRoutingNode()        const {}
+
+            void onRotateRight()            const {}
+            void onRotateLeft()             const {}
+            void onRotateRightOverLeft()    const {}
+            void onRotateLeftOverRight()    const {}
+
+            void onRotateAfterRightRotation() const {}
+            void onRemoveAfterRightRotation() const {}
+            void onDamageAfterRightRotation() const {}
+
+            void onRotateAfterLeftRotation()  const {}
+            void onRemoveAfterLeftRotation()  const {}
+            void onDamageAfterLeftRotation()  const {}
+
+            void onRotateAfterRLRotation()    const {}
+            void onRemoveAfterRLRotation()    const {}
+            void onRotateAfterLRRotation()    const {}
+            void onRemoveAfterLRRotation()    const {}
+
+            void onInsertRebalanceRequired() const {}
+            void onRemoveRebalanceRequired() const {}
+            //@endcond
+        };
+
+        /// Option to allow relaxed insert into \ref cds_container_BronsonAVLTreeMap_rcu "Bronson et al AVL-tree"
+        /**
+            By default, this option is disabled and the new node is created under its parent lock.
+            In this case, it is guaranteed the new node will be attached to its parent.
+            On the other hand, constructing of the new node can be too complex to make it under the lock,
+            that can lead to lock contention.
+
+            When this option is enabled, the new node is created before locking the parent node.
+            After that, the parent is locked and checked whether the new node can be attached to the parent.
+            In this case, false node creating can be performed, but locked section can be significantly small.
+        */
+        template <bool Enable>
+        struct relaxed_insert {
+            //@cond
+            template <typename Base> struct pack : public Base
+            {
+                enum { relaxed_insert = Enable };
+            };
+            //@endcond
+        };
+
+        /// \p BronsonAVLTreeMap traits
+        /**
+            Note that there are two main specialization of Bronson et al AVL-tree:
+            - \ref cds_container_BronsonAVLTreeMap_rcu_ptr "pointer-oriented" - the tree node stores an user-provided pointer to value
+            - \ref cds_container_BronsonAVLTreeMap_rcu "data-oriented" - the tree node contains a copy of values
+
+            Depends on tree specialization, different traits member can be used.
+        */
+        struct traits
+        {
+            /// Key comparison functor
+            /**
+                No default functor is provided. If the option is not specified, the \p less is used.
+
+                See \p cds::opt::compare option description for functor interface.
+
+                You should provide \p compare or \p less functor.
+            */
+            typedef opt::none                       compare;
+
+            /// Specifies binary predicate used for key compare.
+            /**
+                See \p cds::opt::less option description for predicate interface.
+
+                You should provide \p compare or \p less functor.
+            */
+            typedef opt::none                       less;
+
+            /// Allocator for internal node
+            typedef CDS_DEFAULT_ALLOCATOR           node_allocator;
+
+            /// Allocator for node's value (not used in \p BronsonAVLTreeMap<RCU, Key, T*, Traits> specialisation)
+            typedef CDS_DEFAULT_ALLOCATOR           allocator;
+
+            /// Disposer (only for pointer-oriented tree specialization)
+            /**
+                The functor used for dispose removed values.
+                The user-provided disposer is used only for pointer-oriented tree specialization
+                like \p BronsonAVLTreeMap<GC, Key, T*, Traits>. When the node becomes the routing node without value,
+                the disposer will be called to signal that the memory for the value can be safely freed.
+                Default is \ref cds::intrusive::opt::delete_disposer "cds::container::opt::v::delete_disposer<>" which calls \p delete operator.
+            */
+            typedef opt::v::delete_disposer<>       disposer;
+
+            /// @ref cds_sync_monitor "Synchronization monitor" type for node-level locking
+            typedef cds::sync::injecting_monitor<cds::sync::spin>  sync_monitor;
+
+            /// Enable relaxed insertion.
+            /**
+                About relaxed insertion see \p bronson_avltree::relaxed_insert option.
+                By default, this option is disabled.
+            */
+            static bool const relaxed_insert = false;
+
+            /// Item counter
+            /**
+                The type for item counter, by default it is disabled (\p atomicity::empty_item_counter).
+                To enable it use \p atomicity::item_counter or \p atomicity::cache_friendly_item_counter.
+            */
+            typedef atomicity::empty_item_counter     item_counter;
+
+            /// C++ memory ordering model
+            /**
+                List of available memory ordering see \p opt::memory_model
+            */
+            typedef opt::v::relaxed_ordering        memory_model;
+
+            /// Internal statistics
+            /**
+                By default, internal statistics is disabled (\p bronson_avltree::empty_stat).
+                To enable it use \p bronson_avltree::stat.
+            */
+            typedef empty_stat                      stat;
+
+            /// Back-off strategy
+            typedef cds::backoff::empty             back_off;
+
+            /// RCU deadlock checking policy
+            /**
+                List of available options see \p opt::rcu_check_deadlock
+            */
+            typedef cds::opt::v::rcu_throw_deadlock      rcu_check_deadlock;
+        };
+
+        /// Metafunction converting option list to BronsonAVLTreeMap traits
+        /**
+            Note that there are two main specialization of Bronson et al AVL-tree:
+            - \ref cds_container_BronsonAVLTreeMap_rcu_ptr "pointer-oriented" - the tree node stores an user-provided pointer to value
+            - \ref cds_container_BronsonAVLTreeMap_rcu "data-oriented" - the tree node contains a copy of values
+
+            Depends on tree specialization, different options can be specified.
+
+            \p Options are:
+            - \p opt::compare - key compare functor. No default functor is provided.
+                If the option is not specified, \p %opt::less is used.
+            - \p opt::less - specifies binary predicate used for key compare. At least \p %opt::compare or \p %opt::less should be defined.
+            - \p opt::node_allocator - the allocator for internal nodes. Default is \ref CDS_DEFAULT_ALLOCATOR.
+            - \p opt::allocator - the allocator for node's value. Default is \ref CDS_DEFAULT_ALLOCATOR.
+                This option is not used in \p BronsonAVLTreeMap<RCU, Key, T*, Traits> specialisation
+            - \p cds::intrusive::opt::disposer - the functor used for dispose removed values.
+                The user-provided disposer is used only for pointer-oriented tree specialization
+                like \p BronsonAVLTreeMap<GC, Key, T*, Traits>. When the node becomes the rounting node without value,
+                the disposer will be called to signal that the memory for the value can be safely freed.
+                Default is \p cds::intrusive::opt::delete_disposer which calls \p delete operator.
+                Due the nature of GC schema the disposer may be called asynchronously.
+            - \p opt::sync_monitor -  @ref cds_sync_monitor "synchronization monitor" type for node-level locking,
+                default is \p cds::sync::injecting_monitor<cds::sync::spin>
+            - \p bronson_avltree::relaxed_insert - enable (\p true) or disable (\p false, the default)
+                @ref bronson_avltree::relaxed_insert "relaxed insertion"
+            - \p opt::item_counter - the type of item counting feature, by default it is disabled (\p atomicity::empty_item_counter)
+                To enable it use \p atomicity::item_counter or \p atomicity::cache_friendly_item_counter
+            - \p opt::memory_model - C++ memory ordering model. Can be \p opt::v::relaxed_ordering (relaxed memory model, the default)
+                or \p opt::v::sequential_consistent (sequentially consisnent memory model).
