1 |
// 30.10.2002 corrected qpel chroma rounding |
/***************************************************************************** |
2 |
// 04.10.2002 added qpel support to MBMotionCompensation |
* |
3 |
// 01.05.2002 updated MBMotionCompensationBVOP |
* XVID MPEG-4 VIDEO CODEC |
4 |
// 14.04.2002 bframe compensation |
* - Motion Compensation module - |
5 |
|
* |
6 |
|
* Copyright(C) 2002 Peter Ross <pross@xvid.org> |
7 |
|
* |
8 |
|
* This file is part of XviD, a free MPEG-4 video encoder/decoder |
9 |
|
* |
10 |
|
* XviD is free software; you can redistribute it and/or modify it |
11 |
|
* under the terms of the GNU General Public License as published by |
12 |
|
* the Free Software Foundation; either version 2 of the License, or |
13 |
|
* (at your option) any later version. |
14 |
|
* |
15 |
|
* This program is distributed in the hope that it will be useful, |
16 |
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
17 |
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
18 |
|
* GNU General Public License for more details. |
19 |
|
* |
20 |
|
* You should have received a copy of the GNU General Public License |
21 |
|
* along with this program; if not, write to the Free Software |
22 |
|
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
|
* |
24 |
|
* Under section 8 of the GNU General Public License, the copyright |
25 |
|
* holders of XVID explicitly forbid distribution in the following |
26 |
|
* countries: |
27 |
|
* |
28 |
|
* - Japan |
29 |
|
* - United States of America |
30 |
|
* |
31 |
|
* Linking XviD statically or dynamically with other modules is making a |
32 |
|
* combined work based on XviD. Thus, the terms and conditions of the |
33 |
|
* GNU General Public License cover the whole combination. |
34 |
|
* |
35 |
|
* As a special exception, the copyright holders of XviD give you |
36 |
|
* permission to link XviD with independent modules that communicate with |
37 |
|
* XviD solely through the VFW1.1 and DShow interfaces, regardless of the |
38 |
|
* license terms of these independent modules, and to copy and distribute |
39 |
|
* the resulting combined work under terms of your choice, provided that |
40 |
|
* every copy of the combined work is accompanied by a complete copy of |
41 |
|
* the source code of XviD (the version of XviD used to produce the |
42 |
|
* combined work), being distributed under the terms of the GNU General |
43 |
|
* Public License plus this exception. An independent module is a module |
44 |
|
* which is not derived from or based on XviD. |
45 |
|
* |
46 |
|
* Note that people who make modified versions of XviD are not obligated |
47 |
|
* to grant this special exception for their modified versions; it is |
48 |
|
* their choice whether to do so. The GNU General Public License gives |
49 |
|
* permission to release a modified version without this exception; this |
50 |
|
* exception also makes it possible to release a modified version which |
51 |
|
* carries forward this exception. |
52 |
|
* |
53 |
|
* $Id$ |
54 |
|
* |
55 |
|
*************************************************************************/ |
56 |
|
|
57 |
#include "../encoder.h" |
#include "../encoder.h" |
58 |
#include "../utils/mbfunctions.h" |
#include "../utils/mbfunctions.h" |
59 |
#include "../image/interpolate8x8.h" |
#include "../image/interpolate8x8.h" |
|
#include "../image/reduced.h" |
|
60 |
#include "../utils/timer.h" |
