1 |
|
// 30.10.2002 corrected qpel chroma rounding |
2 |
// 04.10.2002 added qpel support to MBMotionCompensation |
// 04.10.2002 added qpel support to MBMotionCompensation |
3 |
// 01.05.2002 updated MBMotionCompensationBVOP |
// 01.05.2002 updated MBMotionCompensationBVOP |
4 |
// 14.04.2002 bframe compensation |
// 14.04.2002 bframe compensation |
12 |
#define ABS(X) (((X)>0)?(X):-(X)) |
#define ABS(X) (((X)>0)?(X):-(X)) |
13 |
#define SIGN(X) (((X)>0)?1:-1) |
#define SIGN(X) (((X)>0)?1:-1) |
14 |
|
|
15 |
|
static __inline void |
16 |
|
compensate16x16_interpolate(int16_t * const dct_codes, |
17 |
|
uint8_t * const cur, |
18 |
|
const uint8_t * const ref, |
19 |
|
const uint8_t * const refh, |
20 |
|
const uint8_t * const refv, |
21 |
|
const uint8_t * const refhv, |
22 |
|
const uint32_t x, |
23 |
|
const uint32_t y, |
24 |
|
const int32_t dx, |
25 |
|
const int32_t dy, |
26 |
|
const uint32_t stride, |
27 |
|
const uint32_t quarterpel, |
28 |
|
const uint32_t rounding) |
29 |
|
{ |
30 |
|
if(quarterpel) { |
31 |
|
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
32 |
|
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
33 |
|
|
34 |
|
transfer_8to16sub(dct_codes, cur + y*stride + x, |
35 |
|
refv + y*stride + x, stride); |
36 |
|
transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, |
37 |
|
refv + y*stride + x + 8, stride); |
38 |
|
transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, |
39 |
|
refv + y*stride + x + 8*stride, stride); |
40 |
|
transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, |
41 |
|
refv + y*stride + x + 8*stride+8, stride); |
42 |
|
|
43 |
|
} |
44 |
|
else |
45 |
|
{ |
46 |
|
const uint8_t * reference; |
47 |
|
|
48 |
|
switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) |
49 |
|
{ |
50 |
|
case 0: reference = ref + ((y + dy / 2) * stride + x + dx / 2); break; |
51 |
|
case 1: reference = refv + ((y + (dy-1) / 2) * stride + x + dx / 2); break; |
52 |
|
case 2: reference = refh + ((y + dy / 2) * stride + x + (dx-1) / 2); break; |
53 |
|
default: // case 3: |
54 |
|
reference = refhv + ((y + (dy-1) / 2) * stride + x + (dx-1) / 2); break; |
55 |
|
} |
56 |
|
transfer_8to16sub(dct_codes, cur + y * stride + x, |
57 |
|
reference, stride); |
58 |
|
transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
59 |
|
reference + 8, stride); |
60 |
|
transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, |
61 |
|
reference + 8*stride, stride); |
62 |
|
transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, |
63 |
|
reference + 8*stride + 8, stride); |
64 |
|
} |
65 |
|
} |
66 |
|
|
67 |
static __inline void |
static __inline void |
68 |
compensate8x8_interpolate(int16_t * const dct_codes, |
compensate8x8_interpolate(int16_t * const dct_codes, |
119 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
120 |
const uint32_t rounding) |
const uint32_t rounding) |
121 |
{ |
{ |
122 |
if (mb->mode == MODE_NOT_CODED) { |
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
123 |
|
|
124 |
|
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
125 |
|
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
126 |
|
|
127 |
|
if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
128 |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width), |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width), |
129 |
ref->y + 16 * (i + j * edged_width), |
ref->y + 16 * (i + j * edged_width), |
130 |
edged_width); |
edged_width); |
146 |
edged_width / 2); |
edged_width / 2); |
147 |
return; |
return; |
148 |
} |
} |
149 |
|
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
150 |
|
|
151 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
152 |
int32_t dx = mb->mvs[0].x; |
refv->y, refhv->y, 16 * i, 16 * j, dx, |
|
int32_t dy = mb->mvs[0].y; |
|
|
|
|
|
if(quarterpel) { |
|
|
dx = mb->qmvs[0].x; |
|
|
dy = mb->qmvs[0].y; |
|
|
} |
|
|
|
|
|
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
|
|
edged_width, quarterpel, rounding); |
|
|
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i + 8, 16 * j, dx, dy, |
|
|
edged_width, quarterpel, rounding); |
|
|
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i, 16 * j + 8, dx, dy, |
|
|
edged_width, quarterpel, rounding); |
|
|
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, dx, |
|
153 |
dy, edged_width, quarterpel, rounding); |
dy, edged_width, quarterpel, rounding); |
154 |
|
|
155 |
if (quarterpel) |
if (quarterpel) |
156 |
{ |
{ |
157 |
dx = (dx >> 1) | (dx & 1); |
dx /= 2; |
158 |
dy = (dy >> 1) | (dy & 1); |
dy /= 2; |
159 |
} |
} |
160 |
|
|
161 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
162 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
163 |
|
|
164 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
165 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
196 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
197 |
mvs[3].x, mvs[3].y, edged_width, quarterpel, rounding); |
mvs[3].x, mvs[3].y, edged_width, quarterpel, rounding); |
198 |
|
|
|
sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
|
|
|
|
199 |
if(quarterpel) |
if(quarterpel) |
200 |
sum /= 2; |
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
201 |
|
else |
202 |
dx = (sum ? SIGN(sum) * |
sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
|
203 |
|
|
204 |
sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
205 |
|
|
206 |
if(quarterpel) |
if(quarterpel) |
207 |
sum /= 2; |
sum = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); |
208 |
|
else |
209 |
|
sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
210 |
|
|
211 |
dy = (sum ? SIGN(sum) * |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
|
212 |
|
|
213 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
214 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
221 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
222 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
223 |
dx, dy, edged_width / 2, rounding), |
dx, dy, edged_width / 2, rounding), |
|
|
|
224 |
edged_width / 2); |
edged_width / 2); |
225 |
} |
} |
226 |
} |
} |