FFmpeg
intrax8dsp.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
19 /**
20  * @file
21  *@brief IntraX8 frame subdecoder image manipulation routines
22  */
23 
24 #include "intrax8dsp.h"
25 #include "libavutil/common.h"
26 
27 /*
28  * area positions, #3 is 1 pixel only, other are 8 pixels
29  * |66666666|
30  * 3|44444444|55555555|
31  * - -+--------+--------+
32  * 1 2|XXXXXXXX|
33  * 1 2|XXXXXXXX|
34  * 1 2|XXXXXXXX|
35  * 1 2|XXXXXXXX|
36  * 1 2|XXXXXXXX|
37  * 1 2|XXXXXXXX|
38  * 1 2|XXXXXXXX|
39  * 1 2|XXXXXXXX|
40  * ^-start
41  */
42 
43 #define area1 (0)
44 #define area2 (8)
45 #define area3 (8 + 8)
46 #define area4 (8 + 8 + 1)
47 #define area5 (8 + 8 + 1 + 8)
48 #define area6 (8 + 8 + 1 + 16)
49 
50 /**
51  Collect statistics and prepare the edge pixels required by the other spatial compensation functions.
52 
53  * @param src pointer to the beginning of the processed block
54  * @param dst pointer to emu_edge, edge pixels are stored the way other compensation routines do.
55  * @param linesize byte offset between 2 vertical pixels in the source image
56  * @param range pointer to the variable where the edge pixel range is to be stored (max-min values)
57  * @param psum pointer to the variable where the edge pixel sum is to be stored
58  * @param edges Informs this routine that the block is on an image border, so it has to interpolate the missing edge pixels.
59  and some of the edge pixels should be interpolated, the flag has the following meaning:
60  1 - mb_x==0 - first block in the row, interpolate area #1,#2,#3;
61  2 - mb_y==0 - first row, interpolate area #3,#4,#5,#6;
62  note: 1|2 - mb_x==mb_y==0 - first block, use 0x80 value for all areas;
63  4 - mb_x>= (mb_width-1) last block in the row, interpolate area #5;
64 -*/
65 static void x8_setup_spatial_compensation(uint8_t *src, uint8_t *dst,
66  ptrdiff_t stride, int *range,
67  int *psum, int edges)
68 {
69  uint8_t *ptr;
70  int sum;
71  int i;
72  int min_pix, max_pix;
73  uint8_t c;
74 
75  if ((edges & 3) == 3) {
76  *psum = 0x80 * (8 + 1 + 8 + 2);
77  *range = 0;
78  memset(dst, 0x80, 16 + 1 + 16 + 8);
79  /* this triggers flat_dc for sure. flat_dc avoids all (other)
80  * prediction modes, but requires dc_level decoding. */
81  return;
82  }
83 
84  min_pix = 256;
85  max_pix = -1;
86 
87  sum = 0;
88 
89  if (!(edges & 1)) { // (mb_x != 0) // there is previous block on this row
90  ptr = src - 1; // left column, area 2
91  for (i = 7; i >= 0; i--) {
92  c = *(ptr - 1); // area1, same mb as area2, no need to check
93  dst[area1 + i] = c;
94  c = *ptr;
95 
96  sum += c;
97  min_pix = FFMIN(min_pix, c);
98  max_pix = FFMAX(max_pix, c);
99  dst[area2 + i] = c;
100 
101  ptr += stride;
102  }
103  }
104 
105  if (!(edges & 2)) { // (mb_y != 0) // there is row above
106  ptr = src - stride; // top line
107  for (i = 0; i < 8; i++) {
108  c = *(ptr + i);
109  sum += c;
110  min_pix = FFMIN(min_pix, c);
111  max_pix = FFMAX(max_pix, c);
112  }
113  if (edges & 4) { // last block on the row?
