4 files changed, 2193 insertions, 0 deletions

diff --git a/contrib/libs/mimalloc/include/mimalloc-atomic.h b/contrib/libs/mimalloc/include/mimalloc-atomic.h
new file mode 100644
index 0000000000..dc48f0a2f0
--- /dev/null
+++ b/contrib/libs/mimalloc/include/mimalloc-atomic.h
@@ -0,0 +1,332 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2021 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
+#elif defined(_MSC_VER)
+// Use MSVC C wrapper for C11 atomics
+#define  _Atomic(tp)            tp
+#define  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
+#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))
+
+
+
+#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);
+}
+
+// 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);
+}
+
+// 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__))
+#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__)
+static inline void mi_atomic_yield(void) {
+  __asm__ __volatile__ ("or 27,27,27" ::: "memory");
+}
+#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/contrib/libs/mimalloc/include/mimalloc-internal.h b/contrib/libs/mimalloc/include/mimalloc-internal.h
new file mode 100644
index 0000000000..1e1a79665c
--- /dev/null
+++ b/contrib/libs/mimalloc/include/mimalloc-internal.h
@@ -0,0 +1,924 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2021, 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
+
+#include "mimalloc-types.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)
+#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))  // 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
+
+// "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_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  _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);
+bool       _mi_preloading();  // true while the C runtime is not ready
+
+// os.c
+size_t     _mi_os_page_size(void);
+void       _mi_os_init(void);                                      // called from process init
+void*      _mi_os_alloc(size_t size, mi_stats_t* stats);           // to allocate thread local data
+void       _mi_os_free(void* p, size_t size, mi_stats_t* stats);   // to free thread local data
+size_t     _mi_os_good_alloc_size(size_t size);
+
+// memory.c
+void*      _mi_mem_alloc_aligned(size_t size, size_t alignment, bool* commit, bool* large, bool* is_pinned, bool* is_zero, size_t* id, mi_os_tld_t* tld);
+void       _mi_mem_free(void* p, size_t size, size_t id, bool fully_committed, bool any_reset, mi_os_tld_t* tld);
+
+bool       _mi_mem_reset(void* p, size_t size, mi_os_tld_t* tld);
+bool       _mi_mem_unreset(void* p, size_t size, bool* is_zero, mi_os_tld_t* tld);
+bool       _mi_mem_commit(void* p, size_t size, bool* is_zero, mi_os_tld_t* tld);
+bool       _mi_mem_protect(void* addr, size_t size);
+bool       _mi_mem_unprotect(void* addr, size_t size);
+
+void        _mi_mem_collect(mi_os_tld_t* tld);
+
+// "segment.c"
+mi_page_t* _mi_segment_page_alloc(mi_heap_t* heap, size_t block_wsize, 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);
+uint8_t*   _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t block_size, size_t* page_size, size_t* pre_size); // page start for any page
+void       _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block_t* block);
+
+void       _mi_segment_thread_collect(mi_segments_tld_t* tld);
+void       _mi_abandoned_reclaim_all(mi_heap_t* heap, mi_segments_tld_t* tld);
+void       _mi_abandoned_await_readers(void);
+
+
+
+// "page.c"
+void*      _mi_malloc_generic(mi_heap_t* heap, size_t size)  mi_attr_noexcept mi_attr_malloc;
+
+void       _mi_page_retire(mi_page_t* page);                                  // 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(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);
+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);
+
+// "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) mi_attr_noexcept;  // called from `_mi_malloc_generic`
+void*       _mi_heap_malloc_zero(mi_heap_t* heap, size_t size, bool zero);
+void*       _mi_heap_realloc_zero(mi_heap_t* heap, void* p, size_t newsize, bool zero);
+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_block_zero_init(const mi_page_t* page, void* p, size_t size);
+
+#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),0)
+#define mi_likely(x)       __builtin_expect((x),1)
+#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 UNUSED(x)     (void)(x)
+#if (MI_DEBUG>0)
+#define UNUSED_RELEASE(x)
+#else
+#define UNUSED_RELEASE(x)  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)
+
+
+// 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);
+}
+
+// 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);
+  }
+}
+
+// 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(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);
+}
+
+// Does malloc satisfy the alignment constraints already?
