clarify to use as the PR branch

atomic: fix mi_atomic_yield for big-endian arm32

.gitignore

@@ -1,3 +1,4 @@
+build
 ide/vs20??/*.db
 ide/vs20??/*.opendb
 ide/vs20??/*.user
@@ -8,3 +9,5 @@ docs/
 *.zip
 *.tar
 *.gz
+.vscode
+.DS_STore

CMakeLists.txt

@@ -14,7 +14,7 @@ option(MI_TRACK_VALGRIND    "Compile with Valgrind support (adds a small overhea
 option(MI_TRACK_ASAN        "Compile with address sanitizer support (adds a small overhead)" OFF)
 option(MI_TRACK_ETW         "Compile with Windows event tracing (ETW) support (adds a small overhead)" OFF)
 option(MI_USE_CXX           "Use the C++ compiler to compile the library (instead of the C compiler)" OFF)
-option(MI_OPT_ARCH          "Only for optimized builds: turn on architecture specific optimizations (for arm64: '-march=armv8.1-a' (2016))" ON)
+option(MI_OPT_ARCH          "Only for optimized builds: turn on architecture specific optimizations (for arm64: '-march=armv8.1-a' (2016))" OFF)
 option(MI_SEE_ASM           "Generate assembly files" OFF)
 option(MI_OSX_INTERPOSE     "Use interpose to override standard malloc on macOS" ON)
 option(MI_OSX_ZONE          "Use malloc zone to override standard malloc on macOS" ON)
@@ -35,6 +35,10 @@ option(MI_INSTALL_TOPLEVEL  "Install directly into $CMAKE_INSTALL_PREFIX instead
 option(MI_NO_THP            "Disable transparent huge pages support on Linux/Android for the mimalloc process only" OFF)
 option(MI_EXTRA_CPPDEFS     "Extra pre-processor definitions (use as `-DMI_EXTRA_CPPDEFS=\"opt1=val1;opt2=val2\"`)" "")
 
+# negated options for vcpkg features
+option(MI_NO_USE_CXX        "Use plain C compilation (has priority over MI_USE_CXX)" OFF)
+option(MI_NO_OPT_ARCH       "Do not use architecture specific optimizations (like '-march=armv8.1-a' for example) (has priority over MI_OPT_ARCH)" OFF)
+
 # deprecated options
 option(MI_WIN_USE_FLS       "Use Fiber local storage on Windows to detect thread termination (deprecated)" OFF)
 option(MI_CHECK_FULL        "Use full internal invariant checking in DEBUG mode (deprecated, use MI_DEBUG_FULL instead)" OFF)
@@ -74,6 +78,19 @@ else()
  set(mi_defines "")
 endif()
 
+# pass git revision as a define
+if(EXISTS "${CMAKE_SOURCE_DIR}/.git/index")
+  find_package(Git)
+  if(GIT_FOUND)
+    execute_process(COMMAND ${GIT_EXECUTABLE} "describe" OUTPUT_VARIABLE mi_git_describe RESULT_VARIABLE mi_git_res ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
+    if(mi_git_res EQUAL "0")
+      list(APPEND mi_defines "MI_GIT_DESCRIBE=${mi_git_describe}")
+      # add to dependencies so we rebuild if the git head commit changes
+      set_property(GLOBAL APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS "${CMAKE_SOURCE_DIR}/.git/index")
+    endif()
+  endif()
+endif()
+
 # -----------------------------------------------------------------------------
 # Convenience: set default build type and compiler depending on the build directory
 # -----------------------------------------------------------------------------
@@ -99,13 +116,55 @@ if("${CMAKE_BINARY_DIR}" MATCHES ".*(S|s)ecure$")
   set(MI_SECURE "ON")
 endif()
 
+
+# Determine architecture
+set(MI_OPT_ARCH_FLAGS "")
+set(MI_ARCH "unknown")
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86|i[3456]86)$" OR CMAKE_GENERATOR_PLATFORM MATCHES "^(x86|Win32)$")
+  set(MI_ARCH "x86")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|x64|amd64|AMD64)$" OR CMAKE_GENERATOR_PLATFORM STREQUAL "x64" OR "x86_64" IN_LIST CMAKE_OSX_ARCHITECTURES) # must be before arm64
+  set(MI_ARCH "x64")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|armv[89].?|ARM64)$" OR CMAKE_GENERATOR_PLATFORM STREQUAL "ARM64" OR "arm64" IN_LIST CMAKE_OSX_ARCHITECTURES)
+  set(MI_ARCH "arm64")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm|armv[34567]|ARM)$")
+  set(MI_ARCH "arm32")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(riscv|riscv32|riscv64)$")
+  if(CMAKE_SIZEOF_VOID_P==4)
+    set(MI_ARCH "riscv32")
+  else()
+    set(MI_ARCH "riscv64")
+  endif()
+else()
+  set(MI_ARCH ${CMAKE_SYSTEM_PROCESSOR})
+endif()
+message(STATUS "Architecture: ${MI_ARCH}") # (${CMAKE_SYSTEM_PROCESSOR}, ${CMAKE_GENERATOR_PLATFORM}, ${CMAKE_GENERATOR})")
+
+# negative overrides (mainly to support vcpkg features)
+if(MI_NO_USE_CXX)
+  set(MI_USE_CXX "OFF")
+endif()
+if(MI_NO_OPT_ARCH)
+  set(MI_OPT_ARCH "OFF")
+elseif(MI_ARCH STREQUAL "arm64")
+  set(MI_OPT_ARCH "ON")  # enable armv8.1-a by default on arm64 unless MI_NO_OPT_ARCH is set
+endif()
+
+
 # -----------------------------------------------------------------------------
 # Process options
 # -----------------------------------------------------------------------------
 
+if(CMAKE_C_COMPILER_ID STREQUAL "Clang" AND CMAKE_CXX_COMPILER_FRONTEND_VARIANT STREQUAL "MSVC")
+  set(MI_CLANG_CL "ON")
+endif()
+
 # put -Wall early so other warnings can be disabled selectively
 if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang")
-  list(APPEND mi_cflags -Wall -Wextra -Wpedantic)
+  if (MI_CLANG_CL)
+    list(APPEND mi_cflags -W)
+  else()
+    list(APPEND mi_cflags -Wall -Wextra -Wpedantic)
+  endif()
 endif()
 if(CMAKE_C_COMPILER_ID MATCHES "GNU")
     list(APPEND mi_cflags -Wall -Wextra)
@@ -118,14 +177,6 @@ if(CMAKE_C_COMPILER_ID MATCHES "MSVC|Intel")
   set(MI_USE_CXX "ON")
 endif()
 
-if(CMAKE_BUILD_TYPE MATCHES "Release|RelWithDebInfo")
-  if (NOT MI_OPT_ARCH)
-    message(STATUS "Architecture specific optimizations are disabled (MI_OPT_ARCH=OFF)")
-  endif()
-else()
-  set(MI_OPT_ARCH OFF)
-endif()
-
 if(MI_OVERRIDE)
   message(STATUS "Override standard malloc (MI_OVERRIDE=ON)")
   if(APPLE)
@@ -332,28 +383,6 @@ if(MI_WIN_USE_FIXED_TLS)
   list(APPEND mi_defines MI_WIN_USE_FIXED_TLS=1)
 endif()
 
-# Determine architecture
-set(MI_OPT_ARCH_FLAGS "")
-set(MI_ARCH "unknown")
-if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86|i[3456]86)$" OR CMAKE_GENERATOR_PLATFORM MATCHES "^(x86|Win32)$")
-  set(MI_ARCH "x86")
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|x64|amd64|AMD64)$" OR CMAKE_GENERATOR_PLATFORM STREQUAL "x64") # must be before arm64
-  set(MI_ARCH "x64")
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|armv8.?|ARM64)$" OR CMAKE_GENERATOR_PLATFORM STREQUAL "ARM64")
-  set(MI_ARCH "arm64")
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(arm|armv[34567]|ARM)$")
-  set(MI_ARCH "arm32")
-elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(riscv|riscv32|riscv64)$")
-  if(CMAKE_SIZEOF_VOID_P==4)
-    set(MI_ARCH "riscv32")
-  else()
-    set(MI_ARCH "riscv64")
-  endif()
-else()
-  set(MI_ARCH ${CMAKE_SYSTEM_PROCESSOR})
-endif()
-message(STATUS "Architecture: ${MI_ARCH}") # (${CMAKE_SYSTEM_PROCESSOR}, ${CMAKE_GENERATOR_PLATFORM}, ${CMAKE_GENERATOR})")
-
 # Check /proc/cpuinfo for an SV39 MMU and limit the virtual address bits.
 # (this will skip the aligned hinting in that case. Issue #939, #949)
 if (EXISTS /proc/cpuinfo)
@@ -371,7 +400,7 @@ endif()
 # endif()
 
 # Compiler flags
-if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang|GNU")
+if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang|GNU" AND NOT MI_CLANG_CL)
   list(APPEND mi_cflags -Wno-unknown-pragmas -fvisibility=hidden)
   if(NOT MI_USE_CXX)
     list(APPEND mi_cflags -Wstrict-prototypes)
@@ -385,7 +414,7 @@ if(CMAKE_C_COMPILER_ID MATCHES "Intel")
   list(APPEND mi_cflags -fvisibility=hidden)
 endif()
 
-if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang|GNU|Intel" AND NOT CMAKE_SYSTEM_NAME MATCHES "Haiku")
+if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang|GNU|Intel" AND NOT CMAKE_SYSTEM_NAME MATCHES "Haiku" AND NOT MI_CLANG_CL)
   if(MI_LOCAL_DYNAMIC_TLS)
     list(APPEND mi_cflags -ftls-model=local-dynamic)
   else()
@@ -401,16 +430,23 @@ if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang|GNU|Intel" AND NOT CMAKE_SYSTEM
   if(MI_OVERRIDE)
     list(APPEND mi_cflags -fno-builtin-malloc)
   endif()
+endif()
+
+if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang|GNU|Intel" AND NOT CMAKE_SYSTEM_NAME MATCHES "Haiku")
   if(MI_OPT_ARCH)
-    if(MI_ARCH STREQUAL "arm64")
-      set(MI_OPT_ARCH_FLAGS "-march=armv8.1-a")         # fast atomics
+    if(APPLE AND CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang" AND CMAKE_OSX_ARCHITECTURES)   # to support multi-arch binaries (#999)
+      if("arm64" IN_LIST CMAKE_OSX_ARCHITECTURES)
+        list(APPEND MI_OPT_ARCH_FLAGS "-Xarch_arm64;-march=armv8.1-a")
+      endif()
+    elseif(MI_ARCH STREQUAL "arm64")
+      set(MI_OPT_ARCH_FLAGS "-march=armv8.1-a")  # fast atomics
     endif()
   endif()
 endif()
 
 if (MSVC AND MSVC_VERSION GREATER_EQUAL 1914)
   list(APPEND mi_cflags /Zc:__cplusplus)
-  if(MI_OPT_ARCH)
+  if(MI_OPT_ARCH AND NOT MI_CLANG_CL)
     if(MI_ARCH STREQUAL "arm64")
       set(MI_OPT_ARCH_FLAGS "/arch:armv8.1")           # fast atomics
     endif()
@@ -418,7 +454,7 @@ if (MSVC AND MSVC_VERSION GREATER_EQUAL 1914)
 endif()
 
 if(MINGW)
-  add_definitions(-D_WIN32_WINNT=0x601)                # issue #976
+  add_definitions(-D_WIN32_WINNT=0x600)                # issue #976
 endif()
 
 if(MI_OPT_ARCH_FLAGS)
@@ -464,13 +500,13 @@ else()
   endif()
 endif()
 
+
 # -----------------------------------------------------------------------------
 # Install and output names
 # -----------------------------------------------------------------------------
 
 # dynamic/shared library and symlinks always go to /usr/local/lib equivalent
-set(mi_install_libdir       "${CMAKE_INSTALL_LIBDIR}")
-set(mi_install_bindir       "${CMAKE_INSTALL_BINDIR}")
+# we use ${CMAKE_INSTALL_BINDIR} and ${CMAKE_INSTALL_LIBDIR}.
 
 # static libraries and object files, includes, and cmake config files
 # are either installed at top level, or use versioned directories for side-by-side installation (default)
@@ -484,19 +520,22 @@ else()
   set(mi_install_cmakedir   "${CMAKE_INSTALL_LIBDIR}/cmake/mimalloc-${mi_version}") # for cmake package info
 endif()
 
-set(mi_basename "mimalloc")
+set(mi_libname "mimalloc")
 if(MI_SECURE)
-  set(mi_basename "${mi_basename}-secure")
+  set(mi_libname "${mi_libname}-secure")
 endif()
 if(MI_TRACK_VALGRIND)
-  set(mi_basename "${mi_basename}-valgrind")
+  set(mi_libname "${mi_libname}-valgrind")
 endif()
 if(MI_TRACK_ASAN)
-  set(mi_basename "${mi_basename}-asan")
+  set(mi_libname "${mi_libname}-asan")
 endif()
 string(TOLOWER "${CMAKE_BUILD_TYPE}" CMAKE_BUILD_TYPE_LC)
-if(NOT(CMAKE_BUILD_TYPE_LC MATCHES "^(release|relwithdebinfo|minsizerel|none)$"))
-  set(mi_basename "${mi_basename}-${CMAKE_BUILD_TYPE_LC}") #append build type (e.g. -debug) if not a release version
+list(APPEND mi_defines "MI_CMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE_LC}")  #todo: multi-config project needs $<CONFIG> ?
+if(CMAKE_BUILD_TYPE_LC MATCHES "^(release|relwithdebinfo|minsizerel|none)$")
+  list(APPEND mi_defines MI_BUILD_RELEASE)
+else()
+  set(mi_libname "${mi_libname}-${CMAKE_BUILD_TYPE_LC}") #append build type (e.g. -debug) if not a release version
 endif()
 
 if(MI_BUILD_SHARED)
@@ -513,7 +552,7 @@ if(MI_BUILD_TESTS)
 endif()
 
 message(STATUS "")
-message(STATUS "Library base name: ${mi_basename}")
+message(STATUS "Library name     : ${mi_libname}")
 message(STATUS "Version          : ${mi_version}.${mi_version_patch}")
 message(STATUS "Build type       : ${CMAKE_BUILD_TYPE_LC}")
 if(MI_USE_CXX)
@@ -534,7 +573,7 @@ message(STATUS "")
 # shared library
 if(MI_BUILD_SHARED)
   add_library(mimalloc SHARED ${mi_sources})
-  set_target_properties(mimalloc PROPERTIES VERSION ${mi_version} SOVERSION ${mi_version_major} OUTPUT_NAME ${mi_basename} )
+  set_target_properties(mimalloc PROPERTIES VERSION ${mi_version} SOVERSION ${mi_version_major} OUTPUT_NAME ${mi_libname} )
   target_compile_definitions(mimalloc PRIVATE ${mi_defines} MI_SHARED_LIB MI_SHARED_LIB_EXPORT)
   target_compile_options(mimalloc PRIVATE ${mi_cflags} ${mi_cflags_dynamic})
   target_link_libraries(mimalloc PRIVATE ${mi_libraries})
@@ -542,15 +581,30 @@ if(MI_BUILD_SHARED)
       $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
       $<INSTALL_INTERFACE:${mi_install_incdir}>
   )
+  install(TARGETS mimalloc EXPORT mimalloc ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR} RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR} LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR})
+  install(EXPORT mimalloc DESTINATION ${mi_install_cmakedir})
+
+  if(WIN32 AND NOT MINGW)
+    # On windows, the import library name for the dll would clash with the static mimalloc.lib library
+    # so we postfix the dll import library with `.dll.lib` (and also the .pdb debug file)
+    set_property(TARGET mimalloc PROPERTY ARCHIVE_OUTPUT_NAME "${mi_libname}.dll" )
+    install(FILES "$<TARGET_FILE_DIR:mimalloc>/${mi_libname}.dll.lib" DESTINATION ${CMAKE_INSTALL_LIBDIR})
+    set_property(TARGET mimalloc PROPERTY PDB_NAME "${mi_libname}.dll")
+    # don't try to install the pdb since it may not be generated depending on the configuration
+    # install(FILES "$<TARGET_FILE_DIR:mimalloc>/${mi_libname}.dll.pdb" DESTINATION ${CMAKE_INSTALL_LIBDIR})
+  endif()
   if(WIN32 AND MI_WIN_REDIRECT)
+    if(MINGW)
+      set_property(TARGET mimalloc PROPERTY PREFIX "")
+    endif()
     # On windows, link and copy the mimalloc redirection dll too.
     if(CMAKE_GENERATOR_PLATFORM STREQUAL "arm64ec")
       set(MIMALLOC_REDIRECT_SUFFIX "-arm64ec")
     elseif(MI_ARCH STREQUAL "x64")
       set(MIMALLOC_REDIRECT_SUFFIX "")
       if(CMAKE_SYSTEM_PROCESSOR STREQUAL "ARM64")
-        message(STATUS "Note: x64 code emulated on Windows for arm64 should use an arm64ec build of 'mimalloc-override.dll'")
-        message(STATUS "      with 'mimalloc-redirect-arm64ec.dll'. See the 'bin\\readme.md' for more information.")
+        message(STATUS "Note: x64 code emulated on Windows for arm64 should use an arm64ec build of 'mimalloc.dll'")
+        message(STATUS "      together with 'mimalloc-redirect-arm64ec.dll'. See the 'bin\\readme.md' for more information.")
       endif()
     elseif(MI_ARCH STREQUAL "x86")
       set(MIMALLOC_REDIRECT_SUFFIX "32")
@@ -558,20 +612,19 @@ if(MI_BUILD_SHARED)
       set(MIMALLOC_REDIRECT_SUFFIX "-${MI_ARCH}")  # -arm64 etc.
     endif()
 
-    target_link_libraries(mimalloc PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/bin/mimalloc-redirect${MIMALLOC_REDIRECT_SUFFIX}.lib)
+    target_link_libraries(mimalloc PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/bin/mimalloc-redirect${MIMALLOC_REDIRECT_SUFFIX}.lib)  # the DLL import library
     add_custom_command(TARGET mimalloc POST_BUILD
       COMMAND "${CMAKE_COMMAND}" -E copy "${CMAKE_CURRENT_SOURCE_DIR}/bin/mimalloc-redirect${MIMALLOC_REDIRECT_SUFFIX}.dll" $<TARGET_FILE_DIR:mimalloc>
       COMMENT "Copy mimalloc-redirect${MIMALLOC_REDIRECT_SUFFIX}.dll to output directory")
-    install(FILES "$<TARGET_FILE_DIR:mimalloc>/mimalloc-redirect${MIMALLOC_REDIRECT_SUFFIX}.dll" DESTINATION ${mi_install_bindir})
+    install(FILES "$<TARGET_FILE_DIR:mimalloc>/mimalloc-redirect${MIMALLOC_REDIRECT_SUFFIX}.dll" DESTINATION ${CMAKE_INSTALL_BINDIR})
   endif()
-
-  install(TARGETS mimalloc EXPORT mimalloc ARCHIVE DESTINATION ${mi_install_libdir} RUNTIME DESTINATION ${mi_install_bindir} LIBRARY DESTINATION ${mi_install_libdir})
-  install(EXPORT mimalloc DESTINATION ${mi_install_cmakedir})
 endif()
 
+
 # static library
 if (MI_BUILD_STATIC)
   add_library(mimalloc-static STATIC ${mi_sources})
+  set_property(TARGET mimalloc-static PROPERTY OUTPUT_NAME ${mi_libname})
   set_property(TARGET mimalloc-static PROPERTY POSITION_INDEPENDENT_CODE ON)
   target_compile_definitions(mimalloc-static PRIVATE ${mi_defines} MI_STATIC_LIB)
   target_compile_options(mimalloc-static PRIVATE ${mi_cflags} ${mi_cflags_static})
@@ -580,15 +633,6 @@ if (MI_BUILD_STATIC)
       $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
       $<INSTALL_INTERFACE:${mi_install_incdir}>
   )
-  if(WIN32)
-    # When building both static and shared libraries on Windows, a static library should use a
-    # different output name to avoid the conflict with the import library of a shared one.
-    string(REPLACE "mimalloc" "mimalloc-static" mi_output_name ${mi_basename})
-    set_target_properties(mimalloc-static PROPERTIES OUTPUT_NAME ${mi_output_name})
-  else()
-    set_target_properties(mimalloc-static PROPERTIES OUTPUT_NAME ${mi_basename})
-  endif()
-
   install(TARGETS mimalloc-static EXPORT mimalloc DESTINATION ${mi_install_objdir} LIBRARY)
   install(EXPORT mimalloc DESTINATION ${mi_install_cmakedir})
 endif()
@@ -597,6 +641,7 @@ endif()
 install(FILES include/mimalloc.h DESTINATION ${mi_install_incdir})
 install(FILES include/mimalloc-override.h DESTINATION ${mi_install_incdir})
 install(FILES include/mimalloc-new-delete.h DESTINATION ${mi_install_incdir})
+install(FILES include/mimalloc-stats.h DESTINATION ${mi_install_incdir})
 install(FILES cmake/mimalloc-config.cmake DESTINATION ${mi_install_cmakedir})
 install(FILES cmake/mimalloc-config-version.cmake DESTINATION ${mi_install_cmakedir})
 
@@ -613,12 +658,15 @@ if (MI_BUILD_OBJECT)
   )
 
   # Copy the generated object file (`static.o`) to the output directory (as `mimalloc.o`)
-  if(NOT WIN32)
+  if(CMAKE_GENERATOR MATCHES "^Visual Studio.*$")
+    set(mimalloc-obj-static "${CMAKE_CURRENT_BINARY_DIR}/mimalloc-obj.dir/$<CONFIG>/static${CMAKE_C_OUTPUT_EXTENSION}")
+  else()
     set(mimalloc-obj-static "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/mimalloc-obj.dir/src/static.c${CMAKE_C_OUTPUT_EXTENSION}")
-    set(mimalloc-obj-out    "${CMAKE_CURRENT_BINARY_DIR}/${mi_basename}${CMAKE_C_OUTPUT_EXTENSION}")
-    add_custom_command(OUTPUT ${mimalloc-obj-out} DEPENDS mimalloc-obj COMMAND "${CMAKE_COMMAND}" -E copy "${mimalloc-obj-static}" "${mimalloc-obj-out}")
-    add_custom_target(mimalloc-obj-target ALL DEPENDS ${mimalloc-obj-out})
   endif()
+  set(mimalloc-obj-out "${CMAKE_CURRENT_BINARY_DIR}/${mi_libname}${CMAKE_C_OUTPUT_EXTENSION}")
+  add_custom_command(OUTPUT ${mimalloc-obj-out} DEPENDS mimalloc-obj COMMAND "${CMAKE_COMMAND}" -E copy "${mimalloc-obj-static}" "${mimalloc-obj-out}")
+  add_custom_target(mimalloc-obj-target ALL DEPENDS ${mimalloc-obj-out})
+
 
   # the following seems to lead to cmake warnings/errors on some systems, disable for now :-(
   # install(TARGETS mimalloc-obj EXPORT mimalloc DESTINATION ${mi_install_objdir})
@@ -627,22 +675,23 @@ if (MI_BUILD_OBJECT)
   # but that fails cmake versions less than 3.10 so we leave it as is for now
   install(FILES ${mimalloc-obj-static}
           DESTINATION ${mi_install_objdir}
-          RENAME ${mi_basename}${CMAKE_C_OUTPUT_EXTENSION} )
+          RENAME ${mi_libname}${CMAKE_C_OUTPUT_EXTENSION} )
 endif()
 
+
 # pkg-config file support
-set(pc_libraries "")
+set(mi_pc_libraries "")
 foreach(item IN LISTS mi_libraries)
   if(item MATCHES " *[-].*")
-    set(pc_libraries "${pc_libraries} ${item}")
+    set(mi_pc_libraries "${mi_pc_libraries} ${item}")
   else()
-    set(pc_libraries "${pc_libraries} -l${item}")
+    set(mi_pc_libraries "${mi_pc_libraries} -l${item}")
   endif()
 endforeach()
 
 include("cmake/JoinPaths.cmake")
-join_paths(includedir_for_pc_file "\${prefix}" "${CMAKE_INSTALL_INCLUDEDIR}")
-join_paths(libdir_for_pc_file "\${prefix}" "${CMAKE_INSTALL_LIBDIR}")
+join_paths(mi_pc_includedir "\${prefix}" "${CMAKE_INSTALL_INCLUDEDIR}")
+join_paths(mi_pc_libdir "\${prefix}" "${CMAKE_INSTALL_LIBDIR}")
 
 configure_file(mimalloc.pc.in mimalloc.pc @ONLY)
 install(FILES "${CMAKE_CURRENT_BINARY_DIR}/mimalloc.pc"
@@ -657,15 +706,41 @@ install(FILES "${CMAKE_CURRENT_BINARY_DIR}/mimalloc.pc"
 if (MI_BUILD_TESTS)
   enable_testing()
 
+  # static link tests
   foreach(TEST_NAME api api-fill stress)
     add_executable(mimalloc-test-${TEST_NAME} test/test-${TEST_NAME}.c)
     target_compile_definitions(mimalloc-test-${TEST_NAME} PRIVATE ${mi_defines})
     target_compile_options(mimalloc-test-${TEST_NAME} PRIVATE ${mi_cflags})
     target_include_directories(mimalloc-test-${TEST_NAME} PRIVATE include)
-    target_link_libraries(mimalloc-test-${TEST_NAME} PRIVATE mimalloc ${mi_libraries})
-
+    if(MI_BUILD_SHARED AND (MI_TRACK_ASAN OR MI_DEBUG_TSAN OR MI_DEBUG_UBSAN))
+      target_link_libraries(mimalloc-test-${TEST_NAME} PRIVATE mimalloc ${mi_libraries})
+    else()
+      target_link_libraries(mimalloc-test-${TEST_NAME} PRIVATE mimalloc-static ${mi_libraries})
+    endif()
     add_test(NAME test-${TEST_NAME} COMMAND mimalloc-test-${TEST_NAME})
   endforeach()
+
+  # dynamic override test
+  if(MI_BUILD_SHARED AND NOT (MI_TRACK_ASAN OR MI_DEBUG_TSAN OR MI_DEBUG_UBSAN))
+    add_executable(mimalloc-test-stress-dynamic test/test-stress.c)
+    target_compile_definitions(mimalloc-test-stress-dynamic PRIVATE ${mi_defines} "USE_STD_MALLOC=1")
+    if(WIN32)
+      target_compile_definitions(mimalloc-test-stress-dynamic PRIVATE "MI_LINK_VERSION=1")
+    endif()
+    target_compile_options(mimalloc-test-stress-dynamic PRIVATE ${mi_cflags})
+    target_include_directories(mimalloc-test-stress-dynamic PRIVATE include)
+    target_link_libraries(mimalloc-test-stress-dynamic PRIVATE mimalloc ${mi_libraries})  # mi_version
+    if(WIN32)
+      add_test(NAME test-stress-dynamic COMMAND ${CMAKE_COMMAND} -E env MIMALLOC_SHOW_STATS=1 $<TARGET_FILE:mimalloc-test-stress-dynamic>)
+    else()
+      if(APPLE)
+        set(LD_PRELOAD "DYLD_INSERT_LIBRARIES")
+      else()
+        set(LD_PRELOAD "LD_PRELOAD")
+      endif()
+      add_test(NAME test-stress-dynamic COMMAND ${CMAKE_COMMAND} -E env MIMALLOC_SHOW_STATS=1 ${LD_PRELOAD}=$<TARGET_FILE:mimalloc> $<TARGET_FILE:mimalloc-test-stress-dynamic>)
+    endif()
+  endif()
 endif()
 
 # -----------------------------------------------------------------------------

azure-pipelines.yml

@@ -7,9 +7,9 @@ trigger:
   branches:
     include:
     - master
-    - dev
-    - dev2
     - dev3
+    - dev2
+    - dev
   tags:
     include:
     - v*
@@ -34,6 +34,14 @@ jobs:
         BuildType: secure
         cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Release -DMI_SECURE=ON
         MSBuildConfiguration: Release
+      Debug x86:
+        BuildType: debug
+        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Debug -DMI_DEBUG_FULL=ON -A Win32
+        MSBuildConfiguration: Debug
+      Release x86:
+        BuildType: release
+        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Release -A Win32
+        MSBuildConfiguration: Release
   steps:
   - task: CMake@1
     inputs:
@@ -161,6 +169,7 @@ jobs:
   - script: ctest --verbose --timeout 240
     workingDirectory: $(BuildType)
     displayName: CTest
+    
 #  - upload: $(Build.SourcesDirectory)/$(BuildType)
 #    artifact: mimalloc-macos-$(BuildType)
 
@@ -240,49 +249,6 @@ jobs:
     workingDirectory: $(BuildType)
     displayName: CTest
 
-- job:
-  displayName: Ubuntu 20.04
-  pool:
-    vmImage:
-      ubuntu-20.04
-  strategy:
-    matrix:
-      Debug:
-        CC: gcc
-        CXX: g++
-        BuildType: debug
-        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Debug -DMI_DEBUG_FULL=ON
-      Debug++:
-        CC: gcc
-        CXX: g++
-        BuildType: debug-cxx
-        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Debug -DMI_DEBUG_FULL=ON -DMI_USE_CXX=ON
-      Debug Clang:
-        CC: clang
-        CXX: clang++
-        BuildType: debug-clang
-        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Debug -DMI_DEBUG_FULL=ON
-      Debug++ Clang:
-        CC: clang
-        CXX: clang++
-        BuildType: debug-clang-cxx
-        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Debug -DMI_DEBUG_FULL=ON -DMI_USE_CXX=ON
-      Release Clang:
-        CC: clang
-        CXX: clang++
-        BuildType: release-clang
-        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Release
-  steps:
-  - task: CMake@1
-    inputs:
-      workingDirectory: $(BuildType)
-      cmakeArgs: .. $(cmakeExtraArgs)
-  - script: make -j$(nproc) -C $(BuildType)
-    displayName: Make
-  - script: ctest --verbose --timeout 240
-    workingDirectory: $(BuildType)
-    displayName: CTest
-
 - job:
   displayName: macOS 15 (Sequoia)
   pool:

bin/readme.md

@@ -1,15 +1,14 @@
 # Windows Override
 
-<span id="override_on_windows">Dynamically overriding on mimalloc on Windows</span> 
-is robust and has the particular advantage to be able to redirect all malloc/free calls 
-that go through the (dynamic) C runtime allocator, including those from other DLL's or 
-libraries. As it intercepts all allocation calls on a low level, it can be used reliably 
-on large programs that include other 3rd party components.
+<span id="override_on_windows">We use a separate redirection DLL to override mimalloc on Windows</span> 
+such that we redirect all malloc/free calls that go through the (dynamic) C runtime allocator, 
+including those from other DLL's or libraries. As it intercepts all allocation calls on a low level, 
+it can be used on large programs that include other 3rd party components.
 There are four requirements to make the overriding work well:
 
 1. Use the C-runtime library as a DLL (using the `/MD` or `/MDd` switch).
 
-2. Link your program explicitly with the `mimalloc.lib` export library for
+2. Link your program explicitly with the `mimalloc.dll.lib` export library for
    the `mimalloc.dll` -- which contains all mimalloc functionality.
    To ensure the `mimalloc.dll` is actually loaded at run-time it is easiest 
    to insert some call to the mimalloc API in the `main` function, like `mi_version()`
@@ -26,6 +25,23 @@ There are four requirements to make the overriding work well:
    list of the final executable (so it can intercept all potential allocations).
    You can use `minject -l <exe>` to check this if needed.
 
+```csharp
+┌──────────────┐                                                    
+│ Your Program │                                                    
+└────┬─────────┘                                                    
+     │                                                              
+     │ mi_version()  ┌───────────────┐     ┌───────────────────────┐
+     ├──────────────►│ mimalloc.dll  ├────►│ mimalloc-redirect.dll │
+     │               └──────┬────────┘     └───────────────────────┘
+     │                      ▼                                       
+     │ malloc() etc. ┌──────────────┐                               
+     ├──────────────►│ ucrtbase.dll │                               
+     │               └──────────────┘                               
+     │                                                              
+     │                                                              
+     └──────────────► ...                                           
+```
+
 For best performance on Windows with C++, it
 is also recommended to also override the `new`/`delete` operations (by including
 [`mimalloc-new-delete.h`](../include/mimalloc-new-delete.h) 

cmake/mimalloc-config-version.cmake

@@ -1,6 +1,6 @@
-set(mi_version_major 1)
-set(mi_version_minor 8)
-set(mi_version_patch 9)
+set(mi_version_major 2)
+set(mi_version_minor 2)
+set(mi_version_patch 3)
 set(mi_version ${mi_version_major}.${mi_version_minor})
 
 set(PACKAGE_VERSION ${mi_version})

contrib/vcpkg/portfile.cmake

@@ -0,0 +1,63 @@
+vcpkg_from_github(
+  OUT_SOURCE_PATH SOURCE_PATH
+  REPO microsoft/mimalloc
+  HEAD_REF master
+
+  # The "REF" can be a commit hash, branch name (dev2), or a version (v2.2.1).
+  # REF "v${VERSION}"
+  REF e2db21e9ba9fb9172b7b0aa0fe9b8742525e8774
+
+  # The sha512 is the hash of the tar.gz bundle.
+  # (To get the sha512, run `vcpkg install mimalloc[override] --overlay-ports=<dir of this file>` and copy the sha from the error message.)
+  SHA512 8cbb601fdf8b46dd6a9c0d314d6da9d4960699853829e96d2470753867f90689fb4caeaf30d628943fd388670dc11902dbecc9cc7c329b99a510524a09bdb612
+)
+
+vcpkg_check_features(OUT_FEATURE_OPTIONS FEATURE_OPTIONS
+  FEATURES
+    c           MI_NO_USE_CXX
+    guarded     MI_GUARDED
+    secure      MI_SECURE
+    override    MI_OVERRIDE
+    optarch     MI_OPT_ARCH
+    optsimd     MI_OPT_SIMD
+    xmalloc     MI_XMALLOC
+    asm         MI_SEE_ASM
+)
+string(COMPARE EQUAL "${VCPKG_LIBRARY_LINKAGE}" "static" MI_BUILD_STATIC)
+string(COMPARE EQUAL "${VCPKG_LIBRARY_LINKAGE}" "dynamic" MI_BUILD_SHARED)
+
+vcpkg_cmake_configure(
+  SOURCE_PATH "${SOURCE_PATH}"
+  OPTIONS
+    -DMI_USE_CXX=ON
+    -DMI_BUILD_TESTS=OFF
+    -DMI_BUILD_OBJECT=ON
+    -DMI_BUILD_STATIC=${MI_BUILD_STATIC}
+    -DMI_BUILD_SHARED=${MI_BUILD_SHARED}
+    -DMI_INSTALL_TOPLEVEL=ON
+    ${FEATURE_OPTIONS}
+)
+
+vcpkg_cmake_install()
+vcpkg_copy_pdbs()
+
+file(COPY
+  "${CMAKE_CURRENT_LIST_DIR}/vcpkg-cmake-wrapper.cmake"
+  "${CMAKE_CURRENT_LIST_DIR}/usage"
+  DESTINATION "${CURRENT_PACKAGES_DIR}/share/${PORT}"
+)
+vcpkg_cmake_config_fixup(CONFIG_PATH lib/cmake/mimalloc)
+
+if(VCPKG_LIBRARY_LINKAGE STREQUAL "dynamic")
+  # todo: why is this needed?
+  vcpkg_replace_string(
+    "${CURRENT_PACKAGES_DIR}/include/mimalloc.h"
+    "!defined(MI_SHARED_LIB)"
+    "0 // !defined(MI_SHARED_LIB)"
+  )
+endif()
+file(REMOVE_RECURSE "${CURRENT_PACKAGES_DIR}/debug/include")
+file(REMOVE_RECURSE "${CURRENT_PACKAGES_DIR}/debug/share")
+
+vcpkg_fixup_pkgconfig()
+vcpkg_install_copyright(FILE_LIST "${SOURCE_PATH}/LICENSE")

contrib/vcpkg/readme.md

@@ -0,0 +1,40 @@
+# Vcpkg support
+
+This directory is meant to provide the sources for the official [vcpkg port] 
+of mimalloc, but can also be used to override the official port with
+your own variant.
+
+For example, you can edit the [`portfile.cmake`](portfile.cmake) 
+to check out a specific commit, version, or branch of mimalloc, or set further options. 
+You can install such custom port as:
+
+```sh
+$ vcpkg install "mimalloc[override]" --recurse --overlay-ports=./contrib/vcpkg
+```
+
+This will also show the correct sha512 hash if you use a custom version.
+Another way is to refer to the overlay from the [vcpkg-configuration.json](https://learn.microsoft.com/en-us/vcpkg/reference/vcpkg-configuration-json) file.
+See also the vcpkg [documentation](https://learn.microsoft.com/en-us/vcpkg/produce/update-package-version) for more information.
+
+
+# Using mimalloc from vcpkg
+
+When using [cmake with vcpkg](https://learn.microsoft.com/en-us/vcpkg/get_started/get-started?pivots=shell-powershell), 
+you can use mimalloc from the `CMakeLists.txt` as:
+
+```cmake
+find_package(mimalloc CONFIG REQUIRED)
+target_link_libraries(main PRIVATE mimalloc)
+```
+
+See [`test/CMakeLists.txt](../../test/CMakeLists.txt) for more examples.
+
+
+# Acknowledgements
+
+The original port for vckpg was contributed by many people, including: @vicroms, @myd7349, @PhoubeHui, @LilyWangL,
+@JonLiu1993, @RT2Code, Remy Tassoux, @wangao, @BillyONeal, @jiayuehua, @dg0yt, @gerar-ryan-immersaview, @nickdademo, 
+and @jimwang118 -- Thank you so much!
+
+
+[vcpkg port]: https://github.com/microsoft/vcpkg/tree/master/ports/mimalloc

contrib/vcpkg/usage

@@ -0,0 +1,20 @@
+Use the following CMake targets to import mimalloc:
+
+  find_package(mimalloc CONFIG REQUIRED)
+  target_link_libraries(main PRIVATE mimalloc)
+
+And use mimalloc in your sources as:
+
+  #include <mimalloc.h>
+  #include <stdio.h>
+  int main(int argc, char** argv) {
+    int* p = mi_malloc_tp(int);
+    *p = mi_version();
+    printf("mimalloc version: %d\n", *p);
+    mi_free(p);
+    return 0;
+  }
+
+When dynamically overriding on Windows, ensure `mimalloc.dll` is linked through some call to
+mimalloc (e.g. `mi_version()`), and that the `mimalloc-redirect.dll` is in the same directory.
+See https://github.com/microsoft/mimalloc/blob/dev/bin/readme.md for detailed information.