114  memset(dst + area5, c, 8); // set with last pixel fr
115  memcpy(dst + area4, ptr, 8);
116  } else {
117  memcpy(dst + area4, ptr, 16); // both area4 and 5
118  }
119  // area6 always present in the above block
120  memcpy(dst + area6, ptr - stride, 8);
121  }
122  // now calculate the stuff we need
123  if (edges & 3) { // mb_x ==0 || mb_y == 0) {
124  int avg = (sum + 4) >> 3;
125 
126  if (edges & 1) // (mb_x == 0) { // implies mb_y !=0
127  memset(dst + area1, avg, 8 + 8 + 1); // areas 1, 2, 3 are averaged
128  else // implies y == 0 x != 0
129  memset(dst + area3, avg, 1 + 16 + 8); // areas 3, 4, 5, 6
130 
131  sum += avg * 9;
132  } else {
133  // the edge pixel, in the top line and left column
134  uint8_t c = *(src - 1 - stride);
135  dst[area3] = c;
136  sum += c;
137  // edge pixel is not part of min/max
138  }
139  *range = max_pix - min_pix;
140  sum += *(dst + area5) + *(dst + area5 + 1);
141  *psum = sum;
142 }
143 
144 static const uint16_t zero_prediction_weights[64 * 2] = {
145  640, 640, 669, 480, 708, 354, 748, 257,
146  792, 198, 760, 143, 808, 101, 772, 72,
147  480, 669, 537, 537, 598, 416, 661, 316,
148  719, 250, 707, 185, 768, 134, 745, 97,
149  354, 708, 416, 598, 488, 488, 564, 388,
150  634, 317, 642, 241, 716, 179, 706, 132,
151  257, 748, 316, 661, 388, 564, 469, 469,
152  543, 395, 571, 311, 655, 238, 660, 180,
153  198, 792, 250, 719, 317, 634, 395, 543,
154  469, 469, 507, 380, 597, 299, 616, 231,
155  161, 855, 206, 788, 266, 710, 340, 623,
156  411, 548, 455, 455, 548, 366, 576, 288,
157  122, 972, 159, 914, 211, 842, 276, 758,
158  341, 682, 389, 584, 483, 483, 520, 390,
159  110, 1172, 144, 1107, 193, 1028, 254, 932,
160  317, 846, 366, 731, 458, 611, 499, 499,
161 };
162 
163 static void spatial_compensation_0(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
164 {
165  int i, j;
166  int x, y;
167  unsigned int p; // power divided by 2
168  int a;
169  uint16_t left_sum[2][8] = { { 0 } };
170  uint16_t top_sum[2][8] = { { 0 } };
171 
172  for (i = 0; i < 8; i++) {
173  a = src[area2 + 7 - i] << 4;
174  for (j = 0; j < 8; j++) {
175  p = abs(i - j);
176  left_sum[p & 1][j] += a >> (p >> 1);
177  }
178  }
179 
180  for (i = 0; i < 8; i++) {
181  a = src[area4 + i] << 4;
182  for (j = 0; j < 8; j++) {
183  p = abs(i - j);
184  top_sum[p & 1][j] += a >> (p >> 1);
185  }
186  }
187  for (; i < 10; i++) {
188  a = src[area4 + i] << 4;
189  for (j = 5; j < 8; j++) {
190  p = abs(i - j);
191  top_sum[p & 1][j] += a >> (p >> 1);
192  }
193  }
194  for (; i < 12; i++) {
195  a = src[area4 + i] << 4;
196  for (j = 7; j < 8; j++) {
197  p = abs(i - j);
198  top_sum[p & 1][j] += a >> (p >> 1);
199  }
200  }
201 
202  for (i = 0; i < 8; i++) {
203  top_sum[0][i] += (top_sum[1][i] * 181 + 128) >> 8; // 181 is sqrt(2)/2
204  left_sum[0][i] += (left_sum[1][i] * 181 + 128) >> 8;
205  }
206  for (y = 0; y < 8; y++) {
207  for (x = 0; x < 8; x++)
208  dst[x] = ((uint32_t) top_sum[0][x] * zero_prediction_weights[y * 16 + x * 2 + 0] +
209  (uint32_t) left_sum[0][y] * zero_prediction_weights[y * 16 + x * 2 + 1] +
210  0x8000) >> 16;
211  dst += stride;
212  }
213 }
214 
215 static void spatial_compensation_1(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
216 {
217  int x, y;
218 
219  for (y = 0; y < 8; y++) {
220  for (x = 0; x < 8; x++)
221  dst[x] = src[area4 + FFMIN(2 * y + x + 2, 15)];
222  dst += stride;
223  }
224 }
225 
226 static void spatial_compensation_2(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
227 {
228  int x, y;
229 
