wip: new segment allocation with flexible large objects

This commit is contained in:
daan 2019-08-15 00:46:45 -07:00
parent 6d11e59250
commit bbd81bbbd1
15 changed files with 843 additions and 439 deletions

View File

@ -95,7 +95,7 @@
<SDLCheck>true</SDLCheck>
<ConformanceMode>true</ConformanceMode>
<AdditionalIncludeDirectories>../../include</AdditionalIncludeDirectories>
<PreprocessorDefinitions>MI_SHARED_LIB;MI_SHARED_LIB_EXPORT;MI_MALLOC_OVERRIDE;%(PreprocessorDefinitions);</PreprocessorDefinitions>
<PreprocessorDefinitions>MI_DEBUG=3;MI_SHARED_LIB;MI_SHARED_LIB_EXPORT;MI_MALLOC_OVERRIDE;%(PreprocessorDefinitions);</PreprocessorDefinitions>
<RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
<SupportJustMyCode>false</SupportJustMyCode>
<CompileAs>Default</CompileAs>
@ -118,7 +118,7 @@
<SDLCheck>true</SDLCheck>
<ConformanceMode>true</ConformanceMode>
<AdditionalIncludeDirectories>../../include</AdditionalIncludeDirectories>
<PreprocessorDefinitions>MI_SHARED_LIB;MI_SHARED_LIB_EXPORT;MI_MALLOC_OVERRIDE;%(PreprocessorDefinitions);</PreprocessorDefinitions>
<PreprocessorDefinitions>MI_DEBUG=3;MI_SHARED_LIB;MI_SHARED_LIB_EXPORT;MI_MALLOC_OVERRIDE;%(PreprocessorDefinitions);</PreprocessorDefinitions>
<RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
<SupportJustMyCode>false</SupportJustMyCode>
<CompileAs>Default</CompileAs>
@ -225,7 +225,6 @@
<ClCompile Include="..\..\src\alloc.c" />
<ClCompile Include="..\..\src\heap.c" />
<ClCompile Include="..\..\src\init.c" />
<ClCompile Include="..\..\src\memory.c" />
<ClCompile Include="..\..\src\options.c" />
<ClCompile Include="..\..\src\os.c" />
<ClCompile Include="..\..\src\page-queue.c">

View File

@ -58,9 +58,6 @@
<ClCompile Include="..\..\src\init.c">
<Filter>Source Files</Filter>
</ClCompile>
<ClCompile Include="..\..\src\memory.c">
<Filter>Source Files</Filter>
</ClCompile>
<ClCompile Include="..\..\src\alloc-override.c">
<Filter>Source Files</Filter>
</ClCompile>

View File

@ -227,7 +227,6 @@
<ClCompile Include="..\..\src\alloc.c" />
<ClCompile Include="..\..\src\heap.c" />
<ClCompile Include="..\..\src\init.c" />
<ClCompile Include="..\..\src\memory.c" />
<ClCompile Include="..\..\src\options.c" />
<ClCompile Include="..\..\src\page-queue.c">
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">true</ExcludedFromBuild>

View File

@ -47,9 +47,6 @@
<ClCompile Include="..\..\src\init.c">
<Filter>Source Files</Filter>
</ClCompile>
<ClCompile Include="..\..\src\memory.c">
<Filter>Source Files</Filter>
</ClCompile>
<ClCompile Include="..\..\src\alloc-posix.c">
<Filter>Source Files</Filter>
</ClCompile>

