gh-90815: Add mimalloc memory allocator (#109914)

* Add mimalloc v2.12

Modified src/alloc.c to remove include of alloc-override.c and not
compile new handler.

Did not include the following files:

 - include/mimalloc-new-delete.h
 - include/mimalloc-override.h
 - src/alloc-override-osx.c
 - src/alloc-override.c
 - src/static.c
 - src/region.c

mimalloc is thread safe and shares a single heap across all runtimes,
therefore finalization and getting global allocated blocks across all
runtimes is different.

* mimalloc: minimal changes for use in Python:

 - remove debug spam for freeing large allocations
 - use same bytes (0xDD) for freed allocations in CPython and mimalloc
   This is important for the test_capi debug memory tests

* Don't export mimalloc symbol in libpython.
* Enable mimalloc as Python allocator option.
* Add mimalloc MIT license.
* Log mimalloc in Lib/test/pythoninfo.py.
* Document new mimalloc support.
* Use macro defs for exports as done in:
  https://github.com/python/cpython/pull/31164/

Co-authored-by: Sam Gross <colesbury@gmail.com>
Co-authored-by: Christian Heimes <christian@python.org>
Co-authored-by: Victor Stinner <vstinner@python.org>

8 files changed, 3091 insertions, 2 deletions

diff --git a/Include/internal/mimalloc/mimalloc.h b/Include/internal/mimalloc/mimalloc.h
new file mode 100644
index 0000000..821129e
--- /dev/null
+++ b/Include/internal/mimalloc/mimalloc.h
@@ -0,0 +1,565 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2023, 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_H
+#define MIMALLOC_H
+
+#define MI_MALLOC_VERSION 212   // major + 2 digits minor
+
+// ------------------------------------------------------
+// Compiler specific attributes
+// ------------------------------------------------------
+
+#ifdef __cplusplus
+  #if (__cplusplus >= 201103L) || (_MSC_VER > 1900)  // C++11
+    #define mi_attr_noexcept   noexcept
+  #else
+    #define mi_attr_noexcept   throw()
+  #endif
+#else
+  #define mi_attr_noexcept
+#endif
+
+#if defined(__cplusplus) && (__cplusplus >= 201703)
+  #define mi_decl_nodiscard    [[nodiscard]]
+#elif (defined(__GNUC__) && (__GNUC__ >= 4)) || defined(__clang__)  // includes clang, icc, and clang-cl
+  #define mi_decl_nodiscard    __attribute__((warn_unused_result))
+#elif defined(_HAS_NODISCARD)
+  #define mi_decl_nodiscard    _NODISCARD
+#elif (_MSC_VER >= 1700)
+  #define mi_decl_nodiscard    _Check_return_
+#else
+  #define mi_decl_nodiscard
+#endif
+
+#if defined(_MSC_VER) || defined(__MINGW32__)
+  #if !defined(MI_SHARED_LIB)
+    #define mi_decl_export
+  #elif defined(MI_SHARED_LIB_EXPORT)
+    #define mi_decl_export              __declspec(dllexport)
+  #else
+    #define mi_decl_export              __declspec(dllimport)
+  #endif
+  #if defined(__MINGW32__)
+    #define mi_decl_restrict
+    #define mi_attr_malloc              __attribute__((malloc))
+  #else
+    #if (_MSC_VER >= 1900) && !defined(__EDG__)
+      #define mi_decl_restrict          __declspec(allocator) __declspec(restrict)
+    #else
+      #define mi_decl_restrict          __declspec(restrict)
+    #endif
+    #define mi_attr_malloc
+  #endif
+  #define mi_cdecl                      __cdecl
+  #define mi_attr_alloc_size(s)
+  #define mi_attr_alloc_size2(s1,s2)
+  #define mi_attr_alloc_align(p)
+#elif defined(__GNUC__)                 // includes clang and icc
+  #if defined(MI_SHARED_LIB) && defined(MI_SHARED_LIB_EXPORT)
+    #define mi_decl_export              __attribute__((visibility("default")))
+  #else
+    #define mi_decl_export
+  #endif
+  #define mi_cdecl                      // leads to warnings... __attribute__((cdecl))
+  #define mi_decl_restrict
+  #define mi_attr_malloc                __attribute__((malloc))
+  #if (defined(__clang_major__) && (__clang_major__ < 4)) || (__GNUC__ < 5)
+    #define mi_attr_alloc_size(s)
+    #define mi_attr_alloc_size2(s1,s2)
+    #define mi_attr_alloc_align(p)
+  #elif defined(__INTEL_COMPILER)
+    #define mi_attr_alloc_size(s)       __attribute__((alloc_size(s)))
+    #define mi_attr_alloc_size2(s1,s2)  __attribute__((alloc_size(s1,s2)))
+    #define mi_attr_alloc_align(p)
+  #else
+    #define mi_attr_alloc_size(s)       __attribute__((alloc_size(s)))
+    #define mi_attr_alloc_size2(s1,s2)  __attribute__((alloc_size(s1,s2)))
+    #define mi_attr_alloc_align(p)      __attribute__((alloc_align(p)))
+  #endif
+#else
+  #define mi_cdecl
+  #define mi_decl_export
+  #define mi_decl_restrict
+  #define mi_attr_malloc
+  #define mi_attr_alloc_size(s)
+  #define mi_attr_alloc_size2(s1,s2)
+  #define mi_attr_alloc_align(p)
+#endif
+
+// ------------------------------------------------------
+// Includes
+// ------------------------------------------------------
+
+#include <stddef.h>     // size_t
+#include <stdbool.h>    // bool
+#include <stdint.h>     // INTPTR_MAX
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// ------------------------------------------------------
+// Standard malloc interface
+// ------------------------------------------------------
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc(size_t size)  mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc(size_t count, size_t size)  mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2);
+mi_decl_nodiscard mi_decl_export void* mi_realloc(void* p, size_t newsize)      mi_attr_noexcept mi_attr_alloc_size(2);
+mi_decl_export void* mi_expand(void* p, size_t newsize)                         mi_attr_noexcept mi_attr_alloc_size(2);
+
+mi_decl_export void mi_free(void* p) mi_attr_noexcept;
+mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_strdup(const char* s) mi_attr_noexcept mi_attr_malloc;
+mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_strndup(const char* s, size_t n) mi_attr_noexcept mi_attr_malloc;
+mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_realpath(const char* fname, char* resolved_name) mi_attr_noexcept mi_attr_malloc;
+
+// ------------------------------------------------------
+// Extended functionality
+// ------------------------------------------------------
+#define MI_SMALL_WSIZE_MAX  (128)
+#define MI_SMALL_SIZE_MAX   (MI_SMALL_WSIZE_MAX*sizeof(void*))
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_small(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_small(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc(size_t size)       mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_mallocn(size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2);
+mi_decl_nodiscard mi_decl_export void* mi_reallocn(void* p, size_t count, size_t size)        mi_attr_noexcept mi_attr_alloc_size2(2,3);
+mi_decl_nodiscard mi_decl_export void* mi_reallocf(void* p, size_t newsize)                   mi_attr_noexcept mi_attr_alloc_size(2);
+
+mi_decl_nodiscard mi_decl_export size_t mi_usable_size(const void* p) mi_attr_noexcept;
+mi_decl_nodiscard mi_decl_export size_t mi_good_size(size_t size)     mi_attr_noexcept;
+
+
+// ------------------------------------------------------
+// Internals
+// ------------------------------------------------------
+
+typedef void (mi_cdecl mi_deferred_free_fun)(bool force, unsigned long long heartbeat, void* arg);
+mi_decl_export void mi_register_deferred_free(mi_deferred_free_fun* deferred_free, void* arg) mi_attr_noexcept;
+
+typedef void (mi_cdecl mi_output_fun)(const char* msg, void* arg);
+mi_decl_export void mi_register_output(mi_output_fun* out, void* arg) mi_attr_noexcept;
+
+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 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_process_init(void)     mi_attr_noexcept;
+mi_decl_export void mi_thread_init(void)      mi_attr_noexcept;
+mi_decl_export void mi_thread_done(void)      mi_attr_noexcept;
+mi_decl_export void mi_thread_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept;
+
+mi_decl_export void mi_process_info(size_t* elapsed_msecs, size_t* user_msecs, size_t* system_msecs,
+                                    size_t* current_rss, size_t* peak_rss,
+                                    size_t* current_commit, size_t* peak_commit, size_t* page_faults) mi_attr_noexcept;
+
+// -------------------------------------------------------------------------------------
+// Aligned allocation
+// Note that `alignment` always follows `size` for consistency with unaligned
+// allocation, but unfortunately this differs from `posix_memalign` and `aligned_alloc`.
+// -------------------------------------------------------------------------------------
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_aligned(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_aligned(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc_aligned(size_t count, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2) mi_attr_alloc_align(3);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc_aligned_at(size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2);
+mi_decl_nodiscard mi_decl_export void* mi_realloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(2) mi_attr_alloc_align(3);
+mi_decl_nodiscard mi_decl_export void* mi_realloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(2);
+
+
+// -------------------------------------------------------------------------------------
+// Heaps: first-class, but can only allocate from the same thread that created it.
+// -------------------------------------------------------------------------------------
+
+struct mi_heap_s;
+typedef struct mi_heap_s mi_heap_t;
+
+mi_decl_nodiscard mi_decl_export mi_heap_t* mi_heap_new(void);
+mi_decl_export void       mi_heap_delete(mi_heap_t* heap);
+mi_decl_export void       mi_heap_destroy(mi_heap_t* heap);
+mi_decl_export mi_heap_t* mi_heap_set_default(mi_heap_t* heap);
+mi_decl_export mi_heap_t* mi_heap_get_default(void);
+mi_decl_export mi_heap_t* mi_heap_get_backing(void);
+mi_decl_export void       mi_heap_collect(mi_heap_t* heap, bool force) mi_attr_noexcept;
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_mallocn(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_small(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2);
+
+mi_decl_nodiscard mi_decl_export void* mi_heap_realloc(mi_heap_t* heap, void* p, size_t newsize)              mi_attr_noexcept mi_attr_alloc_size(3);
+mi_decl_nodiscard mi_decl_export void* mi_heap_reallocn(mi_heap_t* heap, void* p, size_t count, size_t size)  mi_attr_noexcept mi_attr_alloc_size2(3,4);
+mi_decl_nodiscard mi_decl_export void* mi_heap_reallocf(mi_heap_t* heap, void* p, size_t newsize)             mi_attr_noexcept mi_attr_alloc_size(3);
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_strdup(mi_heap_t* heap, const char* s)            mi_attr_noexcept mi_attr_malloc;
+mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_strndup(mi_heap_t* heap, const char* s, size_t n) mi_attr_noexcept mi_attr_malloc;
+mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_realpath(mi_heap_t* heap, const char* fname, char* resolved_name) mi_attr_noexcept mi_attr_malloc;
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(3);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(3);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc_aligned(mi_heap_t* heap, size_t count, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3) mi_attr_alloc_align(4);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc_aligned_at(mi_heap_t* heap, size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3);
+mi_decl_nodiscard mi_decl_export void* mi_heap_realloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(3) mi_attr_alloc_align(4);
+mi_decl_nodiscard mi_decl_export void* mi_heap_realloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(3);
+
+
+// --------------------------------------------------------------------------------
+// Zero initialized re-allocation.
+// Only valid on memory that was originally allocated with zero initialization too.
+// e.g. `mi_calloc`, `mi_zalloc`, `mi_zalloc_aligned` etc.
+// see <https://github.com/microsoft/mimalloc/issues/63#issuecomment-508272992>
+// --------------------------------------------------------------------------------
+
+mi_decl_nodiscard mi_decl_export void* mi_rezalloc(void* p, size_t newsize)                mi_attr_noexcept mi_attr_alloc_size(2);
+mi_decl_nodiscard mi_decl_export void* mi_recalloc(void* p, size_t newcount, size_t size)  mi_attr_noexcept mi_attr_alloc_size2(2,3);
+
+mi_decl_nodiscard mi_decl_export void* mi_rezalloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(2) mi_attr_alloc_align(3);
+mi_decl_nodiscard mi_decl_export void* mi_rezalloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(2);
+mi_decl_nodiscard mi_decl_export void* mi_recalloc_aligned(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept mi_attr_alloc_size2(2,3) mi_attr_alloc_align(4);
+mi_decl_nodiscard mi_decl_export void* mi_recalloc_aligned_at(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size2(2,3);
+
+mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc(mi_heap_t* heap, void* p, size_t newsize)                mi_attr_noexcept mi_attr_alloc_size(3);
+mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc(mi_heap_t* heap, void* p, size_t newcount, size_t size)  mi_attr_noexcept mi_attr_alloc_size2(3,4);
+
+mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(3) mi_attr_alloc_align(4);
+mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(3);
+mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc_aligned(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept mi_attr_alloc_size2(3,4) mi_attr_alloc_align(5);
+mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc_aligned_at(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size2(3,4);
+
+
+// ------------------------------------------------------
+// Analysis
+// ------------------------------------------------------
+
+mi_decl_export bool mi_heap_contains_block(mi_heap_t* heap, const void* p);
+mi_decl_export bool mi_heap_check_owned(mi_heap_t* heap, const void* p);
+mi_decl_export bool mi_check_owned(const void* p);
+
+// An area of heap space contains blocks of a single size.
+typedef struct mi_heap_area_s {
+  void*  blocks;      // start of the area containing heap blocks
+  size_t reserved;    // bytes reserved for this area (virtual)
+  size_t committed;   // current available bytes for this area
+  size_t used;        // number of allocated blocks
+  size_t block_size;  // size in bytes of each block
+  size_t full_block_size; // size in bytes of a full block including padding and metadata.
+} mi_heap_area_t;
+
+typedef bool (mi_cdecl mi_block_visit_fun)(const mi_heap_t* heap, const mi_heap_area_t* area, void* block, size_t block_size, void* arg);
+
+mi_decl_export bool mi_heap_visit_blocks(const mi_heap_t* heap, bool visit_all_blocks, mi_block_visit_fun* visitor, void* arg);
+
+// Experimental
+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_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(void) mi_attr_noexcept;
+
+// Experimental: heaps associated with specific memory arena's
+typedef int mi_arena_id_t;
+mi_decl_export void* mi_arena_area(mi_arena_id_t arena_id, size_t* size);
+mi_decl_export int   mi_reserve_huge_os_pages_at_ex(size_t pages, int numa_node, size_t timeout_msecs, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept;
+mi_decl_export int   mi_reserve_os_memory_ex(size_t size, bool commit, bool allow_large, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept;
+mi_decl_export bool  mi_manage_os_memory_ex(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept;
+
+#if MI_MALLOC_VERSION >= 182
+// Create a heap that only allocates in the specified arena
+mi_decl_nodiscard mi_decl_export mi_heap_t* mi_heap_new_in_arena(mi_arena_id_t arena_id);
+#endif
+
+// deprecated
+mi_decl_export int  mi_reserve_huge_os_pages(size_t pages, double max_secs, size_t* pages_reserved) mi_attr_noexcept;
+
+
+// ------------------------------------------------------
+// Convenience
+// ------------------------------------------------------
+
+#define mi_malloc_tp(tp)                ((tp*)mi_malloc(sizeof(tp)))
+#define mi_zalloc_tp(tp)                ((tp*)mi_zalloc(sizeof(tp)))
+#define mi_calloc_tp(tp,n)              ((tp*)mi_calloc(n,sizeof(tp)))
+#define mi_mallocn_tp(tp,n)             ((tp*)mi_mallocn(n,sizeof(tp)))
+#define mi_reallocn_tp(p,tp,n)          ((tp*)mi_reallocn(p,n,sizeof(tp)))
+#define mi_recalloc_tp(p,tp,n)          ((tp*)mi_recalloc(p,n,sizeof(tp)))
+
+#define mi_heap_malloc_tp(hp,tp)        ((tp*)mi_heap_malloc(hp,sizeof(tp)))
+#define mi_heap_zalloc_tp(hp,tp)        ((tp*)mi_heap_zalloc(hp,sizeof(tp)))
+#define mi_heap_calloc_tp(hp,tp,n)      ((tp*)mi_heap_calloc(hp,n,sizeof(tp)))
+#define mi_heap_mallocn_tp(hp,tp,n)     ((tp*)mi_heap_mallocn(hp,n,sizeof(tp)))
+#define mi_heap_reallocn_tp(hp,p,tp,n)  ((tp*)mi_heap_reallocn(hp,p,n,sizeof(tp)))
+#define mi_heap_recalloc_tp(hp,p,tp,n)  ((tp*)mi_heap_recalloc(hp,p,n,sizeof(tp)))
+
+
+// ------------------------------------------------------
+// Options
+// ------------------------------------------------------
+
+typedef enum mi_option_e {
+  // stable options
+  mi_option_show_errors,              // print error messages
+  mi_option_show_stats,               // print statistics on termination
+  mi_option_verbose,                  // print verbose messages
+  // the following options are experimental (see src/options.h)
+  mi_option_eager_commit,             // eager commit segments? (after `eager_commit_delay` segments) (=1)
+  mi_option_arena_eager_commit,       // eager commit arenas? Use 2 to enable just on overcommit systems (=2)
+  mi_option_purge_decommits,          // should a memory purge decommit (or only reset) (=1)
+  mi_option_allow_large_os_pages,     // allow large (2MiB) OS pages, implies eager commit
+  mi_option_reserve_huge_os_pages,    // reserve N huge OS pages (1GiB/page) at startup
+  mi_option_reserve_huge_os_pages_at, // reserve huge OS pages at a specific NUMA node
+  mi_option_reserve_os_memory,        // reserve specified amount of OS memory in an arena at startup
+  mi_option_deprecated_segment_cache,
+  mi_option_deprecated_page_reset,
+  mi_option_abandoned_page_purge,     // immediately purge delayed purges on thread termination
+  mi_option_deprecated_segment_reset,
+  mi_option_eager_commit_delay,
+  mi_option_purge_delay,              // memory purging is delayed by N milli seconds; use 0 for immediate purging or -1 for no purging at all.