+            - \p opt::stat - internal statistics, by default it is disabled (\p bronson_avltree::empty_stat)
+                To enable statistics use \p \p bronson_avltree::stat
+            - \p opt::backoff - back-off strategy, by default no strategy is used (\p cds::backoff::empty)
+            - \p opt::rcu_check_deadlock - a deadlock checking policy for RCU-based tree, default is \p opt::v::rcu_throw_deadlock
+        */
+        template <typename... Options>
+        struct make_traits {
+#   ifdef CDS_DOXYGEN_INVOKED
+            typedef implementation_defined type ;   ///< Metafunction result
+#   else
+            typedef typename cds::opt::make_options<
+                typename cds::opt::find_type_traits< traits, Options... >::type
+                ,Options...
+            >::type   type;
+#   endif
+        };
+    } // namespace bronson_avltree
+
+    // Forwards
+    template < class GC, typename Key, typename T, class Traits = bronson_avltree::traits >
+    class BronsonAVLTreeMap;
+
+}} // namespace cds::container
+
+#endif // #ifndef CDSLIB_CONTAINER_DETAILS_BRONSON_AVLTREE_BASE_H

extern/libcds/cds/container/details/cuckoo_base.h

@@ -0,0 +1,244 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_DETAILS_CUCKOO_BASE_H
+#define CDSLIB_CONTAINER_DETAILS_CUCKOO_BASE_H
+
+#include <cds/intrusive/cuckoo_set.h>
+
+namespace cds { namespace container {
+
+    /// CuckooSet and CuckooMap related definitions
+    /** @ingroup cds_nonintrusive_helper
+    */
+    namespace cuckoo {
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Lock striping concurrent access policy. This is typedef for intrusive::cuckoo::striping template
+        class striping
+        {};
+#else
+        using intrusive::cuckoo::striping;
+#endif
+
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Refinable concurrent access policy. This is typedef for intrusive::cuckoo::refinable template
+        class refinable
+        {};
+#else
+        using intrusive::cuckoo::refinable;
+#endif
+
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Striping internal statistics. This is typedef for intrusive::cuckoo::striping_stat
+        class striping_stat
+        {};
+#else
+        using intrusive::cuckoo::striping_stat;
+#endif
+
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Empty striping internal statistics. This is typedef for intrusive::cuckoo::empty_striping_stat
+        class empty_striping_stat
+        {};
+#else
+        using intrusive::cuckoo::empty_striping_stat;
+#endif
+
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Refinable internal statistics. This is typedef for intrusive::cuckoo::refinable_stat
+        class refinable_stat
+        {};
+#else
+        using intrusive::cuckoo::refinable_stat;
+#endif
+
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Empty refinable internal statistics. This is typedef for intrusive::cuckoo::empty_refinable_stat
+        class empty_refinable_stat
+        {};
+#else
+        using intrusive::cuckoo::empty_refinable_stat;
+#endif
+
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Cuckoo statistics. This is typedef for intrusive::cuckoo::stat
+        class stat
+        {};
+#else
+        using intrusive::cuckoo::stat;
+#endif
+
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Cuckoo empty statistics.This is typedef for intrusive::cuckoo::empty_stat
+        class empty_stat
+        {};
+#else
+        using intrusive::cuckoo::empty_stat;
+#endif
+
+        /// Option specifying whether to store hash values in the node
+        /**
+             This option reserves additional space in the hook to store the hash value of the object once it's introduced in the container.
+             When this option is used, the unordered container will store the calculated hash value in the hook and rehashing operations won't need
+             to recalculate the hash of the value. This option will improve the performance of unordered containers
+             when rehashing is frequent or hashing the value is a slow operation
+
+             The \p Enable template parameter toggles the feature:
+             - the value \p true enables storing the hash values
+             - the value \p false disables storing the hash values
+        */
+        template <bool Enable>
+        struct store_hash
+        {
+            //@cond
+            template <typename Base>
+            struct pack: public Base {
+                static bool const store_hash = Enable;
+            };
+            //@endcond
+        };
+
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Probe set type option
+        /**
+            @copydetails cds::intrusive::cuckoo::probeset_type
+        */
+        template <typename Type>
+        struct probeset_type
+        {};
+#else
+        using intrusive::cuckoo::probeset_type;
+#endif
+
+        using intrusive::cuckoo::list;
+        using intrusive::cuckoo::vector;
+
+        /// Type traits for CuckooSet and CuckooMap classes
+        struct traits
+        {
+            /// Hash functors tuple
+            /**
+                This is mandatory type and has no predefined one.
+
+                At least, two hash functors should be provided. All hash functor
+                should be orthogonal (different): for each <tt> i,j: i != j => h[i](x) != h[j](x) </tt>.
+                The hash functors are defined as <tt> std::tuple< H1, H2, ... Hn > </tt>:
+                \@code cds::opt::hash< std::tuple< h1, h2 > > \@endcode
+                The number of hash functors specifies the number \p k - the count of hash tables in cuckoo hashing.
+
+                To specify hash tuple in traits you should use \p cds::opt::hash_tuple:
+                \code
+                struct my_traits: public cds::container::cuckoo::traits {
+                    typedef cds::opt::hash_tuple< hash1, hash2 > hash;
+                };
+                \endcode
+            */
+            typedef cds::opt::none      hash;
+
+            /// Concurrent access policy
+            /**
+                Available opt::mutex_policy types:
+                - cuckoo::striping - simple, but the lock array is not resizable
+                - cuckoo::refinable - resizable lock array, but more complex access to set data.
+
+                Default is cuckoo::striping.
+            */
+            typedef cuckoo::striping<>               mutex_policy;
+
+            /// Key equality functor
+            /**
+                Default is <tt>std::equal_to<T></tt>
+            */
+            typedef opt::none                       equal_to;
+
+            /// Key comparison functor
+            /**
+                No default functor is provided. If the option is not specified, the \p less is used.
+            */
+            typedef opt::none                       compare;
+
+            /// specifies binary predicate used for key comparison.
+            /**
+                Default is \p std::less<T>.
+            */
+            typedef opt::none                       less;
+
+            /// Item counter
+            /**
+                The type for item counting feature.
+                Default is cds::atomicity::item_counter
+
+                Only atomic item counter type is allowed.
+            */
+            typedef cds::intrusive::cuckoo::traits::item_counter   item_counter;
+
+            /// Allocator type
+            /**
+                The allocator type for allocating bucket tables.
+                Default is \p CDS_DEFAULT_ALLOCATOR that is \p std::allocator
+            */
+            typedef CDS_DEFAULT_ALLOCATOR       allocator;
+
+            /// Node allocator type
+            /**
+                If this type is not set explicitly, the \ref allocator type is used.
