merge from dev

This commit is contained in:
daanx
2024-11-17 23:29:21 -08:00
24 changed files with 572 additions and 257 deletions

View File

@@ -20,14 +20,36 @@ static bool mi_malloc_is_naturally_aligned( size_t size, size_t alignment ) {
mi_assert_internal(_mi_is_power_of_two(alignment) && (alignment > 0));
if (alignment > size) return false;
if (alignment <= MI_MAX_ALIGN_SIZE) return true;
#if MI_DEBUG_GUARDED
return false;
#else
const size_t bsize = mi_good_size(size);
return (bsize <= MI_MAX_ALIGN_GUARANTEE && (bsize & (alignment-1)) == 0);
#endif
}
#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;
}
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;
heap->guarded_sample_rate = 0;
void* p = _mi_heap_malloc_zero(heap, size, zero);
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
{
@@ -48,6 +70,7 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_overalloc(mi_heap_t
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;
@@ -55,11 +78,11 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_overalloc(mi_heap_t
else {
// otherwise over-allocate
oversize = size + alignment - 1;
p = _mi_heap_malloc_zero(heap, oversize, zero);
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;
@@ -68,6 +91,13 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_overalloc(mi_heap_t
void* aligned_p = (void*)((uintptr_t)p + adjust);
if (aligned_p != 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 ?
@@ -76,8 +106,10 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_overalloc(mi_heap_t
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_GUARDED
mi_assert_internal(p == _mi_page_ptr_unalign(_mi_ptr_page(aligned_p), aligned_p));
#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
@@ -91,6 +123,9 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_overalloc(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;
}
@@ -100,27 +135,27 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_generic(mi_heap_t*
{
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_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(heap, size, zero);
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;
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);
mi_free(p);
}
}
@@ -128,6 +163,7 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_generic(mi_heap_t*
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
{
@@ -138,12 +174,17 @@ static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t
#endif
return NULL;
}
#if !MI_DEBUG_GUARDED
#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);
}
#endif
// try first if there happens to be a small block available with just the right alignment
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;
const size_t padsize = size + MI_PADDING_SIZE;
mi_page_t* page = _mi_heap_get_free_small_page(heap, padsize);
if mi_likely(page->free != NULL) {
const bool is_aligned = (((uintptr_t)page->free + offset) & align_mask)==0;
@@ -160,7 +201,6 @@ static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t
}
}
}
#endif
// fallback to generic aligned allocation
return mi_heap_malloc_zero_aligned_at_generic(heap, size, alignment, offset, zero);
@@ -318,3 +358,5 @@ 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);
}