+  mi_option_use_numa_nodes,           // 0 = use all available numa nodes, otherwise use at most N nodes.
+  mi_option_limit_os_alloc,           // 1 = do not use OS memory for allocation (but only programmatically reserved arenas)
+  mi_option_os_tag,                   // tag used for OS logging (macOS only for now)
+  mi_option_max_errors,               // issue at most N error messages
+  mi_option_max_warnings,             // issue at most N warning messages
+  mi_option_max_segment_reclaim,
+  mi_option_destroy_on_exit,          // if set, release all memory on exit; sometimes used for dynamic unloading but can be unsafe.
+  mi_option_arena_reserve,            // initial memory size in KiB for arena reservation (1GiB on 64-bit)
+  mi_option_arena_purge_mult,
+  mi_option_purge_extend_delay,
+  _mi_option_last,
+  // legacy option names
+  mi_option_large_os_pages = mi_option_allow_large_os_pages,
+  mi_option_eager_region_commit = mi_option_arena_eager_commit,
+  mi_option_reset_decommits = mi_option_purge_decommits,
+  mi_option_reset_delay = mi_option_purge_delay,
+  mi_option_abandoned_page_reset = mi_option_abandoned_page_purge
+} mi_option_t;
+
+
+mi_decl_nodiscard mi_decl_export bool mi_option_is_enabled(mi_option_t option);
+mi_decl_export void mi_option_enable(mi_option_t option);
+mi_decl_export void mi_option_disable(mi_option_t option);
+mi_decl_export void mi_option_set_enabled(mi_option_t option, bool enable);
+mi_decl_export void mi_option_set_enabled_default(mi_option_t option, bool enable);
+
+mi_decl_nodiscard mi_decl_export long   mi_option_get(mi_option_t option);
+mi_decl_nodiscard mi_decl_export long   mi_option_get_clamp(mi_option_t option, long min, long max);
+mi_decl_nodiscard mi_decl_export size_t mi_option_get_size(mi_option_t option);
+mi_decl_export void mi_option_set(mi_option_t option, long value);
+mi_decl_export void mi_option_set_default(mi_option_t option, long value);
+
+
+// -------------------------------------------------------------------------------------------------------
+// "mi" prefixed implementations of various posix, Unix, Windows, and C++ allocation functions.
+// (This can be convenient when providing overrides of these functions as done in `mimalloc-override.h`.)
+// note: we use `mi_cfree` as "checked free" and it checks if the pointer is in our heap before free-ing.
+// -------------------------------------------------------------------------------------------------------
+
+mi_decl_export void  mi_cfree(void* p) mi_attr_noexcept;
+mi_decl_export void* mi__expand(void* p, size_t newsize) mi_attr_noexcept;
+mi_decl_nodiscard mi_decl_export size_t mi_malloc_size(const void* p)        mi_attr_noexcept;
+mi_decl_nodiscard mi_decl_export size_t mi_malloc_good_size(size_t size)     mi_attr_noexcept;
+mi_decl_nodiscard mi_decl_export size_t mi_malloc_usable_size(const void *p) mi_attr_noexcept;
+
+mi_decl_export int mi_posix_memalign(void** p, size_t alignment, size_t size)   mi_attr_noexcept;
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_memalign(size_t alignment, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_valloc(size_t size)  mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_pvalloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_aligned_alloc(size_t alignment, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(1);
+
+mi_decl_nodiscard mi_decl_export void* mi_reallocarray(void* p, size_t count, size_t size) mi_attr_noexcept mi_attr_alloc_size2(2,3);
+mi_decl_nodiscard mi_decl_export int   mi_reallocarr(void* p, size_t count, size_t size) mi_attr_noexcept;
+mi_decl_nodiscard mi_decl_export void* mi_aligned_recalloc(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept;
+mi_decl_nodiscard mi_decl_export void* mi_aligned_offset_recalloc(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept;
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict unsigned short* mi_wcsdup(const unsigned short* s) mi_attr_noexcept mi_attr_malloc;
+mi_decl_nodiscard mi_decl_export mi_decl_restrict unsigned char*  mi_mbsdup(const unsigned char* s)  mi_attr_noexcept mi_attr_malloc;
+mi_decl_export int mi_dupenv_s(char** buf, size_t* size, const char* name)                      mi_attr_noexcept;
+mi_decl_export int mi_wdupenv_s(unsigned short** buf, size_t* size, const unsigned short* name) mi_attr_noexcept;
+
+mi_decl_export void mi_free_size(void* p, size_t size)                           mi_attr_noexcept;
+mi_decl_export void mi_free_size_aligned(void* p, size_t size, size_t alignment) mi_attr_noexcept;
+mi_decl_export void mi_free_aligned(void* p, size_t alignment)                   mi_attr_noexcept;
+
+// The `mi_new` wrappers implement C++ semantics on out-of-memory instead of directly returning `NULL`.
+// (and call `std::get_new_handler` and potentially raise a `std::bad_alloc` exception).
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new(size_t size)                   mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_aligned(size_t size, size_t alignment) mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_nothrow(size_t size)           mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_aligned_nothrow(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_n(size_t count, size_t size)   mi_attr_malloc mi_attr_alloc_size2(1, 2);
+mi_decl_nodiscard mi_decl_export void* mi_new_realloc(void* p, size_t newsize)                mi_attr_alloc_size(2);
+mi_decl_nodiscard mi_decl_export void* mi_new_reallocn(void* p, size_t newcount, size_t size) mi_attr_alloc_size2(2, 3);
+
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_alloc_new(mi_heap_t* heap, size_t size)                mi_attr_malloc mi_attr_alloc_size(2);
+mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_alloc_new_n(mi_heap_t* heap, size_t count, size_t size) mi_attr_malloc mi_attr_alloc_size2(2, 3);
+
+#ifdef __cplusplus
+}
+#endif
+
+// ---------------------------------------------------------------------------------------------
+// Implement the C++ std::allocator interface for use in STL containers.
+// (note: see `mimalloc-new-delete.h` for overriding the new/delete operators globally)
+// ---------------------------------------------------------------------------------------------
+#ifdef __cplusplus
+
+#include <cstddef>     // std::size_t
+#include <cstdint>     // PTRDIFF_MAX
+#if (__cplusplus >= 201103L) || (_MSC_VER > 1900)  // C++11
+#include <type_traits> // std::true_type
+#include <utility>     // std::forward
+#endif
+
+template<class T> struct _mi_stl_allocator_common {
+  typedef T                 value_type;
+  typedef std::size_t       size_type;
+  typedef std::ptrdiff_t    difference_type;
+  typedef value_type&       reference;
+  typedef value_type const& const_reference;
+  typedef value_type*       pointer;
+  typedef value_type const* const_pointer;
+
+  #if ((__cplusplus >= 201103L) || (_MSC_VER > 1900))  // C++11
+  using propagate_on_container_copy_assignment = std::true_type;
+  using propagate_on_container_move_assignment = std::true_type;
+  using propagate_on_container_swap            = std::true_type;
+  template <class U, class ...Args> void construct(U* p, Args&& ...args) { ::new(p) U(std::forward<Args>(args)...); }
+  template <class U> void destroy(U* p) mi_attr_noexcept { p->~U(); }
+  #else
+  void construct(pointer p, value_type const& val) { ::new(p) value_type(val); }
+  void destroy(pointer p) { p->~value_type(); }
+  #endif
+
+  size_type     max_size() const mi_attr_noexcept { return (PTRDIFF_MAX/sizeof(value_type)); }
+  pointer       address(reference x) const        { return &x; }
+  const_pointer address(const_reference x) const  { return &x; }
+};
+
+template<class T> struct mi_stl_allocator : public _mi_stl_allocator_common<T> {
+  using typename _mi_stl_allocator_common<T>::size_type;
+  using typename _mi_stl_allocator_common<T>::value_type;
+  using typename _mi_stl_allocator_common<T>::pointer;
+  template <class U> struct rebind { typedef mi_stl_allocator<U> other; };
+
+  mi_stl_allocator()                                             mi_attr_noexcept = default;
+  mi_stl_allocator(const mi_stl_allocator&)                      mi_attr_noexcept = default;
+  template<class U> mi_stl_allocator(const mi_stl_allocator<U>&) mi_attr_noexcept { }
+  mi_stl_allocator  select_on_container_copy_construction() const { return *this; }
+  void              deallocate(T* p, size_type) { mi_free(p); }
+
+  #if (__cplusplus >= 201703L)  // C++17
+  mi_decl_nodiscard T* allocate(size_type count) { return static_cast<T*>(mi_new_n(count, sizeof(T))); }
+  mi_decl_nodiscard T* allocate(size_type count, const void*) { return allocate(count); }
+  #else
+  mi_decl_nodiscard pointer allocate(size_type count, const void* = 0) { return static_cast<pointer>(mi_new_n(count, sizeof(value_type))); }
+  #endif
+
+  #if ((__cplusplus >= 201103L) || (_MSC_VER > 1900))  // C++11
+  using is_always_equal = std::true_type;
+  #endif
+};
+
+template<class T1,class T2> bool operator==(const mi_stl_allocator<T1>& , const mi_stl_allocator<T2>& ) mi_attr_noexcept { return true; }
+template<class T1,class T2> bool operator!=(const mi_stl_allocator<T1>& , const mi_stl_allocator<T2>& ) mi_attr_noexcept { return false; }
+
+
+#if (__cplusplus >= 201103L) || (_MSC_VER >= 1900)  // C++11
+#define MI_HAS_HEAP_STL_ALLOCATOR 1
+
+#include <memory>      // std::shared_ptr
+
+// Common base class for STL allocators in a specific heap
+template<class T, bool _mi_destroy> struct _mi_heap_stl_allocator_common : public _mi_stl_allocator_common<T> {
+  using typename _mi_stl_allocator_common<T>::size_type;
+  using typename _mi_stl_allocator_common<T>::value_type;
+  using typename _mi_stl_allocator_common<T>::pointer;
+
+  _mi_heap_stl_allocator_common(mi_heap_t* hp) : heap(hp) { }    /* will not delete nor destroy the passed in heap */
+
+  #if (__cplusplus >= 201703L)  // C++17
+  mi_decl_nodiscard T* allocate(size_type count) { return static_cast<T*>(mi_heap_alloc_new_n(this->heap.get(), count, sizeof(T))); }
+  mi_decl_nodiscard T* allocate(size_type count, const void*) { return allocate(count); }
+  #else
+  mi_decl_nodiscard pointer allocate(size_type count, const void* = 0) { return static_cast<pointer>(mi_heap_alloc_new_n(this->heap.get(), count, sizeof(value_type))); }
+  #endif
+
+  #if ((__cplusplus >= 201103L) || (_MSC_VER > 1900))  // C++11
+  using is_always_equal = std::false_type;
+  #endif
+
+  void collect(bool force) { mi_heap_collect(this->heap.get(), force); }
+  template<class U> bool is_equal(const _mi_heap_stl_allocator_common<U, _mi_destroy>& x) const { return (this->heap == x.heap); }
+
+protected:
+  std::shared_ptr<mi_heap_t> heap;
+  template<class U, bool D> friend struct _mi_heap_stl_allocator_common;
+
+  _mi_heap_stl_allocator_common() {
+    mi_heap_t* hp = mi_heap_new();
+    this->heap.reset(hp, (_mi_destroy ? &heap_destroy : &heap_delete));  /* calls heap_delete/destroy when the refcount drops to zero */
+  }
+  _mi_heap_stl_allocator_common(const _mi_heap_stl_allocator_common& x) mi_attr_noexcept : heap(x.heap) { }
+  template<class U> _mi_heap_stl_allocator_common(const _mi_heap_stl_allocator_common<U, _mi_destroy>& x) mi_attr_noexcept : heap(x.heap) { }
+
+private:
+  static void heap_delete(mi_heap_t* hp)  { if (hp != NULL) { mi_heap_delete(hp); } }
+  static void heap_destroy(mi_heap_t* hp) { if (hp != NULL) { mi_heap_destroy(hp); } }
+};
+
+// STL allocator allocation in a specific heap
+template<class T> struct mi_heap_stl_allocator : public _mi_heap_stl_allocator_common<T, false> {
+  using typename _mi_heap_stl_allocator_common<T, false>::size_type;
+  mi_heap_stl_allocator() : _mi_heap_stl_allocator_common<T, false>() { } // creates fresh heap that is deleted when the destructor is called
+  mi_heap_stl_allocator(mi_heap_t* hp) : _mi_heap_stl_allocator_common<T, false>(hp) { }  // no delete nor destroy on the passed in heap
+  template<class U> mi_heap_stl_allocator(const mi_heap_stl_allocator<U>& x) mi_attr_noexcept : _mi_heap_stl_allocator_common<T, false>(x) { }
+
+  mi_heap_stl_allocator select_on_container_copy_construction() const { return *this; }
+  void deallocate(T* p, size_type) { mi_free(p); }
+  template<class U> struct rebind { typedef mi_heap_stl_allocator<U> other; };
+};
+
+template<class T1, class T2> bool operator==(const mi_heap_stl_allocator<T1>& x, const mi_heap_stl_allocator<T2>& y) mi_attr_noexcept { return (x.is_equal(y)); }
+template<class T1, class T2> bool operator!=(const mi_heap_stl_allocator<T1>& x, const mi_heap_stl_allocator<T2>& y) mi_attr_noexcept { return (!x.is_equal(y)); }
+
+
+// STL allocator allocation in a specific heap, where `free` does nothing and
+// the heap is destroyed in one go on destruction -- use with care!
+template<class T> struct mi_heap_destroy_stl_allocator : public _mi_heap_stl_allocator_common<T, true> {
+  using typename _mi_heap_stl_allocator_common<T, true>::size_type;
+  mi_heap_destroy_stl_allocator() : _mi_heap_stl_allocator_common<T, true>() { } // creates fresh heap that is destroyed when the destructor is called
+  mi_heap_destroy_stl_allocator(mi_heap_t* hp) : _mi_heap_stl_allocator_common<T, true>(hp) { }  // no delete nor destroy on the passed in heap
+  template<class U> mi_heap_destroy_stl_allocator(const mi_heap_destroy_stl_allocator<U>& x) mi_attr_noexcept : _mi_heap_stl_allocator_common<T, true>(x) { }
+
+  mi_heap_destroy_stl_allocator select_on_container_copy_construction() const { return *this; }
+  void deallocate(T*, size_type) { /* do nothing as we destroy the heap on destruct. */ }
+  template<class U> struct rebind { typedef mi_heap_destroy_stl_allocator<U> other; };
+};
+
+template<class T1, class T2> bool operator==(const mi_heap_destroy_stl_allocator<T1>& x, const mi_heap_destroy_stl_allocator<T2>& y) mi_attr_noexcept { return (x.is_equal(y)); }
+template<class T1, class T2> bool operator!=(const mi_heap_destroy_stl_allocator<T1>& x, const mi_heap_destroy_stl_allocator<T2>& y) mi_attr_noexcept { return (!x.is_equal(y)); }
+
+#endif // C++11
+
+#endif // __cplusplus
+
+#endif
diff --git a/Include/internal/mimalloc/mimalloc/atomic.h b/Include/internal/mimalloc/mimalloc/atomic.h
new file mode 100644
index 0000000..c6b8146
--- /dev/null
+++ b/Include/internal/mimalloc/mimalloc/atomic.h
@@ -0,0 +1,385 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2023 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_ATOMIC_H
+#define MIMALLOC_ATOMIC_H
+
+// --------------------------------------------------------------------------------------------
+// Atomics
+// We need to be portable between C, C++, and MSVC.
