diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0d780fa1..5cc7ec5c 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -434,7 +434,7 @@ endif()
 
 if(CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang|GNU|Intel" AND NOT CMAKE_SYSTEM_NAME MATCHES "Haiku")
   if(MI_OPT_ARCH)
-    if(APPLE AND CMAKE_C_COMPILER_ID STREQUAL "AppleClang" AND CMAKE_OSX_ARCHITECTURES)   # to support multi-arch binaries (#999)
+    if(APPLE AND CMAKE_C_COMPILER_ID MATCHES "AppleClang|Clang" AND CMAKE_OSX_ARCHITECTURES)   # to support multi-arch binaries (#999)
       if("arm64" IN_LIST CMAKE_OSX_ARCHITECTURES)
         list(APPEND MI_OPT_ARCH_FLAGS "-Xarch_arm64;-march=armv8.1-a")
       endif()
@@ -532,7 +532,9 @@ if(MI_TRACK_ASAN)
 endif()
 string(TOLOWER "${CMAKE_BUILD_TYPE}" CMAKE_BUILD_TYPE_LC)
 list(APPEND mi_defines "MI_CMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE_LC}")  #todo: multi-config project needs $<CONFIG> ?
-if(NOT(CMAKE_BUILD_TYPE_LC MATCHES "^(release|relwithdebinfo|minsizerel|none)$"))
+if(CMAKE_BUILD_TYPE_LC MATCHES "^(release|relwithdebinfo|minsizerel|none)$")
+  list(APPEND mi_defines MI_BUILD_RELEASE)
+else()
   set(mi_libname "${mi_libname}-${CMAKE_BUILD_TYPE_LC}") #append build type (e.g. -debug) if not a release version
 endif()
 
@@ -582,7 +584,7 @@ if(MI_BUILD_SHARED)
   install(TARGETS mimalloc EXPORT mimalloc ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR} RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR} LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR})
   install(EXPORT mimalloc DESTINATION ${mi_install_cmakedir})
 
-  if(WIN32)
+  if(WIN32 AND NOT MINGW)
     # On windows, the import library name for the dll would clash with the static mimalloc.lib library
     # so we postfix the dll import library with `.dll.lib` (and also the .pdb debug file)
     set_property(TARGET mimalloc PROPERTY ARCHIVE_OUTPUT_NAME "${mi_libname}.dll" )
@@ -592,6 +594,9 @@ if(MI_BUILD_SHARED)
     # install(FILES "$<TARGET_FILE_DIR:mimalloc>/${mi_libname}.dll.pdb" DESTINATION ${CMAKE_INSTALL_LIBDIR})
   endif()
   if(WIN32 AND MI_WIN_REDIRECT)
+    if(MINGW)
+      set_property(TARGET mimalloc PROPERTY PREFIX "")
+    endif()
     # On windows, link and copy the mimalloc redirection dll too.
     if(CMAKE_GENERATOR_PLATFORM STREQUAL "arm64ec")
       set(MIMALLOC_REDIRECT_SUFFIX "-arm64ec")
diff --git a/azure-pipelines.yml b/azure-pipelines.yml
index f4a77baa..df608b17 100644
--- a/azure-pipelines.yml
+++ b/azure-pipelines.yml
@@ -7,9 +7,9 @@ trigger:
   branches:
     include:
     - master
-    - dev
-    - dev2
     - dev3
+    - dev2
+    - dev
   tags:
     include:
     - v*
@@ -34,6 +34,14 @@ jobs:
         BuildType: secure
         cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Release -DMI_SECURE=ON
         MSBuildConfiguration: Release
+      Debug x86:
+        BuildType: debug
+        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Debug -DMI_DEBUG_FULL=ON -A Win32
+        MSBuildConfiguration: Debug
+      Release x86:
+        BuildType: release
+        cmakeExtraArgs: -DCMAKE_BUILD_TYPE=Release -A Win32
+        MSBuildConfiguration: Release
   steps:
   - task: CMake@1
     inputs:
diff --git a/bin/mimalloc-redirect-arm64.dll b/bin/mimalloc-redirect-arm64.dll
old mode 100644
new mode 100755
index e6360285..27172d2c
Binary files a/bin/mimalloc-redirect-arm64.dll and b/bin/mimalloc-redirect-arm64.dll differ
diff --git a/bin/mimalloc-redirect-arm64.lib b/bin/mimalloc-redirect-arm64.lib
old mode 100644
new mode 100755
index 11d71ef9..dca80b9b
Binary files a/bin/mimalloc-redirect-arm64.lib and b/bin/mimalloc-redirect-arm64.lib differ
diff --git a/bin/mimalloc-redirect-arm64ec.dll b/bin/mimalloc-redirect-arm64ec.dll
old mode 100644
new mode 100755
index f5ee4e47..a228af39
Binary files a/bin/mimalloc-redirect-arm64ec.dll and b/bin/mimalloc-redirect-arm64ec.dll differ
diff --git a/bin/mimalloc-redirect-arm64ec.lib b/bin/mimalloc-redirect-arm64ec.lib
old mode 100644
new mode 100755
index b88e8fc1..0ce77436
Binary files a/bin/mimalloc-redirect-arm64ec.lib and b/bin/mimalloc-redirect-arm64ec.lib differ
diff --git a/bin/mimalloc-redirect.dll b/bin/mimalloc-redirect.dll
old mode 100644
new mode 100755
index 9e015cc6..ec035f1b
Binary files a/bin/mimalloc-redirect.dll and b/bin/mimalloc-redirect.dll differ
diff --git a/bin/mimalloc-redirect.lib b/bin/mimalloc-redirect.lib
old mode 100644
new mode 100755
index 1d710c01..785fa475
Binary files a/bin/mimalloc-redirect.lib and b/bin/mimalloc-redirect.lib differ
diff --git a/bin/mimalloc-redirect32.dll b/bin/mimalloc-redirect32.dll
old mode 100644
new mode 100755
index 32799ffe..92578f24
Binary files a/bin/mimalloc-redirect32.dll and b/bin/mimalloc-redirect32.dll differ
diff --git a/bin/mimalloc-redirect32.lib b/bin/mimalloc-redirect32.lib
old mode 100644
new mode 100755
index e2927250..bf649787
Binary files a/bin/mimalloc-redirect32.lib and b/bin/mimalloc-redirect32.lib differ
diff --git a/cmake/mimalloc-config-version.cmake b/cmake/mimalloc-config-version.cmake
index 286ec0ba..daac7a5d 100644
--- a/cmake/mimalloc-config-version.cmake
+++ b/cmake/mimalloc-config-version.cmake
@@ -1,6 +1,6 @@
 set(mi_version_major 2)
 set(mi_version_minor 2)
-set(mi_version_patch 2)
+set(mi_version_patch 3)
 set(mi_version ${mi_version_major}.${mi_version_minor})
 
