|
| 1 | +/* ---------------------------------------------------------------------------- |
| 2 | +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen |
| 3 | +This is free software; you can redistribute it and/or modify it under the |
| 4 | +terms of the MIT license. A copy of the license can be found in the file |
| 5 | +"LICENSE" at the root of this distribution. |
| 6 | +-----------------------------------------------------------------------------*/ |
| 7 | + |
| 8 | +#include "mimalloc.h" |
| 9 | +#include "mimalloc/internal.h" |
| 10 | +#include "mimalloc/prim.h" // mi_prim_get_default_heap |
| 11 | + |
| 12 | +#include <string.h> // memset |
| 13 | + |
| 14 | +// ------------------------------------------------------ |
| 15 | +// Aligned Allocation |
| 16 | +// ------------------------------------------------------ |
| 17 | + |
| 18 | +// Fallback primitive aligned allocation -- split out for better codegen |
| 19 | +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 |
| 20 | +{ |
| 21 | + mi_assert_internal(size <= PTRDIFF_MAX); |
| 22 | + mi_assert_internal(alignment != 0 && _mi_is_power_of_two(alignment)); |
| 23 | + |
| 24 | + const uintptr_t align_mask = alignment - 1; // for any x, `(x & align_mask) == (x % alignment)` |
| 25 | + const size_t padsize = size + MI_PADDING_SIZE; |
| 26 | + |
| 27 | + // use regular allocation if it is guaranteed to fit the alignment constraints |
| 28 | + if (offset==0 && alignment<=padsize && padsize<=MI_MAX_ALIGN_GUARANTEE && (padsize&align_mask)==0) { |
| 29 | + void* p = _mi_heap_malloc_zero(heap, size, zero); |
| 30 | + mi_assert_internal(p == NULL || ((uintptr_t)p % alignment) == 0); |
| 31 | + return p; |
| 32 | + } |
| 33 | + |
| 34 | + void* p; |
| 35 | + size_t oversize; |
| 36 | + if mi_unlikely(alignment > MI_ALIGNMENT_MAX) { |
| 37 | + // use OS allocation for very large alignment and allocate inside a huge page (dedicated segment with 1 page) |
| 38 | + // This can support alignments >= MI_SEGMENT_SIZE by ensuring the object can be aligned at a point in the |
| 39 | + // 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) |
| 40 | + if mi_unlikely(offset != 0) { |
| 41 | + // todo: cannot support offset alignment for very large alignments yet |
| 42 | + #if MI_DEBUG > 0 |
| 43 | + _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); |
| 44 | + #endif |
| 45 | + return NULL; |
| 46 | + } |
| 47 | + oversize = (size <= MI_SMALL_SIZE_MAX ? MI_SMALL_SIZE_MAX + 1 /* ensure we use generic malloc path */ : size); |
| 48 | + 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 |
| 49 | + // zero afterwards as only the area from the aligned_p may be committed! |
| 50 | + if (p == NULL) return NULL; |
| 51 | + } |
| 52 | + else { |
| 53 | + // otherwise over-allocate |
| 54 | + oversize = size + alignment - 1; |
| 55 | + p = _mi_heap_malloc_zero(heap, oversize, zero); |
| 56 | + if (p == NULL) return NULL; |
| 57 | + } |
| 58 | + |
| 59 | + // .. and align within the allocation |
| 60 | + const uintptr_t poffset = ((uintptr_t)p + offset) & align_mask; |
| 61 | + const uintptr_t adjust = (poffset == 0 ? 0 : alignment - poffset); |
| 62 | + mi_assert_internal(adjust < alignment); |
| 63 | + void* aligned_p = (void*)((uintptr_t)p + adjust); |
| 64 | + if (aligned_p != p) { |
| 65 | + mi_page_t* page = _mi_ptr_page(p); |
| 66 | + mi_page_set_has_aligned(page, true); |
| 67 | + _mi_padding_shrink(page, (mi_block_t*)p, adjust + size); |
| 68 | + } |
| 69 | + // todo: expand padding if overallocated ? |
| 70 | + |
| 71 | + mi_assert_internal(mi_page_usable_block_size(_mi_ptr_page(p)) >= adjust + size); |
| 72 | + mi_assert_internal(p == _mi_page_ptr_unalign(_mi_ptr_segment(aligned_p), _mi_ptr_page(aligned_p), aligned_p)); |
| 73 | + mi_assert_internal(((uintptr_t)aligned_p + offset) % alignment == 0); |
| 74 | + mi_assert_internal(mi_usable_size(aligned_p)>=size); |
| 75 | + mi_assert_internal(mi_usable_size(p) == mi_usable_size(aligned_p)+adjust); |
| 76 | + |
| 77 | + // now zero the block if needed |