+            */
+            typedef opt::none                   node_allocator;
+
+            /// Store hash value into items. See cuckoo::store_hash for explanation
+            static bool const store_hash = false;
+
+            /// Probe-set type. See \ref probeset_type option for explanation
+            typedef cuckoo::list                probeset_type;
+
+            /// Internal statistics
+            typedef empty_stat                  stat;
+        };
+
+        /// Metafunction converting option list to CuckooSet/CuckooMap traits
+        /**
+            Template argument list \p Options... are:
+            - \p opt::hash - hash functor tuple, mandatory option. At least, two hash functors should be provided. All hash functor
+                should be orthogonal (different): for each <tt> i,j: i != j => h[i](x) != h[j](x) </tt>.
+                The hash functors are passed as <tt> std::tuple< H1, H2, ... Hn > </tt>. The number of hash functors specifies
+                the number \p k - the count of hash tables in cuckoo hashing.
+            - \p opt::mutex_policy - concurrent access policy.
+                Available policies: \p cuckoo::striping, \p cuckoo::refinable.
+                Default is \p %cuckoo::striping.
+            - \p opt::equal_to - key equality functor like \p std::equal_to.
+                If this functor is defined then the probe-set will be unordered.
+                If \p %opt::compare or \p %opt::less option is specified too, then the probe-set will be ordered
+                and \p %opt::equal_to will be ignored.
+            - \p opt::compare - key comparison functor. No default functor is provided.
+                If the option is not specified, the \p %opt::less is used.
+                If \p %opt::compare or \p %opt::less option is specified, then the probe-set will be ordered.
+            - \p opt::less - specifies binary predicate used for key comparison. Default is \p std::less<T>.
+                If \p %opt::compare or \p %opt::less option is specified, then the probe-set will be ordered.
+            - \p opt::item_counter - the type of item counting feature. Default is \p opt::v::sequential_item_counter.
+            - \p opt::allocator - the allocator type using for allocating bucket tables.
+                Default is \ref CDS_DEFAULT_ALLOCATOR
+            - \p opt::node_allocator - the allocator type using for allocating set's items. If this option
+                is not specified then the type defined in \p %opt::allocator option is used.
+            - \p cuckoo::store_hash - this option reserves additional space in the node to store the hash value
+                of the object once it's introduced in the container. When this option is used,
+                the unordered container will store the calculated hash value in the node and rehashing operations won't need
+                to recalculate the hash of the value. This option will improve the performance of unordered containers
+                when rehashing is frequent or hashing the value is a slow operation. Default value is \p false.
+            - \ref intrusive::cuckoo::probeset_type "cuckoo::probeset_type" - type of probe set, may be \p cuckoo::list or <tt>cuckoo::vector<Capacity></tt>,
+                Default is \p cuckoo::list.
+            - \p opt::stat - internal statistics. Possibly types: \p cuckoo::stat, \p cuckoo::empty_stat.
+                Default is \p %cuckoo::empty_stat
+        */
+        template <typename... Options>
+        struct make_traits {
+            typedef typename cds::opt::make_options<
+                typename cds::opt::find_type_traits< cuckoo::traits, Options... >::type
+                ,Options...
+            >::type   type ;    ///< Result of metafunction
+        };
+    }   // namespace cuckoo
+}} // namespace cds::container
+
+#endif  // #ifndef CDSLIB_CONTAINER_DETAILS_CUCKOO_BASE_H

extern/libcds/cds/container/details/ellen_bintree_base.h

@@ -0,0 +1,435 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_DETAILS_ELLEN_BINTREE_BASE_H
+#define CDSLIB_CONTAINER_DETAILS_ELLEN_BINTREE_BASE_H
+
+#include <cds/intrusive/details/ellen_bintree_base.h>
+#include <cds/container/details/base.h>
+#include <cds/opt/compare.h>
+#include <cds/details/binary_functor_wrapper.h>
+
+
+namespace cds { namespace container {
+    /// EllenBinTree related definitions
+    /** @ingroup cds_nonintrusive_helper
+    */
+    namespace ellen_bintree {
+#ifdef CDS_DOXYGEN_INVOKED
+        /// Typedef for \p cds::intrusive::ellen_bintree::update_desc
+        typedef cds::intrusive::ellen_bintree::update_desc update_desc;
+
+        /// Typedef for \p cds::intrusive::ellen_bintree::internal_node
+        typedef cds::intrusive::ellen_bintree::internal_node internal_node;
+
+        /// Typedef for \p cds::intrusive::ellen_bintree::key_extractor
+        typedef cds::intrusive::ellen_bintree::key_extractor key_extractor;
+
+        /// Typedef for \p cds::intrusive::ellen_bintree::update_desc_allocator
+        typedef cds::intrusive::ellen_bintree::update_desc_allocator update_desc_allocator;
+#else
+        using cds::intrusive::ellen_bintree::update_desc;
+        using cds::intrusive::ellen_bintree::internal_node;
+        using cds::intrusive::ellen_bintree::key_extractor;
+        using cds::intrusive::ellen_bintree::update_desc_allocator;
+        using cds::intrusive::ellen_bintree::node_types;
+#endif
+        /// EllenBinTree internal statistics
+        template <typename Counter = cds::intrusive::ellen_bintree::stat<>::event_counter >
+        using stat = cds::intrusive::ellen_bintree::stat< Counter >;
+
+        /// EllenBinTree empty internal statistics
+        typedef cds::intrusive::ellen_bintree::empty_stat empty_stat;
+
+        /// EllenBinTree leaf node
+        template <typename GC, typename T>
+        struct node: public cds::intrusive::ellen_bintree::node<GC>
+        {
+            typedef T   value_type  ;   ///< Value type
+
+            T   m_Value ;   ///< Value
+
+            /// Default ctor
+            node()
+            {}
+
+            /// Initializing ctor
+            template <typename Q>
+            node(Q const& v)
+                : m_Value(v)
+            {}
+
+            /// Copy constructor
+            template <typename... Args>
+            node( Args const&... args )
+                : m_Value( args... )
+            {}
+
+            /// Move constructor
+            template <typename... Args>
+            node( Args&&... args )
+                : m_Value( std::forward<Args>( args )... )
+            {}
+        };
+
+        /// EllenBinTreeMap leaf node
+        template <typename GC, typename Key, typename T>
+        struct map_node: public cds::intrusive::ellen_bintree::node< GC >
+        {
+            typedef Key     key_type    ;   ///< key type
+            typedef T       mapped_type ;   ///< value type
+            typedef std::pair<key_type const, mapped_type> value_type  ;   ///< key-value pair stored in the map
+
+            value_type  m_Value     ;   ///< Key-value pair stored in map leaf node
+
+            /// Initializes key field, value if default-constructed
+            template <typename K>
+            map_node( K const& key )
+                : m_Value( std::make_pair( key_type(key), mapped_type()))
+            {}
+
+            /// Initializes key and value fields
+            template <typename K, typename Q>
+            map_node( K const& key, Q const& v )
+                : m_Value( std::make_pair(key_type(key), mapped_type(v)))
+            {}
+        };
+
+        /// Type traits for \p EllenBinTreeSet and \p EllenBinTreeMap
+        struct traits
+        {
+            /// Key extracting functor (only for \p EllenBinTreeSet)
+            /**
+                This is mandatory functor for \p %EllenBinTreeSet.
