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 |
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 |
|
|
15 |
|
|
16 |
|
|
17 |
static __inline void |
static __inline void |
18 |
compensate8x8_interpolate(int16_t * const dct_codes, |
compensate16x16_interpolate(int16_t * const dct_codes, |
19 |
uint8_t * const cur, |
uint8_t * const cur, |
20 |
const uint8_t * const ref, |
const uint8_t * const ref, |
21 |
const uint8_t * const refh, |
const uint8_t * const refh, |
22 |
const uint8_t * const refv, |
uint8_t * const refv, |
23 |
const uint8_t * const refhv, |
const uint8_t * const refhv, |
24 |
const uint32_t x, |
uint32_t x, |
25 |
const uint32_t y, |
uint32_t y, |
26 |
const int32_t dx, |
const int32_t dx, |
27 |
const int32_t dy, |
const int32_t dy, |
28 |
const uint32_t stride, |
const uint32_t stride, |
29 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
30 |
|
const int reduced_resolution, |
31 |
const uint32_t rounding) |
const uint32_t rounding) |
32 |
{ |
{ |
33 |
|
|
34 |
|
if (reduced_resolution) { |
35 |
|
const uint8_t * reference; |
36 |
|
x*=2; y*=2; |
37 |
|
|
38 |
|
reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
39 |
|
|
40 |
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
41 |
|
filter_diff_18x18_to_8x8(dct_codes, reference, stride); |
42 |
|
|
43 |
|
filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride); |
44 |
|
filter_diff_18x18_to_8x8(dct_codes+64, reference + 16, stride); |
45 |
|
|
46 |
|
filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride); |
47 |
|
filter_diff_18x18_to_8x8(dct_codes+128, reference + 16*stride, stride); |
48 |
|
|
49 |
|
filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride); |
50 |
|
filter_diff_18x18_to_8x8(dct_codes+192, reference + 16*stride + 16, stride); |
51 |
|
|
52 |
|
transfer32x32_copy(cur + y*stride + x, reference, stride); |
53 |
|
|
54 |
|
} else { |
55 |
if(quarterpel) { |
if(quarterpel) { |
56 |
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
const uint8_t * ptr; |
57 |
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
if (dx&3 | dy&3) { |
58 |
|
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refv + 32, |
59 |
|
(uint8_t *) refv + 64, (uint8_t *) refv + 96, x, y, dx, dy, stride, rounding); |
60 |
|
ptr = refv + y*stride + x; |
61 |
|
} else ptr = ref + (y + dy/4)*stride + x + dx/4; // fullpixel position |
62 |
|
|
63 |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
64 |
refv + y*stride + x, stride); |
ptr, stride); |
65 |
|
transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, |
66 |
|
ptr + 8, stride); |
67 |
|
transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, |
68 |
|
ptr + 8*stride, stride); |
69 |
|
transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, |
70 |
|
ptr + 8*stride+8, stride); |
71 |
|
|
72 |
|
} else { |
73 |
|
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
74 |
|
|
75 |
|
transfer_8to16sub(dct_codes, cur + y * stride + x, |
76 |
|
reference, stride); |
77 |
|
transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
78 |
|
reference + 8, stride); |
79 |
|
transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, |
80 |
|
reference + 8*stride, stride); |
81 |
|
transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, |
82 |
|
reference + 8*stride + 8, stride); |
83 |
} |
} |
84 |
else |
} |
85 |
|
} |
86 |
|
|
87 |
|
static __inline void |
88 |
|
compensate8x8_interpolate(int16_t * const dct_codes, |
89 |
|
uint8_t * const cur, |
90 |
|
const uint8_t * const ref, |
91 |
|
const uint8_t * const refh, |
92 |
|
const uint8_t * const refv, |
93 |
|
const uint8_t * const refhv, |
94 |
|
uint32_t x, |
95 |
|
uint32_t y, |
96 |
|
const int32_t dx, |
97 |
|
const int32_t dy, |
98 |
|
const uint32_t stride, |
99 |
|
const uint32_t quarterpel, |
100 |
|
const int reduced_resolution, |
101 |
|
const uint32_t rounding) |
102 |
{ |
{ |
103 |
|
if (reduced_resolution) { |
104 |