| 78 | + if (alignment > MI_ALIGNMENT_MAX) { |
| 79 | + // for the tracker, on huge aligned allocations only from the start of the large block is defined |
| 80 | + mi_track_mem_undefined(aligned_p, size); |
| 81 | + if (zero) { |
| 82 | + _mi_memzero_aligned(aligned_p, mi_usable_size(aligned_p)); |
| 83 | + } |
| 84 | + } |
| 85 | + |
| 86 | + if (p != aligned_p) { |
| 87 | + mi_track_align(p,aligned_p,adjust,mi_usable_size(aligned_p)); |
| 88 | + } |
| 89 | + return aligned_p; |
| 90 | +} |
| 91 | + |
| 92 | +// Primitive aligned allocation |
| 93 | +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 |
| 94 | +{ |
| 95 | + // note: we don't require `size > offset`, we just guarantee that the address at offset is aligned regardless of the allocated size. |
| 96 | + if mi_unlikely(alignment == 0 || !_mi_is_power_of_two(alignment)) { // require power-of-two (see <https://en.cppreference.com/w/c/memory/aligned_alloc>) |
| 97 | + #if MI_DEBUG > 0 |
| 98 | + _mi_error_message(EOVERFLOW, "aligned allocation requires the alignment to be a power-of-two (size %zu, alignment %zu)\n", size, alignment); |
| 99 | + #endif |
| 100 | + return NULL; |
| 101 | + } |
| 102 | + |
| 103 | + if mi_unlikely(size > PTRDIFF_MAX) { // we don't allocate more than PTRDIFF_MAX (see <https://sourceware.org/ml/libc-announce/2019/msg00001.html>) |
| 104 | + #if MI_DEBUG > 0 |
| 105 | + _mi_error_message(EOVERFLOW, "aligned allocation request is too large (size %zu, alignment %zu)\n", size, alignment); |
| 106 | + #endif |
| 107 | + return NULL; |
| 108 | + } |
| 109 | + const uintptr_t align_mask = alignment-1; // for any x, `(x & align_mask) == (x % alignment)` |
| 110 | + const size_t padsize = size + MI_PADDING_SIZE; // note: cannot overflow due to earlier size > PTRDIFF_MAX check |
| 111 | + |
| 112 | + // try first if there happens to be a small block available with just the right alignment |
| 113 | + if mi_likely(padsize <= MI_SMALL_SIZE_MAX && alignment <= padsize) { |
| 114 | + mi_page_t* page = _mi_heap_get_free_small_page(heap, padsize); |
| 115 | + const bool is_aligned = (((uintptr_t)page->free+offset) & align_mask)==0; |
| 116 | + if mi_likely(page->free != NULL && is_aligned) |
| 117 | + { |
| 118 | + #if MI_STAT>1 |
| 119 | + mi_heap_stat_increase(heap, malloc, size); |
| 120 | + #endif |
| 121 | + void* p = _mi_page_malloc(heap, page, padsize, zero); // TODO: inline _mi_page_malloc |
| 122 | + mi_assert_internal(p != NULL); |
| 123 | + mi_assert_internal(((uintptr_t)p + offset) % alignment == 0); |
| 124 | + mi_track_malloc(p,size,zero); |
| 125 | + return p; |
| 126 | + } |
| 127 | + } |
| 128 | + // fallback |
| 129 | + return mi_heap_malloc_zero_aligned_at_fallback(heap, size, alignment, offset, zero); |
| 130 | +} |
| 131 | + |
| 132 | + |
| 133 | +// ------------------------------------------------------ |
| 134 | +// Optimized mi_heap_malloc_aligned / mi_malloc_aligned |
| 135 | +// ------------------------------------------------------ |
| 136 | + |
| 137 | +mi_decl_nodiscard mi_decl_restrict void* mi_heap_malloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { |
| 138 | + return mi_heap_malloc_zero_aligned_at(heap, size, alignment, offset, false); |
| 139 | +} |
| 140 | + |
| 141 | +mi_decl_nodiscard mi_decl_restrict void* mi_heap_malloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept { |
| 142 | + if mi_unlikely(alignment == 0 || !_mi_is_power_of_two(alignment)) return NULL; |
| 143 | + #if !MI_PADDING |
| 144 | + // without padding, any small sized allocation is naturally aligned (see also `_mi_segment_page_start`) |
| 145 | + if mi_likely(_mi_is_power_of_two(size) && size >= alignment && size <= MI_SMALL_SIZE_MAX) |
| 146 | + #else |
| 147 | + // with padding, we can only guarantee this for fixed alignments |
| 148 | + if mi_likely((alignment == sizeof(void*) || (alignment == MI_MAX_ALIGN_SIZE && size > (MI_MAX_ALIGN_SIZE/2))) |
| 149 | + && size <= MI_SMALL_SIZE_MAX) |
| 150 | + #endif |
| 151 | + { |
| 152 | + // fast path for common alignment and size |
| 153 | + return mi_heap_malloc_small(heap, size); |
