Add support for the TBB allocator on windows.

The heap on windows is single threaded causing it to lag behind linux in performance in allocation heavy multithreaded scenarios, BVH building is a prime example.

See https://developer.blender.org/D6218 for benchmark results

for testing with the allocator enabled/disabled you can set the environment variable TBB_MALLOC_DISABLE_REPLACEMENT=1 to disable the TBB allocator.

Reviewed By: @sergey

Differential Revision: https://developer.blender.org/D6218

CMakeLists.txt

@@ -170,6 +170,14 @@ option_defaults_init(
   _init_OPENSUBDIV
 )
 
+# TBB malloc is only supported on for windows currently
+if(WIN32)
+  set(_init_TBB_MALLOC_PROXY ON)
+else()
+  set(_init_TBB_MALLOC_PROXY OFF)
+endif()
+
+
 # customize...
 if(UNIX AND NOT APPLE)
   # some of these libraries are problematic on Linux
@@ -463,6 +471,7 @@ mark_as_advanced(WITH_ASSERT_ABORT)
 
 option(WITH_BOOST "Enable features depending on boost" ON)
 option(WITH_TBB   "Enable features depending on TBB (OpenVDB, OpenImageDenoise, sculpt multithreading)" ON)
+option(WITH_TBB_MALLOC_PROXY "Enable the TBB malloc replacement" ${_init_TBB_MALLOC_PROXY})
 
 # Unit testsing
 option(WITH_GTESTS "Enable GTest unit testing" OFF)

build_files/build_environment/cmake/tbb.cmake

@@ -15,13 +15,21 @@
 # Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
 #
 # ***** END GPL LICENSE BLOCK *****
-
+if(WIN32)
-set(TBB_EXTRA_ARGS
+  set(TBB_EXTRA_ARGS
-  -DTBB_BUILD_SHARED=Off
+    -DTBB_BUILD_SHARED=On
-  -DTBB_BUILD_TBBMALLOC=On
+    -DTBB_BUILD_TBBMALLOC=On
-  -DTBB_BUILD_TBBMALLOC_PROXY=Off
+    -DTBB_BUILD_TBBMALLOC_PROXY=On
-  -DTBB_BUILD_STATIC=On
+    -DTBB_BUILD_STATIC=On
 )
+else()
+  set(TBB_EXTRA_ARGS
+    -DTBB_BUILD_SHARED=Off
+    -DTBB_BUILD_TBBMALLOC=On
+    -DTBB_BUILD_TBBMALLOC_PROXY=Off
+    -DTBB_BUILD_STATIC=On
+  )
+endif()
 
 # CMake script for TBB from https://github.com/wjakob/tbb/blob/master/CMakeLists.txt
 ExternalProject_Add(external_tbb
@@ -39,6 +47,10 @@ if(WIN32)
   if(BUILD_MODE STREQUAL Release)
     ExternalProject_Add_Step(external_tbb after_install
       COMMAND ${CMAKE_COMMAND} -E copy ${LIBDIR}/tbb/lib/tbb_static.lib ${HARVEST_TARGET}/tbb/lib/tbb.lib
+      COMMAND ${CMAKE_COMMAND} -E copy ${LIBDIR}/tbb/lib/tbbmalloc.lib ${HARVEST_TARGET}/tbb/lib/tbbmalloc.lib
+      COMMAND ${CMAKE_COMMAND} -E copy ${LIBDIR}/tbb/lib/tbbmalloc.dll ${HARVEST_TARGET}/tbb/lib/tbbmalloc.dll
+      COMMAND ${CMAKE_COMMAND} -E copy ${LIBDIR}/tbb/lib/tbbmalloc_proxy.lib ${HARVEST_TARGET}/tbb/lib/tbbmalloc_proxy.lib
+      COMMAND ${CMAKE_COMMAND} -E copy ${LIBDIR}/tbb/lib/tbbmalloc_proxy.dll ${HARVEST_TARGET}/tbb/lib/tbbmalloc_proxy.dll
       COMMAND ${CMAKE_COMMAND} -E copy_directory ${LIBDIR}/tbb/include/ ${HARVEST_TARGET}/tbb/include/
       DEPENDEES install
     )
@@ -46,6 +58,7 @@ if(WIN32)
   if(BUILD_MODE STREQUAL Debug)
     ExternalProject_Add_Step(external_tbb after_install
       COMMAND ${CMAKE_COMMAND} -E copy ${LIBDIR}/tbb/lib/tbb_static.lib ${HARVEST_TARGET}/tbb/lib/tbb_debug.lib
+      COMMAND ${CMAKE_COMMAND} -E copy ${LIBDIR}/tbb/lib/tbbmalloc_proxy.lib ${HARVEST_TARGET}/tbb/lib/tbbmalloc_proxy_debug.lib
       DEPENDEES install
     )
   endif()