View File

@ -39,10 +39,20 @@ bool _mi_preloading(); // true while the C runtime is not ready
// os.c
size_t _mi_os_page_size(void);
size_t _mi_os_large_page_size();
void _mi_os_init(void); // called from process init
void* _mi_os_alloc(size_t size, mi_stats_t* stats); // to allocate thread local data
void _mi_os_free(void* p, size_t size, mi_stats_t* stats); // to free thread local data
bool _mi_os_protect(void* addr, size_t size);
bool _mi_os_unprotect(void* addr, size_t size);
bool _mi_os_commit(void* p, size_t size, mi_stats_t* stats);
bool _mi_os_decommit(void* p, size_t size, mi_stats_t* stats);
bool _mi_os_reset(void* p, size_t size, mi_stats_t* stats);
bool _mi_os_unreset(void* p, size_t size, mi_stats_t* stats);
void* _mi_os_alloc_aligned(size_t size, size_t alignment, bool commit, mi_os_tld_t* tld);
/*
// memory.c
void* _mi_mem_alloc_aligned(size_t size, size_t alignment, bool commit, size_t* id, mi_os_tld_t* tld);
void* _mi_mem_alloc(size_t size, bool commit, size_t* id, mi_os_tld_t* tld);
@ -55,6 +65,7 @@ bool _mi_mem_protect(void* addr, size_t size);
bool _mi_mem_unprotect(void* addr, size_t size);
void _mi_mem_collect(mi_stats_t* stats);
*/
// "segment.c"
mi_page_t* _mi_segment_page_alloc(size_t block_wsize, mi_segments_tld_t* tld, mi_os_tld_t* os_tld);
@ -62,7 +73,7 @@ void _mi_segment_page_free(mi_page_t* page, bool force, mi_segments_tld_t*
void _mi_segment_page_abandon(mi_page_t* page, mi_segments_tld_t* tld);
bool _mi_segment_try_reclaim_abandoned( mi_heap_t* heap, bool try_all, mi_segments_tld_t* tld);
void _mi_segment_thread_collect(mi_segments_tld_t* tld);
uint8_t* _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t block_size, size_t* page_size); // page start for any page
uint8_t* _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size); // page start for any page
// "page.c"
void* _mi_malloc_generic(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc;
@ -233,27 +244,47 @@ static inline mi_segment_t* _mi_ptr_segment(const void* p) {
return (mi_segment_t*)((uintptr_t)p & ~MI_SEGMENT_MASK);
}
static inline mi_page_t* mi_slice_to_page(mi_slice_t* s) {
mi_assert_internal(s->slice_offset== 0 && s->slice_count > 0);
return (mi_page_t*)(s);
}
static inline mi_slice_t* mi_page_to_slice(mi_page_t* p) {
mi_assert_internal(p->slice_offset== 0 && p->slice_count > 0);
return (mi_slice_t*)(p);
}
static size_t mi_slice_index(const mi_slice_t* slice) {
mi_segment_t* segment = _mi_ptr_segment(slice);
ptrdiff_t index = slice - segment->slices;
mi_assert_internal(index >= 0 && index < (ptrdiff_t)segment->slice_count);
return index;
}
// Segment belonging to a page
static inline mi_segment_t* _mi_page_segment(const mi_page_t* page) {
mi_segment_t* segment = _mi_ptr_segment(page);
mi_assert_internal(segment == NULL || page == &segment->pages[page->segment_idx]);
mi_assert_internal(segment == NULL || page == mi_slice_to_page(&segment->slices[mi_slice_index(mi_page_to_slice((mi_page_t*)page))]));
return segment;
}
// Get the page containing the pointer
static inline mi_page_t* _mi_segment_page_of(const mi_segment_t* segment, const void* p) {
// if (segment->page_size > MI_SEGMENT_SIZE) return &segment->pages[0]; // huge pages
ptrdiff_t diff = (uint8_t*)p - (uint8_t*)segment;
mi_assert_internal(diff >= 0 && diff < MI_SEGMENT_SIZE);
uintptr_t idx = (uintptr_t)diff >> segment->page_shift;
mi_assert_internal(idx < segment->capacity);
mi_assert_internal(segment->page_kind <= MI_PAGE_MEDIUM || idx == 0);
return &((mi_segment_t*)segment)->pages[idx];
uintptr_t idx = (uintptr_t)diff >> MI_SEGMENT_SLICE_SHIFT;
mi_assert_internal(idx < segment->slice_count);
mi_slice_t* slice0 = (mi_slice_t*)&segment->slices[idx];
mi_slice_t* slice = slice0 - slice0->slice_offset; // adjust to the block that holds the page data
mi_assert_internal(slice->slice_count > slice0->slice_offset);
mi_assert_internal(slice->slice_offset == 0);
mi_assert_internal(slice >= segment->slices && slice < segment->slices + segment->slice_count);
return mi_slice_to_page(slice);
}
// Quick page start for initialized pages
static inline uint8_t* _mi_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size) {
return _mi_segment_page_start(segment, page, page->block_size, page_size);
return _mi_segment_page_start(segment, page, page_size);
}
// Get the page containing the pointer