230  for (y = 0; y < 8; y++) {
231  for (x = 0; x < 8; x++)
232  dst[x] = src[area4 + 1 + y + x];
233  dst += stride;
234  }
235 }
236 
237 static void spatial_compensation_3(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
238 {
239  int x, y;
240 
241  for (y = 0; y < 8; y++) {
242  for (x = 0; x < 8; x++)
243  dst[x] = src[area4 + ((y + 1) >> 1) + x];
244  dst += stride;
245  }
246 }
247 
248 static void spatial_compensation_4(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
249 {
250  int x, y;
251 
252  for (y = 0; y < 8; y++) {
253  for (x = 0; x < 8; x++)
254  dst[x] = (src[area4 + x] + src[area6 + x] + 1) >> 1;
255  dst += stride;
256  }
257 }
258 
259 static void spatial_compensation_5(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
260 {
261  int x, y;
262 
263  for (y = 0; y < 8; y++) {
264  for (x = 0; x < 8; x++) {
265  if (2 * x - y < 0)
266  dst[x] = src[area2 + 9 + 2 * x - y];
267  else
268  dst[x] = src[area4 + x - ((y + 1) >> 1)];
269  }
270  dst += stride;
271  }
272 }
273 
274 static void spatial_compensation_6(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
275 {
276  int x, y;
277 
278  for (y = 0; y < 8; y++) {
279  for (x = 0; x < 8; x++)
280  dst[x] = src[area3 + x - y];
281  dst += stride;
282  }
283 }
284 
285 static void spatial_compensation_7(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
286 {
287  int x, y;
288 
289  for (y = 0; y < 8; y++) {
290  for (x = 0; x < 8; x++) {
291  if (x - 2 * y > 0)
292  dst[x] = (src[area3 - 1 + x - 2 * y] + src[area3 + x - 2 * y] + 1) >> 1;
293  else
294  dst[x] = src[area2 + 8 - y + (x >> 1)];
295  }
296  dst += stride;
297  }
298 }
299 
300 static void spatial_compensation_8(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
301 {
302  int x, y;
303 
304  for (y = 0; y < 8; y++) {
305  for (x = 0; x < 8; x++)
306  dst[x] = (src[area1 + 7 - y] + src[area2 + 7 - y] + 1) >> 1;
307  dst += stride;
308  }
309 }
310 
311 static void spatial_compensation_9(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
312 {
313  int x, y;
314 
315  for (y = 0; y < 8; y++) {
316  for (x = 0; x < 8; x++)
317  dst[x] = src[area2 + 6 - FFMIN(x + y, 6)];
318  dst += stride;
319  }
320 }
321 
322 static void spatial_compensation_10(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
323 {
324  int x, y;
325 
326  for (y = 0; y < 8; y++) {
327  for (x = 0; x < 8; x++)
328  dst[x] = (src[area2 + 7 - y] * (8 - x) + src[area4 + x] * x + 4) >> 3;
329  dst += stride;
330  }
331 }
332 
333 static void spatial_compensation_11(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
334 {
335  int x, y;
336 
337  for (y = 0; y < 8; y++) {
338  for (x = 0; x < 8; x++)
339  dst[x] = (src[area2 + 7 - y] * y + src[area4 + x] * (8 - y) + 4) >> 3;
340  dst += stride;
341  }
342 }
343 
344 static void x8_loop_filter(uint8_t *ptr, const ptrdiff_t a_stride,
345  const ptrdiff_t b_stride, int quant)
346 {
347  int i, t;
348  int p0, p1, p2, p3, p4, p5, p6, p7, p8, p9;
349  int ql = (quant + 10) >> 3;
350 
351  for (i = 0; i < 8; i++, ptr += b_stride) {
352  p0 = ptr[-5 * a_stride];
353  p1 = ptr[-4 * a_stride];
354  p2 = ptr[-3 * a_stride];
355  p3 = ptr[-2 * a_stride];
356  p4 = ptr[-1 * a_stride];
357  p5 = ptr[0];
358  p6 = ptr[1 * a_stride];
359  p7 = ptr[2 * a_stride];
360  p8 = ptr[3 * a_stride];
361  p9 = ptr[4 * a_stride];
362 
363  t = (FFABS(p1 - p2) <= ql) +
364  (FFABS(p2 - p3) <= ql) +
365  (FFABS(p3 - p4) <= ql) +
366  (FFABS(p4 - p5) <= ql);
367 
368  // You need at least 1 to be able to reach a total score of 6.
