1 |
/************************************************************************** |
/***************************************************************************** |
2 |
* |
* |
3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* motion estimation |
* - Motion Estimation related code - |
5 |
* |
* |
6 |
* This program is an implementation of a part of one or more MPEG-4 |
* Copyright(C) 2002 Christoph Lampert <gruel@web.de> |
7 |
* Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
* 2002 Michael Militzer <michael@xvid.org> |
8 |
* to use this software module in hardware or software products are |
* 2002-2003 Radoslaw Czyz <xvid@syskin.cjb.net> |
|
* advised that its use may infringe existing patents or copyrights, and |
|
|
* any such use would be at such party's own risk. The original |
|
|
* developer of this software module and his/her company, and subsequent |
|
|
* editors and their companies, will have no liability for use of this |
|
|
* software or modifications or derivatives thereof. |
|
9 |
* |
* |
10 |
* This program is free software; you can redistribute it and/or modify |
* This program is free software; you can redistribute it and/or modify |
11 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
19 |
* |
* |
20 |
* You should have received a copy of the GNU General Public License |
* You should have received a copy of the GNU General Public License |
21 |
* along with this program; if not, write to the Free Software |
* along with this program; if not, write to the Free Software |
22 |
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
|
* |
24 |
|
* $Id: motion_est.c,v 1.58.2.17 2003-06-09 13:54:37 edgomez Exp $ |
25 |
* |
* |
26 |
*************************************************************************/ |
****************************************************************************/ |
27 |
|
|
28 |
#include <assert.h> |
#include <assert.h> |
29 |
#include <stdio.h> |
#include <stdio.h> |
30 |
#include <stdlib.h> |
#include <stdlib.h> |
31 |
#include <string.h> // memcpy |
#include <string.h> /* memcpy */ |
32 |
#include <math.h> // lrint |
#include <math.h> /* lrint */ |
33 |
|
|
34 |
#include "../encoder.h" |
#include "../encoder.h" |
35 |
#include "../utils/mbfunctions.h" |
#include "../utils/mbfunctions.h" |
78 |
static __inline uint32_t |
static __inline uint32_t |
79 |
d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv) |
d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv) |
80 |
{ |
{ |
81 |
int xb, yb; |
int bits; |
82 |
x = qpel ? x<<1 : x; |
const int q = (1 << (iFcode - 1)) - 1; |
83 |
y = qpel ? y<<1 : y; |
|
84 |
|
x <<= qpel; |
85 |
|
y <<= qpel; |
86 |
if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
87 |
|
|
88 |
x -= pred.x; |
x -= pred.x; |
89 |
y -= pred.y; |
bits = (x != 0 ? iFcode:0); |
90 |
|
x = abs(x); |
91 |
if (x) { |
x += q; |
|
x = ABS(x); |
|
|
x += (1 << (iFcode - 1)) - 1; |
|
92 |
x >>= (iFcode - 1); |
x >>= (iFcode - 1); |
93 |
if (x > 32) x = 32; |
bits += mvtab[x]; |
94 |
xb = mvtab[x] + iFcode; |
|
95 |
} else xb = 1; |
y -= pred.y; |
96 |
|
bits += (y != 0 ? iFcode:0); |
97 |
if (y) { |
y = abs(y); |
98 |
y = ABS(y); |
y += q; |
|
y += (1 << (iFcode - 1)) - 1; |
|
99 |
y >>= (iFcode - 1); |
y >>= (iFcode - 1); |
100 |
if (y > 32) y = 32; |
bits += mvtab[y]; |
101 |
yb = mvtab[y] + iFcode; |
|
102 |
} else yb = 1; |
return bits; |
|
return xb + yb; |
|
103 |
} |
} |
104 |
|
|
105 |
static int32_t ChromaSAD2(int fx, int fy, int bx, int by, const SearchData * const data) |
static int32_t ChromaSAD2(const int fx, const int fy, const int bx, const int by, |
106 |
