6 |
#include "../encoder.h" |
#include "../encoder.h" |
7 |
#include "../utils/mbfunctions.h" |
#include "../utils/mbfunctions.h" |
8 |
#include "../image/interpolate8x8.h" |
#include "../image/interpolate8x8.h" |
9 |
|
#include "../image/reduced.h" |
10 |
#include "../utils/timer.h" |
#include "../utils/timer.h" |
11 |
#include "motion.h" |
#include "motion.h" |
12 |
|
|
13 |
#define ABS(X) (((X)>0)?(X):-(X)) |
#define ABS(X) (((X)>0)?(X):-(X)) |
14 |
#define SIGN(X) (((X)>0)?1:-1) |
#define SIGN(X) (((X)>0)?1:-1) |
15 |
|
|
16 |
|
|
17 |
|
// decode an inter macroblock |
18 |
|
|
19 |
|
static void |
20 |
|
rrv_mv_scaleup(VECTOR * mv) |
21 |
|
{ |
22 |
|
if (mv->x > 0) { |
23 |
|
mv->x = 2*mv->x - 1; |
24 |
|
} else if (mv->x < 0) { |
25 |
|
mv->x = 2*mv->x + 1; |
26 |
|
} |
27 |
|
|
28 |
|
if (mv->y > 0) { |
29 |
|
mv->y = 2*mv->y - 1; |
30 |
|
} else if (mv->y < 0) { |
31 |
|
mv->y = 2*mv->y + 1; |
32 |
|
} |
33 |
|
} |
34 |
|
|
35 |
|
|
36 |
static __inline void |
static __inline void |
37 |
compensate16x16_interpolate(int16_t * const dct_codes, |
compensate16x16_interpolate(int16_t * const dct_codes, |
38 |
uint8_t * const cur, |
uint8_t * const cur, |
39 |
const uint8_t * const ref, |
const uint8_t * const ref, |
40 |
const uint8_t * const refh, |
const uint8_t * const refh, |
41 |
const uint8_t * const refv, |
uint8_t * const refv, |
42 |
const uint8_t * const refhv, |
const uint8_t * const refhv, |
43 |
const uint32_t x, |
uint32_t x, |
44 |
const uint32_t y, |
uint32_t y, |
45 |
const int32_t dx, |
const int32_t dx, |
46 |
const int32_t dy, |
const int32_t dy, |
47 |
const uint32_t stride, |
const uint32_t stride, |
48 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
49 |
|
const int reduced_resolution, |
50 |
const uint32_t rounding) |
const uint32_t rounding) |
51 |
{ |
{ |
52 |
|
|
53 |
|
if (reduced_resolution) |
54 |
|
{ |
55 |
|
|
56 |
|
/* XXX: todo */ |
57 |
|
|
58 |
|
VECTOR mv; |
59 |
|
int i,j; |
60 |
|
uint8_t tmp[18*18]; |
61 |
|
|
62 |
|
x*=2; |
63 |
|
y*=2; |
64 |
|
|
65 |
|
mv.x = dx; |
66 |
|
mv.y = dy; |
67 |
|
rrv_mv_scaleup(&mv); |
68 |
|
|
69 |
|
interpolate32x32_switch( |
70 |
|
refv, ref, x, y, mv.x, mv.y, stride, rounding); |
71 |
|
|
72 |
|
for (j = 0; j < 32; j++) |
73 |
|
for (i = 0; i < 32; i++) |
74 |
|
cur[(y+j)*stride + x + i] -= refv[(y+j)*stride + x + i]; |
75 |
|
|
76 |
|
filter_18x18_to_8x8(dct_codes, cur + y*stride + x, stride); |
77 |
|
filter_18x18_to_8x8(dct_codes+64, cur + y*stride + x + 16, stride); |
78 |
|
filter_18x18_to_8x8(dct_codes+128, cur + (y+16)*stride + x, stride); |
79 |
|
filter_18x18_to_8x8(dct_codes+192, cur + (y+16)*stride + x + 16, stride); |
80 |
|
|
81 |
|
/* |
82 |
|
for (j = 0; j < 16; j++) |
83 |
|
for (i = 0; i < 16; i++) |
84 |
|
tmp[(j+1)*18 + i+1] = refv[ (y+j)*stride + x+i]; |
85 |
|
|
86 |
|
for (i = 1; i < 17; i++) |
87 |
|
{ |
88 |
|
tmp[ 0*18 + i] = tmp[ 1*18 + i]; |
89 |
|
tmp[17*18 + i] = tmp[16*18 + i]; |
90 |
|
} |
91 |
|
|
92 |
|
for (i = 0; i < 18; i++) |
93 |
|
{ |
94 |
|
tmp[ i*18 + 0] = tmp[i*18 + 1]; |
95 |
|
tmp[ i*18 + 17] = tmp[i*18 + 16]; |
96 |
|
} |
97 |
|
filter_18x18_to_8x8(dct_codes, tmp, 18); |
98 |
|
|
99 |
|
|
100 |
|
for (j = 0; j < 16; j++) |
101 |
|
for (i = 0; i < 16; i++) |
102 |
|
tmp[(j+1)*18 + i+1] = refv[ (y+j)*stride + x+i + 16]; |
103 |
|
|
104 |
|
for (i = 1; i < 17; i++) |
105 |
|
{ |
106 |
|
tmp[ 0*18 + i] = tmp[ 1*18 + i]; |
107 |
|
tmp[17*18 + i] = tmp[16*18 + i]; |
108 |
|
} |
109 |
|
|
110 |
|
for (i = 0; i < 18; i++) |
111 |
|
{ |
112 |
|
tmp[ i*18 + 0] = tmp[i*18 + 1]; |
113 |
|
tmp[ i*18 + 17] = tmp[i*18 + 16]; |
114 |
|
} |
115 |
|
filter_18x18_to_8x8(dct_codes+64, tmp, 18); |
