1 |
|
// 30.10.2002 corrected qpel chroma rounding |
2 |
|
// 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 |
5 |
|
|
13 |
#define SIGN(X) (((X)>0)?1:-1) |
#define SIGN(X) (((X)>0)?1:-1) |
14 |
|
|
15 |
static __inline void |
static __inline void |
16 |
compensate8x8_halfpel(int16_t * const dct_codes, |
compensate16x16_interpolate(int16_t * const dct_codes, |
17 |
uint8_t * const cur, |
uint8_t * const cur, |
18 |
const uint8_t * const ref, |
const uint8_t * const ref, |
19 |
const uint8_t * const refh, |
const uint8_t * const refh, |
22 |
const uint32_t x, |
const uint32_t x, |
23 |
const uint32_t y, |
const uint32_t y, |
24 |
const int32_t dx, |
const int32_t dx, |
25 |
const int dy, |
const int32_t dy, |
26 |
const uint32_t stride) |
const uint32_t stride, |
27 |
|
const uint32_t quarterpel, |
28 |
|
const uint32_t rounding) |
29 |
{ |
{ |
30 |
int32_t ddx, ddy; |
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) |
switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) |
49 |
{ |
{ |
50 |
case 0: |
case 0: reference = ref + ((y + dy / 2) * stride + x + dx / 2); break; |
51 |
ddx = dx / 2; |
case 1: reference = refv + ((y + (dy-1) / 2) * stride + x + dx / 2); break; |
52 |
ddy = dy / 2; |
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, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
57 |
ref + (int) ((y + ddy) * stride + x + ddx), stride); |
reference, stride); |
58 |
break; |
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 |
case 1: |
static __inline void |
68 |
ddx = dx / 2; |
compensate8x8_interpolate(int16_t * const dct_codes, |
69 |
ddy = (dy - 1) / 2; |
uint8_t * const cur, |
70 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
const uint8_t * const ref, |
71 |
refv + (int) ((y + ddy) * stride + x + ddx), stride); |
const uint8_t * const refh, |
72 |
break; |
const uint8_t * const refv, |
73 |
|
const uint8_t * const refhv, |
74 |
|
const uint32_t x, |
75 |
|
const uint32_t y, |
76 |
|
const int32_t dx, |
77 |
|
const int32_t dy, |
78 |
|
const uint32_t stride, |
79 |
|
const uint32_t quarterpel, |
80 |
|
const uint32_t rounding) |
81 |
|
{ |
82 |
|
if(quarterpel) { |
83 |
|
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
84 |
|
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
85 |
|
|
|
case 2: |
|
|
ddx = (dx - 1) / 2; |
|
|
ddy = dy / 2; |
|
86 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
87 |
refh + (int) ((y + ddy) * stride + x + ddx), stride); |
refv + y*stride + x, stride); |
88 |
break; |
} |
89 |
|
else |
90 |
|
{ |
91 |
|
const uint8_t * reference; |
92 |
|
|
93 |
|
switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) |
94 |
|
{ |
95 |
|
case 0: reference = ref + ((y + dy / 2) * stride + x + dx / 2); break; |
96 |
|
case 1: reference = refv + ((y + (dy-1) / 2) * stride + x + dx / 2); break; |
97 |
|
case 2: reference = refh + ((y + dy / 2) * stride + x + (dx-1) / 2); break; |
98 |
default: // case 3: |
default: // case 3: |
99 |
ddx = (dx - 1) / 2; |
reference = refhv + ((y + (dy-1) / 2) * stride + x + (dx-1) / 2); break; |
100 |
ddy = (dy - 1) / 2; |
} |
101 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
102 |
refhv + (int) ((y + ddy) * stride + x + ddx), stride); |
reference, stride); |
|
break; |
|
103 |
} |
} |
104 |
} |
} |
105 |
|
|
|
|
|
|
|
|
106 |
void |
void |
107 |
MBMotionCompensation(MACROBLOCK * const mb, |
MBMotionCompensation(MACROBLOCK * const mb, |
108 |
const uint32_t i, |
const uint32_t i, |
116 |
const uint32_t width, |
const uint32_t width, |
117 |
const uint32_t height, |
const uint32_t height, |
118 |
const uint32_t edged_width, |
const uint32_t edged_width, |
119 |
|
const uint32_t quarterpel, |
120 |
const uint32_t rounding) |
const uint32_t rounding) |
121 |
{ |
{ |
122 |
static const uint32_t roundtab[16] = |