+// We base the primitives on the C/C++ atomics and create a mimimal wrapper for MSVC in C compilation mode.
+// This is why we try to use only `uintptr_t` and `<type>*` as atomic types.
+// To gain better insight in the range of used atomics, we use explicitly named memory order operations
+// instead of passing the memory order as a parameter.
+// -----------------------------------------------------------------------------------------------
+
+#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 !defined(ATOMIC_VAR_INIT) || (__cplusplus >= 202002L) // c++20, see issue #571
+ #define MI_ATOMIC_VAR_INIT(x)  x
+#else
+ #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
+#else
+// Use C11 atomics
+#include <stdatomic.h>
+#define  mi_atomic(name)        atomic_##name
+#define  mi_memory_order(name)  memory_order_##name
+#if !defined(ATOMIC_VAR_INIT) || (__STDC_VERSION__ >= 201710L) // c17, see issue #735
+ #define MI_ATOMIC_VAR_INIT(x) x
+#else
+ #define MI_ATOMIC_VAR_INIT(x) ATOMIC_VAR_INIT(x)
+#endif
+#endif
+
+// Various defines for all used memory orders in mimalloc
+#define mi_atomic_cas_weak(p,expected,desired,mem_success,mem_fail)  \
+  mi_atomic(compare_exchange_weak_explicit)(p,expected,desired,mem_success,mem_fail)
+
+#define mi_atomic_cas_strong(p,expected,desired,mem_success,mem_fail)  \
+  mi_atomic(compare_exchange_strong_explicit)(p,expected,desired,mem_success,mem_fail)
+
+#define mi_atomic_load_acquire(p)                mi_atomic(load_explicit)(p,mi_memory_order(acquire))
+#define mi_atomic_load_relaxed(p)                mi_atomic(load_explicit)(p,mi_memory_order(relaxed))
+#define mi_atomic_store_release(p,x)             mi_atomic(store_explicit)(p,x,mi_memory_order(release))
+#define mi_atomic_store_relaxed(p,x)             mi_atomic(store_explicit)(p,x,mi_memory_order(relaxed))
+#define mi_atomic_exchange_release(p,x)          mi_atomic(exchange_explicit)(p,x,mi_memory_order(release))
+#define mi_atomic_exchange_acq_rel(p,x)          mi_atomic(exchange_explicit)(p,x,mi_memory_order(acq_rel))
+#define mi_atomic_cas_weak_release(p,exp,des)    mi_atomic_cas_weak(p,exp,des,mi_memory_order(release),mi_memory_order(relaxed))
+#define mi_atomic_cas_weak_acq_rel(p,exp,des)    mi_atomic_cas_weak(p,exp,des,mi_memory_order(acq_rel),mi_memory_order(acquire))
+#define mi_atomic_cas_strong_release(p,exp,des)  mi_atomic_cas_strong(p,exp,des,mi_memory_order(release),mi_memory_order(relaxed))
+#define mi_atomic_cas_strong_acq_rel(p,exp,des)  mi_atomic_cas_strong(p,exp,des,mi_memory_order(acq_rel),mi_memory_order(acquire))
+
+#define mi_atomic_add_relaxed(p,x)               mi_atomic(fetch_add_explicit)(p,x,mi_memory_order(relaxed))
+#define mi_atomic_sub_relaxed(p,x)               mi_atomic(fetch_sub_explicit)(p,x,mi_memory_order(relaxed))
+#define mi_atomic_add_acq_rel(p,x)               mi_atomic(fetch_add_explicit)(p,x,mi_memory_order(acq_rel))
+#define mi_atomic_sub_acq_rel(p,x)               mi_atomic(fetch_sub_explicit)(p,x,mi_memory_order(acq_rel))
+#define mi_atomic_and_acq_rel(p,x)               mi_atomic(fetch_and_explicit)(p,x,mi_memory_order(acq_rel))
+#define mi_atomic_or_acq_rel(p,x)                mi_atomic(fetch_or_explicit)(p,x,mi_memory_order(acq_rel))
+
+#define mi_atomic_increment_relaxed(p)           mi_atomic_add_relaxed(p,(uintptr_t)1)
+#define mi_atomic_decrement_relaxed(p)           mi_atomic_sub_relaxed(p,(uintptr_t)1)
+#define mi_atomic_increment_acq_rel(p)           mi_atomic_add_acq_rel(p,(uintptr_t)1)
+#define mi_atomic_decrement_acq_rel(p)           mi_atomic_sub_acq_rel(p,(uintptr_t)1)
+
+static inline void mi_atomic_yield(void);
+static inline intptr_t mi_atomic_addi(_Atomic(intptr_t)*p, intptr_t add);
+static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub);
+
+
+#if defined(__cplusplus) || !defined(_MSC_VER)
+
+// In C++/C11 atomics we have polymorphic atomics so can use the typed `ptr` variants (where `tp` is the type of atomic value)
+// We use these macros so we can provide a typed wrapper in MSVC in C compilation mode as well
+#define mi_atomic_load_ptr_acquire(tp,p)                mi_atomic_load_acquire(p)
+#define mi_atomic_load_ptr_relaxed(tp,p)                mi_atomic_load_relaxed(p)
+
+// In C++ we need to add casts to help resolve templates if NULL is passed
+#if defined(__cplusplus)
+#define mi_atomic_store_ptr_release(tp,p,x)             mi_atomic_store_release(p,(tp*)x)
+#define mi_atomic_store_ptr_relaxed(tp,p,x)             mi_atomic_store_relaxed(p,(tp*)x)
+#define mi_atomic_cas_ptr_weak_release(tp,p,exp,des)    mi_atomic_cas_weak_release(p,exp,(tp*)des)
+#define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des)    mi_atomic_cas_weak_acq_rel(p,exp,(tp*)des)
+#define mi_atomic_cas_ptr_strong_release(tp,p,exp,des)  mi_atomic_cas_strong_release(p,exp,(tp*)des)
+#define mi_atomic_exchange_ptr_release(tp,p,x)          mi_atomic_exchange_release(p,(tp*)x)
+#define mi_atomic_exchange_ptr_acq_rel(tp,p,x)          mi_atomic_exchange_acq_rel(p,(tp*)x)
+#else
+#define mi_atomic_store_ptr_release(tp,p,x)             mi_atomic_store_release(p,x)
+#define mi_atomic_store_ptr_relaxed(tp,p,x)             mi_atomic_store_relaxed(p,x)
+#define mi_atomic_cas_ptr_weak_release(tp,p,exp,des)    mi_atomic_cas_weak_release(p,exp,des)
+#define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des)    mi_atomic_cas_weak_acq_rel(p,exp,des)
+#define mi_atomic_cas_ptr_strong_release(tp,p,exp,des)  mi_atomic_cas_strong_release(p,exp,des)
+#define mi_atomic_exchange_ptr_release(tp,p,x)          mi_atomic_exchange_release(p,x)
+#define mi_atomic_exchange_ptr_acq_rel(tp,p,x)          mi_atomic_exchange_acq_rel(p,x)
+#endif
+
+// These are used by the statistics
+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_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 */ };
+}
+
+// Used by timers
+#define mi_atomic_loadi64_acquire(p)            mi_atomic(load_explicit)(p,mi_memory_order(acquire))
+#define mi_atomic_loadi64_relaxed(p)            mi_atomic(load_explicit)(p,mi_memory_order(relaxed))
+#define mi_atomic_storei64_release(p,x)         mi_atomic(store_explicit)(p,x,mi_memory_order(release))
+#define mi_atomic_storei64_relaxed(p,x)         mi_atomic(store_explicit)(p,x,mi_memory_order(relaxed))
+
+#define mi_atomic_casi64_strong_acq_rel(p,e,d)  mi_atomic_cas_strong_acq_rel(p,e,d)
+#define mi_atomic_addi64_acq_rel(p,i)           mi_atomic_add_acq_rel(p,i)
+
+
+#elif defined(_MSC_VER)
+
+// MSVC C compilation wrapper that uses Interlocked operations to model C11 atomics.
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#include <intrin.h>
+#ifdef _WIN64
+typedef LONG64   msc_intptr_t;
+#define MI_64(f) f##64
+#else
+typedef LONG     msc_intptr_t;
+#define MI_64(f) f
+#endif
+
+typedef enum mi_memory_order_e {
+  mi_memory_order_relaxed,
+  mi_memory_order_consume,
+  mi_memory_order_acquire,
+  mi_memory_order_release,
+  mi_memory_order_acq_rel,
+  mi_memory_order_seq_cst
+} mi_memory_order;
+
+static inline uintptr_t mi_atomic_fetch_add_explicit(_Atomic(uintptr_t)*p, uintptr_t add, mi_memory_order mo) {
+  (void)(mo);
+  return (uintptr_t)MI_64(_InterlockedExchangeAdd)((volatile msc_intptr_t*)p, (msc_intptr_t)add);
+}
+static inline uintptr_t mi_atomic_fetch_sub_explicit(_Atomic(uintptr_t)*p, uintptr_t sub, mi_memory_order mo) {
+  (void)(mo);
+  return (uintptr_t)MI_64(_InterlockedExchangeAdd)((volatile msc_intptr_t*)p, -((msc_intptr_t)sub));
+}
+static inline uintptr_t mi_atomic_fetch_and_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) {
+  (void)(mo);
+  return (uintptr_t)MI_64(_InterlockedAnd)((volatile msc_intptr_t*)p, (msc_intptr_t)x);
+}
+static inline uintptr_t mi_atomic_fetch_or_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) {
+  (void)(mo);
+  return (uintptr_t)MI_64(_InterlockedOr)((volatile msc_intptr_t*)p, (msc_intptr_t)x);
+}
+static inline bool mi_atomic_compare_exchange_strong_explicit(_Atomic(uintptr_t)*p, uintptr_t* expected, uintptr_t desired, mi_memory_order mo1, mi_memory_order mo2) {
+  (void)(mo1); (void)(mo2);
+  uintptr_t read = (uintptr_t)MI_64(_InterlockedCompareExchange)((volatile msc_intptr_t*)p, (msc_intptr_t)desired, (msc_intptr_t)(*expected));
+  if (read == *expected) {
+    return true;
+  }
+  else {
+    *expected = read;
+    return false;
+  }
+}
+static inline bool mi_atomic_compare_exchange_weak_explicit(_Atomic(uintptr_t)*p, uintptr_t* expected, uintptr_t desired, mi_memory_order mo1, mi_memory_order mo2) {
+  return mi_atomic_compare_exchange_strong_explicit(p, expected, desired, mo1, mo2);
+}
+static inline uintptr_t mi_atomic_exchange_explicit(_Atomic(uintptr_t)*p, uintptr_t exchange, mi_memory_order mo) {
+  (void)(mo);
+  return (uintptr_t)MI_64(_InterlockedExchange)((volatile msc_intptr_t*)p, (msc_intptr_t)exchange);
+}
+static inline void mi_atomic_thread_fence(mi_memory_order mo) {
+  (void)(mo);
+  _Atomic(uintptr_t) x = 0;
+  mi_atomic_exchange_explicit(&x, 1, mo);
+}
+static inline uintptr_t mi_atomic_load_explicit(_Atomic(uintptr_t) const* p, mi_memory_order mo) {
+  (void)(mo);
+#if defined(_M_IX86) || defined(_M_X64)
+  return *p;
+#else
+  uintptr_t x = *p;
+  if (mo > mi_memory_order_relaxed) {
+    while (!mi_atomic_compare_exchange_weak_explicit(p, &x, x, mo, mi_memory_order_relaxed)) { /* nothing */ };
+  }
+  return x;
+#endif
+}
+static inline void mi_atomic_store_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) {
+  (void)(mo);
+#if defined(_M_IX86) || defined(_M_X64)
+  *p = x;
+#else
+  mi_atomic_exchange_explicit(p, x, mo);
+#endif
+}
+static inline int64_t mi_atomic_loadi64_explicit(_Atomic(int64_t)*p, mi_memory_order mo) {
+  (void)(mo);
+#if defined(_M_X64)
+  return *p;
+#else
+  int64_t old = *p;
+  int64_t x = old;
+  while ((old = InterlockedCompareExchange64(p, x, old)) != x) {
+    x = old;
+  }
+  return x;
+#endif
+}
+static inline void mi_atomic_storei64_explicit(_Atomic(int64_t)*p, int64_t x, mi_memory_order mo) {
+  (void)(mo);
+#if defined(x_M_IX86) || defined(_M_X64)
+  *p = x;
+#else
+  InterlockedExchange64(p, x);
+#endif
+}
+
+// These are used by the statistics
+static inline int64_t mi_atomic_addi64_relaxed(volatile _Atomic(int64_t)*p, int64_t add) {
+#ifdef _WIN64
+  return (int64_t)mi_atomic_addi((int64_t*)p, add);
+#else
+  int64_t current;
+  int64_t sum;
+  do {
+    current = *p;
+    sum = current + add;
+  } while (_InterlockedCompareExchange64(p, sum, current) != current);
+  return current;
+#endif
+}
+static inline void mi_atomic_maxi64_relaxed(volatile _Atomic(int64_t)*p, int64_t x) {
+  int64_t current;
+  do {
+    current = *p;
+  } while (current < x && _InterlockedCompareExchange64(p, x, current) != current);
+}
+
+static inline void mi_atomic_addi64_acq_rel(volatile _Atomic(int64_t*)p, int64_t i) {
+  mi_atomic_addi64_relaxed(p, i);
+}
+
+static inline bool mi_atomic_casi64_strong_acq_rel(volatile _Atomic(int64_t*)p, int64_t* exp, int64_t des) {
+  int64_t read = _InterlockedCompareExchange64(p, des, *exp);
+  if (read == *exp) {
+    return true;
+  }
+  else {
+    *exp = read;
+    return false;
+  }
+}
+
+// The pointer macros cast to `uintptr_t`.
+#define mi_atomic_load_ptr_acquire(tp,p)                (tp*)mi_atomic_load_acquire((_Atomic(uintptr_t)*)(p))
+#define mi_atomic_load_ptr_relaxed(tp,p)                (tp*)mi_atomic_load_relaxed((_Atomic(uintptr_t)*)(p))
+#define mi_atomic_store_ptr_release(tp,p,x)             mi_atomic_store_release((_Atomic(uintptr_t)*)(p),(uintptr_t)(x))
+#define mi_atomic_store_ptr_relaxed(tp,p,x)             mi_atomic_store_relaxed((_Atomic(uintptr_t)*)(p),(uintptr_t)(x))
+#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_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)
+
+#define mi_atomic_loadi64_acquire(p)    mi_atomic(loadi64_explicit)(p,mi_memory_order(acquire))
+#define mi_atomic_loadi64_relaxed(p)    mi_atomic(loadi64_explicit)(p,mi_memory_order(relaxed))
+#define mi_atomic_storei64_release(p,x) mi_atomic(storei64_explicit)(p,x,mi_memory_order(release))
+#define mi_atomic_storei64_relaxed(p,x) mi_atomic(storei64_explicit)(p,x,mi_memory_order(relaxed))
+
+
+#endif
+
+
+// Atomically add a signed value; returns the previous value.
+static inline intptr_t mi_atomic_addi(_Atomic(intptr_t)*p, intptr_t add) {
+  return (intptr_t)mi_atomic_add_acq_rel((_Atomic(uintptr_t)*)p, (uintptr_t)add);
+}
+
+// Atomically subtract a signed value; returns the previous value.
