Upgrade to mimalloc v2.2.3 (#5508)

Git for Windows still carries [mimalloc
v2.1.2](https://github.com/microsoft/mimalloc/releases/tag/v2.1.2),
which is almost 2 years old. Let's upgrade.

Author: dscho

Makefile

@@ -2176,13 +2176,13 @@ ifdef USE_MIMALLOC
 		compat/mimalloc/bitmap.o \
 		compat/mimalloc/heap.o \
 		compat/mimalloc/init.o \
+		compat/mimalloc/libc.o \
 		compat/mimalloc/options.o \
 		compat/mimalloc/os.o \
 		compat/mimalloc/page.o \
 		compat/mimalloc/random.o \
-		compat/mimalloc/prim/windows/prim.o \
+		compat/mimalloc/prim/prim.o \
 		compat/mimalloc/segment.o \
-		compat/mimalloc/segment-cache.o \
 		compat/mimalloc/segment-map.o \
 		compat/mimalloc/stats.o
 
@@ -2194,14 +2194,16 @@ $(MIMALLOC_OBJS): COMPAT_CFLAGS += -DBANNED_H
 $(MIMALLOC_OBJS): COMPAT_CFLAGS += \
 	-Wno-attributes \
 	-Wno-unknown-pragmas \
+	-Wno-unused-function \
 	-Wno-array-bounds
 
 ifdef DEVELOPER
 $(MIMALLOC_OBJS): COMPAT_CFLAGS += \
 	-Wno-pedantic \
 	-Wno-declaration-after-statement \
 	-Wno-old-style-definition \
-	-Wno-missing-prototypes
+	-Wno-missing-prototypes \
+	-Wno-implicit-function-declaration
 endif
 endif
 

compat/.gitattributes

@@ -1 +1,2 @@
 /zlib-uncompress2.c	whitespace=-indent-with-non-tab,-trailing-space
+/mimalloc/**/*	whitespace=-trailing-space

compat/mimalloc/LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2018-2021 Microsoft Corporation, Daan Leijen
+Copyright (c) 2018-2025 Microsoft Corporation, Daan Leijen
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

compat/mimalloc/alloc-aligned.c

@@ -15,68 +15,107 @@ terms of the MIT license. A copy of the license can be found in the file
 // Aligned Allocation
 // ------------------------------------------------------
 
-// Fallback primitive aligned allocation -- split out for better codegen
-static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_fallback(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept
-{
-  mi_assert_internal(size <= PTRDIFF_MAX);
-  mi_assert_internal(alignment != 0 && _mi_is_power_of_two(alignment));
+static bool mi_malloc_is_naturally_aligned( size_t size, size_t alignment ) {
+  // objects up to `MI_MAX_ALIGN_GUARANTEE` are allocated aligned to their size (see `segment.c:_mi_segment_page_start`).
+  mi_assert_internal(_mi_is_power_of_two(alignment) && (alignment > 0));
+  if (alignment > size) return false;
+  if (alignment <= MI_MAX_ALIGN_SIZE) return true;
+  const size_t bsize = mi_good_size(size);
+  return (bsize <= MI_MAX_ALIGN_GUARANTEE && (bsize & (alignment-1)) == 0);
+}
 
-  const uintptr_t align_mask = alignment - 1;  // for any x, `(x & align_mask) == (x % alignment)`
-  const size_t padsize = size + MI_PADDING_SIZE;
+#if MI_GUARDED
+static mi_decl_restrict void* mi_heap_malloc_guarded_aligned(mi_heap_t* heap, size_t size, size_t alignment, bool zero) mi_attr_noexcept {
+  // use over allocation for guarded blocksl
+  mi_assert_internal(alignment > 0 && alignment < MI_BLOCK_ALIGNMENT_MAX);
+  const size_t oversize = size + alignment - 1;
+  void* base = _mi_heap_malloc_guarded(heap, oversize, zero);
+  void* p = mi_align_up_ptr(base, alignment);
+  mi_track_align(base, p, (uint8_t*)p - (uint8_t*)base, size);
+  mi_assert_internal(mi_usable_size(p) >= size);
+  mi_assert_internal(_mi_is_aligned(p, alignment));
+  return p;
+}
 
