Line | Count | Source |
1 | | /* |
2 | | * Copyright © 2020, VideoLAN and dav1d authors |
3 | | * Copyright © 2020, Two Orioles, LLC |
4 | | * All rights reserved. |
5 | | * |
6 | | * Redistribution and use in source and binary forms, with or without |
7 | | * modification, are permitted provided that the following conditions are met: |
8 | | * |
9 | | * 1. Redistributions of source code must retain the above copyright notice, this |
10 | | * list of conditions and the following disclaimer. |
11 | | * |
12 | | * 2. Redistributions in binary form must reproduce the above copyright notice, |
13 | | * this list of conditions and the following disclaimer in the documentation |
14 | | * and/or other materials provided with the distribution. |
15 | | * |
16 | | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND |
17 | | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED |
18 | | * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
19 | | * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR |
20 | | * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES |
21 | | * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
22 | | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND |
23 | | * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
24 | | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
25 | | * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
26 | | */ |
27 | | |
28 | | #include "config.h" |
29 | | |
30 | | #include <stdint.h> |
31 | | |
32 | | #include "src/internal.h" |
33 | | |
34 | | #if TRACK_HEAP_ALLOCATIONS |
35 | | #include <stdio.h> |
36 | | |
37 | | #include "src/log.h" |
38 | | |
39 | | #define DEFAULT_ALIGN 16 |
40 | | |
41 | | typedef struct { |
42 | | size_t sz; |
43 | | unsigned align; |
44 | | enum AllocationType type; |
45 | | } Dav1dAllocationData; |
46 | | |
47 | | typedef struct { |
48 | | size_t curr_sz; |
49 | | size_t peak_sz; |
50 | | unsigned num_allocs; |
51 | | unsigned num_reuses; |
52 | | } AllocStats; |
53 | | |
54 | | static AllocStats tracked_allocs[N_ALLOC_TYPES]; |
55 | | static size_t curr_total_sz; |
56 | | static size_t peak_total_sz; |
57 | | static pthread_mutex_t track_alloc_mutex = PTHREAD_MUTEX_INITIALIZER; |
58 | | |
59 | | static void *track_alloc(const enum AllocationType type, char *ptr, |
60 | | const size_t sz, const size_t align) |
61 | | { |
62 | | assert(align >= sizeof(Dav1dAllocationData)); |
63 | | if (ptr) { |
64 | | ptr += align; |
65 | | Dav1dAllocationData *const d = &((Dav1dAllocationData*)ptr)[-1]; |
66 | | AllocStats *const s = &tracked_allocs[type]; |
67 | | |
68 | | d->sz = sz; |
69 | | d->align = (unsigned)align; |
70 | | d->type = type; |
71 | | |
72 | | pthread_mutex_lock(&track_alloc_mutex); |
73 | | s->num_allocs++; |
74 | | s->curr_sz += sz; |
75 | | if (s->curr_sz > s->peak_sz) |
76 | | s->peak_sz = s->curr_sz; |
77 | | |
78 | | curr_total_sz += sz; |
79 | | if (curr_total_sz > peak_total_sz) |
80 | | peak_total_sz = curr_total_sz; |
81 | | pthread_mutex_unlock(&track_alloc_mutex); |
82 | | } |
83 | | return ptr; |
84 | | } |
85 | | |
86 | | static void *track_free(char *const ptr) { |
87 | | const Dav1dAllocationData *const d = &((Dav1dAllocationData*)ptr)[-1]; |
88 | | const size_t sz = d->sz; |
89 | | |
90 | | pthread_mutex_lock(&track_alloc_mutex); |
91 | | tracked_allocs[d->type].curr_sz -= sz; |
92 | | curr_total_sz -= sz; |
93 | | pthread_mutex_unlock(&track_alloc_mutex); |
94 | | |
95 | | return ptr - d->align; |
96 | | } |
97 | | |
98 | | static void dav1d_track_reuse(const enum AllocationType type) { |
99 | | pthread_mutex_lock(&track_alloc_mutex); |
100 | | tracked_allocs[type].num_reuses++; |
101 | | pthread_mutex_unlock(&track_alloc_mutex); |
102 | | } |
103 | | |
104 | | void *dav1d_malloc(const enum AllocationType type, const size_t sz) { |
105 | | void *const ptr = malloc(sz + DEFAULT_ALIGN); |