 set(PACKAGE_VERSION ${mi_version})
diff --git a/include/mimalloc.h b/include/mimalloc.h
index ae6ae262..ff6f0568 100644
--- a/include/mimalloc.h
+++ b/include/mimalloc.h
@@ -8,7 +8,7 @@ terms of the MIT license. A copy of the license can be found in the file
 #ifndef MIMALLOC_H
 #define MIMALLOC_H
 
-#define MI_MALLOC_VERSION 222   // major + 2 digits minor
+#define MI_MALLOC_VERSION 223   // major + 2 digits minor
 
 // ------------------------------------------------------
 // Compiler specific attributes
diff --git a/include/mimalloc/atomic.h b/include/mimalloc/atomic.h
index 6eaa6f99..39ff5c90 100644
--- a/include/mimalloc/atomic.h
+++ b/include/mimalloc/atomic.h
@@ -266,6 +266,13 @@ static inline int64_t mi_atomic_addi64_relaxed(volatile _Atomic(int64_t)*p, int6
   return current;
 #endif
 }
+static inline void mi_atomic_void_addi64_relaxed(volatile int64_t* p, const volatile int64_t* padd) {
+  const int64_t add = *padd;
+  if (add != 0) {
+    mi_atomic_addi64_relaxed((volatile _Atomic(int64_t)*)p, add);
+  }
+}
+
 static inline void mi_atomic_maxi64_relaxed(volatile _Atomic(int64_t)*p, int64_t x) {
   int64_t current;
   do {
@@ -363,8 +370,9 @@ static inline void mi_atomic_yield(void) {
   _mm_pause();
 }
 #elif (defined(__GNUC__) || defined(__clang__)) && \
-      (defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__armel__) || defined(__ARMEL__) || \
-       defined(__aarch64__) || defined(__powerpc__) || defined(__ppc__) || defined(__PPC__)) || defined(__POWERPC__)
+      (defined(__x86_64__) || defined(__i386__) || \
+       defined(__aarch64__) || defined(__arm__) || \
+       defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__))
 #if defined(__x86_64__) || defined(__i386__)
 static inline void mi_atomic_yield(void) {
   __asm__ volatile ("pause" ::: "memory");
@@ -373,10 +381,16 @@ static inline void mi_atomic_yield(void) {
 static inline void mi_atomic_yield(void) {
   __asm__ volatile("wfe");
 }
-#elif (defined(__arm__) && __ARM_ARCH__ >= 7)
+#elif defined(__arm__)
+#if __ARM_ARCH >= 7
 static inline void mi_atomic_yield(void) {
   __asm__ volatile("yield" ::: "memory");
 }
+#else
+static inline void mi_atomic_yield(void) {
+  __asm__ volatile ("nop" ::: "memory");
+}
+#endif
 #elif defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__)
 #ifdef __APPLE__
 static inline void mi_atomic_yield(void) {
@@ -387,10 +401,6 @@ static inline void mi_atomic_yield(void) {
   __asm__ __volatile__ ("or 27,27,27" ::: "memory");
 }
 #endif
-#elif defined(__armel__) || defined(__ARMEL__)
-static inline void mi_atomic_yield(void) {
-  __asm__ volatile ("nop" ::: "memory");
-}
 #endif
 #elif defined(__sun)
 // Fallback for other archs
diff --git a/include/mimalloc/internal.h b/include/mimalloc/internal.h
index c53c841a..eae85ab6 100644
--- a/include/mimalloc/internal.h
+++ b/include/mimalloc/internal.h
@@ -127,6 +127,7 @@ bool        _mi_os_has_virtual_reserve(void);
 
 bool        _mi_os_reset(void* addr, size_t size);
 bool        _mi_os_commit(void* p, size_t size, bool* is_zero);
+bool        _mi_os_commit_ex(void* addr, size_t size, bool* is_zero, size_t stat_size);
 bool        _mi_os_decommit(void* addr, size_t size);
 bool        _mi_os_protect(void* addr, size_t size);
 bool        _mi_os_unprotect(void* addr, size_t size);
@@ -1030,6 +1031,21 @@ static inline size_t mi_bsr(size_t x) {
   return (x==0 ? MI_SIZE_BITS : MI_SIZE_BITS - 1 - mi_clz(x));
 }
 
+size_t _mi_popcount_generic(size_t x);
+
+static inline size_t mi_popcount(size_t x) {
+  if (x<=1) return x;
+  if (x==SIZE_MAX) return MI_SIZE_BITS;
+  #if defined(__GNUC__)
+    #if (SIZE_MAX == ULONG_MAX)
+      return __builtin_popcountl(x);
+    #else
+      return __builtin_popcountll(x);
+    #endif
+  #else
+    return _mi_popcount_generic(x);
+  #endif
+}
 
 // ---------------------------------------------------------------------------------
 // Provide our own `_mi_memcpy` for potential performance optimizations.
diff --git a/include/mimalloc/types.h b/include/mimalloc/types.h
index 1f4adc07..5a3f5fe2 100644
--- a/include/mimalloc/types.h
+++ b/include/mimalloc/types.h
@@ -67,10 +67,10 @@ terms of the MIT license. A copy of the license can be found in the file
 // #define MI_DEBUG 2  // + internal assertion checks
 // #define MI_DEBUG 3  // + extensive internal invariant checking (cmake -DMI_DEBUG_FULL=ON)
 #if !defined(MI_DEBUG)
-#if !defined(NDEBUG) || defined(_DEBUG)
-#define MI_DEBUG 2
-#else
+#if defined(MI_BUILD_RELEASE) || defined(NDEBUG)
 #define MI_DEBUG 0
+#else
+#define MI_DEBUG 2
 #endif
 #endif
 