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
|
{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; |
|
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_INTER || mb->mode == MODE_INTER_Q) { |
if ( (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
128 |
int32_t dx = mb->mvs[0].x; |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width), |
129 |
int32_t dy = mb->mvs[0].y; |
ref->y + 16 * (i + j * edged_width), |
|
|
|
|
compensate8x8_halfpel(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
|
130 |
edged_width); |
edged_width); |
131 |
compensate8x8_halfpel(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8, |
132 |
refv->y, refhv->y, 16 * i + 8, 16 * j, dx, dy, |
ref->y + 16 * (i + j * edged_width) + 8, |
133 |
edged_width); |
edged_width); |
134 |
compensate8x8_halfpel(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * edged_width, |
135 |
refv->y, refhv->y, 16 * i, 16 * j + 8, dx, dy, |
ref->y + 16 * (i + j * edged_width) + 8 * edged_width, |
136 |
|
edged_width); |
137 |
|
transfer8x8_copy(cur->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), |
138 |
|
ref->y + 16 * (i + j * edged_width) + 8 * (edged_width+1), |
139 |
edged_width); |
edged_width); |
|
compensate8x8_halfpel(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
|
|
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, dx, |
|
|
dy, edged_width); |
|
|
|
|
|
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
|
|
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
|
140 |
|
|
141 |
/* u,v-image-based compensation |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
142 |
compensate8x8_halfpel(&dct_codes[4 * 64], cur->u, ref->u, refh->u, |
ref->u + 8 * (i + j * edged_width/2), |
|
refv->u, refhv->u, 8 * i, 8 * j, dx, dy, |
|
143 |
edged_width / 2); |
edged_width / 2); |
144 |
compensate8x8_halfpel(&dct_codes[5 * 64], cur->v, ref->v, refh->v, |
transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), |
145 |
refv->v, refhv->v, 8 * i, 8 * j, dx, dy, |
ref->v + 8 * (i + j * edged_width/2), |
146 |
edged_width / 2); |
edged_width / 2); |
147 |
*/ |
return; |
148 |
|
} |
149 |
|
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
150 |
|
|
151 |
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
152 |
|
refv->y, refhv->y, 16 * i, 16 * j, dx, |
153 |
|
dy, edged_width, quarterpel, rounding); |
154 |
|
|
155 |
/* --- u,v-block-based interpolation & compensation --- */ |
if (quarterpel) |
156 |
|
{ |
157 |
|
dx /= 2; |
158 |
|
dy /= 2; |
159 |
|
} |
160 |
|
|
161 |
interpolate8x8_switch(refv->u, ref->u, 8 * i, 8 * j, dx, dy, |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
162 |
edged_width / 2, rounding); |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
163 |
|
|
164 |
|
/* uv-block-based compensation */ |
165 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
166 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
167 |
refv->u + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
168 |
|
dx, dy, edged_width / 2, rounding), |
169 |
edged_width / 2); |
edged_width / 2); |
170 |
|
|
|
interpolate8x8_switch(refv->v, ref->v, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2, rounding); |
|
171 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
172 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
173 |
refv->v + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
174 |
|
dx, dy, edged_width / 2, rounding), |
175 |
edged_width / 2); |
edged_width / 2); |
|
/* */ |
|
176 |
|
|
177 |
} else // mode == MODE_INTER4V |
} else { // mode == MODE_INTER4V |
|
{ |
|
178 |
int32_t sum, dx, dy; |
int32_t sum, dx, dy; |
179 |
|
VECTOR *mvs; |
180 |
|
|
181 |
compensate8x8_halfpel(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
if(quarterpel) |
182 |