| 154 | + } |
| 155 | + else { |
| 156 | + return mi_heap_malloc_aligned_at(heap, size, alignment, 0); |
| 157 | + } |
| 158 | +} |
| 159 | + |
| 160 | +// ensure a definition is emitted |
| 161 | +#if defined(__cplusplus) |
| 162 | +static void* _mi_heap_malloc_aligned = (void*)&mi_heap_malloc_aligned; |
| 163 | +#endif |
| 164 | + |
| 165 | +// ------------------------------------------------------ |
| 166 | +// Aligned Allocation |
| 167 | +// ------------------------------------------------------ |
| 168 | + |
| 169 | +mi_decl_nodiscard mi_decl_restrict void* mi_heap_zalloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { |
| 170 | + return mi_heap_malloc_zero_aligned_at(heap, size, alignment, offset, true); |
| 171 | +} |
| 172 | + |
| 173 | +mi_decl_nodiscard mi_decl_restrict void* mi_heap_zalloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept { |
| 174 | + return mi_heap_zalloc_aligned_at(heap, size, alignment, 0); |
| 175 | +} |
| 176 | + |
| 177 | +mi_decl_nodiscard mi_decl_restrict void* mi_heap_calloc_aligned_at(mi_heap_t* heap, size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { |
| 178 | + size_t total; |
| 179 | + if (mi_count_size_overflow(count, size, &total)) return NULL; |
| 180 | + return mi_heap_zalloc_aligned_at(heap, total, alignment, offset); |
| 181 | +} |
| 182 | + |
| 183 | +mi_decl_nodiscard mi_decl_restrict void* mi_heap_calloc_aligned(mi_heap_t* heap, size_t count, size_t size, size_t alignment) mi_attr_noexcept { |
| 184 | + return mi_heap_calloc_aligned_at(heap,count,size,alignment,0); |
| 185 | +} |
| 186 | + |
| 187 | +mi_decl_nodiscard mi_decl_restrict void* mi_malloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept { |
| 188 | + return mi_heap_malloc_aligned_at(mi_prim_get_default_heap(), size, alignment, offset); |
| 189 | +} |
| 190 | + |
| 191 | +mi_decl_nodiscard mi_decl_restrict void* mi_malloc_aligned(size_t size, size_t alignment) mi_attr_noexcept { |
| 192 | + return mi_heap_malloc_aligned(mi_prim_get_default_heap(), size, alignment); |
| 193 | +} |
| 194 | + |
| 195 | +mi_decl_nodiscard mi_decl_restrict void* mi_zalloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept { |
| 196 | + return mi_heap_zalloc_aligned_at(mi_prim_get_default_heap(), size, alignment, offset); |
| 197 | +} |
| 198 | + |
| 199 | +mi_decl_nodiscard mi_decl_restrict void* mi_zalloc_aligned(size_t size, size_t alignment) mi_attr_noexcept { |
| 200 | + return mi_heap_zalloc_aligned(mi_prim_get_default_heap(), size, alignment); |
| 201 | +} |
| 202 | + |
| 203 | +mi_decl_nodiscard mi_decl_restrict void* mi_calloc_aligned_at(size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { |
| 204 | + return mi_heap_calloc_aligned_at(mi_prim_get_default_heap(), count, size, alignment, offset); |
| 205 | +} |
| 206 | + |
| 207 | +mi_decl_nodiscard mi_decl_restrict void* mi_calloc_aligned(size_t count, size_t size, size_t alignment) mi_attr_noexcept { |
| 208 | + return mi_heap_calloc_aligned(mi_prim_get_default_heap(), count, size, alignment); |
| 209 | +} |
| 210 | + |
| 211 | + |
| 212 | +// ------------------------------------------------------ |
| 213 | +// Aligned re-allocation |
| 214 | +// ------------------------------------------------------ |
| 215 | + |
| 216 | +static void* mi_heap_realloc_zero_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset, bool zero) mi_attr_noexcept { |
| 217 | + mi_assert(alignment > 0); |
| 218 | + if (alignment <= sizeof(uintptr_t)) return _mi_heap_realloc_zero(heap,p,newsize,zero); |
| 219 | + if (p == NULL) return mi_heap_malloc_zero_aligned_at(heap,newsize,alignment,offset,zero); |
| 220 | + size_t size = mi_usable_size(p); |
| 221 | + if (newsize <= size && newsize >= (size - (size / 2)) |
| 222 | + && (((uintptr_t)p + offset) % alignment) == 0) { |
| 223 | + return p; // reallocation still fits, is aligned and not more than 50% waste |
| 224 | + } |
| 225 | + else { |
| 226 | + // note: we don't zero allocate upfront so we only zero initialize the expanded part |