@@ -683,22 +683,25 @@ void _mi_assert_fail(const char* assertion, const char* fname, unsigned int line
 // add to stat keeping track of the peak
 void _mi_stat_increase(mi_stat_count_t* stat, size_t amount);
 void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount);
+void _mi_stat_adjust_decrease(mi_stat_count_t* stat, size_t amount);
 // counters can just be increased
 void _mi_stat_counter_increase(mi_stat_counter_t* stat, size_t amount);
 
 #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_adjust_decrease(stat,amount)  _mi_stat_adjust_decrease( &(stat), amount)
 #define mi_stat_counter_increase(stat,amount) _mi_stat_counter_increase( &(stat), amount)
 #else
 #define mi_stat_increase(stat,amount)         ((void)0)
 #define mi_stat_decrease(stat,amount)         ((void)0)
+#define mi_stat_adjust_decrease(stat,amount)  ((void)0)
 #define mi_stat_counter_increase(stat,amount) ((void)0)
 #endif
 
 #define mi_heap_stat_counter_increase(heap,stat,amount)  mi_stat_counter_increase( (heap)->tld->stats.stat, amount)
 #define mi_heap_stat_increase(heap,stat,amount)  mi_stat_increase( (heap)->tld->stats.stat, amount)
 #define mi_heap_stat_decrease(heap,stat,amount)  mi_stat_decrease( (heap)->tld->stats.stat, amount)
-
+#define mi_heap_stat_adjust_decrease(heap,stat,amount)  mi_stat_adjust_decrease( (heap)->tld->stats.stat, amount)
 
 #endif
diff --git a/readme.md b/readme.md
index 81f2057e..87cee98d 100644
--- a/readme.md
+++ b/readme.md
@@ -12,9 +12,9 @@ is a general purpose allocator with excellent [performance](#performance) charac
 Initially developed by Daan Leijen for the runtime systems of the
 [Koka](https://koka-lang.github.io) and [Lean](https://github.com/leanprover/lean) languages.
 
-Latest release   : `v3.0.2` (beta) (2025-03-06).  
-Latest v2 release: `v2.2.2` (2025-03-06).  
-Latest v1 release: `v1.9.2` (2024-03-06).
+Latest release   : `v3.0.3` (beta) (2025-03-28).  
+Latest v2 release: `v2.2.3` (2025-03-28).  
+Latest v1 release: `v1.9.3` (2024-03-28).
 
 mimalloc is a drop-in replacement for `malloc` and can be used in other programs
 without code changes, for example, on dynamically linked ELF-based systems (Linux, BSD, etc.) you can use it as:
@@ -84,6 +84,9 @@ Enjoy!
 
 ### Releases
 
+* 2025-03-28, `v1.9.3`, `v2.2.3`, `v3.0.3-beta`: Various small bug and build fixes, including:
+  fix arm32 pre v7 builds, fix mingw build, get runtime statistics, improve statistic commit counts, 
+  fix execution on non BMI1 x64 systems. 
 * 2025-03-06, `v1.9.2`, `v2.2.2`, `v3.0.2-beta`: Various small bug and build fixes. 
   Add `mi_options_print`, `mi_arenas_print`, and the experimental `mi_stat_get` and `mi_stat_get_json`. 
   Add `mi_thread_set_in_threadpool` and `mi_heap_set_numa_affinity` (v3 only). Add vcpkg portfile. 
diff --git a/src/alloc-aligned.c b/src/alloc-aligned.c
index bb9e9ecb..3d3202eb 100644
--- a/src/alloc-aligned.c
+++ b/src/alloc-aligned.c
@@ -191,9 +191,6 @@ static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t
       const bool is_aligned = (((uintptr_t)page->free + offset) & align_mask)==0;
       if mi_likely(is_aligned)
       {
-        #if MI_STAT>1
-        mi_heap_stat_increase(heap, malloc_requested, size);
-        #endif
         void* p = (zero ? _mi_page_malloc_zeroed(heap,page,padsize) : _mi_page_malloc(heap,page,padsize)); // call specific page malloc for better codegen
         mi_assert_internal(p != NULL);
         mi_assert_internal(((uintptr_t)p + offset) % alignment == 0);
diff --git a/src/alloc.c b/src/alloc.c
index 2350dc9c..0fed5e75 100644
--- a/src/alloc.c
+++ b/src/alloc.c
@@ -30,6 +30,7 @@ terms of the MIT license. A copy of the license can be found in the file
 // Note: in release mode the (inlined) routine is about 7 instructions with a single test.
 extern inline void* _mi_page_malloc_zero(mi_heap_t* heap, mi_page_t* page, size_t size, bool zero) mi_attr_noexcept
 {
+  mi_assert_internal(size >= MI_PADDING_SIZE);
   mi_assert_internal(page->block_size == 0 /* empty heap */ || mi_page_block_size(page) >= size);
 
   // check the free list
@@ -88,6 +89,7 @@ extern inline void* _mi_page_malloc_zero(mi_heap_t* heap, mi_page_t* page, size_
     #if (MI_STAT>1)
     const size_t bin = _mi_bin(bsize);
     mi_heap_stat_increase(heap, malloc_bins[bin], 1);
+    mi_heap_stat_increase(heap, malloc_requested, size - MI_PADDING_SIZE);
     #endif
   }
   #endif
@@ -146,12 +148,6 @@ static inline mi_decl_restrict void* mi_heap_malloc_small_zero(mi_heap_t* heap,
   void* const p = _mi_page_malloc_zero(heap, page, size + MI_PADDING_SIZE, zero);
   mi_track_malloc(p,size,zero);
 
-  #if MI_STAT>1
-  if (p != NULL) {
-    if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
-    mi_heap_stat_increase(heap, malloc_requested, mi_usable_size(p));
-  }
-  #endif
   #if MI_DEBUG>3
   if (p != NULL && zero) {
     mi_assert_expensive(mi_mem_is_zero(p, size));
@@ -188,12 +184,6 @@ extern inline void* _mi_heap_malloc_zero_ex(mi_heap_t* heap, size_t size, bool z
     void* const p = _mi_malloc_generic(heap, size + MI_PADDING_SIZE, zero, huge_alignment);  // note: size can overflow but it is detected in malloc_generic
     mi_track_malloc(p,size,zero);
 