#include "../utils/timer.h" |
61 |
#include "motion.h" |
#include "motion.h" |
62 |
|
|
63 |
#define ABS(X) (((X)>0)?(X):-(X)) |
#define ABS(X) (((X)>0)?(X):-(X)) |
64 |
#define SIGN(X) (((X)>0)?1:-1) |
#define SIGN(X) (((X)>0)?1:-1) |
65 |
|
|
|
|
|
66 |
static __inline void |
static __inline void |
67 |
compensate16x16_interpolate(int16_t * const dct_codes, |
compensate8x8_halfpel(int16_t * const dct_codes, |
68 |
uint8_t * const cur, |
uint8_t * const cur, |
69 |
const uint8_t * const ref, |
const uint8_t * const ref, |
70 |
const uint8_t * const refh, |
const uint8_t * const refh, |
71 |
uint8_t * const refv, |
const uint8_t * const refv, |
72 |
const uint8_t * const refhv, |
const uint8_t * const refhv, |
73 |
uint32_t x, |
const uint32_t x, |
74 |
uint32_t y, |
const uint32_t y, |
75 |
const int32_t dx, |
const int32_t dx, |
76 |
const int32_t dy, |
const int dy, |
77 |
const uint32_t stride, |
const uint32_t stride) |
|
const uint32_t quarterpel, |
|
|
const int reduced_resolution, |
|
|
const uint32_t rounding) |
|
78 |
{ |
{ |
79 |
|
int32_t ddx, ddy; |
80 |
|
|
81 |
if (reduced_resolution) |
switch (((dx & 1) << 1) + (dy & 1)) /* ((dx%2)?2:0)+((dy%2)?1:0) */ |
82 |
{ |
{ |
83 |
const uint8_t * reference; |
case 0: |
84 |
x*=2; y*=2; |
ddx = dx / 2; |
85 |
|
ddy = dy / 2; |
|
reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
|
|
|
|
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes, reference, stride); |
|
|
|
|
|
filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes+64, reference + 16, stride); |
|
|
|
|
|
filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes+128, reference + 16*stride, stride); |
|
|
|
|
|
filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes+192, reference + 16*stride + 16, stride); |
|
|
|
|
|
transfer32x32_copy(cur + y*stride + x, reference, stride); |
|
|
|
|
|
}else{ |
|
|
if(quarterpel) { |
|
|
interpolate16x16_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
|
|
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
|
|
|
|
86 |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
87 |
refv + y*stride + x, stride); |
ref + (int) ((y + ddy) * stride + x + ddx), stride); |
88 |
transfer_8to16sub(dct_codes+64, cur + y*stride + x + 8, |
break; |
|
refv + y*stride + x + 8, stride); |
|
|
transfer_8to16sub(dct_codes+128, cur + y*stride + x + 8*stride, |
|
|
refv + y*stride + x + 8*stride, stride); |
|
|
transfer_8to16sub(dct_codes+192, cur + y*stride + x + 8*stride + 8, |
|
|
refv + y*stride + x + 8*stride+8, stride); |
|
|
|
|
|
} |
|
|
else |
|
|
{ |
|
|
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
|
89 |
|
|
90 |
|
case 1: |
91 |
|
ddx = dx / 2; |
92 |
|
ddy = (dy - 1) / 2; |
93 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
94 |
reference, stride); |
refv + (int) ((y + ddy) * stride + x + ddx), stride); |
95 |
transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8, |
break; |
|
reference + 8, stride); |
|
|
transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride, |
|
|
reference + 8*stride, stride); |
|
|
transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8, |
|
|
reference + 8*stride + 8, stride); |
|
|
} |
|
|
} |
|
|
} |
|
|
|
|
|
static __inline void |
|
|
compensate8x8_interpolate(int16_t * const dct_codes, |
|
|
uint8_t * const cur, |
|
|
const uint8_t * const ref, |
|
|
const uint8_t * const refh, |
|
|
const uint8_t * const refv, |