-  // use regular allocation if it is guaranteed to fit the alignment constraints
-  if (offset==0 && alignment<=padsize && padsize<=MI_MAX_ALIGN_GUARANTEE && (padsize&align_mask)==0) {
-    void* p = _mi_heap_malloc_zero(heap, size, zero);
-    mi_assert_internal(p == NULL || ((uintptr_t)p % alignment) == 0);
-    return p;
-  }
+static void* mi_heap_malloc_zero_no_guarded(mi_heap_t* heap, size_t size, bool zero) {
+  const size_t rate = heap->guarded_sample_rate;
+  // only write if `rate!=0` so we don't write to the constant `_mi_heap_empty`
+  if (rate != 0) { heap->guarded_sample_rate = 0; }
+  void* p = _mi_heap_malloc_zero(heap, size, zero);
+  if (rate != 0) { heap->guarded_sample_rate = rate; }
+  return p;
+}
+#else
+static void* mi_heap_malloc_zero_no_guarded(mi_heap_t* heap, size_t size, bool zero) {
+  return _mi_heap_malloc_zero(heap, size, zero);
+}
+#endif
+
+// Fallback aligned allocation that over-allocates -- split out for better codegen
+static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_overalloc(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept
+{
+  mi_assert_internal(size <= (MI_MAX_ALLOC_SIZE - MI_PADDING_SIZE));
+  mi_assert_internal(alignment != 0 && _mi_is_power_of_two(alignment));
 
   void* p;
   size_t oversize;
-  if mi_unlikely(alignment > MI_ALIGNMENT_MAX) {
+  if mi_unlikely(alignment > MI_BLOCK_ALIGNMENT_MAX) {
     // use OS allocation for very large alignment and allocate inside a huge page (dedicated segment with 1 page)
     // This can support alignments >= MI_SEGMENT_SIZE by ensuring the object can be aligned at a point in the
     // first (and single) page such that the segment info is `MI_SEGMENT_SIZE` bytes before it (so it can be found by aligning the pointer down)
     if mi_unlikely(offset != 0) {
       // todo: cannot support offset alignment for very large alignments yet
-      #if MI_DEBUG > 0
+#if MI_DEBUG > 0
       _mi_error_message(EOVERFLOW, "aligned allocation with a very large alignment cannot be used with an alignment offset (size %zu, alignment %zu, offset %zu)\n", size, alignment, offset);
-      #endif
+#endif
       return NULL;
     }
     oversize = (size <= MI_SMALL_SIZE_MAX ? MI_SMALL_SIZE_MAX + 1 /* ensure we use generic malloc path */ : size);
+    // note: no guarded as alignment > 0
     p = _mi_heap_malloc_zero_ex(heap, oversize, false, alignment); // the page block size should be large enough to align in the single huge page block
     // zero afterwards as only the area from the aligned_p may be committed!
     if (p == NULL) return NULL;
   }
   else {
     // otherwise over-allocate
-    oversize = size + alignment - 1;
-    p = _mi_heap_malloc_zero(heap, oversize, zero);
+    oversize = (size < MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : size) + alignment - 1;  // adjust for size <= 16; with size 0 and aligment 64k, we would allocate a 64k block and pointing just beyond that.
+    p = mi_heap_malloc_zero_no_guarded(heap, oversize, zero);
     if (p == NULL) return NULL;
   }
+  mi_page_t* page = _mi_ptr_page(p);
 
   // .. and align within the allocation
+  const uintptr_t align_mask = alignment - 1;  // for any x, `(x & align_mask) == (x % alignment)`
   const uintptr_t poffset = ((uintptr_t)p + offset) & align_mask;
   const uintptr_t adjust  = (poffset == 0 ? 0 : alignment - poffset);
   mi_assert_internal(adjust < alignment);
   void* aligned_p = (void*)((uintptr_t)p + adjust);
   if (aligned_p != p) {
-    mi_page_t* page = _mi_ptr_page(p);
     mi_page_set_has_aligned(page, true);
+    #if MI_GUARDED
+    // set tag to aligned so mi_usable_size works with guard pages
+    if (adjust >= sizeof(mi_block_t)) {
+      mi_block_t* const block = (mi_block_t*)p;
+      block->next = MI_BLOCK_TAG_ALIGNED;
+    }
+    #endif
     _mi_padding_shrink(page, (mi_block_t*)p, adjust + size);
   }
   // todo: expand padding if overallocated ?
 