+                It has the following prototype:
+                \code
+                struct key_extractor {
+                    void operator ()( Key& dest, T const& src );
+                };
+                \endcode
+                It should initialize \p dest key from \p src data.
+                The functor is used to initialize internal nodes of \p %EllenBinTreeSet
+            */
+            typedef opt::none           key_extractor;
+
+            /// Key comparison functor
+            /**
+                No default functor is provided. If the option is not specified, the \p less is used.
+
+                See \p cds::opt::compare option description for functor interface.
+
+                You should provide \p compare or \p less functor.
+                See \ref cds_container_EllenBinTreeSet_rcu_less "predicate requirements".
+            */
+            typedef opt::none                       compare;
+
+            /// Specifies binary predicate used for key compare.
+            /**
+                See \p cds::opt::less option description.
+
+                You should provide \p compare or \p less functor.
+                See \ref cds_container_EllenBinTreeSet_rcu_less "predicate requirements".
+            */
+            typedef opt::none                       less;
+
+            /// Item counter
+            /**
+                The type for item counter, by default it is disabled (\p atomicity::empty_item_counter).
+                To enable it use \p atomicity::item_counter or \p atomicity::cache_friendly_item_counter
+            */
+            typedef atomicity::empty_item_counter     item_counter;
+
+            /// C++ memory ordering model
+            /**
+                List of available memory ordering see \p opt::memory_model
+            */
+            typedef opt::v::relaxed_ordering        memory_model;
+
+            /// Allocator for update descriptors
+            /**
+                The allocator type is used for \p ellen_bintree::update_desc.
+
+                Update descriptor is helping data structure with short lifetime and it is good candidate
+                for pooling. The number of simultaneously existing descriptors is a small number
+                limited the number of threads working with the tree.
+                Therefore, a bounded lock-free container like \p cds::container::VyukovMPMCCycleQueue
+                is good choice for the free-list of update descriptors,
+                see \p cds::memory::vyukov_queue_pool free-list implementation.
+
+                Also notice that size of update descriptor is not dependent on the type of data
+                stored in the tree so single free-list object can be used for several \p EllenBinTree object.
+            */
+            typedef CDS_DEFAULT_ALLOCATOR           update_desc_allocator;
+
+            /// Allocator for internal nodes
+            /**
+                The allocator type is used for \p ellen_bintree::internal_node.
+            */
+            typedef CDS_DEFAULT_ALLOCATOR           node_allocator;
+
+            /// Allocator for leaf nodes
+            /**
+                Each leaf node contains data stored in the container.
+            */
+            typedef CDS_DEFAULT_ALLOCATOR           allocator;
+
+            /// Internal statistics
+            /**
+                By default, internal statistics is disabled (\p ellen_bintree::empty_stat).
+                To enable it use \p ellen_bintree::stat.
+            */
+            typedef empty_stat                      stat;
+
+            /// Back-off strategy
+            typedef cds::backoff::empty             back_off;
+
+            /// RCU deadlock checking policy (only for RCU-based EllenBinTree<i>XXX</i> classes)
+            /**
+                List of available options see \p opt::rcu_check_deadlock
+            */
+            typedef cds::opt::v::rcu_throw_deadlock      rcu_check_deadlock;
+
+            /// Key copy policy (for \p EllenBinTreeMap)
+            /**
+                The key copy policy defines a functor to copy leaf node's key to internal node.
+                This policy is used only in \p EllenBinTreeMap.
+                By default, assignment operator is used.
+
+                The copy functor interface is:
+                \code
+                struct copy_functor {
+                    void operator()( Key& dest, Key const& src );
+                };
+                \endcode
+            */
+            typedef opt::none                           copy_policy;
+        };
+
+
+        /// Metafunction converting option list to \p EllenBinTreeSet traits
+        /**
+            \p Options are:
+            - \p ellen_bintree::key_extractor - key extracting functor, mandatory option. The functor has the following prototype:
+                \code
+                    struct key_extractor {
+                        void operator ()( Key& dest, T const& src );
+                    };
+                \endcode
+                It should initialize \p dest key from \p src data. The functor is used to initialize internal nodes.
+            - \p opt::compare - key compare functor. No default functor is provided.
+                If the option is not specified, \p %opt::less is used.
+            - \p opt::less - specifies binary predicate used for key compare. At least \p %opt::compare or \p %opt::less should be defined.
+            - \p opt::item_counter - the type of item counter, default is disabled (\p atomicity::empty_item_counter).
+                To enable it use \p atomicity::item_counter or \p atomicity::cache_friendly_item_counter
+            - \p opt::memory_model - C++ memory ordering model. Can be \p opt::v::relaxed_ordering (relaxed memory model, the default)
+                or \p opt::v::sequential_consistent (sequentially consisnent memory model).
+            - \p opt::allocator - the allocator for \ref ellen_bintree::node "leaf nodes" which contains data.
+                Default is \ref CDS_DEFAULT_ALLOCATOR.
+            - \p opt::node_allocator - the allocator for internal nodes. Default is \ref CDS_DEFAULT_ALLOCATOR.
+            - \p ellen_bintree::update_desc_allocator - an allocator of \ref ellen_bintree::update_desc "update descriptors",
+                default is \ref CDS_DEFAULT_ALLOCATOR.
+                Note that update descriptor is helping data structure with short lifetime and it is good candidate for pooling.
+                The number of simultaneously existing descriptors is a relatively small number limited the number of threads
+                working with the tree and RCU buffer size.
+                Therefore, a bounded lock-free container like \p cds::container::VyukovMPMCCycleQueue is good choice for the free-list
+                of update descriptors, see \p cds::memory::vyukov_queue_pool free-list implementation.
+                Also notice that size of update descriptor is not dependent on the type of data
+                stored in the tree so single free-list object can be used for several EllenBinTree-based object.
+            - \p opt::stat - internal statistics, by default disabled (\p ellen_bintree::empty_stat). To enable
+                it use \p ellen_bintree::stat.
+            - \p opt::backoff - back-off strategy, by default no strategy is used (\p cds::backoff::empty)
+            - \p opt::rcu_check_deadlock - a deadlock checking policy, only for RCU-based tree.
+                Default is \p opt::v::rcu_throw_deadlock.
+        */
+        template <typename... Options>
+        struct make_set_traits {
+#   ifdef CDS_DOXYGEN_INVOKED
+            typedef implementation_defined type ;   ///< Metafunction result
+#   else
+            typedef typename cds::opt::make_options<
+                typename cds::opt::find_type_traits< traits, Options... >::type
+                ,Options...
+            >::type   type;
+#   endif
+        };
+
+        /// Metafunction converting option list to \p EllenBinTreeMap traits
+        /**
+            \p Options are:
+            - \p opt::compare - key compare functor. No default functor is provided.