|
|
const uint8_t * const refhv, |
|
|
uint32_t x, |
|
|
uint32_t y, |
|
|
const int32_t dx, |
|
|
const int32_t dy, |
|
|
const uint32_t stride, |
|
|
const uint32_t quarterpel, |
|
|
const int reduced_resolution, |
|
|
const uint32_t rounding) |
|
|
{ |
|
|
if (reduced_resolution) |
|
|
{ |
|
|
const uint8_t * reference; |
|
|
x*=2; y*=2; |
|
|
|
|
|
reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
|
|
|
|
|
filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes, reference, stride); |
|
|
|
|
|
transfer16x16_copy(cur + y*stride + x, reference, stride); |
|
|
|
|
|
} else { |
|
|
|
|
|
if(quarterpel) { |
|
|
interpolate8x8_quarterpel((uint8_t *) refv, (uint8_t *) ref, (uint8_t *) refh, |
|
|
(uint8_t *) refh + 64, (uint8_t *) refhv, x, y, dx, dy, stride, rounding); |
|
96 |
|
|
97 |
|
case 2: |
98 |
|
ddx = (dx - 1) / 2; |
99 |
|
ddy = dy / 2; |
100 |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
transfer_8to16sub(dct_codes, cur + y*stride + x, |
101 |
refv + y*stride + x, stride); |
refh + (int) ((y + ddy) * stride + x + ddx), stride); |
102 |
} |
break; |
|
else |
|
|
{ |
|
|
const uint8_t * reference = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride); |
|
103 |
|
|
104 |
|
default: /* case 3: */ |
105 |
|
ddx = (dx - 1) / 2; |
106 |
|
ddy = (dy - 1) / 2; |
107 |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
transfer_8to16sub(dct_codes, cur + y * stride + x, |
108 |
reference, stride); |
refhv + (int) ((y + ddy) * stride + x + ddx), stride); |
109 |
} |
break; |
110 |
} |
} |
111 |
} |
} |
112 |
|
|
113 |
|
|
114 |
|
|
|
/* XXX: slow, inelegant... */ |
|
|
static void |
|
|
interpolate18x18_switch(uint8_t * const cur, |
|
|
const uint8_t * const refn, |
|
|
const uint32_t x, |
|
|
const uint32_t y, |
|
|
const int32_t dx, |
|
|
const int dy, |
|
|
const uint32_t stride, |
|
|
const uint32_t rounding) |
|
|
{ |
|
|
interpolate8x8_switch(cur, refn, x-1, y-1, dx, dy, stride, rounding); |
|
|
interpolate8x8_switch(cur, refn, x+7, y-1, dx, dy, stride, rounding); |
|
|
interpolate8x8_switch(cur, refn, x+9, y-1, dx, dy, stride, rounding); |
|
|
|
|
|
interpolate8x8_switch(cur, refn, x-1, y+7, dx, dy, stride, rounding); |
|
|
interpolate8x8_switch(cur, refn, x+7, y+7, dx, dy, stride, rounding); |
|
|
interpolate8x8_switch(cur, refn, x+9, y+7, dx, dy, stride, rounding); |
|
|
|
|
|
interpolate8x8_switch(cur, refn, x-1, y+9, dx, dy, stride, rounding); |
|
|
interpolate8x8_switch(cur, refn, x+7, y+9, dx, dy, stride, rounding); |
|
|
interpolate8x8_switch(cur, refn, x+9, y+9, dx, dy, stride, rounding); |
|
|
} |
|
|
|
|
|
|
|
115 |
void |
void |
116 |
MBMotionCompensation(MACROBLOCK * const mb, |
MBMotionCompensation(MACROBLOCK * const mb, |
117 |
const uint32_t i, |
const uint32_t i, |
125 |
const uint32_t width, |
const uint32_t width, |
126 |
const uint32_t height, |
const uint32_t height, |
127 |
const uint32_t edged_width, |
const uint32_t edged_width, |
|
const uint32_t quarterpel, |
|
|
const int reduced_resolution, |
|
128 |
const uint32_t rounding) |
const uint32_t rounding) |
129 |
{ |
{ |
130 |
|
static const uint32_t roundtab[16] = |
131 |
|
{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; |
132 |
|
|
|
if (mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
|
|
|
|
|
int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x); |
|
|
int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y); |