| 227 | + void* newp = mi_heap_malloc_aligned_at(heap,newsize,alignment,offset); |
| 228 | + if (newp != NULL) { |
| 229 | + if (zero && newsize > size) { |
| 230 | + // also set last word in the previous allocation to zero to ensure any padding is zero-initialized |
| 231 | + size_t start = (size >= sizeof(intptr_t) ? size - sizeof(intptr_t) : 0); |
| 232 | + _mi_memzero((uint8_t*)newp + start, newsize - start); |
| 233 | + } |
| 234 | + _mi_memcpy_aligned(newp, p, (newsize > size ? size : newsize)); |
| 235 | + mi_free(p); // only free if successful |
| 236 | + } |
| 237 | + return newp; |
| 238 | + } |
| 239 | +} |
| 240 | + |
| 241 | +static void* mi_heap_realloc_zero_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, bool zero) mi_attr_noexcept { |
| 242 | + mi_assert(alignment > 0); |
| 243 | + if (alignment <= sizeof(uintptr_t)) return _mi_heap_realloc_zero(heap,p,newsize,zero); |
| 244 | + size_t offset = ((uintptr_t)p % alignment); // use offset of previous allocation (p can be NULL) |
| 245 | + return mi_heap_realloc_zero_aligned_at(heap,p,newsize,alignment,offset,zero); |
| 246 | +} |
| 247 | + |
| 248 | +mi_decl_nodiscard void* mi_heap_realloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { |
| 249 | + return mi_heap_realloc_zero_aligned_at(heap,p,newsize,alignment,offset,false); |
| 250 | +} |
| 251 | + |
| 252 | +mi_decl_nodiscard void* mi_heap_realloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept { |
| 253 | + return mi_heap_realloc_zero_aligned(heap,p,newsize,alignment,false); |
| 254 | +} |
| 255 | + |
| 256 | +mi_decl_nodiscard void* mi_heap_rezalloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { |
| 257 | + return mi_heap_realloc_zero_aligned_at(heap, p, newsize, alignment, offset, true); |
| 258 | +} |
| 259 | + |
| 260 | +mi_decl_nodiscard void* mi_heap_rezalloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept { |
| 261 | + return mi_heap_realloc_zero_aligned(heap, p, newsize, alignment, true); |
| 262 | +} |
| 263 | + |
| 264 | +mi_decl_nodiscard void* mi_heap_recalloc_aligned_at(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { |
| 265 | + size_t total; |
| 266 | + if (mi_count_size_overflow(newcount, size, &total)) return NULL; |
| 267 | + return mi_heap_rezalloc_aligned_at(heap, p, total, alignment, offset); |
| 268 | +} |
| 269 | + |
| 270 | +mi_decl_nodiscard void* mi_heap_recalloc_aligned(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept { |
| 271 | + size_t total; |
| 272 | + if (mi_count_size_overflow(newcount, size, &total)) return NULL; |
| 273 | + return mi_heap_rezalloc_aligned(heap, p, total, alignment); |
| 274 | +} |
| 275 | + |
| 276 | +mi_decl_nodiscard void* mi_realloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { |
| 277 | + return mi_heap_realloc_aligned_at(mi_prim_get_default_heap(), p, newsize, alignment, offset); |
| 278 | +} |
| 279 | + |
| 280 | +mi_decl_nodiscard void* mi_realloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept { |
| 281 | + return mi_heap_realloc_aligned(mi_prim_get_default_heap(), p, newsize, alignment); |
| 282 | +} |
| 283 | + |
| 284 | +mi_decl_nodiscard void* mi_rezalloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { |
| 285 | + return mi_heap_rezalloc_aligned_at(mi_prim_get_default_heap(), p, newsize, alignment, offset); |
| 286 | +} |
| 287 | + |
| 288 | +mi_decl_nodiscard void* mi_rezalloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept { |
| 289 | + return mi_heap_rezalloc_aligned(mi_prim_get_default_heap(), p, newsize, alignment); |
| 290 | +} |
| 291 | + |
| 292 | +mi_decl_nodiscard void* mi_recalloc_aligned_at(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { |
| 293 | + return mi_heap_recalloc_aligned_at(mi_prim_get_default_heap(), p, newcount, size, alignment, offset); |
| 294 | +} |
| 295 | + |
| 296 | +mi_decl_nodiscard void* mi_recalloc_aligned(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept { |
| 297 | + return mi_heap_recalloc_aligned(mi_prim_get_default_heap(), p, newcount, size, alignment); |
| 298 | +} |
0 commit comments