const uint8_t * reference; |
const uint8_t * reference; |
105 |
|
x*=2; y*=2; |
106 |
|
|
107 |
|
reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
108 |
|
|
109 |
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
110 |
|
filter_diff_18x18_to_8x8(dct_codes, reference, stride); |
111 |
|
|
112 |
|
transfer16x16_copy(cur + y*stride + x, reference, stride); |
113 |
|
|
114 |
|
} else { |
115 |
|
|
116 |
|
if(quarterpel) { |
117 |
|
const uint8_t * ptr; |
118 |
|
if (dx&3 | dy&3) { |
119 |
|
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refv+32, |
120 |
|
(uint8_t *) refv + 64, (uint8_t *) refv+96, x, y, dx, dy, stride, rounding); |
121 |
|
ptr = refv + y*stride + x; |
122 |
|
} else ptr = ref + (y + dy/4)*stride + x + dx/4; // fullpixel position |
123 |
|
|
124 |
|
|
125 |
|
transfer_8to16sub(dct_codes, cur + y*stride + x, |
126 |
|
ptr, stride); |
127 |
|
} else { |
128 |
|
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
129 |
|
|
|
switch (((dx & 1) << 1) + (dy & 1)) // ((dx%2)?2:0)+((dy%2)?1:0) |
|
|
{ |
|
|
case 0: reference = ref + ((y + dy / 2) * stride + x + dx / 2); break; |
|
|
case 1: reference = refv + ((y + (dy-1) / 2) * stride + x + dx / 2); break; |
|
|
case 2: reference = refh + ((y + dy / 2) * stride + x + (dx-1) / 2); break; |
|
|
default: // case 3: |
|
|
reference = refhv + ((y + (dy-1) / 2) * stride + x + (dx-1) / 2); break; |
|
|
} |
|
130 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
131 |
reference, stride); |
reference, stride); |
132 |
} |
} |
133 |
} |
} |
134 |
|
} |
135 |
|
|
136 |
|
|
137 |
|
|
138 |
|
/* XXX: slow, inelegant... */ |
139 |
|
static void |
140 |
|
interpolate18x18_switch(uint8_t * const cur, |
141 |
|
const uint8_t * const refn, |
142 |
|
const uint32_t x, |
143 |
|
const uint32_t y, |
144 |
|
const int32_t dx, |
145 |
|
const int dy, |
146 |
|
const uint32_t stride, |
147 |
|
const uint32_t rounding) |
148 |
|
{ |
149 |
|
interpolate8x8_switch(cur, refn, x-1, y-1, dx, dy, stride, rounding); |
150 |
|
interpolate8x8_switch(cur, refn, x+7, y-1, dx, dy, stride, rounding); |
151 |
|
interpolate8x8_switch(cur, refn, x+9, y-1, dx, dy, stride, rounding); |
152 |
|
|
153 |
|
interpolate8x8_switch(cur, refn, x-1, y+7, dx, dy, stride, rounding); |
154 |
|
interpolate8x8_switch(cur, refn, x+7, y+7, dx, dy, stride, rounding); |
155 |
|
interpolate8x8_switch(cur, refn, x+9, y+7, dx, dy, stride, rounding); |
156 |
|
|
157 |
|
interpolate8x8_switch(cur, refn, x-1, y+9, dx, dy, stride, rounding); |
158 |
|
interpolate8x8_switch(cur, refn, x+7, y+9, dx, dy, stride, rounding); |
159 |
|
interpolate8x8_switch(cur, refn, x+9, y+9, dx, dy, stride, rounding); |
160 |
|
} |
161 |
|
|
162 |
|
|
163 |
void |
void |
164 |
MBMotionCompensation(MACROBLOCK * const mb, |
MBMotionCompensation(MACROBLOCK * const mb, |
174 |
const uint32_t height, |
const uint32_t height, |
175 |
const uint32_t edged_width, |
const uint32_t edged_width, |
176 |
const uint32_t quarterpel, |
const uint32_t quarterpel, |
177 |
|
const int reduced_resolution, |
178 |
const uint32_t rounding) |
const uint32_t rounding) |
179 |
{ |
{ |
180 |
if (mb->mode == MODE_NOT_CODED) { |
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
181 |
transfer8x8_copy(cur->y + 16 * (i + j * edged_width), |
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
182 |
|
|
183 |
|
if ( (!reduced_resolution) && (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
184 |
|
transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
185 |
ref->y + 16 * (i + j * edged_width), |
ref->y + 16 * (i + j * edged_width), |
186 |
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); |
|
187 |
|
|
188 |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
189 |
ref->u + 8 * (i + j * edged_width/2), |
ref->u + 8 * (i + j * edged_width/2), |
194 |
return; |
return; |
195 |
} |
} |
196 |
|
|
197 |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