|
|
|
|
|
if ( (!reduced_resolution) && (mb->mode == MODE_NOT_CODED) && (dx==0) && (dy==0) ) { /* quick copy */ |
|
|
transfer16x16_copy(cur->y + 16 * (i + j * edged_width), |
|
|
ref->y + 16 * (i + j * edged_width), |
|
|
edged_width); |
|
133 |
|
|
134 |
transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2), |
if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { |
135 |
ref->u + 8 * (i + j * edged_width/2), |
int32_t dx = mb->mvs[0].x; |
136 |
edged_width / 2); |
int32_t dy = mb->mvs[0].y; |
|
transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2), |
|
|
ref->v + 8 * (i + j * edged_width/2), |
|
|
edged_width / 2); |
|
|
return; |
|
|
} |
|
|
/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */ |
|
137 |
|
|
138 |
if (reduced_resolution) |
compensate8x8_halfpel(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
|
{ |
|
|
dx = RRV_MV_SCALEUP(dx); |
|
|
dy = RRV_MV_SCALEUP(dy); |
|
|
} |
|
|
|
|
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
|
139 |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
refv->y, refhv->y, 16 * i, 16 * j, dx, dy, |
140 |
edged_width, quarterpel, reduced_resolution, rounding); |
edged_width); |
141 |
|
compensate8x8_halfpel(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
142 |
if (quarterpel) |
refv->y, refhv->y, 16 * i + 8, 16 * j, dx, dy, |
143 |
{ |
edged_width); |
144 |
dx /= 2; |
compensate8x8_halfpel(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
145 |
dy /= 2; |
refv->y, refhv->y, 16 * i, 16 * j + 8, dx, dy, |
146 |
} |
edged_width); |
147 |
|
compensate8x8_halfpel(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
148 |
|
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, dx, |
149 |
|
dy, edged_width); |
150 |
|
|
151 |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2; |
152 |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2; |
153 |
|
|
154 |
/* uv-block-based compensation */ |
/* uv-image-based compensation */ |
|
if (reduced_resolution) |
|
|
{ |
|
|
const stride = edged_width/2; |
|
|
uint8_t * current, * reference; |
|
155 |
|
|
156 |
current = cur->u + 16*j*stride + 16*i; |
interpolate8x8_switch(refv->u, ref->u, 8 * i, 8 * j, dx, dy, |
157 |
reference = refv->u + 16*j*stride + 16*i; |
edged_width / 2, rounding); |
|
interpolate18x18_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
|
|
filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); |
|
|
transfer16x16_copy(current, reference, stride); |
|
|
|
|
|
current = cur->v + 16*j*stride + 16*i; |
|
|
reference = refv->v + 16*j*stride + 16*i; |
|
|
interpolate18x18_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
|
|
filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); |
|
|
transfer16x16_copy(current, reference, stride); |
|
|
}else{ |
|
158 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
159 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
160 |
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
refv->u + 8 * j * edged_width / 2 + 8 * i, |
|
dx, dy, edged_width / 2, rounding), |
|
161 |
edged_width / 2); |
edged_width / 2); |
162 |
|
|
163 |
|
interpolate8x8_switch(refv->v, ref->v, 8 * i, 8 * j, dx, dy, |
164 |
|
edged_width / 2, rounding); |
165 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
166 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
167 |
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
refv->v + 8 * j * edged_width / 2 + 8 * i, |
