• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3  *
4  * This source code is subject to the terms of the BSD 2 Clause License and
5  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6  * was not distributed with this source code in the LICENSE file, you can
7  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8  * Media Patent License 1.0 was not distributed with this source code in the
9  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10  */
11 
12 #include <assert.h>
13 
14 #include "aom_mem/aom_mem.h"
15 #include "aom_ports/mem.h"
16 #include "aom_scale/yv12config.h"
17 #include "av1/common/enums.h"
18 
19 /****************************************************************************
20  *  Exports
21  ****************************************************************************/
22 
23 /****************************************************************************
24  *
25  ****************************************************************************/
26 #define yv12_align_addr(addr, align) \
27   (void *)(((size_t)(addr) + ((align)-1)) & (size_t) - (align))
28 
29 // TODO(jkoleszar): Maybe replace this with struct aom_image
30 
aom_free_frame_buffer(YV12_BUFFER_CONFIG * ybf)31 int aom_free_frame_buffer(YV12_BUFFER_CONFIG *ybf) {
32   if (ybf) {
33     if (ybf->buffer_alloc_sz > 0) {
34       aom_free(ybf->buffer_alloc);
35     }
36     if (ybf->y_buffer_8bit) aom_free(ybf->y_buffer_8bit);
37 
38     /* buffer_alloc isn't accessed by most functions.  Rather y_buffer,
39       u_buffer and v_buffer point to buffer_alloc and are used.  Clear out
40       all of this so that a freed pointer isn't inadvertently used */
41     memset(ybf, 0, sizeof(YV12_BUFFER_CONFIG));
42   } else {
43     return -1;
44   }
45 
46   return 0;
47 }
48 
realloc_frame_buffer_aligned(YV12_BUFFER_CONFIG * ybf,int width,int height,int ss_x,int ss_y,int use_highbitdepth,int border,int byte_alignment,aom_codec_frame_buffer_t * fb,aom_get_frame_buffer_cb_fn_t cb,void * cb_priv,const int y_stride,const uint64_t yplane_size,const uint64_t uvplane_size,const int aligned_width,const int aligned_height,const int uv_width,const int uv_height,const int uv_stride,const int uv_border_w,const int uv_border_h)49 static int realloc_frame_buffer_aligned(
50     YV12_BUFFER_CONFIG *ybf, int width, int height, int ss_x, int ss_y,
51     int use_highbitdepth, int border, int byte_alignment,
52     aom_codec_frame_buffer_t *fb, aom_get_frame_buffer_cb_fn_t cb,
53     void *cb_priv, const int y_stride, const uint64_t yplane_size,
54     const uint64_t uvplane_size, const int aligned_width,
55     const int aligned_height, const int uv_width, const int uv_height,
56     const int uv_stride, const int uv_border_w, const int uv_border_h) {
57   if (ybf) {
58     const int aom_byte_align = (byte_alignment == 0) ? 1 : byte_alignment;
59     const uint64_t frame_size =
60         (1 + use_highbitdepth) * (yplane_size + 2 * uvplane_size);
61 
62     uint8_t *buf = NULL;
63 
64 #if defined AOM_MAX_ALLOCABLE_MEMORY
65     // The size of ybf->buffer_alloc.
66     uint64_t alloc_size = frame_size;
67     // The size of ybf->y_buffer_8bit.
68     if (use_highbitdepth) alloc_size += yplane_size;
69     // The decoder may allocate REF_FRAMES frame buffers in the frame buffer
70     // pool. Bound the total amount of allocated memory as if these REF_FRAMES
71     // frame buffers were allocated in a single allocation.
72     if (alloc_size > AOM_MAX_ALLOCABLE_MEMORY / REF_FRAMES) return -1;
73 #endif
74 
75     if (cb != NULL) {
76       const int align_addr_extra_size = 31;
77       const uint64_t external_frame_size = frame_size + align_addr_extra_size;
78 
79       assert(fb != NULL);
80 
81       if (external_frame_size != (size_t)external_frame_size) return -1;
82 
83       // Allocation to hold larger frame, or first allocation.
84       if (cb(cb_priv, (size_t)external_frame_size, fb) < 0) return -1;
85 
86       if (fb->data == NULL || fb->size < external_frame_size) return -1;
87 
88       ybf->buffer_alloc = (uint8_t *)yv12_align_addr(fb->data, 32);
89 
90 #if defined(__has_feature)
91 #if __has_feature(memory_sanitizer)
92       // This memset is needed for fixing the issue of using uninitialized
93       // value in msan test. It will cause a perf loss, so only do this for
94       // msan test.
95       memset(ybf->buffer_alloc, 0, (size_t)frame_size);
96 #endif
97 #endif
98     } else if (frame_size > ybf->buffer_alloc_sz) {
99       // Allocation to hold larger frame, or first allocation.