+static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub) {
+  return (intptr_t)mi_atomic_addi(p, -sub);
+}
+
+typedef _Atomic(uintptr_t) mi_atomic_once_t;
+
+// Returns true only on the first invocation
+static inline bool mi_atomic_once( mi_atomic_once_t* once ) {
+  if (mi_atomic_load_relaxed(once) != 0) return false;     // quick test
+  uintptr_t expected = 0;
+  return mi_atomic_cas_strong_acq_rel(once, &expected, (uintptr_t)1); // try to set to 1
+}
+
+typedef _Atomic(uintptr_t) mi_atomic_guard_t;
+
+// Allows only one thread to execute at a time
+#define mi_atomic_guard(guard) \
+  uintptr_t _mi_guard_expected = 0; \
+  for(bool _mi_guard_once = true; \
+      _mi_guard_once && mi_atomic_cas_strong_acq_rel(guard,&_mi_guard_expected,(uintptr_t)1); \
+      (mi_atomic_store_release(guard,(uintptr_t)0), _mi_guard_once = false) )
+
+
+
+// Yield
+#if defined(__cplusplus)
+#include <thread>
+static inline void mi_atomic_yield(void) {
+  std::this_thread::yield();
+}
+#elif defined(_WIN32)
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+static inline void mi_atomic_yield(void) {
+  YieldProcessor();
+}
+#elif defined(__SSE2__)
+#include <emmintrin.h>
+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__)
+#if defined(__x86_64__) || defined(__i386__)
+static inline void mi_atomic_yield(void) {
+  __asm__ volatile ("pause" ::: "memory");
+}
+#elif defined(__aarch64__)
+static inline void mi_atomic_yield(void) {
+  __asm__ volatile("wfe");
+}
+#elif (defined(__arm__) && __ARM_ARCH__ >= 7)
+static inline void mi_atomic_yield(void) {
+  __asm__ volatile("yield" ::: "memory");
+}
+#elif defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__)
+#ifdef __APPLE__
+static inline void mi_atomic_yield(void) {
+  __asm__ volatile ("or r27,r27,r27" ::: "memory");
+}
+#else
+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 <synch.h>
+static inline void mi_atomic_yield(void) {
+  smt_pause();
+}
+#elif defined(__wasi__)
+#include <sched.h>
+static inline void mi_atomic_yield(void) {
+  sched_yield();
+}
+#else
+#include <unistd.h>
+static inline void mi_atomic_yield(void) {
+  sleep(0);
+}
+#endif
+
+
+#endif // __MIMALLOC_ATOMIC_H
diff --git a/Include/internal/mimalloc/mimalloc/internal.h b/Include/internal/mimalloc/mimalloc/internal.h
new file mode 100644
index 0000000..f076bc6
--- /dev/null
+++ b/Include/internal/mimalloc/mimalloc/internal.h
@@ -0,0 +1,979 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2023, 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_INTERNAL_H
+#define MIMALLOC_INTERNAL_H
+
+
+// --------------------------------------------------------------------------
+// This file contains the interal API's of mimalloc and various utility
+// functions and macros.
+// --------------------------------------------------------------------------
+
+#include "mimalloc/types.h"
+#include "mimalloc/track.h"
+
+#if (MI_DEBUG>0)
+#define mi_trace_message(...)  _mi_trace_message(__VA_ARGS__)
+#else
+#define mi_trace_message(...)
+#endif
+
+#define MI_CACHE_LINE          64
+#if defined(_MSC_VER)
+#pragma warning(disable:4127)   // suppress constant conditional warning (due to MI_SECURE paths)
+#pragma warning(disable:26812)  // unscoped enum warning
+#define mi_decl_noinline        __declspec(noinline)
+#define mi_decl_thread          __declspec(thread)
+#define mi_decl_cache_align     __declspec(align(MI_CACHE_LINE))
+#elif (defined(__GNUC__) && (__GNUC__ >= 3)) || defined(__clang__) // includes clang and icc
+#define mi_decl_noinline        __attribute__((noinline))
+#define mi_decl_thread          __thread
+#define mi_decl_cache_align     __attribute__((aligned(MI_CACHE_LINE)))
+#else
+#define mi_decl_noinline
+#define mi_decl_thread          __thread        // hope for the best :-)
+#define mi_decl_cache_align
+#endif
+
+#if defined(__EMSCRIPTEN__) && !defined(__wasi__)
+#define __wasi__
+#endif
+
+#if defined(__cplusplus)
+#define mi_decl_externc       extern "C"
+#else
+#define mi_decl_externc
+#endif
+
+// pthreads
+#if !defined(_WIN32) && !defined(__wasi__)
+#define  MI_USE_PTHREADS
+#include <pthread.h>
+#endif
+
+// "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_warning_message(const char* fmt, ...);
+void       _mi_verbose_message(const char* fmt, ...);
+void       _mi_trace_message(const char* fmt, ...);
+void       _mi_options_init(void);
+void       _mi_error_message(int err, const char* fmt, ...);
+
+// random.c
+void       _mi_random_init(mi_random_ctx_t* ctx);
+void       _mi_random_init_weak(mi_random_ctx_t* ctx);
+void       _mi_random_reinit_if_weak(mi_random_ctx_t * ctx);
+void       _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* new_ctx);
+uintptr_t  _mi_random_next(mi_random_ctx_t* ctx);
+uintptr_t  _mi_heap_random_next(mi_heap_t* heap);
+uintptr_t  _mi_os_random_weak(uintptr_t extra_seed);
+static inline uintptr_t _mi_random_shuffle(uintptr_t x);
+
+// init.c
+extern mi_decl_cache_align mi_stats_t       _mi_stats_main;
+extern mi_decl_cache_align const mi_page_t  _mi_page_empty;
+bool       _mi_is_main_thread(void);
+size_t     _mi_current_thread_count(void);
+bool       _mi_preloading(void);           // true while the C runtime is not initialized yet
+mi_threadid_t _mi_thread_id(void) mi_attr_noexcept;
+mi_heap_t*    _mi_heap_main_get(void);     // statically allocated main backing heap
+void       _mi_thread_done(mi_heap_t* heap);
+void       _mi_thread_data_collect(void);
+
+// os.c
+void       _mi_os_init(void);                                            // called from process init
+void*      _mi_os_alloc(size_t size, mi_memid_t* memid, mi_stats_t* stats);
+void       _mi_os_free(void* p, size_t size, mi_memid_t memid, mi_stats_t* stats);
+void       _mi_os_free_ex(void* p, size_t size, bool still_committed, mi_memid_t memid, mi_stats_t* stats);
+
+size_t     _mi_os_page_size(void);
+size_t     _mi_os_good_alloc_size(size_t size);
+bool       _mi_os_has_overcommit(void);
+bool       _mi_os_has_virtual_reserve(void);
+
+bool       _mi_os_purge(void* p, size_t size, mi_stats_t* stats);
+bool       _mi_os_reset(void* addr, size_t size, mi_stats_t* tld_stats);
+bool       _mi_os_commit(void* p, size_t size, bool* is_zero, mi_stats_t* stats);
+bool       _mi_os_decommit(void* addr, size_t size, mi_stats_t* stats);
+bool       _mi_os_protect(void* addr, size_t size);
+bool       _mi_os_unprotect(void* addr, size_t size);
+bool       _mi_os_purge(void* p, size_t size, mi_stats_t* stats);
+bool       _mi_os_purge_ex(void* p, size_t size, bool allow_reset, mi_stats_t* stats);
+
+void*      _mi_os_alloc_aligned(size_t size, size_t alignment, bool commit, bool allow_large, mi_memid_t* memid, mi_stats_t* stats);
+void*      _mi_os_alloc_aligned_at_offset(size_t size, size_t alignment, size_t align_offset, bool commit, bool allow_large, mi_memid_t* memid, mi_stats_t* tld_stats);
+
+void*      _mi_os_get_aligned_hint(size_t try_alignment, size_t size);
+bool       _mi_os_use_large_page(size_t size, size_t alignment);
+size_t     _mi_os_large_page_size(void);
+
+void*      _mi_os_alloc_huge_os_pages(size_t pages, int numa_node, mi_msecs_t max_secs, size_t* pages_reserved, size_t* psize, mi_memid_t* memid);
+
+// arena.c
+mi_arena_id_t _mi_arena_id_none(void);
+void       _mi_arena_free(void* p, size_t size, size_t still_committed_size, mi_memid_t memid, mi_stats_t* stats);
+void*      _mi_arena_alloc(size_t size, bool commit, bool allow_large, mi_arena_id_t req_arena_id, mi_memid_t* memid, mi_os_tld_t* tld);
+void*      _mi_arena_alloc_aligned(size_t size, size_t alignment, size_t align_offset, bool commit, bool allow_large, mi_arena_id_t req_arena_id, mi_memid_t* memid, mi_os_tld_t* tld);
+bool       _mi_arena_memid_is_suitable(mi_memid_t memid, mi_arena_id_t request_arena_id);
+bool       _mi_arena_contains(const void* p);
+void       _mi_arena_collect(bool force_purge, mi_stats_t* stats);
+void       _mi_arena_unsafe_destroy_all(mi_stats_t* stats);
+
+// "segment-map.c"
+void       _mi_segment_map_allocated_at(const mi_segment_t* segment);
+void       _mi_segment_map_freed_at(const mi_segment_t* segment);
+
+// "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, mi_os_tld_t* os_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_thread_collect(mi_segments_tld_t* tld);
+
+#if MI_HUGE_PAGE_ABANDON
+void       _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block_t* block);
+#else
+void       _mi_segment_huge_page_reset(mi_segment_t* segment, mi_page_t* page, mi_block_t* block);
+#endif
+
+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_await_readers(void);
+void       _mi_abandoned_collect(mi_heap_t* heap, bool force, mi_segments_tld_t* tld);
+
+// "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;
+
+void       _mi_page_retire(mi_page_t* page) mi_attr_noexcept;                  // free the page if there are no other pages with many free blocks
+void       _mi_page_unfull(mi_page_t* page);
+void       _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force);   // free the page
+void       _mi_page_abandon(mi_page_t* page, mi_page_queue_t* pq);            // abandon the page, to be picked up by another thread...
+void       _mi_heap_delayed_free_all(mi_heap_t* heap);
+bool       _mi_heap_delayed_free_partial(mi_heap_t* heap);
+void       _mi_heap_collect_retired(mi_heap_t* heap, bool force);
+
+void       _mi_page_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never);
+bool       _mi_page_try_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never);
+size_t     _mi_page_queue_append(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_queue_t* append);
+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
+
+// "heap.c"
+void       _mi_heap_destroy_pages(mi_heap_t* heap);
+void       _mi_heap_collect_abandon(mi_heap_t* heap);
+void       _mi_heap_set_default_direct(mi_heap_t* heap);
+bool       _mi_heap_memid_is_suitable(mi_heap_t* heap, mi_memid_t memid);
+void       _mi_heap_unsafe_destroy_all(void);
+
+// "stats.c"
+void       _mi_stats_done(mi_stats_t* stats);
+mi_msecs_t  _mi_clock_now(void);
+mi_msecs_t  _mi_clock_end(mi_msecs_t start);
+mi_msecs_t  _mi_clock_start(void);
+
+// "alloc.c"
+void*       _mi_page_malloc(mi_heap_t* heap, mi_page_t* page, size_t size, bool zero) mi_attr_noexcept;  // called from `_mi_malloc_generic`
+void*       _mi_heap_malloc_zero(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept;
+void*       _mi_heap_malloc_zero_ex(mi_heap_t* heap, size_t size, bool zero, size_t huge_alignment) mi_attr_noexcept;     // called from `_mi_heap_malloc_aligned`
+void*       _mi_heap_realloc_zero(mi_heap_t* heap, void* p, size_t newsize, bool zero) mi_attr_noexcept;
+mi_block_t* _mi_page_ptr_unalign(const mi_segment_t* segment, const mi_page_t* page, const void* p);
+bool        _mi_free_delayed_block(mi_block_t* block);
+void        _mi_free_generic(const mi_segment_t* segment, mi_page_t* page, bool is_local, void* p) mi_attr_noexcept;  // for runtime integration
+void        _mi_padding_shrink(const mi_page_t* page, const mi_block_t* block, const size_t min_size);
+
+// option.c, c primitives
+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);
+void        _mi_strlcat(char* dest, const char* src, size_t dest_size);
+size_t      _mi_strlen(const char* s);
+size_t      _mi_strnlen(const char* s, size_t max_len);
+
+
+#if MI_DEBUG>1
+bool        _mi_page_is_valid(mi_page_t* page);
+#endif
+
+
+// ------------------------------------------------------
+// Branches
+// ------------------------------------------------------
+
+#if defined(__GNUC__) || defined(__clang__)
+#define mi_unlikely(x)     (__builtin_expect(!!(x),false))
+#define mi_likely(x)       (__builtin_expect(!!(x),true))
+#elif (defined(__cplusplus) && (__cplusplus >= 202002L)) || (defined(_MSVC_LANG) && _MSVC_LANG >= 202002L)
+#define mi_unlikely(x)     (x) [[unlikely]]
+#define mi_likely(x)       (x) [[likely]]
+#else
+#define mi_unlikely(x)     (x)
+#define mi_likely(x)       (x)
+#endif
+
+#ifndef __has_builtin
+#define __has_builtin(x)  0
+#endif
+
+
+/* -----------------------------------------------------------
+  Error codes passed to `_mi_fatal_error`
+  All are recoverable but EFAULT is a serious error and aborts by default in secure mode.
+  For portability define undefined error codes using common Unix codes:
+  <https://www-numi.fnal.gov/offline_software/srt_public_context/WebDocs/Errors/unix_system_errors.html>
+----------------------------------------------------------- */
+#include <errno.h>
+#ifndef EAGAIN         // double free
+#define EAGAIN (11)
+#endif
+#ifndef ENOMEM         // out of memory
+#define ENOMEM (12)
+#endif
+#ifndef EFAULT         // corrupted free-list or meta-data
+#define EFAULT (14)
+#endif
+#ifndef EINVAL         // trying to free an invalid pointer
+#define EINVAL (22)
+#endif
+#ifndef EOVERFLOW      // count*size overflow
+#define EOVERFLOW (75)
+#endif
+
+
+/* -----------------------------------------------------------
+  Inlined definitions
+----------------------------------------------------------- */
+#define MI_UNUSED(x)     (void)(x)
+#if (MI_DEBUG>0)
+#define MI_UNUSED_RELEASE(x)
+#else
+#define MI_UNUSED_RELEASE(x)  MI_UNUSED(x)
+#endif
+
+#define MI_INIT4(x)   x(),x(),x(),x()
+#define MI_INIT8(x)   MI_INIT4(x),MI_INIT4(x)
+#define MI_INIT16(x)  MI_INIT8(x),MI_INIT8(x)
+#define MI_INIT32(x)  MI_INIT16(x),MI_INIT16(x)
+#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)
+
+
+#include <string.h>
+// initialize a local variable to zero; use memset as compilers optimize constant sized memset's
+#define _mi_memzero_var(x)  memset(&x,0,sizeof(x))
+
+// Is `x` a power of two? (0 is considered a power of two)
+static inline bool _mi_is_power_of_two(uintptr_t x) {
+  return ((x & (x - 1)) == 0);
+}
+
+// Is a pointer aligned?
+static inline bool _mi_is_aligned(void* p, size_t alignment) {
+  mi_assert_internal(alignment != 0);
+  return (((uintptr_t)p % alignment) == 0);
+}
+
+// Align upwards
+static inline uintptr_t _mi_align_up(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) & ~mask);
+  }
+  else {
+    return (((sz + mask)/alignment)*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);
+  }
+}
+
+// Divide upwards: `s <= _mi_divide_up(s,d)*d < s+d`.
+static inline uintptr_t _mi_divide_up(uintptr_t size, size_t divider) {
+  mi_assert_internal(divider != 0);
+  return (divider == 0 ? size : ((size + divider - 1) / divider));
+}
+
+// Is memory zero initialized?
+static inline bool mi_mem_is_zero(const void* p, size_t size) {
+  for (size_t i = 0; i < size; i++) {
+    if (((uint8_t*)p)[i] != 0) return false;
+  }
+  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) {
+  mi_assert_internal(size <= SIZE_MAX - sizeof(uintptr_t));
+  return (size + sizeof(uintptr_t) - 1) / sizeof(uintptr_t);
+}
+
+// Overflow detecting multiply
+#if __has_builtin(__builtin_umul_overflow) || (defined(__GNUC__) && (__GNUC__ >= 5))
+#include <limits.h>      // UINT_MAX, ULONG_MAX
+#if defined(_CLOCK_T)    // for Illumos
+#undef _CLOCK_T
+#endif
+static inline bool mi_mul_overflow(size_t count, size_t size, size_t* total) {
+  #if (SIZE_MAX == ULONG_MAX)
+    return __builtin_umull_overflow(count, size, (unsigned long *)total);
+  #elif (SIZE_MAX == UINT_MAX)
+    return __builtin_umul_overflow(count, size, (unsigned int *)total);
+  #else
+    return __builtin_umulll_overflow(count, size, (unsigned long long *)total);
+  #endif
+}
+#else /* __builtin_umul_overflow is unavailable */
+static inline bool mi_mul_overflow(size_t count, size_t size, size_t* total) {
+  #define MI_MUL_NO_OVERFLOW ((size_t)1 << (4*sizeof(size_t)))  // sqrt(SIZE_MAX)
+  *total = count * size;
+  // note: gcc/clang optimize this to directly check the overflow flag
+  return ((size >= MI_MUL_NO_OVERFLOW || count >= MI_MUL_NO_OVERFLOW) && size > 0 && (SIZE_MAX / size) < count);
+}
+#endif
+
+// Safe multiply `count*size` into `total`; return `true` on overflow.