refv->y, refhv->y, 16 * i, 16 * j, mb->mvs[0].x, |
mvs = mb->qmvs; |
183 |
mb->mvs[0].y, edged_width); |
else |
184 |
compensate8x8_halfpel(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
mvs = mb->mvs; |
185 |
|
|
186 |
|
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
187 |
|
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
188 |
|
mvs[0].y, edged_width, quarterpel, rounding); |
189 |
|
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
190 |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
191 |
mb->mvs[1].x, mb->mvs[1].y, edged_width); |
mvs[1].x, mvs[1].y, edged_width, quarterpel, rounding); |
192 |
compensate8x8_halfpel(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
193 |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
194 |
mb->mvs[2].x, mb->mvs[2].y, edged_width); |
mvs[2].x, mvs[2].y, edged_width, quarterpel, rounding); |
195 |
compensate8x8_halfpel(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
196 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
197 |
mb->mvs[3].x, mb->mvs[3].y, edged_width); |
mvs[3].x, mvs[3].y, edged_width, quarterpel, rounding); |
198 |
|
|
199 |
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
if(quarterpel) |
200 |
dx = (sum ? SIGN(sum) * |
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
201 |
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
else |
202 |
|
sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
203 |
|
|
204 |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
dy = (sum ? SIGN(sum) * |
|
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
|
205 |
|
|
206 |
/* --- uv-image-based compensation -- - |
if(quarterpel) |
207 |
compensate8x8_halfpel(&dct_codes[4 * 64], cur->u, ref->u, refh->u, |
sum = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); |
208 |
refv->u, refhv->u, 8 * i, 8 * j, dx, dy, |
else |
209 |
edged_width / 2); |
sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
|
compensate8x8_halfpel(&dct_codes[5 * 64], cur->v, ref->v, refh->v, |
|
|
refv->v, refhv->v, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2); |
|
|
*/ |
|
210 |
|
|
211 |
/* --- uv-block-based compensation --- |
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
WARNING: these ditry the refv->u and refv->v images */ |
|
212 |
|
|
213 |
interpolate8x8_switch(refv->u, ref->u, 8 * i, 8 * j, dx, dy, |
/* uv-block-based compensation */ |
|
edged_width / 2, rounding); |
|
214 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
215 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
216 |
refv->u + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
217 |
|
dx, dy, edged_width / 2, rounding), |
218 |
edged_width / 2); |
edged_width / 2); |
219 |
|
|
|
interpolate8x8_switch(refv->v, ref->v, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2, rounding); |
|
220 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
221 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
222 |
refv->v + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
223 |
|
dx, dy, edged_width / 2, rounding), |
224 |
edged_width / 2); |
edged_width / 2); |
|
/* */ |
|
225 |
} |
} |
226 |
} |
} |
227 |
|
|
251 |
int k,sum; |
int k,sum; |
252 |
int x = i; |
int x = i; |
253 |
int y = j; |
int y = j; |
254 |
|
uint32_t quarterpel = pParam->m_quarterpel; |
255 |
|
|
256 |
switch (mb->mode) { |
switch (mb->mode) { |
257 |
case MODE_FORWARD: |
case MODE_FORWARD: |
258 |
|
|
259 |
|
if (quarterpel) { |
260 |
|
dx = mb->qmvs[0].x; |
261 |
|
dy = mb->qmvs[0].y; |
262 |
|
} else { |
263 |
dx = mb->mvs[0].x; |
dx = mb->mvs[0].x; |
264 |
dy = mb->mvs[0].y; |
dy = mb->mvs[0].y; |
265 |
|
} |
266 |
|
|
267 |
transfer_8to16sub_c(&dct_codes[0 * 64], |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
268 |
cur->y + (j * 16) * edged_width + (i * 16), |