View File

@ -74,27 +74,28 @@ terms of the MIT license. A copy of the license can be found in the file
// Main tuning parameters for segment and page sizes
// Sizes for 64-bit, divide by two for 32-bit
#define MI_SMALL_PAGE_SHIFT (13 + MI_INTPTR_SHIFT) // 64kb
#define MI_MEDIUM_PAGE_SHIFT ( 3 + MI_SMALL_PAGE_SHIFT) // 512kb
#define MI_LARGE_PAGE_SHIFT ( 3 + MI_MEDIUM_PAGE_SHIFT) // 4mb
#define MI_SEGMENT_SHIFT ( MI_LARGE_PAGE_SHIFT) // 4mb
#define MI_SEGMENT_SLICE_SHIFT (13 + MI_INTPTR_SHIFT) // 64kb
#define MI_SEGMENT_SHIFT (10 + MI_SEGMENT_SLICE_SHIFT) // 64mb
#define MI_SMALL_PAGE_SHIFT (MI_SEGMENT_SLICE_SHIFT) // 64kb
#define MI_MEDIUM_PAGE_SHIFT ( 3 + MI_SEGMENT_SLICE_SHIFT) // 512kb
// Derived constants
#define MI_SEGMENT_SIZE (1<<MI_SEGMENT_SHIFT)
#define MI_SEGMENT_MASK ((uintptr_t)MI_SEGMENT_SIZE - 1)
#define MI_SEGMENT_SIZE ((size_t)1<<MI_SEGMENT_SHIFT)
#define MI_SEGMENT_MASK (MI_SEGMENT_SIZE - 1)
#define MI_SEGMENT_SLICE_SIZE ((size_t)1 << MI_SEGMENT_SLICE_SHIFT)
#define MI_SLICES_PER_SEGMENT (MI_SEGMENT_SIZE / MI_SEGMENT_SLICE_SIZE) // 1024
#define MI_SMALL_PAGE_SIZE (1<<MI_SMALL_PAGE_SHIFT)
#define MI_MEDIUM_PAGE_SIZE (1<<MI_MEDIUM_PAGE_SHIFT)
#define MI_LARGE_PAGE_SIZE (1<<MI_LARGE_PAGE_SHIFT)
#define MI_SMALL_PAGES_PER_SEGMENT (MI_SEGMENT_SIZE/MI_SMALL_PAGE_SIZE)
#define MI_MEDIUM_PAGES_PER_SEGMENT (MI_SEGMENT_SIZE/MI_MEDIUM_PAGE_SIZE)
#define MI_LARGE_PAGES_PER_SEGMENT (MI_SEGMENT_SIZE/MI_LARGE_PAGE_SIZE)
#define MI_MEDIUM_SIZE_MAX (MI_MEDIUM_PAGE_SIZE/8) // 64kb on 64-bit
#define MI_MEDIUM_WSIZE_MAX (MI_MEDIUM_SIZE_MAX/MI_INTPTR_SIZE) // 64kb on 64-bit
#define MI_LARGE_SIZE_MAX (MI_SEGMENT_SIZE/4) // 16mb on 64-bit
#define MI_LARGE_WSIZE_MAX (MI_LARGE_SIZE_MAX/MI_INTPTR_SIZE)
#define MI_MEDIUM_SIZE_MAX (MI_MEDIUM_PAGE_SIZE/4) // 128kb on 64-bit
#define MI_LARGE_SIZE_MAX (MI_LARGE_PAGE_SIZE/4) // 1Mb on 64-bit
#define MI_LARGE_WSIZE_MAX (MI_LARGE_SIZE_MAX>>MI_INTPTR_SHIFT)
#define MI_HUGE_SIZE_MAX (2*MI_INTPTR_SIZE*MI_SEGMENT_SIZE) // (must match MI_REGION_MAX_ALLOC_SIZE in memory.c)
// Minimal alignment necessary. On most platforms 16 bytes are needed
// due to SSE registers for example. This must be at least `MI_INTPTR_SIZE`
@ -103,7 +104,7 @@ terms of the MIT license. A copy of the license can be found in the file
// Maximum number of size classes. (spaced exponentially in 12.5% increments)
#define MI_BIN_HUGE (73U)
#if (MI_LARGE_WSIZE_MAX >= 655360)
#if (MI_MEDIUM_WSIZE_MAX >= 655360)
#error "define more bins"
#endif
@ -154,20 +155,20 @@ typedef uintptr_t mi_thread_free_t;
// - using `uint16_t` does not seem to slow things down
typedef struct mi_page_s {
// "owned" by the segment
uint8_t segment_idx; // index in the segment `pages` array, `page == &segment->pages[page->segment_idx]`
bool segment_in_use:1; // `true` if the segment allocated this page
bool is_reset:1; // `true` if the page memory was reset
bool is_committed:1; // `true` if the page virtual memory is committed
size_t slice_count; // slices in this page (0 if not a page)
uint16_t slice_offset; // distance from the actual page data slice (0 if a page)
bool is_reset; // `true` if the page memory was reset
bool is_committed; // `true` if the page virtual memory is committed
// layout like this to optimize access in `mi_malloc` and `mi_free`
uint16_t capacity; // number of blocks committed
uint16_t reserved; // number of blocks reserved in memory
// 16 bits padding
mi_block_t* free; // list of available free blocks (`malloc` allocates from this list)
#if MI_SECURE
uintptr_t cookie; // random cookie to encode the free lists
#endif
mi_page_flags_t flags; // threadid:62 | has_aligned:1 | in_full:1
mi_page_flags_t flags;
size_t used; // number of blocks in use (including blocks in `local_free` and `thread_free`)
mi_block_t* local_free; // list of deferred free blocks by this thread (migrates to `free`)
@ -182,7 +183,7 @@ typedef struct mi_page_s {
// improve page index calculation
#if (MI_INTPTR_SIZE==8 && MI_SECURE==0)
void* padding[1]; // 12 words on 64-bit
// void* padding[1]; // 12 words on 64-bit
#elif MI_INTPTR_SIZE==4
// void* padding[1]; // 12 words on 32-bit
#endif
@ -193,30 +194,37 @@ typedef struct mi_page_s {
typedef enum mi_page_kind_e {
MI_PAGE_SMALL, // small blocks go into 64kb pages inside a segment
MI_PAGE_MEDIUM, // medium blocks go into 512kb pages inside a segment
MI_PAGE_LARGE, // larger blocks go into a single page spanning a whole segment
MI_PAGE_HUGE // huge blocks (>512kb) are put into a single page in a segment of the exact size (but still 2mb aligned)
MI_PAGE_LARGE, // larger blocks go into a page of just one block
MI_PAGE_HUGE, // huge blocks (>16mb) are put into a single page in a single segment.
} mi_page_kind_t;
typedef enum mi_segment_kind_e {
MI_SEGMENT_NORMAL, // MI_SEGMENT_SIZE size with pages inside.
MI_SEGMENT_HUGE, // > MI_LARGE_SIZE_MAX segment with just one huge page inside.
} mi_segment_kind_t;
typedef mi_page_t mi_slice_t;
// Segments are large allocated memory blocks (2mb on 64 bit) from
// the OS. Inside segments we allocated fixed size _pages_ that
// contain blocks.
typedef struct mi_segment_s {
struct mi_segment_s* next;
struct mi_segment_s* prev;
struct mi_segment_s* abandoned_next;
struct mi_segment_s* abandoned_next; // abandoned segment stack: `used == abandoned`
size_t abandoned; // abandoned pages (i.e. the original owning thread stopped) (`abandoned <= used`)
size_t used; // count of pages in use (`used <= capacity`)
size_t capacity; // count of available pages (`#free + used`)
size_t used; // count of pages in use
size_t segment_size;// for huge pages this may be different from `MI_SEGMENT_SIZE`
size_t segment_info_size; // space we are using from the first page for segment meta-data and possible guard pages.
uintptr_t cookie; // verify addresses in debug mode: `mi_ptr_cookie(segment) == segment->cookie`
size_t memid; // id for the os-level memory manager
bool all_committed;
// layout like this to optimize access in `mi_free`
size_t page_shift; // `1 << page_shift` == the page sizes == `page->block_size * page->reserved` (unless the first page, then `-segment_info_size`).
volatile uintptr_t thread_id; // unique id of the thread owning this segment
mi_page_kind_t page_kind; // kind of pages: small, large, or huge
mi_page_t pages[1]; // up to `MI_SMALL_PAGES_PER_SEGMENT` pages
mi_segment_kind_t kind;
uintptr_t thread_id;
size_t slice_count; // slices in this segment (at most MI_SLICES_PER_SEGMENT)
mi_slice_t slices[MI_SLICES_PER_SEGMENT];
} mi_segment_t;
@ -326,13 +334,13 @@ typedef struct mi_stats_s {
mi_stat_count_t commit_calls;
mi_stat_count_t threads;
mi_stat_count_t huge;
mi_stat_count_t giant;
mi_stat_count_t large;
mi_stat_count_t malloc;
mi_stat_count_t segments_cache;
mi_stat_counter_t page_no_retire;
mi_stat_counter_t searches;
mi_stat_counter_t huge_count;
mi_stat_counter_t giant_count;
mi_stat_counter_t large_count;
#if MI_STAT>1
mi_stat_count_t normal[MI_BIN_HUGE+1];
#endif
@ -367,11 +375,11 @@ typedef struct mi_segment_queue_s {
mi_segment_t* last;
} mi_segment_queue_t;
#define MI_SEGMENT_BIN_MAX (35) // 35 == mi_segment_bin(MI_SEGMENT_SIZE)
// Segments thread local data
typedef struct mi_segments_tld_s {
mi_segment_queue_t small_free; // queue of segments with free small pages
mi_segment_queue_t medium_free; // queue of segments with free medium pages
mi_page_queue_t pages[MI_SEGMENT_BIN_MAX+1]; // free pages inside segments
size_t count; // current number of segments;
size_t peak_count; // peak number of segments
size_t current_size; // current size of all segments