contrib/vcpkg/vcpkg-cmake-wrapper.cmake

@@ -0,0 +1,20 @@
+_find_package(${ARGS})
+
+if(CMAKE_CURRENT_LIST_DIR STREQUAL "${MIMALLOC_CMAKE_DIR}/${MIMALLOC_VERSION_DIR}")
+    set(MIMALLOC_INCLUDE_DIR "${VCPKG_INSTALLED_DIR}/${VCPKG_TARGET_TRIPLET}/include")
+    # As in vcpkg.cmake
+    if(NOT DEFINED CMAKE_BUILD_TYPE OR CMAKE_BUILD_TYPE MATCHES "^[Dd][Ee][Bb][Uu][Gg]$")
+        set(MIMALLOC_LIBRARY_DIR "${VCPKG_INSTALLED_DIR}/${VCPKG_TARGET_TRIPLET}/debug/lib")
+    else()
+        set(MIMALLOC_LIBRARY_DIR "${VCPKG_INSTALLED_DIR}/${VCPKG_TARGET_TRIPLET}/lib")
+    endif()
+    set(MIMALLOC_OBJECT_DIR "${MIMALLOC_LIBRARY_DIR}")
+    set(MIMALLOC_TARGET_DIR "${MIMALLOC_LIBRARY_DIR}")
+endif()
+
+# vcpkg always configures either a static or dynamic library.
+# ensure to always expose the mimalloc target as either the static or dynamic build.
+if(TARGET mimalloc-static AND NOT TARGET mimalloc)
+  add_library(mimalloc INTERFACE IMPORTED)
+  set_target_properties(mimalloc PROPERTIES INTERFACE_LINK_LIBRARIES mimalloc-static)
+endif()

contrib/vcpkg/vcpkg.json

@@ -0,0 +1,45 @@
+{
+  "name": "mimalloc",
+  "version": "2.2.2",
+  "port-version": 1,
+  "description": "Compact general purpose allocator with excellent performance",
+  "homepage": "https://github.com/microsoft/mimalloc",
+  "license": "MIT",
+  "supports": "!uwp",
+  "dependencies": [
+    {
+      "name": "vcpkg-cmake",
+      "host": true
+    },
+    {
+      "name": "vcpkg-cmake-config",
+      "host": true
+    }
+  ],
+  "features": {
+    "c": {
+      "description": "Use C11 compilation (this can still override new/delete)"
+    },
+    "override": {
+      "description": "Override the standard malloc/free interface"
+    },
+    "secure": {
+      "description": "Use full security mitigations (like guard pages and randomization)"
+    },
+    "guarded": {
+      "description": "Use build that support guard pages after objects controlled with MIMALLOC_GUARDED_SAMPLE_RATE"
+    },
+    "xmalloc": {
+      "description": "If out-of-memory, call abort() instead of returning NULL"
+    },
+    "optarch": {
+      "description": "Use architecture specific optimizations (on x64: '-march=haswell;-mavx2', on arm64: '-march=armv8.1-a')"
+    },
+    "optsimd": {
+      "description": "Allow use of SIMD instructions (avx2 or neon) (requires 'optarch' to be enabled)"
+    },
+    "asm": {
+      "description": "Generate assembly files"
+    }
+  }
+}

ide/vs2022/mimalloc-lib.vcxproj

@@ -120,49 +120,52 @@
     <OutDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
     <TargetExt>.lib</TargetExt>
-    <TargetName>mimalloc-static</TargetName>
+    <TargetName>mimalloc</TargetName>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
     <OutDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
     <TargetExt>.lib</TargetExt>
-    <TargetName>mimalloc-static</TargetName>
+    <TargetName>mimalloc</TargetName>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
     <OutDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
     <TargetExt>.lib</TargetExt>
-    <TargetName>mimalloc-static</TargetName>
+    <TargetName>mimalloc</TargetName>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'">
     <OutDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
     <TargetExt>.lib</TargetExt>
-    <TargetName>mimalloc-static</TargetName>
+    <TargetName>mimalloc</TargetName>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'">
     <OutDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
     <TargetExt>.lib</TargetExt>
-    <TargetName>mimalloc-static</TargetName>
+    <TargetName>mimalloc</TargetName>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
     <OutDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
     <TargetExt>.lib</TargetExt>
-    <TargetName>mimalloc-static</TargetName>
+    <TargetName>mimalloc</TargetName>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'">
     <OutDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
     <TargetExt>.lib</TargetExt>
-    <TargetName>mimalloc-static</TargetName>
+    <TargetName>mimalloc</TargetName>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'">
     <OutDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(SolutionDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
     <TargetExt>.lib</TargetExt>
-    <TargetName>mimalloc-static</TargetName>
+    <TargetName>mimalloc</TargetName>
+  </PropertyGroup>
+  <PropertyGroup Label="Vcpkg">
+    <VcpkgEnabled>false</VcpkgEnabled>
   </PropertyGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
     <ClCompile>
@@ -421,15 +424,15 @@
     </ClCompile>
     <ClCompile Include="..\..\src\alloc-posix.c" />
     <ClCompile Include="..\..\src\alloc.c" />
-    <ClCompile Include="..\..\src\arena-abandoned.c">
+    <ClCompile Include="..\..\src\arena-abandon.c">
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'">true</ExcludedFromBuild>
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">true</ExcludedFromBuild>
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">true</ExcludedFromBuild>
-      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'">true</ExcludedFromBuild>
-      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'">true</ExcludedFromBuild>
-      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'">true</ExcludedFromBuild>
-      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
     </ClCompile>
     <ClCompile Include="..\..\src\arena.c" />
     <ClCompile Include="..\..\src\bitmap.c">
@@ -483,6 +486,7 @@
     <ClInclude Include="$(ProjectDir)..\..\include\mimalloc.h" />
     <ClInclude Include="$(ProjectDir)..\..\include\mimalloc-override.h" />
     <ClInclude Include="..\..\include\mimalloc-new-delete.h" />
+    <ClInclude Include="..\..\include\mimalloc-stats.h" />
     <ClInclude Include="..\..\include\mimalloc\atomic.h" />
     <ClInclude Include="..\..\include\mimalloc\internal.h" />
     <ClInclude Include="..\..\include\mimalloc\prim.h" />

ide/vs2022/mimalloc-lib.vcxproj.filters

@@ -16,9 +16,6 @@
     <ClCompile Include="..\..\src\arena.c">
       <Filter>Sources</Filter>
     </ClCompile>
-    <ClCompile Include="..\..\src\arena-abandoned.c">
-      <Filter>Sources</Filter>
-    </ClCompile>
     <ClCompile Include="..\..\src\bitmap.c">
       <Filter>Sources</Filter>
     </ClCompile>
@@ -64,6 +61,9 @@
     <ClCompile Include="..\..\src\stats.c">
       <Filter>Sources</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\src\arena-abandon.c">
+      <Filter>Sources</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\..\include\mimalloc\atomic.h">
@@ -93,6 +93,9 @@
     <ClInclude Include="..\..\include\mimalloc\prim.h">
       <Filter>Headers</Filter>
     </ClInclude>
+    <ClInclude Include="..\..\include\mimalloc-stats.h">
+      <Filter>Headers</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <Filter Include="Headers">

ide/vs2022/mimalloc-override-dll.vcxproj

@@ -160,6 +160,9 @@
     <TargetExt>.dll</TargetExt>
     <TargetName>mimalloc</TargetName>
   </PropertyGroup>
+  <PropertyGroup Label="Vcpkg">
+    <VcpkgEnabled>false</VcpkgEnabled>
+  </PropertyGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
     <ClCompile>
       <WarningLevel>Level3</WarningLevel>
@@ -180,6 +183,8 @@
       </ModuleDefinitionFile>
       <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
       <IgnoreAllDefaultLibraries>false</IgnoreAllDefaultLibraries>
+      <ImportLibrary>$(OutDir)$(TargetName).dll.lib</ImportLibrary>
+      <ProgramDatabaseFile>$(OutDir)$(TargetName).dll.pdb</ProgramDatabaseFile>
     </Link>
     <PostBuildEvent>
       <Command>COPY /Y "$(ProjectDir)..\..\bin\mimalloc-redirect32.dll" "$(OutputPath)"</Command>
@@ -208,6 +213,8 @@
       </ModuleDefinitionFile>
       <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
       <IgnoreAllDefaultLibraries>false</IgnoreAllDefaultLibraries>
+      <ImportLibrary>$(OutDir)$(TargetName).dll.lib</ImportLibrary>
+      <ProgramDatabaseFile>$(OutDir)$(TargetName).dll.pdb</ProgramDatabaseFile>
     </Link>
     <PostBuildEvent>
       <Command>COPY /Y "$(ProjectDir)..\..\bin\mimalloc-redirect.dll" "$(OutputPath)"</Command>
@@ -236,6 +243,8 @@
       </ModuleDefinitionFile>
       <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
       <IgnoreAllDefaultLibraries>false</IgnoreAllDefaultLibraries>
+      <ImportLibrary>$(OutDir)$(TargetName).dll.lib</ImportLibrary>
+      <ProgramDatabaseFile>$(OutDir)$(TargetName).dll.pdb</ProgramDatabaseFile>
     </Link>
     <PostBuildEvent>
       <Command>COPY /Y "$(ProjectDir)..\..\bin\mimalloc-redirect-arm64.dll" "$(OutputPath)"</Command>
@@ -264,6 +273,8 @@
       </ModuleDefinitionFile>
       <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
       <IgnoreAllDefaultLibraries>false</IgnoreAllDefaultLibraries>
+      <ImportLibrary>$(OutDir)$(TargetName).dll.lib</ImportLibrary>
+      <ProgramDatabaseFile>$(OutDir)$(TargetName).dll.pdb</ProgramDatabaseFile>
     </Link>
     <PostBuildEvent>
       <Command>COPY /Y "$(ProjectDir)..\..\bin\mimalloc-redirect-arm64ec.dll" "$(OutputPath)"</Command>
@@ -296,6 +307,8 @@
       </ModuleDefinitionFile>
       <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
       <IgnoreAllDefaultLibraries>false</IgnoreAllDefaultLibraries>
+      <ImportLibrary>$(OutDir)$(TargetName).dll.lib</ImportLibrary>
+      <ProgramDatabaseFile>$(OutDir)$(TargetName).dll.pdb</ProgramDatabaseFile>
     </Link>
     <PostBuildEvent>
       <Command>COPY /Y "$(ProjectDir)..\..\bin\mimalloc-redirect32.dll" "$(OutputPath)"</Command>
@@ -328,6 +341,8 @@
       </ModuleDefinitionFile>
       <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
       <IgnoreAllDefaultLibraries>false</IgnoreAllDefaultLibraries>
+      <ImportLibrary>$(OutDir)$(TargetName).dll.lib</ImportLibrary>
+      <ProgramDatabaseFile>$(OutDir)$(TargetName).dll.pdb</ProgramDatabaseFile>
     </Link>
     <PostBuildEvent>
       <Command>COPY /Y "$(ProjectDir)..\..\bin\mimalloc-redirect.dll" "$(OutputPath)"</Command>
@@ -361,6 +376,8 @@
       </ModuleDefinitionFile>
       <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
       <IgnoreAllDefaultLibraries>false</IgnoreAllDefaultLibraries>
+      <ImportLibrary>$(OutDir)$(TargetName).dll.lib</ImportLibrary>
+      <ProgramDatabaseFile>$(OutDir)$(TargetName).dll.pdb</ProgramDatabaseFile>
     </Link>
     <PostBuildEvent>
       <Command>COPY /Y "$(ProjectDir)..\..\bin\mimalloc-redirect-arm64.dll" "$(OutputPath)"</Command>
@@ -394,6 +411,8 @@
       </ModuleDefinitionFile>
       <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
       <IgnoreAllDefaultLibraries>false</IgnoreAllDefaultLibraries>
+      <ImportLibrary>$(OutDir)$(TargetName).dll.lib</ImportLibrary>
+      <ProgramDatabaseFile>$(OutDir)$(TargetName).dll.pdb</ProgramDatabaseFile>
     </Link>
     <PostBuildEvent>
       <Command>COPY /Y "$(ProjectDir)..\..\bin\mimalloc-redirect-arm64ec.dll" "$(OutputPath)"</Command>
@@ -408,6 +427,7 @@
     <ClInclude Include="..\..\include\mimalloc-etw.h" />
     <ClInclude Include="..\..\include\mimalloc-new-delete.h" />
     <ClInclude Include="..\..\include\mimalloc-override.h" />
+    <ClInclude Include="..\..\include\mimalloc-stats.h" />
     <ClInclude Include="..\..\include\mimalloc\atomic.h" />
     <ClInclude Include="..\..\include\mimalloc\internal.h" />
     <ClInclude Include="..\..\include\mimalloc\prim.h" />
@@ -438,15 +458,15 @@
     </ClCompile>
     <ClCompile Include="..\..\src\alloc-posix.c" />
     <ClCompile Include="..\..\src\alloc.c" />
-    <ClCompile Include="..\..\src\arena-abandoned.c">
+    <ClCompile Include="..\..\src\arena-abandon.c">
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'">true</ExcludedFromBuild>
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">true</ExcludedFromBuild>
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">true</ExcludedFromBuild>
-      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'">true</ExcludedFromBuild>
-      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'">true</ExcludedFromBuild>
-      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
       <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'">true</ExcludedFromBuild>
-      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
     </ClCompile>
     <ClCompile Include="..\..\src\arena.c" />
     <ClCompile Include="..\..\src\bitmap.c" />
@@ -482,9 +502,6 @@
     <ClCompile Include="..\..\src\segment.c" />
     <ClCompile Include="..\..\src\stats.c" />
   </ItemGroup>
-  <ItemGroup>
-    <None Include="..\..\include\mimalloc-etw-gen.man" />
-  </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets">
   </ImportGroup>

ide/vs2022/mimalloc-override-dll.vcxproj.filters

@@ -16,9 +16,6 @@
     <ClCompile Include="..\..\src\arena.c">
       <Filter>Sources</Filter>
     </ClCompile>
-    <ClCompile Include="..\..\src\arena-abandoned.c">
-      <Filter>Sources</Filter>
-    </ClCompile>
     <ClCompile Include="..\..\src\bitmap.c">
       <Filter>Sources</Filter>
     </ClCompile>
@@ -61,6 +58,9 @@
     <ClCompile Include="..\..\src\stats.c">
       <Filter>Sources</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\src\arena-abandon.c">
+      <Filter>Sources</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\..\include\mimalloc\atomic.h">
@@ -96,6 +96,9 @@
     <ClInclude Include="..\..\include\mimalloc\prim.h">
       <Filter>Headers</Filter>
     </ClInclude>
+    <ClInclude Include="..\..\include\mimalloc-stats.h">
+      <Filter>Headers</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <Filter Include="Headers">
@@ -105,9 +108,4 @@
       <UniqueIdentifier>{94b40bdc-a741-45dd-81aa-c05fabcd2970}</UniqueIdentifier>
     </Filter>
   </ItemGroup>
-  <ItemGroup>
-    <None Include="..\..\include\mimalloc-etw-gen.man">
-      <Filter>Sources</Filter>
-    </None>
-  </ItemGroup>
 </Project>

ide/vs2022/mimalloc-override-test-dep.vcxproj

@@ -0,0 +1,355 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="15.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ItemGroup Label="ProjectConfigurations">
+    <ProjectConfiguration Include="Debug|ARM64">
+      <Configuration>Debug</Configuration>
+      <Platform>ARM64</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Debug|ARM64EC">
+      <Configuration>Debug</Configuration>
+      <Platform>ARM64EC</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Debug|Win32">
+      <Configuration>Debug</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|ARM64">
+      <Configuration>Release</Configuration>
+      <Platform>ARM64</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|ARM64EC">
+      <Configuration>Release</Configuration>
+      <Platform>ARM64EC</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|Win32">
+      <Configuration>Release</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Debug|x64">
+      <Configuration>Debug</Configuration>
+      <Platform>x64</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|x64">
+      <Configuration>Release</Configuration>
+      <Platform>x64</Platform>
+    </ProjectConfiguration>
+  </ItemGroup>
+  <PropertyGroup Label="Globals">
+    <VCProjectVersion>15.0</VCProjectVersion>
+    <ProjectGuid>{FEF7869F-750E-4C21-A04D-22707CC66879}</ProjectGuid>
+    <RootNamespace>mimalloc-test-override-dep</RootNamespace>
+    <WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
+    <ProjectName>mimalloc-test-override-dep</ProjectName>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+  <ImportGroup Label="ExtensionSettings">
+  </ImportGroup>
+  <ImportGroup Label="Shared">
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'" Label="PropertySheets">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'" Label="PropertySheets">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'" Label="PropertySheets">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'" Label="PropertySheets">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <PropertyGroup Label="UserMacros" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
+    <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
+    <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
+    <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'">
+    <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
+    <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'">
+    <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
+    <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
+    <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'">
+    <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
+    <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'">
+    <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
+    <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
+  </PropertyGroup>
+  <PropertyGroup Label="Vcpkg">
+    <VcpkgEnabled>false</VcpkgEnabled>
+  </PropertyGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <SDLCheck>true</SDLCheck>
+      <ConformanceMode>true</ConformanceMode>
+      <AdditionalIncludeDirectories>..\..\include</AdditionalIncludeDirectories>
+      <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
+      <ExceptionHandling>Sync</ExceptionHandling>
+      <CompileAs>Default</CompileAs>
+      <SupportJustMyCode>false</SupportJustMyCode>
+    </ClCompile>
+    <Link>
+      <SubSystem>Console</SubSystem>
+      <AdditionalDependencies>kernel32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+    <PostBuildEvent />
+    <PostBuildEvent>
+      <Command>
+      </Command>
+    </PostBuildEvent>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <SDLCheck>true</SDLCheck>
+      <ConformanceMode>true</ConformanceMode>
+      <AdditionalIncludeDirectories>..\..\include</AdditionalIncludeDirectories>
+      <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
+      <ExceptionHandling>Sync</ExceptionHandling>
+      <CompileAs>Default</CompileAs>
+      <SupportJustMyCode>false</SupportJustMyCode>
+    </ClCompile>
+    <Link>
+      <SubSystem>Console</SubSystem>
+      <EntryPointSymbol>
+      </EntryPointSymbol>
+      <AdditionalDependencies>kernel32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+    <PostBuildEvent />
+    <PostBuildEvent>
+      <Command>
+      </Command>
+    </PostBuildEvent>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <SDLCheck>true</SDLCheck>
+      <ConformanceMode>true</ConformanceMode>
+      <AdditionalIncludeDirectories>..\..\include</AdditionalIncludeDirectories>
+      <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
+      <ExceptionHandling>Sync</ExceptionHandling>
+      <CompileAs>Default</CompileAs>
+      <SupportJustMyCode>false</SupportJustMyCode>
+    </ClCompile>
+    <Link>
+      <SubSystem>Console</SubSystem>
+      <EntryPointSymbol>
+      </EntryPointSymbol>
+      <AdditionalDependencies>kernel32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+    <PostBuildEvent />
+    <PostBuildEvent>
+      <Command>
+      </Command>
+    </PostBuildEvent>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|ARM64EC'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <SDLCheck>true</SDLCheck>
+      <ConformanceMode>true</ConformanceMode>
+      <AdditionalIncludeDirectories>..\..\include</AdditionalIncludeDirectories>
+      <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
+      <ExceptionHandling>Sync</ExceptionHandling>
+      <CompileAs>Default</CompileAs>
+      <SupportJustMyCode>false</SupportJustMyCode>
+    </ClCompile>
+    <Link>
+      <SubSystem>Console</SubSystem>
+      <EntryPointSymbol>
+      </EntryPointSymbol>
+      <AdditionalDependencies>kernel32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+    <PostBuildEvent />
+    <PostBuildEvent>
+      <Command>
+      </Command>
+    </PostBuildEvent>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <SDLCheck>true</SDLCheck>
+      <ConformanceMode>true</ConformanceMode>
+      <AdditionalIncludeDirectories>..\..\include</AdditionalIncludeDirectories>
+      <PreprocessorDefinitions>_MBCS;%(PreprocessorDefinitions);NDEBUG</PreprocessorDefinitions>
+      <RuntimeLibrary>MultiThreadedDLL</RuntimeLibrary>
+    </ClCompile>
+    <Link>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <SubSystem>Console</SubSystem>
+      <AdditionalDependencies>kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+    <PostBuildEvent>
+      <Command>
+      </Command>
+    </PostBuildEvent>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <SDLCheck>true</SDLCheck>
+      <ConformanceMode>true</ConformanceMode>
+      <AdditionalIncludeDirectories>..\..\include</AdditionalIncludeDirectories>
+      <PreprocessorDefinitions>_MBCS;%(PreprocessorDefinitions);NDEBUG</PreprocessorDefinitions>
+      <RuntimeLibrary>MultiThreadedDLL</RuntimeLibrary>
+    </ClCompile>
+    <Link>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <SubSystem>Console</SubSystem>
+      <EntryPointSymbol>
+      </EntryPointSymbol>
+      <AdditionalDependencies>kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+    <PostBuildEvent>
+      <Command>
+      </Command>
+    </PostBuildEvent>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <SDLCheck>true</SDLCheck>
+      <ConformanceMode>true</ConformanceMode>
+      <AdditionalIncludeDirectories>..\..\include</AdditionalIncludeDirectories>
+      <PreprocessorDefinitions>_MBCS;%(PreprocessorDefinitions);NDEBUG</PreprocessorDefinitions>
+      <RuntimeLibrary>MultiThreadedDLL</RuntimeLibrary>
+    </ClCompile>
+    <Link>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <SubSystem>Console</SubSystem>
+      <EntryPointSymbol>
+      </EntryPointSymbol>
+      <AdditionalDependencies>kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+    <PostBuildEvent>
+      <Command>
+      </Command>
+    </PostBuildEvent>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|ARM64EC'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <SDLCheck>true</SDLCheck>
+      <ConformanceMode>true</ConformanceMode>
+      <AdditionalIncludeDirectories>..\..\include</AdditionalIncludeDirectories>
+      <PreprocessorDefinitions>_MBCS;%(PreprocessorDefinitions);NDEBUG</PreprocessorDefinitions>
+      <RuntimeLibrary>MultiThreadedDLL</RuntimeLibrary>
+    </ClCompile>
+    <Link>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <SubSystem>Console</SubSystem>
+      <EntryPointSymbol>
+      </EntryPointSymbol>
+      <AdditionalDependencies>kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+    <PostBuildEvent>
+      <Command>
+      </Command>
+    </PostBuildEvent>
+  </ItemDefinitionGroup>
+  <ItemGroup>
+    <ClCompile Include="..\..\test\main-override-dep.cpp" />
+  </ItemGroup>
+  <ItemGroup>
+    <ClInclude Include="..\..\test\main-override-dep.h" />
+  </ItemGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+  <ImportGroup Label="ExtensionTargets">
+  </ImportGroup>
+</Project>

ide/vs2022/mimalloc-override-test.vcxproj

@@ -148,6 +148,9 @@
     <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
   </PropertyGroup>
+  <PropertyGroup Label="Vcpkg">
+    <VcpkgEnabled>false</VcpkgEnabled>
+  </PropertyGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
     <ClCompile>
       <WarningLevel>Level3</WarningLevel>
@@ -347,6 +350,9 @@
     <ProjectReference Include="mimalloc-override-dll.vcxproj">
       <Project>{abb5eae7-b3e6-432e-b636-333449892ea7}</Project>
     </ProjectReference>
+    <ProjectReference Include="mimalloc-override-test-dep.vcxproj">
+      <Project>{fef7869f-750e-4c21-a04d-22707cc66879}</Project>
+    </ProjectReference>
   </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets">

ide/vs2022/mimalloc-test-api.vcxproj

@@ -148,6 +148,9 @@
     <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
   </PropertyGroup>
+  <PropertyGroup Label="Vcpkg">
+    <VcpkgEnabled>false</VcpkgEnabled>
+  </PropertyGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
     <ClCompile>
       <WarningLevel>Level3</WarningLevel>

ide/vs2022/mimalloc-test-stress.vcxproj

@@ -148,6 +148,9 @@
     <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
   </PropertyGroup>
+  <PropertyGroup Label="Vcpkg">
+    <VcpkgEnabled>false</VcpkgEnabled>
+  </PropertyGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
     <ClCompile>
       <WarningLevel>Level3</WarningLevel>

ide/vs2022/mimalloc-test.vcxproj

@@ -148,6 +148,9 @@
     <OutDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(Configuration)\</OutDir>
     <IntDir>$(ProjectDir)..\..\out\msvc-$(Platform)\$(ProjectName)\$(Configuration)\</IntDir>
   </PropertyGroup>
+  <PropertyGroup Label="Vcpkg">
+    <VcpkgEnabled>false</VcpkgEnabled>
+  </PropertyGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
     <ClCompile>
       <WarningLevel>Level3</WarningLevel>

ide/vs2022/mimalloc.sln

@@ -5,11 +5,13 @@ VisualStudioVersion = 17.12.35527.113
 MinimumVisualStudioVersion = 10.0.40219.1
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mimalloc-lib", "mimalloc-lib.vcxproj", "{ABB5EAE7-B3E6-432E-B636-333449892EA6}"
 EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mimalloc-test", "mimalloc-test.vcxproj", "{FEF7858F-750E-4C21-A04D-22707CC66878}"
+Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mimalloc-test-static", "mimalloc-test.vcxproj", "{FEF7858F-750E-4C21-A04D-22707CC66878}"
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mimalloc-override-dll", "mimalloc-override-dll.vcxproj", "{ABB5EAE7-B3E6-432E-B636-333449892EA7}"
 EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mimalloc-override-test", "mimalloc-override-test.vcxproj", "{FEF7868F-750E-4C21-A04D-22707CC66879}"
+Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mimalloc-test-override-dep", "mimalloc-override-test-dep.vcxproj", "{FEF7869F-750E-4C21-A04D-22707CC66879}"
+EndProject
+Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mimalloc-test-override", "mimalloc-override-test.vcxproj", "{FEF7868F-750E-4C21-A04D-22707CC66879}"
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mimalloc-test-stress", "mimalloc-test-stress.vcxproj", "{FEF7958F-750E-4C21-A04D-22707CC66878}"
 EndProject
@@ -75,6 +77,22 @@ Global
 		{ABB5EAE7-B3E6-432E-B636-333449892EA7}.Release|x64.Build.0 = Release|x64
 		{ABB5EAE7-B3E6-432E-B636-333449892EA7}.Release|x86.ActiveCfg = Release|Win32
 		{ABB5EAE7-B3E6-432E-B636-333449892EA7}.Release|x86.Build.0 = Release|Win32
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Debug|ARM64.ActiveCfg = Debug|ARM64
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Debug|ARM64.Build.0 = Debug|ARM64
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Debug|ARM64EC.ActiveCfg = Debug|ARM64EC
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Debug|ARM64EC.Build.0 = Debug|ARM64EC
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Debug|x64.ActiveCfg = Debug|x64
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Debug|x64.Build.0 = Debug|x64
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Debug|x86.ActiveCfg = Debug|Win32
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Debug|x86.Build.0 = Debug|Win32
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Release|ARM64.ActiveCfg = Release|ARM64
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Release|ARM64.Build.0 = Release|ARM64
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Release|ARM64EC.ActiveCfg = Release|ARM64EC
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Release|ARM64EC.Build.0 = Release|ARM64EC
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Release|x64.ActiveCfg = Release|x64
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Release|x64.Build.0 = Release|x64
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Release|x86.ActiveCfg = Release|Win32
+		{FEF7869F-750E-4C21-A04D-22707CC66879}.Release|x86.Build.0 = Release|Win32
 		{FEF7868F-750E-4C21-A04D-22707CC66879}.Debug|ARM64.ActiveCfg = Debug|ARM64
 		{FEF7868F-750E-4C21-A04D-22707CC66879}.Debug|ARM64.Build.0 = Debug|ARM64
 		{FEF7868F-750E-4C21-A04D-22707CC66879}.Debug|ARM64EC.ActiveCfg = Debug|ARM64EC

include/mimalloc-stats.h

@@ -0,0 +1,103 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2025, Microsoft Research, Daan Leijen
+This is free software; you can redistribute it and/or modify it under the
+terms of the MIT license. A copy of the license can be found in the file
+"LICENSE" at the root of this distribution.
+-----------------------------------------------------------------------------*/
+#pragma once
+#ifndef MIMALLOC_STATS_H
+#define MIMALLOC_STATS_H
+
+#include <mimalloc.h>
+#include <stdint.h>
+
+#define MI_STAT_VERSION   1   // increased on every backward incompatible change
+
+// count allocation over time
+typedef struct mi_stat_count_s {
+  int64_t total;                              // total allocated
+  int64_t peak;                               // peak allocation
+  int64_t current;                            // current allocation
+} mi_stat_count_t;
+
+// counters only increase
+typedef struct mi_stat_counter_s {
+  int64_t total;                              // total count
+} mi_stat_counter_t;
+
+#define MI_STAT_FIELDS() \
+  MI_STAT_COUNT(pages)                      /* count of mimalloc pages */ \
+  MI_STAT_COUNT(reserved)                   /* reserved memory bytes */ \
+  MI_STAT_COUNT(committed)                  /* committed bytes */ \
+  MI_STAT_COUNT(reset)                      /* reset bytes */ \
+  MI_STAT_COUNT(purged)                     /* purged bytes */ \
+  MI_STAT_COUNT(page_committed)             /* committed memory inside pages */ \
+  MI_STAT_COUNT(pages_abandoned)            /* abandonded pages count */ \
+  MI_STAT_COUNT(threads)                    /* number of threads */ \
+  MI_STAT_COUNT(malloc_normal)              /* allocated bytes <= MI_LARGE_OBJ_SIZE_MAX */ \
+  MI_STAT_COUNT(malloc_huge)                /* allocated bytes in huge pages */ \
+  MI_STAT_COUNT(malloc_requested)           /* malloc requested bytes */ \
+  \
+  MI_STAT_COUNTER(mmap_calls) \
+  MI_STAT_COUNTER(commit_calls) \
+  MI_STAT_COUNTER(reset_calls) \
+  MI_STAT_COUNTER(purge_calls) \
+  MI_STAT_COUNTER(arena_count)              /* number of memory arena's */ \
+  MI_STAT_COUNTER(malloc_normal_count)      /* number of blocks <= MI_LARGE_OBJ_SIZE_MAX */ \
+  MI_STAT_COUNTER(malloc_huge_count)        /* number of huge bloks */ \
+  MI_STAT_COUNTER(malloc_guarded_count)     /* number of allocations with guard pages */ \
+  \
+  /* internal statistics */ \
+  MI_STAT_COUNTER(arena_rollback_count) \
+  MI_STAT_COUNTER(arena_purges) \
+  MI_STAT_COUNTER(pages_extended)           /* number of page extensions */ \
+  MI_STAT_COUNTER(pages_retire)             /* number of pages that are retired */ \
+  MI_STAT_COUNTER(page_searches)            /* searches for a fresh page */ \
+  /* only on v1 and v2 */ \
+  MI_STAT_COUNT(segments) \
+  MI_STAT_COUNT(segments_abandoned) \
+  MI_STAT_COUNT(segments_cache) \
+  MI_STAT_COUNT(_segments_reserved) \
+  /* only on v3 */ \
+  MI_STAT_COUNTER(pages_reclaim_on_alloc) \
+  MI_STAT_COUNTER(pages_reclaim_on_free) \
+  MI_STAT_COUNTER(pages_reabandon_full) \
+  MI_STAT_COUNTER(pages_unabandon_busy_wait) \
+
+
+// Define the statistics structure
+#define MI_BIN_HUGE             (73U)   // see types.h
+#define MI_STAT_COUNT(stat)     mi_stat_count_t stat;
+#define MI_STAT_COUNTER(stat)   mi_stat_counter_t stat;
+
+typedef struct mi_stats_s
+{
+  int version;
+
+  MI_STAT_FIELDS()
+
+  // future extension
+  mi_stat_count_t   _stat_reserved[4];
+  mi_stat_counter_t _stat_counter_reserved[4];
+
+  // size segregated statistics
+  mi_stat_count_t   malloc_bins[MI_BIN_HUGE+1];   // allocation per size bin
+  mi_stat_count_t   page_bins[MI_BIN_HUGE+1];     // pages allocated per size bin
+} mi_stats_t;
+
+#undef MI_STAT_COUNT
+#undef MI_STAT_COUNTER
+
+// Exported definitions
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+mi_decl_export void  mi_stats_get( size_t stats_size, mi_stats_t* stats ) mi_attr_noexcept;
+mi_decl_export char* mi_stats_get_json( size_t buf_size, char* buf ) mi_attr_noexcept;    // use mi_free to free the result if the input buf == NULL
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // MIMALLOC_STATS_H

include/mimalloc.h

@@ -1,5 +1,5 @@
 /* ----------------------------------------------------------------------------
-Copyright (c) 2018-2023, Microsoft Research, Daan Leijen
+Copyright (c) 2018-2025, Microsoft Research, Daan Leijen
 This is free software; you can redistribute it and/or modify it under the
 terms of the MIT license. A copy of the license can be found in the file
 "LICENSE" at the root of this distribution.
@@ -8,7 +8,7 @@ terms of the MIT license. A copy of the license can be found in the file
 #ifndef MIMALLOC_H
 #define MIMALLOC_H
 
-#define MI_MALLOC_VERSION 189   // major + 2 digits minor
+#define MI_MALLOC_VERSION 223   // major + 2 digits minor
 
 // ------------------------------------------------------
 // Compiler specific attributes
@@ -97,6 +97,7 @@ terms of the MIT license. A copy of the license can be found in the file
 
 #include <stddef.h>     // size_t
 #include <stdbool.h>    // bool
+#include <stdint.h>     // INTPTR_MAX
 
 #ifdef __cplusplus
 extern "C" {
@@ -148,12 +149,12 @@ typedef void (mi_cdecl mi_error_fun)(int err, void* arg);
 mi_decl_export void mi_register_error(mi_error_fun* fun, void* arg);
 
 mi_decl_export void mi_collect(bool force)    mi_attr_noexcept;
-mi_decl_export void mi_collect_reduce(size_t target_thread_owned) mi_attr_noexcept;
 mi_decl_export int  mi_version(void)          mi_attr_noexcept;
 mi_decl_export void mi_stats_reset(void)      mi_attr_noexcept;
 mi_decl_export void mi_stats_merge(void)      mi_attr_noexcept;
 mi_decl_export void mi_stats_print(void* out) mi_attr_noexcept;  // backward compatibility: `out` is ignored and should be NULL
 mi_decl_export void mi_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept;
+mi_decl_export void mi_options_print(void)    mi_attr_noexcept;
 
 mi_decl_export void mi_process_init(void)     mi_attr_noexcept;
 mi_decl_export void mi_thread_init(void)      mi_attr_noexcept;
@@ -270,13 +271,14 @@ mi_decl_export bool mi_heap_visit_blocks(const mi_heap_t* heap, bool visit_block
 mi_decl_nodiscard mi_decl_export bool mi_is_in_heap_region(const void* p) mi_attr_noexcept;
 mi_decl_nodiscard mi_decl_export bool mi_is_redirected(void) mi_attr_noexcept;
 
-mi_decl_export int mi_reserve_huge_os_pages_interleave(size_t pages, size_t numa_nodes, size_t timeout_msecs) mi_attr_noexcept;
-mi_decl_export int mi_reserve_huge_os_pages_at(size_t pages, int numa_node, size_t timeout_msecs) mi_attr_noexcept;
+mi_decl_export int   mi_reserve_huge_os_pages_interleave(size_t pages, size_t numa_nodes, size_t timeout_msecs) mi_attr_noexcept;
+mi_decl_export int   mi_reserve_huge_os_pages_at(size_t pages, int numa_node, size_t timeout_msecs) mi_attr_noexcept;
 
-mi_decl_export int  mi_reserve_os_memory(size_t size, bool commit, bool allow_large) mi_attr_noexcept;
-mi_decl_export bool mi_manage_os_memory(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node) mi_attr_noexcept;
+mi_decl_export int   mi_reserve_os_memory(size_t size, bool commit, bool allow_large) mi_attr_noexcept;
+mi_decl_export bool  mi_manage_os_memory(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node) mi_attr_noexcept;
 
-mi_decl_export void mi_debug_show_arenas(bool show_inuse) mi_attr_noexcept;
+mi_decl_export void  mi_debug_show_arenas(void) mi_attr_noexcept;
+mi_decl_export void  mi_arenas_print(void) mi_attr_noexcept;
 
 // Experimental: heaps associated with specific memory arena's
 typedef int mi_arena_id_t;
@@ -291,17 +293,26 @@ mi_decl_nodiscard mi_decl_export mi_heap_t* mi_heap_new_in_arena(mi_arena_id_t a
 #endif
 