+static inline bool mi_malloc_satisfies_alignment(size_t alignment, size_t size) {
+  return (alignment == sizeof(void*) || (alignment == MI_MAX_ALIGN_SIZE && size > (MI_MAX_ALIGN_SIZE/2)));
+}
+
+// Overflow detecting multiply
+#if __has_builtin(__builtin_umul_overflow) || __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 == UINT_MAX)
+    return __builtin_umul_overflow(count, size, total);
+  #elif (SIZE_MAX == ULONG_MAX)
+    return __builtin_umull_overflow(count, size, total);
+  #else
+    return __builtin_umulll_overflow(count, size, 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;
+  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))) {
+    _mi_error_message(EOVERFLOW, "allocation request is too large (%zu * %zu bytes)\n", count, size);
+    *total = SIZE_MAX;
+    return true;
+  }
+  else return false;
+}
+
+
+/* ----------------------------------------------------------------------------------------
+The thread local default heap: `_mi_get_default_heap` returns the thread local heap.
+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: the uniqueid use is buggy but kept for reference.
+------------------------------------------------------------------------------------------- */
+
+extern const mi_heap_t _mi_heap_empty;  // read-only empty heap, initial value of the thread local default heap
+extern bool _mi_process_is_initialized;
+mi_heap_t*  _mi_heap_main_get(void);    // statically allocated main backing heap
+
+#if defined(MI_MALLOC_OVERRIDE)
+#if defined(__APPLE__) // macOS
+#define MI_TLS_SLOT               89  // seems unused? 
+// 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>
+#endif
+#endif
+
+#if defined(MI_TLS_SLOT)
+static inline void* mi_tls_slot(size_t slot) mi_attr_noexcept;   // forward declaration
+#elif defined(MI_TLS_PTHREAD_SLOT_OFS)
+#include <pthread.h>
+static inline mi_heap_t** mi_tls_pthread_heap_slot(void) {
+  pthread_t self = pthread_self();
+  #if defined(__DragonFly__)
+  if (self==NULL) {
+    mi_heap_t* pheap_main = _mi_heap_main_get();
+    return &pheap_main;
+  }
+  #endif
+  return (mi_heap_t**)((uint8_t*)self + MI_TLS_PTHREAD_SLOT_OFS);
+}
+#elif defined(MI_TLS_PTHREAD)
+#include <pthread.h>
+extern pthread_key_t _mi_heap_default_key;
+#endif
+
+// Default heap to allocate from (if not using TLS- or pthread slots).
+// Do not use this directly but use through `mi_heap_get_default()` (or the unchecked `mi_get_default_heap`).
+// This thread local variable is only used when neither MI_TLS_SLOT, MI_TLS_PTHREAD, or MI_TLS_PTHREAD_SLOT_OFS are defined.
+// However, on the Apple M1 we do use the address of this variable as the unique thread-id (issue #356).
+extern mi_decl_thread mi_heap_t* _mi_heap_default;  // default heap to allocate from
+
+static inline mi_heap_t* mi_get_default_heap(void) {
+#if defined(MI_TLS_SLOT)
+  mi_heap_t* heap = (mi_heap_t*)mi_tls_slot(MI_TLS_SLOT);
+  return (mi_unlikely(heap == NULL) ? (mi_heap_t*)&_mi_heap_empty : heap);
+#elif defined(MI_TLS_PTHREAD_SLOT_OFS)
+  mi_heap_t* heap = *mi_tls_pthread_heap_slot();
+  return (mi_unlikely(heap == NULL) ? (mi_heap_t*)&_mi_heap_empty : heap);
+#elif defined(MI_TLS_PTHREAD)
+  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
+  #if defined(MI_TLS_RECURSE_GUARD)
+  if (mi_unlikely(!_mi_process_is_initialized)) return _mi_heap_main_get();
+  #endif
+  return _mi_heap_default;
+#endif
+}
+
+static inline bool mi_heap_is_default(const mi_heap_t* heap) {
+  return (heap == mi_get_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];
+}
+
+// Get the page belonging to a certain size class
+static inline mi_page_t* _mi_get_free_small_page(size_t size) {
+  return _mi_heap_get_free_small_page(mi_get_default_heap(), size);
+}
+
+// Segment that contains the pointer
+static inline mi_segment_t* _mi_ptr_segment(const void* p) {
+  // mi_assert_internal(p != NULL);
+  return (mi_segment_t*)((uintptr_t)p & ~MI_SEGMENT_MASK);
+}
+
+// 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 || page == &segment->pages[page->segment_idx]);
+  return segment;
+}
+
+// used internally
+static inline uintptr_t _mi_segment_page_idx_of(const mi_segment_t* segment, const void* p) {
+  // if (segment->page_size > MI_SEGMENT_SIZE) return &segment->pages[0];  // huge pages
+  ptrdiff_t diff = (uint8_t*)p - (uint8_t*)segment;
+  mi_assert_internal(diff >= 0 && (size_t)diff < MI_SEGMENT_SIZE);
+  uintptr_t idx = (uintptr_t)diff >> segment->page_shift;
+  mi_assert_internal(idx < segment->capacity);
+  mi_assert_internal(segment->page_kind <= MI_PAGE_MEDIUM || idx == 0);
+  return idx;
+}
+
+// Get the page containing the pointer
+static inline mi_page_t* _mi_segment_page_of(const mi_segment_t* segment, const void* p) {
+  uintptr_t idx = _mi_segment_page_idx_of(segment, p);