-  mi_assert_internal(mi_page_usable_block_size(_mi_ptr_page(p)) >= adjust + size);
-  mi_assert_internal(p == _mi_page_ptr_unalign(_mi_ptr_segment(aligned_p), _mi_ptr_page(aligned_p), aligned_p));
+  mi_assert_internal(mi_page_usable_block_size(page) >= adjust + size);
   mi_assert_internal(((uintptr_t)aligned_p + offset) % alignment == 0);
   mi_assert_internal(mi_usable_size(aligned_p)>=size);
   mi_assert_internal(mi_usable_size(p) == mi_usable_size(aligned_p)+adjust);
+  #if MI_DEBUG > 1
+  mi_page_t* const apage = _mi_ptr_page(aligned_p);
+  void* unalign_p = _mi_page_ptr_unalign(apage, aligned_p);
+  mi_assert_internal(p == unalign_p);
+  #endif
 
   // now zero the block if needed
-  if (alignment > MI_ALIGNMENT_MAX) {
-    // for the tracker, on huge aligned allocations only from the start of the large block is defined
+  if (alignment > MI_BLOCK_ALIGNMENT_MAX) {
+    // for the tracker, on huge aligned allocations only the memory from the start of the large block is defined
     mi_track_mem_undefined(aligned_p, size);
     if (zero) {
       _mi_memzero_aligned(aligned_p, mi_usable_size(aligned_p));
@@ -85,10 +124,47 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_fallback(mi_heap_t*
 
   if (p != aligned_p) {
     mi_track_align(p,aligned_p,adjust,mi_usable_size(aligned_p));
+    #if MI_GUARDED
+    mi_track_mem_defined(p, sizeof(mi_block_t));
+    #endif
   }
   return aligned_p;
 }
 
+// Generic primitive aligned allocation -- split out for better codegen
+static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_generic(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept
+{
+  mi_assert_internal(alignment != 0 && _mi_is_power_of_two(alignment));
+  // we don't allocate more than MI_MAX_ALLOC_SIZE (see <https://sourceware.org/ml/libc-announce/2019/msg00001.html>)
+  if mi_unlikely(size > (MI_MAX_ALLOC_SIZE - MI_PADDING_SIZE)) {
+    #if MI_DEBUG > 0
+    _mi_error_message(EOVERFLOW, "aligned allocation request is too large (size %zu, alignment %zu)\n", size, alignment);
+    #endif
+    return NULL;
+  }
+
+  // use regular allocation if it is guaranteed to fit the alignment constraints.
+  // this is important to try as the fast path in `mi_heap_malloc_zero_aligned` only works when there exist
+  // a page with the right block size, and if we always use the over-alloc fallback that would never happen.
+  if (offset == 0 && mi_malloc_is_naturally_aligned(size,alignment)) {
+    void* p = mi_heap_malloc_zero_no_guarded(heap, size, zero);
+    mi_assert_internal(p == NULL || ((uintptr_t)p % alignment) == 0);
+    const bool is_aligned_or_null = (((uintptr_t)p) & (alignment-1))==0;
+    if mi_likely(is_aligned_or_null) {
+      return p;
+    }
+    else {
+      // this should never happen if the `mi_malloc_is_naturally_aligned` check is correct..
+      mi_assert(false);
+      mi_free(p);
+    }
+  }
+
+  // fall back to over-allocation
+  return mi_heap_malloc_zero_aligned_at_overalloc(heap,size,alignment,offset,zero);
+}
+
+
 // Primitive aligned allocation
 static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept
 {
@@ -100,33 +176,32 @@ static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t
     return NULL;
   }
 
-  if mi_unlikely(size > PTRDIFF_MAX) {          // we don't allocate more than PTRDIFF_MAX (see <https://sourceware.org/ml/libc-announce/2019/msg00001.html>)
-    #if MI_DEBUG > 0
-    _mi_error_message(EOVERFLOW, "aligned allocation request is too large (size %zu, alignment %zu)\n", size, alignment);
-    #endif
-    return NULL;
+  #if MI_GUARDED
+  if (offset==0 && alignment < MI_BLOCK_ALIGNMENT_MAX && mi_heap_malloc_use_guarded(heap,size)) {
+    return mi_heap_malloc_guarded_aligned(heap, size, alignment, zero);
   }
-  const uintptr_t align_mask = alignment-1;       // for any x, `(x & align_mask) == (x % alignment)`
-  const size_t padsize = size + MI_PADDING_SIZE;  // note: cannot overflow due to earlier size > PTRDIFF_MAX check
+  #endif
 