 
-// Experimental: allow sub-processes whose memory segments stay separated (and no reclamation between them)
-// Used for example for separate interpreter's in one process.
+// Experimental: allow sub-processes whose memory areas stay separated (and no reclamation between them)
+// Used for example for separate interpreters in one process.
 typedef void* mi_subproc_id_t;
 mi_decl_export mi_subproc_id_t mi_subproc_main(void);
 mi_decl_export mi_subproc_id_t mi_subproc_new(void);
 mi_decl_export void mi_subproc_delete(mi_subproc_id_t subproc);
 mi_decl_export void mi_subproc_add_current_thread(mi_subproc_id_t subproc); // this should be called right after a thread is created (and no allocation has taken place yet)
 
-// Experimental: visit abandoned heap areas (from threads that have been terminated)
+// Experimental: visit abandoned heap areas (that are not owned by a specific heap)
 mi_decl_export bool mi_abandoned_visit_blocks(mi_subproc_id_t subproc_id, int heap_tag, bool visit_blocks, mi_block_visit_fun* visitor, void* arg);
 
+// Experimental: objects followed by a guard page.
+// A sample rate of 0 disables guarded objects, while 1 uses a guard page for every object.
+// A seed of 0 uses a random start point. Only objects within the size bound are eligable for guard pages.
+mi_decl_export void mi_heap_guarded_set_sample_rate(mi_heap_t* heap, size_t sample_rate, size_t seed);
+mi_decl_export void mi_heap_guarded_set_size_bound(mi_heap_t* heap, size_t min, size_t max);
+
+// Experimental: communicate that the thread is part of a threadpool
+mi_decl_export void mi_thread_set_in_threadpool(void) mi_attr_noexcept;
+
 // Experimental: create a new heap with a specified heap tag. Set `allow_destroy` to false to allow the thread
 // to reclaim abandoned memory (with a compatible heap_tag and arena_id) but in that case `mi_heap_destroy` will
 // fall back to `mi_heap_delete`.
@@ -309,12 +320,8 @@ mi_decl_nodiscard mi_decl_export mi_heap_t* mi_heap_new_ex(int heap_tag, bool al
 
 // deprecated
 mi_decl_export int mi_reserve_huge_os_pages(size_t pages, double max_secs, size_t* pages_reserved) mi_attr_noexcept;
+mi_decl_export void mi_collect_reduce(size_t target_thread_owned) mi_attr_noexcept;
 
-// Experimental: objects followed by a guard page.
-// A sample rate of 0 disables guarded objects, while 1 uses a guard page for every object.
-// A seed of 0 uses a random start point. Only objects within the size bound are eligable for guard pages.
-mi_decl_export void mi_heap_guarded_set_sample_rate(mi_heap_t* heap, size_t sample_rate, size_t seed);
-mi_decl_export void mi_heap_guarded_set_size_bound(mi_heap_t* heap, size_t min, size_t max);
 
 
 // ------------------------------------------------------
@@ -379,6 +386,7 @@ typedef enum mi_option_e {
   mi_option_guarded_sample_rate,        // 1 out of N allocations in the min/max range will be guarded (=1000)
   mi_option_guarded_sample_seed,        // can be set to allow for a (more) deterministic re-execution when a guard page is triggered (=0)
   mi_option_target_segments_per_thread, // experimental (=0)
+  mi_option_generic_collect,            // collect heaps every N (=10000) generic allocation calls
   _mi_option_last,
   // legacy option names
   mi_option_large_os_pages = mi_option_allow_large_os_pages,

include/mimalloc/atomic.h

@@ -31,33 +31,33 @@ terms of the MIT license. A copy of the license can be found in the file
 #if defined(__cplusplus)
 // Use C++ atomics
 #include <atomic>
-#define  _Atomic(tp)            std::atomic<tp>
-#define  mi_atomic(name)        std::atomic_##name
-#define  mi_memory_order(name)  std::memory_order_##name
-#if (__cplusplus >= 202002L)    // c++20, see issue #571
- #define MI_ATOMIC_VAR_INIT(x)  x
+#define  _Atomic(tp)              std::atomic<tp>
+#define  mi_atomic(name)          std::atomic_##name
+#define  mi_memory_order(name)    std::memory_order_##name
+#if (__cplusplus >= 202002L)      // c++20, see issue #571
+ #define MI_ATOMIC_VAR_INIT(x)    x
 #elif !defined(ATOMIC_VAR_INIT)
- #define MI_ATOMIC_VAR_INIT(x)  x
+ #define MI_ATOMIC_VAR_INIT(x)    x
 #else
- #define MI_ATOMIC_VAR_INIT(x)  ATOMIC_VAR_INIT(x)
+ #define MI_ATOMIC_VAR_INIT(x)    ATOMIC_VAR_INIT(x)
 #endif
 #elif defined(_MSC_VER)
 // Use MSVC C wrapper for C11 atomics
-#define  _Atomic(tp)            tp
-#define  MI_ATOMIC_VAR_INIT(x)  x
-#define  mi_atomic(name)        mi_atomic_##name
-#define  mi_memory_order(name)  mi_memory_order_##name
+#define  _Atomic(tp)              tp
+#define  MI_ATOMIC_VAR_INIT(x)    x
+#define  mi_atomic(name)          mi_atomic_##name
+#define  mi_memory_order(name)    mi_memory_order_##name
 #else
 // Use C11 atomics
 #include <stdatomic.h>
-#define  mi_atomic(name)        atomic_##name
-#define  mi_memory_order(name)  memory_order_##name
+#define  mi_atomic(name)          atomic_##name
+#define  mi_memory_order(name)    memory_order_##name
 #if (__STDC_VERSION__ >= 201710L) // c17, see issue #735
- #define MI_ATOMIC_VAR_INIT(x) x
+ #define MI_ATOMIC_VAR_INIT(x)    x
 #elif !defined(ATOMIC_VAR_INIT)
- #define MI_ATOMIC_VAR_INIT(x) x
+ #define MI_ATOMIC_VAR_INIT(x)    x
 #else
- #define MI_ATOMIC_VAR_INIT(x) ATOMIC_VAR_INIT(x)
+ #define MI_ATOMIC_VAR_INIT(x)    ATOMIC_VAR_INIT(x)
 #endif
 #endif
 
@@ -129,6 +129,12 @@ static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub);
 static inline int64_t mi_atomic_addi64_relaxed(volatile int64_t* p, int64_t add) {
   return mi_atomic(fetch_add_explicit)((_Atomic(int64_t)*)p, add, mi_memory_order(relaxed));
 }
+static inline void mi_atomic_void_addi64_relaxed(volatile int64_t* p, const volatile int64_t* padd) {
+  const int64_t add = mi_atomic_load_relaxed((_Atomic(int64_t)*)padd);
+  if (add != 0) {
+    mi_atomic(fetch_add_explicit)((_Atomic(int64_t)*)p, add, mi_memory_order(relaxed));
+  }
+}
 static inline void mi_atomic_maxi64_relaxed(volatile int64_t* p, int64_t x) {
   int64_t current = mi_atomic_load_relaxed((_Atomic(int64_t)*)p);
   while (current < x && !mi_atomic_cas_weak_release((_Atomic(int64_t)*)p, &current, x)) { /* nothing */ };
@@ -260,6 +266,13 @@ static inline int64_t mi_atomic_addi64_relaxed(volatile _Atomic(int64_t)*p, int6
   return current;
 #endif
 }
+static inline void mi_atomic_void_addi64_relaxed(volatile int64_t* p, const volatile int64_t* padd) {
+  const int64_t add = *padd;
+  if (add != 0) {
+    mi_atomic_addi64_relaxed((volatile _Atomic(int64_t)*)p, add);
+  }
+}
+
 static inline void mi_atomic_maxi64_relaxed(volatile _Atomic(int64_t)*p, int64_t x) {
   int64_t current;
   do {
@@ -290,6 +303,7 @@ static inline bool mi_atomic_casi64_strong_acq_rel(volatile _Atomic(int64_t*)p,
 #define mi_atomic_cas_ptr_weak_release(tp,p,exp,des)    mi_atomic_cas_weak_release((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des)
 #define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des)    mi_atomic_cas_weak_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des)
 #define mi_atomic_cas_ptr_strong_release(tp,p,exp,des)  mi_atomic_cas_strong_release((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des)
+#define mi_atomic_exchange_ptr_relaxed(tp,p,x)          (tp*)mi_atomic_exchange_relaxed((_Atomic(uintptr_t)*)(p),(uintptr_t)x)
 #define mi_atomic_exchange_ptr_release(tp,p,x)          (tp*)mi_atomic_exchange_release((_Atomic(uintptr_t)*)(p),(uintptr_t)x)
 #define mi_atomic_exchange_ptr_acq_rel(tp,p,x)          (tp*)mi_atomic_exchange_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t)x)
 
@@ -356,8 +370,9 @@ static inline void mi_atomic_yield(void) {
   _mm_pause();
 }
 #elif (defined(__GNUC__) || defined(__clang__)) && \
-      (defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__armel__) || defined(__ARMEL__) || \
-       defined(__aarch64__) || defined(__powerpc__) || defined(__ppc__) || defined(__PPC__)) || defined(__POWERPC__)
+      (defined(__x86_64__) || defined(__i386__) || \
+       defined(__aarch64__) || defined(__arm__) || \
+       defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__))
 #if defined(__x86_64__) || defined(__i386__)
 static inline void mi_atomic_yield(void) {
   __asm__ volatile ("pause" ::: "memory");
@@ -366,10 +381,16 @@ static inline void mi_atomic_yield(void) {
 static inline void mi_atomic_yield(void) {
   __asm__ volatile("wfe");
 }
-#elif (defined(__arm__) && __ARM_ARCH__ >= 7)
+#elif defined(__arm__)
+#if __ARM_ARCH >= 7
 static inline void mi_atomic_yield(void) {
   __asm__ volatile("yield" ::: "memory");
 }
+#else
+static inline void mi_atomic_yield(void) {
+  __asm__ volatile ("nop" ::: "memory");
+}
+#endif
 #elif defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__)
 #ifdef __APPLE__
 static inline void mi_atomic_yield(void) {
@@ -380,10 +401,6 @@ static inline void mi_atomic_yield(void) {
   __asm__ __volatile__ ("or 27,27,27" ::: "memory");
 }
 #endif
-#elif defined(__armel__) || defined(__ARMEL__)
-static inline void mi_atomic_yield(void) {
-  __asm__ volatile ("nop" ::: "memory");
-}
 #endif
 #elif defined(__sun)
 // Fallback for other archs

include/mimalloc/internal.h

@@ -64,8 +64,8 @@ terms of the MIT license. A copy of the license can be found in the file
 
 // "libc.c"
 #include    <stdarg.h>
-void        _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args);
-void        _mi_snprintf(char* buf, size_t buflen, const char* fmt, ...);
+int         _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args);
+int         _mi_snprintf(char* buf, size_t buflen, const char* fmt, ...);
 char        _mi_toupper(char c);
 int         _mi_strnicmp(const char* s, const char* t, size_t n);
 void        _mi_strlcpy(char* dest, const char* src, size_t dest_size);
@@ -77,6 +77,7 @@ bool        _mi_getenv(const char* name, char* result, size_t result_size);
 // "options.c"
 void        _mi_fputs(mi_output_fun* out, void* arg, const char* prefix, const char* message);
 void        _mi_fprintf(mi_output_fun* out, void* arg, const char* fmt, ...);
+void        _mi_message(const char* fmt, ...);
 void        _mi_warning_message(const char* fmt, ...);
 void        _mi_verbose_message(const char* fmt, ...);
 void        _mi_trace_message(const char* fmt, ...);
@@ -126,6 +127,7 @@ bool        _mi_os_has_virtual_reserve(void);
 
 bool        _mi_os_reset(void* addr, size_t size);
 bool        _mi_os_commit(void* p, size_t size, bool* is_zero);
+bool        _mi_os_commit_ex(void* addr, size_t size, bool* is_zero, size_t stat_size);
 bool        _mi_os_decommit(void* addr, size_t size);
 bool        _mi_os_protect(void* addr, size_t size);
 bool        _mi_os_unprotect(void* addr, size_t size);
@@ -176,10 +178,11 @@ void        _mi_segment_map_freed_at(const mi_segment_t* segment);
 void        _mi_segment_map_unsafe_destroy(void);
 
 // "segment.c"
-mi_page_t*  _mi_segment_page_alloc(mi_heap_t* heap, size_t block_size, size_t page_alignment, mi_segments_tld_t* tld);
-void        _mi_segment_page_free(mi_page_t* page, bool force, mi_segments_tld_t* tld);
-void        _mi_segment_page_abandon(mi_page_t* page, mi_segments_tld_t* tld);
-uint8_t*    _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size);
+mi_page_t* _mi_segment_page_alloc(mi_heap_t* heap, size_t block_size, size_t page_alignment, mi_segments_tld_t* tld);
+void       _mi_segment_page_free(mi_page_t* page, bool force, mi_segments_tld_t* tld);
+void       _mi_segment_page_abandon(mi_page_t* page, mi_segments_tld_t* tld);
+bool       _mi_segment_try_reclaim_abandoned( mi_heap_t* heap, bool try_all, mi_segments_tld_t* tld);
+void       _mi_segment_collect(mi_segment_t* segment, bool force);
 
 #if MI_HUGE_PAGE_ABANDON
 void        _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block_t* block);
@@ -187,10 +190,11 @@ void        _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, m
 void        _mi_segment_huge_page_reset(mi_segment_t* segment, mi_page_t* page, mi_block_t* block);
 #endif
 
-void        _mi_segments_collect(bool force, mi_segments_tld_t* tld);
-void        _mi_abandoned_reclaim_all(mi_heap_t* heap, mi_segments_tld_t* tld);
-bool        _mi_segment_attempt_reclaim(mi_heap_t* heap, mi_segment_t* segment);
-bool        _mi_segment_visit_blocks(mi_segment_t* segment, int heap_tag, bool visit_blocks, mi_block_visit_fun* visitor, void* arg);
+uint8_t*   _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size); // page start for any page
+void       _mi_abandoned_reclaim_all(mi_heap_t* heap, mi_segments_tld_t* tld);
+void       _mi_abandoned_collect(mi_heap_t* heap, bool force, mi_segments_tld_t* tld);
+bool       _mi_segment_attempt_reclaim(mi_heap_t* heap, mi_segment_t* segment);
+bool       _mi_segment_visit_blocks(mi_segment_t* segment, int heap_tag, bool visit_blocks, mi_block_visit_fun* visitor, void* arg);
 
 // "page.c"
 void*       _mi_malloc_generic(mi_heap_t* heap, size_t size, bool zero, size_t huge_alignment)  mi_attr_noexcept mi_attr_malloc;
@@ -213,8 +217,8 @@ void        _mi_deferred_free(mi_heap_t* heap, bool force);
 void        _mi_page_free_collect(mi_page_t* page,bool force);
 void        _mi_page_reclaim(mi_heap_t* heap, mi_page_t* page);   // callback from segments
 
-size_t      _mi_bin_size(uint8_t bin);           // for stats
-uint8_t     _mi_bin(size_t size);                // for stats
+size_t      _mi_bin_size(size_t bin);            // for stats
+size_t      _mi_bin(size_t size);                // for stats
 
 // "heap.c"
 void        _mi_heap_init(mi_heap_t* heap, mi_tld_t* tld, mi_arena_id_t arena_id, bool noreclaim, uint8_t tag);
@@ -311,7 +315,7 @@ bool        _mi_page_is_valid(mi_page_t* page);
 #define MI_INIT64(x)  MI_INIT32(x),MI_INIT32(x)
 #define MI_INIT128(x) MI_INIT64(x),MI_INIT64(x)
 #define MI_INIT256(x) MI_INIT128(x),MI_INIT128(x)
-
+#define MI_INIT74(x)  MI_INIT64(x),MI_INIT8(x),x(),x()
 
 #include <string.h>
 // initialize a local variable to zero; use memset as compilers optimize constant sized memset's
@@ -340,12 +344,28 @@ static inline uintptr_t _mi_align_up(uintptr_t sz, size_t alignment) {
   }
 }
 
+// Align downwards
+static inline uintptr_t _mi_align_down(uintptr_t sz, size_t alignment) {
+  mi_assert_internal(alignment != 0);
+  uintptr_t mask = alignment - 1;
+  if ((alignment & mask) == 0) { // power of two?
+    return (sz & ~mask);
+  }
+  else {
+    return ((sz / alignment) * alignment);
+  }
+}
 
 // Align a pointer upwards
 static inline void* mi_align_up_ptr(void* p, size_t alignment) {
   return (void*)_mi_align_up((uintptr_t)p, alignment);
 }
 
+// Align a pointer downwards
+static inline void* mi_align_down_ptr(void* p, size_t alignment) {
+  return (void*)_mi_align_down((uintptr_t)p, alignment);
+}
+
 
 // Divide upwards: `s <= _mi_divide_up(s,d)*d < s+d`.
 static inline uintptr_t _mi_divide_up(uintptr_t size, size_t divider) {
@@ -369,6 +389,7 @@ static inline bool mi_mem_is_zero(const void* p, size_t size) {
   return true;
 }
 
+
 // Align a byte size to a size in _machine words_,
 // i.e. byte size == `wsize*sizeof(void*)`.
 static inline size_t _mi_wsize_from_size(size_t size) {
@@ -463,29 +484,44 @@ static inline mi_segment_t* _mi_ptr_segment(const void* p) {
   #endif
 }
 
+static inline mi_page_t* mi_slice_to_page(mi_slice_t* s) {
+  mi_assert_internal(s->slice_offset== 0 && s->slice_count > 0);
+  return (mi_page_t*)(s);
+}
+
+static inline mi_slice_t* mi_page_to_slice(mi_page_t* p) {
+  mi_assert_internal(p->slice_offset== 0 && p->slice_count > 0);
+  return (mi_slice_t*)(p);
+}
+
 // Segment belonging to a page
 static inline mi_segment_t* _mi_page_segment(const mi_page_t* page) {
   mi_assert_internal(page!=NULL);
   mi_segment_t* segment = _mi_ptr_segment(page);
-  mi_assert_internal(segment == NULL || page == &segment->pages[page->segment_idx]);
+  mi_assert_internal(segment == NULL || ((mi_slice_t*)page >= segment->slices && (mi_slice_t*)page < segment->slices + segment->slice_entries));
   return segment;
 }
 
-// used internally
-static inline size_t _mi_segment_page_idx_of(const mi_segment_t* segment, const void* p) {
-  // if (segment->page_size > MI_SEGMENT_SIZE) return &segment->pages[0];  // huge pages
-  ptrdiff_t diff = (uint8_t*)p - (uint8_t*)segment;
-  mi_assert_internal(diff >= 0 && (size_t)diff <= MI_SEGMENT_SIZE /* for huge alignment it can be equal */);
-  size_t idx = (size_t)diff >> segment->page_shift;
-  mi_assert_internal(idx < segment->capacity);
-  mi_assert_internal(segment->page_kind <= MI_PAGE_MEDIUM || idx == 0);
-  return idx;
+static inline mi_slice_t* mi_slice_first(const mi_slice_t* slice) {
+  mi_slice_t* start = (mi_slice_t*)((uint8_t*)slice - slice->slice_offset);
+  mi_assert_internal(start >= _mi_ptr_segment(slice)->slices);
+  mi_assert_internal(start->slice_offset == 0);
+  mi_assert_internal(start + start->slice_count > slice);
+  return start;
 }
 
-// Get the page containing the pointer
+// Get the page containing the pointer (performance critical as it is called in mi_free)
 static inline mi_page_t* _mi_segment_page_of(const mi_segment_t* segment, const void* p) {
-  size_t idx = _mi_segment_page_idx_of(segment, p);
-  return &((mi_segment_t*)segment)->pages[idx];
+  mi_assert_internal(p > (void*)segment);
+  ptrdiff_t diff = (uint8_t*)p - (uint8_t*)segment;
+  mi_assert_internal(diff > 0 && diff <= (ptrdiff_t)MI_SEGMENT_SIZE);
+  size_t idx = (size_t)diff >> MI_SEGMENT_SLICE_SHIFT;
+  mi_assert_internal(idx <= segment->slice_entries);
+  mi_slice_t* slice0 = (mi_slice_t*)&segment->slices[idx];
+  mi_slice_t* slice = mi_slice_first(slice0);  // adjust to the block that holds the page data
+  mi_assert_internal(slice->slice_offset == 0);
+  mi_assert_internal(slice >= segment->slices && slice < segment->slices + segment->slice_entries);
+  return mi_slice_to_page(slice);
 }
 
 // Quick page start for initialized pages
@@ -508,8 +544,8 @@ static inline size_t mi_page_block_size(const mi_page_t* page) {
 }
 
 static inline bool mi_page_is_huge(const mi_page_t* page) {
-  mi_assert_internal((page->is_huge && _mi_page_segment(page)->page_kind == MI_PAGE_HUGE) ||
-                     (!page->is_huge && _mi_page_segment(page)->page_kind != MI_PAGE_HUGE));
+  mi_assert_internal((page->is_huge && _mi_page_segment(page)->kind == MI_SEGMENT_HUGE) ||
+                     (!page->is_huge && _mi_page_segment(page)->kind != MI_SEGMENT_HUGE));
   return page->is_huge;
 }
 
@@ -521,7 +557,11 @@ static inline size_t mi_page_usable_block_size(const mi_page_t* page) {
 
 // size of a segment
 static inline size_t mi_segment_size(mi_segment_t* segment) {
-  return segment->segment_size;
+  return segment->segment_slices * MI_SEGMENT_SLICE_SIZE;
+}
+
+static inline uint8_t* mi_segment_end(mi_segment_t* segment) {
+  return (uint8_t*)segment + mi_segment_size(segment);
 }
 
 // Thread free access
@@ -676,12 +716,13 @@ static inline bool mi_is_in_same_segment(const void* p, const void* q) {
 }
 
 static inline bool mi_is_in_same_page(const void* p, const void* q) {
-  mi_segment_t* segmentp = _mi_ptr_segment(p);
-  mi_segment_t* segmentq = _mi_ptr_segment(q);
-  if (segmentp != segmentq) return false;
-  size_t idxp = _mi_segment_page_idx_of(segmentp, p);
-  size_t idxq = _mi_segment_page_idx_of(segmentq, q);
-  return (idxp == idxq);
+  mi_segment_t* segment = _mi_ptr_segment(p);
+  if (_mi_ptr_segment(q) != segment) return false;
+  // assume q may be invalid // return (_mi_segment_page_of(segment, p) == _mi_segment_page_of(segment, q));
+  mi_page_t* page = _mi_segment_page_of(segment, p);
+  size_t psize;
+  uint8_t* start = _mi_segment_page_start(segment, page, &psize);
+  return (start <= (uint8_t*)q && (uint8_t*)q < start + psize);
 }
 
 static inline uintptr_t mi_rotl(uintptr_t x, uintptr_t shift) {
@@ -763,6 +804,50 @@ static inline void mi_block_set_next(const mi_page_t* page, mi_block_t* block, c
 }
 
 
+// -------------------------------------------------------------------
+// commit mask
+// -------------------------------------------------------------------
+
+static inline void mi_commit_mask_create_empty(mi_commit_mask_t* cm) {
+  for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) {
+    cm->mask[i] = 0;
+  }
+}
+
+static inline void mi_commit_mask_create_full(mi_commit_mask_t* cm) {
+  for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) {
+    cm->mask[i] = ~((size_t)0);
+  }
+}
+
+static inline bool mi_commit_mask_is_empty(const mi_commit_mask_t* cm) {
+  for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) {
+    if (cm->mask[i] != 0) return false;
+  }
+  return true;
+}
+
+static inline bool mi_commit_mask_is_full(const mi_commit_mask_t* cm) {
+  for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) {
+    if (cm->mask[i] != ~((size_t)0)) return false;
+  }
+  return true;
+}
+
+// defined in `segment.c`:
+size_t _mi_commit_mask_committed_size(const mi_commit_mask_t* cm, size_t total);
+size_t _mi_commit_mask_next_run(const mi_commit_mask_t* cm, size_t* idx);
+
+#define mi_commit_mask_foreach(cm,idx,count) \
+  idx = 0; \
+  while ((count = _mi_commit_mask_next_run(cm,&idx)) > 0) {
+
+#define mi_commit_mask_foreach_end() \
+    idx += count; \
+  }
+
+
+
 /* -----------------------------------------------------------
   memory id's
 ----------------------------------------------------------- */
@@ -839,21 +924,21 @@ static inline size_t _mi_os_numa_node_count(void) {
 
 #include <limits.h>       // LONG_MAX
 #define MI_HAVE_FAST_BITSCAN
-static inline size_t mi_clz(uintptr_t x) {
-  if (x==0) return MI_INTPTR_BITS;
-#if (INTPTR_MAX == LONG_MAX)
-  return __builtin_clzl(x);
-#else
-  return __builtin_clzll(x);
-#endif
+static inline size_t mi_clz(size_t x) {
+  if (x==0) return MI_SIZE_BITS;
+  #if (SIZE_MAX == ULONG_MAX)
+    return __builtin_clzl(x);
+  #else
+    return __builtin_clzll(x);
+  #endif
 }
-static inline size_t mi_ctz(uintptr_t x) {
-  if (x==0) return MI_INTPTR_BITS;
-#if (INTPTR_MAX == LONG_MAX)
-  return __builtin_ctzl(x);
-#else
-  return __builtin_ctzll(x);
-#endif
+static inline size_t mi_ctz(size_t x) {
+  if (x==0) return MI_SIZE_BITS;
+  #if (SIZE_MAX == ULONG_MAX)
+    return __builtin_ctzl(x);
+  #else
+    return __builtin_ctzll(x);
+  #endif
 }
 
 #elif defined(_MSC_VER)
@@ -861,24 +946,24 @@ static inline size_t mi_ctz(uintptr_t x) {
 #include <limits.h>       // LONG_MAX
 #include <intrin.h>       // BitScanReverse64
 #define MI_HAVE_FAST_BITSCAN
-static inline size_t mi_clz(uintptr_t x) {
-  if (x==0) return MI_INTPTR_BITS;
+static inline size_t mi_clz(size_t x) {
+  if (x==0) return MI_SIZE_BITS;
   unsigned long idx;
-#if (INTPTR_MAX == LONG_MAX)
-  _BitScanReverse(&idx, x);
-#else
-  _BitScanReverse64(&idx, x);
-#endif
-  return ((MI_INTPTR_BITS - 1) - idx);
+  #if (SIZE_MAX == ULONG_MAX)
+    _BitScanReverse(&idx, x);
+  #else
+    _BitScanReverse64(&idx, x);
+  #endif
+  return ((MI_SIZE_BITS - 1) - idx);
 }
-static inline size_t mi_ctz(uintptr_t x) {
-  if (x==0) return MI_INTPTR_BITS;
+static inline size_t mi_ctz(size_t x) {
+  if (x==0) return MI_SIZE_BITS;
   unsigned long idx;
-#if (INTPTR_MAX == LONG_MAX)
-  _BitScanForward(&idx, x);
-#else
-  _BitScanForward64(&idx, x);
-#endif
+  #if (SIZE_MAX == ULONG_MAX)
+    _BitScanForward(&idx, x);
+  #else
+    _BitScanForward64(&idx, x);
+  #endif
   return idx;
 }
 
@@ -941,11 +1026,26 @@ static inline size_t mi_clz(size_t x) {
 
 #endif
 
-// "bit scan reverse": Return index of the highest bit (or MI_INTPTR_BITS if `x` is zero)
-static inline size_t mi_bsr(uintptr_t x) {
-  return (x==0 ? MI_INTPTR_BITS : MI_INTPTR_BITS - 1 - mi_clz(x));
+// "bit scan reverse": Return index of the highest bit (or MI_SIZE_BITS if `x` is zero)
+static inline size_t mi_bsr(size_t x) {
+  return (x==0 ? MI_SIZE_BITS : MI_SIZE_BITS - 1 - mi_clz(x));
 }
 
+size_t _mi_popcount_generic(size_t x);
+
+static inline size_t mi_popcount(size_t x) {
+  if (x<=1) return x;
+  if (x==SIZE_MAX) return MI_SIZE_BITS;
+  #if defined(__GNUC__)
+    #if (SIZE_MAX == ULONG_MAX)
+      return __builtin_popcountl(x);
+    #else
+      return __builtin_popcountll(x);
+    #endif
+  #else
+    return _mi_popcount_generic(x);
+  #endif
+}
 
 // ---------------------------------------------------------------------------------
 // Provide our own `_mi_memcpy` for potential performance optimizations.

include/mimalloc/prim.h

@@ -22,14 +22,14 @@ terms of the MIT license. A copy of the license can be found in the file
 
 // OS memory configuration
 typedef struct mi_os_mem_config_s {
-  size_t  page_size;            // default to 4KiB
-  size_t  large_page_size;      // 0 if not supported, usually 2MiB (4MiB on Windows)
-  size_t  alloc_granularity;    // smallest allocation size (usually 4KiB, on Windows 64KiB)
-  size_t  physical_memory;      // physical memory size
-  size_t  virtual_address_bits; // usually 48 or 56 bits on 64-bit systems. (used to determine secure randomization)
-  bool    has_overcommit;       // can we reserve more memory than can be actually committed?
-  bool    has_partial_free;     // can allocated blocks be freed partially? (true for mmap, false for VirtualAlloc)
-  bool    has_virtual_reserve;  // supports virtual address space reservation? (if true we can reserve virtual address space without using commit or physical memory)
+  size_t  page_size;              // default to 4KiB
+  size_t  large_page_size;        // 0 if not supported, usually 2MiB (4MiB on Windows)
+  size_t  alloc_granularity;      // smallest allocation size (usually 4KiB, on Windows 64KiB)
+  size_t  physical_memory_in_kib; // physical memory size in KiB
+  size_t  virtual_address_bits;   // usually 48 or 56 bits on 64-bit systems. (used to determine secure randomization)
+  bool    has_overcommit;         // can we reserve more memory than can be actually committed?
+  bool    has_partial_free;       // can allocated blocks be freed partially? (true for mmap, false for VirtualAlloc)
+  bool    has_virtual_reserve;    // supports virtual address space reservation? (if true we can reserve virtual address space without using commit or physical memory)
 } mi_os_mem_config_t;
 
 // Initialize
@@ -124,7 +124,7 @@ void _mi_prim_thread_associate_default_heap(mi_heap_t* heap);
 //-------------------------------------------------------------------
 // Access to TLS (thread local storage) slots.
 // We need fast access to both a unique thread id (in `free.c:mi_free`) and
-// to a thread-local heap pointer (in `alloc.c:mi_malloc`). 
+// to a thread-local heap pointer (in `alloc.c:mi_malloc`).
 // To achieve this we use specialized code for various platforms.
 //-------------------------------------------------------------------
 

include/mimalloc/types.h

@@ -13,14 +13,16 @@ terms of the MIT license. A copy of the license can be found in the file
 // mi_heap_t      : all data for a thread-local heap, contains
 //                  lists of all managed heap pages.
 // mi_segment_t   : a larger chunk of memory (32GiB) from where pages
-//                  are allocated.
-// mi_page_t      : a mimalloc page (usually 64KiB or 512KiB) from
+//                  are allocated. A segment is divided in slices (64KiB) from
+//                  which pages are allocated.
+// mi_page_t      : a "mimalloc" page (usually 64KiB or 512KiB) from
 //                  where objects are allocated.
 //                  Note: we write "OS page" for OS memory pages while
 //                  using plain "page" for mimalloc pages (`mi_page_t`).
 // --------------------------------------------------------------------------
 
 
+#include <mimalloc-stats.h>
 #include <stddef.h>   // ptrdiff_t
 #include <stdint.h>   // uintptr_t, uint16_t, etc
 #include "atomic.h"   // _Atomic
@@ -65,10 +67,10 @@ terms of the MIT license. A copy of the license can be found in the file
 // #define MI_DEBUG 2  // + internal assertion checks
 // #define MI_DEBUG 3  // + extensive internal invariant checking (cmake -DMI_DEBUG_FULL=ON)
 #if !defined(MI_DEBUG)
-#if !defined(NDEBUG) || defined(_DEBUG)
-#define MI_DEBUG 2
-#else
+#if defined(MI_BUILD_RELEASE) || defined(NDEBUG)
 #define MI_DEBUG 0
+#else
+#define MI_DEBUG 2
 #endif
 #endif
 
@@ -166,55 +168,68 @@ typedef int32_t  mi_ssize_t;
 // ------------------------------------------------------
 
 // Main tuning parameters for segment and page sizes
-// Sizes for 64-bit, divide by two for 32-bit
+// Sizes for 64-bit (usually divide by two for 32-bit)
+#ifndef MI_SEGMENT_SLICE_SHIFT
+#define MI_SEGMENT_SLICE_SHIFT            (13 + MI_INTPTR_SHIFT)         // 64KiB  (32KiB on 32-bit)
+#endif
+
+#ifndef MI_SEGMENT_SHIFT
+#if MI_INTPTR_SIZE > 4
+#define MI_SEGMENT_SHIFT                  ( 9 + MI_SEGMENT_SLICE_SHIFT)  // 32MiB
+#else
+#define MI_SEGMENT_SHIFT                  ( 7 + MI_SEGMENT_SLICE_SHIFT)  // 4MiB on 32-bit
+#endif
+#endif
+
 #ifndef MI_SMALL_PAGE_SHIFT
-#define MI_SMALL_PAGE_SHIFT               (13 + MI_INTPTR_SHIFT)      // 64KiB
+#define MI_SMALL_PAGE_SHIFT               (MI_SEGMENT_SLICE_SHIFT)       // 64KiB
 #endif
 #ifndef MI_MEDIUM_PAGE_SHIFT
-#define MI_MEDIUM_PAGE_SHIFT              ( 3 + MI_SMALL_PAGE_SHIFT)  // 512KiB
-#endif
-#ifndef MI_LARGE_PAGE_SHIFT
-#define MI_LARGE_PAGE_SHIFT               ( 3 + MI_MEDIUM_PAGE_SHIFT) // 4MiB
-#endif
-#ifndef MI_SEGMENT_SHIFT
-#define MI_SEGMENT_SHIFT                  ( MI_LARGE_PAGE_SHIFT)      // 4MiB -- must be equal to `MI_LARGE_PAGE_SHIFT`
+#define MI_MEDIUM_PAGE_SHIFT              ( 3 + MI_SMALL_PAGE_SHIFT)     // 512KiB
 #endif
 
 // Derived constants
 #define MI_SEGMENT_SIZE                   (MI_ZU(1)<<MI_SEGMENT_SHIFT)
-#define MI_SEGMENT_ALIGN                  (MI_SEGMENT_SIZE)
+#define MI_SEGMENT_ALIGN                  MI_SEGMENT_SIZE
 #define MI_SEGMENT_MASK                   ((uintptr_t)(MI_SEGMENT_ALIGN - 1))
+#define MI_SEGMENT_SLICE_SIZE             (MI_ZU(1)<< MI_SEGMENT_SLICE_SHIFT)
+#define MI_SLICES_PER_SEGMENT             (MI_SEGMENT_SIZE / MI_SEGMENT_SLICE_SIZE) // 1024
 
 #define MI_SMALL_PAGE_SIZE                (MI_ZU(1)<<MI_SMALL_PAGE_SHIFT)
 #define MI_MEDIUM_PAGE_SIZE               (MI_ZU(1)<<MI_MEDIUM_PAGE_SHIFT)
-#define MI_LARGE_PAGE_SIZE                (MI_ZU(1)<<MI_LARGE_PAGE_SHIFT)
 
-#define MI_SMALL_PAGES_PER_SEGMENT        (MI_SEGMENT_SIZE/MI_SMALL_PAGE_SIZE)
-#define MI_MEDIUM_PAGES_PER_SEGMENT       (MI_SEGMENT_SIZE/MI_MEDIUM_PAGE_SIZE)
-#define MI_LARGE_PAGES_PER_SEGMENT        (MI_SEGMENT_SIZE/MI_LARGE_PAGE_SIZE)
-
-// The max object size are checked to not waste more than 12.5% internally over the page sizes.
-// (Except for large pages since huge objects are allocated in 4MiB chunks)
-#define MI_SMALL_OBJ_SIZE_MAX             (MI_SMALL_PAGE_SIZE/4)   // 16KiB
-#define MI_MEDIUM_OBJ_SIZE_MAX            (MI_MEDIUM_PAGE_SIZE/4)  // 128KiB
-#define MI_LARGE_OBJ_SIZE_MAX             (MI_LARGE_PAGE_SIZE/2)   // 2MiB
+#define MI_SMALL_OBJ_SIZE_MAX             (MI_SMALL_PAGE_SIZE/8)   // 8 KiB on 64-bit
+#define MI_MEDIUM_OBJ_SIZE_MAX            (MI_MEDIUM_PAGE_SIZE/8)  // 64 KiB on 64-bit
+#define MI_MEDIUM_OBJ_WSIZE_MAX           (MI_MEDIUM_OBJ_SIZE_MAX/MI_INTPTR_SIZE)
+#define MI_LARGE_OBJ_SIZE_MAX             (MI_SEGMENT_SIZE/2)      // 16 MiB on 64-bit
 #define MI_LARGE_OBJ_WSIZE_MAX            (MI_LARGE_OBJ_SIZE_MAX/MI_INTPTR_SIZE)
 
 // Maximum number of size classes. (spaced exponentially in 12.5% increments)
-#define MI_BIN_HUGE  (73U)
+#if MI_BIN_HUGE != 73U
+#error "mimalloc internal: expecting 73 bins"
+#endif
 
-#if (MI_LARGE_OBJ_WSIZE_MAX >= 655360)
+#if (MI_MEDIUM_OBJ_WSIZE_MAX >= 655360)
 #error "mimalloc internal: define more bins"
 #endif
 
 // Maximum block size for which blocks are guaranteed to be block size aligned. (see `segment.c:_mi_segment_page_start`)
-#define MI_MAX_ALIGN_GUARANTEE   (MI_MEDIUM_OBJ_SIZE_MAX)
+#define MI_MAX_ALIGN_GUARANTEE            (MI_MEDIUM_OBJ_SIZE_MAX)
 