106 | | return track_alloc(type, ptr, sz, DEFAULT_ALIGN); |
107 | | } |
108 | | |
109 | | void *dav1d_alloc_aligned(const enum AllocationType type, |
110 | | const size_t sz, const size_t align) |
111 | | { |
112 | | void *const ptr = dav1d_alloc_aligned_internal(sz + align, align); |
113 | | return track_alloc(type, ptr, sz, align); |
114 | | } |
115 | | |
116 | | void *dav1d_realloc(const enum AllocationType type, |
117 | | void *ptr, const size_t sz) |
118 | | { |
119 | | if (!ptr) |
120 | | return dav1d_malloc(type, sz); |
121 | | ptr = realloc((char*)ptr - DEFAULT_ALIGN, sz + DEFAULT_ALIGN); |
122 | | if (ptr) |
123 | | ptr = track_free((char*)ptr + DEFAULT_ALIGN); |
124 | | return track_alloc(type, ptr, sz, DEFAULT_ALIGN); |
125 | | } |
126 | | |
127 | | void dav1d_free(void *ptr) { |
128 | | if (ptr) |
129 | | free(track_free(ptr)); |
130 | | } |
131 | | |
132 | | void dav1d_free_aligned(void *ptr) { |
133 | | if (ptr) { |
134 | | dav1d_free_aligned_internal(track_free(ptr)); |
135 | | } |
136 | | } |
137 | | |
138 | | static COLD int cmp_stats(const void *const a, const void *const b) { |
139 | | const size_t a_sz = ((const AllocStats*)a)->peak_sz; |
140 | | const size_t b_sz = ((const AllocStats*)b)->peak_sz; |
141 | | return a_sz < b_sz ? -1 : a_sz > b_sz; |
142 | | } |
143 | | |
144 | | /* Insert spaces as thousands separators for better readability */ |
145 | | static COLD int format_tsep(char *const s, const size_t n, const size_t value) { |
146 | | if (value < 1000) |
147 | | return snprintf(s, n, "%u", (unsigned)value); |
148 | | |
149 | | const int len = format_tsep(s, n, value / 1000); |
150 | | assert((size_t)len < n); |
151 | | return len + snprintf(s + len, n - len, " %03u", (unsigned)(value % 1000)); |
152 | | } |
153 | | |
154 | | COLD void dav1d_log_alloc_stats(Dav1dContext *const c) { |
155 | | static const char *const type_names[N_ALLOC_TYPES] = { |
156 | | [ALLOC_BLOCK ] = "Block data", |
157 | | [ALLOC_CDEF ] = "CDEF line buffers", |
158 | | [ALLOC_CDF ] = "CDF contexts", |
159 | | [ALLOC_COEF ] = "Coefficient data", |
160 | | [ALLOC_COMMON_CTX] = "Common context data", |
161 | | [ALLOC_DAV1DDATA ] = "Dav1dData", |
162 | | [ALLOC_IPRED ] = "Intra pred edges", |
163 | | [ALLOC_LF ] = "Loopfilter data", |
164 | | [ALLOC_LR ] = "Looprestoration data", |
165 | | [ALLOC_OBU_HDR ] = "OBU headers", |
166 | | [ALLOC_OBU_META ] = "OBU metadata", |
167 | | [ALLOC_PAL ] = "Palette data", |
168 | | [ALLOC_PIC ] = "Picture buffers", |
169 | | [ALLOC_PIC_CTX ] = "Picture context data", |
170 | | [ALLOC_REFMVS ] = "Reference mv data", |
171 | | [ALLOC_SEGMAP ] = "Segmentation maps", |
172 | | [ALLOC_THREAD_CTX] = "Thread context data", |
173 | | [ALLOC_TILE ] = "Tile data", |
174 | | }; |
175 | | |
176 | | struct { |
177 | | AllocStats stats; |
178 | | enum AllocationType type; |
179 | | } data[N_ALLOC_TYPES]; |
180 | | unsigned total_allocs = 0; |
181 | | unsigned total_reuses = 0; |
182 | | |
183 | | pthread_mutex_lock(&track_alloc_mutex); |
184 | | for (int i = 0; i < N_ALLOC_TYPES; i++) { |
185 | | AllocStats *const s = &data[i].stats; |
186 | | *s = tracked_allocs[i]; |
187 | | data[i].type = i; |
188 | | total_allocs += s->num_allocs; |
189 | | total_reuses += s->num_reuses; |
190 | | } |
191 | | size_t total_sz = peak_total_sz; |
192 | | pthread_mutex_unlock(&track_alloc_mutex); |
193 | | |
194 | | /* Sort types by memory usage */ |
195 | | qsort(&data, N_ALLOC_TYPES, sizeof(*data), cmp_stats); |
196 | | |
197 | | const double inv_total_share = 100.0 / total_sz; |
198 | | char total_sz_buf[32]; |
199 | | const int sz_len = 4 + format_tsep(total_sz_buf, sizeof(total_sz_buf), total_sz); |
200 | | |
201 | | dav1d_log(c, "\n Type Allocs Reuses Share Peak size\n" |
202 | | "---------------------------------------------------------------------\n"); |