-    #if MI_STAT>1
-    if (p != NULL) {
-      if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
-      mi_heap_stat_increase(heap, malloc_requested, mi_usable_size(p));
-    }
-    #endif
     #if MI_DEBUG>3
     if (p != NULL && zero) {
       mi_assert_expensive(mi_mem_is_zero(p, size));
@@ -666,7 +656,8 @@ mi_decl_restrict void* _mi_heap_malloc_guarded(mi_heap_t* heap, size_t size, boo
   if (p != NULL) {
     if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); }
     #if MI_STAT>1
-    mi_heap_stat_increase(heap, malloc_requested, mi_usable_size(p));
+    mi_heap_stat_adjust_decrease(heap, malloc_requested, req_size);
+    mi_heap_stat_increase(heap, malloc_requested, size);
     #endif
     _mi_stat_counter_increase(&heap->tld->stats.malloc_guarded_count, 1);
   }
diff --git a/src/arena.c b/src/arena.c
index 51382e22..01102c27 100644
--- a/src/arena.c
+++ b/src/arena.c
@@ -259,7 +259,7 @@ static mi_decl_noinline void* mi_arena_try_alloc_at(mi_arena_t* arena, size_t ar
 
   // set the dirty bits (todo: no need for an atomic op here?)
   if (arena->memid.initially_zero && arena->blocks_dirty != NULL) {
-    memid->initially_zero = _mi_bitmap_claim_across(arena->blocks_dirty, arena->field_count, needed_bcount, bitmap_index, NULL);
+    memid->initially_zero = _mi_bitmap_claim_across(arena->blocks_dirty, arena->field_count, needed_bcount, bitmap_index, NULL, NULL);
   }
 
   // set commit state
@@ -271,10 +271,14 @@ static mi_decl_noinline void* mi_arena_try_alloc_at(mi_arena_t* arena, size_t ar
     // commit requested, but the range may not be committed as a whole: ensure it is committed now
     memid->initially_committed = true;
     bool any_uncommitted;
-    _mi_bitmap_claim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &any_uncommitted);
+    size_t already_committed = 0;
+    _mi_bitmap_claim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &any_uncommitted, &already_committed);
     if (any_uncommitted) {
+      mi_assert_internal(already_committed < needed_bcount);
+      const size_t commit_size = mi_arena_block_size(needed_bcount);
+      const size_t stat_commit_size = commit_size - mi_arena_block_size(already_committed);
       bool commit_zero = false;
-      if (!_mi_os_commit(p, mi_arena_block_size(needed_bcount), &commit_zero)) {
+      if (!_mi_os_commit_ex(p, commit_size, &commit_zero, stat_commit_size)) {
         memid->initially_committed = false;
       }
       else {
@@ -284,7 +288,14 @@ static mi_decl_noinline void* mi_arena_try_alloc_at(mi_arena_t* arena, size_t ar
   }
   else {
     // no need to commit, but check if already fully committed
-    memid->initially_committed = _mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index);
+    size_t already_committed = 0;
+    memid->initially_committed = _mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &already_committed);
+    if (!memid->initially_committed && already_committed > 0) {
+      // partially committed: as it will be committed at some time, adjust the stats and pretend the range is fully uncommitted.
+      mi_assert_internal(already_committed < needed_bcount);
+      _mi_stat_decrease(&_mi_stats_main.committed, mi_arena_block_size(already_committed));
+      _mi_bitmap_unclaim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index);
+    }
   }
 
   return p;
@@ -468,17 +479,19 @@ static void mi_arena_purge(mi_arena_t* arena, size_t bitmap_idx, size_t blocks)
   const size_t size = mi_arena_block_size(blocks);
   void* const p = mi_arena_block_start(arena, bitmap_idx);
   bool needs_recommit;
-  if (_mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx)) {
+  size_t already_committed = 0;
+  if (_mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx, &already_committed)) {
     // all blocks are committed, we can purge freely
+    mi_assert_internal(already_committed == blocks);
     needs_recommit = _mi_os_purge(p, size);
   }
   else {
     // some blocks are not committed -- this can happen when a partially committed block is freed
     // in `_mi_arena_free` and it is conservatively marked as uncommitted but still scheduled for a purge
-    // we need to ensure we do not try to reset (as that may be invalid for uncommitted memory),
-    // and also undo the decommit stats (as it was already adjusted)
+    // we need to ensure we do not try to reset (as that may be invalid for uncommitted memory).
+    mi_assert_internal(already_committed < blocks);
     mi_assert_internal(mi_option_is_enabled(mi_option_purge_decommits));
-    needs_recommit = _mi_os_purge_ex(p, size, false /* allow reset? */, 0);    
+    needs_recommit = _mi_os_purge_ex(p, size, false /* allow reset? */, mi_arena_block_size(already_committed));    
   }
 
   // clear the purged blocks
@@ -512,7 +525,7 @@ static void mi_arena_schedule_purge(mi_arena_t* arena, size_t bitmap_idx, size_t
     else {
       // already an expiration was set
     }
-    _mi_bitmap_claim_across(arena->blocks_purge, arena->field_count, blocks, bitmap_idx, NULL);
+    _mi_bitmap_claim_across(arena->blocks_purge, arena->field_count, blocks, bitmap_idx, NULL, NULL);
   }
 }
 
@@ -652,15 +665,16 @@ void _mi_arena_free(void* p, size_t size, size_t committed_size, mi_memid_t memi
   if (p==NULL) return;
   if (size==0) return;
   const bool all_committed = (committed_size == size);
+  const size_t decommitted_size = (committed_size <= size ? size - committed_size : 0);
 