|
dx, dy, edged_width / 2, rounding), |
|
168 |
edged_width / 2); |
edged_width / 2); |
|
} |
|
|
|
|
|
} else { // mode == MODE_INTER4V |
|
|
int k; |
|
|
int32_t sum, dx, dy; |
|
|
VECTOR mvs[4]; |
|
169 |
|
|
170 |
if(quarterpel) |
} else /* mode == MODE_INTER4V */ |
|
for (k = 0; k < 4; k++) mvs[k] = mb->qmvs[k]; |
|
|
else |
|
|
for (k = 0; k < 4; k++) mvs[k] = mb->mvs[k]; |
|
|
|
|
|
if (reduced_resolution) |
|
|
{ |
|
|
for (k = 0; k < 4; k++) |
|
171 |
{ |
{ |
172 |
mvs[k].x = RRV_MV_SCALEUP(mvs[k].x); |
int32_t sum, dx, dy; |
|
mvs[k].y = RRV_MV_SCALEUP(mvs[k].y); |
|
|
} |
|
|
} |
|
173 |
|
|
174 |
compensate8x8_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
compensate8x8_halfpel(&dct_codes[0 * 64], cur->y, ref->y, refh->y, |
175 |
refv->y, refhv->y, 16 * i, 16 * j, mvs[0].x, |
refv->y, refhv->y, 16 * i, 16 * j, mb->mvs[0].x, |
176 |
mvs[0].y, edged_width, quarterpel, reduced_resolution, rounding); |
mb->mvs[0].y, edged_width); |
177 |
compensate8x8_interpolate(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
compensate8x8_halfpel(&dct_codes[1 * 64], cur->y, ref->y, refh->y, |
178 |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
refv->y, refhv->y, 16 * i + 8, 16 * j, |
179 |
mvs[1].x, mvs[1].y, edged_width, quarterpel, reduced_resolution, rounding); |
mb->mvs[1].x, mb->mvs[1].y, edged_width); |
180 |
compensate8x8_interpolate(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
compensate8x8_halfpel(&dct_codes[2 * 64], cur->y, ref->y, refh->y, |
181 |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
refv->y, refhv->y, 16 * i, 16 * j + 8, |
182 |
mvs[2].x, mvs[2].y, edged_width, quarterpel, reduced_resolution, rounding); |
mb->mvs[2].x, mb->mvs[2].y, edged_width); |
183 |
compensate8x8_interpolate(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
compensate8x8_halfpel(&dct_codes[3 * 64], cur->y, ref->y, refh->y, |
184 |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
refv->y, refhv->y, 16 * i + 8, 16 * j + 8, |
185 |
mvs[3].x, mvs[3].y, edged_width, quarterpel, reduced_resolution, rounding); |
mb->mvs[3].x, mb->mvs[3].y, edged_width); |
|
|
|
|
if(quarterpel) |
|
|
sum = (mvs[0].x / 2) + (mvs[1].x / 2) + (mvs[2].x / 2) + (mvs[3].x / 2); |
|
|
else |
|
|
sum = mvs[0].x + mvs[1].x + mvs[2].x + mvs[3].x; |
|
|
|
|
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
|
|
|
if(quarterpel) |
|
|
sum = (mvs[0].y / 2) + (mvs[1].y / 2) + (mvs[2].y / 2) + (mvs[3].y / 2); |
|
|
else |
|
|
sum = mvs[0].y + mvs[1].y + mvs[2].y + mvs[3].y; |
|
|
|
|
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
|
|
|
|
|
|
/* uv-block-based compensation */ |
|
|
if (reduced_resolution) |
|
|
{ |
|
|
const stride = edged_width/2; |
|
|
uint8_t * current, * reference; |
|
|
|
|
|
current = cur->u + 16*j*stride + 16*i; |
|
|
reference = refv->u + 16*j*stride + 16*i; |
|
|
interpolate18x18_switch(refv->u, ref->u, 16*i, 16*j, dx, dy, stride, rounding); |
|
|
filter_18x18_to_8x8(dct_codes + 4*64, current, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes + 4*64, reference, stride); |
|
|
transfer16x16_copy(current, reference, stride); |
|
|
|
|
|
current = cur->v + 16*j*stride + 16*i; |
|
|
reference = refv->v + 16*j*stride + 16*i; |
|
|
interpolate18x18_switch(refv->v, ref->v, 16*i, 16*j, dx, dy, stride, rounding); |
|
|
filter_18x18_to_8x8(dct_codes + 5*64, current, stride); |
|
|
filter_diff_18x18_to_8x8(dct_codes + 5*64, reference, stride); |