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
269 |
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
dy, edged_width, quarterpel, 0); |
270 |
i * 16, j * 16, 1, dx, dy, edged_width), |
|
271 |
edged_width); |
if (quarterpel) { |
272 |
|
dx /= 2; |
273 |
transfer_8to16sub(&dct_codes[1 * 64], |
dy /= 2; |
274 |
cur->y + (j * 16) * edged_width + (i * 16 + 8), |
} |
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
|
|
i * 16 + 8, j * 16, 1, dx, dy, edged_width), |
|
|
edged_width); |
|
|
|
|
|
transfer_8to16sub_c(&dct_codes[2 * 64], |
|
|
cur->y + (j * 16 + 8) * edged_width + (i * 16), |
|
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
|
|
i * 16, j * 16 + 8, 1, dx, dy, |
|
|
edged_width), edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[3 * 64], |
|
|
cur->y + (j * 16 + 8) * edged_width + (i * 16 + 8), |
|
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
|
|
i * 16 + 8, j * 16 + 8, 1, dx, dy, |
|
|
edged_width), edged_width); |
|
|
|
|
275 |
|
|
276 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
277 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
278 |
|
|
279 |
/* --- uv-image-based compensation --- |
/* uv-block-based compensation */ |
|
compensate8x8_halfpel(&dct_codes[4 * 64], cur->u, f_ref->u, f_refh->u, |
|
|
f_refv->u, f_refhv->u, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2); |
|
|
compensate8x8_halfpel(&dct_codes[5 * 64], cur->v, f_ref->v, f_refh->v, |
|
|
f_refv->v, f_refhv->v, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2); |
|
|
*/ |
|
|
|
|
|
/* --- u,v-block-based interpolation & compensation --- */ |
|
|
interpolate8x8_switch(f_refv->u, f_ref->u, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2, 0); |
|
280 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
281 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
282 |
f_refv->u + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, |
283 |
|
dx, dy, edged_width / 2, 0), |
284 |
|
|
285 |
edged_width / 2); |
edged_width / 2); |
286 |
|
|
|
interpolate8x8_switch(f_refv->v, f_ref->v, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2, 0); |
|
287 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
288 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
289 |
f_refv->v + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, |
290 |
|
dx, dy, edged_width / 2, 0), |
291 |
|
|
292 |
edged_width / 2); |
edged_width / 2); |
|
/* */ |
|
293 |
|
|
294 |
break; |
break; |
295 |
|
|
296 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
297 |
|
if (quarterpel) { |
298 |
|
b_dx = mb->b_qmvs[0].x; |
299 |
|
b_dy = mb->b_qmvs[0].y; |
300 |
|
} else { |
301 |
b_dx = mb->b_mvs[0].x; |
b_dx = mb->b_mvs[0].x; |
302 |
b_dy = mb->b_mvs[0].y; |
b_dy = mb->b_mvs[0].y; |
303 |
|
} |
304 |
|
|
305 |
transfer_8to16sub_c(&dct_codes[0 * 64], |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
306 |
cur->y + (j * 16) * edged_width + (i * 16), |
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
307 |
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
b_dy, edged_width, quarterpel, 0); |
308 |
i * 16, j * 16, 1, b_dx, b_dy, |
|
309 |
edged_width), edged_width); |
if (quarterpel) { |
310 |
|
b_dx /= 2; |
311 |
transfer_8to16sub(&dct_codes[1 * 64], |
b_dy /= 2; |
312 |
cur->y + (j * 16) * edged_width + (i * 16 + 8), |
} |
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
|
|
i * 16 + 8, j * 16, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
|
|
|
|
transfer_8to16sub_c(&dct_codes[2 * 64], |
|
|
cur->y + (j * 16 + 8) * edged_width + (i * 16), |
|
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
|
|
i * 16, j * 16 + 8, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[3 * 64], |
|
|
cur->y + (j * 16 + 8) * edged_width + (i * 16 + 8), |
|