   // need to set all memory to undefined as some parts may still be marked as no_access (like padding etc.)
   mi_track_mem_undefined(p,size);
 
   if (mi_memkind_is_os(memid.memkind)) {
     // was a direct OS allocation, pass through
-    if (!all_committed && committed_size > 0) {
-      // if partially committed, adjust the committed stats (as `_mi_os_free` will increase decommit by the full size)
-      _mi_stat_decrease(&_mi_stats_main.committed, committed_size);
+    if (!all_committed && decommitted_size > 0) {
+      // if partially committed, adjust the committed stats (as `_mi_os_free` will decrease commit by the full size)
+      _mi_stat_increase(&_mi_stats_main.committed, decommitted_size);
     }
     _mi_os_free(p, size, memid);
   }
@@ -694,14 +708,14 @@ void _mi_arena_free(void* p, size_t size, size_t committed_size, mi_memid_t memi
       mi_assert_internal(arena->blocks_purge != NULL);
 
       if (!all_committed) {
-        // mark the entire range as no longer committed (so we recommit the full range when re-using)
+        // mark the entire range as no longer committed (so we will recommit the full range when re-using)
         _mi_bitmap_unclaim_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx);
         mi_track_mem_noaccess(p,size);
-        if (committed_size > 0) {
+        //if (committed_size > 0) {
           // if partially committed, adjust the committed stats (is it will be recommitted when re-using)
-          // in the delayed purge, we now need to not count a decommit if the range is not marked as committed.
+          // in the delayed purge, we do no longer decrease the commit if the range is not marked entirely as committed.
           _mi_stat_decrease(&_mi_stats_main.committed, committed_size);
-        }
+        //}
         // note: if not all committed, it may be that the purge will reset/decommit the entire range
         // that contains already decommitted parts. Since purge consistently uses reset or decommit that
         // works (as we should never reset decommitted parts).
diff --git a/src/bitmap.c b/src/bitmap.c
index 0f67b0ca..32d1e954 100644
--- a/src/bitmap.c
+++ b/src/bitmap.c
@@ -369,7 +369,7 @@ bool _mi_bitmap_unclaim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t
 
 // Set `count` bits at `bitmap_idx` to 1 atomically
 // Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit.
-bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero) {
+bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero, size_t* already_set) {
   size_t idx = mi_bitmap_index_field(bitmap_idx);
   size_t pre_mask;
   size_t mid_mask;
@@ -377,28 +377,31 @@ bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t co
   size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask);
   bool all_zero = true;
   bool any_zero = false;
+  size_t one_count = 0;
   _Atomic(size_t)*field = &bitmap[idx];
   size_t prev = mi_atomic_or_acq_rel(field++, pre_mask);
-  if ((prev & pre_mask) != 0) all_zero = false;
+  if ((prev & pre_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & pre_mask); }
   if ((prev & pre_mask) != pre_mask) any_zero = true;
   while (mid_count-- > 0) {
     prev = mi_atomic_or_acq_rel(field++, mid_mask);
-    if ((prev & mid_mask) != 0) all_zero = false;
+    if ((prev & mid_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & mid_mask); }
     if ((prev & mid_mask) != mid_mask) any_zero = true;
   }
   if (post_mask!=0) {
     prev = mi_atomic_or_acq_rel(field, post_mask);
-    if ((prev & post_mask) != 0) all_zero = false;
+    if ((prev & post_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & post_mask); }
     if ((prev & post_mask) != post_mask) any_zero = true;
   }
   if (pany_zero != NULL) { *pany_zero = any_zero; }
+  if (already_set != NULL) { *already_set = one_count; };
+  mi_assert_internal(all_zero ? one_count == 0 : one_count <= count);
   return all_zero;
 }
 
 
 // Returns `true` if all `count` bits were 1.
 // `any_ones` is `true` if there was at least one bit set to one.
-static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_ones) {
+static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_ones, size_t* already_set) {
   size_t idx = mi_bitmap_index_field(bitmap_idx);
   size_t pre_mask;
   size_t mid_mask;
@@ -406,30 +409,33 @@ static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_field
   size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask);
   bool all_ones = true;
   bool any_ones = false;
+  size_t one_count = 0;
   mi_bitmap_field_t* field = &bitmap[idx];
   size_t prev = mi_atomic_load_relaxed(field++);
   if ((prev & pre_mask) != pre_mask) all_ones = false;
-  if ((prev & pre_mask) != 0) any_ones = true;
+  if ((prev & pre_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & pre_mask); }
   while (mid_count-- > 0) {
     prev = mi_atomic_load_relaxed(field++);
     if ((prev & mid_mask) != mid_mask) all_ones = false;
-    if ((prev & mid_mask) != 0) any_ones = true;
+    if ((prev & mid_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & mid_mask); }
   }
   if (post_mask!=0) {
     prev = mi_atomic_load_relaxed(field);
     if ((prev & post_mask) != post_mask) all_ones = false;
-    if ((prev & post_mask) != 0) any_ones = true;
+    if ((prev & post_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & post_mask); }
   }
   if (pany_ones != NULL) { *pany_ones = any_ones; }
+  if (already_set != NULL) { *already_set = one_count; }
+  mi_assert_internal(all_ones ? one_count == count : one_count < count);
   return all_ones;
 }
 
-bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) {
-  return mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, NULL);
+bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, size_t* already_set) {
+  return mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, NULL, already_set);
 }
 
 bool _mi_bitmap_is_any_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) {
   bool any_ones;
-  mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, &any_ones);
+  mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, &any_ones, NULL);
   return any_ones;
 }
diff --git a/src/bitmap.h b/src/bitmap.h
index f098dd8f..0f4744f4 100644
--- a/src/bitmap.h
+++ b/src/bitmap.h
@@ -111,9 +111,9 @@ bool _mi_bitmap_unclaim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t
 
 // Set `count` bits at `bitmap_idx` to 1 atomically
 // Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit.
-bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero);
+bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero, size_t* already_set);
 
-bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx);
+bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, size_t* already_set);
 bool _mi_bitmap_is_any_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx);
 