|
|
transfer16x16_copy(current, reference, stride); |
|
|
|
|
|
}else{ |
|
|
transfer_8to16sub(&dct_codes[4 * 64], |
|
|
cur->u + 8 * j * edged_width / 2 + 8 * i, |
|
|
interpolate8x8_switch2(refv->u, ref->u, 8 * i, 8 * j, |
|
|
dx, dy, edged_width / 2, rounding), |
|
|
edged_width / 2); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[5 * 64], |
|
|
cur->v + 8 * j * edged_width / 2 + 8 * i, |
|
|
interpolate8x8_switch2(refv->u, ref->v, 8 * i, 8 * j, |
|
|
dx, dy, edged_width / 2, rounding), |
|
|
edged_width / 2); |
|
|
} |
|
|
} |
|
|
} |
|
|
|
|
|
|
|
|
void |
|
|
MBMotionCompensationBVOP(MBParam * pParam, |
|
|
MACROBLOCK * const mb, |
|
|
const uint32_t i, |
|
|
const uint32_t j, |
|
|
IMAGE * const cur, |
|
|
const IMAGE * const f_ref, |
|
|
const IMAGE * const f_refh, |
|
|
const IMAGE * const f_refv, |
|
|
const IMAGE * const f_refhv, |
|
|
const IMAGE * const b_ref, |
|
|
const IMAGE * const b_refh, |
|
|
const IMAGE * const b_refv, |
|
|
const IMAGE * const b_refhv, |
|
|
int16_t * dct_codes) |
|
|
{ |
|
|
static const uint32_t roundtab[16] = |
|
|
{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; |
|
|
|
|
|
const int32_t edged_width = pParam->edged_width; |
|
|
int32_t dx, dy; |
|
|
int32_t b_dx, b_dy; |
|
|
int k,sum; |
|
|
int x = i; |
|
|
int y = j; |
|
|
uint32_t quarterpel = pParam->m_quarterpel; |
|
|
|
|
|
switch (mb->mode) { |
|
|
case MODE_FORWARD: |
|
|
|
|
|
if (quarterpel) { |
|
|
dx = mb->qmvs[0].x; |
|
|
dy = mb->qmvs[0].y; |
|
|
} else { |
|
|
dx = mb->mvs[0].x; |
|
|
dy = mb->mvs[0].y; |
|
|
} |
|
|
|
|
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y, |
|
|
f_refv->y, f_refhv->y, 16 * i, 16 * j, dx, |
|
|
dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
|
|
|
|
|
if (quarterpel) { |
|
|
dx /= 2; |
|
|
dy /= 2; |
|
|
} |
|
|
|
|
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
|
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
|
|
|
|
|
/* uv-block-based compensation */ |
|
|
transfer_8to16sub(&dct_codes[4 * 64], |
|
|
cur->u + 8 * j * edged_width / 2 + 8 * i, |
|
|
interpolate8x8_switch2(f_refv->u, f_ref->u, 8 * i, 8 * j, |
|
|
dx, dy, edged_width / 2, 0), |
|
|
|
|
|
edged_width / 2); |
|
|
|
|
|
transfer_8to16sub(&dct_codes[5 * 64], |
|
|
cur->v + 8 * j * edged_width / 2 + 8 * i, |
|
|
interpolate8x8_switch2(f_refv->u, f_ref->v, 8 * i, 8 * j, |
|
|
dx, dy, edged_width / 2, 0), |
|
|
|
|
|
edged_width / 2); |
|
|
|
|
|
break; |
|
|
|
|
|
case MODE_BACKWARD: |
|
|
if (quarterpel) { |
|
|
b_dx = mb->b_qmvs[0].x; |
|
|
b_dy = mb->b_qmvs[0].y; |
|
|
} else { |
|
|
b_dx = mb->b_mvs[0].x; |
|
|
b_dy = mb->b_mvs[0].y; |
|
|
} |
|
|
|
|
|
compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y, |
|
|
b_refv->y, b_refhv->y, 16 * i, 16 * j, b_dx, |
|
|
b_dy, edged_width, quarterpel, 0 /*reduced_resolution*/, 0); |
|
|
|
|
|
if (quarterpel) { |
|
|
b_dx /= 2; |
|
|
b_dy /= 2; |
|
|
} |
|
186 |
|
|
187 |
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
188 |
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
dx = (sum ? SIGN(sum) * |
189 |
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
190 |
|
|
191 |
|
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
192 |
|
dy = (sum ? SIGN(sum) * |
193 |
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
194 |
|
|
195 |
/* uv-block-based compensation */ |
/* uv-block-based compensation */ |
196 |
|
