kiper220/mimalloc
1
0
Fork 0
mirror of https://github.com/microsoft/mimalloc.git synced 2025-07-06 11:34:38 +03:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'dev3' into dev3-bin

Browse source
This commit is contained in:
daanx 2024-12-31 15:11:24 -08:00
commit ff52ea0553
34 changed files with 220 additions and 2812 deletions
Download patch file Download diff file Expand all files Collapse all files

5
include/mimalloc/atomic.h
View file

@ -22,7 +22,7 @@ terms of the MIT license. A copy of the license can be found in the file
// --------------------------------------------------------------------------------------------
// 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.
// We base the primitives on the C/C++ atomics and create a minimal 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.
@ -72,6 +72,7 @@ terms of the MIT license. A copy of the license can be found in the file
#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_relaxed(p,x) mi_atomic(exchange_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))
@ -115,6 +116,7 @@ static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub);
#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_relaxed(tp,p,x) mi_atomic_exchange_relaxed(p,(tp*)x)
#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
@ -123,6 +125,7 @@ static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub);
#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_relaxed(tp,p,x) mi_atomic_exchange_relaxed(p,x)
#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

4
include/mimalloc/internal.h
View file

@ -10,7 +10,7 @@ terms of the MIT license. A copy of the license can be found in the file
// --------------------------------------------------------------------------
// This file contains the interal API's of mimalloc and various utility
// This file contains the internal API's of mimalloc and various utility
// functions and macros.
// --------------------------------------------------------------------------
@ -206,7 +206,7 @@ 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);
bool _mi_heap_memid_is_suitable(mi_heap_t* heap, mi_memid_t memid);
void _mi_heap_unsafe_destroy_all(void);
void _mi_heap_unsafe_destroy_all(mi_heap_t* heap);
mi_heap_t* _mi_heap_by_tag(mi_heap_t* heap, uint8_t tag);
void _mi_heap_area_init(mi_heap_area_t* area, mi_page_t* page);
bool _mi_heap_area_visit_blocks(const mi_heap_area_t* area, mi_page_t* page, mi_block_visit_fun* visitor, void* arg);

1
include/mimalloc/prim.h
View file

@ -257,6 +257,7 @@ static inline void mi_prim_tls_slot_set(size_t slot, void* value) mi_attr_noexce
// Nevertheless, it seems needed on older graviton platforms (see issue #851).
// For now, we only enable this for specific platforms.
#if !defined(__APPLE__) /* on apple (M1) the wrong register is read (tpidr_el0 instead of tpidrro_el0) so fall back to TLS slot assembly (<https://github.com/microsoft/mimalloc/issues/343#issuecomment-763272369>)*/ \
&& !defined(__CYGWIN__) \
&& !defined(MI_LIBC_MUSL) \
&& (!defined(__clang_major__) || __clang_major__ >= 14) /* older clang versions emit bad code; fall back to using the TLS slot (<https://lore.kernel.org/linux-arm-kernel/202110280952.352F66D8@keescook/T/>) */
#if (defined(__GNUC__) && (__GNUC__ >= 7) && defined(__aarch64__)) /* aarch64 for older gcc versions (issue #851) */ \