 #endif
diff --git a/src/free.c b/src/free.c
index 3aa6e369..3b906738 100644
--- a/src/free.c
+++ b/src/free.c
@@ -522,15 +522,13 @@ static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) {
 // only maintain stats for smaller objects if requested
 #if (MI_STAT>0)
 static void mi_stat_free(const mi_page_t* page, const mi_block_t* block) {
-  #if (MI_STAT < 2)
   MI_UNUSED(block);
-  #endif
   mi_heap_t* const heap = mi_heap_get_default();
   const size_t bsize = mi_page_usable_block_size(page);
-  #if (MI_STAT>1)
-  const size_t usize = mi_page_usable_size_of(page, block);
-  mi_heap_stat_decrease(heap, malloc_requested, usize);
-  #endif
+  // #if (MI_STAT>1)
+  // const size_t usize = mi_page_usable_size_of(page, block);
+  // mi_heap_stat_decrease(heap, malloc_requested, usize);
+  // #endif
   if (bsize <= MI_MEDIUM_OBJ_SIZE_MAX) {
     mi_heap_stat_decrease(heap, malloc_normal, bsize);
     #if (MI_STAT > 1)
diff --git a/src/heap.c b/src/heap.c
index bd3a1e2a..cbfee560 100644
--- a/src/heap.c
+++ b/src/heap.c
@@ -344,17 +344,17 @@ static bool _mi_heap_page_destroy(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_
       mi_heap_stat_decrease(heap, malloc_huge, bsize);
     }
   }
-#if (MI_STAT)
+  #if (MI_STAT>0)
   _mi_page_free_collect(page, false);  // update used count
   const size_t inuse = page->used;
   if (bsize <= MI_LARGE_OBJ_SIZE_MAX) {
     mi_heap_stat_decrease(heap, malloc_normal, bsize * inuse);
-#if (MI_STAT>1)
+    #if (MI_STAT>1)
     mi_heap_stat_decrease(heap, malloc_bins[_mi_bin(bsize)], inuse);
-#endif
+    #endif
   }
-  mi_heap_stat_decrease(heap, malloc_requested, bsize * inuse);  // todo: off for aligned blocks...
-#endif
+  // mi_heap_stat_decrease(heap, malloc_requested, bsize * inuse);  // todo: off for aligned blocks...
+  #endif
 
   /// pretend it is all free now
   mi_assert_internal(mi_page_thread_free(page) == NULL);
diff --git a/src/init.c b/src/init.c
index b0fb2e66..ddded152 100644
--- a/src/init.c
+++ b/src/init.c
@@ -108,7 +108,7 @@ const mi_page_t _mi_page_empty = {
 // may lead to allocation itself on some platforms)
 // --------------------------------------------------------
 
-mi_decl_hidden mi_decl_cache_align const mi_heap_t _mi_heap_empty = {
+mi_decl_cache_align const mi_heap_t _mi_heap_empty = {
   NULL,
   MI_ATOMIC_VAR_INIT(NULL),
   0,                // tid
diff --git a/src/libc.c b/src/libc.c
index 1bd97aa3..52d095eb 100644
--- a/src/libc.c
+++ b/src/libc.c
@@ -275,3 +275,60 @@ int _mi_snprintf(char* buf, size_t buflen, const char* fmt, ...) {
   va_end(args);
   return written;
 }
+
+
+#if MI_SIZE_SIZE == 4
+#define mi_mask_even_bits32      (0x55555555)
+#define mi_mask_even_pairs32     (0x33333333)
+#define mi_mask_even_nibbles32   (0x0F0F0F0F)
+
+// sum of all the bytes in `x` if it is guaranteed that the sum < 256!
+static size_t mi_byte_sum32(uint32_t x) {
+  // perform `x * 0x01010101`: the highest byte contains the sum of all bytes.
+  x += (x << 8);
+  x += (x << 16);
+  return (size_t)(x >> 24);
+}
+
+static size_t mi_popcount_generic32(uint32_t x) {
+  // first count each 2-bit group `a`, where: a==0b00 -> 00, a==0b01 -> 01, a==0b10 -> 01, a==0b11 -> 10
+  // in other words, `a - (a>>1)`; to do this in parallel, we need to mask to prevent spilling a bit pair
+  // into the lower bit-pair:
+  x = x - ((x >> 1) & mi_mask_even_bits32);
+  // add the 2-bit pair results
+  x = (x & mi_mask_even_pairs32) + ((x >> 2) & mi_mask_even_pairs32);
+  // add the 4-bit nibble results
+  x = (x + (x >> 4)) & mi_mask_even_nibbles32;
+  // each byte now has a count of its bits, we can sum them now:
+  return mi_byte_sum32(x);
+}
+
+mi_decl_noinline size_t _mi_popcount_generic(size_t x) {
+  return mi_popcount_generic32(x);
+}
+
+#else
+#define mi_mask_even_bits64      (0x5555555555555555)
+#define mi_mask_even_pairs64     (0x3333333333333333)
+#define mi_mask_even_nibbles64   (0x0F0F0F0F0F0F0F0F)
+
+// sum of all the bytes in `x` if it is guaranteed that the sum < 256!
+static size_t mi_byte_sum64(uint64_t x) {
+  x += (x << 8);
+  x += (x << 16);
+  x += (x << 32);
+  return (size_t)(x >> 56);
+}
+
+static size_t mi_popcount_generic64(uint64_t x) {
+  x = x - ((x >> 1) & mi_mask_even_bits64);
+  x = (x & mi_mask_even_pairs64) + ((x >> 2) & mi_mask_even_pairs64);
+  x = (x + (x >> 4)) & mi_mask_even_nibbles64;
+  return mi_byte_sum64(x);
+}
+
+mi_decl_noinline size_t _mi_popcount_generic(size_t x) {
+  return mi_popcount_generic64(x);
+}
+#endif
+
diff --git a/src/prim/unix/prim.c b/src/prim/unix/prim.c
index c95500e7..ad6ca2a9 100644
--- a/src/prim/unix/prim.c
+++ b/src/prim/unix/prim.c
@@ -31,11 +31,12 @@ terms of the MIT license. A copy of the license can be found in the file
 
 #if defined(__linux__)
   #include <features.h>
+  #include <linux/prctl.h>  // PR_SET_VMA
   //#if defined(MI_NO_THP)
-  #include <sys/prctl.h>  // THP disable
+  #include <sys/prctl.h>    // THP disable
   //#endif
   #if defined(__GLIBC__)
-  #include <linux/mman.h> // linux mmap flags
+  #include <linux/mman.h>   // linux mmap flags
   #else
   #include <sys/mman.h>
   #endif
@@ -205,14 +206,24 @@ static int unix_madvise(void* addr, size_t size, int advice) {
   return (res==0 ? 0 : errno);
 }
 