+                If the option is not specified, \p %opt::less is used.
+            - \p opt::less - specifies binary predicate used for key compare. At least \p %opt::compare or \p %opt::less should be defined.
+            - \p opt::item_counter - the type of item counter, default is disabled (\p atomicity::empty_item_counter).
+                To enable it use \p atomicity::item_counter or \p atomicity::cache_friendly_item_counter
+            - opt::memory_model - C++ memory ordering model. Can be \p opt::v::relaxed_ordering (relaxed memory model, the default)
+                or \p opt::v::sequential_consistent (sequentially consisnent memory model).
+            - \p opt::allocator - the allocator used for \ref ellen_bintree::map_node "leaf nodes" which contains data.
+                Default is \ref CDS_DEFAULT_ALLOCATOR.
+            - \p opt::node_allocator - the allocator used for \ref ellen_bintree::internal_node "internal nodes".
+                Default is \ref CDS_DEFAULT_ALLOCATOR.
+            - \p ellen_bintree::update_desc_allocator - an allocator of \ref ellen_bintree::update_desc "update descriptors",
+                default is \ref CDS_DEFAULT_ALLOCATOR.
+                Note that update descriptor is helping data structure with short lifetime and it is good candidate for pooling.
+                The number of simultaneously existing descriptors is a relatively small number limited the number of threads
+                working with the tree and RCU buffer size.
+                Therefore, a bounded lock-free container like \p cds::container::VyukovMPMCCycleQueue is good choice for the free-list
+                of update descriptors, see \p cds::memory::vyukov_queue_pool free-list implementation.
+                Also notice that size of update descriptor is not dependent on the type of data
+                stored in the tree so single free-list object can be used for several EllenBinTree-based object.
+            - \p opt::stat - internal statistics, by default disabled (\p ellen_bintree::empty_stat). To enable
+                it use \p ellen_bintree::stat.
+            - \p opt::backoff - back-off strategy, by default no strategy is used (\p cds::backoff::empty)
+            - \p opt::rcu_check_deadlock - a deadlock checking policy, only for RCU-based tree. Default is \p opt::v::rcu_throw_deadlock
+            - opt::copy_policy - key copying policy defines a functor to copy leaf node's key to internal node.
+                By default, assignment operator is used.
+                The copy functor interface is:
+                \code
+                struct copy_functor {
+                    void operator()( Key& dest, Key const& src );
+                };
+                \endcode
+        */
+        template <typename... Options>
+        struct make_map_traits {
+#   ifdef CDS_DOXYGEN_INVOKED
+            typedef implementation_defined type ;   ///< Metafunction result
+#   else
+            typedef typename cds::opt::make_options<
+                typename cds::opt::find_type_traits< traits, Options... >::type
+                ,Options...
+            >::type   type;
+#   endif
+        };
+
+        //@cond
+        namespace details {
+
+            template < class GC, typename Key, typename T, class Traits>
+            struct make_ellen_bintree_set
+            {
+                typedef GC      gc;
+                typedef Key     key_type;
+                typedef T       value_type;
+                typedef Traits  original_traits;
+
+                typedef node< gc, value_type >  leaf_node;
+
+                struct intrusive_key_extractor
+                {
+                    void operator()( key_type& dest, leaf_node const& src ) const
+                    {
+                        typename original_traits::key_extractor()( dest, src.m_Value );
+                    }
+                };
+
+                struct value_accessor
+                {
+                    value_type const& operator()( leaf_node const& node ) const
+                    {
+                        return node.m_Value;
+                    }
+                };
+
+                typedef typename cds::opt::details::make_comparator< value_type, original_traits, false >::type key_comparator;
+
+                typedef cds::details::Allocator< leaf_node, typename original_traits::allocator> cxx_leaf_node_allocator;
+                struct leaf_deallocator
+                {
+                    void operator()( leaf_node * p ) const
+                    {
+                        cxx_leaf_node_allocator().Delete( p );
+                    }
+                };
+
+                struct intrusive_traits: public original_traits
+                {
+                    typedef cds::intrusive::ellen_bintree::base_hook< cds::opt::gc< gc >> hook;
+                    typedef intrusive_key_extractor key_extractor;
+                    typedef leaf_deallocator        disposer;
+                    typedef cds::details::compare_wrapper< leaf_node, key_comparator, value_accessor > compare;
+                };
+
+                // Metafunction result
+                typedef cds::intrusive::EllenBinTree< gc, key_type, leaf_node, intrusive_traits > type;
+            };
+
+            template < class GC, typename Key, typename T, class Traits>
+            struct make_ellen_bintree_map
+            {
+                typedef GC      gc;
+                typedef Key     key_type;
+                typedef T       mapped_type;
+                typedef map_node< gc, key_type, mapped_type >   leaf_node;
+                typedef typename leaf_node::value_type          value_type;
+
+                typedef Traits  original_traits;
+
+                struct assignment_copy_policy {
+                    void operator()( key_type& dest, key_type const& src )
+                    {
+                        dest = src;
+                    }
+                };
+                typedef typename std::conditional<
+                    std::is_same< typename original_traits::copy_policy, opt::none >::value,
+                    assignment_copy_policy,
+                    typename original_traits::copy_policy
+                >::type copy_policy;
+
+                struct intrusive_key_extractor
+                {
+                    void operator()( key_type& dest, leaf_node const& src ) const
+                    {
+                        copy_policy()( dest, src.m_Value.first );
+                    }
+                };
+
+                struct key_accessor
+                {
+                    key_type const& operator()( leaf_node const& node ) const
+                    {
+                        return node.m_Value.first;
+                    }
+                };
+
+                typedef typename cds::opt::details::make_comparator< key_type, original_traits, false >::type key_comparator;
+
+                typedef cds::details::Allocator< leaf_node, typename original_traits::allocator>    cxx_leaf_node_allocator;
+                struct leaf_deallocator
+                {
+                    void operator()( leaf_node * p ) const
+                    {
+                        cxx_leaf_node_allocator().Delete( p );
+                    }
+                };
+
+                struct intrusive_traits: public original_traits
+                {
+                    typedef cds::intrusive::ellen_bintree::base_hook< cds::opt::gc< gc > >  hook;
+                    typedef intrusive_key_extractor key_extractor;
+                    typedef leaf_deallocator        disposer;
+                    typedef cds::details::compare_wrapper< leaf_node, key_comparator, key_accessor >    compare;
+                };
+
+                // Metafunction result
+                typedef cds::intrusive::EllenBinTree< gc, key_type, leaf_node, intrusive_traits >    type;
+            };
+
+        } // namespace details
+        //@endcond
+    } // namespace ellen_bintree
+
+    // Forward declarations
+    //@cond
+    template < class GC, typename Key, typename T, class Traits = ellen_bintree::traits >
+    class EllenBinTreeSet;
+