build_files/build_environment/cmake/versions.cmake

@@ -149,9 +149,9 @@ set(PYTHON_SHORT_VERSION_NO_DOTS 37)
 set(PYTHON_URI https://www.python.org/ftp/python/${PYTHON_VERSION}/Python-${PYTHON_VERSION}.tar.xz)
 set(PYTHON_HASH d33e4aae66097051c2eca45ee3604803)
 
-set(TBB_VERSION 2018_U5)
+set(TBB_VERSION 2019_U9)
 set(TBB_URI https://github.com/01org/tbb/archive/${TBB_VERSION}.tar.gz)
-set(TBB_HASH ff3ae09f8c23892fbc3008c39f78288f)
+set(TBB_HASH 584edbec127c508f2cd5b6e79ad200fc)
 
 set(OPENVDB_VERSION 5.1.0)
 set(OPENVDB_URI https://github.com/dreamworksanimation/openvdb/archive/v${OPENVDB_VERSION}.tar.gz)

build_files/cmake/platform/platform_win32.cmake

@@ -573,6 +573,9 @@ if(WITH_TBB)
   set(TBB_LIBRARIES optimized ${LIBDIR}/tbb/lib/tbb.lib debug ${LIBDIR}/tbb/lib/tbb_debug.lib)
   set(TBB_INCLUDE_DIR ${LIBDIR}/tbb/include)
   set(TBB_INCLUDE_DIRS ${TBB_INCLUDE_DIR})
+  if(WITH_TBB_MALLOC_PROXY)
+    add_definitions(-DWITH_TBB_MALLOC)
+  endif()
 else()
   if(WITH_OPENIMAGEDENOISE)
     message(STATUS "TBB disabled, also disabling OpenImageDenoise")

source/creator/CMakeLists.txt

@@ -81,6 +81,12 @@ if(WITH_CODEC_FFMPEG)
   add_definitions(-DWITH_FFMPEG)
 endif()
 
+if(WITH_TBB)
+  blender_include_dirs(${TBB_INCLUDE_DIRS})
+  link_directories(${LIBDIR}/tbb/lib)
+endif()
+
+
 if(WITH_PYTHON)
   blender_include_dirs(../blender/python)
   add_definitions(-DWITH_PYTHON)
@@ -806,6 +812,15 @@ elseif(WIN32)
     )
   endif()
 
+  if(WITH_TBB_MALLOC_PROXY)
+    install(
+      FILES
+        ${LIBDIR}/tbb/lib/tbbmalloc.dll
+        ${LIBDIR}/tbb/lib/tbbmalloc_proxy.dll
+      DESTINATION "."
+    )
+  endif()
+
   if(WITH_CODEC_SNDFILE)
     install(
       FILES ${LIBDIR}/sndfile/lib/libsndfile-1.dll

source/creator/creator.c

@@ -29,6 +29,11 @@
 #  include "utfconv.h"
 #endif
 
+#if defined(WITH_TBB_MALLOC) && defined(_MSC_VER)
+#  pragma comment(lib, "tbbmalloc_proxy.lib")
+#  pragma comment(linker, "/include:__TBB_malloc_proxy")
+#endif
+
 #include "MEM_guardedalloc.h"
 
 #include "CLG_log.h"