+static inline bool mi_count_size_overflow(size_t count, size_t size, size_t* total) {
+  if (count==1) {  // quick check for the case where count is one (common for C++ allocators)
+    *total = size;
+    return false;
+  }
+  else if mi_unlikely(mi_mul_overflow(count, size, total)) {
+    #if MI_DEBUG > 0
+    _mi_error_message(EOVERFLOW, "allocation request is too large (%zu * %zu bytes)\n", count, size);
+    #endif
+    *total = SIZE_MAX;
+    return true;
+  }
+  else return false;
+}
+
+
+/*----------------------------------------------------------------------------------------
+  Heap functions
+------------------------------------------------------------------------------------------- */
+
+extern const mi_heap_t _mi_heap_empty;  // read-only empty heap, initial value of the thread local default heap
+
+static inline bool mi_heap_is_backing(const mi_heap_t* heap) {
+  return (heap->tld->heap_backing == heap);
+}
+
+static inline bool mi_heap_is_initialized(mi_heap_t* heap) {
+  mi_assert_internal(heap != NULL);
+  return (heap != &_mi_heap_empty);
+}
+
+static inline uintptr_t _mi_ptr_cookie(const void* p) {
+  extern mi_heap_t _mi_heap_main;
+  mi_assert_internal(_mi_heap_main.cookie != 0);
+  return ((uintptr_t)p ^ _mi_heap_main.cookie);
+}
+
+/* -----------------------------------------------------------
+  Pages
+----------------------------------------------------------- */
+
+static inline mi_page_t* _mi_heap_get_free_small_page(mi_heap_t* heap, size_t size) {
+  mi_assert_internal(size <= (MI_SMALL_SIZE_MAX + MI_PADDING_SIZE));
+  const size_t idx = _mi_wsize_from_size(size);
+  mi_assert_internal(idx < MI_PAGES_DIRECT);
+  return heap->pages_free_direct[idx];
+}
+
+// Segment that contains the pointer
+// Large aligned blocks may be aligned at N*MI_SEGMENT_SIZE (inside a huge segment > MI_SEGMENT_SIZE),
+// and we need align "down" to the segment info which is `MI_SEGMENT_SIZE` bytes before it;
+// therefore we align one byte before `p`.
+static inline mi_segment_t* _mi_ptr_segment(const void* p) {
+  mi_assert_internal(p != NULL);
+  return (mi_segment_t*)(((uintptr_t)p - 1) & ~MI_SEGMENT_MASK);
+}
+
+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_segment_t* segment = _mi_ptr_segment(page);
+  mi_assert_internal(segment == NULL || ((mi_slice_t*)page >= segment->slices && (mi_slice_t*)page < segment->slices + segment->slice_entries));
+  return segment;
+}
+
+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 (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) {
+  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
+static inline uint8_t* _mi_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size) {
+  return _mi_segment_page_start(segment, page, page_size);
+}
+
+// Get the page containing the pointer
+static inline mi_page_t* _mi_ptr_page(void* p) {
+  return _mi_segment_page_of(_mi_ptr_segment(p), p);
+}
+
+// Get the block size of a page (special case for huge objects)
+static inline size_t mi_page_block_size(const mi_page_t* page) {
+  const size_t bsize = page->xblock_size;
+  mi_assert_internal(bsize > 0);
+  if mi_likely(bsize < MI_HUGE_BLOCK_SIZE) {
+    return bsize;
+  }
+  else {
+    size_t psize;
+    _mi_segment_page_start(_mi_page_segment(page), page, &psize);
+    return psize;
+  }
+}
+
+static inline bool mi_page_is_huge(const mi_page_t* page) {
+  return (_mi_page_segment(page)->kind == MI_SEGMENT_HUGE);
+}
+
+// Get the usable block size of a page without fixed padding.
+// This may still include internal padding due to alignment and rounding up size classes.
+static inline size_t mi_page_usable_block_size(const mi_page_t* page) {
+  return mi_page_block_size(page) - MI_PADDING_SIZE;
+}
+
+// size of a segment
+static inline size_t mi_segment_size(mi_segment_t* segment) {
+  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
+static inline mi_block_t* mi_page_thread_free(const mi_page_t* page) {
+  return (mi_block_t*)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xthread_free) & ~3);
+}
+
+static inline mi_delayed_t mi_page_thread_free_flag(const mi_page_t* page) {
+  return (mi_delayed_t)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xthread_free) & 3);
+}
+
+// Heap access
+static inline mi_heap_t* mi_page_heap(const mi_page_t* page) {
+  return (mi_heap_t*)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xheap));
+}
+
+static inline void mi_page_set_heap(mi_page_t* page, mi_heap_t* heap) {
+  mi_assert_internal(mi_page_thread_free_flag(page) != MI_DELAYED_FREEING);
+  mi_atomic_store_release(&page->xheap,(uintptr_t)heap);
+}
+
+// Thread free flag helpers
+static inline mi_block_t* mi_tf_block(mi_thread_free_t tf) {
+  return (mi_block_t*)(tf & ~0x03);
+}
+static inline mi_delayed_t mi_tf_delayed(mi_thread_free_t tf) {
+  return (mi_delayed_t)(tf & 0x03);
+}
+static inline mi_thread_free_t mi_tf_make(mi_block_t* block, mi_delayed_t delayed) {
+  return (mi_thread_free_t)((uintptr_t)block | (uintptr_t)delayed);
+}
+static inline mi_thread_free_t mi_tf_set_delayed(mi_thread_free_t tf, mi_delayed_t delayed) {
+  return mi_tf_make(mi_tf_block(tf),delayed);
+}
+static inline mi_thread_free_t mi_tf_set_block(mi_thread_free_t tf, mi_block_t* block) {
+  return mi_tf_make(block, mi_tf_delayed(tf));
+}
+
+// are all blocks in a page freed?
+// note: needs up-to-date used count, (as the `xthread_free` list may not be empty). see `_mi_page_collect_free`.
+static inline bool mi_page_all_free(const mi_page_t* page) {
+  mi_assert_internal(page != NULL);
+  return (page->used == 0);
+}
+
+// are there any available blocks?
+static inline bool mi_page_has_any_available(const mi_page_t* page) {
+  mi_assert_internal(page != NULL && page->reserved > 0);
+  return (page->used < page->reserved || (mi_page_thread_free(page) != NULL));
+}
+
+// are there immediately available blocks, i.e. blocks available on the free list.
+static inline bool mi_page_immediate_available(const mi_page_t* page) {
+  mi_assert_internal(page != NULL);
+  return (page->free != NULL);
+}
+
+// is more than 7/8th of a page in use?
+static inline bool mi_page_mostly_used(const mi_page_t* page) {
+  if (page==NULL) return true;
+  uint16_t frac = page->reserved / 8U;
+  return (page->reserved - page->used <= frac);
+}
+
+static inline mi_page_queue_t* mi_page_queue(const mi_heap_t* heap, size_t size) {
+  return &((mi_heap_t*)heap)->pages[_mi_bin(size)];
+}
+
+
+
+//-----------------------------------------------------------
+// Page flags
+//-----------------------------------------------------------
+static inline bool mi_page_is_in_full(const mi_page_t* page) {
+  return page->flags.x.in_full;
+}
+
+static inline void mi_page_set_in_full(mi_page_t* page, bool in_full) {
+  page->flags.x.in_full = in_full;
+}
+
+static inline bool mi_page_has_aligned(const mi_page_t* page) {
+  return page->flags.x.has_aligned;
+}
+
+static inline void mi_page_set_has_aligned(mi_page_t* page, bool has_aligned) {
+  page->flags.x.has_aligned = has_aligned;
+}
+
+
+/* -------------------------------------------------------------------
+Encoding/Decoding the free list next pointers
+
+This is to protect against buffer overflow exploits where the
+free list is mutated. Many hardened allocators xor the next pointer `p`
+with a secret key `k1`, as `p^k1`. This prevents overwriting with known
+values but might be still too weak: if the attacker can guess
+the pointer `p` this  can reveal `k1` (since `p^k1^p == k1`).
+Moreover, if multiple blocks can be read as well, the attacker can
+xor both as `(p1^k1) ^ (p2^k1) == p1^p2` which may reveal a lot
+about the pointers (and subsequently `k1`).
+
+Instead mimalloc uses an extra key `k2` and encodes as `((p^k2)<<<k1)+k1`.
+Since these operations are not associative, the above approaches do not
+work so well any more even if the `p` can be guesstimated. For example,
+for the read case we can subtract two entries to discard the `+k1` term,
+but that leads to `((p1^k2)<<<k1) - ((p2^k2)<<<k1)` at best.
+We include the left-rotation since xor and addition are otherwise linear
+in the lowest bit. Finally, both keys are unique per page which reduces
+the re-use of keys by a large factor.
+
+We also pass a separate `null` value to be used as `NULL` or otherwise
+`(k2<<<k1)+k1` would appear (too) often as a sentinel value.
+------------------------------------------------------------------- */
+
+static inline bool mi_is_in_same_segment(const void* p, const void* q) {
+  return (_mi_ptr_segment(p) == _mi_ptr_segment(q));
+}
+
+static inline bool mi_is_in_same_page(const void* p, const void* q) {
+  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) {
+  shift %= MI_INTPTR_BITS;
+  return (shift==0 ? x : ((x << shift) | (x >> (MI_INTPTR_BITS - shift))));
+}
+static inline uintptr_t mi_rotr(uintptr_t x, uintptr_t shift) {
+  shift %= MI_INTPTR_BITS;
+  return (shift==0 ? x : ((x >> shift) | (x << (MI_INTPTR_BITS - shift))));
+}
+
+static inline void* mi_ptr_decode(const void* null, const mi_encoded_t x, const uintptr_t* keys) {
+  void* p = (void*)(mi_rotr(x - keys[0], keys[0]) ^ keys[1]);
+  return (p==null ? NULL : p);
+}
+
+static inline mi_encoded_t mi_ptr_encode(const void* null, const void* p, const uintptr_t* keys) {
+  uintptr_t x = (uintptr_t)(p==NULL ? null : p);
+  return mi_rotl(x ^ keys[1], keys[0]) + keys[0];
+}
+
+static inline mi_block_t* mi_block_nextx( const void* null, const mi_block_t* block, const uintptr_t* keys ) {
+  mi_track_mem_defined(block,sizeof(mi_block_t));
+  mi_block_t* next;
+  #ifdef MI_ENCODE_FREELIST
+  next = (mi_block_t*)mi_ptr_decode(null, block->next, keys);
+  #else
+  MI_UNUSED(keys); MI_UNUSED(null);
+  next = (mi_block_t*)block->next;
+  #endif
+  mi_track_mem_noaccess(block,sizeof(mi_block_t));
+  return next;
+}
+
+static inline void mi_block_set_nextx(const void* null, mi_block_t* block, const mi_block_t* next, const uintptr_t* keys) {
+  mi_track_mem_undefined(block,sizeof(mi_block_t));
+  #ifdef MI_ENCODE_FREELIST
+  block->next = mi_ptr_encode(null, next, keys);
+  #else
+  MI_UNUSED(keys); MI_UNUSED(null);
+  block->next = (mi_encoded_t)next;
+  #endif
+  mi_track_mem_noaccess(block,sizeof(mi_block_t));
+}
+
+static inline mi_block_t* mi_block_next(const mi_page_t* page, const mi_block_t* block) {
+  #ifdef MI_ENCODE_FREELIST
+  mi_block_t* next = mi_block_nextx(page,block,page->keys);
+  // check for free list corruption: is `next` at least in the same page?
+  // TODO: check if `next` is `page->block_size` aligned?
+  if mi_unlikely(next!=NULL && !mi_is_in_same_page(block, next)) {
+    _mi_error_message(EFAULT, "corrupted free list entry of size %zub at %p: value 0x%zx\n", mi_page_block_size(page), block, (uintptr_t)next);
+    next = NULL;
+  }
+  return next;
+  #else
+  MI_UNUSED(page);
+  return mi_block_nextx(page,block,NULL);
+  #endif
+}
+
+static inline void mi_block_set_next(const mi_page_t* page, mi_block_t* block, const mi_block_t* next) {
+  #ifdef MI_ENCODE_FREELIST
+  mi_block_set_nextx(page,block,next, page->keys);
+  #else
+  MI_UNUSED(page);
+  mi_block_set_nextx(page,block,next,NULL);
+  #endif
+}
+
+
+// -------------------------------------------------------------------
+// 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
+----------------------------------------------------------- */
+
+static inline mi_memid_t _mi_memid_create(mi_memkind_t memkind) {
+  mi_memid_t memid;
+  _mi_memzero_var(memid);
+  memid.memkind = memkind;
+  return memid;
+}
+
+static inline mi_memid_t _mi_memid_none(void) {
+  return _mi_memid_create(MI_MEM_NONE);
+}
+
+static inline mi_memid_t _mi_memid_create_os(bool committed, bool is_zero, bool is_large) {
+  mi_memid_t memid = _mi_memid_create(MI_MEM_OS);
+  memid.initially_committed = committed;
+  memid.initially_zero = is_zero;
+  memid.is_pinned = is_large;
+  return memid;
+}
+
+
+// -------------------------------------------------------------------
+// Fast "random" shuffle
+// -------------------------------------------------------------------
+
+static inline uintptr_t _mi_random_shuffle(uintptr_t x) {
+  if (x==0) { x = 17; }   // ensure we don't get stuck in generating zeros
+#if (MI_INTPTR_SIZE==8)
+  // by Sebastiano Vigna, see: <http://xoshiro.di.unimi.it/splitmix64.c>
+  x ^= x >> 30;
+  x *= 0xbf58476d1ce4e5b9UL;
+  x ^= x >> 27;
+  x *= 0x94d049bb133111ebUL;
+  x ^= x >> 31;
+#elif (MI_INTPTR_SIZE==4)
+  // by Chris Wellons, see: <https://nullprogram.com/blog/2018/07/31/>
+  x ^= x >> 16;
+  x *= 0x7feb352dUL;
+  x ^= x >> 15;
+  x *= 0x846ca68bUL;
+  x ^= x >> 16;
+#endif
+  return x;
+}
+
+// -------------------------------------------------------------------
+// Optimize numa node access for the common case (= one node)
+// -------------------------------------------------------------------
+
+int    _mi_os_numa_node_get(mi_os_tld_t* tld);
+size_t _mi_os_numa_node_count_get(void);
+
+extern _Atomic(size_t) _mi_numa_node_count;
+static inline int _mi_os_numa_node(mi_os_tld_t* tld) {
+  if mi_likely(mi_atomic_load_relaxed(&_mi_numa_node_count) == 1) { return 0; }
+  else return _mi_os_numa_node_get(tld);
+}
+static inline size_t _mi_os_numa_node_count(void) {
+  const size_t count = mi_atomic_load_relaxed(&_mi_numa_node_count);
+  if mi_likely(count > 0) { return count; }
+  else return _mi_os_numa_node_count_get();
+}
+
+
+
+// -----------------------------------------------------------------------
+// Count bits: trailing or leading zeros (with MI_INTPTR_BITS on all zero)
+// -----------------------------------------------------------------------
+
+#if defined(__GNUC__)
+
+#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_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
+}
+
+#elif defined(_MSC_VER)
+
+#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;
+  unsigned long idx;
+#if (INTPTR_MAX == LONG_MAX)
+  _BitScanReverse(&idx, x);
+#else
+  _BitScanReverse64(&idx, x);
+#endif
+  return ((MI_INTPTR_BITS - 1) - idx);
+}
+static inline size_t mi_ctz(uintptr_t x) {
+  if (x==0) return MI_INTPTR_BITS;
+  unsigned long idx;
+#if (INTPTR_MAX == LONG_MAX)
+  _BitScanForward(&idx, x);
+#else
+  _BitScanForward64(&idx, x);
+#endif
+  return idx;
+}
+
+#else
+static inline size_t mi_ctz32(uint32_t x) {
+  // de Bruijn multiplication, see <http://supertech.csail.mit.edu/papers/debruijn.pdf>
+  static const unsigned char debruijn[32] = {
+    0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8,
+    31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9
+  };
+  if (x==0) return 32;
+  return debruijn[((x & -(int32_t)x) * 0x077CB531UL) >> 27];
+}
+static inline size_t mi_clz32(uint32_t x) {
+  // de Bruijn multiplication, see <http://supertech.csail.mit.edu/papers/debruijn.pdf>
+  static const uint8_t debruijn[32] = {
+    31, 22, 30, 21, 18, 10, 29, 2, 20, 17, 15, 13, 9, 6, 28, 1,
+    23, 19, 11, 3, 16, 14, 7, 24, 12, 4, 8, 25, 5, 26, 27, 0
+  };
+  if (x==0) return 32;
+  x |= x >> 1;
+  x |= x >> 2;
+  x |= x >> 4;
+  x |= x >> 8;
+  x |= x >> 16;
+  return debruijn[(uint32_t)(x * 0x07C4ACDDUL) >> 27];
+}
+
+static inline size_t mi_clz(uintptr_t x) {
+  if (x==0) return MI_INTPTR_BITS;
+#if (MI_INTPTR_BITS <= 32)
+  return mi_clz32((uint32_t)x);
+#else
+  size_t count = mi_clz32((uint32_t)(x >> 32));
+  if (count < 32) return count;
+  return (32 + mi_clz32((uint32_t)x));
+#endif
+}
+static inline size_t mi_ctz(uintptr_t x) {
+  if (x==0) return MI_INTPTR_BITS;
+#if (MI_INTPTR_BITS <= 32)
+  return mi_ctz32((uint32_t)x);
+#else
+  size_t count = mi_ctz32((uint32_t)x);
+  if (count < 32) return count;
+  return (32 + mi_ctz32((uint32_t)(x>>32)));
+#endif
+}
+
+#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));
+}
+
+
+// ---------------------------------------------------------------------------------
+// Provide our own `_mi_memcpy` for potential performance optimizations.