interpolate8x8_switch(refv->u, ref->u, 8 * i, 8 * j, dx, dy, |
197 |
|
edged_width / 2, rounding); |
198 |
transfer_8to16sub(&dct_codes[4 * 64], |
transfer_8to16sub(&dct_codes[4 * 64], |
199 |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
cur->u + 8 * j * edged_width / 2 + 8 * i, |
200 |
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
refv->u + 8 * j * edged_width / 2 + 8 * i, |
|
b_dx, b_dy, edged_width / 2, 0), |
|
|
|
|
201 |
edged_width / 2); |
edged_width / 2); |
202 |
|
|
203 |
|
interpolate8x8_switch(refv->v, ref->v, 8 * i, 8 * j, dx, dy, |
204 |
|
edged_width / 2, rounding); |
205 |
transfer_8to16sub(&dct_codes[5 * 64], |
transfer_8to16sub(&dct_codes[5 * 64], |
206 |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
cur->v + 8 * j * edged_width / 2 + 8 * i, |
207 |
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
refv->v + 8 * j * edged_width / 2 + 8 * i, |
|
b_dx, b_dy, edged_width / 2, 0), |
|
|
|
|
|
edged_width / 2); |
|
|
|
|
|
break; |
|
|
|
|
|
case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */ |
|
|
case MODE_DIRECT_NO4V: |
|
|
|
|
|
if (quarterpel) { |
|
|
dx = mb->qmvs[0].x; |
|
|
dy = mb->qmvs[0].y; |
|
|
b_dx = mb->b_qmvs[0].x; |
|
|
b_dy = mb->b_qmvs[0].y; |
|
|
|
|
|
interpolate16x16_quarterpel((uint8_t *) f_refv->y, (uint8_t *) f_ref->y, (uint8_t *) f_refh->y, |
|
|
(uint8_t *) f_refh->y + 64, (uint8_t *) f_refhv->y, 16*i, 16*j, dx, dy, edged_width, 0); |
|
|
interpolate16x16_quarterpel((uint8_t *) b_refv->y, (uint8_t *) b_ref->y, (uint8_t *) b_refh->y, |
|
|
(uint8_t *) b_refh->y + 64, (uint8_t *) b_refhv->y, 16*i, 16*j, b_dx, b_dy, edged_width, 0); |
|
|
|
|
|
for (k = 0; k < 4; k++) { |
|
|
transfer_8to16sub2(&dct_codes[k * 64], |
|
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
|
|
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
|
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
|
|
edged_width); |
|
|
} |
|
|
b_dx /= 2; |
|
|
b_dy /= 2; |
|
|
dx /= 2; |
|
|
dy /= 2; |
|
|
|
|
|
} else { |
|
|
dx = mb->mvs[0].x; |
|
|
dy = mb->mvs[0].y; |
|
|
b_dx = mb->b_mvs[0].x; |
|
|
b_dy = mb->b_mvs[0].y; |
|
|
|
|
|
for (k = 0; k < 4; k++) { |
|
|
transfer_8to16sub2(&dct_codes[k * 64], |
|
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
|
|
get_ref(f_ref->y, f_refh->y, f_refv->y, |
|
|
f_refhv->y, 2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
|
|
edged_width), |
|
|
get_ref(b_ref->y, b_refh->y, b_refv->y, |
|
|
b_refhv->y, 2*i + (k&1), 2 * j+(k>>1), 8, b_dx, b_dy, |
|
|
edged_width), |
|
|
edged_width); |
|
|
} |
|
|
|
|
|
} |
|
|
|
|
|
|
|
|
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
|
|
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
|
|
|
|
|
b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3]; |
|
|
b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3]; |
|
|
|
|
|
transfer_8to16sub2(&dct_codes[4 * 64], |
|
|
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
|
|
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
|
|
b_dx, b_dy, edged_width / 2, 0), |
|
|
interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, |
|
|
dx, dy, edged_width / 2, 0), |
|
|
edged_width / 2); |
|
|
|
|
|
transfer_8to16sub2(&dct_codes[5 * 64], |
|
|
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
|
|
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
|
|
b_dx, b_dy, edged_width / 2, 0), |
|
|
interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, |