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
|
|
i * 16 + 8, j * 16 + 8, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
313 |
|
|
314 |
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
315 |
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
316 |
|
|
|
/* --- uv-image-based compensation --- |
|
|
compensate8x8_halfpel(&dct_codes[4 * 64], cur->u, b_ref->u, b_refh->u, |
|
|
b_refv->u, b_refhv->u, 8 * i, 8 * j, b_dx, b_dy, |
|
|
edged_width / 2); |
|
|
compensate8x8_halfpel(&dct_codes[5 * 64], cur->v, b_ref->v, b_refh->v, |
|
|
b_refv->v, b_refhv->v, 8 * i, 8 * j, b_dx, b_dy, |
|
|
edged_width / 2); |
|
|
*/ |
|
317 |
|
|
318 |
/* --- u,v-block-based interpolation & compensation --- */ |
/* uv-block-based compensation */ |
|
interpolate8x8_switch(b_refv->u, b_ref->u, 8 * i, 8 * j, b_dx, b_dy, |
|
|
edged_width / 2, 0); |
|
319 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
320 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
321 |
b_refv->u + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
322 |
|
b_dx, b_dy, edged_width / 2, 0), |
323 |
|
|
324 |
edged_width / 2); |
edged_width / 2); |
325 |
|
|
|
interpolate8x8_switch(b_refv->v, b_ref->v, 8 * i, 8 * j, b_dx, b_dy, |
|
|
edged_width / 2, 0); |
|
326 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
327 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
328 |
b_refv->v + 8 * j * edged_width / 2 + 8 * i, |
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
329 |
|
b_dx, b_dy, edged_width / 2, 0), |
330 |
|
|
331 |
edged_width / 2); |
edged_width / 2); |
|
/* */ |
|
332 |
|
|
333 |
break; |
break; |
334 |
|
|
|
|
|
335 |
case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
336 |
|
case MODE_DIRECT_NO4V: |
337 |
|
|
338 |
|
if (quarterpel) { |
339 |
|
dx = mb->qmvs[0].x; |
340 |
|
dy = mb->qmvs[0].y; |
341 |
|
b_dx = mb->b_qmvs[0].x; |
342 |
|
b_dy = mb->b_qmvs[0].y; |
343 |
|
|
344 |
|
interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refh->y, |
345 |
|
(uint8_t *) f_refh->y + 64, (uint8_t *) f_refhv->y, 16*i, 16*j, dx, dy, edged_width, 0); |
346 |
|
interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) b_refh->y, |
347 |
|
(uint8_t *) b_refh->y + 64, (uint8_t *) b_refhv->y, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
348 |
|
|
349 |
|
for (k = 0; k < 4; k++) { |
350 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
351 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
352 |
|
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
353 |
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
354 |
|
edged_width); |
355 |
|
} |
356 |
|
b_dx /= 2; |
357 |
|
b_dy /= 2; |
358 |
|
dx /= 2; |
359 |
|
dy /= 2; |
360 |
|
|
361 |
|
} else { |
362 |
dx = mb->mvs[0].x; |
dx = mb->mvs[0].x; |
363 |
dy = mb->mvs[0].y; |
dy = mb->mvs[0].y; |
|
|
|
364 |
b_dx = mb->b_mvs[0].x; |
b_dx = mb->b_mvs[0].x; |
365 |
b_dy = mb->b_mvs[0].y; |
b_dy = mb->b_mvs[0].y; |
366 |
|
|
367 |
for (k=0;k<4;k++) |
for (k = 0; k < 4; k++) { |
368 |
{ |
transfer_8to16sub2(&dct_codes[k * 64], |
|
transfer_8to16sub2_c(&dct_codes[k * 64], |
|
369 |
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
370 |
get_ref(f_ref->y, f_refh->y, f_refv->y, |
get_ref(f_ref->y, f_refh->y, f_refv->y, |
371 |
f_refhv->y, 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
f_refhv->y, 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
376 |
edged_width); |
edged_width); |
377 |
} |
} |
378 |
|
|
379 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
} |
380 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
|
381 |
|
|
382 |
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
383 |
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
384 |
|
|
385 |
/* --- uv-image-based compensation --- |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
386 |