 // Alignments over MI_BLOCK_ALIGNMENT_MAX are allocated in dedicated huge page segments
-#define MI_BLOCK_ALIGNMENT_MAX   (MI_SEGMENT_SIZE >> 1)
+#define MI_BLOCK_ALIGNMENT_MAX            (MI_SEGMENT_SIZE >> 1)
 
-// We never allocate more than PTRDIFF_MAX (see also <https://sourceware.org/ml/libc-announce/2019/msg00001.html>)
+// Maximum slice count (255) for which we can find the page for interior pointers
+#define MI_MAX_SLICE_OFFSET_COUNT         ((MI_BLOCK_ALIGNMENT_MAX / MI_SEGMENT_SLICE_SIZE) - 1)
+
+// we never allocate more than PTRDIFF_MAX (see also <https://sourceware.org/ml/libc-announce/2019/msg00001.html>)
+// on 64-bit+ systems we also limit the maximum allocation size such that the slice count fits in 32-bits. (issue #877)
+#if (PTRDIFF_MAX > INT32_MAX) && (PTRDIFF_MAX >= (MI_SEGMENT_SLIZE_SIZE * UINT32_MAX))
+#define MI_MAX_ALLOC_SIZE   (MI_SEGMENT_SLICE_SIZE * (UINT32_MAX-1))
+#else
 #define MI_MAX_ALLOC_SIZE   PTRDIFF_MAX
+#endif
+
 
 // ------------------------------------------------------
 // Mimalloc pages contain allocated blocks
@@ -293,8 +308,8 @@ typedef uintptr_t mi_thread_free_t;
 // Notes:
 // - Access is optimized for `free.c:mi_free` and `alloc.c:mi_page_alloc`
 // - Using `uint16_t` does not seem to slow things down
-// - The size is 10 words on 64-bit which helps the page index calculations
-//   (and 12 words on 32-bit, and encoded free lists add 2 words)
+// - The size is 12 words on 64-bit which helps the page index calculations
+//   (and 14 words on 32-bit, and encoded free lists add 2 words)
 // - `xthread_free` uses the bottom bits as a delayed-free flags to optimize
 //   concurrent frees where only the first concurrent free adds to the owning
 //   heap `thread_delayed_free` list (see `free.c:mi_free_block_mt`).
@@ -304,12 +319,12 @@ typedef uintptr_t mi_thread_free_t;
 //   will be freed correctly even if only other threads free blocks.
 typedef struct mi_page_s {
   // "owned" by the segment
-  uint8_t               segment_idx;       // index in the segment `pages` array, `page == &segment->pages[page->segment_idx]`
-  uint8_t               segment_in_use:1;  // `true` if the segment allocated this page
+  uint32_t              slice_count;       // slices in this page (0 if not a page)
+  uint32_t              slice_offset;      // distance from the actual page data slice (0 if a page)
   uint8_t               is_committed:1;    // `true` if the page virtual memory is committed
   uint8_t               is_zero_init:1;    // `true` if the page was initially zero initialized
-  uint8_t               is_huge:1;         // `true` if the page is in a huge segment
-
+  uint8_t               is_huge:1;         // `true` if the page is in a huge segment (`segment->kind == MI_SEGMENT_HUGE`)
+                                           // padding
   // layout like this to optimize access in `mi_malloc` and `mi_free`
   uint16_t              capacity;          // number of blocks committed, must be the first field, see `segment.c:page_clear`
   uint16_t              reserved;          // number of blocks reserved in memory
@@ -333,12 +348,11 @@ typedef struct mi_page_s {
   _Atomic(mi_thread_free_t) xthread_free;  // list of deferred free blocks freed by other threads
   _Atomic(uintptr_t)        xheap;
 
-  struct mi_page_s*     next;              // next page owned by the heap with the same `block_size`
-  struct mi_page_s*     prev;              // previous page owned by the heap with the same `block_size`
+  struct mi_page_s*     next;              // next page owned by this thread with the same `block_size`
+  struct mi_page_s*     prev;              // previous page owned by this thread with the same `block_size`
 
-  #if MI_INTPTR_SIZE==4                    // pad to 12 words on 32-bit
+  // 64-bit 11 words, 32-bit 13 words, (+2 for secure)
   void* padding[1];
-  #endif
 } mi_page_t;
 
 
@@ -351,10 +365,44 @@ typedef enum mi_page_kind_e {
   MI_PAGE_SMALL,    // small blocks go into 64KiB pages inside a segment
   MI_PAGE_MEDIUM,   // medium blocks go into 512KiB pages inside a segment
   MI_PAGE_LARGE,    // larger blocks go into a single page spanning a whole segment
-  MI_PAGE_HUGE      // a huge page is a single page in a segment of variable size (but still 2MiB aligned)
-                    // used for blocks `> MI_LARGE_OBJ_SIZE_MAX` or an alignment `> MI_BLOCK_ALIGNMENT_MAX`.
+  MI_PAGE_HUGE      // a huge page is a single page in a segment of variable size
+                    // used for blocks `> MI_LARGE_OBJ_SIZE_MAX` or an aligment `> MI_BLOCK_ALIGNMENT_MAX`.
 } mi_page_kind_t;
 
+typedef enum mi_segment_kind_e {
+  MI_SEGMENT_NORMAL, // MI_SEGMENT_SIZE size with pages inside.
+  MI_SEGMENT_HUGE,   // segment with just one huge page inside.
+} mi_segment_kind_t;
+
+// ------------------------------------------------------
+// A segment holds a commit mask where a bit is set if
+// the corresponding MI_COMMIT_SIZE area is committed.
+// The MI_COMMIT_SIZE must be a multiple of the slice
+// size. If it is equal we have the most fine grained
+// decommit (but setting it higher can be more efficient).
+// The MI_MINIMAL_COMMIT_SIZE is the minimal amount that will
+// be committed in one go which can be set higher than
+// MI_COMMIT_SIZE for efficiency (while the decommit mask
+// is still tracked in fine-grained MI_COMMIT_SIZE chunks)
+// ------------------------------------------------------
+
+#define MI_MINIMAL_COMMIT_SIZE      (1*MI_SEGMENT_SLICE_SIZE)
+#define MI_COMMIT_SIZE              (MI_SEGMENT_SLICE_SIZE)              // 64KiB
+#define MI_COMMIT_MASK_BITS         (MI_SEGMENT_SIZE / MI_COMMIT_SIZE)
+#define MI_COMMIT_MASK_FIELD_BITS    MI_SIZE_BITS
+#define MI_COMMIT_MASK_FIELD_COUNT  (MI_COMMIT_MASK_BITS / MI_COMMIT_MASK_FIELD_BITS)
+
+#if (MI_COMMIT_MASK_BITS != (MI_COMMIT_MASK_FIELD_COUNT * MI_COMMIT_MASK_FIELD_BITS))
+#error "the segment size must be exactly divisible by the (commit size * size_t bits)"
+#endif
+
+typedef struct mi_commit_mask_s {
+  size_t mask[MI_COMMIT_MASK_FIELD_COUNT];
+} mi_commit_mask_t;
+
+typedef mi_page_t  mi_slice_t;
+typedef int64_t    mi_msecs_t;
+
 
 // ---------------------------------------------------------------
 // a memory id tracks the provenance of arena/OS allocated memory
@@ -398,43 +446,57 @@ typedef struct mi_memid_s {
 } mi_memid_t;
 
 
-// ---------------------------------------------------------------
-// Segments contain mimalloc pages
-// ---------------------------------------------------------------
+// -----------------------------------------------------------------------------------------
+// Segments are large allocated memory blocks (32mb on 64 bit) from arenas or the OS.
+//
+// Inside segments we allocated fixed size mimalloc pages (`mi_page_t`) that contain blocks.
+// The start of a segment is this structure with a fixed number of slice entries (`slices`)
+// usually followed by a guard OS page and the actual allocation area with pages.
+// While a page is not allocated, we view it's data as a `mi_slice_t` (instead of a `mi_page_t`).
+// Of any free area, the first slice has the info and `slice_offset == 0`; for any subsequent
+// slices part of the area, the `slice_offset` is the byte offset back to the first slice
+// (so we can quickly find the page info on a free, `internal.h:_mi_segment_page_of`).
+// For slices, the `block_size` field is repurposed to signify if a slice is used (`1`) or not (`0`).
+// Small and medium pages use a fixed amount of slices to reduce slice fragmentation, while
+// large and huge pages span a variable amount of slices.
+
 typedef struct mi_subproc_s mi_subproc_t;
 
-// Segments are large allocated memory blocks (2MiB on 64 bit) from the OS.
-// Inside segments we allocated fixed size _pages_ that contain blocks.
 typedef struct mi_segment_s {
   // constant fields
-  mi_memid_t           memid;            // memory id to track provenance
-  bool                 allow_decommit;
-  bool                 allow_purge;
-  size_t               segment_size;     // for huge pages this may be different from `MI_SEGMENT_SIZE`
-  mi_subproc_t*        subproc;          // segment belongs to sub process
+  mi_memid_t        memid;              // memory id for arena/OS allocation
+  bool              allow_decommit;     // can we decommmit the memory
+  bool              allow_purge;        // can we purge the memory (reset or decommit)
+  size_t            segment_size;
+  mi_subproc_t*     subproc;            // segment belongs to sub process
 
   // segment fields
-  struct mi_segment_s* next;             // must be the first (non-constant) segment field  -- see `segment.c:segment_init`
-  struct mi_segment_s* prev;
-  bool                 was_reclaimed;    // true if it was reclaimed (used to limit reclaim-on-free reclamation)
-  bool                 dont_free;        // can be temporarily true to ensure the segment is not freed
+  mi_msecs_t        purge_expire;       // purge slices in the `purge_mask` after this time
+  mi_commit_mask_t  purge_mask;         // slices that can be purged
+  mi_commit_mask_t  commit_mask;        // slices that are currently committed
 
-  size_t               abandoned;        // abandoned pages (i.e. the original owning thread stopped) (`abandoned <= used`)
-  size_t               abandoned_visits; // count how often this segment is visited for reclaiming (to force reclaim if it is too long)
+  // from here is zero initialized
+  struct mi_segment_s* next;            // the list of freed segments in the cache (must be first field, see `segment.c:mi_segment_init`)
+  bool              was_reclaimed;      // true if it was reclaimed (used to limit on-free reclamation)
+  bool              dont_free;          // can be temporarily true to ensure the segment is not freed
 
-  size_t               used;             // count of pages in use (`used <= capacity`)
-  size_t               capacity;         // count of available pages (`#free + used`)
-  size_t               segment_info_size;// space we are using from the first page for segment meta-data and possible guard pages.
-  uintptr_t            cookie;           // verify addresses in secure mode: `_mi_ptr_cookie(segment) == segment->cookie`
+  size_t            abandoned;          // abandoned pages (i.e. the original owning thread stopped) (`abandoned <= used`)
+  size_t            abandoned_visits;   // count how often this segment is visited during abondoned reclamation (to force reclaim if it takes too long)
+  size_t            used;               // count of pages in use
+  uintptr_t         cookie;             // verify addresses in debug mode: `mi_ptr_cookie(segment) == segment->cookie`
 
   struct mi_segment_s* abandoned_os_next; // only used for abandoned segments outside arena's, and only if `mi_option_visit_abandoned` is enabled
   struct mi_segment_s* abandoned_os_prev;
 
+  size_t            segment_slices;      // for huge segments this may be different from `MI_SLICES_PER_SEGMENT`
+  size_t            segment_info_slices; // initial count of slices that we are using for segment info and possible guard pages.
+
   // layout like this to optimize access in `mi_free`
+  mi_segment_kind_t kind;
+  size_t            slice_entries;       // entries in the `slices` array, at most `MI_SLICES_PER_SEGMENT`
   _Atomic(mi_threadid_t) thread_id;      // unique id of the thread owning this segment
-  size_t               page_shift;       // `1 << page_shift` == the page sizes == `page->block_size * page->reserved` (unless the first page, then `-segment_info_size`).
-  mi_page_kind_t       page_kind;        // kind of pages: small, medium, large, or huge
-  mi_page_t            pages[1];         // up to `MI_SMALL_PAGES_PER_SEGMENT` pages
+
+  mi_slice_t        slices[MI_SLICES_PER_SEGMENT+1];  // one extra final entry for huge blocks with large alignment
 } mi_segment_t;
 
 
@@ -500,6 +562,8 @@ struct mi_heap_s {
   size_t                page_count;                          // total number of pages in the `pages` queues.
   size_t                page_retired_min;                    // smallest retired index (retired pages are fully free, but still in the page queues)
   size_t                page_retired_max;                    // largest retired index into the `pages` array.
+  long                  generic_count;                       // how often is `_mi_malloc_generic` called?
+  long                  generic_collect_count;               // how often is `_mi_malloc_generic` called without collecting?
   mi_heap_t*            next;                                // list of heaps per thread
   bool                  no_reclaim;                          // `true` if this heap should not reclaim abandoned pages
   uint8_t               tag;                                 // custom tag, can be used for separating heaps based on the object types
@@ -515,6 +579,60 @@ struct mi_heap_s {
 };
 
 
+// ------------------------------------------------------
+// Sub processes do not reclaim or visit segments
+// from other sub processes. These are essentially the
+// static variables of a process.
+// ------------------------------------------------------
+
+struct mi_subproc_s {
+  _Atomic(size_t)    abandoned_count;         // count of abandoned segments for this sub-process
+  _Atomic(size_t)    abandoned_os_list_count; // count of abandoned segments in the os-list
+  mi_lock_t          abandoned_os_lock;       // lock for the abandoned os segment list (outside of arena's) (this lock protect list operations)
+  mi_lock_t          abandoned_os_visit_lock; // ensure only one thread per subproc visits the abandoned os list
+  mi_segment_t*      abandoned_os_list;       // doubly-linked list of abandoned segments outside of arena's (in OS allocated memory)
+  mi_segment_t*      abandoned_os_list_tail;  // the tail-end of the list
+  mi_memid_t         memid;                   // provenance of this memory block
+};
+
+
+// ------------------------------------------------------
+// Thread Local data
+// ------------------------------------------------------
+
+// A "span" is is an available range of slices. The span queues keep
+// track of slice spans of at most the given `slice_count` (but more than the previous size class).
+typedef struct mi_span_queue_s {
+  mi_slice_t* first;
+  mi_slice_t* last;
+  size_t      slice_count;
+} mi_span_queue_t;
+
+#define MI_SEGMENT_BIN_MAX (35)     // 35 == mi_segment_bin(MI_SLICES_PER_SEGMENT)
+
+// Segments thread local data
+typedef struct mi_segments_tld_s {
+  mi_span_queue_t     spans[MI_SEGMENT_BIN_MAX+1];  // free slice spans inside segments
+  size_t              count;        // current number of segments;
+  size_t              peak_count;   // peak number of segments
+  size_t              current_size; // current size of all segments
+  size_t              peak_size;    // peak size of all segments
+  size_t              reclaim_count;// number of reclaimed (abandoned) segments
+  mi_subproc_t*       subproc;      // sub-process this thread belongs to.
+  mi_stats_t*         stats;        // points to tld stats
+} mi_segments_tld_t;
+
+// Thread local data
+struct mi_tld_s {
+  unsigned long long  heartbeat;     // monotonic heartbeat count
+  bool                recurse;       // true if deferred was called; used to prevent infinite recursion.
+  mi_heap_t*          heap_backing;  // backing heap of this thread (cannot be deleted)
+  mi_heap_t*          heaps;         // list of heaps in this thread (so we can abandon all when the thread terminates)
+  mi_segments_tld_t   segments;      // segment tld
+  mi_stats_t          stats;         // statistics
+};
+
+
 
 // ------------------------------------------------------
 // Debug
@@ -550,10 +668,10 @@ void _mi_assert_fail(const char* assertion, const char* fname, unsigned int line
 #define mi_assert_expensive(x)
 #endif
 
+
 // ------------------------------------------------------
 // Statistics
 // ------------------------------------------------------
-
 #ifndef MI_STAT
 #if (MI_DEBUG>0)
 #define MI_STAT 2
@@ -562,58 +680,9 @@ void _mi_assert_fail(const char* assertion, const char* fname, unsigned int line
 #endif
 #endif
 
-typedef struct mi_stat_count_s {
-  int64_t allocated;
-  int64_t freed;
-  int64_t peak;
-  int64_t current;
-} mi_stat_count_t;
-
-typedef struct mi_stat_counter_s {
-  int64_t total;
-  int64_t count;
-} mi_stat_counter_t;
-
-typedef struct mi_stats_s {
-  mi_stat_count_t segments;
-  mi_stat_count_t pages;
-  mi_stat_count_t reserved;
-  mi_stat_count_t committed;
-  mi_stat_count_t reset;
-  mi_stat_count_t purged;
-  mi_stat_count_t page_committed;
-  mi_stat_count_t segments_abandoned;
-  mi_stat_count_t pages_abandoned;
-  mi_stat_count_t threads;
-  mi_stat_count_t normal;
-  mi_stat_count_t huge;
-  mi_stat_count_t giant;
-  mi_stat_count_t malloc;
-  mi_stat_count_t segments_cache;
-  mi_stat_counter_t pages_extended;
-  mi_stat_counter_t mmap_calls;
-  mi_stat_counter_t commit_calls;
-  mi_stat_counter_t reset_calls;
-  mi_stat_counter_t purge_calls;
-  mi_stat_counter_t page_no_retire;
-  mi_stat_counter_t searches;
-  mi_stat_counter_t normal_count;
-  mi_stat_counter_t huge_count;
-  mi_stat_counter_t arena_count;
-  mi_stat_counter_t arena_crossover_count;
-  mi_stat_counter_t arena_rollback_count;
-  mi_stat_counter_t guarded_alloc_count;
-#if MI_STAT>1
-  mi_stat_count_t normal_bins[MI_BIN_HUGE+1];
-#endif
-} mi_stats_t;
-
-
 // add to stat keeping track of the peak
 void _mi_stat_increase(mi_stat_count_t* stat, size_t amount);
 void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount);
-// adjust stat in special cases to compensate for double counting
-void _mi_stat_adjust_increase(mi_stat_count_t* stat, size_t amount);
 void _mi_stat_adjust_decrease(mi_stat_count_t* stat, size_t amount);
 // counters can just be increased
 void _mi_stat_counter_increase(mi_stat_counter_t* stat, size_t amount);
@@ -621,72 +690,18 @@ void _mi_stat_counter_increase(mi_stat_counter_t* stat, size_t amount);
 #if (MI_STAT)
 #define mi_stat_increase(stat,amount)         _mi_stat_increase( &(stat), amount)
 #define mi_stat_decrease(stat,amount)         _mi_stat_decrease( &(stat), amount)
-#define mi_stat_counter_increase(stat,amount) _mi_stat_counter_increase( &(stat), amount)
-#define mi_stat_adjust_increase(stat,amount)  _mi_stat_adjust_increase( &(stat), amount)
 #define mi_stat_adjust_decrease(stat,amount)  _mi_stat_adjust_decrease( &(stat), amount)
+#define mi_stat_counter_increase(stat,amount) _mi_stat_counter_increase( &(stat), amount)
 #else
 #define mi_stat_increase(stat,amount)         ((void)0)
 #define mi_stat_decrease(stat,amount)         ((void)0)
-#define mi_stat_counter_increase(stat,amount) ((void)0)
-#define mi_stat_adjuct_increase(stat,amount)  ((void)0)
 #define mi_stat_adjust_decrease(stat,amount)  ((void)0)
+#define mi_stat_counter_increase(stat,amount) ((void)0)
 #endif
 
 #define mi_heap_stat_counter_increase(heap,stat,amount)  mi_stat_counter_increase( (heap)->tld->stats.stat, amount)
 #define mi_heap_stat_increase(heap,stat,amount)  mi_stat_increase( (heap)->tld->stats.stat, amount)
 #define mi_heap_stat_decrease(heap,stat,amount)  mi_stat_decrease( (heap)->tld->stats.stat, amount)
-
-
-// ------------------------------------------------------
-// Sub processes do not reclaim or visit segments
-// from other sub processes
-// ------------------------------------------------------
-
-struct mi_subproc_s {
-  _Atomic(size_t)    abandoned_count;         // count of abandoned segments for this sub-process
-  _Atomic(size_t)    abandoned_os_list_count; // count of abandoned segments in the os-list
-  mi_lock_t          abandoned_os_lock;       // lock for the abandoned os segment list (outside of arena's) (this lock protect list operations)
-  mi_lock_t          abandoned_os_visit_lock; // ensure only one thread per subproc visits the abandoned os list
-  mi_segment_t*      abandoned_os_list;       // doubly-linked list of abandoned segments outside of arena's (in OS allocated memory)
-  mi_segment_t*      abandoned_os_list_tail;  // the tail-end of the list
-  mi_memid_t         memid;                   // provenance of this memory block
-};
-
-// ------------------------------------------------------
-// Thread Local data
-// ------------------------------------------------------
-
-// Milliseconds as in `int64_t` to avoid overflows
-typedef int64_t  mi_msecs_t;
-
-// Queue of segments
-typedef struct mi_segment_queue_s {
-  mi_segment_t* first;
-  mi_segment_t* last;
-} mi_segment_queue_t;
-
-// Segments thread local data
-typedef struct mi_segments_tld_s {
-  mi_segment_queue_t  small_free;   // queue of segments with free small pages
-  mi_segment_queue_t  medium_free;  // queue of segments with free medium pages
-  mi_page_queue_t     pages_purge;  // queue of freed pages that are delay purged
-  size_t              count;        // current number of segments;
-  size_t              peak_count;   // peak number of segments
-  size_t              current_size; // current size of all segments
-  size_t              peak_size;    // peak size of all segments
-  size_t              reclaim_count;// number of reclaimed (abandoned) segments
-  mi_subproc_t*       subproc;      // sub-process this thread belongs to.
-  mi_stats_t*         stats;        // points to tld stats
-} mi_segments_tld_t;
-
-// Thread local data
-struct mi_tld_s {
-  unsigned long long  heartbeat;     // monotonic heartbeat count
-  bool                recurse;       // true if deferred was called; used to prevent infinite recursion.
-  mi_heap_t*          heap_backing;  // backing heap of this thread (cannot be deleted)
-  mi_heap_t*          heaps;         // list of heaps in this thread (so we can abandon all when the thread terminates)
-  mi_segments_tld_t   segments;      // segment tld
-  mi_stats_t          stats;         // statistics
-};
+#define mi_heap_stat_adjust_decrease(heap,stat,amount)  mi_stat_adjust_decrease( (heap)->tld->stats.stat, amount)
 
 #endif

mimalloc.pc.in

@@ -1,11 +1,11 @@
 prefix=@CMAKE_INSTALL_PREFIX@
-libdir=@libdir_for_pc_file@
-includedir=@includedir_for_pc_file@
+libdir=@mi_pc_libdir@
+includedir=@mi_pc_includedir@
 
 Name: @PROJECT_NAME@
 Description: A compact general purpose allocator with excellent performance
 Version: @PACKAGE_VERSION@
 URL: https://github.com/microsoft/mimalloc/
-Libs: -L${libdir} -lmimalloc
-Libs.private: @pc_libraries@
+Libs: -L${libdir} -l@mi_libname@
+Libs.private: @mi_pc_libraries@
 Cflags: -I${includedir}

readme.md

@@ -12,15 +12,17 @@ is a general purpose allocator with excellent [performance](#performance) charac
 Initially developed by Daan Leijen for the runtime systems of the
 [Koka](https://koka-lang.github.io) and [Lean](https://github.com/leanprover/lean) languages.
 
-Latest release tag: `v2.1.9` (2025-01-03).  
-Latest v1 tag: `v1.8.9` (2024-01-03).
+Latest release   : `v3.0.3` (beta) (2025-03-28).  
+Latest v2 release: `v2.2.3` (2025-03-28).  
+Latest v1 release: `v1.9.3` (2024-03-28).
 
 mimalloc is a drop-in replacement for `malloc` and can be used in other programs
 without code changes, for example, on dynamically linked ELF-based systems (Linux, BSD, etc.) you can use it as:
 ```
 > LD_PRELOAD=/usr/lib/libmimalloc.so  myprogram
 ```
-It also includes a robust way to override the default allocator in [Windows](#override_on_windows). Notable aspects of the design include:
+It also includes a way to dynamically override the default allocator in [Windows](#override_on_windows). 
+Notable aspects of the design include:
 
 - __small and consistent__: the library is about 10k LOC using simple and
   consistent data structures. This makes it very suitable
@@ -70,22 +72,31 @@ Enjoy!
 
 ### Branches
 
-* `master`: latest stable release (based on `dev2`).
-* `dev`: development branch for mimalloc v1. Use this branch for submitting PR's.
+* `master`: latest stable release (still based on `dev2`).
+* `dev`:  development branch for mimalloc v1. **Use this branch for submitting PR's**.
 * `dev2`: development branch for mimalloc v2. This branch is downstream of `dev` 
-         (and is essentially equal to `dev` except for `src/segment.c`). Uses larger sliced segments to manage
-         mimalloc pages what can reduce fragmentation.
-* `dev3`: development branch for mimalloc v3-alpha. This branch is downstream of `dev`. This is still experimental,
-          but simplifies previous versions by having no segments any more. This improves sharing of memory 
-          between threads, and on certain large workloads uses less memory with less fragmentation.
+          (and is essentially equal to `dev` except for `src/segment.c`). Uses larger sliced segments to manage
+          mimalloc pages that can reduce fragmentation.
+* `dev3`: development branch for mimalloc v3-beta. This branch is downstream of `dev`. This version 
+          simplifies the lock-free ownership of previous versions, has no thread-local segments any more. 
+          This improves sharing of memory between threads, and on certain large workloads may use less memory 
+          with less fragmentation.
 
 ### Releases
 
-* 2025-01-03, `v1.8.9`, `v2.1.9`, `v3.0-alpha`: Interim release. Support Windows arm64. New [guarded](#guarded) build that can place OS 
+* 2025-03-28, `v1.9.3`, `v2.2.3`, `v3.0.3` (beta): Various small bug and build fixes, including:
+  fix arm32 pre v7 builds, fix mingw build, get runtime statistics, improve statistic commit counts, 
+  fix execution on non BMI1 x64 systems. 
+* 2025-03-06, `v1.9.2`, `v2.2.2`, `v3.0.2-beta`: Various small bug and build fixes. 
+  Add `mi_options_print`, `mi_arenas_print`, and the experimental `mi_stat_get` and `mi_stat_get_json`. 
+  Add `mi_thread_set_in_threadpool` and `mi_heap_set_numa_affinity` (v3 only). Add vcpkg portfile. 
+  Upgrade mimalloc-redirect to v1.3.2. `MI_OPT_ARCH` is off by default now but still assumes armv8.1-a on arm64
+  for fast atomic operations. Add QNX support.
+* 2025-01-03, `v1.8.9`, `v2.1.9`, `v3.0.1-alpha`: Interim release. Support Windows arm64. New [guarded](#guarded) build that can place OS 
   guard pages behind objects to catch buffer overflows as they occur. 
   Many small fixes: build on Windows arm64, cygwin, riscV, and dragonfly; fix Windows static library initialization to account for
   thread local destructors (in Rust/C++); macOS tag change; macOS TLS slot fix; improve stats; 
-  consistent mimalloc.dll on Windows (instead of mimalloc-override.dll); fix mimalloc-redirect on Win11 H2; 
+  consistent `mimalloc.dll` on Windows (instead of `mimalloc-override.dll`); fix mimalloc-redirect on Win11 H2; 
   add 0-byte to canary; upstream CPython fixes; reduce .bss size; allow fixed TLS slot on Windows for improved performance.
 * 2024-05-21, `v1.8.7`, `v2.1.7`: Fix build issues on less common platforms. Started upstreaming patches
   from the CPython [integration](https://github.com/python/cpython/issues/113141#issuecomment-2119255217). Upstream `vcpkg` patches.
@@ -165,8 +176,8 @@ mimalloc is used in various large scale low-latency services and programs, for e
 ## Windows
 
 Open `ide/vs2022/mimalloc.sln` in Visual Studio 2022 and build.
-The `mimalloc` project builds a static library (in `out/msvc-x64`), while the
-`mimalloc-override` project builds a DLL for overriding malloc
+The `mimalloc-lib` project builds a static library (in `out/msvc-x64`), while the
+`mimalloc-override-dll` project builds DLL for overriding malloc
 in the entire program.
 
