diff --git a/include/mimalloc/types.h b/include/mimalloc/types.h
index 91d8eecb..5a3f5fe2 100644
--- a/include/mimalloc/types.h
+++ b/include/mimalloc/types.h
@@ -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/src/alloc-aligned.c b/src/alloc-aligned.c
index bb9e9ecb..ee4e129f 100644
--- a/src/alloc-aligned.c
+++ b/src/alloc-aligned.c
@@ -191,10 +191,11 @@ 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)
       {
+        void* p = (zero ? _mi_page_malloc_zeroed(heap,page,padsize) : _mi_page_malloc(heap,page,padsize)); // call specific page malloc for better codegen
         #if MI_STAT>1
+        mi_heap_stat_adjust_decrease(heap, malloc_requested, padsize);
         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);
         mi_track_malloc(p,size,zero);
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/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/stats.c b/src/stats.c
index 7bfd091a..dec74f70 100644
--- a/src/stats.c
+++ b/src/stats.c
@@ -62,6 +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) {
@@ -199,6 +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);
@@ -215,7 +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
@@ -284,10 +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 };
@@ -297,7 +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++) {