27
include/mimalloc/types.h
View file

@ -100,9 +100,10 @@ terms of the MIT license. A copy of the license can be found in the file
#endif
// ------------------------------------------------------
// --------------------------------------------------------------
// Sizes of internal data-structures
// ------------------------------------------------------
// (comments specify sizes on 64-bit, usually 32-bit is halved)
// --------------------------------------------------------------
// Sizes are for 64-bit
#ifndef MI_ARENA_SLICE_SHIFT
@ -116,19 +117,19 @@ terms of the MIT license. A copy of the license can be found in the file
#define MI_BCHUNK_BITS_SHIFT (6 + MI_SIZE_SHIFT) // optimized for 512 bits per chunk (avx512)
#endif
#define MI_BCHUNK_BITS (1 << MI_BCHUNK_BITS_SHIFT)
#define MI_ARENA_SLICE_SIZE (MI_ZU(1) << MI_ARENA_SLICE_SHIFT)
#define MI_BCHUNK_BITS (1 << MI_BCHUNK_BITS_SHIFT) // sub-bitmaps are "bchunks" of 512 bits
#define MI_ARENA_SLICE_SIZE (MI_ZU(1) << MI_ARENA_SLICE_SHIFT) // arena's allocate in slices of 64 KiB
#define MI_ARENA_SLICE_ALIGN (MI_ARENA_SLICE_SIZE)
#define MI_ARENA_MIN_OBJ_SLICES (1)
#define MI_ARENA_MAX_OBJ_SLICES (MI_BCHUNK_BITS) // 32 MiB (for now, cannot cross chunk boundaries)
#define MI_ARENA_MIN_OBJ_SLICES (1)
#define MI_ARENA_MAX_OBJ_SLICES (MI_BCHUNK_BITS) // 32 MiB (for now, cannot cross chunk boundaries)
#define MI_ARENA_MIN_OBJ_SIZE (MI_ARENA_MIN_OBJ_SLICES * MI_ARENA_SLICE_SIZE)
#define MI_ARENA_MAX_OBJ_SIZE (MI_ARENA_MAX_OBJ_SLICES * MI_ARENA_SLICE_SIZE)
#define MI_SMALL_PAGE_SIZE MI_ARENA_MIN_OBJ_SIZE
#define MI_MEDIUM_PAGE_SIZE (8*MI_SMALL_PAGE_SIZE) // 512 KiB (=byte in the bitmap)
#define MI_LARGE_PAGE_SIZE (MI_SIZE_SIZE*MI_MEDIUM_PAGE_SIZE) // 4 MiB (=word in the bitmap)
#define MI_SMALL_PAGE_SIZE MI_ARENA_MIN_OBJ_SIZE // 64 KiB
#define MI_MEDIUM_PAGE_SIZE (8*MI_SMALL_PAGE_SIZE) // 512 KiB (=byte in the bchunk bitmap)
#define MI_LARGE_PAGE_SIZE (MI_SIZE_SIZE*MI_MEDIUM_PAGE_SIZE) // 4 MiB (=word in the bchunk bitmap)
// Maximum number of size classes. (spaced exponentially in 12.5% increments)
#define MI_BIN_HUGE (73U)
@ -272,7 +273,7 @@ typedef uint8_t mi_heaptag_t;
//
// Notes:
// - Non-atomic fields can only be accessed if having ownership (low bit of `xthread_free`).
// - If a page is not part of a heap it is called "abandoned" -- in
// - If a page is not part of a heap it is called "abandoned" (`heap==NULL`) -- in
// that case the `xthreadid` is 0 or 1 (1 is for abandoned pages that
// are in the abandoned page lists of an arena, these are called "mapped" abandoned pages).
// - The layout is optimized for `free.c:mi_free` and `alloc.c:mi_page_alloc`
@ -304,7 +305,7 @@ typedef struct mi_page_s {
mi_heap_t* heap; // the heap owning this page (or NULL for abandoned pages)
struct mi_page_s* next; // next page owned by the heap with the same `block_size`
struct mi_page_s* prev; // previous page owned by the heap with the same `block_size`
size_t slice_committed; // committed size relative to the first arena slice of the page data
size_t slice_committed; // committed size relative to the first arena slice of the page data (or 0 if the page is fully committed already)
mi_memid_t memid; // provenance of the page memory
} mi_page_t;
@ -315,7 +316,7 @@ typedef struct mi_page_s {
#define MI_PAGE_ALIGN MI_ARENA_SLICE_ALIGN // pages must be aligned on this for the page map.
#define MI_PAGE_MIN_START_BLOCK_ALIGN MI_MAX_ALIGN_SIZE // minimal block alignment for the first block in a page (16b)
#define MI_PAGE_MAX_START_BLOCK_ALIGN2 MI_KiB // maximal block alignment for "power of 2"-sized blocks
#define MI_PAGE_MAX_START_BLOCK_ALIGN2 MI_KiB // maximal block alignment for "power of 2"-sized blocks (such that we guarantee natural alignment)
#define MI_PAGE_MAX_OVERALLOC_ALIGN MI_ARENA_SLICE_SIZE // (64 KiB) limit for which we overallocate in arena pages, beyond this use OS allocation
#if (MI_ENCODE_FREELIST || MI_PADDING) && MI_SIZE_SIZE == 8
@ -328,7 +329,7 @@ typedef struct mi_page_s {
// (Except for large pages since huge objects are allocated in 4MiB chunks)
#define MI_SMALL_MAX_OBJ_SIZE ((MI_SMALL_PAGE_SIZE-MI_PAGE_INFO_SIZE)/8) // < 8 KiB
#define MI_MEDIUM_MAX_OBJ_SIZE ((MI_MEDIUM_PAGE_SIZE-MI_PAGE_INFO_SIZE)/8) // < 64 KiB
#define MI_LARGE_MAX_OBJ_SIZE (MI_LARGE_PAGE_SIZE/4) // <= 512 KiB // note: this must be a nice power of 2 or we get rounding issues with _mi_bin
#define MI_LARGE_MAX_OBJ_SIZE (MI_LARGE_PAGE_SIZE/4) // <= 512 KiB // note: this must be a nice power of 2 or we get rounding issues with `_mi_bin`
#define MI_LARGE_MAX_OBJ_WSIZE (MI_LARGE_MAX_OBJ_SIZE/MI_SIZE_SIZE)