+  return &((mi_segment_t*)segment)->pages[idx];
+}
+
+// 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) {
+  const size_t bsize = page->xblock_size;
+  mi_assert_internal(bsize > 0 && (bsize%sizeof(void*)) == 0);
+  return _mi_segment_page_start(segment, page, bsize, page_size, NULL);
+}
+
+// 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 cased 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, bsize, &psize, NULL);
+    return psize;
+  }
+}
+
+// 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;
+}
+
+
+// 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* segmentp = _mi_ptr_segment(p);
+  mi_segment_t* segmentq = _mi_ptr_segment(q);
+  if (segmentp != segmentq) return false;
+  uintptr_t idxp = _mi_segment_page_idx_of(segmentp, p);
+  uintptr_t idxq = _mi_segment_page_idx_of(segmentq, q);
+  return (idxp == idxq);
+}
+
+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 (mi_unlikely(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)(mi_unlikely(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 ) {
+  #ifdef MI_ENCODE_FREELIST
+  return (mi_block_t*)mi_ptr_decode(null, block->next, keys);
+  #else
+  UNUSED(keys); UNUSED(null);
+  return (mi_block_t*)block->next;
+  #endif
+}
+
+static inline void mi_block_set_nextx(const void* null, mi_block_t* block, const mi_block_t* next, const uintptr_t* keys) {
+  #ifdef MI_ENCODE_FREELIST
+  block->next = mi_ptr_encode(null, next, keys);
+  #else
+  UNUSED(keys); UNUSED(null);
+  block->next = (mi_encoded_t)next;
+  #endif
+}
+
+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
+  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
+  UNUSED(page);
+  mi_block_set_nextx(page,block,next,NULL);
+  #endif
+}
+
+// -------------------------------------------------------------------
+// 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();
+}
+
+
+// -------------------------------------------------------------------
+// Getting the thread id should be performant as it is called in the
+// fast path of `_mi_free` and we specialize for various platforms.
+// -------------------------------------------------------------------
+#if defined(_WIN32)
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+static inline uintptr_t _mi_thread_id(void) mi_attr_noexcept {
+  // Windows: works on Intel and ARM in both 32- and 64-bit
+  return (uintptr_t)NtCurrentTeb();
+}
+
+#elif defined(__GNUC__) && \
+      (defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__aarch64__))
+
+// TLS register on x86 is in the FS or GS register, see: https://akkadia.org/drepper/tls.pdf
+static inline void* mi_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)) : );  // 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; 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; UNUSED(ofs);
+#if defined(__APPLE__) // M1, issue #343
+  __asm__ volatile ("mrs %0, tpidrro_el0" : "=r" (tcb));
+  tcb = (void**)((uintptr_t)tcb & ~0x07UL);  // clear lower 3 bits
+#else
+  __asm__ volatile ("mrs %0, tpidr_el0" : "=r" (tcb));
+#endif
+  res = tcb[slot];
+#endif
+  return res;
+}
+
+// setting is only used on macOSX for now
+static inline void mi_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 macOSX uses GS
+#elif defined(__x86_64__) && (MI_INTPTR_SIZE==4)
+  __asm__("movl %1,%%fs:%1" : "=m" (*((void**)ofs)) : "rn" (value) : );  // x32 ABI
+#elif defined(__x86_64__)
+  __asm__("movq %1,%%fs:%1" : "=m" (*((void**)ofs)) : "rn" (value) : );  // x86_64 Linux, BSD uses FS
+#elif defined(__arm__)
+  void** tcb; 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; UNUSED(ofs);
+#if defined(__APPLE__) // M1, issue #343
+  __asm__ volatile ("mrs %0, tpidrro_el0" : "=r" (tcb));
+  tcb = (void**)((uintptr_t)tcb & ~0x07UL);  // clear lower 3 bits
+#else
+  __asm__ volatile ("mrs %0, tpidr_el0" : "=r" (tcb));
+#endif
+  tcb[slot] = value;
+#endif
+}
+
+static inline uintptr_t _mi_thread_id(void) mi_attr_noexcept {
+#if defined(__BIONIC__) && (defined(__arm__) || defined(__aarch64__))
+  // on Android, slot 1 is the thread ID (pointer to pthread internal struct)
+  return (uintptr_t)mi_tls_slot(1);
+#else
+  // in all our other targets, slot 0 is the pointer to the thread control block
+  return (uintptr_t)mi_tls_slot(0);
+#endif
+}
+#else
+// otherwise use standard C
+static inline uintptr_t _mi_thread_id(void) mi_attr_noexcept {
+  return (uintptr_t)&_mi_heap_default;
+}
+#endif
+
+// -----------------------------------------------------------------------
+// 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
+#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 defined(_WIN32) && (defined(_M_IX86) || defined(_M_X64))
+#include <intrin.h>
+#include <string.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); // todo: use noinline?