|
|
dx, dy, edged_width / 2, 0), |
|
208 |
edged_width / 2); |
edged_width / 2); |
|
|
|
|
break; |
|
|
|
|
|
case MODE_DIRECT: |
|
|
if (quarterpel) { |
|
|
for (k=0;k<4;k++) { |
|
|
|
|
|
dx = mb->qmvs[k].x; |
|
|
dy = mb->qmvs[k].y; |
|
|
b_dx = mb->b_qmvs[k].x; |
|
|
b_dy = mb->b_qmvs[k].y; |
|
|
|
|
|
interpolate8x8_quarterpel((uint8_t *) f_refv->y, |
|
|
(uint8_t *) f_ref->y, |
|
|
(uint8_t *) f_refh->y, |
|
|
(uint8_t *) f_refh->y + 64, |
|
|
(uint8_t *) f_refhv->y, |
|
|
16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0); |
|
|
interpolate8x8_quarterpel((uint8_t *) b_refv->y, |
|
|
(uint8_t *) b_ref->y, |
|
|
(uint8_t *) b_refh->y, |
|
|
(uint8_t *) b_refh->y + 64, |
|
|
(uint8_t *) b_refhv->y, |
|
|
16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0); |
|
|
|
|
|
|
|
|
transfer_8to16sub2(&dct_codes[k * 64], |
|
|
cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
|
|
f_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
|
|
b_refv->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width, |
|
|
edged_width); |
|
|
} |
|
|
sum = mb->qmvs[0].y/2 + mb->qmvs[1].y/2 + mb->qmvs[2].y/2 + mb->qmvs[3].y/2; |
|
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
sum = mb->qmvs[0].x/2 + mb->qmvs[1].x/2 + mb->qmvs[2].x/2 + mb->qmvs[3].x/2; |
|
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
|
|
|
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; |
|
|
b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
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; |
|
|
b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
|
|
|
} else { |
|
|
for (k=0;k<4;k++) { |
|
|
dx = mb->mvs[k].x; |
|
|
dy = mb->mvs[k].y; |
|
|
|
|
|
b_dx = mb->b_mvs[k].x; |
|
|
b_dy = mb->b_mvs[k].y; |
|
|
|
|
|
transfer_8to16sub2(&dct_codes[k * 64], |
|
|
cur->y + (i*16 + (k&1)*8) + (j*16 + (k>>1)*8 ) * edged_width, |
|
|
get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, |
|
|
2*i + (k&1), 2*j + (k>>1), 8, dx, dy, |
|
|
edged_width), |
|
|
get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, |
|
|
2*i + (k&1), 2*j + (k>>1), 8, b_dx, b_dy, |
|
|
edged_width), |
|
|
edged_width); |
|
|
} |
|
|
|
|
|
sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; |
|
|
dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
|
|
|
sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; |
|
|
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
|
|
|
sum = mb->b_mvs[0].x + mb->b_mvs[1].x + mb->b_mvs[2].x + mb->b_mvs[3].x; |
|
|
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
|
|
|
sum = mb->b_mvs[0].y + mb->b_mvs[1].y + mb->b_mvs[2].y + mb->b_mvs[3].y; |
|
|
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
|
|
|
} |
|
|
transfer_8to16sub2(&dct_codes[4 * 64], |
|
|
cur->u + (y * 8) * edged_width / 2 + (x * 8), |
|
|
interpolate8x8_switch2(f_refv->u, b_ref->u, 8 * i, 8 * j, |
|
|
b_dx, b_dy, edged_width / 2, 0), |
|
|
interpolate8x8_switch2(f_refv->u + 8, f_ref->u, 8 * i, 8 * j, dx, dy, |
|
|
edged_width / 2, 0), |
|
|
edged_width / 2); |
|
|
|
|
|
transfer_8to16sub2(&dct_codes[5 * 64], |
|
|
cur->v + (y * 8) * edged_width / 2 + (x * 8), |
|
|
interpolate8x8_switch2(f_refv->u, b_ref->v, 8 * i, 8 * j, |
|
|
b_dx, b_dy, edged_width / 2, 0), |
|
|
interpolate8x8_switch2(f_refv->u + 8, f_ref->v, 8 * i, 8 * j, |
|
|
dx, dy, edged_width / 2, 0), |
|
|
edged_width / 2); |
|
|
|
|
|
|
|
|
break; |
|
209 |
} |
} |
210 |
} |
} |