if ((mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) /*&& !quarterpel*/) { |
|
int32_t dx = mb->mvs[0].x; |
|
|
int32_t dy = mb->mvs[0].y; |
|
198 |
|
|
199 |
if(quarterpel) { |
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
200 |
dx = mb->qmvs[0].x; |
|
201 |
dy = mb->qmvs[0].y; |
if (reduced_resolution) { |
202 |
|
dx = RRV_MV_SCALEUP(dx); |
203 |
|
dy = RRV_MV_SCALEUP(dy); |
204 |
} |
} |
205 |
|
|
206 |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
207 |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
208 |
edged_width, quarterpel, rounding); |
edged_width, quarterpel, reduced_resolution, 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, |
|
|
dy, edged_width, quarterpel, rounding); |
|
209 |
|
|
210 |
if (quarterpel) |
if (quarterpel) { |
211 |
{ |
dx /= 2; |
212 |
dx = (dx >> 1) | (dx & 1); |
dy /= 2; |
|
dy = (dy >> 1) | (dy & 1); |
|
213 |
} |
} |
214 |
|
|
215 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
216 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
217 |
|
|
218 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
219 |
|
if (reduced_resolution) { |
220 |
|
const int stride = edged_width/2; |
221 |
|
uint8_t * current, * reference; |
222 |
|
|
223 |
|
current = cur->u + 16*j*stride + 16*i; |
224 |
|
reference = refv->u + 16*j*stride + 16*i; |
225 |
|
interpolate18x18_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
226 |
|
filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
227 |
|
filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); |
228 |
|
transfer16x16_copy(current, reference, stride); |
229 |
|
|
230 |
|
current = cur->v + 16*j*stride + 16*i; |
231 |
|
reference = refv->v + 16*j*stride + 16*i; |
232 |
|
interpolate18x18_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
233 |
|
filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
234 |
|
filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); |
235 |
|
transfer16x16_copy(current, reference, stride); |
236 |
|
} else { |
237 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
238 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
239 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
245 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
246 |
dx, dy, edged_width / 2, rounding), |
dx, dy, edged_width / 2, rounding), |
247 |
edged_width / 2); |
edged_width / 2); |
248 |
|
} |
249 |
|
|
250 |
} else { // mode == MODE_INTER4V |
} else { // mode == MODE_INTER4V |
251 |
int32_t sum, dx, dy; |
int k; |
252 |
VECTOR *mvs; |
VECTOR mvs[4]; |
253 |
|
|
254 |
if(quarterpel) |
if(quarterpel) |
255 |
mvs = mb->qmvs; |
for (k = 0; k < 4; k++) mvs[k] = mb->qmvs[k]; |
256 |
else |
else |
257 |
mvs = mb->mvs; |
for (k = 0; k < 4; k++) mvs[k] = mb->mvs[k]; |
258 |
|
|
259 |
|
if (reduced_resolution) |
260 |
|
for (k = 0; k < 4; k++) { |
261 |
|
mvs[k].x = RRV_MV_SCALEUP(mvs[k].x); |
262 |
|
mvs[k].y = RRV_MV_SCALEUP(mvs[k].y); |
263 |
|
} |
264 |
|
|
265 |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
266 |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
267 |
mvs[0].y, edged_width, quarterpel, rounding); |
mvs[0].y, edged_width, quarterpel, reduced_resolution, rounding); |
268 |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
269 |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
270 |
mvs[1].x, mvs[1].y, edged_width, quarterpel, rounding); |
mvs[1].x, mvs[1].y, edged_width, quarterpel, reduced_resolution, rounding); |
271 |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
272 |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
273 |
mvs[2].x, mvs[2].y, edged_width, quarterpel, rounding); |
mvs[2].x, mvs[2].y, edged_width, quarterpel, reduced_resolution, rounding); |
274 |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
275 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
276 |
mvs[3].x, mvs[3].y, edged_width, quarterpel, rounding); |
mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); |