-static void* unix_mmap_prim(void* addr, size_t size, size_t try_alignment, int protect_flags, int flags, int fd) {
+static void* unix_mmap_prim(void* addr, size_t size, int protect_flags, int flags, int fd) {
+  void* p = mmap(addr, size, protect_flags, flags, fd, 0 /* offset */);
+  #if (defined(__linux__) && defined(PR_SET_VMA))
+  if (p!=MAP_FAILED && p!=NULL) {
+    prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, p, size, "mimalloc");
+  }
+  #endif
+  return p;
+}
+
+static void* unix_mmap_prim_aligned(void* addr, size_t size, size_t try_alignment, int protect_flags, int flags, int fd) {
   MI_UNUSED(try_alignment);
   void* p = NULL;
   #if defined(MAP_ALIGNED)  // BSD
   if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) {
     size_t n = mi_bsr(try_alignment);
     if (((size_t)1 << n) == try_alignment && n >= 12 && n <= 30) {  // alignment is a power of 2 and 4096 <= alignment <= 1GiB
-      p = mmap(addr, size, protect_flags, flags | MAP_ALIGNED(n), fd, 0);
+      p = unix_mmap_prim(addr, size, protect_flags, flags | MAP_ALIGNED(n), fd);
       if (p==MAP_FAILED || !_mi_is_aligned(p,try_alignment)) {
         int err = errno;
         _mi_trace_message("unable to directly request aligned OS memory (error: %d (0x%x), size: 0x%zx bytes, alignment: 0x%zx, hint address: %p)\n", err, err, size, try_alignment, addr);
@@ -223,7 +234,7 @@ static void* unix_mmap_prim(void* addr, size_t size, size_t try_alignment, int p
   }
   #elif defined(MAP_ALIGN)  // Solaris
   if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) {
-    p = mmap((void*)try_alignment, size, protect_flags, flags | MAP_ALIGN, fd, 0);  // addr parameter is the required alignment
+    p = unix_mmap_prim((void*)try_alignment, size, protect_flags, flags | MAP_ALIGN, fd);  // addr parameter is the required alignment
     if (p!=MAP_FAILED) return p;
     // fall back to regular mmap
   }
@@ -233,7 +244,7 @@ static void* unix_mmap_prim(void* addr, size_t size, size_t try_alignment, int p
   if (addr == NULL) {
     void* hint = _mi_os_get_aligned_hint(try_alignment, size);
     if (hint != NULL) {
-      p = mmap(hint, size, protect_flags, flags, fd, 0);
+      p = unix_mmap_prim(hint, size, protect_flags, flags, fd);
       if (p==MAP_FAILED || !_mi_is_aligned(p,try_alignment)) {
         #if MI_TRACK_ENABLED  // asan sometimes does not instrument errno correctly?
         int err = 0;
@@ -248,7 +259,7 @@ static void* unix_mmap_prim(void* addr, size_t size, size_t try_alignment, int p
   }
   #endif
   // regular mmap
-  p = mmap(addr, size, protect_flags, flags, fd, 0);
+  p = unix_mmap_prim(addr, size, protect_flags, flags, fd);
   if (p!=MAP_FAILED) return p;
   // failed to allocate
   return NULL;
@@ -319,7 +330,7 @@ static void* unix_mmap(void* addr, size_t size, size_t try_alignment, int protec
       if (large_only || lflags != flags) {
         // try large OS page allocation
         *is_large = true;
-        p = unix_mmap_prim(addr, size, try_alignment, protect_flags, lflags, lfd);
+        p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, lflags, lfd);
         #ifdef MAP_HUGE_1GB
         if (p == NULL && (lflags & MAP_HUGE_1GB) == MAP_HUGE_1GB) {
           mi_huge_pages_available = false; // don't try huge 1GiB pages again
@@ -327,7 +338,7 @@ static void* unix_mmap(void* addr, size_t size, size_t try_alignment, int protec
             _mi_warning_message("unable to allocate huge (1GiB) page, trying large (2MiB) pages instead (errno: %i)\n", errno);
           }
           lflags = ((lflags & ~MAP_HUGE_1GB) | MAP_HUGE_2MB);
-          p = unix_mmap_prim(addr, size, try_alignment, protect_flags, lflags, lfd);
+          p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, lflags, lfd);
         }
         #endif
         if (large_only) return p;
@@ -340,7 +351,7 @@ static void* unix_mmap(void* addr, size_t size, size_t try_alignment, int protec
   // regular allocation
   if (p == NULL) {
     *is_large = false;
-    p = unix_mmap_prim(addr, size, try_alignment, protect_flags, flags, fd);
+    p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, flags, fd);
     if (p != NULL) {
       #if defined(MADV_HUGEPAGE)
       // Many Linux systems don't allow MAP_HUGETLB but they support instead
diff --git a/src/random.c b/src/random.c
index 4fc8b2f8..f17698ba 100644
--- a/src/random.c
+++ b/src/random.c
@@ -143,13 +143,17 @@ void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* ctx_new) {
 
 uintptr_t _mi_random_next(mi_random_ctx_t* ctx) {
   mi_assert_internal(mi_random_is_initialized(ctx));
-  #if MI_INTPTR_SIZE <= 4
-    return chacha_next32(ctx);
-  #elif MI_INTPTR_SIZE == 8
-    return (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx));
-  #else
-  # error "define mi_random_next for this platform"
-  #endif
+  uintptr_t r;
+  do {
+    #if MI_INTPTR_SIZE <= 4
+    r = chacha_next32(ctx);
+    #elif MI_INTPTR_SIZE == 8
+    r = (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx));
+    #else
+    # error "define mi_random_next for this platform"
+    #endif
+  } while (r==0);
+  return r;
 }
 