+//
+// For now, only on Windows with msvc/clang-cl we optimize to `rep movsb` if
+// we happen to run on x86/x64 cpu's that have "fast short rep movsb" (FSRM) support
+// (AMD Zen3+ (~2020) or Intel Ice Lake+ (~2017). See also issue #201 and pr #253.
+// ---------------------------------------------------------------------------------
+
+#if !MI_TRACK_ENABLED && defined(_WIN32) && (defined(_M_IX86) || defined(_M_X64))
+#include <intrin.h>
+extern bool _mi_cpu_has_fsrm;
+static inline void _mi_memcpy(void* dst, const void* src, size_t n) {
+  if (_mi_cpu_has_fsrm) {
+    __movsb((unsigned char*)dst, (const unsigned char*)src, n);
+  }
+  else {
+    memcpy(dst, src, n);
+  }
+}
+static inline void _mi_memzero(void* dst, size_t n) {
+  if (_mi_cpu_has_fsrm) {
+    __stosb((unsigned char*)dst, 0, n);
+  }
+  else {
+    memset(dst, 0, n);
+  }
+}
+#else
+static inline void _mi_memcpy(void* dst, const void* src, size_t n) {
+  memcpy(dst, src, n);
+}
+static inline void _mi_memzero(void* dst, size_t n) {
+  memset(dst, 0, n);
+}
+#endif
+
+// -------------------------------------------------------------------------------
+// The `_mi_memcpy_aligned` can be used if the pointers are machine-word aligned
+// This is used for example in `mi_realloc`.
+// -------------------------------------------------------------------------------
+
+#if (defined(__GNUC__) && (__GNUC__ >= 4)) || defined(__clang__)
+// On GCC/CLang we provide a hint that the pointers are word aligned.
+static inline void _mi_memcpy_aligned(void* dst, const void* src, size_t n) {
+  mi_assert_internal(((uintptr_t)dst % MI_INTPTR_SIZE == 0) && ((uintptr_t)src % MI_INTPTR_SIZE == 0));
+  void* adst = __builtin_assume_aligned(dst, MI_INTPTR_SIZE);
+  const void* asrc = __builtin_assume_aligned(src, MI_INTPTR_SIZE);
+  _mi_memcpy(adst, asrc, n);
+}
+
+static inline void _mi_memzero_aligned(void* dst, size_t n) {
+  mi_assert_internal((uintptr_t)dst % MI_INTPTR_SIZE == 0);
+  void* adst = __builtin_assume_aligned(dst, MI_INTPTR_SIZE);
+  _mi_memzero(adst, n);
+}
+#else
+// Default fallback on `_mi_memcpy`
+static inline void _mi_memcpy_aligned(void* dst, const void* src, size_t n) {
+  mi_assert_internal(((uintptr_t)dst % MI_INTPTR_SIZE == 0) && ((uintptr_t)src % MI_INTPTR_SIZE == 0));
+  _mi_memcpy(dst, src, n);
+}
+
+static inline void _mi_memzero_aligned(void* dst, size_t n) {
+  mi_assert_internal((uintptr_t)dst % MI_INTPTR_SIZE == 0);
+  _mi_memzero(dst, n);
+}
+#endif
+
+
+#endif
diff --git a/Include/internal/mimalloc/mimalloc/prim.h b/Include/internal/mimalloc/mimalloc/prim.h
new file mode 100644
index 0000000..4b9e4dc
--- /dev/null
+++ b/Include/internal/mimalloc/mimalloc/prim.h
@@ -0,0 +1,323 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2023, 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_PRIM_H
+#define MIMALLOC_PRIM_H
+
+
+// --------------------------------------------------------------------------
+// This file specifies the primitive portability API.
+// Each OS/host needs to implement these primitives, see `src/prim`
+// for implementations on Window, macOS, WASI, and Linux/Unix.
+//
+// note: on all primitive functions, we always have result parameters != NUL, and:
+//  addr != NULL and page aligned
+//  size > 0     and page aligned
+//  return value is an error code an int where 0 is success.
+// --------------------------------------------------------------------------
+
+// OS memory configuration
+typedef struct mi_os_mem_config_s {
+  size_t  page_size;            // 4KiB
+  size_t  large_page_size;      // 2MiB
+  size_t  alloc_granularity;    // smallest allocation size (on Windows 64KiB)
+  bool    has_overcommit;       // can we reserve more memory than can be actually committed?
+  bool    must_free_whole;      // must allocated blocks be freed as a whole (false for mmap, true 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
+void _mi_prim_mem_init( mi_os_mem_config_t* config );
+
+// Free OS memory
+int _mi_prim_free(void* addr, size_t size );
+
+// Allocate OS memory. Return NULL on error.
+// The `try_alignment` is just a hint and the returned pointer does not have to be aligned.
+// If `commit` is false, the virtual memory range only needs to be reserved (with no access)
+// which will later be committed explicitly using `_mi_prim_commit`.
+// `is_zero` is set to true if the memory was zero initialized (as on most OS's)
+// pre: !commit => !allow_large
+//      try_alignment >= _mi_os_page_size() and a power of 2
+int _mi_prim_alloc(size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero, void** addr);
+
+// Commit memory. Returns error code or 0 on success.
+// For example, on Linux this would make the memory PROT_READ|PROT_WRITE.
+// `is_zero` is set to true if the memory was zero initialized (e.g. on Windows)
+int _mi_prim_commit(void* addr, size_t size, bool* is_zero);
+
+// Decommit memory. Returns error code or 0 on success. The `needs_recommit` result is true
+// if the memory would need to be re-committed. For example, on Windows this is always true,
+// but on Linux we could use MADV_DONTNEED to decommit which does not need a recommit.
+// pre: needs_recommit != NULL
+int _mi_prim_decommit(void* addr, size_t size, bool* needs_recommit);
+
+// Reset memory. The range keeps being accessible but the content might be reset.
+// Returns error code or 0 on success.
+int _mi_prim_reset(void* addr, size_t size);
+
+// Protect memory. Returns error code or 0 on success.
+int _mi_prim_protect(void* addr, size_t size, bool protect);
+
+// Allocate huge (1GiB) pages possibly associated with a NUMA node.
+// `is_zero` is set to true if the memory was zero initialized (as on most OS's)
+// pre: size > 0  and a multiple of 1GiB.
+//      numa_node is either negative (don't care), or a numa node number.
+int _mi_prim_alloc_huge_os_pages(void* hint_addr, size_t size, int numa_node, bool* is_zero, void** addr);
+
+// Return the current NUMA node
+size_t _mi_prim_numa_node(void);
+
+// Return the number of logical NUMA nodes
+size_t _mi_prim_numa_node_count(void);
+
+// Clock ticks
+mi_msecs_t _mi_prim_clock_now(void);
+
+// Return process information (only for statistics)
+typedef struct mi_process_info_s {
+  mi_msecs_t  elapsed;
+  mi_msecs_t  utime;
+  mi_msecs_t  stime;
+  size_t      current_rss;
+  size_t      peak_rss;
+  size_t      current_commit;
+  size_t      peak_commit;
+  size_t      page_faults;
+} mi_process_info_t;
+
+void _mi_prim_process_info(mi_process_info_t* pinfo);
+
+// Default stderr output. (only for warnings etc. with verbose enabled)
+// msg != NULL && _mi_strlen(msg) > 0
+void _mi_prim_out_stderr( const char* msg );
+
+// Get an environment variable. (only for options)
+// name != NULL, result != NULL, result_size >= 64
+bool _mi_prim_getenv(const char* name, char* result, size_t result_size);
+
+
+// Fill a buffer with strong randomness; return `false` on error or if
+// there is no strong randomization available.
+bool _mi_prim_random_buf(void* buf, size_t buf_len);
+
+// Called on the first thread start, and should ensure `_mi_thread_done` is called on thread termination.
+void _mi_prim_thread_init_auto_done(void);
+
+// Called on process exit and may take action to clean up resources associated with the thread auto done.
+void _mi_prim_thread_done_auto_done(void);
+
+// Called when the default heap for a thread changes
+void _mi_prim_thread_associate_default_heap(mi_heap_t* heap);
+
+
+//-------------------------------------------------------------------
+// Thread id: `_mi_prim_thread_id()`
+//
+// Getting the thread id should be performant as it is called in the
+// fast path of `_mi_free` and we specialize for various platforms as
+// inlined definitions. Regular code should call `init.c:_mi_thread_id()`.
+// We only require _mi_prim_thread_id() to return a unique id
+// for each thread (unequal to zero).
+//-------------------------------------------------------------------
+
+// defined in `init.c`; do not use these directly
+extern mi_decl_thread mi_heap_t* _mi_heap_default;  // default heap to allocate from
+extern bool _mi_process_is_initialized;             // has mi_process_init been called?
+
+static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept;
+
+#if defined(_WIN32)
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept {
+  // Windows: works on Intel and ARM in both 32- and 64-bit
+  return (uintptr_t)NtCurrentTeb();
+}
+
+// We use assembly for a fast thread id on the main platforms. The TLS layout depends on
+// both the OS and libc implementation so we use specific tests for each main platform.
+// If you test on another platform and it works please send a PR :-)
+// see also https://akkadia.org/drepper/tls.pdf for more info on the TLS register.
+#elif defined(__GNUC__) && ( \
+           (defined(__GLIBC__)   && (defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__aarch64__))) \
+        || (defined(__APPLE__)   && (defined(__x86_64__) || defined(__aarch64__))) \
+        || (defined(__BIONIC__)  && (defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__aarch64__))) \
+        || (defined(__FreeBSD__) && (defined(__x86_64__) || defined(__i386__) || defined(__aarch64__))) \
+        || (defined(__OpenBSD__) && (defined(__x86_64__) || defined(__i386__) || defined(__aarch64__))) \
+      )
+
+static inline void* mi_prim_tls_slot(size_t slot) mi_attr_noexcept {
+  void* res;
+  const size_t ofs = (slot*sizeof(void*));
+  #if defined(__i386__)
+    __asm__("movl %%gs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : );  // x86 32-bit always uses GS
+  #elif defined(__APPLE__) && defined(__x86_64__)
+    __asm__("movq %%gs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : );  // x86_64 macOSX uses GS
+  #elif defined(__x86_64__) && (MI_INTPTR_SIZE==4)
+    __asm__("movl %%fs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : );  // x32 ABI
+  #elif defined(__x86_64__)
+    __asm__("movq %%fs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : );  // x86_64 Linux, BSD uses FS
+  #elif defined(__arm__)
+    void** tcb; MI_UNUSED(ofs);
+    __asm__ volatile ("mrc p15, 0, %0, c13, c0, 3\nbic %0, %0, #3" : "=r" (tcb));
+    res = tcb[slot];
+  #elif defined(__aarch64__)
+    void** tcb; MI_UNUSED(ofs);
+    #if defined(__APPLE__) // M1, issue #343
+    __asm__ volatile ("mrs %0, tpidrro_el0\nbic %0, %0, #7" : "=r" (tcb));
+    #else
+    __asm__ volatile ("mrs %0, tpidr_el0" : "=r" (tcb));
+    #endif
+    res = tcb[slot];
+  #endif
+  return res;
+}
+
+// setting a tls slot is only used on macOS for now
+static inline void mi_prim_tls_slot_set(size_t slot, void* value) mi_attr_noexcept {
+  const size_t ofs = (slot*sizeof(void*));
+  #if defined(__i386__)
+    __asm__("movl %1,%%gs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : );  // 32-bit always uses GS
+  #elif defined(__APPLE__) && defined(__x86_64__)
+    __asm__("movq %1,%%gs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : );  // x86_64 macOS uses GS
+  #elif defined(__x86_64__) && (MI_INTPTR_SIZE==4)
+    __asm__("movl %1,%%fs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : );  // x32 ABI
+  #elif defined(__x86_64__)
+    __asm__("movq %1,%%fs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : );  // x86_64 Linux, BSD uses FS
+  #elif defined(__arm__)
+    void** tcb; MI_UNUSED(ofs);
+    __asm__ volatile ("mrc p15, 0, %0, c13, c0, 3\nbic %0, %0, #3" : "=r" (tcb));
+    tcb[slot] = value;
+  #elif defined(__aarch64__)
+    void** tcb; MI_UNUSED(ofs);
+    #if defined(__APPLE__) // M1, issue #343
+    __asm__ volatile ("mrs %0, tpidrro_el0\nbic %0, %0, #7" : "=r" (tcb));
+    #else
+    __asm__ volatile ("mrs %0, tpidr_el0" : "=r" (tcb));
+    #endif
+    tcb[slot] = value;
+  #endif
+}
+
+static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept {
+  #if defined(__BIONIC__)
+    // issue #384, #495: on the Bionic libc (Android), slot 1 is the thread id
+    // see: https://github.com/aosp-mirror/platform_bionic/blob/c44b1d0676ded732df4b3b21c5f798eacae93228/libc/platform/bionic/tls_defines.h#L86
+    return (uintptr_t)mi_prim_tls_slot(1);
+  #else
+    // in all our other targets, slot 0 is the thread id
+    // glibc: https://sourceware.org/git/?p=glibc.git;a=blob_plain;f=sysdeps/x86_64/nptl/tls.h
+    // apple: https://github.com/apple/darwin-xnu/blob/main/libsyscall/os/tsd.h#L36
+    return (uintptr_t)mi_prim_tls_slot(0);
+  #endif
+}
+
+#else
+
+// otherwise use portable C, taking the address of a thread local variable (this is still very fast on most platforms).
+static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept {
+  return (uintptr_t)&_mi_heap_default;
+}
+
+#endif
+
+
+
+/* ----------------------------------------------------------------------------------------
+The thread local default heap: `_mi_prim_get_default_heap()`
+This is inlined here as it is on the fast path for allocation functions.
+
+On most platforms (Windows, Linux, FreeBSD, NetBSD, etc), this just returns a
+__thread local variable (`_mi_heap_default`). With the initial-exec TLS model this ensures
+that the storage will always be available (allocated on the thread stacks).
+
+On some platforms though we cannot use that when overriding `malloc` since the underlying
+TLS implementation (or the loader) will call itself `malloc` on a first access and recurse.
+We try to circumvent this in an efficient way:
+- macOSX : we use an unused TLS slot from the OS allocated slots (MI_TLS_SLOT). On OSX, the
+           loader itself calls `malloc` even before the modules are initialized.
+- OpenBSD: we use an unused slot from the pthread block (MI_TLS_PTHREAD_SLOT_OFS).
+- DragonFly: defaults are working but seem slow compared to freeBSD (see PR #323)
+------------------------------------------------------------------------------------------- */
+
+static inline mi_heap_t* mi_prim_get_default_heap(void);
+
+#if defined(MI_MALLOC_OVERRIDE)
+#if defined(__APPLE__) // macOS
+  #define MI_TLS_SLOT               89  // seems unused?
+  // #define MI_TLS_RECURSE_GUARD 1
+  // other possible unused ones are 9, 29, __PTK_FRAMEWORK_JAVASCRIPTCORE_KEY4 (94), __PTK_FRAMEWORK_GC_KEY9 (112) and __PTK_FRAMEWORK_OLDGC_KEY9 (89)
+  // see <https://github.com/rweichler/substrate/blob/master/include/pthread_machdep.h>
+#elif defined(__OpenBSD__)
+  // use end bytes of a name; goes wrong if anyone uses names > 23 characters (ptrhread specifies 16)
+  // see <https://github.com/openbsd/src/blob/master/lib/libc/include/thread_private.h#L371>
+  #define MI_TLS_PTHREAD_SLOT_OFS   (6*sizeof(int) + 4*sizeof(void*) + 24)
+  // #elif defined(__DragonFly__)
+  // #warning "mimalloc is not working correctly on DragonFly yet."