|
|
|
|
sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
|
|
|
|
|
if(quarterpel) |
|
|
sum /= 2; |
|
277 |
|
|
278 |
dx = (sum ? SIGN(sum) * |
if (quarterpel) { |
279 |
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
dx = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
280 |
|
dy = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); |
281 |
sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
} else { |
282 |
|
dx = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
283 |
|
dy = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
284 |
|
} |
285 |
|
|
286 |
if(quarterpel) |
dx = (dx >> 3) + roundtab_76[dx & 0xf]; |
287 |
sum /= 2; |
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
288 |
|
|
|
dy = (sum ? SIGN(sum) * |
|
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
|
289 |
|
|
290 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
291 |
|
if (reduced_resolution) { |
292 |
|
const int stride = edged_width/2; |
293 |
|
uint8_t * current, * reference; |
294 |
|
|
295 |
|
current = cur->u + 16*j*stride + 16*i; |
296 |
|
reference = refv->u + 16*j*stride + 16*i; |
297 |
|
interpolate18x18_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
298 |
|
filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
299 |
|
filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); |
300 |
|
transfer16x16_copy(current, reference, stride); |
301 |
|
|
302 |
|
current = cur->v + 16*j*stride + 16*i; |
303 |
|
reference = refv->v + 16*j*stride + 16*i; |
304 |
|
interpolate18x18_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
305 |
|
filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
306 |
|
filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); |
307 |
|
transfer16x16_copy(current, reference, stride); |
308 |
|
|
309 |
|
} else { |
310 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
311 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
312 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
317 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
318 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
319 |
dx, dy, edged_width / 2, rounding), |
dx, dy, edged_width / 2, rounding), |
|
|
|
320 |
edged_width / 2); |
edged_width / 2); |
321 |
} |
} |
322 |
} |
} |
323 |
|
} |
324 |
|
|
325 |
|
|
326 |
void |
void |
348 |
int k,sum; |
int k,sum; |
349 |
int x = i; |
int x = i; |
350 |
int y = j; |
int y = j; |
351 |
|
const uint32_t quarterpel = pParam->m_quarterpel; |
352 |
|
const uint8_t * ptr1, * ptr2; |
353 |
|
|
354 |
switch (mb->mode) { |
switch (mb->mode) { |
355 |
case MODE_FORWARD: |
case MODE_FORWARD: |
356 |
|
|
357 |
|
if (quarterpel) { |
358 |
|
dx = mb->qmvs[0].x; |
359 |
|
dy = mb->qmvs[0].y; |
360 |
|
} else { |
361 |
dx = mb->mvs[0].x; |
dx = mb->mvs[0].x; |
362 |
dy = mb->mvs[0].y; |
dy = mb->mvs[0].y; |
363 |
|
} |
364 |
|
|
365 |
transfer_8to16sub(&dct_codes[0 * 64], |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
366 |
cur->y + (j * 16) * edged_width + (i * 16), |
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
367 |
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
dy, edged_width, quarterpel, 0, 0); |
|
i * 16, j * 16, 1, dx, dy, edged_width), |
|
|
edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[1 * 64], |
|
|
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(&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); |
|
368 |
|
|
369 |
|
if (quarterpel) { |
370 |
|
dx /= 2; |
371 |
|
dy /= 2; |
372 |
|
} |
373 |
|
|
374 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
375 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
376 |
|
|
377 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
378 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
379 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
380 |
interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, |
381 |
dx, dy, edged_width / 2, 0), |
dx, dy, edged_width / 2, 0), |
|