View File

@ -43,7 +43,7 @@ static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* heap, size_t size, size_t
if (p == NULL) return NULL;
// .. and align within the allocation
mi_page_set_has_aligned( _mi_ptr_page(p), true );
mi_page_set_has_aligned(_mi_ptr_page(p), true);
uintptr_t adjust = alignment - (((uintptr_t)p + offset) % alignment);
mi_assert_internal(adjust % sizeof(uintptr_t) == 0);
void* aligned_p = (adjust == alignment ? p : (void*)((uintptr_t)p + adjust));

View File

@ -226,7 +226,7 @@ void mi_free(void* p) mi_attr_noexcept
#endif
mi_page_t* const page = _mi_segment_page_of(segment, p);
#if (MI_STAT>1)
mi_heap_t* heap = mi_heap_get_default();
mi_heap_stat_decrease( heap, malloc, mi_usable_size(p));
@ -235,9 +235,9 @@ void mi_free(void* p) mi_attr_noexcept
}
// huge page stat is accounted for in `_mi_page_retire`
#endif
const uintptr_t tid = _mi_thread_id();
if (mi_likely(tid == page->flags)) { // if equal, the thread id matches and it is not a full page, nor has aligned blocks
uintptr_t tid = _mi_thread_id();
if (mi_likely(page->flags == tid)) {
// local, and not full or aligned
mi_block_t* block = (mi_block_t*)p;
mi_block_set_next(page, block, page->local_free);

View File

@ -150,7 +150,7 @@ static void mi_heap_collect_ex(mi_heap_t* heap, mi_collect_t collect)
// collect regions
if (collect >= FORCE && _mi_is_main_thread()) {
_mi_mem_collect(&heap->tld->stats);
// _mi_mem_collect(&heap->tld->stats);
}
}
@ -245,9 +245,9 @@ static bool _mi_heap_page_destroy(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_
_mi_page_use_delayed_free(page, MI_NEVER_DELAYED_FREE);
// stats
if (page->block_size > MI_LARGE_SIZE_MAX) {
if (page->block_size > MI_HUGE_SIZE_MAX) {
_mi_stat_decrease(&heap->tld->stats.giant,page->block_size);
if (page->block_size > MI_MEDIUM_SIZE_MAX) {
if (page->block_size <= MI_LARGE_SIZE_MAX) {
_mi_stat_decrease(&heap->tld->stats.large,page->block_size);
}
else {
_mi_stat_decrease(&heap->tld->stats.huge, page->block_size);
@ -255,7 +255,7 @@ static bool _mi_heap_page_destroy(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_
}
#if (MI_STAT>1)
size_t inuse = page->used - page->thread_freed;
if (page->block_size <= MI_LARGE_SIZE_MAX) {
if (page->block_size <= MI_MEDIUM_SIZE_MAX) {
mi_heap_stat_decrease(heap,normal[_mi_bin(page->block_size)], inuse);
}
mi_heap_stat_decrease(heap,malloc, page->block_size * inuse); // todo: off for aligned blocks...