369  if (t > 0) {
370  t += (FFABS(p5 - p6) <= ql) +
371  (FFABS(p6 - p7) <= ql) +
372  (FFABS(p7 - p8) <= ql) +
373  (FFABS(p8 - p9) <= ql) +
374  (FFABS(p0 - p1) <= ql);
375  if (t >= 6) {
376  int min, max;
377 
378  min = max = p1;
379  min = FFMIN(min, p3);
380  max = FFMAX(max, p3);
381  min = FFMIN(min, p5);
382  max = FFMAX(max, p5);
383  min = FFMIN(min, p8);
384  max = FFMAX(max, p8);
385  if (max - min < 2 * quant) { // early stop
386  min = FFMIN(min, p2);
387  max = FFMAX(max, p2);
388  min = FFMIN(min, p4);
389  max = FFMAX(max, p4);
390  min = FFMIN(min, p6);
391  max = FFMAX(max, p6);
392  min = FFMIN(min, p7);
393  max = FFMAX(max, p7);
394  if (max - min < 2 * quant) {
395  ptr[-2 * a_stride] = (4 * p2 + 3 * p3 + 1 * p7 + 4) >> 3;
396  ptr[-1 * a_stride] = (3 * p2 + 3 * p4 + 2 * p7 + 4) >> 3;
397  ptr[0] = (2 * p2 + 3 * p5 + 3 * p7 + 4) >> 3;
398  ptr[1 * a_stride] = (1 * p2 + 3 * p6 + 4 * p7 + 4) >> 3;
399  continue;
400  }
401  }
402  }
403  }
404  {
405  int x, x0, x1, x2;
406  int m;
407 
408  x0 = (2 * p3 - 5 * p4 + 5 * p5 - 2 * p6 + 4) >> 3;
409  if (FFABS(x0) < quant) {
410  x1 = (2 * p1 - 5 * p2 + 5 * p3 - 2 * p4 + 4) >> 3;
411  x2 = (2 * p5 - 5 * p6 + 5 * p7 - 2 * p8 + 4) >> 3;
412 
413  x = FFABS(x0) - FFMIN(FFABS(x1), FFABS(x2));
414  m = p4 - p5;
415 
416  if (x > 0 && (m ^ x0) < 0) {
417  int32_t sign;
418 
419  sign = m >> 31;
420  m = (m ^ sign) - sign; // abs(m)
421  m >>= 1;
422 
423  x = 5 * x >> 3;
424 
425  if (x > m)
426  x = m;
427 
428  x = (x ^ sign) - sign;
429 
430  ptr[-1 * a_stride] -= x;
431  ptr[0] += x;
432  }
433  }
434  }
435  }
436 }
437 
438 static void x8_h_loop_filter(uint8_t *src, ptrdiff_t stride, int qscale)
439 {
440  x8_loop_filter(src, stride, 1, qscale);
441 }
442 
443 static void x8_v_loop_filter(uint8_t *src, ptrdiff_t stride, int qscale)
444 {
445  x8_loop_filter(src, 1, stride, qscale);
446 }
447 
449 {
465 }
IntraX8DSPContext::setup_spatial_compensation
void(* setup_spatial_compensation)(uint8_t *src, uint8_t *dst, ptrdiff_t stride, int *range, int *sum, int edges)
Definition: intrax8dsp.h:31
spatial_compensation_3
static void spatial_compensation_3(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:237
IntraX8DSPContext
Definition: intrax8dsp.h:25
x8_setup_spatial_compensation
static void x8_setup_spatial_compensation(uint8_t *src, uint8_t *dst, ptrdiff_t stride, int *range, int *psum, int edges)
Collect statistics and prepare the edge pixels required by the other spatial compensation functions.