 ## Linux, macOS, BSD, etc.
@@ -231,13 +242,21 @@ The cmake build type is specified when actually building, for example:
 > cmake --build . --config=Release
 ```
 
-## Single source
+You can also install the [LLVM toolset](https://learn.microsoft.com/en-us/cpp/build/clang-support-msbuild?view=msvc-170#install-1) 
+on Windows to build with the `clang-cl` compiler directly:
+
+```
+> cmake ../.. -G "Visual Studio 17 2022" -T ClangCl
+```
+
+
+## Single Source
 
 You can also directly build the single `src/static.c` file as part of your project without
 needing `cmake` at all. Make sure to also add the mimalloc `include` directory to the include path.
 
 
-# Using the library
+# Using the Library
 
 The preferred usage is including `<mimalloc.h>`, linking with
 the shared- or static library, and using the `mi_malloc` API exclusively for allocation. For example,
@@ -465,18 +484,17 @@ Note that certain security restrictions may apply when doing this from
 the [shell](https://stackoverflow.com/questions/43941322/dyld-insert-libraries-ignored-when-calling-application-through-bash).
 
 
-# Windows Override
+### Dynamic Override on Windows
 
-<span id="override_on_windows">Dynamically overriding on mimalloc on Windows</span> 
-is robust and has the particular advantage to be able to redirect all malloc/free calls 
-that go through the (dynamic) C runtime allocator, including those from other DLL's or 
-libraries. As it intercepts all allocation calls on a low level, it can be used reliably 
-on large programs that include other 3rd party components.
+<span id="override_on_windows">We use a separate redirection DLL to override mimalloc on Windows</span> 
+such that we redirect all malloc/free calls that go through the (dynamic) C runtime allocator, 
+including those from other DLL's or libraries. As it intercepts all allocation calls on a low level, 
+it can be used on large programs that include other 3rd party components.
 There are four requirements to make the overriding work well:
 
 1. Use the C-runtime library as a DLL (using the `/MD` or `/MDd` switch).
 
-2. Link your program explicitly with the `mimalloc.lib` export library for the `mimalloc.dll`.
+2. Link your program explicitly with the `mimalloc.dll.lib` export library for the `mimalloc.dll`.
    (which must be compiled with `-DMI_OVERRIDE=ON`, which is the default though).
    To ensure the `mimalloc.dll` is actually loaded at run-time it is easiest 
    to insert some call to the mimalloc API in the `main` function, like `mi_version()`
@@ -493,9 +511,8 @@ There are four requirements to make the overriding work well:
    list of the final executable (so it can intercept all potential allocations).
    You can use `minject -l <exe>` to check this if needed.
 
-For best performance on Windows with C++, it
-is also recommended to also override the `new`/`delete` operations (by including
-[`mimalloc-new-delete.h`](include/mimalloc-new-delete.h)
+For best performance on Windows with C++, it is also recommended to also override 
+the `new`/`delete` operations (by including [`mimalloc-new-delete.h`](include/mimalloc-new-delete.h)
 a single(!) source file in your project).
 
 The environment variable `MIMALLOC_DISABLE_REDIRECT=1` can be used to disable dynamic

src/alloc-aligned.c

@@ -115,7 +115,7 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_overalloc(mi_heap_t
 
   // now zero the block if needed
   if (alignment > MI_BLOCK_ALIGNMENT_MAX) {
-    // for the tracker, on huge aligned allocations only from the start of the large block is defined
+    // for the tracker, on huge aligned allocations only the memory from the start of the large block is defined
     mi_track_mem_undefined(aligned_p, size);
     if (zero) {
       _mi_memzero_aligned(aligned_p, mi_usable_size(aligned_p));
@@ -191,9 +191,6 @@ static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t
       const bool is_aligned = (((uintptr_t)page->free + offset) & align_mask)==0;
       if mi_likely(is_aligned)
       {
-        #if MI_STAT>1
-        mi_heap_stat_increase(heap, malloc, size);
-        #endif
         void* p = (zero ? _mi_page_malloc_zeroed(heap,page,padsize) : _mi_page_malloc(heap,page,padsize)); // call specific page malloc for better codegen
         mi_assert_internal(p != NULL);
         mi_assert_internal(((uintptr_t)p + offset) % alignment == 0);
@@ -220,6 +217,11 @@ mi_decl_nodiscard mi_decl_restrict void* mi_heap_malloc_aligned(mi_heap_t* heap,
   return mi_heap_malloc_aligned_at(heap, size, alignment, 0);
 }
 
+// ensure a definition is emitted
+#if defined(__cplusplus)
+void* _mi_extern_heap_malloc_aligned = (void*)&mi_heap_malloc_aligned;
+#endif
+
 // ------------------------------------------------------
 // Aligned Allocation
 // ------------------------------------------------------

src/alloc.c

@@ -30,6 +30,7 @@ terms of the MIT license. A copy of the license can be found in the file
 // Note: in release mode the (inlined) routine is about 7 instructions with a single test.
 extern inline void* _mi_page_malloc_zero(mi_heap_t* heap, mi_page_t* page, size_t size, bool zero) mi_attr_noexcept
 {
+  mi_assert_internal(size >= MI_PADDING_SIZE);
   mi_assert_internal(page->block_size == 0 /* empty heap */ || mi_page_block_size(page) >= size);
 
   // check the free list
@@ -82,12 +83,13 @@ extern inline void* _mi_page_malloc_zero(mi_heap_t* heap, mi_page_t* page, size_
 
   #if (MI_STAT>0)
   const size_t bsize = mi_page_usable_block_size(page);
-  if (bsize <= MI_LARGE_OBJ_SIZE_MAX) {
-    mi_heap_stat_increase(heap, normal, bsize);
-    mi_heap_stat_counter_increase(heap, normal_count, 1);
+  if (bsize <= MI_MEDIUM_OBJ_SIZE_MAX) {
+    mi_heap_stat_increase(heap, malloc_normal, bsize);
+    mi_heap_stat_counter_increase(heap, malloc_normal_count, 1);
     #if (MI_STAT>1)
     const size_t bin = _mi_bin(bsize);
-    mi_heap_stat_increase(heap, normal_bins[bin], 1);
+    mi_heap_stat_increase(heap, malloc_bins[bin], 1);
+    mi_heap_stat_increase(heap, malloc_requested, size - MI_PADDING_SIZE);
     #endif
   }
   #endif
@@ -146,12 +148,6 @@ static inline mi_decl_restrict void* mi_heap_malloc_small_zero(mi_heap_t* heap,
   void* const p = _mi_page_malloc_zero(heap, page, size + MI_PADDING_SIZE, zero);
   mi_track_malloc(p,size,zero);
 
-  #if MI_STAT>1
-  if (p != NULL) {
-    if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
-    mi_heap_stat_increase(heap, malloc, mi_usable_size(p));
-  }
-  #endif
   #if MI_DEBUG>3
   if (p != NULL && zero) {
     mi_assert_expensive(mi_mem_is_zero(p, size));
@@ -188,12 +184,6 @@ extern inline void* _mi_heap_malloc_zero_ex(mi_heap_t* heap, size_t size, bool z
     void* const p = _mi_malloc_generic(heap, size + MI_PADDING_SIZE, zero, huge_alignment);  // note: size can overflow but it is detected in malloc_generic
     mi_track_malloc(p,size,zero);
 
-    #if MI_STAT>1
-    if (p != NULL) {
-      if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
-      mi_heap_stat_increase(heap, malloc, mi_usable_size(p));
-    }
-    #endif
     #if MI_DEBUG>3
     if (p != NULL && zero) {
       mi_assert_expensive(mi_mem_is_zero(p, size));
@@ -640,7 +630,7 @@ static void* mi_block_ptr_set_guarded(mi_block_t* block, size_t obj_size) {
     // give up to place it right in front of the guard page if the offset is too large for unalignment
     offset = MI_BLOCK_ALIGNMENT_MAX;
   }
-  void* p = (uint8_t*)block + offset;  
+  void* p = (uint8_t*)block + offset;
   mi_track_align(block, p, offset, obj_size);
   mi_track_mem_defined(block, sizeof(mi_block_t));
   return p;
@@ -662,13 +652,14 @@ mi_decl_restrict void* _mi_heap_malloc_guarded(mi_heap_t* heap, size_t size, boo
   void* const p   = mi_block_ptr_set_guarded(block, obj_size);
 
   // stats
-  mi_track_malloc(p, size, zero);  
+  mi_track_malloc(p, size, zero);
   if (p != NULL) {
     if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
     #if MI_STAT>1
-    mi_heap_stat_increase(heap, malloc, mi_usable_size(p));
+    mi_heap_stat_adjust_decrease(heap, malloc_requested, req_size);
+    mi_heap_stat_increase(heap, malloc_requested, size);
     #endif
-    _mi_stat_counter_increase(&heap->tld->stats.guarded_alloc_count, 1);
+    _mi_stat_counter_increase(&heap->tld->stats.malloc_guarded_count, 1);
   }
   #if MI_DEBUG>3
   if (p != NULL && zero) {
@@ -694,7 +685,7 @@ void* _mi_externs[] = {
   (void*)&mi_zalloc_small,
   (void*)&mi_heap_malloc,
   (void*)&mi_heap_zalloc,
-  (void*)&mi_heap_malloc_small
+  (void*)&mi_heap_malloc_small,
   // (void*)&mi_heap_alloc_new,
   // (void*)&mi_heap_alloc_new_n
 };

src/arena.c

@@ -99,6 +99,10 @@ bool _mi_arena_memid_is_suitable(mi_memid_t memid, mi_arena_id_t request_arena_i
   }
 }
 
+bool _mi_arena_memid_is_os_allocated(mi_memid_t memid) {
+  return (memid.memkind == MI_MEM_OS);
+}
+
 size_t mi_arena_get_count(void) {
   return mi_atomic_load_relaxed(&mi_arena_count);
 }
@@ -255,7 +259,7 @@ static mi_decl_noinline void* mi_arena_try_alloc_at(mi_arena_t* arena, size_t ar
 
   // set the dirty bits (todo: no need for an atomic op here?)
   if (arena->memid.initially_zero && arena->blocks_dirty != NULL) {
-    memid->initially_zero = _mi_bitmap_claim_across(arena->blocks_dirty, arena->field_count, needed_bcount, bitmap_index, NULL);
+    memid->initially_zero = _mi_bitmap_claim_across(arena->blocks_dirty, arena->field_count, needed_bcount, bitmap_index, NULL, NULL);
   }
 
   // set commit state
@@ -267,10 +271,14 @@ static mi_decl_noinline void* mi_arena_try_alloc_at(mi_arena_t* arena, size_t ar
     // commit requested, but the range may not be committed as a whole: ensure it is committed now
     memid->initially_committed = true;
     bool any_uncommitted;
-    _mi_bitmap_claim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &any_uncommitted);
+    size_t already_committed = 0;
+    _mi_bitmap_claim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &any_uncommitted, &already_committed);
     if (any_uncommitted) {
+      mi_assert_internal(already_committed < needed_bcount);
+      const size_t commit_size = mi_arena_block_size(needed_bcount);
+      const size_t stat_commit_size = commit_size - mi_arena_block_size(already_committed);
       bool commit_zero = false;
-      if (!_mi_os_commit(p, mi_arena_block_size(needed_bcount), &commit_zero)) {
+      if (!_mi_os_commit_ex(p, commit_size, &commit_zero, stat_commit_size)) {
         memid->initially_committed = false;
       }
       else {
@@ -280,7 +288,14 @@ static mi_decl_noinline void* mi_arena_try_alloc_at(mi_arena_t* arena, size_t ar
   }
   else {
     // no need to commit, but check if already fully committed
-    memid->initially_committed = _mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index);
+    size_t already_committed = 0;
+    memid->initially_committed = _mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &already_committed);
+    if (!memid->initially_committed && already_committed > 0) {
+      // partially committed: as it will be committed at some time, adjust the stats and pretend the range is fully uncommitted.
+      mi_assert_internal(already_committed < needed_bcount);
+      _mi_stat_decrease(&_mi_stats_main.committed, mi_arena_block_size(already_committed));
+      _mi_bitmap_unclaim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index);
+    }
   }
 
   return p;
@@ -464,17 +479,19 @@ static void mi_arena_purge(mi_arena_t* arena, size_t bitmap_idx, size_t blocks)
   const size_t size = mi_arena_block_size(blocks);
   void* const p = mi_arena_block_start(arena, bitmap_idx);
   bool needs_recommit;
-  if (_mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx)) {
+  size_t already_committed = 0;
+  if (_mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx, &already_committed)) {
     // all blocks are committed, we can purge freely
+    mi_assert_internal(already_committed == blocks);
     needs_recommit = _mi_os_purge(p, size);
   }
   else {
     // some blocks are not committed -- this can happen when a partially committed block is freed
     // in `_mi_arena_free` and it is conservatively marked as uncommitted but still scheduled for a purge
-    // we need to ensure we do not try to reset (as that may be invalid for uncommitted memory),
-    // and also undo the decommit stats (as it was already adjusted)
+    // we need to ensure we do not try to reset (as that may be invalid for uncommitted memory).
+    mi_assert_internal(already_committed < blocks);
     mi_assert_internal(mi_option_is_enabled(mi_option_purge_decommits));
-    needs_recommit = _mi_os_purge_ex(p, size, false /* allow reset? */, 0);    
+    needs_recommit = _mi_os_purge_ex(p, size, false /* allow reset? */, mi_arena_block_size(already_committed));    
   }
 
   // clear the purged blocks
@@ -508,7 +525,7 @@ static void mi_arena_schedule_purge(mi_arena_t* arena, size_t bitmap_idx, size_t
     else {
       // already an expiration was set
     }
-    _mi_bitmap_claim_across(arena->blocks_purge, arena->field_count, blocks, bitmap_idx, NULL);
+    _mi_bitmap_claim_across(arena->blocks_purge, arena->field_count, blocks, bitmap_idx, NULL, NULL);
   }
 }
 
@@ -550,7 +567,8 @@ static bool mi_arena_try_purge(mi_arena_t* arena, mi_msecs_t now, bool force)
 
   // reset expire (if not already set concurrently)
   mi_atomic_casi64_strong_acq_rel(&arena->purge_expire, &expire, (mi_msecs_t)0);
-  
+  _mi_stat_counter_increase(&_mi_stats_main.arena_purges, 1);
+
   // potential purges scheduled, walk through the bitmap
   bool any_purged = false;
   bool full_purge = true;
@@ -603,7 +621,7 @@ static void mi_arenas_try_purge( bool force, bool visit_all )
 
   // check if any arena needs purging?
   const mi_msecs_t now = _mi_clock_now();
-  mi_msecs_t arenas_expire = mi_atomic_load_acquire(&mi_arenas_purge_expire);
+  mi_msecs_t arenas_expire = mi_atomic_loadi64_acquire(&mi_arenas_purge_expire);
   if (!force && (arenas_expire == 0 || arenas_expire < now)) return;
 
   const size_t max_arena = mi_atomic_load_acquire(&mi_arena_count);
@@ -614,7 +632,7 @@ static void mi_arenas_try_purge( bool force, bool visit_all )
   mi_atomic_guard(&purge_guard)
   {
     // increase global expire: at most one purge per delay cycle
-    mi_atomic_store_release(&mi_arenas_purge_expire, now + mi_arena_purge_delay());  
+    mi_atomic_storei64_release(&mi_arenas_purge_expire, now + mi_arena_purge_delay());  
     size_t max_purge_count = (visit_all ? max_arena : 2);
     bool all_visited = true;
     for (size_t i = 0; i < max_arena; i++) {
@@ -631,7 +649,7 @@ static void mi_arenas_try_purge( bool force, bool visit_all )
     }
     if (all_visited) {
       // all arena's were visited and purged: reset global expire
-      mi_atomic_store_release(&mi_arenas_purge_expire, 0);
+      mi_atomic_storei64_release(&mi_arenas_purge_expire, 0);
     }
   }
 }
@@ -647,15 +665,16 @@ void _mi_arena_free(void* p, size_t size, size_t committed_size, mi_memid_t memi
   if (p==NULL) return;
   if (size==0) return;
   const bool all_committed = (committed_size == size);
+  const size_t decommitted_size = (committed_size <= size ? size - committed_size : 0);
 
   // need to set all memory to undefined as some parts may still be marked as no_access (like padding etc.)
   mi_track_mem_undefined(p,size);
 
   if (mi_memkind_is_os(memid.memkind)) {
     // was a direct OS allocation, pass through
-    if (!all_committed && committed_size > 0) {
-      // if partially committed, adjust the committed stats (as `_mi_os_free` will increase decommit by the full size)
-      _mi_stat_decrease(&_mi_stats_main.committed, committed_size);
+    if (!all_committed && decommitted_size > 0) {
+      // if partially committed, adjust the committed stats (as `_mi_os_free` will decrease commit by the full size)
+      _mi_stat_increase(&_mi_stats_main.committed, decommitted_size);
     }
     _mi_os_free(p, size, memid);
   }
@@ -689,14 +708,14 @@ void _mi_arena_free(void* p, size_t size, size_t committed_size, mi_memid_t memi
       mi_assert_internal(arena->blocks_purge != NULL);
 
       if (!all_committed) {
-        // mark the entire range as no longer committed (so we recommit the full range when re-using)
+        // mark the entire range as no longer committed (so we will recommit the full range when re-using)
         _mi_bitmap_unclaim_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx);
         mi_track_mem_noaccess(p,size);
-        if (committed_size > 0) {
+        //if (committed_size > 0) {
           // if partially committed, adjust the committed stats (is it will be recommitted when re-using)
-          // in the delayed purge, we now need to not count a decommit if the range is not marked as committed.
+          // in the delayed purge, we do no longer decrease the commit if the range is not marked entirely as committed.
           _mi_stat_decrease(&_mi_stats_main.committed, committed_size);
-        }
+        //}
         // note: if not all committed, it may be that the purge will reset/decommit the entire range
         // that contains already decommitted parts. Since purge consistently uses reset or decommit that
         // works (as we should never reset decommitted parts).
@@ -900,7 +919,7 @@ int mi_reserve_os_memory(size_t size, bool commit, bool allow_large) mi_attr_noe
 ----------------------------------------------------------- */
 
 static size_t mi_debug_show_bitmap(const char* prefix, const char* header, size_t block_count, mi_bitmap_field_t* fields, size_t field_count ) {
-  _mi_verbose_message("%s%s:\n", prefix, header);
+  _mi_message("%s%s:\n", prefix, header);
   size_t bcount = 0;
   size_t inuse_count = 0;
   for (size_t i = 0; i < field_count; i++) {
@@ -917,13 +936,14 @@ static size_t mi_debug_show_bitmap(const char* prefix, const char* header, size_
       }
     }
     buf[MI_BITMAP_FIELD_BITS] = 0;
-    _mi_verbose_message("%s  %s\n", prefix, buf);
+    _mi_message("%s  %s\n", prefix, buf);
   }
-  _mi_verbose_message("%s  total ('x'): %zu\n", prefix, inuse_count);
+  _mi_message("%s  total ('x'): %zu\n", prefix, inuse_count);
   return inuse_count;
 }
 
-void mi_debug_show_arenas(bool show_inuse) mi_attr_noexcept {
+void mi_debug_show_arenas(void) mi_attr_noexcept {
+  const bool show_inuse = true;
   size_t max_arenas = mi_atomic_load_relaxed(&mi_arena_count);
   size_t inuse_total = 0;
   //size_t abandoned_total = 0;
@@ -931,7 +951,7 @@ void mi_debug_show_arenas(bool show_inuse) mi_attr_noexcept {
   for (size_t i = 0; i < max_arenas; i++) {
     mi_arena_t* arena = mi_atomic_load_ptr_relaxed(mi_arena_t, &mi_arenas[i]);
     if (arena == NULL) break;
-    _mi_verbose_message("arena %zu: %zu blocks of size %zuMiB (in %zu fields) %s\n", i, arena->block_count, MI_ARENA_BLOCK_SIZE / MI_MiB, arena->field_count, (arena->memid.is_pinned ? ", pinned" : ""));
+    _mi_message("arena %zu: %zu blocks of size %zuMiB (in %zu fields) %s\n", i, arena->block_count, MI_ARENA_BLOCK_SIZE / MI_MiB, arena->field_count, (arena->memid.is_pinned ? ", pinned" : ""));
     if (show_inuse) {
       inuse_total += mi_debug_show_bitmap("  ", "inuse blocks", arena->block_count, arena->blocks_inuse, arena->field_count);
     }
@@ -945,9 +965,14 @@ void mi_debug_show_arenas(bool show_inuse) mi_attr_noexcept {
     //  purge_total += mi_debug_show_bitmap("  ", "purgeable blocks", arena->block_count, arena->blocks_purge, arena->field_count);
     //}
   }
-  if (show_inuse)     _mi_verbose_message("total inuse blocks    : %zu\n", inuse_total);
-  //if (show_abandoned) _mi_verbose_message("total abandoned blocks: %zu\n", abandoned_total);
-  //if (show_purge)     _mi_verbose_message("total purgeable blocks: %zu\n", purge_total);
+  if (show_inuse)     _mi_message("total inuse blocks    : %zu\n", inuse_total);
+  //if (show_abandoned) _mi_message("total abandoned blocks: %zu\n", abandoned_total);
+  //if (show_purge)     _mi_message("total purgeable blocks: %zu\n", purge_total);
+}
+
+
+void mi_arenas_print(void) mi_attr_noexcept {
+  mi_debug_show_arenas();
 }
 
 

src/bitmap.c

@@ -34,17 +34,17 @@ static inline size_t mi_bitmap_mask_(size_t count, size_t bitidx) {
 }
 
 
-
 /* -----------------------------------------------------------
   Claim a bit sequence atomically
 ----------------------------------------------------------- */
 
 // Try to atomically claim a sequence of `count` bits in a single
 // field at `idx` in `bitmap`. Returns `true` on success.
-bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_t count, mi_bitmap_index_t* bitmap_idx)
+inline bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_t count, mi_bitmap_index_t* bitmap_idx)
 {
   mi_assert_internal(bitmap_idx != NULL);
   mi_assert_internal(count <= MI_BITMAP_FIELD_BITS);
+  mi_assert_internal(count > 0);
   mi_bitmap_field_t* field = &bitmap[idx];
   size_t map  = mi_atomic_load_relaxed(field);
   if (map==MI_BITMAP_FIELD_FULL) return false; // short cut
@@ -81,7 +81,7 @@ bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_
       // on to the next bit range
 #ifdef MI_HAVE_FAST_BITSCAN
       mi_assert_internal(mapm != 0);
-      const size_t shift = (count == 1 ? 1 : (MI_INTPTR_BITS - mi_clz(mapm) - bitidx));
+      const size_t shift = (count == 1 ? 1 : (MI_SIZE_BITS - mi_clz(mapm) - bitidx));
       mi_assert_internal(shift > 0 && shift <= count);
 #else
       const size_t shift = 1;
@@ -94,9 +94,9 @@ bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_
   return false;
 }
 
-
+// Find `count` bits of 0 and set them to 1 atomically; returns `true` on success.
 // Starts at idx, and wraps around to search in all `bitmap_fields` fields.
-// For now, `count` can be at most MI_BITMAP_FIELD_BITS and will never cross fields.
+// `count` can be at most MI_BITMAP_FIELD_BITS and will never cross fields.
 bool _mi_bitmap_try_find_from_claim(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx) {
   size_t idx = start_field_idx;
   for (size_t visited = 0; visited < bitmap_fields; visited++, idx++) {
@@ -108,6 +108,24 @@ bool _mi_bitmap_try_find_from_claim(mi_bitmap_t bitmap, const size_t bitmap_fiel
   return false;
 }
 
+// Like _mi_bitmap_try_find_from_claim but with an extra predicate that must be fullfilled
+bool _mi_bitmap_try_find_from_claim_pred(mi_bitmap_t bitmap, const size_t bitmap_fields, 
+            const size_t start_field_idx, const size_t count, 
+            mi_bitmap_pred_fun_t pred_fun, void* pred_arg,            
+            mi_bitmap_index_t* bitmap_idx) {
+  size_t idx = start_field_idx;
+  for (size_t visited = 0; visited < bitmap_fields; visited++, idx++) {
+    if (idx >= bitmap_fields) idx = 0; // wrap
+    if (_mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx)) {
+      if (pred_fun == NULL || pred_fun(*bitmap_idx, pred_arg)) { 
+        return true;
+      }
+      // predicate returned false, unclaim and look further
+      _mi_bitmap_unclaim(bitmap, bitmap_fields, count, *bitmap_idx);
+    }
+  }
+  return false;
+}
 
 // Set `count` bits at `bitmap_idx` to 0 atomically
 // Returns `true` if all `count` bits were 1 previously.
@@ -146,7 +164,7 @@ static bool mi_bitmap_is_claimedx(mi_bitmap_t bitmap, size_t bitmap_fields, size
   return ((field & mask) == mask);
 }
 
-// Try to set `count` bits at `bitmap_idx` from 0 to 1 atomically. 
+// Try to set `count` bits at `bitmap_idx` from 0 to 1 atomically.
 // Returns `true` if successful when all previous `count` bits were 0.
 bool _mi_bitmap_try_claim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) {
   const size_t idx = mi_bitmap_index_field(bitmap_idx);
@@ -154,9 +172,9 @@ bool _mi_bitmap_try_claim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count
   const size_t mask = mi_bitmap_mask_(count, bitidx);
   mi_assert_internal(bitmap_fields > idx); MI_UNUSED(bitmap_fields);
   size_t expected = mi_atomic_load_relaxed(&bitmap[idx]);
-  do  {    
+  do  {
     if ((expected & mask) != 0) return false;
-  } 
+  }
   while (!mi_atomic_cas_strong_acq_rel(&bitmap[idx], &expected, expected | mask));
   mi_assert_internal((expected & mask) == 0);
   return true;
@@ -194,7 +212,7 @@ static bool mi_bitmap_try_find_claim_field_across(mi_bitmap_t bitmap, size_t bit
   if (initial == 0)     return false;
   if (initial >= count) return _mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx);    // no need to cross fields (this case won't happen for us)
   if (_mi_divide_up(count - initial, MI_BITMAP_FIELD_BITS) >= (bitmap_fields - idx)) return false; // not enough entries
-  
+
   // scan ahead
   size_t found = initial;
   size_t mask = 0;     // mask bits for the final field
@@ -228,7 +246,7 @@ static bool mi_bitmap_try_find_claim_field_across(mi_bitmap_t bitmap, size_t bit
 
   // intermediate fields
   while (++field < final_field) {
-    newmap = mi_bitmap_mask_(MI_BITMAP_FIELD_BITS, 0);
+    newmap = MI_BITMAP_FIELD_FULL;
     map = 0;
     if (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)) { goto rollback; }
   }
@@ -242,7 +260,6 @@ static bool mi_bitmap_try_find_claim_field_across(mi_bitmap_t bitmap, size_t bit
   } while (!mi_atomic_cas_strong_acq_rel(field, &map, newmap));
 
   // claimed!
-  mi_stat_counter_increase(_mi_stats_main.arena_crossover_count,1);
   *bitmap_idx = mi_bitmap_index_create(idx, initial_idx);
   return true;
 
@@ -251,7 +268,7 @@ rollback:
   // (we just failed to claim `field` so decrement first)
   while (--field > initial_field) {
     newmap = 0;
-    map = mi_bitmap_mask_(MI_BITMAP_FIELD_BITS, 0);
+    map = MI_BITMAP_FIELD_FULL;
     mi_assert_internal(mi_atomic_load_relaxed(field) == map);
     mi_atomic_store_release(field, newmap);
   }
@@ -352,7 +369,7 @@ bool _mi_bitmap_unclaim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t
 
 // Set `count` bits at `bitmap_idx` to 1 atomically
 // Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit.
-bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero) {
+bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero, size_t* already_set) {
   size_t idx = mi_bitmap_index_field(bitmap_idx);
   size_t pre_mask;
   size_t mid_mask;
@@ -360,28 +377,31 @@ bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t co
   size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask);
   bool all_zero = true;
   bool any_zero = false;
+  size_t one_count = 0;
   _Atomic(size_t)*field = &bitmap[idx];
   size_t prev = mi_atomic_or_acq_rel(field++, pre_mask);
-  if ((prev & pre_mask) != 0) all_zero = false;
+  if ((prev & pre_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & pre_mask); }
   if ((prev & pre_mask) != pre_mask) any_zero = true;
   while (mid_count-- > 0) {
     prev = mi_atomic_or_acq_rel(field++, mid_mask);
-    if ((prev & mid_mask) != 0) all_zero = false;
+    if ((prev & mid_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & mid_mask); }
     if ((prev & mid_mask) != mid_mask) any_zero = true;
   }
   if (post_mask!=0) {
     prev = mi_atomic_or_acq_rel(field, post_mask);
-    if ((prev & post_mask) != 0) all_zero = false;
+    if ((prev & post_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & post_mask); }
     if ((prev & post_mask) != post_mask) any_zero = true;
   }
   if (pany_zero != NULL) { *pany_zero = any_zero; }
+  if (already_set != NULL) { *already_set = one_count; };
+  mi_assert_internal(all_zero ? one_count == 0 : one_count <= count);
   return all_zero;
 }
 
 
 // Returns `true` if all `count` bits were 1.
 // `any_ones` is `true` if there was at least one bit set to one.
-static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_ones) {
+static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_ones, size_t* already_set) {
   size_t idx = mi_bitmap_index_field(bitmap_idx);
   size_t pre_mask;
   size_t mid_mask;
@@ -389,30 +409,33 @@ static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_field
   size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask);
   bool all_ones = true;
   bool any_ones = false;
+  size_t one_count = 0;
   mi_bitmap_field_t* field = &bitmap[idx];
   size_t prev = mi_atomic_load_relaxed(field++);
   if ((prev & pre_mask) != pre_mask) all_ones = false;
-  if ((prev & pre_mask) != 0) any_ones = true;
+  if ((prev & pre_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & pre_mask); }
   while (mid_count-- > 0) {
     prev = mi_atomic_load_relaxed(field++);
     if ((prev & mid_mask) != mid_mask) all_ones = false;
-    if ((prev & mid_mask) != 0) any_ones = true;
+    if ((prev & mid_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & mid_mask); }
   }
   if (post_mask!=0) {
     prev = mi_atomic_load_relaxed(field);
     if ((prev & post_mask) != post_mask) all_ones = false;
-    if ((prev & post_mask) != 0) any_ones = true;
+    if ((prev & post_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & post_mask); }
   }
   if (pany_ones != NULL) { *pany_ones = any_ones; }
+  if (already_set != NULL) { *already_set = one_count; }
+  mi_assert_internal(all_ones ? one_count == count : one_count < count);
   return all_ones;
 }
 
-bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) {
-  return mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, NULL);
+bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, size_t* already_set) {
+  return mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, NULL, already_set);
 }
 
 bool _mi_bitmap_is_any_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) {
   bool any_ones;
-  mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, &any_ones);
+  mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, &any_ones, NULL);
   return any_ones;
 }

src/bitmap.h

@@ -44,6 +44,11 @@ static inline mi_bitmap_index_t mi_bitmap_index_create(size_t idx, size_t bitidx
   return mi_bitmap_index_create_ex(idx,bitidx);
 }
 
+// Create a bit index.
+static inline mi_bitmap_index_t mi_bitmap_index_create_from_bit(size_t full_bitidx) {  
+  return mi_bitmap_index_create(full_bitidx / MI_BITMAP_FIELD_BITS, full_bitidx % MI_BITMAP_FIELD_BITS);
+}
+
 // Get the field index from a bit index.
 static inline size_t mi_bitmap_index_field(mi_bitmap_index_t bitmap_idx) {
   return (bitmap_idx / MI_BITMAP_FIELD_BITS);
@@ -71,6 +76,10 @@ bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_
 // For now, `count` can be at most MI_BITMAP_FIELD_BITS and will never cross fields.
 bool _mi_bitmap_try_find_from_claim(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx);
 
+// Like _mi_bitmap_try_find_from_claim but with an extra predicate that must be fullfilled
+typedef bool (mi_cdecl *mi_bitmap_pred_fun_t)(mi_bitmap_index_t bitmap_idx, void* pred_arg);
+bool _mi_bitmap_try_find_from_claim_pred(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_pred_fun_t pred_fun, void* pred_arg, mi_bitmap_index_t* bitmap_idx);
+
 // Set `count` bits at `bitmap_idx` to 0 atomically
 // Returns `true` if all `count` bits were 1 previously.
 bool _mi_bitmap_unclaim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx);
@@ -102,9 +111,9 @@ bool _mi_bitmap_unclaim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t
 
 // Set `count` bits at `bitmap_idx` to 1 atomically
 // Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit.
-bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero);
+bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero, size_t* already_set);
 
-bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx);
+bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, size_t* already_set);
 bool _mi_bitmap_is_any_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx);
 
 #endif

src/free.c

@@ -35,7 +35,9 @@ static inline void mi_free_block_local(mi_page_t* page, mi_block_t* block, bool
   mi_check_padding(page, block);
   if (track_stats) { mi_stat_free(page, block); }
   #if (MI_DEBUG>0) && !MI_TRACK_ENABLED  && !MI_TSAN && !MI_GUARDED
-  memset(block, MI_DEBUG_FREED, mi_page_block_size(page));
+  if (!mi_page_is_huge(page)) {   // huge page content may be already decommitted
+    memset(block, MI_DEBUG_FREED, mi_page_block_size(page));
+  }
   #endif
   if (track_stats) { mi_track_free_size(block, mi_page_usable_size_of(page, block)); } // faster then mi_usable_size as we already know the page and that p is unaligned
 
@@ -121,10 +123,16 @@ static inline mi_segment_t* mi_checked_ptr_segment(const void* p, const char* ms
 
   #if (MI_DEBUG>0)
   if mi_unlikely(!mi_is_in_heap_region(p)) {
-    _mi_warning_message("%s: pointer might not point to a valid heap region: %p\n"
-      "(this may still be a valid very large allocation (over 64MiB))\n", msg, p);
-    if mi_likely(_mi_ptr_cookie(segment) == segment->cookie) {
-      _mi_warning_message("(yes, the previous pointer %p was valid after all)\n", p);
+  #if (MI_INTPTR_SIZE == 8 && defined(__linux__))
+    if (((uintptr_t)p >> 40) != 0x7F) { // linux tends to align large blocks above 0x7F000000000 (issue #640)
+  #else
+    {
+  #endif
+      _mi_warning_message("%s: pointer might not point to a valid heap region: %p\n"
+        "(this may still be a valid very large allocation (over 64MiB))\n", msg, p);
+      if mi_likely(_mi_ptr_cookie(segment) == segment->cookie) {
+        _mi_warning_message("(yes, the previous pointer %p was valid after all)\n", p);
+      }
     }
   }
   #endif
@@ -272,7 +280,7 @@ static void mi_decl_noinline mi_free_block_mt(mi_page_t* page, mi_segment_t* seg
   // for small size, ensure we can fit the delayed thread pointers without triggering overflow detection
   _mi_padding_shrink(page, block, sizeof(mi_block_t));
 
-  if (segment->page_kind == MI_PAGE_HUGE) {
+  if (segment->kind == MI_SEGMENT_HUGE) {
     #if MI_HUGE_PAGE_ABANDON
     // huge page segments are always abandoned and can be freed immediately
     _mi_segment_huge_page_free(segment, page, block);
@@ -514,24 +522,24 @@ static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) {
 // only maintain stats for smaller objects if requested
 #if (MI_STAT>0)
 static void mi_stat_free(const mi_page_t* page, const mi_block_t* block) {
-#if (MI_STAT < 2)
   MI_UNUSED(block);
-#endif
   mi_heap_t* const heap = mi_heap_get_default();
   const size_t bsize = mi_page_usable_block_size(page);