View File

@ -21,7 +21,7 @@ const mi_page_t _mi_page_empty = {
NULL, 0, 0,
0, NULL, NULL, NULL
#if (MI_INTPTR_SIZE==8 && MI_SECURE==0)
, { NULL }
// , { NULL }
#endif
};
@ -43,8 +43,8 @@ const mi_page_t _mi_page_empty = {
QNULL( 10240), QNULL( 12288), QNULL( 14336), QNULL( 16384), QNULL( 20480), QNULL( 24576), QNULL( 28672), QNULL( 32768), /* 56 */ \
QNULL( 40960), QNULL( 49152), QNULL( 57344), QNULL( 65536), QNULL( 81920), QNULL( 98304), QNULL(114688), QNULL(131072), /* 64 */ \
QNULL(163840), QNULL(196608), QNULL(229376), QNULL(262144), QNULL(327680), QNULL(393216), QNULL(458752), QNULL(524288), /* 72 */ \
QNULL(MI_LARGE_WSIZE_MAX + 1 /* 655360, Huge queue */), \
QNULL(MI_LARGE_WSIZE_MAX + 2) /* Full queue */ }
QNULL(MI_MEDIUM_WSIZE_MAX + 1 /* 655360, Huge queue */), \
QNULL(MI_MEDIUM_WSIZE_MAX + 2) /* Full queue */ }
#define MI_STAT_COUNT_NULL() {0,0,0,0}
@ -91,14 +91,23 @@ const mi_heap_t _mi_heap_empty = {
mi_decl_thread mi_heap_t* _mi_heap_default = (mi_heap_t*)&_mi_heap_empty;
// Empty page queues for every bin
#define MI_SEGMENT_PAGE_QUEUES_EMPTY \
{ QNULL(0), \
QNULL( 1), QNULL( 2), QNULL( 3), QNULL( 4), QNULL( 5), QNULL( 6), QNULL( 7), QNULL( 8), /* 8 */ \
QNULL( 10), QNULL( 12), QNULL( 14), QNULL( 16), QNULL( 20), QNULL( 24), QNULL( 28), QNULL( 32), /* 16 */ \
QNULL( 40), QNULL( 48), QNULL( 56), QNULL( 64), QNULL( 80), QNULL( 96), QNULL( 112), QNULL( 128), /* 24 */ \
QNULL( 160), QNULL( 192), QNULL( 224), /* 27 */ }
#define tld_main_stats ((mi_stats_t*)((uint8_t*)&tld_main + offsetof(mi_tld_t,stats)))
static mi_tld_t tld_main = {
0,
&_mi_heap_main,
{ { NULL, NULL }, {NULL ,NULL}, 0, 0, 0, 0, 0, 0, NULL, tld_main_stats }, // segments
{ 0, tld_main_stats }, // os
{ MI_STATS_NULL } // stats
{ MI_SEGMENT_PAGE_QUEUES_EMPTY, 0, 0, 0, 0, 0, 0, NULL, tld_main_stats }, // segments
{ 0, tld_main_stats }, // os
{ MI_STATS_NULL } // stats
};
mi_heap_t _mi_heap_main = {