   // try first if there happens to be a small block available with just the right alignment
-  if mi_likely(padsize <= MI_SMALL_SIZE_MAX && alignment <= padsize) {
+  if mi_likely(size <= MI_SMALL_SIZE_MAX && alignment <= size) {
+    const uintptr_t align_mask = alignment-1;       // for any x, `(x & align_mask) == (x % alignment)`
+    const size_t padsize = size + MI_PADDING_SIZE;
     mi_page_t* page = _mi_heap_get_free_small_page(heap, padsize);
-    const bool is_aligned = (((uintptr_t)page->free+offset) & align_mask)==0;
-    if mi_likely(page->free != NULL && is_aligned)
-    {
-      #if MI_STAT>1
-      mi_heap_stat_increase(heap, malloc, size);
-      #endif
-      void* p = _mi_page_malloc(heap, page, padsize, zero); // TODO: inline _mi_page_malloc
-      mi_assert_internal(p != NULL);
-      mi_assert_internal(((uintptr_t)p + offset) % alignment == 0);
-      mi_track_malloc(p,size,zero);
-      return p;
+    if mi_likely(page->free != NULL) {
+      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
+        mi_assert_internal(p != NULL);
+        mi_assert_internal(((uintptr_t)p + offset) % alignment == 0);
+        mi_track_malloc(p,size,zero);
+        return p;
+      }
     }
   }
-  // fallback
-  return mi_heap_malloc_zero_aligned_at_fallback(heap, size, alignment, offset, zero);
+
+  // fallback to generic aligned allocation
+  return mi_heap_malloc_zero_aligned_at_generic(heap, size, alignment, offset, zero);
 }
 
 
@@ -139,27 +214,12 @@ mi_decl_nodiscard mi_decl_restrict void* mi_heap_malloc_aligned_at(mi_heap_t* he
 }
 
 mi_decl_nodiscard mi_decl_restrict void* mi_heap_malloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept {
-  if mi_unlikely(alignment == 0 || !_mi_is_power_of_two(alignment)) return NULL;
-  #if !MI_PADDING
-  // without padding, any small sized allocation is naturally aligned (see also `_mi_segment_page_start`)
-  if mi_likely(_mi_is_power_of_two(size) && size >= alignment && size <= MI_SMALL_SIZE_MAX)
-  #else
-  // with padding, we can only guarantee this for fixed alignments
-  if mi_likely((alignment == sizeof(void*) || (alignment == MI_MAX_ALIGN_SIZE && size > (MI_MAX_ALIGN_SIZE/2)))
-		&& size <= MI_SMALL_SIZE_MAX)
-  #endif
-  {
-    // fast path for common alignment and size
-    return mi_heap_malloc_small(heap, size);
-  }
-  else {
-    return mi_heap_malloc_aligned_at(heap, size, alignment, 0);
-  }
+  return mi_heap_malloc_aligned_at(heap, size, alignment, 0);
 }
 
 // ensure a definition is emitted
 #if defined(__cplusplus)
-static void* _mi_heap_malloc_aligned = (void*)&mi_heap_malloc_aligned;
+void* _mi_extern_heap_malloc_aligned = (void*)&mi_heap_malloc_aligned;
 #endif
 
 // ------------------------------------------------------
@@ -227,9 +287,9 @@ static void* mi_heap_realloc_zero_aligned_at(mi_heap_t* heap, void* p, size_t ne
     void* newp = mi_heap_malloc_aligned_at(heap,newsize,alignment,offset);
     if (newp != NULL) {
       if (zero && newsize > size) {
-	// also set last word in the previous allocation to zero to ensure any padding is zero-initialized
-	size_t start = (size >= sizeof(intptr_t) ? size - sizeof(intptr_t) : 0);
-	_mi_memzero((uint8_t*)newp + start, newsize - start);
+        // also set last word in the previous allocation to zero to ensure any padding is zero-initialized
+        size_t start = (size >= sizeof(intptr_t) ? size - sizeof(intptr_t) : 0);
+        _mi_memzero((uint8_t*)newp + start, newsize - start);
       }
       _mi_memcpy_aligned(newp, p, (newsize > size ? size : newsize));
       mi_free(p); // only free if successful
@@ -296,3 +356,5 @@ mi_decl_nodiscard void* mi_recalloc_aligned_at(void* p, size_t newcount, size_t
 mi_decl_nodiscard void* mi_recalloc_aligned(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept {
   return mi_heap_recalloc_aligned(mi_prim_get_default_heap(), p, newcount, size, alignment);
 }
+
+