-#if (MI_STAT>1)
-  const size_t usize = mi_page_usable_size_of(page, block);
-  mi_heap_stat_decrease(heap, malloc, usize);
-#endif
-  if (bsize <= MI_LARGE_OBJ_SIZE_MAX) {
-    mi_heap_stat_decrease(heap, normal, bsize);
-#if (MI_STAT > 1)
-    mi_heap_stat_decrease(heap, normal_bins[_mi_bin(bsize)], 1);
-#endif
+  // #if (MI_STAT>1)
+  // const size_t usize = mi_page_usable_size_of(page, block);
+  // mi_heap_stat_decrease(heap, malloc_requested, usize);
+  // #endif
+  if (bsize <= MI_MEDIUM_OBJ_SIZE_MAX) {
+    mi_heap_stat_decrease(heap, malloc_normal, bsize);
+    #if (MI_STAT > 1)
+    mi_heap_stat_decrease(heap, malloc_bins[_mi_bin(bsize)], 1);
+    #endif
   }
+  //else if (bsize <= MI_LARGE_OBJ_SIZE_MAX) {
+  //  mi_heap_stat_decrease(heap, malloc_large, bsize);
+  //}
   else {
-    const size_t bpsize = mi_page_block_size(page);  // match stat in page.c:mi_huge_page_alloc
-    mi_heap_stat_decrease(heap, huge, bpsize);
+    mi_heap_stat_decrease(heap, malloc_huge, bsize);
   }
 }
 #else

src/heap.c

@@ -95,6 +95,11 @@ static bool mi_heap_page_collect(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t
   mi_assert_internal(mi_heap_page_is_valid(heap, pq, page, NULL, NULL));
   mi_collect_t collect = *((mi_collect_t*)arg_collect);
   _mi_page_free_collect(page, collect >= MI_FORCE);
+  if (collect == MI_FORCE) {
+    // note: call before a potential `_mi_page_free` as the segment may be freed if this was the last used page in that segment.
+    mi_segment_t* segment = _mi_page_segment(page);
+    _mi_segment_collect(segment, true /* force? */);
+  }
   if (mi_page_all_free(page)) {
     // no more used blocks, free the page.
     // note: this will free retired pages as well.
@@ -127,14 +132,15 @@ static void mi_heap_collect_ex(mi_heap_t* heap, mi_collect_t collect)
   const bool is_main_thread = (_mi_is_main_thread() && heap->thread_id == _mi_thread_id());
 
   // note: never reclaim on collect but leave it to threads that need storage to reclaim
-  if (
-  #ifdef NDEBUG
+  const bool force_main =
+    #ifdef NDEBUG
       collect == MI_FORCE
-  #else
+    #else
       collect >= MI_FORCE
-  #endif
-    && is_main_thread && mi_heap_is_backing(heap) && !heap->no_reclaim)
-  {
+    #endif
+      && is_main_thread && mi_heap_is_backing(heap) && !heap->no_reclaim;
+
+  if (force_main) {
     // the main thread is abandoned (end-of-program), try to reclaim all abandoned segments.
     // if all memory is freed by now, all segments should be freed.
     // note: this only collects in the current subprocess
@@ -157,8 +163,9 @@ static void mi_heap_collect_ex(mi_heap_t* heap, mi_collect_t collect)
   mi_heap_visit_pages(heap, &mi_heap_page_collect, &collect, NULL);
   mi_assert_internal( collect != MI_ABANDON || mi_atomic_load_ptr_acquire(mi_block_t,&heap->thread_delayed_free) == NULL );
 
-  // collect segments (purge pages, this can be expensive so don't force on abandonment)
-  _mi_segments_collect(collect == MI_FORCE, &heap->tld->segments);
+  // collect abandoned segments (in particular, purge expired parts of segments in the abandoned segment list)
+  // note: forced purge can be quite expensive if many threads are created/destroyed so we do not force on abandonment
+  _mi_abandoned_collect(heap, collect == MI_FORCE /* force? */, &heap->tld->segments);
 
   // if forced, collect thread data cache on program-exit (or shared library unload)
   if (force && is_main_thread && mi_heap_is_backing(heap)) {
@@ -167,6 +174,11 @@ static void mi_heap_collect_ex(mi_heap_t* heap, mi_collect_t collect)
 
   // collect arenas (this is program wide so don't force purges on abandonment of threads)
   _mi_arenas_collect(collect == MI_FORCE /* force purge? */);
+
+  // merge statistics
+  if (collect <= MI_FORCE) {
+    mi_stats_merge();
+  }
 }
 
 void _mi_heap_collect_abandon(mi_heap_t* heap) {
@@ -323,20 +335,26 @@ static bool _mi_heap_page_destroy(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_
 
   // stats
   const size_t bsize = mi_page_block_size(page);
-  if (bsize > MI_LARGE_OBJ_SIZE_MAX) {
-    mi_heap_stat_decrease(heap, huge, bsize);
+  if (bsize > MI_MEDIUM_OBJ_SIZE_MAX) {
+    //if (bsize <= MI_LARGE_OBJ_SIZE_MAX) {
+    //  mi_heap_stat_decrease(heap, malloc_large, bsize);
+    //}
+    //else 
+    {
+      mi_heap_stat_decrease(heap, malloc_huge, bsize);
+    }
   }
-#if (MI_STAT)
+  #if (MI_STAT>0)
   _mi_page_free_collect(page, false);  // update used count
   const size_t inuse = page->used;
   if (bsize <= MI_LARGE_OBJ_SIZE_MAX) {
-    mi_heap_stat_decrease(heap, normal, bsize * inuse);
-#if (MI_STAT>1)
-    mi_heap_stat_decrease(heap, normal_bins[_mi_bin(bsize)], inuse);
-#endif
+    mi_heap_stat_decrease(heap, malloc_normal, bsize * inuse);
+    #if (MI_STAT>1)
+    mi_heap_stat_decrease(heap, malloc_bins[_mi_bin(bsize)], inuse);
+    #endif
   }
-  mi_heap_stat_decrease(heap, malloc, bsize * inuse);  // todo: off for aligned blocks...
-#endif
+  // mi_heap_stat_decrease(heap, malloc_requested, bsize * inuse);  // todo: off for aligned blocks...
+  #endif
 
   /// pretend it is all free now
   mi_assert_internal(mi_page_thread_free(page) == NULL);
@@ -552,7 +570,7 @@ void _mi_heap_area_init(mi_heap_area_t* area, mi_page_t* page) {
 
 static void mi_get_fast_divisor(size_t divisor, uint64_t* magic, size_t* shift) {
   mi_assert_internal(divisor > 0 && divisor <= UINT32_MAX);
-  *shift = MI_INTPTR_BITS - mi_clz(divisor - 1);
+  *shift = MI_SIZE_BITS - mi_clz(divisor - 1);
   *magic = ((((uint64_t)1 << 32) * (((uint64_t)1 << *shift) - divisor)) / divisor + 1);
 }
 

src/init.c

@@ -34,13 +34,12 @@ const mi_page_t _mi_page_empty = {
   MI_ATOMIC_VAR_INIT(0), // xthread_free
   MI_ATOMIC_VAR_INIT(0), // xheap
   NULL, NULL
-  #if MI_INTPTR_SIZE==4
-  , { NULL }
-  #endif
+  , { 0 }  // padding
 };
 
 #define MI_PAGE_EMPTY() ((mi_page_t*)&_mi_page_empty)
 
+#if (MI_SMALL_WSIZE_MAX==128)
 #if (MI_PADDING>0) && (MI_INTPTR_SIZE >= 8)
 #define MI_SMALL_PAGES_EMPTY  { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY(), MI_PAGE_EMPTY() }
 #elif (MI_PADDING>0)
@@ -48,7 +47,9 @@ const mi_page_t _mi_page_empty = {
 #else
 #define MI_SMALL_PAGES_EMPTY  { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY() }
 #endif
-
+#else
+#error "define right initialization sizes corresponding to MI_SMALL_WSIZE_MAX"
+#endif
 
 // Empty page queues for every bin
 #define QNULL(sz)  { NULL, NULL, (sz)*sizeof(uintptr_t) }
@@ -63,32 +64,40 @@ const mi_page_t _mi_page_empty = {
     QNULL( 10240), QNULL( 12288), QNULL( 14336), QNULL( 16384), QNULL( 20480), QNULL( 24576), QNULL( 28672), QNULL( 32768), /* 56 */ \
     QNULL( 40960), QNULL( 49152), QNULL( 57344), QNULL( 65536), QNULL( 81920), QNULL( 98304), QNULL(114688), QNULL(131072), /* 64 */ \
     QNULL(163840), QNULL(196608), QNULL(229376), QNULL(262144), QNULL(327680), QNULL(393216), QNULL(458752), QNULL(524288), /* 72 */ \
-    QNULL(MI_LARGE_OBJ_WSIZE_MAX + 1  /* 655360, Huge queue */), \
-    QNULL(MI_LARGE_OBJ_WSIZE_MAX + 2) /* Full queue */ }
+    QNULL(MI_MEDIUM_OBJ_WSIZE_MAX + 1  /* 655360, Huge queue */), \
+    QNULL(MI_MEDIUM_OBJ_WSIZE_MAX + 2) /* Full queue */ }
 
-#define MI_STAT_COUNT_NULL()  {0,0,0,0}
+#define MI_STAT_COUNT_NULL()  {0,0,0}
 
 // Empty statistics
-#if MI_STAT>1
-#define MI_STAT_COUNT_END_NULL()  , { MI_STAT_COUNT_NULL(), MI_INIT32(MI_STAT_COUNT_NULL) }
-#else
-#define MI_STAT_COUNT_END_NULL()
-#endif
-
 #define MI_STATS_NULL  \
-  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
-  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
-  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
-  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
-  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
-  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
-  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
-  MI_STAT_COUNT_NULL(), \
-  { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, \
-  { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, \
-  { 0, 0 }, { 0, 0 }, { 0, 0 }, { 0, 0 }, \
-  { 0, 0 } \
-  MI_STAT_COUNT_END_NULL()
+  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
+  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
+  MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \
+  { 0 }, { 0 }, { 0 }, { 0 }, \
+  { 0 }, { 0 }, { 0 }, { 0 }, \
+  \
+  { 0 }, { 0 }, { 0 }, { 0 }, { 0 }, \
+  MI_INIT4(MI_STAT_COUNT_NULL), \
+  { 0 }, { 0 }, { 0 }, { 0 },  \
+  \
+  { MI_INIT4(MI_STAT_COUNT_NULL) }, \
+  { { 0 }, { 0 }, { 0 }, { 0 } }, \
+  \
+  { MI_INIT74(MI_STAT_COUNT_NULL) }, \
+  { MI_INIT74(MI_STAT_COUNT_NULL) }
+
+
+// Empty slice span queues for every bin
+#define SQNULL(sz)  { NULL, NULL, sz }
+#define MI_SEGMENT_SPAN_QUEUES_EMPTY \
+  { SQNULL(1), \
+    SQNULL(     1), SQNULL(     2), SQNULL(     3), SQNULL(     4), SQNULL(     5), SQNULL(     6), SQNULL(     7), SQNULL(    10), /*  8 */ \
+    SQNULL(    12), SQNULL(    14), SQNULL(    16), SQNULL(    20), SQNULL(    24), SQNULL(    28), SQNULL(    32), SQNULL(    40), /* 16 */ \
+    SQNULL(    48), SQNULL(    56), SQNULL(    64), SQNULL(    80), SQNULL(    96), SQNULL(   112), SQNULL(   128), SQNULL(   160), /* 24 */ \
+    SQNULL(   192), SQNULL(   224), SQNULL(   256), SQNULL(   320), SQNULL(   384), SQNULL(   448), SQNULL(   512), SQNULL(   640), /* 32 */ \
+    SQNULL(   768), SQNULL(   896), SQNULL(  1024) /* 35 */ }
+
 
 // --------------------------------------------------------
 // Statically allocate an empty heap as the initial
@@ -109,6 +118,7 @@ mi_decl_cache_align const mi_heap_t _mi_heap_empty = {
   { {0}, {0}, 0, true }, // random
   0,                // page count
   MI_BIN_FULL, 0,   // page retired min/max
+  0, 0,             // generic count
   NULL,             // next
   false,            // can reclaim
   0,                // tag
@@ -119,6 +129,17 @@ mi_decl_cache_align const mi_heap_t _mi_heap_empty = {
   MI_PAGE_QUEUES_EMPTY
 };
 
+static mi_decl_cache_align mi_subproc_t mi_subproc_default;
+
+#define tld_empty_stats  ((mi_stats_t*)((uint8_t*)&tld_empty + offsetof(mi_tld_t,stats)))
+
+mi_decl_cache_align static const mi_tld_t tld_empty = {
+  0,
+  false,
+  NULL, NULL,
+  { MI_SEGMENT_SPAN_QUEUES_EMPTY, 0, 0, 0, 0, 0, &mi_subproc_default, tld_empty_stats }, // segments
+  { MI_STAT_VERSION, MI_STATS_NULL }       // stats
+};
 
 mi_threadid_t _mi_thread_id(void) mi_attr_noexcept {
   return _mi_prim_thread_id();
@@ -127,18 +148,13 @@ mi_threadid_t _mi_thread_id(void) mi_attr_noexcept {
 // the thread-local default heap for allocation
 mi_decl_thread mi_heap_t* _mi_heap_default = (mi_heap_t*)&_mi_heap_empty;
 
-extern mi_heap_t _mi_heap_main;
-
-static mi_decl_cache_align mi_subproc_t mi_subproc_default;
+extern mi_decl_hidden mi_heap_t _mi_heap_main;
 
 static mi_decl_cache_align mi_tld_t tld_main = {
   0, false,
-  &_mi_heap_main, &_mi_heap_main,
-  { { NULL, NULL }, {NULL ,NULL}, {NULL ,NULL, 0},
-    0, 0, 0, 0, 0, &mi_subproc_default,
-    &tld_main.stats
-  }, // segments
-  { MI_STATS_NULL }       // stats
+  &_mi_heap_main, & _mi_heap_main,
+  { MI_SEGMENT_SPAN_QUEUES_EMPTY, 0, 0, 0, 0, 0, &mi_subproc_default, &tld_main.stats }, // segments
+  { MI_STAT_VERSION, MI_STATS_NULL }       // stats
 };
 
 mi_decl_cache_align mi_heap_t _mi_heap_main = {
@@ -151,6 +167,7 @@ mi_decl_cache_align mi_heap_t _mi_heap_main = {
   { {0x846ca68b}, {0}, 0, true },  // random
   0,                // page count
   MI_BIN_FULL, 0,   // page retired min/max
+  0, 0,             // generic count
   NULL,             // next heap
   false,            // can reclaim
   0,                // tag
@@ -163,7 +180,7 @@ mi_decl_cache_align mi_heap_t _mi_heap_main = {
 
 bool _mi_process_is_initialized = false;  // set to `true` in `mi_process_init`.
 
-mi_stats_t _mi_stats_main = { MI_STATS_NULL };
+mi_stats_t _mi_stats_main = { MI_STAT_VERSION, MI_STATS_NULL };
 
 #if MI_GUARDED
 mi_decl_export void mi_heap_guarded_set_sample_rate(mi_heap_t* heap, size_t sample_rate, size_t seed) {
@@ -393,7 +410,7 @@ static bool _mi_thread_heap_init(void) {
 
 // initialize thread local data
 void _mi_tld_init(mi_tld_t* tld, mi_heap_t* bheap) {
-  _mi_memzero_aligned(tld,sizeof(mi_tld_t));
+  _mi_memcpy_aligned(tld, &tld_empty, sizeof(mi_tld_t));
   tld->heap_backing = bheap;
   tld->heaps = NULL;
   tld->segments.subproc = &mi_subproc_default;
@@ -434,7 +451,10 @@ static bool _mi_thread_heap_done(mi_heap_t* heap) {
 
   // free if not the main thread
   if (heap != &_mi_heap_main) {
-    mi_assert_internal(heap->tld->segments.count == 0 || heap->thread_id != _mi_thread_id());
+    // the following assertion does not always hold for huge segments as those are always treated
+    // as abondened: one may allocate it in one thread, but deallocate in another in which case
+    // the count can be too large or negative. todo: perhaps not count huge segments? see issue #363
+    // mi_assert_internal(heap->tld->segments.count == 0 || heap->thread_id != _mi_thread_id());
     mi_thread_data_free((mi_thread_data_t*)heap);
   }
   else {
@@ -547,6 +567,9 @@ void _mi_heap_set_default_direct(mi_heap_t* heap)  {
   _mi_prim_thread_associate_default_heap(heap);
 }
 
+void mi_thread_set_in_threadpool(void) mi_attr_noexcept {
+  // nothing
+}
 
 // --------------------------------------------------------
 // Run functions on process init/done, and thread init/done
@@ -558,6 +581,11 @@ bool mi_decl_noinline _mi_preloading(void) {
   return os_preloading;
 }
 
+// Returns true if mimalloc was redirected
+mi_decl_nodiscard bool mi_is_redirected(void) mi_attr_noexcept {
+  return _mi_is_redirected();
+}
+
 // Called once by the process loader from `src/prim/prim.c`
 void _mi_process_load(void) {
   mi_heap_main_init();
@@ -594,7 +622,7 @@ static void mi_detect_cpu_features(void) {
   int32_t cpu_info[4];
   __cpuid(cpu_info, 7);
   _mi_cpu_has_fsrm = ((cpu_info[3] & (1 << 4)) != 0); // bit 4 of EDX : see <https://en.wikipedia.org/wiki/CPUID#EAX=7,_ECX=0:_Extended_Features>
-  _mi_cpu_has_erms = ((cpu_info[2] & (1 << 9)) != 0); // bit 9 of ECX : see <https://en.wikipedia.org/wiki/CPUID#EAX=7,_ECX=0:_Extended_Features>
+  _mi_cpu_has_erms = ((cpu_info[1] & (1 << 9)) != 0); // bit 9 of EBX : see <https://en.wikipedia.org/wiki/CPUID#EAX=7,_ECX=0:_Extended_Features>
 }
 #else
 static void mi_detect_cpu_features(void) {
@@ -617,14 +645,6 @@ void mi_process_init(void) mi_attr_noexcept {
   mi_detect_cpu_features();
   _mi_os_init();
   mi_heap_main_init();
-  #if MI_DEBUG
-  _mi_verbose_message("debug level : %d\n", MI_DEBUG);
-  #endif
-  _mi_verbose_message("secure level: %d\n", MI_SECURE);
-  _mi_verbose_message("mem tracking: %s\n", MI_TRACK_TOOL);
-  #if MI_TSAN
-  _mi_verbose_message("thread santizer enabled\n");
-  #endif
   mi_thread_init();
 
   #if defined(_WIN32)
@@ -649,7 +669,7 @@ void mi_process_init(void) mi_attr_noexcept {
   if (mi_option_is_enabled(mi_option_reserve_os_memory)) {
     long ksize = mi_option_get(mi_option_reserve_os_memory);
     if (ksize > 0) {
-      mi_reserve_os_memory((size_t)ksize*MI_KiB, true, true);
+      mi_reserve_os_memory((size_t)ksize*MI_KiB, true /* commit? */, true /* allow large pages? */);
     }
   }
 }

src/libc.c

@@ -7,7 +7,7 @@ terms of the MIT license. A copy of the license can be found in the file
 
 // --------------------------------------------------------
 // This module defines various std libc functions to reduce
-// the dependency on libc, and also prevent errors caused 
+// the dependency on libc, and also prevent errors caused
 // by some libc implementations when called before `main`
 // executes (due to malloc redirection)
 // --------------------------------------------------------
@@ -83,7 +83,7 @@ bool _mi_getenv(const char* name, char* result, size_t result_size) {
 // Define our own limited `_mi_vsnprintf` and `_mi_snprintf`
 // This is mostly to avoid calling these when libc is not yet
 // initialized (and to reduce dependencies)
-// 
+//
 // format:      d i, p x u, s
 // prec:        z l ll L
 // width:       10
@@ -130,7 +130,7 @@ static void mi_out_alignright(char fill, char* start, size_t len, size_t extra,
 }
 
 
-static void mi_out_num(uintmax_t x, size_t base, char prefix, char** out, char* end) 
+static void mi_out_num(uintmax_t x, size_t base, char prefix, char** out, char* end)
 {
   if (x == 0 || base == 0 || base > 16) {
     if (prefix != 0) { mi_outc(prefix, out, end); }
@@ -144,8 +144,8 @@ static void mi_out_num(uintmax_t x, size_t base, char prefix, char** out, char*
       mi_outc((digit <= 9 ? '0' + digit : 'A' + digit - 10),out,end);
       x = x / base;
     }
-    if (prefix != 0) { 
-      mi_outc(prefix, out, end); 
+    if (prefix != 0) {
+      mi_outc(prefix, out, end);
     }
     size_t len = *out - start;
     // and reverse in-place
@@ -160,8 +160,8 @@ static void mi_out_num(uintmax_t x, size_t base, char prefix, char** out, char*
 
 #define MI_NEXTC()  c = *in; if (c==0) break; in++;
 
-void _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args) {
-  if (buf == NULL || bufsize == 0 || fmt == NULL) return;
+int _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args) {
+  if (buf == NULL || bufsize == 0 || fmt == NULL) return 0;
   buf[bufsize - 1] = 0;
   char* const end = buf + (bufsize - 1);
   const char* in = fmt;
@@ -181,7 +181,7 @@ void _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args) {
       size_t width = 0;
       char   numtype = 'd';
       char   numplus = 0;
-      bool   alignright = true; 
+      bool   alignright = true;
       if (c == '+' || c == ' ') { numplus = c; MI_NEXTC(); }
       if (c == '-') { alignright = false; MI_NEXTC(); }
       if (c == '0') { fill = '0'; MI_NEXTC(); }
@@ -191,7 +191,7 @@ void _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args) {
           width = (10 * width) + (c - '0'); MI_NEXTC();
         }
         if (c == 0) break;  // extra check due to while
-      }      
+      }
       if (c == 'z' || c == 't' || c == 'L') { numtype = c; MI_NEXTC(); }
       else if (c == 'l') {
         numtype = c; MI_NEXTC();
@@ -265,11 +265,70 @@ void _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args) {
   }
   mi_assert_internal(out <= end);
   *out = 0;
+  return (int)(out - buf);
 }
 
-void _mi_snprintf(char* buf, size_t buflen, const char* fmt, ...) {
+int _mi_snprintf(char* buf, size_t buflen, const char* fmt, ...) {
   va_list args;
   va_start(args, fmt);
-  _mi_vsnprintf(buf, buflen, fmt, args);
+  const int written = _mi_vsnprintf(buf, buflen, fmt, args);
   va_end(args);
+  return written;
 }
+
+
+#if MI_SIZE_SIZE == 4
+#define mi_mask_even_bits32      (0x55555555)
+#define mi_mask_even_pairs32     (0x33333333)
+#define mi_mask_even_nibbles32   (0x0F0F0F0F)
+
+// sum of all the bytes in `x` if it is guaranteed that the sum < 256!
+static size_t mi_byte_sum32(uint32_t x) {
+  // perform `x * 0x01010101`: the highest byte contains the sum of all bytes.
+  x += (x << 8);
+  x += (x << 16);
+  return (size_t)(x >> 24);
+}
+
+static size_t mi_popcount_generic32(uint32_t x) {
+  // first count each 2-bit group `a`, where: a==0b00 -> 00, a==0b01 -> 01, a==0b10 -> 01, a==0b11 -> 10
+  // in other words, `a - (a>>1)`; to do this in parallel, we need to mask to prevent spilling a bit pair
+  // into the lower bit-pair:
+  x = x - ((x >> 1) & mi_mask_even_bits32);
+  // add the 2-bit pair results
+  x = (x & mi_mask_even_pairs32) + ((x >> 2) & mi_mask_even_pairs32);
+  // add the 4-bit nibble results
+  x = (x + (x >> 4)) & mi_mask_even_nibbles32;
+  // each byte now has a count of its bits, we can sum them now:
+  return mi_byte_sum32(x);
+}
+
+mi_decl_noinline size_t _mi_popcount_generic(size_t x) {
+  return mi_popcount_generic32(x);
+}
+
+#else
+#define mi_mask_even_bits64      (0x5555555555555555)
+#define mi_mask_even_pairs64     (0x3333333333333333)
+#define mi_mask_even_nibbles64   (0x0F0F0F0F0F0F0F0F)
+
+// sum of all the bytes in `x` if it is guaranteed that the sum < 256!
+static size_t mi_byte_sum64(uint64_t x) {
+  x += (x << 8);
+  x += (x << 16);
+  x += (x << 32);
+  return (size_t)(x >> 56);
+}
+
+static size_t mi_popcount_generic64(uint64_t x) {
+  x = x - ((x >> 1) & mi_mask_even_bits64);
+  x = (x & mi_mask_even_pairs64) + ((x >> 2) & mi_mask_even_pairs64);
+  x = (x + (x >> 4)) & mi_mask_even_nibbles64;
+  return mi_byte_sum64(x);
+}
+
+mi_decl_noinline size_t _mi_popcount_generic(size_t x) {
+  return mi_popcount_generic64(x);
+}
+#endif
+

src/options.c

@@ -106,11 +106,11 @@ typedef struct mi_option_desc_s {
 static mi_option_desc_t options[_mi_option_last] =
 {
   // stable options
-#if MI_DEBUG || defined(MI_SHOW_ERRORS)
+  #if MI_DEBUG || defined(MI_SHOW_ERRORS)
   { 1, UNINIT, MI_OPTION(show_errors) },
-#else
+  #else
   { 0, UNINIT, MI_OPTION(show_errors) },
-#endif
+  #endif
   { 0, UNINIT, MI_OPTION(show_stats) },
   { MI_DEFAULT_VERBOSE, UNINIT, MI_OPTION(verbose) },
 
@@ -129,7 +129,7 @@ static mi_option_desc_t options[_mi_option_last] =
        UNINIT, MI_OPTION(reserve_os_memory)     },      // reserve N KiB OS memory in advance (use `option_get_size`)
   { 0, UNINIT, MI_OPTION(deprecated_segment_cache) },   // cache N segments per thread
   { 0, UNINIT, MI_OPTION(deprecated_page_reset) },      // reset page memory on free
-  { 0, UNINIT, MI_OPTION(abandoned_page_purge) },       // purge free page memory when a thread terminates
+  { 0, UNINIT, MI_OPTION_LEGACY(abandoned_page_purge,abandoned_page_reset) },       // reset free page memory when a thread terminates
   { 0, UNINIT, MI_OPTION(deprecated_segment_reset) },   // reset segment memory on free (needs eager commit)
 #if defined(__NetBSD__)
   { 0, UNINIT, MI_OPTION(eager_commit_delay) },         // the first N segments per thread are not eagerly committed
@@ -162,6 +162,7 @@ static mi_option_desc_t options[_mi_option_last] =
          UNINIT, MI_OPTION(guarded_sample_rate)},       // 1 out of N allocations in the min/max range will be guarded (=4000)
   { 0,   UNINIT, MI_OPTION(guarded_sample_seed)},
   { 0,   UNINIT, MI_OPTION(target_segments_per_thread) }, // abandon segments beyond this point, or 0 to disable.
+  { 10000, UNINIT, MI_OPTION(generic_collect) },          // collect heaps every N (=10000) generic allocation calls
 };
 
 static void mi_option_init(mi_option_desc_t* desc);
@@ -176,11 +177,6 @@ void _mi_options_init(void) {
   for(int i = 0; i < _mi_option_last; i++ ) {
     mi_option_t option = (mi_option_t)i;
     long l = mi_option_get(option); MI_UNUSED(l); // initialize
-    // if (option != mi_option_verbose)
-    {
-      mi_option_desc_t* desc = &options[option];
-      _mi_verbose_message("option '%s': %ld %s\n", desc->name, desc->value, (mi_option_has_size_in_kib(option) ? "KiB" : ""));
-    }
   }
   mi_max_error_count = mi_option_get(mi_option_max_errors);
   mi_max_warning_count = mi_option_get(mi_option_max_warnings);
@@ -191,7 +187,50 @@ void _mi_options_init(void) {
       _mi_warning_message("option 'allow_large_os_pages' is disabled to allow for guarded objects\n");
     }
   }
-  _mi_verbose_message("guarded build: %s\n", mi_option_get(mi_option_guarded_sample_rate) != 0 ? "enabled" : "disabled");
+  #endif
+  if (mi_option_is_enabled(mi_option_verbose)) { mi_options_print(); }
+}
+
+#define mi_stringifyx(str)  #str                // and stringify
+#define mi_stringify(str)   mi_stringifyx(str)  // expand
+
+void mi_options_print(void) mi_attr_noexcept
+{
+  // show version
+  const int vermajor = MI_MALLOC_VERSION/100;
+  const int verminor = (MI_MALLOC_VERSION%100)/10;
+  const int verpatch = (MI_MALLOC_VERSION%10);
+  _mi_message("v%i.%i.%i%s%s (built on %s, %s)\n", vermajor, verminor, verpatch,
+      #if defined(MI_CMAKE_BUILD_TYPE)
+      ", " mi_stringify(MI_CMAKE_BUILD_TYPE)
+      #else
+      ""
+      #endif
+      ,
+      #if defined(MI_GIT_DESCRIBE)
+      ", git " mi_stringify(MI_GIT_DESCRIBE)
+      #else
+      ""
+      #endif
+      , __DATE__, __TIME__);
+
+  // show options
+  for (int i = 0; i < _mi_option_last; i++) {
+    mi_option_t option = (mi_option_t)i;
+    long l = mi_option_get(option); MI_UNUSED(l); // possibly initialize
+    mi_option_desc_t* desc = &options[option];
+    _mi_message("option '%s': %ld %s\n", desc->name, desc->value, (mi_option_has_size_in_kib(option) ? "KiB" : ""));
+  }
+
+  // show build configuration
+  _mi_message("debug level : %d\n", MI_DEBUG );
+  _mi_message("secure level: %d\n", MI_SECURE );
+  _mi_message("mem tracking: %s\n", MI_TRACK_TOOL);
+  #if MI_GUARDED
+  _mi_message("guarded build: %s\n", mi_option_get(mi_option_guarded_sample_rate) != 0 ? "enabled" : "disabled");
+  #endif
+  #if MI_TSAN
+  _mi_message("thread santizer enabled\n");
   #endif
 }
 
@@ -442,6 +481,13 @@ static void mi_vfprintf_thread(mi_output_fun* out, void* arg, const char* prefix
   }
 }
 
+void _mi_message(const char* fmt, ...) {
+  va_list args;
+  va_start(args, fmt);
+  mi_vfprintf_thread(NULL, NULL, "mimalloc: ", fmt, args);
+  va_end(args);
+}
+
 void _mi_trace_message(const char* fmt, ...) {
   if (mi_option_get(mi_option_verbose) <= 1) return;  // only with verbose level 2 or higher
   va_list args;

src/os.c

@@ -18,17 +18,17 @@ terms of the MIT license. A copy of the license can be found in the file
 ----------------------------------------------------------- */
 #ifndef MI_DEFAULT_VIRTUAL_ADDRESS_BITS
 #if MI_INTPTR_SIZE < 8
-#define MI_DEFAULT_VIRTUAL_ADDRESS_BITS   32
+#define MI_DEFAULT_VIRTUAL_ADDRESS_BITS     32
 #else
-#define MI_DEFAULT_VIRTUAL_ADDRESS_BITS   48
+#define MI_DEFAULT_VIRTUAL_ADDRESS_BITS     48
 #endif
 #endif
 
-#ifndef MI_DEFAULT_PHYSICAL_MEMORY
+#ifndef MI_DEFAULT_PHYSICAL_MEMORY_IN_KIB
 #if MI_INTPTR_SIZE < 8
-#define MI_DEFAULT_PHYSICAL_MEMORY    4*MI_GiB
+#define MI_DEFAULT_PHYSICAL_MEMORY_IN_KIB   4*MI_MiB    // 4 GiB
 #else
-#define MI_DEFAULT_PHYSICAL_MEMORY    32*MI_GiB
+#define MI_DEFAULT_PHYSICAL_MEMORY_IN_KIB   32*MI_MiB   // 32 GiB
 #endif
 #endif
 
@@ -36,7 +36,7 @@ static mi_os_mem_config_t mi_os_mem_config = {
   4096,     // page size
   0,        // large page size (usually 2MiB)
   4096,     // allocation granularity
-  MI_DEFAULT_PHYSICAL_MEMORY,
+  MI_DEFAULT_PHYSICAL_MEMORY_IN_KIB,
   MI_DEFAULT_VIRTUAL_ADDRESS_BITS,
   true,     // has overcommit?  (if true we use MAP_NORESERVE on mmap systems)
   false,    // can we partially free allocated blocks? (on mmap systems we can free anywhere in a mapped range, but on Windows we must free the entire span)
@@ -91,21 +91,6 @@ void _mi_os_init(void) {
 bool _mi_os_decommit(void* addr, size_t size);
 bool _mi_os_commit(void* addr, size_t size, bool* is_zero);
 
-static inline uintptr_t _mi_align_down(uintptr_t sz, size_t alignment) {
-  mi_assert_internal(alignment != 0);
-  uintptr_t mask = alignment - 1;
-  if ((alignment & mask) == 0) { // power of two?
-    return (sz & ~mask);
-  }
-  else {
-    return ((sz / alignment) * alignment);
-  }
-}
-
-static void* mi_align_down_ptr(void* p, size_t alignment) {
-  return (void*)_mi_align_down((uintptr_t)p, alignment);
-}
-
 
 /* -----------------------------------------------------------
   aligned hinting
@@ -519,7 +504,7 @@ bool _mi_os_purge_ex(void* p, size_t size, bool allow_reset, size_t stat_size)
   mi_os_stat_increase(purged, size);
 
   if (mi_option_is_enabled(mi_option_purge_decommits) &&   // should decommit?
-    !_mi_preloading())                                     // don't decommit during preloading (unsafe)
+      !_mi_preloading())                                   // don't decommit during preloading (unsafe)
   {
     bool needs_recommit = true;
     mi_os_decommit_ex(p, size, &needs_recommit, stat_size);
@@ -539,7 +524,6 @@ bool _mi_os_purge(void* p, size_t size) {
   return _mi_os_purge_ex(p, size, true, size);
 }
 
-
 // Protect a region in memory to be not accessible.
 static  bool mi_os_protectx(void* addr, size_t size, bool protect) {
   // page align conservatively within the range

src/page-queue.c

@@ -12,7 +12,7 @@ terms of the MIT license. A copy of the license can be found in the file
 #ifndef MI_IN_PAGE_C
 #error "this file should be included from 'page.c'"
 // include to help an IDE
-#include "mimalloc.h"     
+#include "mimalloc.h"
 #include "mimalloc/internal.h"
 #include "mimalloc/atomic.h"
 #endif
@@ -38,15 +38,15 @@ terms of the MIT license. A copy of the license can be found in the file
 
 
 static inline bool mi_page_queue_is_huge(const mi_page_queue_t* pq) {
-  return (pq->block_size == (MI_LARGE_OBJ_SIZE_MAX+sizeof(uintptr_t)));
+  return (pq->block_size == (MI_MEDIUM_OBJ_SIZE_MAX+sizeof(uintptr_t)));
 }
 
 static inline bool mi_page_queue_is_full(const mi_page_queue_t* pq) {
-  return (pq->block_size == (MI_LARGE_OBJ_SIZE_MAX+(2*sizeof(uintptr_t))));
+  return (pq->block_size == (MI_MEDIUM_OBJ_SIZE_MAX+(2*sizeof(uintptr_t))));
 }
 
 static inline bool mi_page_queue_is_special(const mi_page_queue_t* pq) {
-  return (pq->block_size > MI_LARGE_OBJ_SIZE_MAX);
+  return (pq->block_size > MI_MEDIUM_OBJ_SIZE_MAX);
 }
 
 /* -----------------------------------------------------------
@@ -57,27 +57,23 @@ static inline bool mi_page_queue_is_special(const mi_page_queue_t* pq) {
 // Returns MI_BIN_HUGE if the size is too large.
 // We use `wsize` for the size in "machine word sizes",
 // i.e. byte size == `wsize*sizeof(void*)`.
-static inline uint8_t mi_bin(size_t size) {
+static inline size_t mi_bin(size_t size) {
   size_t wsize = _mi_wsize_from_size(size);
-  uint8_t bin;
-  if (wsize <= 1) {
-    bin = 1;
+#if defined(MI_ALIGN4W)
+  if mi_likely(wsize <= 4) {
+    return (wsize <= 1 ? 1 : (wsize+1)&~1); // round to double word sizes
   }
-  #if defined(MI_ALIGN4W)
-  else if (wsize <= 4) {
-    bin = (uint8_t)((wsize+1)&~1); // round to double word sizes
+#elif defined(MI_ALIGN2W)
+  if mi_likely(wsize <= 8) {
+    return (wsize <= 1 ? 1 : (wsize+1)&~1); // round to double word sizes
   }
-  #elif defined(MI_ALIGN2W)
-  else if (wsize <= 8) {
-    bin = (uint8_t)((wsize+1)&~1); // round to double word sizes
+#else
+  if mi_likely(wsize <= 8) {
+    return (wsize == 0 ? 1 : wsize);
   }
-  #else
-  else if (wsize <= 8) {
-    bin = (uint8_t)wsize;
-  }
-  #endif
-  else if (wsize > MI_LARGE_OBJ_WSIZE_MAX) {
-    bin = MI_BIN_HUGE;
+#endif
+  else if mi_unlikely(wsize > MI_MEDIUM_OBJ_WSIZE_MAX) {
+    return MI_BIN_HUGE;
   }
   else {
     #if defined(MI_ALIGN4W)
@@ -85,15 +81,14 @@ static inline uint8_t mi_bin(size_t size) {
     #endif
     wsize--;
     // find the highest bit
-    uint8_t b = (uint8_t)mi_bsr(wsize);  // note: wsize != 0
+    const size_t b = (MI_SIZE_BITS - 1 - mi_clz(wsize));  // note: wsize != 0
     // and use the top 3 bits to determine the bin (~12.5% worst internal fragmentation).
     // - adjust with 3 because we use do not round the first 8 sizes
     //   which each get an exact bin
-    bin = ((b << 2) + (uint8_t)((wsize >> (b - 2)) & 0x03)) - 3;
-    mi_assert_internal(bin < MI_BIN_HUGE);
+    const size_t bin = ((b << 2) + ((wsize >> (b - 2)) & 0x03)) - 3;
+    mi_assert_internal(bin > 0 && bin < MI_BIN_HUGE);
+    return bin;
   }
-  mi_assert_internal(bin > 0 && bin <= MI_BIN_HUGE);
-  return bin;
 }
 
 
@@ -102,17 +97,17 @@ static inline uint8_t mi_bin(size_t size) {
   Queue of pages with free blocks
 ----------------------------------------------------------- */
 
-uint8_t _mi_bin(size_t size) {
+size_t _mi_bin(size_t size) {
   return mi_bin(size);
 }
 
-size_t _mi_bin_size(uint8_t bin) {
+size_t _mi_bin_size(size_t bin) {
   return _mi_heap_empty.pages[bin].block_size;
 }
 
 // Good size for allocation
 size_t mi_good_size(size_t size) mi_attr_noexcept {
-  if (size <= MI_LARGE_OBJ_SIZE_MAX) {
+  if (size <= MI_MEDIUM_OBJ_SIZE_MAX) {
     return _mi_bin_size(mi_bin(size + MI_PADDING_SIZE));
   }
   else {
@@ -141,13 +136,22 @@ static bool mi_heap_contains_queue(const mi_heap_t* heap, const mi_page_queue_t*
 }
 #endif
 
+static inline bool mi_page_is_large_or_huge(const mi_page_t* page) {
+  return (mi_page_block_size(page) > MI_MEDIUM_OBJ_SIZE_MAX || mi_page_is_huge(page));
+}
+
+static size_t mi_page_bin(const mi_page_t* page) {
+  const size_t bin = (mi_page_is_in_full(page) ? MI_BIN_FULL : (mi_page_is_huge(page) ? MI_BIN_HUGE : mi_bin(mi_page_block_size(page))));
+  mi_assert_internal(bin <= MI_BIN_FULL);
+  return bin;
+}
+
 static mi_page_queue_t* mi_heap_page_queue_of(mi_heap_t* heap, const mi_page_t* page) {
   mi_assert_internal(heap!=NULL);
-  uint8_t bin = (mi_page_is_in_full(page) ? MI_BIN_FULL : (mi_page_is_huge(page) ? MI_BIN_HUGE : mi_bin(mi_page_block_size(page))));
-  mi_assert_internal(bin <= MI_BIN_FULL);
+  const size_t bin = mi_page_bin(page);
   mi_page_queue_t* pq = &heap->pages[bin];
   mi_assert_internal((mi_page_block_size(page) == pq->block_size) ||
-                       (mi_page_is_huge(page) && mi_page_queue_is_huge(pq)) ||
+                       (mi_page_is_large_or_huge(page) && mi_page_queue_is_huge(pq)) ||
                          (mi_page_is_in_full(page) && mi_page_queue_is_full(pq)));
   return pq;
 }
@@ -185,7 +189,7 @@ static inline void mi_heap_queue_first_update(mi_heap_t* heap, const mi_page_que
   }
   else {
     // find previous size; due to minimal alignment upto 3 previous bins may need to be skipped
-    uint8_t bin = mi_bin(size);
+    size_t bin = mi_bin(size);
     const mi_page_queue_t* prev = pq - 1;
     while( bin == mi_bin(prev->block_size) && prev > &heap->pages[0]) {
       prev--;
@@ -210,10 +214,11 @@ static bool mi_page_queue_is_empty(mi_page_queue_t* queue) {
 static void mi_page_queue_remove(mi_page_queue_t* queue, mi_page_t* page) {
   mi_assert_internal(page != NULL);
   mi_assert_expensive(mi_page_queue_contains(queue, page));
-  mi_assert_internal(mi_page_block_size(page) == queue->block_size || 
-                      (mi_page_is_huge(page) && mi_page_queue_is_huge(queue)) || 
+  mi_assert_internal(mi_page_block_size(page) == queue->block_size ||
+                      (mi_page_is_large_or_huge(page) && mi_page_queue_is_huge(queue)) ||
                         (mi_page_is_in_full(page) && mi_page_queue_is_full(queue)));