|
|
382 |
edged_width / 2); |
edged_width / 2); |
383 |
|
|
384 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
385 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
386 |
interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, |
interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, |
387 |
dx, dy, edged_width / 2, 0), |
dx, dy, edged_width / 2, 0), |
|
|
|
388 |
edged_width / 2); |
edged_width / 2); |
389 |
|
|
390 |
break; |
break; |
391 |
|
|
392 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
393 |
|
if (quarterpel) { |
394 |
|
b_dx = mb->b_qmvs[0].x; |
395 |
|
b_dy = mb->b_qmvs[0].y; |
396 |
|
} else { |
397 |
b_dx = mb->b_mvs[0].x; |
b_dx = mb->b_mvs[0].x; |
398 |
b_dy = mb->b_mvs[0].y; |
b_dy = mb->b_mvs[0].y; |
399 |
|
} |
400 |
|
|
401 |
transfer_8to16sub(&dct_codes[0 * 64], |
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
402 |
cur->y + (j * 16) * edged_width + (i * 16), |
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
403 |
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
b_dy, edged_width, quarterpel, 0, 0); |
|
i * 16, j * 16, 1, b_dx, b_dy, |
|
|
edged_width), edged_width); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[1 * 64], |
|
|
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); |
|
404 |
|
|
405 |
transfer_8to16sub(&dct_codes[2 * 64], |
if (quarterpel) { |
406 |
cur->y + (j * 16 + 8) * edged_width + (i * 16), |
b_dx /= 2; |
407 |
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
b_dy /= 2; |
408 |
i * 16, j * 16 + 8, 1, b_dx, b_dy, |
} |
|
edged_width), edged_width); |
|
409 |
|
|
410 |
transfer_8to16sub(&dct_codes[3 * 64], |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
411 |
cur->y + (j * 16 + 8) * edged_width + (i * 16 + 8), |
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
|
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); |
|
412 |
|
|
|
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
|
|
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
|
413 |
|
|
414 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
415 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
428 |
|
|
429 |
break; |
break; |
430 |
|
|
|
|
|
431 |
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) */ |
432 |
case MODE_DIRECT_NO4V: |
case MODE_DIRECT_NO4V: |
433 |
|
|
434 |
|
if (quarterpel) { |
435 |
|
dx = mb->qmvs[0].x; |
436 |
|
dy = mb->qmvs[0].y; |
437 |
|
b_dx = mb->b_qmvs[0].x; |
438 |
|
b_dy = mb->b_qmvs[0].y; |
439 |
|
|
440 |
|
if (dx&3 | dy&3) { |
441 |
|
interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refv->y + 32, |
442 |
|
(uint8_t *) f_refv->y + 64, (uint8_t *) f_refv->y + 96, 16*i, 16*j, dx, dy, edged_width, 0); |
443 |
|
ptr1 = f_refv->y + i * 16 + j * 16 * edged_width; |
444 |
|
} else ptr1 = f_ref->y + (16*j + dy/4)*edged_width + 16*i + dx/4; // fullpixel position |
445 |
|
|
446 |
|
if (b_dx&3 | b_dy&3) { |
447 |
|
interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) f_refv->y + 32, |
448 |
|
(uint8_t *) f_refv->y + 64, (uint8_t *) f_refv->y + 96, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
449 |
|
ptr2 = b_refv->y + i * 16 + j * 16 * edged_width; |
450 |
|
} else ptr2 = b_ref->y + (16*j + b_dy/4)*edged_width + 16*i + b_dx/4; // fullpixel position |
451 |
|
|
452 |
|
for (k = 0; k < 4; k++) { |
453 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
454 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
455 |
|
ptr1 + (k&1)*8 + (k>>1)*8*edged_width, |
456 |
|
ptr2 + (k&1)*8 + (k>>1)*8*edged_width, edged_width); |
457 |
|
} |
458 |
|
b_dx /= 2; |
459 |
|
b_dy /= 2; |
460 |
|
dx /= 2; |
461 |
|
dy /= 2; |
462 |
|
|
463 |
|
} else { |
464 |
dx = mb->mvs[0].x; |
dx = mb->mvs[0].x; |
465 |
dy = mb->mvs[0].y; |
dy = mb->mvs[0].y; |
|
|
|
466 |
b_dx = mb->b_mvs[0].x; |
b_dx = mb->b_mvs[0].x; |
467 |
b_dy = mb->b_mvs[0].y; |
b_dy = mb->b_mvs[0].y; |
468 |
|
|
478 |
edged_width); |
edged_width); |
479 |
} |
} |
480 |
|
