Windows port - improve performance by using segment heap

When sysbench is run with many threads, a lot of CPU is spent on memory
allocation and freeing. Profiler shows as much as 80 percent in malloc/free
code, when oltp_read_only is run with 1000 threads.

This does not happen when using SegmentHeap. There, at most 25% time is
spend on allocation and freeing, and the OLTP tps/qps numbers are better.

This patch uses windows application manifest to define heapType=SegmentHeap
for the sysbench executable. If not desired, -DWITH_SEGMENT_HEAP=0 cmake
parameter will disable the segment heap.

src/win/CMakeLists.txt

@@ -1,4 +1,8 @@
 target_sources(sysbench PRIVATE
  ${CMAKE_CURRENT_SOURCE_DIR}/sb_win_posix.h
  ${CMAKE_CURRENT_SOURCE_DIR}/sb_win_posix.c)
+option(WITH_SEGMENT_HEAP "Use Windows segment heap, for better performance" ON)
+if(WITH_SEGMENT_HEAP)
+  target_sources(sysbench PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/segment_heap.manifest)
+endif()
 target_include_directories(sysbench PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})

src/win/segment_heap.manifest

@@ -0,0 +1,7 @@
+<assembly xmlns="urn:schemas-microsoft-com:asm.v1" manifestVersion="1.0" xmlns:asmv3="urn:schemas-microsoft-com:asm.v3">
+  <asmv3:application>
+    <asmv3:windowsSettings xmlns="http://schemas.microsoft.com/SMI/2020/WindowsSettings">
+      <heapType>SegmentHeap</heapType>
+    </asmv3:windowsSettings>
+  </asmv3:application>
+</assembly>