+  }
+}
+#else
+#include <string.h>
+static inline void _mi_memcpy(void* dst, const void* src, size_t n) {
+  memcpy(dst, src, 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 (__GNUC__ >= 4) || defined(__clang__)
+// On GCC/CLang we provide a hint that the pointers are word aligned.
+#include <string.h>
+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);
+  memcpy(adst, asrc, 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);
+}
+#endif
+
+
+#endif
diff --git a/contrib/libs/mimalloc/include/mimalloc-types.h b/contrib/libs/mimalloc/include/mimalloc-types.h
new file mode 100644
index 0000000000..caf161d63f
--- /dev/null
+++ b/contrib/libs/mimalloc/include/mimalloc-types.h
@@ -0,0 +1,501 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2021, 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
+
+#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 `MI_INTPTR_SIZE`
+#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_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 byte-precise buffer overflow on free.
+#if !defined(MI_PADDING) && (MI_DEBUG>=1)
+#define MI_PADDING  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 || MI_PADDING > 0)
+#define MI_ENCODE_FREELIST  1
+#endif
+
+// ------------------------------------------------------
+// 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 == 9223372036854775807LL
+# define MI_INTPTR_SHIFT (3)
+#elif INTPTR_MAX == 2147483647LL
+# define MI_INTPTR_SHIFT (2)
+#else
+#error platform must be 32 or 64 bits
+#endif
+
+#define MI_INTPTR_SIZE  (1<<MI_INTPTR_SHIFT)
+#define MI_INTPTR_BITS  (MI_INTPTR_SIZE*8)
+
+#define KiB     ((size_t)1024)
+#define MiB     (KiB*KiB)
+#define GiB     (MiB*KiB)
+
+
+// ------------------------------------------------------
+// Main internal data-structures
+// ------------------------------------------------------
+
+// Main tuning parameters for segment and page sizes
+// Sizes for 64-bit, divide by two for 32-bit
+#define MI_SMALL_PAGE_SHIFT               (13 + MI_INTPTR_SHIFT)      // 64kb
+#define MI_MEDIUM_PAGE_SHIFT              ( 3 + MI_SMALL_PAGE_SHIFT)  // 512kb
+#define MI_LARGE_PAGE_SHIFT               ( 3 + MI_MEDIUM_PAGE_SHIFT) // 4mb
+#define MI_SEGMENT_SHIFT                  ( MI_LARGE_PAGE_SHIFT)      // 4mb
+
+// Derived constants
+#define MI_SEGMENT_SIZE                   (1UL<<MI_SEGMENT_SHIFT)
+#define MI_SEGMENT_MASK                   ((uintptr_t)MI_SEGMENT_SIZE - 1)
+
+#define MI_SMALL_PAGE_SIZE                (1UL<<MI_SMALL_PAGE_SHIFT)
+#define MI_MEDIUM_PAGE_SIZE               (1UL<<MI_MEDIUM_PAGE_SHIFT)
+#define MI_LARGE_PAGE_SIZE                (1UL<<MI_LARGE_PAGE_SHIFT)
+
+#define MI_SMALL_PAGES_PER_SEGMENT        (MI_SEGMENT_SIZE/MI_SMALL_PAGE_SIZE)
+#define MI_MEDIUM_PAGES_PER_SEGMENT       (MI_SEGMENT_SIZE/MI_MEDIUM_PAGE_SIZE)
+#define MI_LARGE_PAGES_PER_SEGMENT        (MI_SEGMENT_SIZE/MI_LARGE_PAGE_SIZE)
+
+// The max object size are checked to not waste more than 12.5% internally over the page sizes.