   mi_heap_t* heap = mi_page_heap(page);
+
   if (page->prev != NULL) page->prev->next = page->next;
   if (page->next != NULL) page->next->prev = page->prev;
   if (page == queue->last)  queue->last = page->prev;
@@ -235,10 +240,10 @@ static void mi_page_queue_push(mi_heap_t* heap, mi_page_queue_t* queue, mi_page_
   mi_assert_internal(mi_page_heap(page) == heap);
   mi_assert_internal(!mi_page_queue_contains(queue, page));
   #if MI_HUGE_PAGE_ABANDON
-  mi_assert_internal(_mi_page_segment(page)->page_kind != MI_PAGE_HUGE);
+  mi_assert_internal(_mi_page_segment(page)->kind != MI_SEGMENT_HUGE);
   #endif
   mi_assert_internal(mi_page_block_size(page) == queue->block_size ||
-                      (mi_page_is_huge(page) && mi_page_queue_is_huge(queue)) ||
+                      (mi_page_is_large_or_huge(page) && mi_page_queue_is_huge(queue)) ||
                         (mi_page_is_in_full(page) && mi_page_queue_is_full(queue)));
 
   mi_page_set_in_full(page, mi_page_queue_is_full(queue));
@@ -277,8 +282,8 @@ static void mi_page_queue_enqueue_from_ex(mi_page_queue_t* to, mi_page_queue_t*
   mi_assert_internal((bsize == to->block_size && bsize == from->block_size) ||
                      (bsize == to->block_size && mi_page_queue_is_full(from)) ||
                      (bsize == from->block_size && mi_page_queue_is_full(to)) ||
-                     (mi_page_is_huge(page) && mi_page_queue_is_huge(to)) ||
-                     (mi_page_is_huge(page) && mi_page_queue_is_full(to)));
+                     (mi_page_is_large_or_huge(page) && mi_page_queue_is_huge(to)) ||
+                     (mi_page_is_large_or_huge(page) && mi_page_queue_is_full(to)));
 
   mi_heap_t* heap = mi_page_heap(page);
 
@@ -317,8 +322,8 @@ static void mi_page_queue_enqueue_from_ex(mi_page_queue_t* to, mi_page_queue_t*
       page->prev = to->first;
       page->next = next;
       to->first->next = page;
-      if (next != NULL) { 
-        next->prev = page; 
+      if (next != NULL) {
+        next->prev = page;
       }
       else {
         to->last = page;

src/page.c

@@ -82,11 +82,9 @@ static bool mi_page_is_valid_init(mi_page_t* page) {
   mi_assert_internal(page->used <= page->capacity);
   mi_assert_internal(page->capacity <= page->reserved);
 
-  // const size_t bsize = mi_page_block_size(page);
-  mi_segment_t* segment = _mi_page_segment(page);
   uint8_t* start = mi_page_start(page);
-  mi_assert_internal(start == _mi_segment_page_start(segment,page,NULL));
-  mi_assert_internal(page->is_huge == (segment->page_kind == MI_PAGE_HUGE));
+  mi_assert_internal(start == _mi_segment_page_start(_mi_page_segment(page), page, NULL));
+  mi_assert_internal(page->is_huge == (_mi_page_segment(page)->kind == MI_SEGMENT_HUGE));
   //mi_assert_internal(start + page->capacity*page->block_size == page->top);
 
   mi_assert_internal(mi_page_list_is_valid(page,page->free));
@@ -123,14 +121,15 @@ bool _mi_page_is_valid(mi_page_t* page) {
   #endif
   if (mi_page_heap(page)!=NULL) {
     mi_segment_t* segment = _mi_page_segment(page);
-    mi_assert_internal(!_mi_process_is_initialized || segment->thread_id == mi_page_heap(page)->thread_id || segment->thread_id==0);
+
+    mi_assert_internal(!_mi_process_is_initialized || segment->thread_id==0 || segment->thread_id == mi_page_heap(page)->thread_id);
     #if MI_HUGE_PAGE_ABANDON
-    if (segment->page_kind != MI_PAGE_HUGE)
+    if (segment->kind != MI_SEGMENT_HUGE)
     #endif
     {
       mi_page_queue_t* pq = mi_page_queue_of(page);
       mi_assert_internal(mi_page_queue_contains(pq, page));
-      mi_assert_internal(pq->block_size==mi_page_block_size(page) || mi_page_block_size(page) > MI_LARGE_OBJ_SIZE_MAX || mi_page_is_in_full(page));
+      mi_assert_internal(pq->block_size==mi_page_block_size(page) || mi_page_block_size(page) > MI_MEDIUM_OBJ_SIZE_MAX || mi_page_is_in_full(page));
       mi_assert_internal(mi_heap_contains_queue(mi_page_heap(page),pq));
     }
   }
@@ -257,10 +256,11 @@ void _mi_page_free_collect(mi_page_t* page, bool force) {
 // called from segments when reclaiming abandoned pages
 void _mi_page_reclaim(mi_heap_t* heap, mi_page_t* page) {
   mi_assert_expensive(mi_page_is_valid_init(page));
+
   mi_assert_internal(mi_page_heap(page) == heap);
   mi_assert_internal(mi_page_thread_free_flag(page) != MI_NEVER_DELAYED_FREE);
   #if MI_HUGE_PAGE_ABANDON
-  mi_assert_internal(_mi_page_segment(page)->page_kind != MI_PAGE_HUGE);
+  mi_assert_internal(_mi_page_segment(page)->kind != MI_SEGMENT_HUGE);
   #endif
 
   // TODO: push on full queue immediately if it is full?
@@ -274,7 +274,7 @@ static mi_page_t* mi_page_fresh_alloc(mi_heap_t* heap, mi_page_queue_t* pq, size
   #if !MI_HUGE_PAGE_ABANDON
   mi_assert_internal(pq != NULL);
   mi_assert_internal(mi_heap_contains_queue(heap, pq));
-  mi_assert_internal(page_alignment > 0 || block_size > MI_LARGE_OBJ_SIZE_MAX || block_size == pq->block_size);
+  mi_assert_internal(page_alignment > 0 || block_size > MI_MEDIUM_OBJ_SIZE_MAX || block_size == pq->block_size);
   #endif
   mi_page_t* page = _mi_segment_page_alloc(heap, block_size, page_alignment, &heap->tld->segments);
   if (page == NULL) {
@@ -284,12 +284,14 @@ static mi_page_t* mi_page_fresh_alloc(mi_heap_t* heap, mi_page_queue_t* pq, size
   #if MI_HUGE_PAGE_ABANDON
   mi_assert_internal(pq==NULL || _mi_page_segment(page)->page_kind != MI_PAGE_HUGE);
   #endif
+  mi_assert_internal(page_alignment >0 || block_size > MI_MEDIUM_OBJ_SIZE_MAX || _mi_page_segment(page)->kind != MI_SEGMENT_HUGE);
   mi_assert_internal(pq!=NULL || mi_page_block_size(page) >= block_size);
   // a fresh page was found, initialize it
   const size_t full_block_size = (pq == NULL || mi_page_is_huge(page) ? mi_page_block_size(page) : block_size); // see also: mi_segment_huge_page_alloc
   mi_assert_internal(full_block_size >= block_size);
   mi_page_init(heap, page, full_block_size, heap->tld);
   mi_heap_stat_increase(heap, pages, 1);
+  mi_heap_stat_increase(heap, page_bins[mi_page_bin(page)], 1);
   if (pq != NULL) { mi_page_queue_push(heap, pq, page); }
   mi_assert_expensive(_mi_page_is_valid(page));
   return page;
@@ -411,7 +413,7 @@ void _mi_page_force_abandon(mi_page_t* page) {
 
   // ensure this page is no longer in the heap delayed free list
   _mi_heap_delayed_free_all(heap);
-  // We can still access the page meta-info even if it is freed as we ensure 
+  // We can still access the page meta-info even if it is freed as we ensure
   // in `mi_segment_force_abandon` that the segment is not freed (yet)
   if (page->capacity == 0) return; // it may have been freed now
 
@@ -425,6 +427,7 @@ void _mi_page_force_abandon(mi_page_t* page) {
   }
 }
 
+
 // Free a page with no more free blocks
 void _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force) {
   mi_assert_internal(page != NULL);
@@ -438,15 +441,17 @@ void _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force) {
 
   // remove from the page list
   // (no need to do _mi_heap_delayed_free first as all blocks are already free)
-  mi_segments_tld_t* segments_tld = &mi_page_heap(page)->tld->segments;
+  mi_heap_t* heap = mi_page_heap(page);
+  mi_segments_tld_t* segments_tld = &heap->tld->segments;
   mi_page_queue_remove(pq, page);
 
   // and free it
+  mi_heap_stat_decrease(heap, page_bins[mi_page_bin(page)], 1);
   mi_page_set_heap(page,NULL);
   _mi_segment_page_free(page, force, segments_tld);
 }
 
-#define MI_MAX_RETIRE_SIZE    MI_LARGE_OBJ_SIZE_MAX   // should be less than size for MI_BIN_HUGE
+#define MI_MAX_RETIRE_SIZE    MI_MEDIUM_OBJ_SIZE_MAX   // should be less than size for MI_BIN_HUGE
 #define MI_RETIRE_CYCLES      (16)
 
 // Retire a page with no more used blocks
@@ -473,7 +478,7 @@ void _mi_page_retire(mi_page_t* page) mi_attr_noexcept {
   const size_t bsize = mi_page_block_size(page);
   if mi_likely( /* bsize < MI_MAX_RETIRE_SIZE && */ !mi_page_queue_is_special(pq)) {  // not full or huge queue?
     if (pq->last==page && pq->first==page) { // the only page in the queue?
-      mi_stat_counter_increase(_mi_stats_main.page_no_retire,1);
+      mi_stat_counter_increase(_mi_stats_main.pages_retire,1);
       page->retire_expire = (bsize <= MI_SMALL_OBJ_SIZE_MAX ? MI_RETIRE_CYCLES : MI_RETIRE_CYCLES/4);
       mi_heap_t* heap = mi_page_heap(page);
       mi_assert_internal(pq >= heap->pages);
@@ -620,7 +625,7 @@ static mi_decl_noinline void mi_page_free_list_extend( mi_page_t* const page, co
 #if (MI_SECURE>0)
 #define MI_MIN_EXTEND         (8*MI_SECURE) // extend at least by this many
 #else
-#define MI_MIN_EXTEND         (1)
+#define MI_MIN_EXTEND         (4)
 #endif
 
 // Extend the capacity (up to reserved) by initializing a free list
@@ -629,6 +634,7 @@ static mi_decl_noinline void mi_page_free_list_extend( mi_page_t* const page, co
 // allocations but this did not speed up any benchmark (due to an
 // extra test in malloc? or cache effects?)
 static void mi_page_extend_free(mi_heap_t* heap, mi_page_t* page, mi_tld_t* tld) {
+  MI_UNUSED(tld);
   mi_assert_expensive(mi_page_is_valid_init(page));
   #if (MI_SECURE<=2)
   mi_assert(page->free == NULL);
@@ -637,9 +643,6 @@ static void mi_page_extend_free(mi_heap_t* heap, mi_page_t* page, mi_tld_t* tld)
   #endif
   if (page->capacity >= page->reserved) return;
 
-  size_t page_size;
-  //uint8_t* page_start =
-  _mi_segment_page_start(_mi_page_segment(page), page, &page_size);
   mi_stat_counter_increase(tld->stats.pages_extended, 1);
 
   // calculate the extend count
@@ -685,6 +688,8 @@ static void mi_page_init(mi_heap_t* heap, mi_page_t* page, size_t block_size, mi
   size_t page_size;
   page->page_start = _mi_segment_page_start(segment, page, &page_size);
   mi_track_mem_noaccess(page->page_start,page_size);
+  mi_assert_internal(mi_page_block_size(page) <= page_size);
+  mi_assert_internal(page_size <= page->slice_count*MI_SEGMENT_SLICE_SIZE);
   mi_assert_internal(page_size / block_size < (1L<<16));
   page->reserved = (uint16_t)(page_size / block_size);
   mi_assert_internal(page->reserved > 0);
@@ -699,6 +704,7 @@ static void mi_page_init(mi_heap_t* heap, mi_page_t* page, size_t block_size, mi
     mi_assert_expensive(mi_mem_is_zero(page->page_start, page_size));
   }
   #endif
+  mi_assert_internal(page->is_committed);
   if (block_size > 0 && _mi_is_power_of_two(block_size)) {
     page->block_size_shift = (uint8_t)(mi_ctz((uintptr_t)block_size));
   }
@@ -809,7 +815,7 @@ static mi_page_t* mi_page_queue_find_free_ex(mi_heap_t* heap, mi_page_queue_t* p
     page = next;
   } // for each page
 
-  mi_heap_stat_counter_increase(heap, searches, count);
+  mi_heap_stat_counter_increase(heap, page_searches, count);
 
   // set the page to the best candidate
   if (page_candidate != NULL) {
@@ -821,7 +827,7 @@ static mi_page_t* mi_page_queue_find_free_ex(mi_heap_t* heap, mi_page_queue_t* p
   }
 
   if (page == NULL) {
-    _mi_heap_collect_retired(heap, false); // perhaps make a page available
+    _mi_heap_collect_retired(heap, false); // perhaps make a page available?
     page = mi_page_fresh(heap, pq);
     if (page == NULL && first_try) {
       // out-of-memory _or_ an abandoned page with free blocks was reclaimed, try once again
@@ -899,31 +905,47 @@ void mi_register_deferred_free(mi_deferred_free_fun* fn, void* arg) mi_attr_noex
   General allocation
 ----------------------------------------------------------- */
 
-// Huge pages contain just one block, and the segment contains just that page.
+// Large and huge page allocation.
+// Huge pages contain just one block, and the segment contains just that page (as `MI_SEGMENT_HUGE`).
 // Huge pages are also use if the requested alignment is very large (> MI_BLOCK_ALIGNMENT_MAX)
 // so their size is not always `> MI_LARGE_OBJ_SIZE_MAX`.
-static mi_page_t* mi_huge_page_alloc(mi_heap_t* heap, size_t size, size_t page_alignment) {
+static mi_page_t* mi_large_huge_page_alloc(mi_heap_t* heap, size_t size, size_t page_alignment) {
   size_t block_size = _mi_os_good_alloc_size(size);
   mi_assert_internal(mi_bin(block_size) == MI_BIN_HUGE || page_alignment > 0);
+  bool is_huge = (block_size > MI_LARGE_OBJ_SIZE_MAX || page_alignment > 0);
   #if MI_HUGE_PAGE_ABANDON
-  mi_page_queue_t* pq = NULL;
+  mi_page_queue_t* pq = (is_huge ? NULL : mi_page_queue(heap, block_size));
   #else
-  mi_page_queue_t* pq = mi_page_queue(heap, MI_LARGE_OBJ_SIZE_MAX+1);  // always in the huge queue regardless of the block size
-  mi_assert_internal(mi_page_queue_is_huge(pq));
+  mi_page_queue_t* pq = mi_page_queue(heap, is_huge ? MI_LARGE_OBJ_SIZE_MAX+1 : block_size);
+  mi_assert_internal(!is_huge || mi_page_queue_is_huge(pq));
   #endif
   mi_page_t* page = mi_page_fresh_alloc(heap, pq, block_size, page_alignment);
   if (page != NULL) {
-    mi_assert_internal(mi_page_block_size(page) >= size);
     mi_assert_internal(mi_page_immediate_available(page));
-    mi_assert_internal(mi_page_is_huge(page));
-    mi_assert_internal(_mi_page_segment(page)->page_kind == MI_PAGE_HUGE);
-    mi_assert_internal(_mi_page_segment(page)->used==1);
-    #if MI_HUGE_PAGE_ABANDON
-    mi_assert_internal(_mi_page_segment(page)->thread_id==0); // abandoned, not in the huge queue
-    mi_page_set_heap(page, NULL);
-    #endif
-    mi_heap_stat_increase(heap, huge, mi_page_block_size(page));
-    mi_heap_stat_counter_increase(heap, huge_count, 1);
+
+    if (is_huge) {
+      mi_assert_internal(mi_page_is_huge(page));
+      mi_assert_internal(_mi_page_segment(page)->kind == MI_SEGMENT_HUGE);
+      mi_assert_internal(_mi_page_segment(page)->used==1);
+      #if MI_HUGE_PAGE_ABANDON
+      mi_assert_internal(_mi_page_segment(page)->thread_id==0); // abandoned, not in the huge queue
+      mi_page_set_heap(page, NULL);
+      #endif
+    }
+    else {
+      mi_assert_internal(!mi_page_is_huge(page));
+    }
+
+    const size_t bsize = mi_page_usable_block_size(page);  // note: not `mi_page_block_size` to account for padding
+    /*if (bsize <= MI_LARGE_OBJ_SIZE_MAX) {
+      mi_heap_stat_increase(heap, malloc_large, bsize);
+      mi_heap_stat_counter_increase(heap, malloc_large_count, 1);
+    }
+    else */
+    {
+      _mi_stat_increase(&heap->tld->stats.malloc_huge, bsize);
+      _mi_stat_counter_increase(&heap->tld->stats.malloc_huge_count, 1);
+    }
   }
   return page;
 }
@@ -934,13 +956,13 @@ static mi_page_t* mi_huge_page_alloc(mi_heap_t* heap, size_t size, size_t page_a
 static mi_page_t* mi_find_page(mi_heap_t* heap, size_t size, size_t huge_alignment) mi_attr_noexcept {
   // huge allocation?
   const size_t req_size = size - MI_PADDING_SIZE;  // correct for padding_size in case of an overflow on `size`
-  if mi_unlikely(req_size > (MI_LARGE_OBJ_SIZE_MAX - MI_PADDING_SIZE) || huge_alignment > 0) {
+  if mi_unlikely(req_size > (MI_MEDIUM_OBJ_SIZE_MAX - MI_PADDING_SIZE) || huge_alignment > 0) {
     if mi_unlikely(req_size > MI_MAX_ALLOC_SIZE) {
       _mi_error_message(EOVERFLOW, "allocation request is too large (%zu bytes)\n", req_size);
       return NULL;
     }
     else {
-      return mi_huge_page_alloc(heap,size,huge_alignment);
+      return mi_large_huge_page_alloc(heap,size,huge_alignment);
     }
   }
   else {
@@ -967,11 +989,23 @@ void* _mi_malloc_generic(mi_heap_t* heap, size_t size, bool zero, size_t huge_al
   }
   mi_assert_internal(mi_heap_is_initialized(heap));
 
-  // call potential deferred free routines
-  _mi_deferred_free(heap, false);
+  // do administrative tasks every N generic mallocs
+  if mi_unlikely(++heap->generic_count >= 100) {
+    heap->generic_collect_count += heap->generic_count;
+    heap->generic_count = 0;
+    // call potential deferred free routines
+    _mi_deferred_free(heap, false);
 
-  // free delayed frees from other threads (but skip contended ones)
-  _mi_heap_delayed_free_partial(heap);
+    // free delayed frees from other threads (but skip contended ones)
+    _mi_heap_delayed_free_partial(heap);
+    
+    // collect every once in a while (10000 by default)
+    const long generic_collect = mi_option_get_clamp(mi_option_generic_collect, 1, 1000000L);    
+    if (heap->generic_collect_count >= generic_collect) {
+      heap->generic_collect_count = 0;
+      mi_heap_collect(heap, false /* force? */);
+    }
+  }
 
   // find (or allocate) a page of the right size
   mi_page_t* page = mi_find_page(heap, size, huge_alignment);
@@ -990,14 +1024,20 @@ void* _mi_malloc_generic(mi_heap_t* heap, size_t size, bool zero, size_t huge_al
   mi_assert_internal(mi_page_block_size(page) >= size);
 
   // and try again, this time succeeding! (i.e. this should never recurse through _mi_page_malloc)
+  void* p;
   if mi_unlikely(zero && mi_page_is_huge(page)) {
     // note: we cannot call _mi_page_malloc with zeroing for huge blocks; we zero it afterwards in that case.
-    void* p = _mi_page_malloc(heap, page, size);
+    p = _mi_page_malloc(heap, page, size);
     mi_assert_internal(p != NULL);
     _mi_memzero_aligned(p, mi_page_usable_block_size(page));
-    return p;
   }
   else {
-    return _mi_page_malloc_zero(heap, page, size, zero);
+    p = _mi_page_malloc_zero(heap, page, size, zero);
+    mi_assert_internal(p != NULL);
   }
+  // move singleton pages to the full queue
+  if (page->reserved == page->used) {
+    mi_page_to_full(page, mi_page_queue_of(page));
+  }
+  return p;
 }

src/prim/unix/prim.c

@@ -31,11 +31,12 @@ terms of the MIT license. A copy of the license can be found in the file
 
 #if defined(__linux__)
   #include <features.h>
+  #include <linux/prctl.h>  // PR_SET_VMA
   //#if defined(MI_NO_THP)
-  #include <sys/prctl.h>  // THP disable
+  #include <sys/prctl.h>    // THP disable
   //#endif
   #if defined(__GLIBC__)
-  #include <linux/mman.h> // linux mmap flags
+  #include <linux/mman.h>   // linux mmap flags
   #else
   #include <sys/mman.h>
   #endif
@@ -57,12 +58,19 @@ terms of the MIT license. A copy of the license can be found in the file
   #include <sys/sysctl.h>
 #endif
 
-#if defined(__linux__) || defined(__FreeBSD__)
+#if (defined(__linux__) && !defined(__ANDROID__)) || defined(__FreeBSD__)
   #define MI_HAS_SYSCALL_H
   #include <sys/syscall.h>
 #endif
 
+#if !defined(MADV_DONTNEED) && defined(POSIX_MADV_DONTNEED)  // QNX
+#define MADV_DONTNEED  POSIX_MADV_DONTNEED
+#endif
+#if !defined(MADV_FREE) && defined(POSIX_MADV_FREE)  // QNX
+#define MADV_FREE  POSIX_MADV_FREE
+#endif
 
+  
 //------------------------------------------------------------------------------------
 // Use syscalls for some primitives to allow for libraries that override open/read/close etc.
 // and do allocation themselves; using syscalls prevents recursion when mimalloc is
@@ -142,8 +150,9 @@ void _mi_prim_mem_init( mi_os_mem_config_t* config )
     config->alloc_granularity = (size_t)psize;
     #if defined(_SC_PHYS_PAGES)
     long pphys = sysconf(_SC_PHYS_PAGES);
-    if (pphys > 0 && (size_t)pphys < (SIZE_MAX/(size_t)psize)) {
-      config->physical_memory = (size_t)pphys * (size_t)psize;
+    const size_t psize_in_kib = (size_t)psize / MI_KiB;
+    if (psize_in_kib > 0 && pphys > 0 && (size_t)pphys <= (SIZE_MAX/psize_in_kib)) {
+      config->physical_memory_in_kib = (size_t)pphys * psize_in_kib;
     }
     #endif
   }
@@ -189,20 +198,32 @@ int _mi_prim_free(void* addr, size_t size ) {
 static int unix_madvise(void* addr, size_t size, int advice) {
   #if defined(__sun)
   int res = madvise((caddr_t)addr, size, advice);  // Solaris needs cast (issue #520)
+  #elif defined(__QNX__)
+  int res = posix_madvise(addr, size, advice);
   #else
   int res = madvise(addr, size, advice);
   #endif
   return (res==0 ? 0 : errno);
 }
 
-static void* unix_mmap_prim(void* addr, size_t size, size_t try_alignment, int protect_flags, int flags, int fd) {
+static void* unix_mmap_prim(void* addr, size_t size, int protect_flags, int flags, int fd) {
+  void* p = mmap(addr, size, protect_flags, flags, fd, 0 /* offset */);
+  #if (defined(__linux__) && defined(PR_SET_VMA))
+  if (p!=MAP_FAILED && p!=NULL) {
+    prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, p, size, "mimalloc");
+  }
+  #endif
+  return p;
+}
+
+static void* unix_mmap_prim_aligned(void* addr, size_t size, size_t try_alignment, int protect_flags, int flags, int fd) {
   MI_UNUSED(try_alignment);
   void* p = NULL;
   #if defined(MAP_ALIGNED)  // BSD
   if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) {
     size_t n = mi_bsr(try_alignment);
     if (((size_t)1 << n) == try_alignment && n >= 12 && n <= 30) {  // alignment is a power of 2 and 4096 <= alignment <= 1GiB
-      p = mmap(addr, size, protect_flags, flags | MAP_ALIGNED(n), fd, 0);
+      p = unix_mmap_prim(addr, size, protect_flags, flags | MAP_ALIGNED(n), fd);
       if (p==MAP_FAILED || !_mi_is_aligned(p,try_alignment)) {
         int err = errno;
         _mi_trace_message("unable to directly request aligned OS memory (error: %d (0x%x), size: 0x%zx bytes, alignment: 0x%zx, hint address: %p)\n", err, err, size, try_alignment, addr);
@@ -213,7 +234,7 @@ static void* unix_mmap_prim(void* addr, size_t size, size_t try_alignment, int p
   }
   #elif defined(MAP_ALIGN)  // Solaris
   if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) {
-    p = mmap((void*)try_alignment, size, protect_flags, flags | MAP_ALIGN, fd, 0);  // addr parameter is the required alignment
+    p = unix_mmap_prim((void*)try_alignment, size, protect_flags, flags | MAP_ALIGN, fd);  // addr parameter is the required alignment
     if (p!=MAP_FAILED) return p;
     // fall back to regular mmap
   }
@@ -223,7 +244,7 @@ static void* unix_mmap_prim(void* addr, size_t size, size_t try_alignment, int p
   if (addr == NULL) {
     void* hint = _mi_os_get_aligned_hint(try_alignment, size);
     if (hint != NULL) {
-      p = mmap(hint, size, protect_flags, flags, fd, 0);
+      p = unix_mmap_prim(hint, size, protect_flags, flags, fd);
       if (p==MAP_FAILED || !_mi_is_aligned(p,try_alignment)) {
         #if MI_TRACK_ENABLED  // asan sometimes does not instrument errno correctly?
         int err = 0;
@@ -238,7 +259,7 @@ static void* unix_mmap_prim(void* addr, size_t size, size_t try_alignment, int p
   }
   #endif
   // regular mmap
-  p = mmap(addr, size, protect_flags, flags, fd, 0);
+  p = unix_mmap_prim(addr, size, protect_flags, flags, fd);
   if (p!=MAP_FAILED) return p;
   // failed to allocate
   return NULL;
@@ -309,7 +330,7 @@ static void* unix_mmap(void* addr, size_t size, size_t try_alignment, int protec
       if (large_only || lflags != flags) {
         // try large OS page allocation
         *is_large = true;
-        p = unix_mmap_prim(addr, size, try_alignment, protect_flags, lflags, lfd);
+        p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, lflags, lfd);
         #ifdef MAP_HUGE_1GB
         if (p == NULL && (lflags & MAP_HUGE_1GB) == MAP_HUGE_1GB) {
           mi_huge_pages_available = false; // don't try huge 1GiB pages again
@@ -317,7 +338,7 @@ static void* unix_mmap(void* addr, size_t size, size_t try_alignment, int protec
             _mi_warning_message("unable to allocate huge (1GiB) page, trying large (2MiB) pages instead (errno: %i)\n", errno);
           }
           lflags = ((lflags & ~MAP_HUGE_1GB) | MAP_HUGE_2MB);
-          p = unix_mmap_prim(addr, size, try_alignment, protect_flags, lflags, lfd);
+          p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, lflags, lfd);
         }
         #endif
         if (large_only) return p;
@@ -330,7 +351,7 @@ static void* unix_mmap(void* addr, size_t size, size_t try_alignment, int protec
   // regular allocation
   if (p == NULL) {
     *is_large = false;
-    p = unix_mmap_prim(addr, size, try_alignment, protect_flags, flags, fd);
+    p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, flags, fd);
     if (p != NULL) {
       #if defined(MADV_HUGEPAGE)
       // Many Linux systems don't allow MAP_HUGETLB but they support instead
@@ -388,10 +409,6 @@ static void unix_mprotect_hint(int err) {
   #endif
 }
 
-
-
-
-
 int _mi_prim_commit(void* start, size_t size, bool* is_zero) {
   // commit: ensure we can access the area
   // note: we may think that *is_zero can be true since the memory

src/prim/windows/prim.c

@@ -62,6 +62,9 @@ static PGetNumaProcessorNodeEx      pGetNumaProcessorNodeEx = NULL;
 static PGetNumaNodeProcessorMaskEx  pGetNumaNodeProcessorMaskEx = NULL;
 static PGetNumaProcessorNode        pGetNumaProcessorNode = NULL;
 
+// Available after Windows XP
+typedef BOOL (__stdcall *PGetPhysicallyInstalledSystemMemory)( PULONGLONG TotalMemoryInKilobytes );
+
 //---------------------------------------------
 // Enable large page support dynamically (if possible)
 //---------------------------------------------
@@ -120,16 +123,10 @@ void _mi_prim_mem_init( mi_os_mem_config_t* config )
   if (si.dwAllocationGranularity > 0) { config->alloc_granularity = si.dwAllocationGranularity; }
   // get virtual address bits
   if ((uintptr_t)si.lpMaximumApplicationAddress > 0) {
-    const size_t vbits = MI_INTPTR_BITS - mi_clz((uintptr_t)si.lpMaximumApplicationAddress);
+    const size_t vbits = MI_SIZE_BITS - mi_clz((uintptr_t)si.lpMaximumApplicationAddress);
     config->virtual_address_bits = vbits;
   }
-  // get physical memory
-  ULONGLONG memInKiB = 0;
-  if (GetPhysicallyInstalledSystemMemory(&memInKiB)) {
-    if (memInKiB > 0 && memInKiB < (SIZE_MAX / MI_KiB)) {
-      config->physical_memory = memInKiB * MI_KiB;
-    }
-  }
+
   // get the VirtualAlloc2 function
   HINSTANCE  hDll;
   hDll = LoadLibrary(TEXT("kernelbase.dll"));
@@ -152,8 +149,19 @@ void _mi_prim_mem_init( mi_os_mem_config_t* config )
     pGetNumaProcessorNodeEx = (PGetNumaProcessorNodeEx)(void (*)(void))GetProcAddress(hDll, "GetNumaProcessorNodeEx");
     pGetNumaNodeProcessorMaskEx = (PGetNumaNodeProcessorMaskEx)(void (*)(void))GetProcAddress(hDll, "GetNumaNodeProcessorMaskEx");
     pGetNumaProcessorNode = (PGetNumaProcessorNode)(void (*)(void))GetProcAddress(hDll, "GetNumaProcessorNode");
+    // Get physical memory (not available on XP, so check dynamically)
+    PGetPhysicallyInstalledSystemMemory pGetPhysicallyInstalledSystemMemory = (PGetPhysicallyInstalledSystemMemory)(void (*)(void))GetProcAddress(hDll,"GetPhysicallyInstalledSystemMemory");
+    if (pGetPhysicallyInstalledSystemMemory != NULL) {
+      ULONGLONG memInKiB = 0;
+      if ((*pGetPhysicallyInstalledSystemMemory)(&memInKiB)) {
+        if (memInKiB > 0 && memInKiB <= SIZE_MAX) {
+          config->physical_memory_in_kib = (size_t)memInKiB;
+        }
+      }
+    }
     FreeLibrary(hDll);
   }
+  // Enable large/huge OS page support?
   if (mi_option_is_enabled(mi_option_allow_large_os_pages) || mi_option_is_enabled(mi_option_reserve_huge_os_pages)) {
     win_enable_large_os_pages(&config->large_page_size);
   }
@@ -173,7 +181,7 @@ int _mi_prim_free(void* addr, size_t size ) {
     // In mi_os_mem_alloc_aligned the fallback path may have returned a pointer inside
     // the memory region returned by VirtualAlloc; in that case we need to free using
     // the start of the region.
-    MEMORY_BASIC_INFORMATION info = { 0 };
+    MEMORY_BASIC_INFORMATION info; _mi_memzero_var(info);
     VirtualQuery(addr, &info, sizeof(info));
     if (info.AllocationBase < addr && ((uint8_t*)addr - (uint8_t*)info.AllocationBase) < (ptrdiff_t)MI_SEGMENT_SIZE) {
       errcode = 0;
@@ -635,7 +643,7 @@ static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) {
   }
   else if (reason==DLL_THREAD_DETACH && !_mi_is_redirected()) {
     _mi_thread_done(NULL);
-  }    
+  }
 }
 
 
@@ -643,7 +651,7 @@ static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) {
   #define MI_PRIM_HAS_PROCESS_ATTACH  1
 
   // Windows DLL: easy to hook into process_init and thread_done
-  __declspec(dllexport) BOOL WINAPI DllMain(HINSTANCE inst, DWORD reason, LPVOID reserved) {
+  BOOL WINAPI DllMain(HINSTANCE inst, DWORD reason, LPVOID reserved) {
     mi_win_main((PVOID)inst,reason,reserved);
     return TRUE;
   }

src/random.c

@@ -143,13 +143,17 @@ void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* ctx_new) {
 
 uintptr_t _mi_random_next(mi_random_ctx_t* ctx) {
   mi_assert_internal(mi_random_is_initialized(ctx));
-  #if MI_INTPTR_SIZE <= 4
-    return chacha_next32(ctx);
-  #elif MI_INTPTR_SIZE == 8
-    return (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx));
-  #else
-  # error "define mi_random_next for this platform"
-  #endif
+  uintptr_t r;
+  do {
+    #if MI_INTPTR_SIZE <= 4
+    r = chacha_next32(ctx);
+    #elif MI_INTPTR_SIZE == 8
+    r = (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx));
+    #else
+    # error "define mi_random_next for this platform"
+    #endif
+  } while (r==0);
+  return r;
 }
 
 
@@ -163,7 +167,7 @@ uintptr_t _mi_os_random_weak(uintptr_t extra_seed) {
   x ^= _mi_prim_clock_now();  
   // and do a few randomization steps
   uintptr_t max = ((x ^ (x >> 17)) & 0x0F) + 1;
-  for (uintptr_t i = 0; i < max; i++) {
+  for (uintptr_t i = 0; i < max || x==0; i++, x++) {
     x = _mi_random_shuffle(x);
   }
   mi_assert_internal(x != 0);
@@ -179,7 +183,7 @@ static void mi_random_init_ex(mi_random_ctx_t* ctx, bool use_weak) {
     if (!use_weak) { _mi_warning_message("unable to use secure randomness\n"); }
     #endif
     uintptr_t x = _mi_os_random_weak(0);
-    for (size_t i = 0; i < 8; i++) {  // key is eight 32-bit words.
+    for (size_t i = 0; i < 8; i++, x++) {  // key is eight 32-bit words.
       x = _mi_random_shuffle(x);
       ((uint32_t*)key)[i] = (uint32_t)x;
     }

src/segment-map.c

@@ -28,8 +28,14 @@ terms of the MIT license. A copy of the license can be found in the file
 #define MI_SEGMENT_MAP_PART_SIZE      (MI_INTPTR_SIZE*MI_KiB - 128)      // 128 > sizeof(mi_memid_t) ! 
 #define MI_SEGMENT_MAP_PART_BITS      (8*MI_SEGMENT_MAP_PART_SIZE)
 #define MI_SEGMENT_MAP_PART_ENTRIES   (MI_SEGMENT_MAP_PART_SIZE / MI_INTPTR_SIZE)
-#define MI_SEGMENT_MAP_PART_BIT_SPAN  (MI_SEGMENT_ALIGN)
+#define MI_SEGMENT_MAP_PART_BIT_SPAN  (MI_SEGMENT_ALIGN)                 // memory area covered by 1 bit
+
+#if (MI_SEGMENT_MAP_PART_BITS < (MI_SEGMENT_MAP_MAX_ADDRESS / MI_SEGMENT_MAP_PART_BIT_SPAN)) // prevent overflow on 32-bit (issue #1017)
 #define MI_SEGMENT_MAP_PART_SPAN      (MI_SEGMENT_MAP_PART_BITS * MI_SEGMENT_MAP_PART_BIT_SPAN)
+#else
+#define MI_SEGMENT_MAP_PART_SPAN      MI_SEGMENT_MAP_MAX_ADDRESS
+#endif
+
 #define MI_SEGMENT_MAP_MAX_PARTS      ((MI_SEGMENT_MAP_MAX_ADDRESS / MI_SEGMENT_MAP_PART_SPAN) + 1)
 
 // A part of the segment map.
@@ -52,7 +58,7 @@ static mi_segmap_part_t* mi_segment_map_index_of(const mi_segment_t* segment, bo
   mi_segmap_part_t* part = mi_atomic_load_ptr_relaxed(mi_segmap_part_t, &mi_segment_map[segindex]);
 
   // allocate on demand to reduce .bss footprint
-  if (part == NULL) {
+  if mi_unlikely(part == NULL) {
     if (!create_on_demand) return NULL;
     mi_memid_t memid;
     part = (mi_segmap_part_t*)_mi_os_alloc(sizeof(mi_segmap_part_t), &memid);

src/stats.c

@@ -30,55 +30,25 @@ static void mi_stat_update(mi_stat_count_t* stat, int64_t amount) {
   {
     // add atomically (for abandoned pages)
     int64_t current = mi_atomic_addi64_relaxed(&stat->current, amount);
+    // if (stat == &_mi_stats_main.committed) { mi_assert_internal(current + amount >= 0); };
     mi_atomic_maxi64_relaxed(&stat->peak, current + amount);
     if (amount > 0) {
-      mi_atomic_addi64_relaxed(&stat->allocated,amount);
-    }
-    else {
-      mi_atomic_addi64_relaxed(&stat->freed, -amount);
+      mi_atomic_addi64_relaxed(&stat->total,amount);
     }
   }
   else {
     // add thread local
     stat->current += amount;
-    if (stat->current > stat->peak) stat->peak = stat->current;
-    if (amount > 0) {
-      stat->allocated += amount;
-    }
-    else {
-      stat->freed += -amount;
-    }
-  }
-}
-
-// Adjust stats to compensate; for example before committing a range,
-// first adjust downwards with parts that were already committed so 
-// we avoid double counting.
-static void mi_stat_adjust(mi_stat_count_t* stat, int64_t amount) {
-  if (amount == 0) return;
-  if mi_unlikely(mi_is_in_main(stat))
-  {
-    // adjust atomically 
-    mi_atomic_addi64_relaxed(&stat->current, amount);
-    mi_atomic_addi64_relaxed(&stat->allocated, amount);
-    mi_atomic_addi64_relaxed(&stat->freed, amount);
-  }
-  else {
-    // don't affect the peak
-    stat->current += amount;    
-    // add to both
-    stat->allocated += amount;
-    stat->freed += amount;    
+    if (stat->current > stat->peak) { stat->peak = stat->current; }
+    if (amount > 0) { stat->total += amount; }
   }
 }
 
 void _mi_stat_counter_increase(mi_stat_counter_t* stat, size_t amount) {
   if (mi_is_in_main(stat)) {
-    mi_atomic_addi64_relaxed( &stat->count, 1 );
     mi_atomic_addi64_relaxed( &stat->total, (int64_t)amount );
   }
   else {
-    stat->count++;
     stat->total += amount;
   }
 }
@@ -91,72 +61,66 @@ void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount) {
   mi_stat_update(stat, -((int64_t)amount));
 }
 
-void _mi_stat_adjust_increase(mi_stat_count_t* stat, size_t amount) {
-  mi_stat_adjust(stat, (int64_t)amount);
+
+static void mi_stat_adjust(mi_stat_count_t* stat, int64_t amount) {
+  if (amount == 0) return;
+  if mi_unlikely(mi_is_in_main(stat))
+  {
+    // adjust atomically 
+    mi_atomic_addi64_relaxed(&stat->current, amount);
+    mi_atomic_addi64_relaxed(&stat->total,amount);
+  }
+  else {
+    // adjust local
+    stat->current += amount;
+    stat->total += amount;
+  }
 }
 
 void _mi_stat_adjust_decrease(mi_stat_count_t* stat, size_t amount) {
   mi_stat_adjust(stat, -((int64_t)amount));
 }
 
+
 // must be thread safe as it is called from stats_merge
-static void mi_stat_add(mi_stat_count_t* stat, const mi_stat_count_t* src, int64_t unit) {
+static void mi_stat_count_add_mt(mi_stat_count_t* stat, const mi_stat_count_t* src) {
   if (stat==src) return;
-  if (src->allocated==0 && src->freed==0) return;
-  mi_atomic_addi64_relaxed( &stat->allocated, src->allocated * unit);
-  mi_atomic_addi64_relaxed( &stat->current, src->current * unit);
-  mi_atomic_addi64_relaxed( &stat->freed, src->freed * unit);