View File

@ -34,15 +34,15 @@ terms of the MIT license. A copy of the license can be found in the file
static inline bool mi_page_queue_is_huge(const mi_page_queue_t* pq) {
return (pq->block_size == (MI_LARGE_SIZE_MAX+sizeof(uintptr_t)));
return (pq->block_size == (MI_MEDIUM_SIZE_MAX+sizeof(uintptr_t)));
}
static inline bool mi_page_queue_is_full(const mi_page_queue_t* pq) {
return (pq->block_size == (MI_LARGE_SIZE_MAX+(2*sizeof(uintptr_t))));
return (pq->block_size == (MI_MEDIUM_SIZE_MAX+(2*sizeof(uintptr_t))));
}
static inline bool mi_page_queue_is_special(const mi_page_queue_t* pq) {
return (pq->block_size > MI_LARGE_SIZE_MAX);
return (pq->block_size > MI_MEDIUM_SIZE_MAX);
}
/* -----------------------------------------------------------
@ -116,7 +116,7 @@ extern inline uint8_t _mi_bin(size_t size) {
bin = (uint8_t)wsize;
}
#endif
else if (wsize > MI_LARGE_WSIZE_MAX) {
else if (wsize > MI_MEDIUM_WSIZE_MAX) {
bin = MI_BIN_HUGE;
}
else {
@ -147,7 +147,7 @@ size_t _mi_bin_size(uint8_t bin) {
// Good size for allocation
size_t mi_good_size(size_t size) mi_attr_noexcept {
if (size <= MI_LARGE_SIZE_MAX) {
if (size <= MI_MEDIUM_SIZE_MAX) {
return _mi_bin_size(_mi_bin(size));
}
else {
@ -245,7 +245,7 @@ static bool mi_page_queue_is_empty(mi_page_queue_t* queue) {
static void mi_page_queue_remove(mi_page_queue_t* queue, mi_page_t* page) {
mi_assert_internal(page != NULL);
mi_assert_expensive(mi_page_queue_contains(queue, page));
mi_assert_internal(page->block_size == queue->block_size || (page->block_size > MI_LARGE_SIZE_MAX && mi_page_queue_is_huge(queue)) || (mi_page_is_in_full(page) && mi_page_queue_is_full(queue)));
mi_assert_internal(page->block_size == queue->block_size || (page->block_size > MI_MEDIUM_SIZE_MAX && mi_page_queue_is_huge(queue)) || (mi_page_is_in_full(page) && mi_page_queue_is_full(queue)));
if (page->prev != NULL) page->prev->next = page->next;
if (page->next != NULL) page->next->prev = page->prev;
if (page == queue->last) queue->last = page->prev;
@ -268,7 +268,7 @@ static void mi_page_queue_push(mi_heap_t* heap, mi_page_queue_t* queue, mi_page_
mi_assert_internal(page->heap == NULL);
mi_assert_internal(!mi_page_queue_contains(queue, page));
mi_assert_internal(page->block_size == queue->block_size ||
(page->block_size > MI_LARGE_SIZE_MAX && mi_page_queue_is_huge(queue)) ||
(page->block_size > MI_MEDIUM_SIZE_MAX && mi_page_queue_is_huge(queue)) ||
(mi_page_is_in_full(page) && mi_page_queue_is_full(queue)));
mi_page_set_in_full(page, mi_page_queue_is_full(queue));
@ -297,8 +297,8 @@ static void mi_page_queue_enqueue_from(mi_page_queue_t* to, mi_page_queue_t* fro
mi_assert_internal((page->block_size == to->block_size && page->block_size == from->block_size) ||
(page->block_size == to->block_size && mi_page_queue_is_full(from)) ||
(page->block_size == from->block_size && mi_page_queue_is_full(to)) ||
(page->block_size > MI_LARGE_SIZE_MAX && mi_page_queue_is_huge(to)) ||
(page->block_size > MI_LARGE_SIZE_MAX && mi_page_queue_is_full(to)));
(page->block_size > MI_MEDIUM_SIZE_MAX && mi_page_queue_is_huge(to)) ||
(page->block_size > MI_MEDIUM_SIZE_MAX && mi_page_queue_is_full(to)));
if (page->prev != NULL) page->prev->next = page->next;
if (page->next != NULL) page->next->prev = page->prev;

View File

@ -74,7 +74,7 @@ static bool mi_page_is_valid_init(mi_page_t* page) {
mi_segment_t* segment = _mi_page_segment(page);
uint8_t* start = _mi_page_start(segment,page,NULL);
mi_assert_internal(start == _mi_segment_page_start(segment,page,page->block_size,NULL));
mi_assert_internal(start == _mi_segment_page_start(segment,page,NULL));
mi_assert_internal(segment->thread_id==0 || segment->thread_id == mi_page_thread_id(page));
//mi_assert_internal(start + page->capacity*page->block_size == page->top);
@ -102,7 +102,7 @@ bool _mi_page_is_valid(mi_page_t* page) {
mi_assert_internal(!_mi_process_is_initialized || segment->thread_id == page->heap->thread_id);
mi_page_queue_t* pq = mi_page_queue_of(page);
mi_assert_internal(mi_page_queue_contains(pq, page));
mi_assert_internal(pq->block_size==page->block_size || page->block_size > MI_LARGE_SIZE_MAX || mi_page_is_in_full(page));
mi_assert_internal(pq->block_size==page->block_size || page->block_size > MI_MEDIUM_SIZE_MAX || mi_page_is_in_full(page));
mi_assert_internal(mi_heap_contains_queue(page->heap,pq));
}
return true;
@ -356,9 +356,9 @@ void _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force) {
mi_page_set_has_aligned(page, false);
// account for huge pages here
if (page->block_size > MI_LARGE_SIZE_MAX) {
if (page->block_size > MI_HUGE_SIZE_MAX) {
_mi_stat_decrease(&page->heap->tld->stats.giant, page->block_size);
if (page->block_size > MI_MEDIUM_SIZE_MAX) {
if (page->block_size <= MI_LARGE_SIZE_MAX) {
_mi_stat_decrease(&page->heap->tld->stats.large, page->block_size);
}
else {
_mi_stat_decrease(&page->heap->tld->stats.huge, page->block_size);
@ -554,7 +554,7 @@ static void mi_page_init(mi_heap_t* heap, mi_page_t* page, size_t block_size, mi
mi_assert_internal(block_size > 0);
// set fields
size_t page_size;
_mi_segment_page_start(segment, page, block_size, &page_size);
_mi_segment_page_start(segment, page, &page_size);
page->block_size = block_size;
mi_assert_internal(page_size / block_size < (1L<<16));
page->reserved = (uint16_t)(page_size / block_size);
@ -702,7 +702,7 @@ void mi_register_deferred_free(mi_deferred_free_fun* fn) mi_attr_noexcept {
----------------------------------------------------------- */
// A huge page is allocated directly without being in a queue
static mi_page_t* mi_huge_page_alloc(mi_heap_t* heap, size_t size) {
static mi_page_t* mi_large_page_alloc(mi_heap_t* heap, size_t size) {
size_t block_size = _mi_wsize_from_size(size) * sizeof(uintptr_t);
mi_assert_internal(_mi_bin(block_size) == MI_BIN_HUGE);
mi_page_queue_t* pq = mi_page_queue(heap,block_size);
@ -711,9 +711,9 @@ static mi_page_t* mi_huge_page_alloc(mi_heap_t* heap, size_t size) {
if (page != NULL) {
mi_assert_internal(mi_page_immediate_available(page));
mi_assert_internal(page->block_size == block_size);
if (page->block_size > MI_HUGE_SIZE_MAX) {
_mi_stat_increase(&heap->tld->stats.giant, block_size);
_mi_stat_counter_increase(&heap->tld->stats.giant_count, 1);
if (page->block_size <= MI_LARGE_SIZE_MAX) {
_mi_stat_increase(&heap->tld->stats.large, block_size);
_mi_stat_counter_increase(&heap->tld->stats.large_count, 1);
}
else {
_mi_stat_increase(&heap->tld->stats.huge, block_size);
@ -744,12 +744,12 @@ void* _mi_malloc_generic(mi_heap_t* heap, size_t size) mi_attr_noexcept
// huge allocation?
mi_page_t* page;
if (mi_unlikely(size > MI_LARGE_SIZE_MAX)) {
if (mi_unlikely(size > MI_MEDIUM_SIZE_MAX)) {
if (mi_unlikely(size >= (SIZE_MAX - MI_MAX_ALIGN_SIZE))) {
page = NULL;
}
else {
page = mi_huge_page_alloc(heap,size);
page = mi_large_page_alloc(heap,size);
}
}
else {