203 | | for (int i = N_ALLOC_TYPES - 1; i >= 0; i--) { |
204 | | const AllocStats *const s = &data[i].stats; |
205 | | if (s->num_allocs) { |
206 | | const double share = s->peak_sz * inv_total_share; |
207 | | char sz_buf[32]; |
208 | | format_tsep(sz_buf, sizeof(sz_buf), s->peak_sz); |
209 | | dav1d_log(c, " %-20s%10u%10u%8.1f%%%*s\n", type_names[data[i].type], |
210 | | s->num_allocs, s->num_reuses, share, sz_len, sz_buf); |
211 | | } |
212 | | } |
213 | | dav1d_log(c, "---------------------------------------------------------------------\n" |
214 | | "%31u%10u %s\n", |
215 | | total_allocs, total_reuses, total_sz_buf); |
216 | | } |
217 | | #endif /* TRACK_HEAP_ALLOCATIONS */ |
218 | | |
219 | 0 | static COLD void mem_pool_destroy(Dav1dMemPool *const pool) { |
220 | 0 | pthread_mutex_destroy(&pool->lock); |
221 | 0 | dav1d_free(pool); |
222 | 0 | } |
223 | | |
224 | 0 | void dav1d_mem_pool_push(Dav1dMemPool *const pool, void *const ptr) { |
225 | 0 | pthread_mutex_lock(&pool->lock); |
226 | 0 | Dav1dMemPoolBuffer *const buf = (Dav1dMemPoolBuffer*)((uintptr_t)ptr - 64); |
227 | 0 | const int ref_cnt = --pool->ref_cnt; |
228 | 0 | if (!pool->end) { |
229 | 0 | buf->next = pool->buf; |
230 | 0 | pool->buf = buf; |
231 | 0 | pthread_mutex_unlock(&pool->lock); |
232 | 0 | assert(ref_cnt > 0); |
233 | 0 | } else { |
234 | 0 | pthread_mutex_unlock(&pool->lock); |
235 | 0 | dav1d_free_aligned(buf); |
236 | 0 | if (!ref_cnt) mem_pool_destroy(pool); |
237 | 0 | } |
238 | 0 | } |
239 | | |
240 | 0 | void *dav1d_mem_pool_pop(Dav1dMemPool *const pool, const size_t size) { |
241 | 0 | pthread_mutex_lock(&pool->lock); |
242 | 0 | Dav1dMemPoolBuffer *buf = pool->buf; |
243 | 0 | pool->ref_cnt++; |
244 | |
|
245 | 0 | if (buf) { |
246 | 0 | pool->buf = buf->next; |
247 | 0 | pthread_mutex_unlock(&pool->lock); |
248 | 0 | if (buf->size != size) { |
249 | | /* Reallocate if the size has changed */ |
250 | 0 | dav1d_free_aligned(buf); |
251 | 0 | goto alloc; |
252 | 0 | } |
253 | | #if TRACK_HEAP_ALLOCATIONS |
254 | | dav1d_track_reuse(pool->type); |
255 | | #endif |
256 | 0 | } else { |
257 | 0 | pthread_mutex_unlock(&pool->lock); |
258 | 0 | alloc: |
259 | 0 | buf = dav1d_alloc_aligned(pool->type, size + 64, 64); |
260 | 0 | if (!buf) { |
261 | 0 | pthread_mutex_lock(&pool->lock); |
262 | 0 | const int ref_cnt = --pool->ref_cnt; |
263 | 0 | pthread_mutex_unlock(&pool->lock); |
264 | 0 | if (!ref_cnt) mem_pool_destroy(pool); |
265 | 0 | return NULL; |
266 | 0 | } |
267 | 0 | buf->size = size; |
268 | 0 | } |
269 | | |
270 | 0 | return (void*)((uintptr_t)buf + 64); |
271 | 0 | } |
272 | | |
273 | | COLD int dav1d_mem_pool_init(const enum AllocationType type, |
274 | | Dav1dMemPool **const ppool) |
275 | 0 | { |
276 | 0 | Dav1dMemPool *const pool = dav1d_malloc(ALLOC_COMMON_CTX, |
277 | 0 | sizeof(Dav1dMemPool)); |
278 | 0 | if (pool) { |
279 | 0 | if (!pthread_mutex_init(&pool->lock, NULL)) { |
280 | 0 | pool->buf = NULL; |
281 | 0 | pool->ref_cnt = 1; |
282 | 0 | pool->end = 0; |
283 | | #if TRACK_HEAP_ALLOCATIONS |
284 | | pool->type = type; |
285 | | #endif |
286 | 0 | *ppool = pool; |
287 | 0 | return 0; |
288 | 0 | } |
289 | 0 | dav1d_free(pool); |
290 | 0 | } |
291 | 0 | *ppool = NULL; |
292 | 0 | return DAV1D_ERR(ENOMEM); |
293 | 0 | } |
294 | | |
295 | 0 | COLD void dav1d_mem_pool_end(Dav1dMemPool *const pool) { |
296 | 0 | if (pool) { |
297 | 0 | pthread_mutex_lock(&pool->lock); |
298 | 0 | Dav1dMemPoolBuffer *buf = pool->buf; |
299 | 0 | const int ref_cnt = --pool->ref_cnt; |
300 | 0 | pool->buf = NULL; |
301 | 0 | pool->end = 1; |
302 | 0 | pthread_mutex_unlock(&pool->lock); |
303 | |
|
304 | 0 | while (buf) { |
305 | 0 | void *const ptr = buf; |
306 | 0 | buf = buf->next; |
307 | 0 | dav1d_free_aligned(ptr); |
308 | 0 | } |
309 | 0 | if (!ref_cnt) mem_pool_destroy(pool); |
310 | 0 | } |
311 | 0 | } |