+    template < class GC, typename Key, typename T, class Traits = ellen_bintree::traits >
+    class EllenBinTreeMap;
+    //@endcond
+
+}} // namespace cds::container
+
+#endif // #ifndef CDSLIB_CONTAINER_DETAILS_ELLEN_BINTREE_BASE_H

extern/libcds/cds/container/details/feldman_hashmap_base.h

@@ -0,0 +1,366 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_DETAILS_FELDMAN_HASHMAP_BASE_H
+#define CDSLIB_CONTAINER_DETAILS_FELDMAN_HASHMAP_BASE_H
+
+#include <cds/intrusive/details/feldman_hashset_base.h>
+#include <cds/container/details/base.h>
+#include <cds/opt/hash.h>
+
+namespace cds { namespace container {
+    /// \p FeldmanHashMap related definitions
+    /** @ingroup cds_nonintrusive_helper
+    */
+    namespace feldman_hashmap {
+        /// \p FeldmanHashMap internal statistics, see cds::intrusive::feldman_hashset::stat
+        template <typename EventCounter = cds::atomicity::event_counter>
+        using stat = cds::intrusive::feldman_hashset::stat< EventCounter >;
+
+        /// \p FeldmanHashMap empty internal statistics
+        typedef cds::intrusive::feldman_hashset::empty_stat empty_stat;
+
+        /// Bit-wise memcmp-based comparator for hash value \p T
+        template <typename T>
+        using bitwise_compare = cds::intrusive::feldman_hashset::bitwise_compare< T >;
+
+        /// \p FeldmanHashMap level statistics
+        typedef cds::intrusive::feldman_hashset::level_statistics level_statistics;
+
+        /// Key size option
+        /**
+            @copydetails cds::container::feldman_hashmap::traits::hash_size
+        */
+        template <size_t Size>
+        using hash_size = cds::intrusive::feldman_hashset::hash_size< Size >;
+
+        /// Hash splitter option
+        /**
+            @copydetails cds::container::feldman_hashmap::traits::hash_splitter
+        */
+        template <typename Splitter>
+        using hash_splitter = cds::intrusive::feldman_hashset::hash_splitter< Splitter >;
+
+
+        /// \p FeldmanHashMap traits
+        struct traits
+        {
+            /// Hash functor, default is \p opt::none
+            /**
+                \p FeldmanHashMap may use any hash functor converting a key to
+                fixed-sized bit-string, for example, <a href="https://en.wikipedia.org/wiki/Secure_Hash_Algorithm">SHA1, SHA2</a>,
+                <a href="https://en.wikipedia.org/wiki/MurmurHash">MurmurHash</a>,
+                <a href="https://en.wikipedia.org/wiki/CityHash">CityHash</a>
+                or its successor <a href="https://code.google.com/p/farmhash/">FarmHash</a>.
+
+                If you use a fixed-sized key you can use it directly instead of a hash.
+                In such case \p %traits::hash should be specified as \p opt::none.
+                However, if you want to use the hash values or if your key type is not fixed-sized
+                you must specify a proper hash functor in your traits.
+                For example:
+                fixed-sized key - IP4 address map
+                @code
+                    // Key - IP address
+                    struct ip4_address {
+                        uint8_t ip[4];
+                    };
+
+                    // IP compare
+                    struct ip4_cmp {
+                        int operator()( ip4_address const& lhs, ip4_address const& rhs ) const
+                        {
+                            return memcmp( &lhs, &rhs, sizeof(lhs));
+                        }
+                    };
+
+                    // Value - statistics for the IP address
+                    struct statistics {
+                        // ...
+                    };
+
+                    // Traits
+                    // Key type (ip4_addr) is fixed-sized so we may use the map without any hash functor
+                    struct ip4_map_traits: public cds::container::multilevl_hashmap::traits
+                    {
+                        typedef ip4_cmp  compare;
+                    };
+
+                    // IP4 address - statistics map
+                    typedef cds::container::FeldmanHashMap< cds::gc::HP, ip4_address, statistics, ip4_map_traits > ip4_map;
+                @endcode
+
+                variable-size key requires a hash functor: URL map
+                @code
+                    // Value - statistics for the URL
+                    struct statistics {
+                        // ...
+                    };
+
+                    // Traits
+                    // Key type (std::string) is variable-sized so we must provide a hash functor in our traits
+                    // We do not specify any comparing predicate (less or compare) so <tt> std::less<std::string> </tt> will be used by default
+                    struct url_map_traits: public cds::container::multilevl_hashmap::traits
+                    {
+                        typedef std::hash<std::string> hash;
+                    };
+
+                    // URL statistics map
+                    typedef cds::container::FeldmanHashMap< cds::gc::HP, std::string, statistics, url_map_traits > url_map;
+                @endcode
+            */
+            typedef opt::none hash;
+
+            /// The size of hash value in bytes
+            /**
+                By default, the size of hash value is <tt>sizeof( hash_type )</tt>
+                where \p hash_type is type of \p hash() result or <tt>sizeof( key )</tt> if you use fixed-sized key.
+
+                Sometimes that size is wrong, for example, for that 6-byte key:
+                \code
+                struct key_type {
+                    uint32_t    key;
+                    uint16_t    subkey;
+                };
+
+                static_assert( sizeof( key_type ) == 6, "Key type size mismatch" );
+                \endcode
+                Here <tt>sizeof( key_type ) == 8</tt> so \p static_assert will be thrown.
+
+                For that case you can specify \p hash_size explicitly.
+
+                Value \p 0 means auto-calculated <tt>sizeof( key_type )</tt>.
+            */
+            static constexpr size_t const hash_size = 0;
+
+            /// Hash splitter
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::hash_splitter
+            */
+            typedef cds::opt::none hash_splitter;
+
+            /// Hash comparing functor
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::compare
+            */
+            typedef cds::opt::none compare;
+
+            /// Specifies binary predicate used for hash compare.
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::less
+            */
+            typedef cds::opt::none less;
+
+            /// Item counter
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::item_counter
+            */
+            typedef cds::atomicity::item_counter item_counter;
+
+            /// Item allocator
+            /**
+                Default is \ref CDS_DEFAULT_ALLOCATOR
+            */
+            typedef CDS_DEFAULT_ALLOCATOR allocator;
+
+            /// Array node allocator
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::node_allocator
+            */
+            typedef CDS_DEFAULT_ALLOCATOR node_allocator;
+
+            /// C++ memory ordering model
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::memory_model
+            */
+            typedef cds::opt::v::relaxed_ordering memory_model;
+
+            /// Back-off strategy
+            typedef cds::backoff::Default back_off;
+
+            /// Internal statistics
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::stat
+            */
+            typedef empty_stat stat;
+
+            /// RCU deadlock checking policy (only for \ref cds_container_FeldmanHashMap_rcu "RCU-based FeldmanHashMap")
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::rcu_check_deadlock
+            */
+            typedef cds::opt::v::rcu_throw_deadlock rcu_check_deadlock;
+        };
+
+        /// Metafunction converting option list to \p feldman_hashmap::traits
+        /**
+            Supported \p Options are:
+            - \p opt::hash - a hash functor, default is \p std::hash
+                @copydetails traits::hash
+            - \p feldman_hashmap::hash_size - the size of hash value in bytes.