|
481 |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
} |
482 |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
|
483 |
|
|
484 |
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
485 |
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
486 |
|
|
487 |
b_dx = (b_dx & 3) ? (b_dx >> 1) | 1 : b_dx / 2; |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
488 |
b_dy = (b_dy & 3) ? (b_dy >> 1) | 1 : b_dy / 2; |
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
489 |
|
|
490 |
transfer_8to16sub2(&dct_codes[4 * 64], |
transfer_8to16sub2(&dct_codes[4 * 64], |
491 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
507 |
|
|
508 |
case MODE_DIRECT: |
case MODE_DIRECT: |
509 |
|
|
510 |
for (k=0;k<4;k++) |
if (quarterpel) { |
511 |
{ |
for (k = 0; k < 4; k++) { |
512 |
|
dx = mb->qmvs[k].x; |
513 |
|
dy = mb->qmvs[k].y; |
514 |
|
b_dx = mb->b_qmvs[k].x; |
515 |
|
b_dy = mb->b_qmvs[k].y; |
516 |
|
|
517 |
|
if (dx&3 | dy&3) { |
518 |
|
interpolate8x8_quarterpel((uint8_t *) f_refv->y, |
519 |
|
(uint8_t *) f_ref->y, |
520 |
|
(uint8_t *) f_refv->y + 32, |
521 |
|
(uint8_t *) f_refv->y + 64, |
522 |
|
(uint8_t *) f_refv->y + 96, |
523 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
524 |
|
ptr1 = f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width; |
525 |
|
} else ptr1 = f_ref->y + (16*j + (k>>1)*8 + dy/4)*edged_width + 16*i + (k&1)*8 + dx/4; |
526 |
|
|
527 |
|
if (b_dx&3 | b_dy&3) { |
528 |
|
interpolate8x8_quarterpel((uint8_t *) b_refv->y, |
529 |
|
(uint8_t *) b_ref->y, |
530 |
|
(uint8_t *) f_refv->y + 32, |
531 |
|
(uint8_t *) f_refv->y + 64, |
532 |
|
(uint8_t *) f_refv->y + 96, |
533 |
|
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
534 |
|
ptr2 = b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width; |
535 |
|
} else ptr2 = b_ref->y + (16*j + (k>>1)*8 + b_dy/4)*edged_width + 16*i + (k&1)*8 + b_dx/4; |
536 |
|
|
537 |
|
|
538 |
|
transfer_8to16sub2(&dct_codes[k * 64], |
539 |
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
540 |
|
ptr1, ptr2, edged_width); |
541 |
|
} |
542 |
|
sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; |
543 |
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
544 |
|
sum = mb->qmvs[0].x/2 + mb->qmvs[1].x/2 + mb->qmvs[2].x/2 + mb->qmvs[3].x/2; |
545 |
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
546 |
|
|
547 |
|
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; |
548 |
|
b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
549 |
|
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; |
550 |
|
b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
551 |
|
|
552 |
|
} else { |
553 |
|
for (k=0;k<4;k++) { |
554 |
dx = mb->mvs[k].x; |
dx = mb->mvs[k].x; |
555 |
dy = mb->mvs[k].y; |
dy = mb->mvs[k].y; |
556 |
|
|
557 |
b_dx = mb->b_mvs[k].x; |
b_dx = mb->b_mvs[k].x; |
558 |
b_dy = mb->b_mvs[k].y; |
b_dy = mb->b_mvs[k].y; |
559 |
|
|
|
// fprintf(stderr,"Direct Vector %d -- %d:%d %d:%d\n",k,dx,dy,b_dx,b_dy); |
|
|
|
|
560 |
transfer_8to16sub2(&dct_codes[k * 64], |
transfer_8to16sub2(&dct_codes[k * 64], |
561 |
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, |
562 |
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, |
574 |
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; |
575 |
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)); |
576 |
|
|
|
|
|
577 |
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; |
578 |
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)); |
579 |
|
|
580 |
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; |
581 |
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)); |
582 |
|
|
583 |
/* // for QPel don't forget to always do |
} |
|
|
|
|
if (quarterpel) |
|
|
sum /= 2; |
|
|
*/ |
|
584 |
transfer_8to16sub2(&dct_codes[4 * 64], |
transfer_8to16sub2(&dct_codes[4 * 64], |
585 |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
586 |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |