b24e7fce |
1 | /* Alloc.c -- Memory allocation functions |
2 | 2018-04-27 : Igor Pavlov : Public domain */ |
3 | |
4 | #include "Precomp.h" |
5 | |
6 | #include <stdio.h> |
7 | |
8 | #ifdef _WIN32 |
9 | #include <windows.h> |
10 | #endif |
11 | #include <stdlib.h> |
12 | |
13 | #include "Alloc.h" |
14 | |
15 | /* #define _SZ_ALLOC_DEBUG */ |
16 | |
17 | /* use _SZ_ALLOC_DEBUG to debug alloc/free operations */ |
18 | #ifdef _SZ_ALLOC_DEBUG |
19 | |
20 | #include <stdio.h> |
21 | int g_allocCount = 0; |
22 | int g_allocCountMid = 0; |
23 | int g_allocCountBig = 0; |
24 | |
25 | |
26 | #define CONVERT_INT_TO_STR(charType, tempSize) \ |
27 | unsigned char temp[tempSize]; unsigned i = 0; \ |
28 | while (val >= 10) { temp[i++] = (unsigned char)('0' + (unsigned)(val % 10)); val /= 10; } \ |
29 | *s++ = (charType)('0' + (unsigned)val); \ |
30 | while (i != 0) { i--; *s++ = temp[i]; } \ |
31 | *s = 0; |
32 | |
33 | static void ConvertUInt64ToString(UInt64 val, char *s) |
34 | { |
35 | CONVERT_INT_TO_STR(char, 24); |
36 | } |
37 | |
38 | #define GET_HEX_CHAR(t) ((char)(((t < 10) ? ('0' + t) : ('A' + (t - 10))))) |
39 | |
40 | static void ConvertUInt64ToHex(UInt64 val, char *s) |
41 | { |
42 | UInt64 v = val; |
43 | unsigned i; |
44 | for (i = 1;; i++) |
45 | { |
46 | v >>= 4; |
47 | if (v == 0) |
48 | break; |
49 | } |
50 | s[i] = 0; |
51 | do |
52 | { |
53 | unsigned t = (unsigned)(val & 0xF); |
54 | val >>= 4; |
55 | s[--i] = GET_HEX_CHAR(t); |
56 | } |
57 | while (i); |
58 | } |
59 | |
60 | #define DEBUG_OUT_STREAM stderr |
61 | |
62 | static void Print(const char *s) |
63 | { |
64 | fputs(s, DEBUG_OUT_STREAM); |
65 | } |
66 | |
67 | static void PrintAligned(const char *s, size_t align) |
68 | { |
69 | size_t len = strlen(s); |
70 | for(;;) |
71 | { |
72 | fputc(' ', DEBUG_OUT_STREAM); |
73 | if (len >= align) |
74 | break; |
75 | ++len; |
76 | } |
77 | Print(s); |
78 | } |
79 | |
80 | static void PrintLn() |
81 | { |
82 | Print("\n"); |
83 | } |
84 | |
85 | static void PrintHex(UInt64 v, size_t align) |
86 | { |
87 | char s[32]; |
88 | ConvertUInt64ToHex(v, s); |
89 | PrintAligned(s, align); |
90 | } |
91 | |
92 | static void PrintDec(UInt64 v, size_t align) |
93 | { |
94 | char s[32]; |
95 | ConvertUInt64ToString(v, s); |
96 | PrintAligned(s, align); |
97 | } |
98 | |
99 | static void PrintAddr(void *p) |
100 | { |
101 | PrintHex((UInt64)(size_t)(ptrdiff_t)p, 12); |
102 | } |
103 | |
104 | |
105 | #define PRINT_ALLOC(name, cnt, size, ptr) \ |
106 | Print(name " "); \ |
107 | PrintDec(cnt++, 10); \ |
108 | PrintHex(size, 10); \ |
109 | PrintAddr(ptr); \ |
110 | PrintLn(); |
111 | |
112 | #define PRINT_FREE(name, cnt, ptr) if (ptr) { \ |
113 | Print(name " "); \ |
114 | PrintDec(--cnt, 10); \ |
115 | PrintAddr(ptr); \ |
116 | PrintLn(); } |
117 | |
118 | #else |
119 | |
120 | #define PRINT_ALLOC(name, cnt, size, ptr) |
121 | #define PRINT_FREE(name, cnt, ptr) |
122 | #define Print(s) |
123 | #define PrintLn() |
124 | #define PrintHex(v, align) |
125 | #define PrintDec(v, align) |
126 | #define PrintAddr(p) |
127 | |
128 | #endif |
129 | |
130 | |
131 | |
132 | void *MyAlloc(size_t size) |
133 | { |
134 | if (size == 0) |
135 | return NULL; |
136 | #ifdef _SZ_ALLOC_DEBUG |
137 | { |
138 | void *p = malloc(size); |
139 | PRINT_ALLOC("Alloc ", g_allocCount, size, p); |
140 | return p; |
141 | } |
142 | #else |
143 | return malloc(size); |
144 | #endif |
145 | } |
146 | |
147 | void MyFree(void *address) |
148 | { |
149 | PRINT_FREE("Free ", g_allocCount, address); |
150 | |
151 | free(address); |
152 | } |
153 | |
154 | #ifdef _WIN32 |
155 | |
156 | void *MidAlloc(size_t size) |
157 | { |
158 | if (size == 0) |
159 | return NULL; |
160 | |
161 | PRINT_ALLOC("Alloc-Mid", g_allocCountMid, size, NULL); |
162 | |
163 | return VirtualAlloc(NULL, size, MEM_COMMIT, PAGE_READWRITE); |
164 | } |
165 | |
166 | void MidFree(void *address) |
167 | { |
168 | PRINT_FREE("Free-Mid", g_allocCountMid, address); |
169 | |
170 | if (!address) |
171 | return; |
172 | VirtualFree(address, 0, MEM_RELEASE); |
173 | } |
174 | |
175 | #ifndef MEM_LARGE_PAGES |
176 | #undef _7ZIP_LARGE_PAGES |
177 | #endif |
178 | |
179 | #ifdef _7ZIP_LARGE_PAGES |
180 | SIZE_T g_LargePageSize = 0; |
181 | typedef SIZE_T (WINAPI *GetLargePageMinimumP)(); |
182 | #endif |
183 | |
184 | void SetLargePageSize() |
185 | { |
186 | #ifdef _7ZIP_LARGE_PAGES |
187 | SIZE_T size; |
188 | GetLargePageMinimumP largePageMinimum = (GetLargePageMinimumP) |
189 | GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")), "GetLargePageMinimum"); |
190 | if (!largePageMinimum) |
191 | return; |
192 | size = largePageMinimum(); |
193 | if (size == 0 || (size & (size - 1)) != 0) |
194 | return; |
195 | g_LargePageSize = size; |
196 | #endif |
197 | } |
198 | |
199 | |
200 | void *BigAlloc(size_t size) |
201 | { |
202 | if (size == 0) |
203 | return NULL; |
204 | |
205 | PRINT_ALLOC("Alloc-Big", g_allocCountBig, size, NULL); |
206 | |
207 | #ifdef _7ZIP_LARGE_PAGES |
208 | { |
209 | SIZE_T ps = g_LargePageSize; |
210 | if (ps != 0 && ps <= (1 << 30) && size > (ps / 2)) |
211 | { |
212 | size_t size2; |
213 | ps--; |
214 | size2 = (size + ps) & ~ps; |
215 | if (size2 >= size) |
216 | { |
217 | void *res = VirtualAlloc(NULL, size2, MEM_COMMIT | MEM_LARGE_PAGES, PAGE_READWRITE); |
218 | if (res) |
219 | return res; |
220 | } |
221 | } |
222 | } |
223 | #endif |
224 | |
225 | return VirtualAlloc(NULL, size, MEM_COMMIT, PAGE_READWRITE); |
226 | } |
227 | |
228 | void BigFree(void *address) |
229 | { |
230 | PRINT_FREE("Free-Big", g_allocCountBig, address); |
231 | |
232 | if (!address) |
233 | return; |
234 | VirtualFree(address, 0, MEM_RELEASE); |
235 | } |
236 | |
237 | #endif |
238 | |
239 | |
240 | static void *SzAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p); return MyAlloc(size); } |
241 | static void SzFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p); MyFree(address); } |
242 | const ISzAlloc g_Alloc = { SzAlloc, SzFree }; |
243 | |
244 | static void *SzMidAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p); return MidAlloc(size); } |
245 | static void SzMidFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p); MidFree(address); } |