 
@@ -163,7 +167,7 @@ uintptr_t _mi_os_random_weak(uintptr_t extra_seed) {
   x ^= _mi_prim_clock_now();  
   // and do a few randomization steps
   uintptr_t max = ((x ^ (x >> 17)) & 0x0F) + 1;
-  for (uintptr_t i = 0; i < max; i++) {
+  for (uintptr_t i = 0; i < max || x==0; i++, x++) {
     x = _mi_random_shuffle(x);
   }
   mi_assert_internal(x != 0);
@@ -179,7 +183,7 @@ static void mi_random_init_ex(mi_random_ctx_t* ctx, bool use_weak) {
     if (!use_weak) { _mi_warning_message("unable to use secure randomness\n"); }
     #endif
     uintptr_t x = _mi_os_random_weak(0);
-    for (size_t i = 0; i < 8; i++) {  // key is eight 32-bit words.
+    for (size_t i = 0; i < 8; i++, x++) {  // key is eight 32-bit words.
       x = _mi_random_shuffle(x);
       ((uint32_t*)key)[i] = (uint32_t)x;
     }
diff --git a/src/stats.c b/src/stats.c
index d6c76fa5..dec74f70 100644
--- a/src/stats.c
+++ b/src/stats.c
@@ -30,6 +30,7 @@ static void mi_stat_update(mi_stat_count_t* stat, int64_t amount) {
   {
     // add atomically (for abandoned pages)
     int64_t current = mi_atomic_addi64_relaxed(&stat->current, amount);
+    // if (stat == &_mi_stats_main.committed) { mi_assert_internal(current + amount >= 0); };
     mi_atomic_maxi64_relaxed(&stat->peak, current + amount);
     if (amount > 0) {
       mi_atomic_addi64_relaxed(&stat->total,amount);
@@ -61,6 +62,25 @@ void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount) {
 }
 
 
+static void mi_stat_adjust(mi_stat_count_t* stat, int64_t amount) {
+  if (amount == 0) return;
+  if mi_unlikely(mi_is_in_main(stat))
+  {
+    // adjust atomically 
+    mi_atomic_addi64_relaxed(&stat->current, amount);
+    mi_atomic_addi64_relaxed(&stat->total,amount);
+  }
+  else {
+    // adjust local
+    stat->current += amount;
+    stat->total += amount;
+  }
+}
+
+void _mi_stat_adjust_decrease(mi_stat_count_t* stat, size_t amount) {
+  mi_stat_adjust(stat, -((int64_t)amount));
+}
+
 
 // must be thread safe as it is called from stats_merge
 static void mi_stat_count_add_mt(mi_stat_count_t* stat, const mi_stat_count_t* src) {
@@ -198,6 +218,13 @@ static void mi_stat_peak_print(const mi_stat_count_t* stat, const char* msg, int
   _mi_fprintf(out, arg, "\n");
 }
 
+static void mi_stat_total_print(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg) {
+  _mi_fprintf(out, arg, "%10s:", msg);
+  _mi_fprintf(out, arg, "%12s", " ");  // no peak
+  mi_print_amount(stat->total, unit, out, arg);
+  _mi_fprintf(out, arg, "\n");
+}
+
 static void mi_stat_counter_print(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg ) {
   _mi_fprintf(out, arg, "%10s:", msg);
   mi_print_amount(stat->total, -1, out, arg);
@@ -214,7 +241,7 @@ static void mi_stat_counter_print_avg(const mi_stat_counter_t* stat, const char*
 
 
 static void mi_print_header(mi_output_fun* out, void* arg ) {
-  _mi_fprintf(out, arg, "%10s: %11s %11s %11s %11s %11s\n", "heap stats", "peak   ", "total   ", "current   ", "unit   ", "total#   ");
+  _mi_fprintf(out, arg, "%10s: %11s %11s %11s %11s %11s\n", "heap stats", "peak   ", "total   ", "current   ", "block   ", "total#   ");
 }
 
 #if MI_STAT>1
@@ -283,10 +310,10 @@ static void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0)
   // and print using that
   mi_print_header(out,arg);
   #if MI_STAT>1
-  mi_stats_print_bins(stats->malloc_bins, MI_BIN_HUGE, "normal",out,arg);
+  mi_stats_print_bins(stats->malloc_bins, MI_BIN_HUGE, "bin",out,arg);
   #endif
   #if MI_STAT
-  mi_stat_print(&stats->malloc_normal, "normal", (stats->malloc_normal_count.total == 0 ? 1 : -1), out, arg);
+  mi_stat_print(&stats->malloc_normal, "binned", (stats->malloc_normal_count.total == 0 ? 1 : -1), out, arg);
   // mi_stat_print(&stats->malloc_large, "large", (stats->malloc_large_count.total == 0 ? 1 : -1), out, arg);
   mi_stat_print(&stats->malloc_huge, "huge", (stats->malloc_huge_count.total == 0 ? 1 : -1), out, arg);
   mi_stat_count_t total = { 0,0,0 };
@@ -296,7 +323,7 @@ static void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0)
   mi_stat_print_ex(&total, "total", 1, out, arg, "");
   #endif
   #if MI_STAT>1
-  mi_stat_print_ex(&stats->malloc_requested, "malloc req", 1, out, arg, "");
+  mi_stat_total_print(&stats->malloc_requested, "malloc req", 1, out, arg);
   _mi_fprintf(out, arg, "\n");
   #endif
   mi_stat_print_ex(&stats->reserved, "reserved", 1, out, arg, "");
diff --git a/test/test-api.c b/test/test-api.c
index 15484544..6f5d6722 100644
--- a/test/test-api.c
+++ b/test/test-api.c
@@ -203,7 +203,11 @@ int main(void) {
   CHECK_BODY("malloc-aligned9") { // test large alignments
     bool ok = true;
     void* p[8];
-    size_t sizes[8] = { 8, 512, 1024 * 1024, MI_BLOCK_ALIGNMENT_MAX, MI_BLOCK_ALIGNMENT_MAX + 1, 2 * MI_BLOCK_ALIGNMENT_MAX, 8 * MI_BLOCK_ALIGNMENT_MAX, 0 };
+    size_t sizes[8] = { 8, 512, 1024 * 1024, MI_BLOCK_ALIGNMENT_MAX, MI_BLOCK_ALIGNMENT_MAX + 1, 
+      #if SIZE_MAX > UINT32_MAX
+      2 * MI_BLOCK_ALIGNMENT_MAX, 8 * MI_BLOCK_ALIGNMENT_MAX, 
+      #endif
+      0 };
     for (int i = 0; i < 28 && ok; i++) {
       int align = (1 << i);
       for (int j = 0; j < 8 && ok; j++) {