+                @copydetails traits::hash_size
+            - \p opt::allocator - item allocator
+                @copydetails traits::allocator
+            - \p opt::node_allocator - array node allocator.
+                @copydetails traits::node_allocator
+            - \p opt::compare - hash comparison functor. No default functor is provided.
+                If the option is not specified, the \p opt::less is used.
+            - \p opt::less - specifies binary predicate used for hash comparison.
+                @copydetails cds::container::feldman_hashmap::traits::less
+            - \p opt::back_off - back-off strategy used. If the option is not specified, the \p cds::backoff::Default is used.
+            - \p opt::item_counter - the type of item counting feature.
+                @copydetails cds::container::feldman_hashmap::traits::item_counter
+            - \p opt::memory_model - C++ memory ordering model. Can be \p opt::v::relaxed_ordering (relaxed memory model, the default)
+                or \p opt::v::sequential_consistent (sequentially consisnent memory model).
+            - \p opt::stat - internal statistics. By default, it is disabled (\p feldman_hashmap::empty_stat).
+                To enable it use \p feldman_hashmap::stat
+            - \p opt::rcu_check_deadlock - a deadlock checking policy for \ref cds_intrusive_FeldmanHashSet_rcu "RCU-based FeldmanHashSet"
+                Default is \p opt::v::rcu_throw_deadlock
+        */
+        template <typename... Options>
+        struct make_traits
+        {
+#   ifdef CDS_DOXYGEN_INVOKED
+            typedef implementation_defined type ;   ///< Metafunction result
+#   else
+            typedef typename cds::opt::make_options<
+                typename cds::opt::find_type_traits< traits, Options... >::type
+                ,Options...
+            >::type   type;
+#   endif
+        };
+    } // namespace feldman_hashmap
+
+    //@cond
+    // Forward declaration
+    template < class GC, typename Key, typename T, class Traits = feldman_hashmap::traits >
+    class FeldmanHashMap;
+    //@endcond
+
+    //@cond
+    namespace details {
+
+        template <typename Key, typename Value, typename Hash>
+        struct hash_selector
+        {
+            typedef Key key_type;
+            typedef Value mapped_type;
+            typedef Hash hasher;
+
+            typedef typename std::decay<
+                typename std::remove_reference<
+                decltype(hasher()(std::declval<key_type>()))
+                >::type
+            >::type hash_type;
+
+            struct node_type
+            {
+                std::pair< key_type const, mapped_type> m_Value;
+                hash_type const m_hash;
+
+                node_type() = delete;
+                node_type(node_type const&) = delete;
+
+                template <typename Q>
+                node_type(hasher& h, Q const& key)
+                    : m_Value( std::move( std::make_pair( key_type( key ), mapped_type())))
+                    , m_hash( h( m_Value.first ))
+                {}
+
+                template <typename Q, typename U >
+                node_type(hasher& h, Q const& key, U const& val)
+                    : m_Value( std::move( std::make_pair( key_type( key ), mapped_type(val))))
+                    , m_hash( h( m_Value.first ))
+                {}
+
+                template <typename Q, typename... Args>
+                node_type(hasher& h, Q&& key, Args&&... args)
+                    : m_Value( std::move(std::make_pair( key_type( std::forward<Q>(key)), std::move( mapped_type(std::forward<Args>(args)...)))))
+                    , m_hash( h( m_Value.first ))
+                {}
+            };
+
+            struct hash_accessor
+            {
+                hash_type const& operator()(node_type const& node) const
+                {
+                    return node.m_hash;
+                }
+            };
+        };
+
+        template <typename Key, typename Value>
+        struct hash_selector<Key, Value, opt::none>
+        {
+            typedef Key key_type;
+            typedef Value mapped_type;
+
+            struct hasher {
+                key_type const& operator()(key_type const& k) const
+                {
+                    return k;
+                }
+            };
+            typedef key_type hash_type;
+
+            struct node_type
+            {
+                std::pair< key_type const, mapped_type> m_Value;
+
+                node_type() = delete;
+                node_type(node_type const&) = delete;
+
+                template <typename Q, typename... Args>
+                node_type( hasher /*h*/, Q&& key, Args&&... args )
+                    : m_Value( std::make_pair( key_type( std::forward<Q>( key )), mapped_type( std::forward<Args>(args)...)))
+                {}
+            };
+
+            struct hash_accessor
+            {
+                hash_type const& operator()(node_type const& node) const
+                {
+                    return node.m_Value.first;
+                }
+            };
+        };
+
+        template <typename GC, typename Key, typename T, typename Traits>
+        struct make_feldman_hashmap
+        {
+            typedef GC      gc;
+            typedef Key     key_type;
+            typedef T       mapped_type;
+            typedef Traits  original_traits;
+
+
+            typedef hash_selector< key_type, mapped_type, typename original_traits::hash > select;
+            typedef typename select::hasher    hasher;
+            typedef typename select::hash_type hash_type;
+            typedef typename select::node_type node_type;
+
+            typedef cds::details::Allocator< node_type, typename original_traits::allocator > cxx_node_allocator;
+
+            struct node_disposer
+            {
+                void operator()( node_type * p ) const
+                {
+                    cxx_node_allocator().Delete( p );
+                }
+            };
+
+            struct intrusive_traits: public original_traits
+            {
+                typedef typename select::hash_accessor hash_accessor;
+                typedef node_disposer disposer;
+            };
+
+            // Metafunction result
+            typedef cds::intrusive::FeldmanHashSet< GC, node_type, intrusive_traits > type;
+        };
+    } // namespace details
+    //@endcond
+
+}} // namespace cds::container
+
+#endif // #ifndef CDSLIB_CONTAINER_DETAILS_FELDMAN_HASHMAP_BASE_H

extern/libcds/cds/container/details/feldman_hashset_base.h

@@ -0,0 +1,205 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_DETAILS_FELDMAN_HASHSET_BASE_H
+#define CDSLIB_CONTAINER_DETAILS_FELDMAN_HASHSET_BASE_H
+
+#include <cds/intrusive/details/feldman_hashset_base.h>
+#include <cds/container/details/base.h>
+
+namespace cds { namespace container {
+    /// \p FeldmanHashSet related definitions
+    /** @ingroup cds_nonintrusive_helper
+    */
+    namespace feldman_hashset {
+        /// Hash accessor option
+        /**
+            @copydetails cds::intrusive::feldman_hashset::traits::hash_accessor
+        */
+        template <typename Accessor>
+        using hash_accessor = cds::intrusive::feldman_hashset::hash_accessor< Accessor >;
+
+        /// Hash size option
+        /**
+            @copydetails cds::intrusive::feldman_hashset::traits::hash_size
+        */
+        template <size_t Size>
+        using hash_size = cds::intrusive::feldman_hashset::hash_size< Size >;
+
+        /// Hash splitter
+        /**
+            @copydetails cds::intrusive::feldman_hashset::traits::hash_splitter
+        */
+        template <typename Splitter>
+        using hash_splitter = cds::intrusive::feldman_hashset::hash_splitter< Splitter >;
+
+        /// \p FeldmanHashSet internal statistics, see cds::intrusive::feldman_hashset::stat
+        template <typename EventCounter = cds::atomicity::event_counter>
+        using stat = cds::intrusive::feldman_hashset::stat< EventCounter >;
+
+        /// \p FeldmanHashSet empty internal statistics
+        typedef cds::intrusive::feldman_hashset::empty_stat empty_stat;
+
+        /// Bit-wise memcmp-based comparator for hash value \p T
+        template <typename T>
+        using bitwise_compare = cds::intrusive::feldman_hashset::bitwise_compare< T >;
+
+        /// \p FeldmanHashSet level statistics
+        typedef cds::intrusive::feldman_hashset::level_statistics level_statistics;
+
+        /// \p FeldmanHashSet traits
+        struct traits
+        {
+            /// Mandatory functor to get hash value from data node
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::hash_accessor
+            */
+            typedef cds::opt::none hash_accessor;
+
+            /// The size of hash value in bytes
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::hash_size
+            */
+            static constexpr size_t const hash_size = 0;
+
+            /// Hash splitter
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::hash_splitter
+            */
+            typedef cds::opt::none hash_splitter;
+
+            /// Hash comparing functor
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::compare
+            */
+            typedef cds::opt::none compare;
+
+            /// Specifies binary predicate used for hash compare.