100       aom_free(ybf->buffer_alloc);
101       ybf->buffer_alloc = NULL;
102       ybf->buffer_alloc_sz = 0;
103 
104       if (frame_size != (size_t)frame_size) return -1;
105 
106       ybf->buffer_alloc = (uint8_t *)aom_memalign(32, (size_t)frame_size);
107       if (!ybf->buffer_alloc) return -1;
108 
109       ybf->buffer_alloc_sz = (size_t)frame_size;
110 
111       // This memset is needed for fixing valgrind error from C loop filter
112       // due to access uninitialized memory in frame border. It could be
113       // removed if border is totally removed.
114       memset(ybf->buffer_alloc, 0, ybf->buffer_alloc_sz);
115     }
116 
117     ybf->y_crop_width = width;
118     ybf->y_crop_height = height;
119     ybf->y_width = aligned_width;
120     ybf->y_height = aligned_height;
121     ybf->y_stride = y_stride;
122 
123     ybf->uv_crop_width = (width + ss_x) >> ss_x;
124     ybf->uv_crop_height = (height + ss_y) >> ss_y;
125     ybf->uv_width = uv_width;
126     ybf->uv_height = uv_height;
127     ybf->uv_stride = uv_stride;
128 
129     ybf->border = border;
130     ybf->frame_size = (size_t)frame_size;
131     ybf->subsampling_x = ss_x;
132     ybf->subsampling_y = ss_y;
133 
134     buf = ybf->buffer_alloc;
135     if (use_highbitdepth) {
136       // Store uint16 addresses when using 16bit framebuffers
137       buf = CONVERT_TO_BYTEPTR(ybf->buffer_alloc);
138       ybf->flags = YV12_FLAG_HIGHBITDEPTH;
139     } else {
140       ybf->flags = 0;
141     }
142 
143     ybf->y_buffer = (uint8_t *)yv12_align_addr(
144         buf + (border * y_stride) + border, aom_byte_align);
145     ybf->u_buffer = (uint8_t *)yv12_align_addr(
146         buf + yplane_size + (uv_border_h * uv_stride) + uv_border_w,
147         aom_byte_align);
148     ybf->v_buffer =
149         (uint8_t *)yv12_align_addr(buf + yplane_size + uvplane_size +
150                                        (uv_border_h * uv_stride) + uv_border_w,
151                                    aom_byte_align);
152 
153     ybf->use_external_reference_buffers = 0;
154 
155     if (use_highbitdepth) {
156       if (ybf->y_buffer_8bit) aom_free(ybf->y_buffer_8bit);
157       ybf->y_buffer_8bit = (uint8_t *)aom_memalign(32, (size_t)yplane_size);
158       if (!ybf->y_buffer_8bit) return -1;
159     } else {
160       if (ybf->y_buffer_8bit) {
161         aom_free(ybf->y_buffer_8bit);
162         ybf->y_buffer_8bit = NULL;
163         ybf->buf_8bit_valid = 0;
164       }
165     }
166 
167     ybf->corrupted = 0; /* assume not corrupted by errors */
168     return 0;
169   }
170   return -2;
171 }
172 
calc_stride_and_planesize(const int ss_x,const int ss_y,const int aligned_width,const int aligned_height,const int border,const int byte_alignment,int * y_stride,int * uv_stride,uint64_t * yplane_size,uint64_t * uvplane_size,const int uv_height)173 static int calc_stride_and_planesize(const int ss_x, const int ss_y,
174                                      const int aligned_width,
175                                      const int aligned_height, const int border,
176                                      const int byte_alignment, int *y_stride,
177                                      int *uv_stride, uint64_t *yplane_size,
178                                      uint64_t *uvplane_size,
179                                      const int uv_height) {
180   /* Only support allocating buffers that have a border that's a multiple
181    * of 32. The border restriction is required to get 16-byte alignment of
182    * the start of the chroma rows without introducing an arbitrary gap
183    * between planes, which would break the semantics of things like
184    * aom_img_set_rect(). */
185   if (border & 0x1f) return -3;
186   *y_stride = ((aligned_width + 2 * border) + 31) & ~31;
187   *yplane_size =
188       (aligned_height + 2 * border) * (uint64_t)(*y_stride) + byte_alignment;
189 
190   *uv_stride = *y_stride >> ss_x;
191   *uvplane_size = (uv_height + 2 * (border >> ss_y)) * (uint64_t)(*uv_stride) +
192                   byte_alignment;
193   return 0;
194 }
195 
aom_realloc_frame_buffer(YV12_BUFFER_CONFIG * ybf,int width,int height,int ss_x,int ss_y,int use_highbitdepth,int border,int byte_alignment,aom_codec_frame_buffer_t * fb,aom_get_frame_buffer_cb_fn_t cb,void * cb_priv)196 int aom_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height,
197                              int ss_x, int ss_y, int use_highbitdepth,
198                              int border, int byte_alignment,