246 | const ISzAlloc g_MidAlloc = { SzMidAlloc, SzMidFree }; |
247 | |
248 | static void *SzBigAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p); return BigAlloc(size); } |
249 | static void SzBigFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p); BigFree(address); } |
250 | const ISzAlloc g_BigAlloc = { SzBigAlloc, SzBigFree }; |
251 | |
252 | |
253 | /* |
254 | uintptr_t : <stdint.h> C99 (optional) |
255 | : unsupported in VS6 |
256 | */ |
257 | |
258 | #ifdef _WIN32 |
259 | typedef UINT_PTR UIntPtr; |
260 | #else |
261 | /* |
262 | typedef uintptr_t UIntPtr; |
263 | */ |
264 | typedef ptrdiff_t UIntPtr; |
265 | #endif |
266 | |
267 | |
268 | #define ADJUST_ALLOC_SIZE 0 |
269 | /* |
270 | #define ADJUST_ALLOC_SIZE (sizeof(void *) - 1) |
271 | */ |
272 | /* |
273 | Use (ADJUST_ALLOC_SIZE = (sizeof(void *) - 1)), if |
274 | MyAlloc() can return address that is NOT multiple of sizeof(void *). |
275 | */ |
276 | |
277 | |
278 | /* |
279 | #define MY_ALIGN_PTR_DOWN(p, align) ((void *)((char *)(p) - ((size_t)(UIntPtr)(p) & ((align) - 1)))) |
280 | */ |
281 | #define MY_ALIGN_PTR_DOWN(p, align) ((void *)((((UIntPtr)(p)) & ~((UIntPtr)(align) - 1)))) |
282 | |
283 | #define MY_ALIGN_PTR_UP_PLUS(p, align) MY_ALIGN_PTR_DOWN(((char *)(p) + (align) + ADJUST_ALLOC_SIZE), align) |
284 | |
285 | |
286 | #if (_POSIX_C_SOURCE >= 200112L) && !defined(_WIN32) |
287 | #define USE_posix_memalign |
288 | #endif |
289 | |
290 | /* |
291 | This posix_memalign() is for test purposes only. |
292 | We also need special Free() function instead of free(), |
293 | if this posix_memalign() is used. |
294 | */ |
295 | |
296 | /* |
297 | static int posix_memalign(void **ptr, size_t align, size_t size) |
298 | { |
299 | size_t newSize = size + align; |
300 | void *p; |
301 | void *pAligned; |
302 | *ptr = NULL; |
303 | if (newSize < size) |
304 | return 12; // ENOMEM |
305 | p = MyAlloc(newSize); |
306 | if (!p) |
307 | return 12; // ENOMEM |
308 | pAligned = MY_ALIGN_PTR_UP_PLUS(p, align); |
309 | ((void **)pAligned)[-1] = p; |
310 | *ptr = pAligned; |
311 | return 0; |
312 | } |
313 | */ |
314 | |
315 | /* |
316 | ALLOC_ALIGN_SIZE >= sizeof(void *) |
317 | ALLOC_ALIGN_SIZE >= cache_line_size |
318 | */ |
319 | |
320 | #define ALLOC_ALIGN_SIZE ((size_t)1 << 7) |
321 | |
322 | static void *SzAlignedAlloc(ISzAllocPtr pp, size_t size) |
323 | { |
324 | #ifndef USE_posix_memalign |
325 | |
326 | void *p; |
327 | void *pAligned; |
328 | size_t newSize; |
329 | UNUSED_VAR(pp); |
330 | |
331 | /* also we can allocate additional dummy ALLOC_ALIGN_SIZE bytes after aligned |
332 | block to prevent cache line sharing with another allocated blocks */ |
333 | |
334 | newSize = size + ALLOC_ALIGN_SIZE * 1 + ADJUST_ALLOC_SIZE; |
335 | if (newSize < size) |
336 | return NULL; |
337 | |
338 | p = MyAlloc(newSize); |
339 | |
340 | if (!p) |
341 | return NULL; |
342 | pAligned = MY_ALIGN_PTR_UP_PLUS(p, ALLOC_ALIGN_SIZE); |
343 | |
344 | Print(" size="); PrintHex(size, 8); |
345 | Print(" a_size="); PrintHex(newSize, 8); |
346 | Print(" ptr="); PrintAddr(p); |
347 | Print(" a_ptr="); PrintAddr(pAligned); |