116 |
|
|
117 |
|
for (j = 0; j < 16; j++) |
118 |
|
for (i = 0; i < 16; i++) |
119 |
|
tmp[(j+1)*18 + i+1] = refv[ (y+16+j)*stride + x+i]; |
120 |
|
|
121 |
|
for (i = 1; i < 17; i++) |
122 |
|
{ |
123 |
|
tmp[ 0*18 + i] = tmp[ 1*18 + i]; |
124 |
|
tmp[17*18 + i] = tmp[16*18 + i]; |
125 |
|
} |
126 |
|
|
127 |
|
for (i = 0; i < 18; i++) |
128 |
|
{ |
129 |
|
tmp[ i*18 + 0] = tmp[i*18 + 1]; |
130 |
|
tmp[ i*18 + 17] = tmp[i*18 + 16]; |
131 |
|
} |
132 |
|
filter_18x18_to_8x8(dct_codes+128, tmp, 18); |
133 |
|
|
134 |
|
for (j = 0; j < 16; j++) |
135 |
|
for (i = 0; i < 16; i++) |
136 |
|
tmp[(j+1)*18 + i+1] = refv[ (y+16+j)*stride + x+i + 16]; |
137 |
|
|
138 |
|
for (i = 1; i < 17; i++) |
139 |
|
{ |
140 |
|
tmp[ 0*18 + i] = tmp[ 1*18 + i]; |
141 |
|
tmp[17*18 + i] = tmp[16*18 + i]; |
142 |
|
} |
143 |
|
|
144 |
|
for (i = 0; i < 18; i++) |
145 |
|
{ |
146 |
|
tmp[ i*18 + 0] = tmp[i*18 + 1]; |
147 |
|
tmp[ i*18 + 17] = tmp[i*18 + 16]; |
148 |
|
} |
149 |
|
filter_18x18_to_8x8(dct_codes+192, tmp, 18); |
150 |
|
*/ |
151 |
|
|
152 |
|
|
153 |
|
//memset(dct_codes, 0, sizeof(uint16_t) * 64 * 4); |
154 |
|
|
155 |
|
}else{ |
156 |
if(quarterpel) { |
if(quarterpel) { |
157 |
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
158 |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
189 |
reference + 8*stride + 8, stride); |
reference + 8*stride + 8, stride); |
190 |
} |
} |
191 |
} |
} |
192 |
|
} |
193 |
|
|
194 |
static __inline void |
static __inline void |
195 |
compensate8x8_interpolate(int16_t * const dct_codes, |
compensate8x8_interpolate(int16_t * const dct_codes, |
204 |
const int32_t dy, |
const int32_t dy, |
205 |
const uint32_t stride, |
const uint32_t stride, |
206 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
207 |
|
const int reduced_resolution, |
208 |
const uint32_t rounding) |
const uint32_t rounding) |
209 |
{ |
{ |
210 |
|
if (reduced_resolution) |
211 |
|
{ |
212 |
|
// XXX: todo |
213 |
|
} else { |
214 |
|
|
215 |
if(quarterpel) { |
if(quarterpel) { |
216 |
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
217 |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
235 |
reference, stride); |
reference, stride); |
236 |
} |
} |
237 |
} |
} |
238 |
|
} |
239 |
|
|
240 |
void |
void |
241 |
MBMotionCompensation(MACROBLOCK * const mb, |
MBMotionCompensation(MACROBLOCK * const mb, |
251 |
const uint32_t height, |
const uint32_t height, |
252 |
const uint32_t edged_width, |
const uint32_t edged_width, |
253 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
254 |
|
const int reduced_resolution, |
255 |
const uint32_t rounding) |
const uint32_t rounding) |
256 |
{ |
{ |
257 |
|
|
258 |
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
259 |
|
|
260 |
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
261 |
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
262 |
|
|
263 |
if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
264 |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width), |
transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
265 |
ref->y + 16 * (i + j * edged_width), |
ref->y + 16 * (i + j * edged_width), |
266 |
edged_width); |
edged_width); |
|
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8, |
|
|
ref->y + 16 * (i + j * edged_width) + 8, |
|
|
edged_width); |
|
|
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * edged_width, |
|
|
ref->y + 16 * (i + j * edged_width) + 8 * edged_width, |
|
|
edged_width); |
|
|
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), |
|
|
ref->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), |
|
|
edged_width); |
|
267 |
|
|
268 |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