199                              aom_codec_frame_buffer_t *fb,
200                              aom_get_frame_buffer_cb_fn_t cb, void *cb_priv) {
201 #if CONFIG_SIZE_LIMIT
202   if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT) return -1;
203 #endif
204 
205   if (ybf) {
206     int y_stride = 0;
207     int uv_stride = 0;
208     uint64_t yplane_size = 0;
209     uint64_t uvplane_size = 0;
210     const int aligned_width = (width + 7) & ~7;
211     const int aligned_height = (height + 7) & ~7;
212     const int uv_width = aligned_width >> ss_x;
213     const int uv_height = aligned_height >> ss_y;
214     const int uv_border_w = border >> ss_x;
215     const int uv_border_h = border >> ss_y;
216 
217     int error = calc_stride_and_planesize(
218         ss_x, ss_y, aligned_width, aligned_height, border, byte_alignment,
219         &y_stride, &uv_stride, &yplane_size, &uvplane_size, uv_height);
220     if (error) return error;
221     return realloc_frame_buffer_aligned(
222         ybf, width, height, ss_x, ss_y, use_highbitdepth, border,
223         byte_alignment, fb, cb, cb_priv, y_stride, yplane_size, uvplane_size,
224         aligned_width, aligned_height, uv_width, uv_height, uv_stride,
225         uv_border_w, uv_border_h);
226   }
227   return -2;
228 }
229 
230 // TODO(anyone): This function allocates memory for
231 // lookahead buffer considering height and width is
232 // aligned to 128. Currently variance calculation of
233 // simple_motion_search_get_best_ref() function is done
234 // for full sb size (i.e integral multiple of max sb
235 // size = 128 or 64). Hence partial sbs need up to 127
236 // pixels beyond frame boundary. 128 aligned limitation of
237 // lookahead buffer can be removed if variance calculation
238 // is adjusted for partial sbs
239 
240 // NOTE: Chroma width and height need not be aligned to
241 // 128 since variance calculation happens only for luma plane
aom_realloc_lookahead_buffer(YV12_BUFFER_CONFIG * ybf,int width,int height,int ss_x,int ss_y,int use_highbitdepth,int border,int byte_alignment,aom_codec_frame_buffer_t * fb,aom_get_frame_buffer_cb_fn_t cb,void * cb_priv)242 int aom_realloc_lookahead_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height,
243                                  int ss_x, int ss_y, int use_highbitdepth,
244                                  int border, int byte_alignment,
245                                  aom_codec_frame_buffer_t *fb,
246                                  aom_get_frame_buffer_cb_fn_t cb,
247                                  void *cb_priv) {
248   if (ybf) {
249     int y_stride = 0;
250     int uv_stride = 0;
251     uint64_t yplane_size = 0;
252     uint64_t uvplane_size = 0;
253     const int aligned_128_width = (width + 127) & ~127;
254     const int aligned_128_height = (height + 127) & ~127;
255     const int aligned_width = (width + 7) & ~7;
256     const int aligned_height = (height + 7) & ~7;
257     const int uv_64_height = aligned_128_height >> ss_y;
258     const int uv_width = aligned_width >> ss_x;
259     const int uv_height = aligned_height >> ss_y;
260     const int uv_border_w = border >> ss_x;
261     const int uv_border_h = border >> ss_y;
262 
263     int error = calc_stride_and_planesize(
264         ss_x, ss_y, aligned_128_width, aligned_128_height, border,
265         byte_alignment, &y_stride, &uv_stride, &yplane_size, &uvplane_size,
266         uv_64_height);
267     if (error) return error;
268 
269     return realloc_frame_buffer_aligned(
270         ybf, width, height, ss_x, ss_y, use_highbitdepth, border,
271         byte_alignment, fb, cb, cb_priv, y_stride, yplane_size, uvplane_size,
272         aligned_width, aligned_height, uv_width, uv_height, uv_stride,
273         uv_border_w, uv_border_h);
274   }
275   return -2;
276 }
277 
aom_alloc_frame_buffer(YV12_BUFFER_CONFIG * ybf,int width,int height,int ss_x,int ss_y,int use_highbitdepth,int border,int byte_alignment)278 int aom_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height,
279                            int ss_x, int ss_y, int use_highbitdepth, int border,
280                            int byte_alignment) {
281   if (ybf) {
282     aom_free_frame_buffer(ybf);
283     return aom_realloc_frame_buffer(ybf, width, height, ss_x, ss_y,
284                                     use_highbitdepth, border, byte_alignment,
285                                     NULL, NULL, NULL);
286   }
287   return -2;
288 }
289