From d3af0b04b263e179fda8284c594bf3e6adb1462d Mon Sep 17 00:00:00 2001
From: Julian Waller <git@julusian.co.uk>
Date: Fri, 24 May 2024 14:37:15 +0100
Subject: [PATCH] Revert "fix: replace tbbmalloc with os native
 implementations"

tbbmalloc is necessary on windows to get satisfactory performance in many cases #1537

This reverts commit 3816e08f5b89cd1139a4c46a47bb1aef2b6b15f4.
---
 src/CMakeModules/Bootstrap_Windows.cmake | 4 ++++
 src/common/memshfl.h                     | 3 ++-
 src/shell/CMakeLists.txt                 | 2 ++
 src/shell/main.cpp                       | 7 +++++++
 4 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/src/CMakeModules/Bootstrap_Windows.cmake b/src/CMakeModules/Bootstrap_Windows.cmake
index a7d22f0d8c..e47d0d5a3f 100644
--- a/src/CMakeModules/Bootstrap_Windows.cmake
+++ b/src/CMakeModules/Bootstrap_Windows.cmake
@@ -109,6 +109,10 @@ set(TBB_BIN_PATH "${SOURCE_DIR}/redist/intel64/vc14")
 link_directories("${SOURCE_DIR}/lib/intel64/vc14")
 casparcg_add_runtime_dependency("${TBB_BIN_PATH}/tbb12.dll" "Release")
 casparcg_add_runtime_dependency("${TBB_BIN_PATH}/tbb12_debug.dll" "Debug")
+casparcg_add_runtime_dependency("${TBB_BIN_PATH}/tbbmalloc.dll" "Release")
+casparcg_add_runtime_dependency("${TBB_BIN_PATH}/tbbmalloc_debug.dll" "Debug")
+casparcg_add_runtime_dependency("${TBB_BIN_PATH}/tbbmalloc_proxy.dll" "Release")
+casparcg_add_runtime_dependency("${TBB_BIN_PATH}/tbbmalloc_proxy_debug.dll" "Debug")
 
 # GLEW
 casparcg_add_external_project(glew)
diff --git a/src/common/memshfl.h b/src/common/memshfl.h
index 3947e9867f..8c69e23302 100644
--- a/src/common/memshfl.h
+++ b/src/common/memshfl.h
@@ -26,6 +26,7 @@
 #include <simde/x86/ssse3.h>
 #else
 #ifdef _MSC_VER
+#include <tbb/scalable_allocator.h>
 #include <intrin.h>
 #else
 #include <tmmintrin.h>
@@ -37,7 +38,7 @@ namespace caspar {
 #ifdef _MSC_VER
 static std::shared_ptr<void> create_aligned_buffer(size_t size)
 {
-    return std::shared_ptr<void>(_aligned_malloc(size, 64), _aligned_free);
+    return std::shared_ptr<void>(scalable_aligned_malloc(size, 64), scalable_aligned_free);
 }
 #else
 static std::shared_ptr<void> create_aligned_buffer(size_t size)
diff --git a/src/shell/CMakeLists.txt b/src/shell/CMakeLists.txt
index a7c4857536..f35aa25dab 100644
--- a/src/shell/CMakeLists.txt
+++ b/src/shell/CMakeLists.txt
@@ -55,7 +55,9 @@ if (MSVC)
 		Winmm.lib
 		Ws2_32.lib
 		optimized tbb.lib
+		optimized tbbmalloc.lib
 		debug tbb_debug.lib
+		debug tbbmalloc_debug.lib
 		OpenGL32.lib
 		glew32.lib
 		openal32.lib
diff --git a/src/shell/main.cpp b/src/shell/main.cpp
index 8c9ebe679f..eea4404c81 100644
--- a/src/shell/main.cpp
+++ b/src/shell/main.cpp
@@ -19,10 +19,17 @@
  * Author: Robert Nagy, ronag89@gmail.com
  */
 
+// tbbmalloc_proxy:
+// Replace the standard memory allocation routines in Microsoft* C/C++ RTL
+// (malloc/free, global new/delete, etc.) with the TBB memory allocator
+// as the default allocator suffers from low performance.
+
 #if defined _DEBUG && defined _MSC_VER
 #define _CRTDBG_MAP_ALLOC
 #include <crtdbg.h>
 #include <stdlib.h>
+#elif defined _MSC_VER
+#include <tbb/tbbmalloc_proxy.h>
 #endif
 
 #include "included_modules.h"