348 | PrintLn(); |
349 | |
350 | ((void **)pAligned)[-1] = p; |
351 | |
352 | return pAligned; |
353 | |
354 | #else |
355 | |
356 | void *p; |
357 | UNUSED_VAR(pp); |
358 | if (posix_memalign(&p, ALLOC_ALIGN_SIZE, size)) |
359 | return NULL; |
360 | |
361 | Print(" posix_memalign="); PrintAddr(p); |
362 | PrintLn(); |
363 | |
364 | return p; |
365 | |
366 | #endif |
367 | } |
368 | |
369 | |
370 | static void SzAlignedFree(ISzAllocPtr pp, void *address) |
371 | { |
372 | UNUSED_VAR(pp); |
373 | #ifndef USE_posix_memalign |
374 | if (address) |
375 | MyFree(((void **)address)[-1]); |
376 | #else |
377 | free(address); |
378 | #endif |
379 | } |
380 | |
381 | |
382 | const ISzAlloc g_AlignedAlloc = { SzAlignedAlloc, SzAlignedFree }; |
383 | |
384 | |
385 | |
386 | #define MY_ALIGN_PTR_DOWN_1(p) MY_ALIGN_PTR_DOWN(p, sizeof(void *)) |
387 | |
388 | /* we align ptr to support cases where CAlignOffsetAlloc::offset is not multiply of sizeof(void *) */ |
389 | #define REAL_BLOCK_PTR_VAR(p) ((void **)MY_ALIGN_PTR_DOWN_1(p))[-1] |
390 | /* |
391 | #define REAL_BLOCK_PTR_VAR(p) ((void **)(p))[-1] |
392 | */ |
393 | |
394 | static void *AlignOffsetAlloc_Alloc(ISzAllocPtr pp, size_t size) |
395 | { |
396 | CAlignOffsetAlloc *p = CONTAINER_FROM_VTBL(pp, CAlignOffsetAlloc, vt); |
397 | void *adr; |
398 | void *pAligned; |
399 | size_t newSize; |
400 | size_t extra; |
401 | size_t alignSize = (size_t)1 << p->numAlignBits; |
402 | |
403 | if (alignSize < sizeof(void *)) |
404 | alignSize = sizeof(void *); |
405 | |
406 | if (p->offset >= alignSize) |
407 | return NULL; |
408 | |
409 | /* also we can allocate additional dummy ALLOC_ALIGN_SIZE bytes after aligned |
410 | block to prevent cache line sharing with another allocated blocks */ |
411 | extra = p->offset & (sizeof(void *) - 1); |
412 | newSize = size + alignSize + extra + ADJUST_ALLOC_SIZE; |
413 | if (newSize < size) |
414 | return NULL; |
415 | |
416 | adr = ISzAlloc_Alloc(p->baseAlloc, newSize); |
417 | |
418 | if (!adr) |
419 | return NULL; |
420 | |
421 | pAligned = (char *)MY_ALIGN_PTR_DOWN((char *)adr + |
422 | alignSize - p->offset + extra + ADJUST_ALLOC_SIZE, alignSize) + p->offset; |
423 | |
424 | PrintLn(); |
425 | Print("- Aligned: "); |
426 | Print(" size="); PrintHex(size, 8); |
427 | Print(" a_size="); PrintHex(newSize, 8); |
428 | Print(" ptr="); PrintAddr(adr); |
429 | Print(" a_ptr="); PrintAddr(pAligned); |
430 | PrintLn(); |
431 | |
432 | REAL_BLOCK_PTR_VAR(pAligned) = adr; |
433 | |
434 | return pAligned; |
435 | } |
436 | |
437 | |
438 | static void AlignOffsetAlloc_Free(ISzAllocPtr pp, void *address) |
439 | { |
440 | if (address) |
441 | { |
442 | CAlignOffsetAlloc *p = CONTAINER_FROM_VTBL(pp, CAlignOffsetAlloc, vt); |
443 | PrintLn(); |
444 | Print("- Aligned Free: "); |
445 | PrintLn(); |
446 | ISzAlloc_Free(p->baseAlloc, REAL_BLOCK_PTR_VAR(address)); |
447 | } |
448 | } |
449 | |
450 | |
451 | void AlignOffsetAlloc_CreateVTable(CAlignOffsetAlloc *p) |
452 | { |
453 | p->vt.Alloc = AlignOffsetAlloc_Alloc; |
454 | p->vt.Free = AlignOffsetAlloc_Free; |
455 | } |