269 |
ref->u + 8 * (i + j * edged_width/2), |
ref->u + 8 * (i + j * edged_width/2), |
276 |
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
277 |
|
|
278 |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
279 |
refv->y, refhv->y, 16 * i, 16 * j, dx, |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
280 |
dy, edged_width, quarterpel, rounding); |
edged_width, quarterpel, reduced_resolution, rounding); |
281 |
|
|
282 |
if (quarterpel) |
if (quarterpel) |
283 |
{ |
{ |
289 |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
290 |
|
|
291 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
292 |
|
if (reduced_resolution) |
293 |
|
{ |
294 |
|
// XXX: todo |
295 |
|
}else{ |
296 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
297 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
298 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
304 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
305 |
dx, dy, edged_width / 2, rounding), |
dx, dy, edged_width / 2, rounding), |
306 |
edged_width / 2); |
edged_width / 2); |
307 |
|
} |
308 |
|
|
309 |
} else { // mode == MODE_INTER4V |
} else { // mode == MODE_INTER4V |
310 |
int32_t sum, dx, dy; |
int32_t sum, dx, dy; |
315 |
else |
else |
316 |
mvs = mb->mvs; |
mvs = mb->mvs; |
317 |
|
|
318 |
|
if (reduced_resolution) |
319 |
|
{ |
320 |
|
///XXX: todo |
321 |
|
}else{ |
322 |
|
|
323 |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
324 |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
325 |
mvs[0].y, edged_width, quarterpel, rounding); |
mvs[0].y, edged_width, quarterpel, reduced_resolution, rounding); |
326 |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
327 |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
328 |
mvs[1].x, mvs[1].y, edged_width, quarterpel, rounding); |
mvs[1].x, mvs[1].y, edged_width, quarterpel, reduced_resolution, rounding); |
329 |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
330 |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
331 |
mvs[2].x, mvs[2].y, edged_width, quarterpel, rounding); |
mvs[2].x, mvs[2].y, edged_width, quarterpel, reduced_resolution, rounding); |
332 |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
333 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
334 |
mvs[3].x, mvs[3].y, edged_width, quarterpel, rounding); |
mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); |
335 |
|
} |
336 |
|
|
337 |
if(quarterpel) |
if(quarterpel) |
338 |
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
349 |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
350 |
|
|
351 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
352 |
|
if (reduced_resolution) |
353 |
|
{ |
354 |
|
//XXX: todo |
355 |
|
}else{ |
356 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
357 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
358 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
366 |
edged_width / 2); |
edged_width / 2); |
367 |
} |
} |
368 |
} |
} |
369 |
|
} |
370 |
|
|
371 |
|
|
372 |
void |
void |
409 |
|
|
410 |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
411 |
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
412 |
dy, edged_width, quarterpel, 0); |
dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
413 |
|
|
414 |
if (quarterpel) { |
if (quarterpel) { |
415 |
dx /= 2; |
dx /= 2; |
447 |
|
|
448 |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
449 |
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
450 |
b_dy, edged_width, quarterpel, 0); |
b_dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
451 |
|
|
452 |
if (quarterpel) { |
if (quarterpel) { |
453 |
b_dx /= 2; |
b_dx /= 2; |