Definition: intrax8dsp.c:65
max
#define max(a, b)
Definition: cuda_runtime.h:33
FFMAX
#define FFMAX(a, b)
Definition: macros.h:47
spatial_compensation_8
static void spatial_compensation_8(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:300
intrax8dsp.h
area6
#define area6
Definition: intrax8dsp.c:48
spatial_compensation_6
static void spatial_compensation_6(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:274
spatial_compensation_2
static void spatial_compensation_2(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:226
area5
#define area5
Definition: intrax8dsp.c:47
zero_prediction_weights
static const uint16_t zero_prediction_weights[64 *2]
Definition: intrax8dsp.c:144
area2
#define area2
Definition: intrax8dsp.c:44
spatial_compensation_0
static void spatial_compensation_0(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:163
quant
static const uint8_t quant[64]
Definition: vmixdec.c:71
av_cold
#define av_cold
Definition: attributes.h:90
area4
#define area4
Definition: intrax8dsp.c:46
spatial_compensation_10
static void spatial_compensation_10(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:322
spatial_compensation_1
static void spatial_compensation_1(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:215
IntraX8DSPContext::v_loop_filter
void(* v_loop_filter)(uint8_t *src, ptrdiff_t stride, int qscale)
Definition: intrax8dsp.h:26
x8_v_loop_filter
static void x8_v_loop_filter(uint8_t *src, ptrdiff_t stride, int qscale)
Definition: intrax8dsp.c:443
FFABS
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:72
spatial_compensation_4
static void spatial_compensation_4(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:248
spatial_compensation_7
static void spatial_compensation_7(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:285
area1
#define area1
Definition: intrax8dsp.c:43
abs
#define abs(x)
Definition: cuda_runtime.h:35
x8_h_loop_filter
static void x8_h_loop_filter(uint8_t *src, ptrdiff_t stride, int qscale)
Definition: intrax8dsp.c:438
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
area3
#define area3
Definition: intrax8dsp.c:45
IntraX8DSPContext::h_loop_filter
void(* h_loop_filter)(uint8_t *src, ptrdiff_t stride, int qscale)
Definition: intrax8dsp.h:27
avg
#define avg(a, b, c, d)
Definition: colorspacedsp_template.c:28
range
enum AVColorRange range
Definition: mediacodec_wrapper.c:2646
x8_loop_filter
static void x8_loop_filter(uint8_t *ptr, const ptrdiff_t a_stride, const ptrdiff_t b_stride, int quant)
Definition: intrax8dsp.c:344
a
The reader does not expect b to be semantically here and if the code is changed by maybe adding a a division or other the signedness will almost certainly be mistaken To avoid this confusion a new type was SUINT is the C unsigned type but it holds a signed int to use the same example SUINT a
Definition: undefined.txt:41
IntraX8DSPContext::spatial_compensation
void(* spatial_compensation[12])(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.h:29
spatial_compensation_9
static void spatial_compensation_9(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:311
spatial_compensation_11
static void spatial_compensation_11(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:333
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:255
common.h
FFMIN
#define FFMIN(a, b)
Definition: macros.h:49
stride
#define stride
Definition: h264pred_template.c:537
src
INIT_CLIP pixel * src
Definition: h264pred_template.c:418
int32_t
int32_t
Definition: audioconvert.c:56
spatial_compensation_5
static void spatial_compensation_5(uint8_t *src, uint8_t *dst, ptrdiff_t stride)
Definition: intrax8dsp.c:259
ff_intrax8dsp_init
av_cold void ff_intrax8dsp_init(IntraX8DSPContext *dsp)
Definition: intrax8dsp.c:448
min
float min
Definition: vorbis_enc_data.h:429