|
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
387 |
|
|
388 |
transfer_8to16sub2_c(&dct_codes[4 * 64], |
transfer_8to16sub2(&dct_codes[4 * 64], |
389 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
390 |
get_ref(f_ref->u, f_refh->u, f_refv->u, |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
391 |
f_refhv->u, i, j, 8, dx, dy, |
b_dx, b_dy, edged_width / 2, 0), |
392 |
edged_width / 2), |
interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, |
393 |
get_ref(b_ref->u, b_refh->u, b_refv->u, |
dx, dy, edged_width / 2, 0), |
|
b_refhv->u, i, j, 8, b_dx, b_dy, |
|
|
edged_width / 2), |
|
394 |
edged_width / 2); |
edged_width / 2); |
395 |
|
|
396 |
transfer_8to16sub2_c(&dct_codes[5 * 64], |
transfer_8to16sub2(&dct_codes[5 * 64], |
397 |
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
398 |
get_ref(f_ref->v, f_refh->v, f_refv->v, |
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
399 |
f_refhv->v, 8 * i, 8 * j, 1, dx, dy, |
b_dx, b_dy, edged_width / 2, 0), |
400 |
edged_width / 2), |
interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, |
401 |
get_ref(b_ref->v, b_refh->v, b_refv->v, |
dx, dy, edged_width / 2, 0), |
|
b_refhv->v, 8 * i, 8 * j, 1, b_dx, b_dy, |
|
|
edged_width / 2), |
|
|
edged_width / 2); |
|
|
*/ |
|
|
|
|
|
/* --- u,v-block-based interpolation & compensation --- */ |
|
|
interpolate8x8_switch(f_refv->u, f_ref->u, 8 * i, 8 * j, dx, dy, edged_width / 2, 0); |
|
|
interpolate8x8_switch(b_refv->u, b_ref->u, 8 * i, 8 * j, b_dx, b_dy, edged_width / 2, 0); |
|
|
transfer_8to16sub2_c(&dct_codes[4 * 64], |
|
|
cur->u + 8 * y * edged_width / 2 + 8 * x, |
|
|
f_refv->u + 8 * j * edged_width / 2 + 8 * i, |
|
|
b_refv->u + 8 * j * edged_width / 2 + 8 * i, |
|
|
edged_width / 2); |
|
|
|
|
|
interpolate8x8_switch(f_refv->v, f_ref->v, 8 * i, 8 * j, dx, dy, edged_width / 2, 0); |
|
|
interpolate8x8_switch(b_refv->v, b_ref->v, 8 * i, 8 * j, b_dx, b_dy, edged_width / 2, 0); |
|
|
transfer_8to16sub2_c(&dct_codes[5 * 64], |
|
|
cur->v + 8 * y * edged_width / 2 + 8 * x, |
|
|
f_refv->v + 8 * j * edged_width / 2 + 8 * i, |
|
|
b_refv->v + 8 * j * edged_width / 2 + 8 * i, |
|
402 |
edged_width / 2); |
edged_width / 2); |
403 |
/* */ |
|
404 |
break; |
break; |
405 |
|
|
406 |
case MODE_DIRECT: |
case MODE_DIRECT: |
407 |
|
if (quarterpel) { |
408 |
|
for (k=0;k<4;k++) { |
409 |
|
|
410 |
for (k=0;k<4;k++) |
dx = mb->qmvs[k].x; |
411 |
{ |
dy = mb->qmvs[k].y; |
412 |
|
b_dx = mb->b_qmvs[k].x; |
413 |
|
b_dy = mb->b_qmvs[k].y; |
414 |
|
|
415 |
|
interpolate8x8_quarterpel((uint8_t *) f_refv->y, |
416 |
|
(uint8_t *) f_ref->y, |
417 |
|
(uint8_t *) f_refh->y, |
418 |
|
(uint8_t *) f_refh->y + 64, |
419 |
|
(uint8_t *) f_refhv->y, |
420 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
421 |
|
interpolate8x8_quarterpel((uint8_t *) b_refv->y, |
422 |
|
(uint8_t *) b_ref->y, |
423 |
|
(uint8_t *) b_refh->y, |
424 |
|
(uint8_t *) b_refh->y + 64, |
425 |
|
(uint8_t *) b_refhv->y, |
426 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
427 |
|
|
428 |
|
|
429 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
430 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
431 |
|
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
432 |
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
433 |
|
edged_width); |
434 |
|
} |
435 |
|
sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; |
436 |
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
437 |
|
sum = mb->qmvs[0].x/2 + mb->qmvs[1].x/2 + mb->qmvs[2].x/2 + mb->qmvs[3].x/2; |
438 |
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
439 |
|
|
440 |
|
sum = mb->b_qmvs[0].y/2 + mb->b_qmvs[1].y/2 + mb->b_qmvs[2].y/2 + mb->b_qmvs[3].y/2; |
441 |
|
b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
442 |
|