+  // #define MI_TLS_PTHREAD_SLOT_OFS   (4 + 1*sizeof(void*))  // offset `uniqueid` (also used by gdb?) <https://github.com/DragonFlyBSD/DragonFlyBSD/blob/master/lib/libthread_xu/thread/thr_private.h#L458>
+#elif defined(__ANDROID__)
+  // See issue #381
+  #define MI_TLS_PTHREAD
+#endif
+#endif
+
+
+#if defined(MI_TLS_SLOT)
+
+static inline mi_heap_t* mi_prim_get_default_heap(void) {
+  mi_heap_t* heap = (mi_heap_t*)mi_prim_tls_slot(MI_TLS_SLOT);
+  if mi_unlikely(heap == NULL) {
+    #ifdef __GNUC__
+    __asm(""); // prevent conditional load of the address of _mi_heap_empty
+    #endif
+    heap = (mi_heap_t*)&_mi_heap_empty;
+  }
+  return heap;
+}
+
+#elif defined(MI_TLS_PTHREAD_SLOT_OFS)
+
+static inline mi_heap_t** mi_prim_tls_pthread_heap_slot(void) {
+  pthread_t self = pthread_self();
+  #if defined(__DragonFly__)
+  if (self==NULL) return NULL;
+  #endif
+  return (mi_heap_t**)((uint8_t*)self + MI_TLS_PTHREAD_SLOT_OFS);
+}
+
+static inline mi_heap_t* mi_prim_get_default_heap(void) {
+  mi_heap_t** pheap = mi_prim_tls_pthread_heap_slot();
+  if mi_unlikely(pheap == NULL) return _mi_heap_main_get();
+  mi_heap_t* heap = *pheap;
+  if mi_unlikely(heap == NULL) return (mi_heap_t*)&_mi_heap_empty;
+  return heap;
+}
+
+#elif defined(MI_TLS_PTHREAD)
+
+extern pthread_key_t _mi_heap_default_key;
+static inline mi_heap_t* mi_prim_get_default_heap(void) {
+  mi_heap_t* heap = (mi_unlikely(_mi_heap_default_key == (pthread_key_t)(-1)) ? _mi_heap_main_get() : (mi_heap_t*)pthread_getspecific(_mi_heap_default_key));
+  return (mi_unlikely(heap == NULL) ? (mi_heap_t*)&_mi_heap_empty : heap);
+}
+
+#else // default using a thread local variable; used on most platforms.
+
+static inline mi_heap_t* mi_prim_get_default_heap(void) {
+  #if defined(MI_TLS_RECURSE_GUARD)
+  if (mi_unlikely(!_mi_process_is_initialized)) return _mi_heap_main_get();
+  #endif
+  return _mi_heap_default;
+}
+
+#endif  // mi_prim_get_default_heap()
+
+
+
+#endif  // MIMALLOC_PRIM_H
diff --git a/Include/internal/mimalloc/mimalloc/track.h b/Include/internal/mimalloc/mimalloc/track.h
new file mode 100644
index 0000000..fa1a048
--- /dev/null
+++ b/Include/internal/mimalloc/mimalloc/track.h
@@ -0,0 +1,147 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2023, 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_TRACK_H
+#define MIMALLOC_TRACK_H
+
+/* ------------------------------------------------------------------------------------------------------
+Track memory ranges with macros for tools like Valgrind address sanitizer, or other memory checkers.
+These can be defined for tracking allocation:
+
+  #define mi_track_malloc_size(p,reqsize,size,zero)
+  #define mi_track_free_size(p,_size)
+
+The macros are set up such that the size passed to `mi_track_free_size`
+always matches the size of `mi_track_malloc_size`. (currently, `size == mi_usable_size(p)`).
+The `reqsize` is what the user requested, and `size >= reqsize`.
+The `size` is either byte precise (and `size==reqsize`) if `MI_PADDING` is enabled,
+or otherwise it is the usable block size which may be larger than the original request.
+Use `_mi_block_size_of(void* p)` to get the full block size that was allocated (including padding etc).
+The `zero` parameter is `true` if the allocated block is zero initialized.
+
+Optional:
+
+  #define mi_track_align(p,alignedp,offset,size)
+  #define mi_track_resize(p,oldsize,newsize)
+  #define mi_track_init()
+
+The `mi_track_align` is called right after a `mi_track_malloc` for aligned pointers in a block.
+The corresponding `mi_track_free` still uses the block start pointer and original size (corresponding to the `mi_track_malloc`).
+The `mi_track_resize` is currently unused but could be called on reallocations within a block.
+`mi_track_init` is called at program start.
+
+The following macros are for tools like asan and valgrind to track whether memory is
+defined, undefined, or not accessible at all:
+
+  #define mi_track_mem_defined(p,size)
+  #define mi_track_mem_undefined(p,size)
+  #define mi_track_mem_noaccess(p,size)
+
+-------------------------------------------------------------------------------------------------------*/
+
+#if MI_TRACK_VALGRIND
+// valgrind tool
+
+#define MI_TRACK_ENABLED      1
+#define MI_TRACK_HEAP_DESTROY 1           // track free of individual blocks on heap_destroy
+#define MI_TRACK_TOOL         "valgrind"
+
+#include <valgrind/valgrind.h>
+#include <valgrind/memcheck.h>
+
+#define mi_track_malloc_size(p,reqsize,size,zero) VALGRIND_MALLOCLIKE_BLOCK(p,size,MI_PADDING_SIZE /*red zone*/,zero)
+#define mi_track_free_size(p,_size)               VALGRIND_FREELIKE_BLOCK(p,MI_PADDING_SIZE /*red zone*/)
+#define mi_track_resize(p,oldsize,newsize)        VALGRIND_RESIZEINPLACE_BLOCK(p,oldsize,newsize,MI_PADDING_SIZE /*red zone*/)
+#define mi_track_mem_defined(p,size)              VALGRIND_MAKE_MEM_DEFINED(p,size)
+#define mi_track_mem_undefined(p,size)            VALGRIND_MAKE_MEM_UNDEFINED(p,size)
+#define mi_track_mem_noaccess(p,size)             VALGRIND_MAKE_MEM_NOACCESS(p,size)
+
+#elif MI_TRACK_ASAN
+// address sanitizer
+
+#define MI_TRACK_ENABLED      1
+#define MI_TRACK_HEAP_DESTROY 0
+#define MI_TRACK_TOOL         "asan"
+
+#include <sanitizer/asan_interface.h>
+
+#define mi_track_malloc_size(p,reqsize,size,zero) ASAN_UNPOISON_MEMORY_REGION(p,size)
+#define mi_track_free_size(p,size)                ASAN_POISON_MEMORY_REGION(p,size)
+#define mi_track_mem_defined(p,size)              ASAN_UNPOISON_MEMORY_REGION(p,size)
+#define mi_track_mem_undefined(p,size)            ASAN_UNPOISON_MEMORY_REGION(p,size)
+#define mi_track_mem_noaccess(p,size)             ASAN_POISON_MEMORY_REGION(p,size)
+
+#elif MI_TRACK_ETW
+// windows event tracing
+
+#define MI_TRACK_ENABLED      1
+#define MI_TRACK_HEAP_DESTROY 1
+#define MI_TRACK_TOOL         "ETW"
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#include "../src/prim/windows/etw.h"
+
+#define mi_track_init()                           EventRegistermicrosoft_windows_mimalloc();
+#define mi_track_malloc_size(p,reqsize,size,zero) EventWriteETW_MI_ALLOC((UINT64)(p), size)
+#define mi_track_free_size(p,size)                EventWriteETW_MI_FREE((UINT64)(p), size)
+
+#else
+// no tracking
+
+#define MI_TRACK_ENABLED      0
+#define MI_TRACK_HEAP_DESTROY 0
+#define MI_TRACK_TOOL         "none"
+
+#define mi_track_malloc_size(p,reqsize,size,zero)
+#define mi_track_free_size(p,_size)
+
+#endif
+
+// -------------------
+// Utility definitions
+
+#ifndef mi_track_resize
+#define mi_track_resize(p,oldsize,newsize)      mi_track_free_size(p,oldsize); mi_track_malloc(p,newsize,false)
+#endif
+
+#ifndef mi_track_align
+#define mi_track_align(p,alignedp,offset,size)  mi_track_mem_noaccess(p,offset)
+#endif
+
+#ifndef mi_track_init
+#define mi_track_init()
+#endif
+
+#ifndef mi_track_mem_defined
+#define mi_track_mem_defined(p,size)
+#endif
+
+#ifndef mi_track_mem_undefined
+#define mi_track_mem_undefined(p,size)
+#endif
+
+#ifndef mi_track_mem_noaccess
+#define mi_track_mem_noaccess(p,size)
+#endif
+
+
+#if MI_PADDING
+#define mi_track_malloc(p,reqsize,zero) \
+  if ((p)!=NULL) { \
+    mi_assert_internal(mi_usable_size(p)==(reqsize)); \
+    mi_track_malloc_size(p,reqsize,reqsize,zero); \
+  }
+#else
+#define mi_track_malloc(p,reqsize,zero) \
+  if ((p)!=NULL) { \
+    mi_assert_internal(mi_usable_size(p)>=(reqsize)); \
+    mi_track_malloc_size(p,reqsize,mi_usable_size(p),zero); \
+  }
+#endif
+
+#endif
diff --git a/Include/internal/mimalloc/mimalloc/types.h b/Include/internal/mimalloc/mimalloc/types.h
new file mode 100644
index 0000000..7616f37
--- /dev/null
+++ b/Include/internal/mimalloc/mimalloc/types.h
@@ -0,0 +1,670 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2023, 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_TYPES_H
+#define MIMALLOC_TYPES_H
+
+// --------------------------------------------------------------------------
+// This file contains the main type definitions for mimalloc:
+// 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
+//                  where objects are allocated.
+// --------------------------------------------------------------------------
+
+
+#include <stddef.h>   // ptrdiff_t
+#include <stdint.h>   // uintptr_t, uint16_t, etc
+#include "mimalloc/atomic.h"  // _Atomic
+
+#ifdef _MSC_VER
+#pragma warning(disable:4214) // bitfield is not int
+#endif
+
+// Minimal alignment necessary. On most platforms 16 bytes are needed
+// due to SSE registers for example. This must be at least `sizeof(void*)`
+#ifndef MI_MAX_ALIGN_SIZE
+#define MI_MAX_ALIGN_SIZE  16   // sizeof(max_align_t)
+#endif
+
+// ------------------------------------------------------
+// Variants
+// ------------------------------------------------------
+
+// Define NDEBUG in the release version to disable assertions.
+// #define NDEBUG
+
+// Define MI_TRACK_<tool> to enable tracking support
+// #define MI_TRACK_VALGRIND 1
+// #define MI_TRACK_ASAN     1
+// #define MI_TRACK_ETW      1
+
+// Define MI_STAT as 1 to maintain statistics; set it to 2 to have detailed statistics (but costs some performance).
+// #define MI_STAT 1
+
+// Define MI_SECURE to enable security mitigations
+// #define MI_SECURE 1  // guard page around metadata
+// #define MI_SECURE 2  // guard page around each mimalloc page
+// #define MI_SECURE 3  // encode free lists (detect corrupted free list (buffer overflow), and invalid pointer free)
+// #define MI_SECURE 4  // checks for double free. (may be more expensive)
+
+#if !defined(MI_SECURE)
+#define MI_SECURE 0
+#endif
+
+// Define MI_DEBUG for debug mode
+// #define MI_DEBUG 1  // basic assertion checks and statistics, check double free, corrupted free list, and invalid pointer free.
+// #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
+#define MI_DEBUG 0
+#endif
+#endif
+
+// Reserve extra padding at the end of each block to be more resilient against heap block overflows.
+// The padding can detect buffer overflow on free.
+#if !defined(MI_PADDING) && (MI_SECURE>=3 || MI_DEBUG>=1 || (MI_TRACK_VALGRIND || MI_TRACK_ASAN || MI_TRACK_ETW))
+#define MI_PADDING  1
+#endif
+
+// Check padding bytes; allows byte-precise buffer overflow detection
+#if !defined(MI_PADDING_CHECK) && MI_PADDING && (MI_SECURE>=3 || MI_DEBUG>=1)
+#define MI_PADDING_CHECK 1
+#endif
+
+
+// Encoded free lists allow detection of corrupted free lists
+// and can detect buffer overflows, modify after free, and double `free`s.
+#if (MI_SECURE>=3 || MI_DEBUG>=1)
+#define MI_ENCODE_FREELIST  1
+#endif
+
+
+// We used to abandon huge pages but to eagerly deallocate if freed from another thread,
+// but that makes it not possible to visit them during a heap walk or include them in a
+// `mi_heap_destroy`. We therefore instead reset/decommit the huge blocks if freed from
+// another thread so most memory is available until it gets properly freed by the owning thread.
+// #define MI_HUGE_PAGE_ABANDON 1
+
+
+// ------------------------------------------------------
+// Platform specific values
+// ------------------------------------------------------
+
+// ------------------------------------------------------
+// Size of a pointer.
+// We assume that `sizeof(void*)==sizeof(intptr_t)`
+// and it holds for all platforms we know of.
+//
+// However, the C standard only requires that:
+//  p == (void*)((intptr_t)p))
+// but we also need:
+//  i == (intptr_t)((void*)i)
+// or otherwise one might define an intptr_t type that is larger than a pointer...
+// ------------------------------------------------------
+
+#if INTPTR_MAX > INT64_MAX
+# define MI_INTPTR_SHIFT (4)  // assume 128-bit  (as on arm CHERI for example)
+#elif INTPTR_MAX == INT64_MAX
+# define MI_INTPTR_SHIFT (3)
+#elif INTPTR_MAX == INT32_MAX
+# define MI_INTPTR_SHIFT (2)
+#else
+#error platform pointers must be 32, 64, or 128 bits
+#endif
+
+#if SIZE_MAX == UINT64_MAX
+# define MI_SIZE_SHIFT (3)
+typedef int64_t  mi_ssize_t;
+#elif SIZE_MAX == UINT32_MAX
+# define MI_SIZE_SHIFT (2)
+typedef int32_t  mi_ssize_t;
+#else
+#error platform objects must be 32 or 64 bits
+#endif
+
+#if (SIZE_MAX/2) > LONG_MAX
+# define MI_ZU(x)  x##ULL
+# define MI_ZI(x)  x##LL
+#else
+# define MI_ZU(x)  x##UL
+# define MI_ZI(x)  x##L
+#endif
+
+#define MI_INTPTR_SIZE  (1<<MI_INTPTR_SHIFT)
+#define MI_INTPTR_BITS  (MI_INTPTR_SIZE*8)
+
+#define MI_SIZE_SIZE  (1<<MI_SIZE_SHIFT)
+#define MI_SIZE_BITS  (MI_SIZE_SIZE*8)
+
+#define MI_KiB     (MI_ZU(1024))
+#define MI_MiB     (MI_KiB*MI_KiB)
+#define MI_GiB     (MI_MiB*MI_KiB)
+
+
+// ------------------------------------------------------
+// Main internal data-structures
+// ------------------------------------------------------
+
+// Main tuning parameters for segment and page sizes
+// Sizes for 64-bit (usually divide by two for 32-bit)
+#define MI_SEGMENT_SLICE_SHIFT            (13 + MI_INTPTR_SHIFT)         // 64KiB  (32KiB on 32-bit)
+
+#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
+
+#define MI_SMALL_PAGE_SHIFT               (MI_SEGMENT_SLICE_SHIFT)       // 64KiB
+#define MI_MEDIUM_PAGE_SHIFT              ( 3 + MI_SMALL_PAGE_SHIFT)     // 512KiB
+
+
+// Derived constants
+#define MI_SEGMENT_SIZE                   (MI_ZU(1)<<MI_SEGMENT_SHIFT)
+#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_SMALL_OBJ_SIZE_MAX             (MI_SMALL_PAGE_SIZE/4)   // 8KiB on 64-bit
+#define MI_MEDIUM_OBJ_SIZE_MAX            (MI_MEDIUM_PAGE_SIZE/4)  // 128KiB 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)      // 32MiB 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_MEDIUM_OBJ_WSIZE_MAX >= 655360)
+#error "mimalloc internal: define more bins"
+#endif
+
+// Maximum slice offset (15)
+#define MI_MAX_SLICE_OFFSET               ((MI_ALIGNMENT_MAX / MI_SEGMENT_SLICE_SIZE) - 1)
+
+// Used as a special value to encode block sizes in 32 bits.