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::less
+            */
+            typedef cds::opt::none less;
+
+            /// Item counter
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::item_counter
+            */
+            typedef cds::atomicity::item_counter item_counter;
+
+            /// Item allocator
+            /**
+                Default is \ref CDS_DEFAULT_ALLOCATOR
+            */
+            typedef CDS_DEFAULT_ALLOCATOR allocator;
+
+            /// Array node allocator
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::node_allocator
+            */
+            typedef CDS_DEFAULT_ALLOCATOR node_allocator;
+
+            /// C++ memory ordering model
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::memory_model
+            */
+            typedef cds::opt::v::relaxed_ordering memory_model;
+
+            /// Back-off strategy
+            typedef cds::backoff::Default back_off;
+
+            /// Internal statistics
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::stat
+            */
+            typedef empty_stat stat;
+
+            /// RCU deadlock checking policy (only for \ref cds_container_FeldmanHashSet_rcu "RCU-based FeldmanHashSet")
+            /**
+                @copydetails cds::intrusive::feldman_hashset::traits::rcu_check_deadlock
+            */
+            typedef cds::opt::v::rcu_throw_deadlock rcu_check_deadlock;
+        };
+
+        /// Metafunction converting option list to \p feldman_hashset::traits
+        /**
+            Supported \p Options are:
+            - \p feldman_hashset::hash_accessor - mandatory option, hash accessor functor.
+                @copydetails traits::hash_accessor
+            - \p feldman_hashset::hash_size - the size of hash value in bytes.
+                @copydetails traits::hash_size
+            - \p feldman_hashset::hash_splitter - a hash splitter algorithm
+                @copydetails traits::hash_splitter
+            - \p opt::allocator - item allocator
+                @copydetails traits::allocator
+            - \p opt::node_allocator - array node allocator.
+                @copydetails traits::node_allocator
+            - \p opt::compare - hash comparison functor. No default functor is provided.
+                If the option is not specified, the \p opt::less is used.
+            - \p opt::less - specifies binary predicate used for hash comparison.
+                @copydetails cds::container::feldman_hashset::traits::less
+            - \p opt::back_off - back-off strategy used. If the option is not specified, the \p cds::backoff::Default is used.
+            - \p opt::item_counter - the type of item counting feature.
+                @copydetails cds::intrusive::feldman_hashset::traits::item_counter
+            - \p opt::memory_model - C++ memory ordering model. Can be \p opt::v::relaxed_ordering (relaxed memory model, the default)
+                or \p opt::v::sequential_consistent (sequentially consisnent memory model).
+            - \p opt::stat - internal statistics. By default, it is disabled (\p feldman_hashset::empty_stat).
+                To enable it use \p feldman_hashset::stat
+            - \p opt::rcu_check_deadlock - a deadlock checking policy for \ref cds_intrusive_FeldmanHashSet_rcu "RCU-based FeldmanHashSet"
+                Default is \p opt::v::rcu_throw_deadlock
+        */
+        template <typename... Options>
+        struct make_traits
+        {
+#   ifdef CDS_DOXYGEN_INVOKED
+            typedef implementation_defined type ;   ///< Metafunction result
+#   else
+            typedef typename cds::opt::make_options<
+                typename cds::opt::find_type_traits< traits, Options... >::type
+                ,Options...
+            >::type   type;
+#   endif
+        };
+    } // namespace feldman_hashset
+
+    //@cond
+    // Forward declaration
+    template < class GC, typename T, class Traits = cds::container::feldman_hashset::traits >
+    class FeldmanHashSet;
+    //@endcond
+
+    //@cond
+    namespace details {
+
+        template <typename GC, typename T, typename Traits>
+        struct make_feldman_hashset
+        {
+            typedef GC      gc;
+            typedef T       value_type;
+            typedef Traits  original_traits;
+
+            typedef cds::details::Allocator< value_type, typename original_traits::allocator > cxx_node_allocator;
+
+            struct node_disposer
+            {
+                void operator()( value_type * p ) const
+                {
+                    cxx_node_allocator().Delete( p );
+                }
+            };
+
+            struct intrusive_traits: public original_traits
+            {
+                typedef node_disposer disposer;
+            };
+
+            // Metafunction result
+            typedef cds::intrusive::FeldmanHashSet< GC, T, intrusive_traits > type;
+        };
+    } // namespace details
+    //@endcond
+
+}} // namespace cds::container
+
+#endif // #ifndef CDSLIB_CONTAINER_DETAILS_FELDMAN_HASHSET_BASE_H

extern/libcds/cds/container/details/guarded_ptr_cast.h

@@ -0,0 +1,33 @@
+// Copyright (c) 2006-2018 Maxim Khizhinsky
+//
+// Distributed under the Boost Software License, Version 1.0. (See accompanying
+// file LICENSE or copy at http://www.boost.org/LICENSE_1_0.txt)
+
+#ifndef CDSLIB_CONTAINER_DETAILS_GUARDED_PTR_CAST_H
+#define CDSLIB_CONTAINER_DETAILS_GUARDED_PTR_CAST_H
+//@cond
+
+#include <cds/details/defs.h>
+
+namespace cds { namespace container { namespace details {
+
+    template <typename Node, typename T>
+    struct guarded_ptr_cast_set {
+        T * operator()(Node* pNode ) const noexcept
+        {
+            return &(pNode->m_Value);
+        }
+    };
+
+    template <typename Node, typename T>
+    struct guarded_ptr_cast_map {
+        T * operator()(Node* pNode ) const noexcept
+        {
+            return &(pNode->m_Data);
+        }
+    };
+
+}}} // namespace cds::container::details
+
+//@endcond
+#endif // #ifndef CDSLIB_CONTAINER_DETAILS_GUARDED_PTR_CAST_H