+// (Except for large pages since huge objects are allocated in 4MiB chunks)
+#define MI_SMALL_OBJ_SIZE_MAX             (MI_SMALL_PAGE_SIZE/4)   // 16kb
+#define MI_MEDIUM_OBJ_SIZE_MAX            (MI_MEDIUM_PAGE_SIZE/4)  // 128kb
+#define MI_LARGE_OBJ_SIZE_MAX             (MI_LARGE_PAGE_SIZE/2)   // 2mb
+#define MI_LARGE_OBJ_WSIZE_MAX            (MI_LARGE_OBJ_SIZE_MAX/MI_INTPTR_SIZE)
+#define MI_HUGE_OBJ_SIZE_MAX              (2*MI_INTPTR_SIZE*MI_SEGMENT_SIZE)        // (must match MI_REGION_MAX_ALLOC_SIZE in memory.c)
+
+// Maximum number of size classes. (spaced exponentially in 12.5% increments)
+#define MI_BIN_HUGE  (73U)
+
+#if (MI_LARGE_OBJ_WSIZE_MAX >= 655360)
+#error "define more bins"
+#endif
+
+// Used as a special value to encode block sizes in 32 bits.
+#define MI_HUGE_BLOCK_SIZE   ((uint32_t)MI_HUGE_OBJ_SIZE_MAX)
+
+// The free lists use encoded next fields
+// (Only actually encodes when MI_ENCODED_FREELIST is defined.)
+typedef uintptr_t mi_encoded_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
+  uint8_t               segment_idx;       // index in the segment `pages` array, `page == &segment->pages[page->segment_idx]`
+  uint8_t               segment_in_use:1;  // `true` if the segment allocated this page
+  uint8_t               is_reset:1;        // `true` if the page memory was reset
+  uint8_t               is_committed:1;    // `true` if the page virtual memory is committed
+  uint8_t               is_zero_init:1;    // `true` if the page was 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               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)
+  #ifdef MI_ENCODE_FREELIST
+  uintptr_t             keys[2];           // two random keys to encode the free lists (see `_mi_block_next`)
+  #endif
+  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`)
+  _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`
+} mi_page_t;
+
+
+
+typedef enum mi_page_kind_e {
+  MI_PAGE_SMALL,    // small blocks go into 64kb pages inside a segment
+  MI_PAGE_MEDIUM,   // medium blocks go into 512kb pages inside a segment
+  MI_PAGE_LARGE,    // larger blocks go into a single page spanning a whole segment
+  MI_PAGE_HUGE      // huge blocks (>512kb) are put into a single page in a segment of the exact size (but still 2mb aligned)
+} mi_page_kind_t;
+
+// Segments are large allocated memory blocks (2mb on 64 bit) from
+// the OS. Inside segments we allocated fixed size _pages_ that
+// contain blocks.
+typedef struct mi_segment_s {
+  // memory fields
+  size_t               memid;            // id for the os-level memory manager
+  bool                 mem_is_pinned;    // `true` if we cannot decommit/reset/protect in this memory (i.e. when allocated using large OS pages)
+  bool                 mem_is_committed; // `true` if the whole segment is eagerly committed  
+
+  // segment fields
+  _Atomic(struct mi_segment_s*) abandoned_next;
+  struct mi_segment_s* next;             // must be the first segment field after abandoned_next -- see `segment.c:segment_init`
+  struct mi_segment_s* prev;
+
+  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 if it is too long)
+
+  size_t               used;             // count of pages in use (`used <= capacity`)
+  size_t               capacity;         // count of available pages (`#free + used`)
+  size_t               segment_size;     // for huge pages this may be different from `MI_SEGMENT_SIZE`
+  size_t               segment_info_size;// space we are using from the first page for segment meta-data and possible guard pages.
+  uintptr_t            cookie;           // verify addresses in secure mode: `_mi_ptr_cookie(segment) == segment->cookie`
+
+  // layout like this to optimize access in `mi_free`
+  size_t               page_shift;       // `1 << page_shift` == the page sizes == `page->block_size * page->reserved` (unless the first page, then `-segment_info_size`).