File diff suppressed because it is too large Load Diff

View File

@ -106,11 +106,11 @@ static void mi_stats_add(mi_stats_t* stats, const mi_stats_t* src) {
mi_stat_add(&stats->malloc, &src->malloc, 1);
mi_stat_add(&stats->segments_cache, &src->segments_cache, 1);
mi_stat_add(&stats->huge, &src->huge, 1);
mi_stat_add(&stats->giant, &src->giant, 1);
mi_stat_add(&stats->large, &src->large, 1);
mi_stat_counter_add(&stats->page_no_retire, &src->page_no_retire, 1);
mi_stat_counter_add(&stats->searches, &src->searches, 1);
mi_stat_counter_add(&stats->huge_count, &src->huge_count, 1);
mi_stat_counter_add(&stats->giant_count, &src->giant_count, 1);
mi_stat_counter_add(&stats->large_count, &src->large_count, 1);
#if MI_STAT>1
for (size_t i = 0; i <= MI_BIN_HUGE; i++) {
if (src->normal[i].allocated > 0 || src->normal[i].freed > 0) {
@ -232,11 +232,11 @@ static void _mi_stats_print(mi_stats_t* stats, double secs, FILE* out) mi_attr_n
mi_stats_print_bins(&normal, stats->normal, MI_BIN_HUGE, "normal",out);
mi_stat_print(&normal, "normal", 1, out);
mi_stat_print(&stats->huge, "huge", (stats->huge_count.count == 0 ? 1 : -(stats->huge.allocated / stats->huge_count.count)), out);
mi_stat_print(&stats->giant, "giant", (stats->giant_count.count == 0 ? 1 : -(stats->giant.allocated / stats->giant_count.count)), out);
mi_stat_print(&stats->large, "giant", (stats->large_count.count == 0 ? 1 : -(stats->large.allocated / stats->large_count.count)), out);
mi_stat_count_t total = { 0,0,0,0 };
mi_stat_add(&total, &normal, 1);
mi_stat_add(&total, &stats->huge, 1);
mi_stat_add(&total, &stats->giant, 1);
mi_stat_add(&total, &stats->large, 1);
mi_stat_print(&total, "total", 1, out);
_mi_fprintf(out, "malloc requested: ");
mi_print_amount(stats->malloc.allocated, 1, out);