sum = mb->b_qmvs[0].x/2 + mb->b_qmvs[1].x/2 + mb->b_qmvs[2].x/2 + mb->b_qmvs[3].x/2; |
443 |
|
b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
444 |
|
|
445 |
|
} else { |
446 |
|
for (k=0;k<4;k++) { |
447 |
dx = mb->mvs[k].x; |
dx = mb->mvs[k].x; |
448 |
dy = mb->mvs[k].y; |
dy = mb->mvs[k].y; |
449 |
|
|
450 |
b_dx = mb->b_mvs[k].x; |
b_dx = mb->b_mvs[k].x; |
451 |
b_dy = mb->b_mvs[k].y; |
b_dy = mb->b_mvs[k].y; |
452 |
|
|
453 |
// fprintf(stderr,"Direct Vector %d -- %d:%d %d:%d\n",k,dx,dy,b_dx,b_dy); |
transfer_8to16sub2(&dct_codes[k * 64], |
|
|
|
|
transfer_8to16sub2_c(&dct_codes[k * 64], |
|
454 |
cur->y + (i*16 + (k&1)*8) + (j*16 + (k>>1)*8 ) * edged_width, |
cur->y + (i*16 + (k&1)*8) + (j*16 + (k>>1)*8 ) * edged_width, |
455 |
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
456 |
2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
467 |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
468 |
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
469 |
|
|
|
|
|
470 |
sum = mb->b_mvs[0].x + mb->b_mvs[1].x + mb->b_mvs[2].x + mb->b_mvs[3].x; |
sum = mb->b_mvs[0].x + mb->b_mvs[1].x + mb->b_mvs[2].x + mb->b_mvs[3].x; |
471 |
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
472 |
|
|
473 |
sum = mb->b_mvs[0].y + mb->b_mvs[1].y + mb->b_mvs[2].y + mb->b_mvs[3].y; |
sum = mb->b_mvs[0].y + mb->b_mvs[1].y + mb->b_mvs[2].y + mb->b_mvs[3].y; |
474 |
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
475 |
|
|
476 |
/* // for QPel don't forget to always do |
} |
477 |
|
transfer_8to16sub2(&dct_codes[4 * 64], |
|
if (quarterpel) |
|
|
sum /= 2; |
|
|
*/ |
|
|
|
|
|
/* --- uv-image-based compensation --- |
|
|
transfer_8to16sub2_c(&dct_codes[4 * 64], |
|
478 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
479 |
get_ref(f_ref->u, f_refh->u, f_refv->u, |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
480 |
f_refhv->u, i, j, 8, dx, dy, |
b_dx, b_dy, edged_width / 2, 0), |
481 |
edged_width / 2), |
interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, dx, dy, |
482 |
get_ref(b_ref->u, b_refh->u, b_refv->u, |
edged_width / 2, 0), |
|
b_refhv->u, i, j, 8, b_dx, b_dy, |
|
|
edged_width / 2), |
|
483 |
edged_width / 2); |
edged_width / 2); |
484 |
|
|
485 |
transfer_8to16sub2_c(&dct_codes[5 * 64], |
transfer_8to16sub2(&dct_codes[5 * 64], |
486 |
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
487 |
get_ref(f_ref->v, f_refh->v, f_refv->v, |
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
488 |
f_refhv->v, i, j, 8, dx, dy, |
b_dx, b_dy, edged_width / 2, 0), |
489 |
edged_width / 2), |
interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, |
490 |
get_ref(b_ref->v, b_refh->v, b_refv->v, |
dx, dy, edged_width / 2, 0), |
|
b_refhv->v, i, j, 8, b_dx, b_dy, |
|
|
edged_width / 2), |
|
|
edged_width / 2); |
|
|
*/ |
|
|
|
|
|
|
|
|
/* --- uv-block-based compensation */ |
|
|
interpolate8x8_switch(f_refv->u, f_ref->u, 8 * i, 8 * j, dx, dy, edged_width / 2, 0); |
|
|
interpolate8x8_switch(b_refv->u, b_ref->u, 8 * i, 8 * j, b_dx, b_dy, edged_width / 2, 0); |
|
|
transfer_8to16sub2_c(&dct_codes[4 * 64], |
|
|
cur->u + 8 * y * edged_width / 2 + 8 * x, |
|
|
f_refv->u + 8 * j * edged_width / 2 + 8 * i, |
|
|
b_refv->u + 8 * j * edged_width / 2 + 8 * i, |
|
|
edged_width / 2); |
|
|
|
|
|
interpolate8x8_switch(f_refv->v, f_ref->v, 8 * i, 8 * j, dx, dy, edged_width / 2, 0); |
|
|
interpolate8x8_switch(b_refv->v, b_ref->v, 8 * i, 8 * j, b_dx, b_dy, edged_width / 2, 0); |
|
|
transfer_8to16sub2_c(&dct_codes[5 * 64], |
|
|
cur->v + 8 * y * edged_width / 2 + 8 * x, |
|
|
f_refv->v + 8 * j * edged_width / 2 + 8 * i, |
|
|
b_refv->v + 8 * j * edged_width / 2 + 8 * i, |
|
491 |
edged_width / 2); |
edged_width / 2); |
492 |
/* */ |
|
493 |
|
|
494 |
break; |
break; |
495 |
} |
} |
496 |
} |
} |