+  _Atomic(uintptr_t)   thread_id;        // unique id of the thread owning this segment
+  mi_page_kind_t       page_kind;        // kind of pages: small, large, or huge
+  mi_page_t            pages[1];         // up to `MI_SMALL_PAGES_PER_SEGMENT` pages
+} 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;
+} mi_random_ctx_t;
+
+
+// In debug mode there is a padding stucture 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;
+  uintptr_t             thread_id;                           // thread this heap belongs too
+  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
+// ------------------------------------------------------
+
+#define MI_DEBUG_UNINIT     (0xD0)
+#define MI_DEBUG_FREED      (0xDF)
+#define MI_DEBUG_PADDING    (0xDE)
+
+#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 page_committed;
+  mi_stat_count_t segments_abandoned;
+  mi_stat_count_t pages_abandoned;
+  mi_stat_count_t threads;
+  mi_stat_count_t normal;
+  mi_stat_count_t huge;
+  mi_stat_count_t giant;
+  mi_stat_count_t malloc;
+  mi_stat_count_t segments_cache;
+  mi_stat_counter_t pages_extended;
+  mi_stat_counter_t mmap_calls;
+  mi_stat_counter_t commit_calls;
+  mi_stat_counter_t page_no_retire;
+  mi_stat_counter_t searches;
+  mi_stat_counter_t normal_count;
+  mi_stat_counter_t huge_count;
+  mi_stat_counter_t giant_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
+// ------------------------------------------------------
+
+typedef int64_t  mi_msecs_t;
+
+// Queue of segments
+typedef struct mi_segment_queue_s {
+  mi_segment_t* first;
+  mi_segment_t* last;
+} mi_segment_queue_t;
+
+// 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_segment_queue_t  small_free;   // queue of segments with free small pages
+  mi_segment_queue_t  medium_free;  // queue of segments with free medium pages
+  mi_page_queue_t     pages_reset;  // queue of freed pages that can be reset
+  size_t              count;        // current number of segments;
+  size_t              peak_count;   // peak number of segments
+  size_t              current_size; // current size of all segments
+  size_t              peak_size;    // peak size of all segments
+  size_t              cache_count;  // number of segments in the cache
+  size_t              cache_size;   // total size of all segments in the cache
+  mi_segment_t*       cache;        // (small) cache of 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/contrib/libs/mimalloc/include/mimalloc.h b/contrib/libs/mimalloc/include/mimalloc.h
new file mode 100644
index 0000000000..fe5aa8f343
--- /dev/null
+++ b/contrib/libs/mimalloc/include/mimalloc.h
@@ -0,0 +1,436 @@
+/* ----------------------------------------------------------------------------
+Copyright (c) 2018-2021, 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 171   // 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 (__GNUC__ >= 4) || defined(__clang__)  // includes clang, icc, and clang-cl
+  #define mi_decl_nodiscard    __attribute__((warn_unused_result))
+#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
+  #define mi_cdecl                      // leads to warnings... __attribute__((cdecl))
+  #define mi_decl_export                __attribute__((visibility("default")))
+  #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
+
+#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;        // bytes in use by allocated blocks
+  size_t block_size;  // size in bytes of each block
+} 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;
+
+
+// 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, all `false` by default
+// ------------------------------------------------------
+
+typedef enum mi_option_e {
+  // stable options
+  mi_option_show_errors,
+  mi_option_show_stats,
+  mi_option_verbose,
+  // the following options are experimental
+  mi_option_eager_commit,
+  mi_option_eager_region_commit,
+  mi_option_reset_decommits,
+  mi_option_large_os_pages,         // implies eager commit
+  mi_option_reserve_huge_os_pages,
+  mi_option_reserve_os_memory,
+  mi_option_segment_cache,
+  mi_option_page_reset,
+  mi_option_abandoned_page_reset,
+  mi_option_segment_reset,
+  mi_option_eager_commit_delay,
+  mi_option_reset_delay,
+  mi_option_use_numa_nodes,
+  mi_option_limit_os_alloc,
+  mi_option_os_tag,
+  mi_option_max_errors,
+  mi_option_max_warnings,
+  _mi_option_last
+} 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_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_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 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);
+
+#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 <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 {
+  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;
+  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 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;
+  using is_always_equal                        = 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 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; }
+#endif // __cplusplus
+
+#endif