+#define MI_HUGE_BLOCK_SIZE                ((uint32_t)(2*MI_GiB))
+
+// blocks up to this size are always allocated aligned
+#define MI_MAX_ALIGN_GUARANTEE            (8*MI_MAX_ALIGN_SIZE)
+
+// Alignments over MI_ALIGNMENT_MAX are allocated in dedicated huge page segments
+#define MI_ALIGNMENT_MAX                  (MI_SEGMENT_SIZE >> 1)
+
+
+// ------------------------------------------------------
+// Mimalloc pages contain allocated blocks
+// ------------------------------------------------------
+
+// The free lists use encoded next fields
+// (Only actually encodes when MI_ENCODED_FREELIST is defined.)
+typedef uintptr_t  mi_encoded_t;
+
+// thread id's
+typedef size_t     mi_threadid_t;
+
+// free lists contain blocks
+typedef struct mi_block_s {
+  mi_encoded_t next;
+} mi_block_t;
+
+
+// The delayed flags are used for efficient multi-threaded free-ing
+typedef enum mi_delayed_e {
+  MI_USE_DELAYED_FREE   = 0, // push on the owning heap thread delayed list
+  MI_DELAYED_FREEING    = 1, // temporary: another thread is accessing the owning heap
+  MI_NO_DELAYED_FREE    = 2, // optimize: push on page local thread free queue if another block is already in the heap thread delayed free list
+  MI_NEVER_DELAYED_FREE = 3  // sticky, only resets on page reclaim
+} mi_delayed_t;
+
+
+// The `in_full` and `has_aligned` page flags are put in a union to efficiently
+// test if both are false (`full_aligned == 0`) in the `mi_free` routine.
+#if !MI_TSAN
+typedef union mi_page_flags_s {
+  uint8_t full_aligned;
+  struct {
+    uint8_t in_full : 1;
+    uint8_t has_aligned : 1;
+  } x;
+} mi_page_flags_t;
+#else
+// under thread sanitizer, use a byte for each flag to suppress warning, issue #130
+typedef union mi_page_flags_s {
+  uint16_t full_aligned;
+  struct {
+    uint8_t in_full;
+    uint8_t has_aligned;
+  } x;
+} mi_page_flags_t;
+#endif
+
+// Thread free list.
+// We use the bottom 2 bits of the pointer for mi_delayed_t flags
+typedef uintptr_t mi_thread_free_t;
+
+// A page contains blocks of one specific size (`block_size`).
+// Each page has three list of free blocks:
+// `free` for blocks that can be allocated,
+// `local_free` for freed blocks that are not yet available to `mi_malloc`
+// `thread_free` for freed blocks by other threads
+// The `local_free` and `thread_free` lists are migrated to the `free` list
+// when it is exhausted. The separate `local_free` list is necessary to
+// implement a monotonic heartbeat. The `thread_free` list is needed for
+// avoiding atomic operations in the common case.
+//
+//
+// `used - |thread_free|` == actual blocks that are in use (alive)
+// `used - |thread_free| + |free| + |local_free| == capacity`
+//
+// We don't count `freed` (as |free|) but use `used` to reduce
+// the number of memory accesses in the `mi_page_all_free` function(s).
+//
+// Notes:
+// - Access is optimized for `mi_free` and `mi_page_alloc` (in `alloc.c`)
+// - Using `uint16_t` does not seem to slow things down
+// - The size is 8 words on 64-bit which helps the page index calculations
+//   (and 10 words on 32-bit, and encoded free lists add 2 words. Sizes 10
+//    and 12 are still good for address calculation)
+// - To limit the structure size, the `xblock_size` is 32-bits only; for
+//   blocks > MI_HUGE_BLOCK_SIZE the size is determined from the segment page size
+// - `thread_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 `alloc.c:mi_free_block_mt`).
+//   The invariant is that no-delayed-free is only set if there is
+//   at least one block that will be added, or as already been added, to
+//   the owning heap `thread_delayed_free` list. This guarantees that pages
+//   will be freed correctly even if only other threads free blocks.
+typedef struct mi_page_s {
+  // "owned" by the segment
+  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
+
+  // 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
+  mi_page_flags_t       flags;             // `in_full` and `has_aligned` flags (8 bits)
+  uint8_t               free_is_zero : 1;  // `true` if the blocks in the free list are zero initialized
+  uint8_t               retire_expire : 7; // expiration count for retired blocks
+
+  mi_block_t*           free;              // list of available free blocks (`malloc` allocates from this list)
+  uint32_t              used;              // number of blocks in use (including blocks in `local_free` and `thread_free`)
+  uint32_t              xblock_size;       // size available in each block (always `>0`)
+  mi_block_t*           local_free;        // list of deferred free blocks by this thread (migrates to `free`)
+
+  #if (MI_ENCODE_FREELIST || MI_PADDING)
+  uintptr_t             keys[2];           // two random keys to encode the free lists (see `_mi_block_next`) or padding canary
+  #endif
+
+  _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 this thread with the same `block_size`
+  struct mi_page_s*     prev;              // previous page owned by this thread with the same `block_size`
+
+  // 64-bit 9 words, 32-bit 12 words, (+2 for secure)
+  #if MI_INTPTR_SIZE==8
+  uintptr_t padding[1];
+  #endif
+} mi_page_t;
+
+
+
+// ------------------------------------------------------
+// Mimalloc segments contain mimalloc pages
+// ------------------------------------------------------
+
+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 medium pages inside a segment
+  MI_PAGE_LARGE,    // larger blocks go into a page of just one block
+  MI_PAGE_HUGE,     // huge blocks (> 16 MiB) are put into a single page in a single segment.
+} mi_page_kind_t;
+
+typedef enum mi_segment_kind_e {
+  MI_SEGMENT_NORMAL, // MI_SEGMENT_SIZE size with pages inside.
+  MI_SEGMENT_HUGE,   // > MI_LARGE_SIZE_MAX 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;
+
+
+// Memory can reside in arena's, direct OS allocated, or statically allocated. The memid keeps track of this.
+typedef enum mi_memkind_e {
+  MI_MEM_NONE,      // not allocated
+  MI_MEM_EXTERNAL,  // not owned by mimalloc but provided externally (via `mi_manage_os_memory` for example)
+  MI_MEM_STATIC,    // allocated in a static area and should not be freed (for arena meta data for example)
+  MI_MEM_OS,        // allocated from the OS
+  MI_MEM_OS_HUGE,   // allocated as huge os pages
+  MI_MEM_OS_REMAP,  // allocated in a remapable area (i.e. using `mremap`)
+  MI_MEM_ARENA      // allocated from an arena (the usual case)
+} mi_memkind_t;
+
+static inline bool mi_memkind_is_os(mi_memkind_t memkind) {
+  return (memkind >= MI_MEM_OS && memkind <= MI_MEM_OS_REMAP);
+}
+
+typedef struct mi_memid_os_info {
+  void*         base;               // actual base address of the block (used for offset aligned allocations)
+  size_t        alignment;          // alignment at allocation
+} mi_memid_os_info_t;
+
+typedef struct mi_memid_arena_info {
+  size_t        block_index;        // index in the arena
+  mi_arena_id_t id;                 // arena id (>= 1)
+  bool          is_exclusive;       // the arena can only be used for specific arena allocations
+} mi_memid_arena_info_t;
+
+typedef struct mi_memid_s {
+  union {
+    mi_memid_os_info_t    os;       // only used for MI_MEM_OS
+    mi_memid_arena_info_t arena;    // only used for MI_MEM_ARENA
+  } mem;
+  bool          is_pinned;          // `true` if we cannot decommit/reset/protect in this memory (e.g. when allocated using large OS pages)
+  bool          initially_committed;// `true` if the memory was originally allocated as committed
+  bool          initially_zero;     // `true` if the memory was originally zero initialized
+  mi_memkind_t  memkind;
+} mi_memid_t;
+
+
+// Segments are large allocated memory blocks (8mb 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 for arena allocation
+  bool              allow_decommit;
+  bool              allow_purge;
+  size_t            segment_size;
+
+  // segment fields
+  mi_msecs_t        purge_expire;
+  mi_commit_mask_t  purge_mask;
+  mi_commit_mask_t  commit_mask;
+
+  _Atomic(struct mi_segment_s*) abandoned_next;
+
+  // 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`)
+
+  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 in the abandoned list (to force reclaim it it is too long)
+  size_t            used;               // count of pages in use
+  uintptr_t         cookie;             // verify addresses in debug mode: `mi_ptr_cookie(segment) == segment->cookie`
+
+  size_t            segment_slices;      // for huge segments this may be different from `MI_SLICES_PER_SEGMENT`
+  size_t            segment_info_slices; // initial slices we are using 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
+
+  mi_slice_t        slices[MI_SLICES_PER_SEGMENT+1];  // one more for huge blocks with large alignment
+} mi_segment_t;
+
+
+// ------------------------------------------------------
+// Heaps
+// Provide first-class heaps to allocate from.
+// A heap just owns a set of pages for allocation and
+// can only be allocate/reallocate from the thread that created it.
+// Freeing blocks can be done from any thread though.
+// Per thread, the segments are shared among its heaps.
+// Per thread, there is always a default heap that is
+// used for allocation; it is initialized to statically
+// point to an empty heap to avoid initialization checks
+// in the fast path.
+// ------------------------------------------------------
+
+// Thread local data
+typedef struct mi_tld_s mi_tld_t;
+
+// Pages of a certain block size are held in a queue.
+typedef struct mi_page_queue_s {
+  mi_page_t* first;
+  mi_page_t* last;
+  size_t     block_size;
+} mi_page_queue_t;
+
+#define MI_BIN_FULL  (MI_BIN_HUGE+1)
+
+// Random context
+typedef struct mi_random_cxt_s {
+  uint32_t input[16];
+  uint32_t output[16];
+  int      output_available;
+  bool     weak;
+} mi_random_ctx_t;
+
+
+// In debug mode there is a padding structure at the end of the blocks to check for buffer overflows
+#if (MI_PADDING)
+typedef struct mi_padding_s {
+  uint32_t canary; // encoded block value to check validity of the padding (in case of overflow)
+  uint32_t delta;  // padding bytes before the block. (mi_usable_size(p) - delta == exact allocated bytes)
+} mi_padding_t;
+#define MI_PADDING_SIZE   (sizeof(mi_padding_t))
+#define MI_PADDING_WSIZE  ((MI_PADDING_SIZE + MI_INTPTR_SIZE - 1) / MI_INTPTR_SIZE)
+#else
+#define MI_PADDING_SIZE   0
+#define MI_PADDING_WSIZE  0
+#endif
+
+#define MI_PAGES_DIRECT   (MI_SMALL_WSIZE_MAX + MI_PADDING_WSIZE + 1)
+
+
+// A heap owns a set of pages.
+struct mi_heap_s {
+  mi_tld_t*             tld;
+  mi_page_t*            pages_free_direct[MI_PAGES_DIRECT];  // optimize: array where every entry points a page with possibly free blocks in the corresponding queue for that size.
+  mi_page_queue_t       pages[MI_BIN_FULL + 1];              // queue of pages for each size class (or "bin")
+  _Atomic(mi_block_t*)  thread_delayed_free;
+  mi_threadid_t         thread_id;                           // thread this heap belongs too
+  mi_arena_id_t         arena_id;                            // arena id if the heap belongs to a specific arena (or 0)
+  uintptr_t             cookie;                              // random cookie to verify pointers (see `_mi_ptr_cookie`)
+  uintptr_t             keys[2];                             // two random keys used to encode the `thread_delayed_free` list
+  mi_random_ctx_t       random;                              // random number context used for secure allocation
+  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.
+  mi_heap_t*            next;                                // list of heaps per thread
+  bool                  no_reclaim;                          // `true` if this heap should not reclaim abandoned pages
+};
+
+
+
+// ------------------------------------------------------
+// Debug
+// ------------------------------------------------------
+
+#if !defined(MI_DEBUG_UNINIT)
+#define MI_DEBUG_UNINIT     (0xD0)
+#endif
+#if !defined(MI_DEBUG_FREED)
+#define MI_DEBUG_FREED      (0xDF)
+#endif
+#if !defined(MI_DEBUG_PADDING)
+#define MI_DEBUG_PADDING    (0xDE)
+#endif
+
+#if (MI_DEBUG)
+// use our own assertion to print without memory allocation
+void _mi_assert_fail(const char* assertion, const char* fname, unsigned int line, const char* func );
+#define mi_assert(expr)     ((expr) ? (void)0 : _mi_assert_fail(#expr,__FILE__,__LINE__,__func__))
+#else
+#define mi_assert(x)
+#endif
+
+#if (MI_DEBUG>1)
+#define mi_assert_internal    mi_assert
+#else
+#define mi_assert_internal(x)
+#endif
+
+#if (MI_DEBUG>2)
+#define mi_assert_expensive   mi_assert
+#else
+#define mi_assert_expensive(x)
+#endif
+
+// ------------------------------------------------------
+// Statistics
+// ------------------------------------------------------
+
+#ifndef MI_STAT
+#if (MI_DEBUG>0)
+#define MI_STAT 2
+#else
+#define MI_STAT 0
+#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 large;
+  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 large_count;
+#if MI_STAT>1
+  mi_stat_count_t normal_bins[MI_BIN_HUGE+1];
+#endif
+} mi_stats_t;
+
+
+void _mi_stat_increase(mi_stat_count_t* stat, size_t amount);
+void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount);
+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)
+#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
+#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)
+
+// ------------------------------------------------------
+// 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)
+
+// OS thread local data
+typedef struct mi_os_tld_s {
+  size_t                region_idx;   // start point for next allocation
+  mi_stats_t*           stats;        // points to tld stats
+} mi_os_tld_t;
+
+
+// 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
+  mi_stats_t*         stats;        // points to tld stats
+  mi_os_tld_t*        os;           // points to os 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_os_tld_t         os;            // os tld
+  mi_stats_t          stats;         // statistics
+};
+
+#endif
diff --git a/Include/internal/pycore_mimalloc.h b/Include/internal/pycore_mimalloc.h
new file mode 100644
index 0000000..c29dc82
--- /dev/null
+++ b/Include/internal/pycore_mimalloc.h
@@ -0,0 +1,19 @@
+#ifndef Py_INTERNAL_MIMALLOC_H
+#define Py_INTERNAL_MIMALLOC_H
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+#if defined(MIMALLOC_H) || defined(MIMALLOC_TYPES_H)
+#  error "pycore_mimalloc.h must be included before mimalloc.h"
+#endif
+
+#include "pycore_pymem.h"
+#define MI_DEBUG_UNINIT     PYMEM_CLEANBYTE
+#define MI_DEBUG_FREED      PYMEM_DEADBYTE
+#define MI_DEBUG_PADDING    PYMEM_FORBIDDENBYTE
+
+#include "mimalloc.h"
+
+#endif // Py_INTERNAL_MIMALLOC_H
diff --git a/Include/internal/pycore_pymem_init.h b/Include/internal/pycore_pymem_init.h
index 119fa16..11fbe16 100644
--- a/Include/internal/pycore_pymem_init.h
+++ b/Include/internal/pycore_pymem_init.h
@@ -18,17 +18,18 @@ extern void * _PyMem_RawRealloc(void *, void *, size_t);
 extern void _PyMem_RawFree(void *, void *);
 #define PYRAW_ALLOC {NULL, _PyMem_RawMalloc, _PyMem_RawCalloc, _PyMem_RawRealloc, _PyMem_RawFree}
 
-#ifdef WITH_PYMALLOC
+#if defined(WITH_PYMALLOC)
 extern void* _PyObject_Malloc(void *, size_t);
 extern void* _PyObject_Calloc(void *, size_t, size_t);
 extern void _PyObject_Free(void *, void *);
 extern void* _PyObject_Realloc(void *, void *, size_t);
 #  define PYOBJ_ALLOC {NULL, _PyObject_Malloc, _PyObject_Calloc, _PyObject_Realloc, _PyObject_Free}
+#  define PYMEM_ALLOC PYOBJ_ALLOC
 #else
 # define PYOBJ_ALLOC PYRAW_ALLOC
+# define PYMEM_ALLOC PYOBJ_ALLOC
 #endif  // WITH_PYMALLOC
 
-#define PYMEM_ALLOC PYOBJ_ALLOC
 
 extern void* _PyMem_DebugRawMalloc(void *, size_t);
 extern void* _PyMem_DebugRawCalloc(void *, size_t, size_t);