-  // peak scores do not work across threads..
-  mi_atomic_addi64_relaxed( &stat->peak, src->peak * unit);
+  mi_atomic_void_addi64_relaxed(&stat->total, &src->total); 
+  mi_atomic_void_addi64_relaxed(&stat->current, &src->current); 
+  // peak scores do really not work across threads .. we just add them
+  mi_atomic_void_addi64_relaxed( &stat->peak, &src->peak);
+  // or, take the max?
+  // mi_atomic_maxi64_relaxed(&stat->peak, src->peak);
 }
 
-static void mi_stat_counter_add(mi_stat_counter_t* stat, const mi_stat_counter_t* src, int64_t unit) {
+static void mi_stat_counter_add_mt(mi_stat_counter_t* stat, const mi_stat_counter_t* src) {
   if (stat==src) return;
-  mi_atomic_addi64_relaxed( &stat->total, src->total * unit);
-  mi_atomic_addi64_relaxed( &stat->count, src->count * unit);
+  mi_atomic_void_addi64_relaxed(&stat->total, &src->total);
 }
 
+#define MI_STAT_COUNT(stat)    mi_stat_count_add_mt(&stats->stat, &src->stat);
+#define MI_STAT_COUNTER(stat)  mi_stat_counter_add_mt(&stats->stat, &src->stat);
+
 // must be thread safe as it is called from stats_merge
 static void mi_stats_add(mi_stats_t* stats, const mi_stats_t* src) {
   if (stats==src) return;
-  mi_stat_add(&stats->segments, &src->segments,1);
-  mi_stat_add(&stats->pages, &src->pages,1);
-  mi_stat_add(&stats->reserved, &src->reserved, 1);
-  mi_stat_add(&stats->committed, &src->committed, 1);
-  mi_stat_add(&stats->reset, &src->reset, 1);
-  mi_stat_add(&stats->purged, &src->purged, 1);
-  mi_stat_add(&stats->page_committed, &src->page_committed, 1);
 
-  mi_stat_add(&stats->pages_abandoned, &src->pages_abandoned, 1);
-  mi_stat_add(&stats->segments_abandoned, &src->segments_abandoned, 1);
-  mi_stat_add(&stats->threads, &src->threads, 1);
+  // copy all fields
+  MI_STAT_FIELDS()
 
-  mi_stat_add(&stats->malloc, &src->malloc, 1);
-  mi_stat_add(&stats->segments_cache, &src->segments_cache, 1);
-  mi_stat_add(&stats->normal, &src->normal, 1);
-  mi_stat_add(&stats->huge, &src->huge, 1);
-  mi_stat_add(&stats->giant, &src->giant, 1);
-
-  mi_stat_counter_add(&stats->pages_extended, &src->pages_extended, 1);
-  mi_stat_counter_add(&stats->mmap_calls, &src->mmap_calls, 1);
-  mi_stat_counter_add(&stats->commit_calls, &src->commit_calls, 1);
-  mi_stat_counter_add(&stats->reset_calls, &src->reset_calls, 1);
-  mi_stat_counter_add(&stats->purge_calls, &src->purge_calls, 1);
-
-  mi_stat_counter_add(&stats->page_no_retire, &src->page_no_retire, 1);
-  mi_stat_counter_add(&stats->searches, &src->searches, 1);
-  mi_stat_counter_add(&stats->normal_count, &src->normal_count, 1);
-  mi_stat_counter_add(&stats->huge_count, &src->huge_count, 1);  
-  mi_stat_counter_add(&stats->guarded_alloc_count, &src->guarded_alloc_count, 1);
-#if MI_STAT>1
+  #if MI_STAT>1
   for (size_t i = 0; i <= MI_BIN_HUGE; i++) {
-    if (src->normal_bins[i].allocated > 0 || src->normal_bins[i].freed > 0) {
-      mi_stat_add(&stats->normal_bins[i], &src->normal_bins[i], 1);
-    }
+    mi_stat_count_add_mt(&stats->malloc_bins[i], &src->malloc_bins[i]);
+  }
+  #endif
+  for (size_t i = 0; i <= MI_BIN_HUGE; i++) {
+    mi_stat_count_add_mt(&stats->page_bins[i], &src->page_bins[i]);
   }
-#endif
 }
 
+#undef MI_STAT_COUNT
+#undef MI_STAT_COUNTER
+
 /* -----------------------------------------------------------
   Display statistics
 ----------------------------------------------------------- */
@@ -207,26 +171,26 @@ static void mi_stat_print_ex(const mi_stat_count_t* stat, const char* msg, int64
   if (unit != 0) {
     if (unit > 0) {
       mi_print_amount(stat->peak, unit, out, arg);
-      mi_print_amount(stat->allocated, unit, out, arg);
-      mi_print_amount(stat->freed, unit, out, arg);
+      mi_print_amount(stat->total, unit, out, arg);
+      // mi_print_amount(stat->freed, unit, out, arg);
       mi_print_amount(stat->current, unit, out, arg);
       mi_print_amount(unit, 1, out, arg);
-      mi_print_count(stat->allocated, unit, out, arg);
+      mi_print_count(stat->total, unit, out, arg);
     }
     else {
       mi_print_amount(stat->peak, -1, out, arg);
-      mi_print_amount(stat->allocated, -1, out, arg);
-      mi_print_amount(stat->freed, -1, out, arg);
+      mi_print_amount(stat->total, -1, out, arg);
+      // mi_print_amount(stat->freed, -1, out, arg);
       mi_print_amount(stat->current, -1, out, arg);
       if (unit == -1) {
         _mi_fprintf(out, arg, "%24s", "");
       }
       else {
         mi_print_amount(-unit, 1, out, arg);
-        mi_print_count((stat->allocated / -unit), 0, out, arg);
+        mi_print_count((stat->total / -unit), 0, out, arg);
       }
     }
-    if (stat->allocated > stat->freed) {
+    if (stat->current != 0) {
       _mi_fprintf(out, arg, "  ");
       _mi_fprintf(out, arg, (notok == NULL ? "not all freed" : notok));
       _mi_fprintf(out, arg, "\n");
@@ -237,7 +201,7 @@ static void mi_stat_print_ex(const mi_stat_count_t* stat, const char* msg, int64
   }
   else {
     mi_print_amount(stat->peak, 1, out, arg);
-    mi_print_amount(stat->allocated, 1, out, arg);
+    mi_print_amount(stat->total, 1, out, arg);
     _mi_fprintf(out, arg, "%11s", " ");  // no freed
     mi_print_amount(stat->current, 1, out, arg);
     _mi_fprintf(out, arg, "\n");
@@ -254,6 +218,13 @@ static void mi_stat_peak_print(const mi_stat_count_t* stat, const char* msg, int
   _mi_fprintf(out, arg, "\n");
 }
 
+static void mi_stat_total_print(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg) {
+  _mi_fprintf(out, arg, "%10s:", msg);
+  _mi_fprintf(out, arg, "%12s", " ");  // no peak
+  mi_print_amount(stat->total, unit, out, arg);
+  _mi_fprintf(out, arg, "\n");
+}
+
 static void mi_stat_counter_print(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg ) {
   _mi_fprintf(out, arg, "%10s:", msg);
   mi_print_amount(stat->total, -1, out, arg);
@@ -262,7 +233,7 @@ static void mi_stat_counter_print(const mi_stat_counter_t* stat, const char* msg
 
 
 static void mi_stat_counter_print_avg(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg) {
-  const int64_t avg_tens = (stat->count == 0 ? 0 : (stat->total*10 / stat->count));
+  const int64_t avg_tens = (stat->total == 0 ? 0 : (stat->total*10 / stat->total));
   const long avg_whole = (long)(avg_tens/10);
   const long avg_frac1 = (long)(avg_tens%10);
   _mi_fprintf(out, arg, "%10s: %5ld.%ld avg\n", msg, avg_whole, avg_frac1);
@@ -270,7 +241,7 @@ static void mi_stat_counter_print_avg(const mi_stat_counter_t* stat, const char*
 
 
 static void mi_print_header(mi_output_fun* out, void* arg ) {
-  _mi_fprintf(out, arg, "%10s: %11s %11s %11s %11s %11s %11s\n", "heap stats", "peak   ", "total   ", "freed   ", "current   ", "unit   ", "count   ");
+  _mi_fprintf(out, arg, "%10s: %11s %11s %11s %11s %11s\n", "heap stats", "peak   ", "total   ", "current   ", "block   ", "total#   ");
 }
 
 #if MI_STAT>1
@@ -278,7 +249,7 @@ static void mi_stats_print_bins(const mi_stat_count_t* bins, size_t max, const c
   bool found = false;
   char buf[64];
   for (size_t i = 0; i <= max; i++) {
-    if (bins[i].allocated > 0) {
+    if (bins[i].total > 0) {
       found = true;
       int64_t unit = _mi_bin_size((uint8_t)i);
       _mi_snprintf(buf, 64, "%s %3lu", fmt, (long)i);
@@ -339,42 +310,44 @@ static void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0)
   // and print using that
   mi_print_header(out,arg);
   #if MI_STAT>1
-  mi_stats_print_bins(stats->normal_bins, MI_BIN_HUGE, "normal",out,arg);
+  mi_stats_print_bins(stats->malloc_bins, MI_BIN_HUGE, "bin",out,arg);
   #endif
   #if MI_STAT
-  mi_stat_print(&stats->normal, "normal", (stats->normal_count.count == 0 ? 1 : -(stats->normal.allocated / stats->normal_count.count)), out, arg);
-  mi_stat_print(&stats->huge, "huge", (stats->huge_count.count == 0 ? 1 : -(stats->huge.allocated / stats->huge_count.count)), out, arg);  
-  mi_stat_count_t total = { 0,0,0,0 };
-  mi_stat_add(&total, &stats->normal, 1);
-  mi_stat_add(&total, &stats->huge, 1);
-  mi_stat_print(&total, "total", 1, out, arg);
+  mi_stat_print(&stats->malloc_normal, "binned", (stats->malloc_normal_count.total == 0 ? 1 : -1), out, arg);
+  // mi_stat_print(&stats->malloc_large, "large", (stats->malloc_large_count.total == 0 ? 1 : -1), out, arg);
+  mi_stat_print(&stats->malloc_huge, "huge", (stats->malloc_huge_count.total == 0 ? 1 : -1), out, arg);
+  mi_stat_count_t total = { 0,0,0 };
+  mi_stat_count_add_mt(&total, &stats->malloc_normal);
+  // mi_stat_count_add(&total, &stats->malloc_large);
+  mi_stat_count_add_mt(&total, &stats->malloc_huge);
+  mi_stat_print_ex(&total, "total", 1, out, arg, "");
   #endif
   #if MI_STAT>1
-  mi_stat_print(&stats->malloc, "malloc req", 1, out, arg);
+  mi_stat_total_print(&stats->malloc_requested, "malloc req", 1, out, arg);
   _mi_fprintf(out, arg, "\n");
   #endif
   mi_stat_print_ex(&stats->reserved, "reserved", 1, out, arg, "");
   mi_stat_print_ex(&stats->committed, "committed", 1, out, arg, "");
   mi_stat_peak_print(&stats->reset, "reset", 1, out, arg );
   mi_stat_peak_print(&stats->purged, "purged", 1, out, arg );
-  mi_stat_print(&stats->page_committed, "touched", 1, out, arg);
+  mi_stat_print_ex(&stats->page_committed, "touched", 1, out, arg, "");
   mi_stat_print(&stats->segments, "segments", -1, out, arg);
   mi_stat_print(&stats->segments_abandoned, "-abandoned", -1, out, arg);
   mi_stat_print(&stats->segments_cache, "-cached", -1, out, arg);
   mi_stat_print(&stats->pages, "pages", -1, out, arg);
   mi_stat_print(&stats->pages_abandoned, "-abandoned", -1, out, arg);
   mi_stat_counter_print(&stats->pages_extended, "-extended", out, arg);
-  mi_stat_counter_print(&stats->page_no_retire, "-noretire", out, arg);
+  mi_stat_counter_print(&stats->pages_retire, "-retire", out, arg);
   mi_stat_counter_print(&stats->arena_count, "arenas", out, arg);
-  mi_stat_counter_print(&stats->arena_crossover_count, "-crossover", out, arg);
+  // mi_stat_counter_print(&stats->arena_crossover_count, "-crossover", out, arg);
   mi_stat_counter_print(&stats->arena_rollback_count, "-rollback", out, arg);
   mi_stat_counter_print(&stats->mmap_calls, "mmaps", out, arg);
   mi_stat_counter_print(&stats->commit_calls, "commits", out, arg);
   mi_stat_counter_print(&stats->reset_calls, "resets", out, arg);
   mi_stat_counter_print(&stats->purge_calls, "purges", out, arg);
-  mi_stat_counter_print(&stats->guarded_alloc_count, "guarded", out, arg);
+  mi_stat_counter_print(&stats->malloc_guarded_count, "guarded", out, arg);
   mi_stat_print(&stats->threads, "threads", -1, out, arg);
-  mi_stat_counter_print_avg(&stats->searches, "searches", out, arg);
+  mi_stat_counter_print_avg(&stats->page_searches, "searches", out, arg);
   _mi_fprintf(out, arg, "%10s: %5zu\n", "numa nodes", _mi_os_numa_node_count());
 
   size_t elapsed;
@@ -483,7 +456,7 @@ mi_decl_export void mi_process_info(size_t* elapsed_msecs, size_t* user_msecs, s
   pinfo.page_faults    = 0;
 
   _mi_prim_process_info(&pinfo);
-  
+
   if (elapsed_msecs!=NULL)  *elapsed_msecs  = (pinfo.elapsed < 0 ? 0 : (pinfo.elapsed < (mi_msecs_t)PTRDIFF_MAX ? (size_t)pinfo.elapsed : PTRDIFF_MAX));
   if (user_msecs!=NULL)     *user_msecs     = (pinfo.utime < 0 ? 0 : (pinfo.utime < (mi_msecs_t)PTRDIFF_MAX ? (size_t)pinfo.utime : PTRDIFF_MAX));
   if (system_msecs!=NULL)   *system_msecs   = (pinfo.stime < 0 ? 0 : (pinfo.stime < (mi_msecs_t)PTRDIFF_MAX ? (size_t)pinfo.stime : PTRDIFF_MAX));
@@ -493,3 +466,164 @@ mi_decl_export void mi_process_info(size_t* elapsed_msecs, size_t* user_msecs, s
   if (peak_commit!=NULL)    *peak_commit    = pinfo.peak_commit;
   if (page_faults!=NULL)    *page_faults    = pinfo.page_faults;
 }
+
+
+// --------------------------------------------------------
+// Return statistics
+// --------------------------------------------------------
+
+void mi_stats_get(size_t stats_size, mi_stats_t* stats) mi_attr_noexcept {
+  if (stats == NULL || stats_size == 0) return;
+  _mi_memzero(stats, stats_size);
+  const size_t size = (stats_size > sizeof(mi_stats_t) ? sizeof(mi_stats_t) : stats_size);
+  _mi_memcpy(stats, &_mi_stats_main, size);
+  stats->version = MI_STAT_VERSION;
+}
+
+
+// --------------------------------------------------------
+// Statics in json format
+// --------------------------------------------------------
+
+typedef struct mi_heap_buf_s {
+  char*   buf;
+  size_t  size;
+  size_t  used;
+  bool    can_realloc;
+} mi_heap_buf_t;
+
+static bool mi_heap_buf_expand(mi_heap_buf_t* hbuf) {
+  if (hbuf==NULL) return false;
+  if (hbuf->buf != NULL && hbuf->size>0) {
+    hbuf->buf[hbuf->size-1] = 0;
+  }
+  if (hbuf->size > SIZE_MAX/2 || !hbuf->can_realloc) return false;
+  const size_t newsize = (hbuf->size == 0 ? 2*MI_KiB : 2*hbuf->size);
+  char* const  newbuf  = (char*)mi_rezalloc(hbuf->buf, newsize);
+  if (newbuf == NULL) return false;
+  hbuf->buf = newbuf;
+  hbuf->size = newsize;
+  return true;
+}
+
+static void mi_heap_buf_print(mi_heap_buf_t* hbuf, const char* msg) {
+  if (msg==NULL || hbuf==NULL) return;
+  if (hbuf->used + 1 >= hbuf->size && !hbuf->can_realloc) return;
+  for (const char* src = msg; *src != 0; src++) {
+    char c = *src;
+    if (hbuf->used + 1 >= hbuf->size) {
+      if (!mi_heap_buf_expand(hbuf)) return;
+    }
+    mi_assert_internal(hbuf->used < hbuf->size);
+    hbuf->buf[hbuf->used++] = c;
+  }
+  mi_assert_internal(hbuf->used < hbuf->size);
+  hbuf->buf[hbuf->used] = 0;
+}
+
+static void mi_heap_buf_print_count_bin(mi_heap_buf_t* hbuf, const char* prefix, mi_stat_count_t* stat, size_t bin, bool add_comma) {
+  const size_t binsize = _mi_bin_size(bin);
+  const size_t pagesize = (binsize <= MI_SMALL_OBJ_SIZE_MAX ? MI_SMALL_PAGE_SIZE :
+                            (binsize <= MI_MEDIUM_OBJ_SIZE_MAX ? MI_MEDIUM_PAGE_SIZE :
+                              #if MI_LARGE_PAGE_SIZE
+                              (binsize <= MI_LARGE_OBJ_SIZE_MAX ? MI_LARGE_PAGE_SIZE : 0)
+                              #else
+                              0
+                              #endif
+                              ));
+  char buf[128];
+  _mi_snprintf(buf, 128, "%s{ \"total\": %lld, \"peak\": %lld, \"current\": %lld, \"block_size\": %zu, \"page_size\": %zu }%s\n", prefix, stat->total, stat->peak, stat->current, binsize, pagesize, (add_comma ? "," : ""));
+  buf[127] = 0;
+  mi_heap_buf_print(hbuf, buf);
+}
+
+static void mi_heap_buf_print_count(mi_heap_buf_t* hbuf, const char* prefix, mi_stat_count_t* stat, bool add_comma) {
+  char buf[128];
+  _mi_snprintf(buf, 128, "%s{ \"total\": %lld, \"peak\": %lld, \"current\": %lld }%s\n", prefix, stat->total, stat->peak, stat->current, (add_comma ? "," : ""));
+  buf[127] = 0;
+  mi_heap_buf_print(hbuf, buf);
+}
+
+static void mi_heap_buf_print_count_value(mi_heap_buf_t* hbuf, const char* name, mi_stat_count_t* stat) {
+  char buf[128];
+  _mi_snprintf(buf, 128, "  \"%s\": ", name);
+  buf[127] = 0;
+  mi_heap_buf_print(hbuf, buf);
+  mi_heap_buf_print_count(hbuf, "", stat, true);
+}
+
+static void mi_heap_buf_print_value(mi_heap_buf_t* hbuf, const char* name, int64_t val) {
+  char buf[128];
+  _mi_snprintf(buf, 128, "  \"%s\": %lld,\n", name, val);
+  buf[127] = 0;
+  mi_heap_buf_print(hbuf, buf);
+}
+
+static void mi_heap_buf_print_size(mi_heap_buf_t* hbuf, const char* name, size_t val, bool add_comma) {
+  char buf[128];
+  _mi_snprintf(buf, 128, "    \"%s\": %zu%s\n", name, val, (add_comma ? "," : ""));
+  buf[127] = 0;
+  mi_heap_buf_print(hbuf, buf);
+}
+
+static void mi_heap_buf_print_counter_value(mi_heap_buf_t* hbuf, const char* name, mi_stat_counter_t* stat) {
+  mi_heap_buf_print_value(hbuf, name, stat->total);
+}
+
+#define MI_STAT_COUNT(stat)    mi_heap_buf_print_count_value(&hbuf, #stat, &stats->stat);
+#define MI_STAT_COUNTER(stat)  mi_heap_buf_print_counter_value(&hbuf, #stat, &stats->stat);
+
+char* mi_stats_get_json(size_t output_size, char* output_buf) mi_attr_noexcept {
+  mi_heap_buf_t hbuf = { NULL, 0, 0, true };
+  if (output_size > 0 && output_buf != NULL) {
+    _mi_memzero(output_buf, output_size);
+    hbuf.buf = output_buf;
+    hbuf.size = output_size;
+    hbuf.can_realloc = false;
+  }
+  else {
+    if (!mi_heap_buf_expand(&hbuf)) return NULL;
+  }
+  mi_heap_buf_print(&hbuf, "{\n");
+  mi_heap_buf_print_value(&hbuf, "version", MI_STAT_VERSION);
+  mi_heap_buf_print_value(&hbuf, "mimalloc_version", MI_MALLOC_VERSION);
+
+  // process info
+  mi_heap_buf_print(&hbuf, "  \"process\": {\n");
+  size_t elapsed;
+  size_t user_time;
+  size_t sys_time;
+  size_t current_rss;
+  size_t peak_rss;
+  size_t current_commit;
+  size_t peak_commit;
+  size_t page_faults;
+  mi_process_info(&elapsed, &user_time, &sys_time, &current_rss, &peak_rss, &current_commit, &peak_commit, &page_faults);
+  mi_heap_buf_print_size(&hbuf, "elapsed_msecs", elapsed, true);
+  mi_heap_buf_print_size(&hbuf, "user_msecs", user_time, true);
+  mi_heap_buf_print_size(&hbuf, "system_msecs", sys_time, true);
+  mi_heap_buf_print_size(&hbuf, "page_faults", page_faults, true);
+  mi_heap_buf_print_size(&hbuf, "rss_current", current_rss, true);
+  mi_heap_buf_print_size(&hbuf, "rss_peak", peak_rss, true);
+  mi_heap_buf_print_size(&hbuf, "commit_current", current_commit, true);
+  mi_heap_buf_print_size(&hbuf, "commit_peak", peak_commit, false);
+  mi_heap_buf_print(&hbuf, "  },\n");
+
+  // statistics
+  mi_stats_t* stats = &_mi_stats_main;
+  MI_STAT_FIELDS()
+
+  // size bins
+  mi_heap_buf_print(&hbuf, "  \"malloc_bins\": [\n");
+  for (size_t i = 0; i <= MI_BIN_HUGE; i++) {
+    mi_heap_buf_print_count_bin(&hbuf, "    ", &stats->malloc_bins[i], i, i!=MI_BIN_HUGE);
+  }
+  mi_heap_buf_print(&hbuf, "  ],\n");
+  mi_heap_buf_print(&hbuf, "  \"page_bins\": [\n");
+  for (size_t i = 0; i <= MI_BIN_HUGE; i++) {
+    mi_heap_buf_print_count_bin(&hbuf, "    ", &stats->page_bins[i], i, i!=MI_BIN_HUGE);
+  }
+  mi_heap_buf_print(&hbuf, "  ]\n");
+  mi_heap_buf_print(&hbuf, "}\n");
+  return hbuf.buf;
+}

test/CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.0)
+cmake_minimum_required(VERSION 3.18)
 project(mimalloc-test C CXX)
 
 set(CMAKE_C_STANDARD 11)
@@ -16,10 +16,12 @@ if (NOT CMAKE_BUILD_TYPE)
 endif()
 
 # Import mimalloc (if installed)
-find_package(mimalloc 1.7 REQUIRED NO_SYSTEM_ENVIRONMENT_PATH)
+find_package(mimalloc 2.2 CONFIG REQUIRED)
 message(STATUS "Found mimalloc installed at: ${MIMALLOC_LIBRARY_DIR} (${MIMALLOC_VERSION_DIR})")
 
-# overriding with a dynamic library
+
+# link with a dynamic shared library
+# use `LD_PRELOAD` to actually override malloc/free at runtime with mimalloc
 add_executable(dynamic-override  main-override.c)
 target_link_libraries(dynamic-override PUBLIC mimalloc)
 
@@ -29,9 +31,9 @@ target_link_libraries(dynamic-override-cxx PUBLIC mimalloc)
 
 # overriding with a static object file works reliable as the symbols in the
 # object file have priority over those in library files
-add_executable(static-override-obj main-override.c ${MIMALLOC_OBJECT_DIR}/mimalloc.o)
+add_executable(static-override-obj main-override.c ${MIMALLOC_OBJECT_DIR}/mimalloc${CMAKE_C_OUTPUT_EXTENSION})
 target_include_directories(static-override-obj PUBLIC ${MIMALLOC_INCLUDE_DIR})
-target_link_libraries(static-override-obj PUBLIC pthread)
+target_link_libraries(static-override-obj PUBLIC mimalloc-static)
 
 
 # overriding with a static library works too if using the `mimalloc-override.h`

test/main-override-dep.cpp

@@ -0,0 +1,15 @@
+// Issue #981: test overriding allocation in a DLL that is compiled independent of mimalloc. 
+// This is imported by the `mimalloc-test-override` project.
+#include <string>
+#include "main-override-dep.h"
+
+std::string TestAllocInDll::GetString()
+{
+	char* test = new char[128];
+	memset(test, 0, 128);
+	const char* t = "test";
+	memcpy(test, t, 4);
+	std::string r = test;
+	delete[] test;
+	return r;
+}

test/main-override-dep.h

@@ -0,0 +1,11 @@
+#pragma once
+// Issue #981: test overriding allocation in a DLL that is compiled independent of mimalloc. 
+// This is imported by the `mimalloc-test-override` project.
+
+#include <string>
+
+class TestAllocInDll
+{
+public:
+	__declspec(dllexport) std::string GetString();
+};

test/main-override-static.c

@@ -10,7 +10,6 @@
 #include <mimalloc.h>
 #include <mimalloc-override.h>  // redefines malloc etc.
 
-static void mi_bins(void);
 
 static void double_free1();
 static void double_free2();
@@ -24,15 +23,19 @@ static void test_reserved(void);
 static void negative_stat(void);
 static void alloc_huge(void);
 static void test_heap_walk(void);
+static void test_heap_arena(void);
+static void test_align(void);
 static void test_canary_leak(void);
 static void test_manage_os_memory(void);
 // static void test_large_pages(void);
 
-
 int main() {
   mi_version();
   mi_stats_reset();
-  test_manage_os_memory();
+  
+  // mi_bins();
+
+  // test_manage_os_memory();
   // test_large_pages();
   // detect double frees and heap corruption
   // double_free1();
@@ -40,17 +43,17 @@ int main() {
   // corrupt_free();
   // block_overflow1();
   // block_overflow2();
-  test_canary_leak();
+  // test_canary_leak();
   // test_aslr();
   // invalid_free();
   // test_reserved();
   // negative_stat();
   // test_heap_walk();
   // alloc_huge();
-
-  // mi_bins();
-
-
+  // test_heap_walk();
+  // test_heap_arena();
+  // test_align();
+  
   void* p1 = malloc(78);
   void* p2 = malloc(24);
   free(p1);
@@ -66,7 +69,7 @@ int main() {
   free(p1);
   free(p2);
   free(s);
-
+  
   /* now test if override worked by allocating/freeing across the api's*/
   //p1 = mi_malloc(32);
   //free(p1);
@@ -81,6 +84,13 @@ int main() {
   return 0;
 }
 
+static void test_align() {
+  void* p = mi_malloc_aligned(256, 256);
+  if (((uintptr_t)p % 256) != 0) {
+    fprintf(stderr, "%p is not 256 alignend!\n", p);
+  }
+}
+
 static void invalid_free() {
   free((void*)0xBADBEEF);
   realloc((void*)0xBADBEEF,10);
@@ -238,6 +248,20 @@ static void test_heap_walk(void) {
   mi_heap_visit_blocks(heap, true, &test_visit, NULL);
 }
 
+static void test_heap_arena(void) {
+  mi_arena_id_t arena_id;
+  int err = mi_reserve_os_memory_ex(100 * 1024 * 1024, false /* commit */, false /* allow large */, true /* exclusive */, &arena_id);
+  if (err) abort();
+  mi_heap_t* heap = mi_heap_new_in_arena(arena_id);
+  for (int i = 0; i < 500000; i++) {
+    void* p = mi_heap_malloc(heap, 1024);
+    if (p == NULL) {
+      printf("out of memory after %d kb (expecting about 100_000kb)\n", i);
+      break;
+    }
+  }
+}
+
 static void test_canary_leak(void) {
   char* p = mi_mallocn_tp(char,23);
   for(int i = 0; i < 23; i++) {
@@ -376,31 +400,34 @@ static inline size_t _mi_wsize_from_size(size_t size) {
   return (size + sizeof(uintptr_t) - 1) / sizeof(uintptr_t);
 }
 
+// #define MI_ALIGN2W
+
 // Return the bin for a given field size.
 // Returns MI_BIN_HUGE if the size is too large.
 // We use `wsize` for the size in "machine word sizes",
 // i.e. byte size == `wsize*sizeof(void*)`.
-extern inline uint8_t _mi_bin8(size_t size) {
-  size_t wsize = _mi_wsize_from_size(size);
-  uint8_t bin;
-  if (wsize <= 1) {
+static inline size_t mi_bin(size_t wsize) {
+  // size_t wsize = _mi_wsize_from_size(size);
+  // size_t bin;
+  /*if (wsize <= 1) {
     bin = 1;
   }
+  */
 #if defined(MI_ALIGN4W)
-  else if (wsize <= 4) {
-    bin = (uint8_t)((wsize+1)&~1); // round to double word sizes
+  if (wsize <= 4) {
+    return (wsize <= 1 ? 1 : (wsize+1)&~1); // round to double word sizes
   }
 #elif defined(MI_ALIGN2W)
-  else if (wsize <= 8) {
-    bin = (uint8_t)((wsize+1)&~1); // round to double word sizes
+  if (wsize <= 8) {
+    return (wsize <= 1 ? 1 : (wsize+1)&~1); // round to double word sizes
   }
 #else
-  else if (wsize <= 8) {
-    bin = (uint8_t)wsize;
+  if (wsize <= 8) {
+    return (wsize == 0 ? 1 : wsize);
   }
 #endif
   else if (wsize > MI_LARGE_WSIZE_MAX) {
-    bin = MI_BIN_HUGE;
+    return MI_BIN_HUGE;
   }
   else {
 #if defined(MI_ALIGN4W)
@@ -408,15 +435,17 @@ extern inline uint8_t _mi_bin8(size_t size) {
 #endif
     wsize--;
     // find the highest bit
-    uint8_t b = mi_bsr32((uint32_t)wsize);
+    const size_t b = _mi_bsr(wsize);  // note: wsize != 0
     // and use the top 3 bits to determine the bin (~12.5% worst internal fragmentation).
     // - adjust with 3 because we use do not round the first 8 sizes
     //   which each get an exact bin
-    bin = ((b << 2) + (uint8_t)((wsize >> (b - 2)) & 0x03)) - 3;
+    const size_t bin = ((b << 2) + ((wsize >> (b - 2)) & 0x03)) - 3;
+    assert(bin > 0 && bin < MI_BIN_HUGE);
+    return bin;
   }
-  return bin;
 }
 
+
 static inline uint8_t _mi_bin4(size_t size) {
   size_t wsize = _mi_wsize_from_size(size);
   uint8_t bin;
@@ -472,7 +501,7 @@ static size_t _mi_binx8(size_t bsize) {
 }
 
 
-static inline size_t mi_bin(size_t wsize) {
+static inline size_t mi_binx(size_t wsize) {
   uint8_t bin;
   if (wsize <= 1) {
     bin = 1;

test/main-override.cpp

@@ -9,16 +9,11 @@
 #include <vector>
 #include <future>
 #include <iostream>
-
 #include <thread>
-//#include <mimalloc.h>
 #include <assert.h>
 
 #ifdef _WIN32
 #include <mimalloc-new-delete.h>
-#endif
-
-#ifdef _WIN32
 #include <windows.h>
 static void msleep(unsigned long msecs) { Sleep(msecs); }
 #else
@@ -32,9 +27,12 @@ static void heap_late_free();         // issue #204
 static void padding_shrink();         // issue #209
 static void various_tests();
 static void test_mt_shutdown();
+static void large_alloc(void);        // issue #363
 static void fail_aslr();              // issue #372
 static void tsan_numa_test();         // issue #414
 static void strdup_test();            // issue #445
+static void bench_alloc_large(void);  // issue #xxx
+//static void test_large_migrate(void); // issue #691
 static void heap_thread_free_huge();
 static void test_std_string();        // issue #697
 static void test_thread_local();      // issue #944
@@ -42,28 +40,38 @@ static void test_thread_local();      // issue #944
 static void test_mixed1();             // issue #942
 static void test_stl_allocators();
 
+#if x_WIN32
+#include "main-override-dep.h"
+static void test_dep();               // issue #981: test overriding in another DLL
+#else
+static void test_dep() { };
+#endif
 
 int main() {
   mi_stats_reset();  // ignore earlier allocations
   various_tests();
   test_mixed1();
-  
+
+  test_dep();
+
   //test_std_string();
   //test_thread_local();
   // heap_thread_free_huge();
   /*
-  heap_thread_free_large();
-  heap_no_delete();
-  heap_late_free();
-  padding_shrink();
-  
-  tsan_numa_test();
-  */
-  /*
-  strdup_test();
-  test_stl_allocators();
-  test_mt_shutdown();
+   heap_thread_free_huge();
+   heap_thread_free_large();
+   heap_no_delete();
+   heap_late_free();
+   padding_shrink();
+   various_tests();
+   large_alloc();
+   tsan_numa_test();
+   strdup_test();
   */
+  // test_stl_allocators();
+  // test_mt_shutdown();
+  // test_large_migrate();
+
   //fail_aslr();
   mi_stats_print(NULL);
   return 0;
@@ -109,8 +117,10 @@ static void various_tests() {
   t->~Test();
   delete[] tbuf;
 
+  #if _WIN32
   const char* ptr = ::_Getdays();  // test _base overrid
   free((void*)ptr);
+  #endif
 }
 
 class Static {
@@ -139,6 +149,16 @@ static bool test_stl_allocator1() {
 
 struct some_struct { int i; int j; double z; };
 
+
+#if x_WIN32
+static void test_dep()
+{
+  TestAllocInDll t;
+  std::string s = t.GetString();
+}
+#endif
+
+
 static bool test_stl_allocator2() {
   std::vector<some_struct, mi_stl_allocator<some_struct> > vec;
   vec.push_back(some_struct());
@@ -223,7 +243,7 @@ static void test_mixed0() {
     std::cout << "Running on " << threads.size() << " threads took " << duration
               << std::endl;
 }
-#endif 
+#endif
 
 void asd() {
   void* p = malloc(128);
@@ -343,7 +363,7 @@ static void heap_thread_free_large_worker() {
 
 static void heap_thread_free_large() {
   for (int i = 0; i < 100; i++) {
-    shared_p = mi_malloc_aligned(2*1024*1024 + 1, 8);
+    shared_p = mi_malloc_aligned(2 * 1024 * 1024 + 1, 8);
     auto t1 = std::thread(heap_thread_free_large_worker);
     t1.join();
   }
@@ -354,14 +374,13 @@ static void heap_thread_free_huge_worker() {
 }
 
 static void heap_thread_free_huge() {
-  for (int i = 0; i < 10; i++) {
+  for (int i = 0; i < 100; i++) {
     shared_p = mi_malloc(1024 * 1024 * 1024);
     auto t1 = std::thread(heap_thread_free_huge_worker);
     t1.join();
   }
 }
 
-
 static void test_mt_shutdown()
 {
   const int threads = 5;
@@ -386,9 +405,21 @@ static void test_mt_shutdown()
   std::cout << "done" << std::endl;
 }
 
+// issue #363
+using namespace std;
+
+void large_alloc(void)
+{
+  char* a = new char[1ull << 25];
+  thread th([&] {
+    delete[] a;
+    });
+  th.join();
+}
+
 // issue #372
 static void fail_aslr() {
-  size_t sz = (4ULL << 40); // 4TiB
+  size_t sz = (size_t)(4ULL << 40); // 4TiB
   void* p = malloc(sz);
   printf("pointer p: %p: area up to %p\n", p, (uint8_t*)p + sz);
   *(int*)0x5FFFFFFF000 = 0;  // should segfault
@@ -406,6 +437,36 @@ static void tsan_numa_test() {
   t1.join();
 }
 
+// issue #?
+#include <chrono>
+#include <random>
+#include <iostream>
+
+static void bench_alloc_large(void) {
+  static constexpr int kNumBuffers = 20;
+  static constexpr size_t kMinBufferSize = 5 * 1024 * 1024;
+  static constexpr size_t kMaxBufferSize = 25 * 1024 * 1024;
+  std::unique_ptr<char[]> buffers[kNumBuffers];
+
+  std::random_device rd;  (void)rd;
+  std::mt19937 gen(42); //rd());
+  std::uniform_int_distribution<> size_distribution(kMinBufferSize, kMaxBufferSize);
+  std::uniform_int_distribution<> buf_number_distribution(0, kNumBuffers - 1);
+
+  static constexpr int kNumIterations = 2000;
+  const auto start = std::chrono::steady_clock::now();
+  for (int i = 0; i < kNumIterations; ++i) {
+    int buffer_idx = buf_number_distribution(gen);
+    size_t new_size = size_distribution(gen);
+    buffers[buffer_idx] = std::make_unique<char[]>(new_size);
+  }
+  const auto end = std::chrono::steady_clock::now();
+  const auto num_ms = std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count();
+  const auto us_per_allocation = std::chrono::duration_cast<std::chrono::microseconds>(end - start).count() / kNumIterations;
+  std::cout << kNumIterations << " allocations Done in " << num_ms << "ms." << std::endl;
+  std::cout << "Avg " << us_per_allocation << " us per allocation" << std::endl;
+}
+
 
 class MTest
 {
@@ -432,4 +493,4 @@ void test_thread_local()
         mi_stats_print(NULL);
     }
     return;
-}
+}

test/test-api.c

@@ -203,7 +203,11 @@ int main(void) {
   CHECK_BODY("malloc-aligned9") { // test large alignments
     bool ok = true;
     void* p[8];
-    size_t sizes[8] = { 8, 512, 1024 * 1024, MI_BLOCK_ALIGNMENT_MAX, MI_BLOCK_ALIGNMENT_MAX + 1, 2 * MI_BLOCK_ALIGNMENT_MAX, 8 * MI_BLOCK_ALIGNMENT_MAX, 0 };
+    size_t sizes[8] = { 8, 512, 1024 * 1024, MI_BLOCK_ALIGNMENT_MAX, MI_BLOCK_ALIGNMENT_MAX + 1, 
+      #if SIZE_MAX > UINT32_MAX
+      2 * MI_BLOCK_ALIGNMENT_MAX, 8 * MI_BLOCK_ALIGNMENT_MAX, 
+      #endif
+      0 };
     for (int i = 0; i < 28 && ok; i++) {
       int align = (1 << i);
       for (int j = 0; j < 8 && ok; j++) {

test/test-stress.c

@@ -61,6 +61,7 @@ static bool   main_participates = false;       // main thread participates as a
 #define custom_free(p)        free(p)
 #else
 #include <mimalloc.h>
+#include <mimalloc-stats.h>
 #define custom_calloc(n,s)    mi_calloc(n,s)
 #define custom_realloc(p,s)   mi_realloc(p,s)
 #define custom_free(p)        mi_free(p)
@@ -116,7 +117,7 @@ static void* alloc_items(size_t items, random_t r) {
     else if (chance(10, r) && allow_large_objects) items *= 1000;  // 0.1% huge
     else items *= 100;                                             // 1% large objects;
   }
-  if (items == 40) items++;              // pthreads uses that size for stack increases
+  if (items>=32 && items<=40) items*=2;              // pthreads uses 320b allocations (this shows that more clearly in the stats)
   if (use_one_size > 0) items = (use_one_size / sizeof(uintptr_t));
   if (items==0) items = 1;
   uintptr_t* p = (uintptr_t*)custom_calloc(items,sizeof(uintptr_t));
@@ -268,16 +269,23 @@ static void test_leak(void) {
 }
 #endif
 
+#if defined(USE_STD_MALLOC) && defined(MI_LINK_VERSION)
+#ifdef __cplusplus
+extern "C"
+#endif
+int mi_version(void);
+#endif
+
 int main(int argc, char** argv) {
+  #ifdef MI_LINK_VERSION
+    mi_version();
+  #endif
   #ifdef HEAP_WALK
     mi_option_enable(mi_option_visit_abandoned);
   #endif
   #if !defined(NDEBUG) && !defined(USE_STD_MALLOC)
     mi_option_set(mi_option_arena_reserve, 32 * 1024 /* in kib = 32MiB */);
   #endif
-  #ifndef USE_STD_MALLOC
-    mi_stats_reset();
-  #endif
 
   // > mimalloc-test-stress [THREADS] [SCALE] [ITER]
   if (argc >= 2) {
@@ -299,6 +307,11 @@ int main(int argc, char** argv) {
     allow_large_objects = true;
   }
   printf("Using %d threads with a %d%% load-per-thread and %d iterations %s\n", THREADS, SCALE, ITER, (allow_large_objects ? "(allow large objects)" : ""));
+
+  #if !defined(NDEBUG) && !defined(USE_STD_MALLOC)
+  mi_stats_reset();
+  #endif
+
   //mi_reserve_os_memory(1024*1024*1024ULL, false, true);
   //int res = mi_reserve_huge_os_pages(4,1);
   //printf("(reserve huge: %i\n)", res);
@@ -307,20 +320,31 @@ int main(int argc, char** argv) {
 
   // Run ITER full iterations where half the objects in the transfer buffer survive to the next round.
   srand(0x7feb352d);
-  // mi_stats_reset();
+  
+  //mi_reserve_os_memory(512ULL << 20, true, true);
+
+  #if !defined(NDEBUG) && !defined(USE_STD_MALLOC)
+  mi_stats_reset();
+  #endif
+
 #ifdef STRESS
-    test_stress();
+  test_stress();
 #else
-    test_leak();
+  test_leak();
 #endif
 
 #ifndef USE_STD_MALLOC
   #ifndef NDEBUG
-  mi_debug_show_arenas(true);
+  mi_debug_show_arenas();
   mi_collect(true);
-  #endif  
+  char* json = mi_stats_get_json(0, NULL);
+  if (json != NULL) {
+    fputs(json,stderr);
+    mi_free(json);
+  }
+  #endif
+  mi_stats_print(NULL);  
 #endif
-  mi_stats_print(NULL);
   //bench_end_program();
   return 0;
 }