View File

@ -6,8 +6,168 @@
#include <mimalloc.h>
#include <mimalloc-override.h> // redefines malloc etc.
#include <stdint.h>
#include <stdbool.h>
#define MI_INTPTR_SIZE 8
#define MI_LARGE_WSIZE_MAX (4*1024*1024 / MI_INTPTR_SIZE)
#define MI_BIN_HUGE 100
//#define MI_ALIGN2W
// Bit scan reverse: return the index of the highest bit.
static inline uint8_t mi_bsr32(uint32_t x);
#if defined(_MSC_VER)
#include <windows.h>
#include <intrin.h>
static inline uint8_t mi_bsr32(uint32_t x) {
uint32_t idx;
_BitScanReverse((DWORD*)&idx, x);
return idx;
}
#elif defined(__GNUC__) || defined(__clang__)
static inline uint8_t mi_bsr32(uint32_t x) {
return (31 - __builtin_clz(x));
}
#else
static inline uint8_t mi_bsr32(uint32_t x) {
// de Bruijn multiplication, see <http://supertech.csail.mit.edu/papers/debruijn.pdf>
static const uint8_t debruijn[32] = {
31, 0, 22, 1, 28, 23, 18, 2, 29, 26, 24, 10, 19, 7, 3, 12,
30, 21, 27, 17, 25, 9, 6, 11, 20, 16, 8, 5, 15, 4, 14, 13,
};
x |= x >> 1;
x |= x >> 2;
x |= x >> 4;
x |= x >> 8;
x |= x >> 16;
x++;
return debruijn[(x*0x076be629) >> 27];
}
#endif
// Bit scan reverse: return the index of the highest bit.
uint8_t _mi_bsr(uintptr_t x) {
if (x == 0) return 0;
#if MI_INTPTR_SIZE==8
uint32_t hi = (x >> 32);
return (hi == 0 ? mi_bsr32((uint32_t)x) : 32 + mi_bsr32(hi));
#elif MI_INTPTR_SIZE==4
return mi_bsr32(x);
#else
# error "define bsr for non-32 or 64-bit platforms"
#endif
}
static inline size_t _mi_wsize_from_size(size_t size) {
return (size + sizeof(uintptr_t) - 1) / sizeof(uintptr_t);
}
// Return the bin for a given field size.
// Returns MI_BIN_HUGE if the size is too large.
// We use `wsize` for the size in "machine word sizes",
// i.e. byte size == `wsize*sizeof(void*)`.
extern inline uint8_t _mi_bin8(size_t size) {
size_t wsize = _mi_wsize_from_size(size);
uint8_t bin;
if (wsize <= 1) {
bin = 1;
}
#if defined(MI_ALIGN4W)
else if (wsize <= 4) {
bin = (uint8_t)((wsize+1)&~1); // round to double word sizes
}
#elif defined(MI_ALIGN2W)
else if (wsize <= 8) {
bin = (uint8_t)((wsize+1)&~1); // round to double word sizes
}
#else
else if (wsize <= 8) {
bin = (uint8_t)wsize;
}
#endif
else if (wsize > MI_LARGE_WSIZE_MAX) {
bin = MI_BIN_HUGE;
}
else {
#if defined(MI_ALIGN4W)
if (wsize <= 16) { wsize = (wsize+3)&~3; } // round to 4x word sizes
#endif
wsize--;
// find the highest bit
uint8_t b = mi_bsr32((uint32_t)wsize);
// and use the top 3 bits to determine the bin (~12.5% worst internal fragmentation).
// - adjust with 3 because we use do not round the first 8 sizes
// which each get an exact bin
bin = ((b << 2) + (uint8_t)((wsize >> (b - 2)) & 0x03)) - 3;
}
return bin;
}
extern inline uint8_t _mi_bin4(size_t size) {
size_t wsize = _mi_wsize_from_size(size);
uint8_t bin;
if (wsize <= 1) {
bin = 1;
}
#if defined(MI_ALIGN4W)
else if (wsize <= 4) {
bin = (uint8_t)((wsize+1)&~1); // round to double word sizes
}
#elif defined(MI_ALIGN2W)
else if (wsize <= 8) {
bin = (uint8_t)((wsize+1)&~1); // round to double word sizes
}
#else
else if (wsize <= 8) {
bin = (uint8_t)wsize;
}
#endif
else if (wsize > MI_LARGE_WSIZE_MAX) {
bin = MI_BIN_HUGE;
}
else {
uint8_t b = mi_bsr32((uint32_t)wsize);
bin = ((b << 1) + (uint8_t)((wsize >> (b - 1)) & 0x01)) + 3;
}
return bin;
}
size_t _mi_binx4(size_t bsize) {
if (bsize==0) return 0;
uint8_t b = mi_bsr32((uint32_t)bsize);
if (b <= 1) return bsize;
size_t bin = ((b << 1) | (bsize >> (b - 1))&0x01);
return bin;
}
size_t _mi_binx8(size_t bsize) {
if (bsize==0) return 0;
uint8_t b = mi_bsr32((uint32_t)bsize);
if (b <= 2) return bsize;
size_t bin = ((b << 2) | (bsize >> (b - 2))&0x03) - 5;
return bin;
}
void mi_bins() {
//printf(" QNULL(1), /* 0 */ \\\n ");
size_t last_bin = 1;
for (size_t bsize = 0; bsize < 8*1024; bsize++) {
size_t size = bsize * 64 * 1024;
size_t bin = _mi_binx8(bsize);
if (bin != last_bin) {
printf("bsize: %6zd, size: %6zd, bin: %6zd\n", bsize, size, bin);
//printf("QNULL(%6zd), ", wsize);
//if (last_bin%8 == 0) printf("/* %i */ \\\n ", last_bin);
last_bin = bin;
}
}
}
int main() {
mi_version();
mi_bins();
void* p1 = malloc(78);
void* p2 = malloc(24);
free(p1);
@ -25,7 +185,7 @@ int main() {
//p1 = mi_malloc(32);
//free(p1);
//p2 = malloc(32);
//mi_free(p2);
//mi_free(p2);
mi_stats_print(NULL);
return 0;
}