|
const SearchData * const data) |
107 |
{ |
{ |
108 |
int sad; |
int sad; |
109 |
const uint32_t stride = data->iEdgedWidth/2; |
const uint32_t stride = data->iEdgedWidth/2; |
111 |
* f_refv = data->RefQ + 8, |
* f_refv = data->RefQ + 8, |
112 |
* b_refu = data->RefQ + 16, |
* b_refu = data->RefQ + 16, |
113 |
* b_refv = data->RefQ + 24; |
* b_refv = data->RefQ + 24; |
114 |
|
int offset = (fx>>1) + (fy>>1)*stride; |
115 |
|
|
116 |
switch (((fx & 1) << 1) | (fy & 1)) { |
switch (((fx & 1) << 1) | (fy & 1)) { |
117 |
case 0: |
case 0: |
118 |
fx = fx / 2; fy = fy / 2; |
f_refu = (uint8_t*)data->RefP[4] + offset; |
119 |
f_refu = (uint8_t*)data->RefCU + fy * stride + fx, stride; |
f_refv = (uint8_t*)data->RefP[5] + offset; |
|
f_refv = (uint8_t*)data->RefCV + fy * stride + fx, stride; |
|
120 |
break; |
break; |
121 |
case 1: |
case 1: |
122 |
fx = fx / 2; fy = (fy - 1) / 2; |
interpolate8x8_halfpel_v(f_refu, data->RefP[4] + offset, stride, data->rounding); |
123 |
interpolate8x8_halfpel_v(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
interpolate8x8_halfpel_v(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_v(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
|
124 |
break; |
break; |
125 |
case 2: |
case 2: |
126 |
fx = (fx - 1) / 2; fy = fy / 2; |
interpolate8x8_halfpel_h(f_refu, data->RefP[4] + offset, stride, data->rounding); |
127 |
interpolate8x8_halfpel_h(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
interpolate8x8_halfpel_h(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_h(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
|
128 |
break; |
break; |
129 |
default: |
default: |
130 |
fx = (fx - 1) / 2; fy = (fy - 1) / 2; |
interpolate8x8_halfpel_hv(f_refu, data->RefP[4] + offset, stride, data->rounding); |
131 |
interpolate8x8_halfpel_hv(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
interpolate8x8_halfpel_hv(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_hv(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
|
132 |
break; |
break; |
133 |
} |
} |
134 |
|
|
135 |
|
offset = (bx>>1) + (by>>1)*stride; |
136 |
switch (((bx & 1) << 1) | (by & 1)) { |
switch (((bx & 1) << 1) | (by & 1)) { |
137 |
case 0: |
case 0: |
138 |
bx = bx / 2; by = by / 2; |
b_refu = (uint8_t*)data->b_RefP[4] + offset; |
139 |
b_refu = (uint8_t*)data->b_RefCU + by * stride + bx, stride; |
b_refv = (uint8_t*)data->b_RefP[5] + offset; |
|
b_refv = (uint8_t*)data->b_RefCV + by * stride + bx, stride; |
|
140 |
break; |
break; |
141 |
case 1: |
case 1: |
142 |
bx = bx / 2; by = (by - 1) / 2; |
interpolate8x8_halfpel_v(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
143 |
interpolate8x8_halfpel_v(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
interpolate8x8_halfpel_v(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_v(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
|
144 |
break; |
break; |
145 |
case 2: |
case 2: |
146 |
bx = (bx - 1) / 2; by = by / 2; |
interpolate8x8_halfpel_h(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
147 |
interpolate8x8_halfpel_h(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
interpolate8x8_halfpel_h(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_h(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
|
148 |
break; |
break; |
149 |
default: |
default: |
150 |
bx = (bx - 1) / 2; by = (by - 1) / 2; |
interpolate8x8_halfpel_hv(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
151 |
interpolate8x8_halfpel_hv(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
interpolate8x8_halfpel_hv(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_hv(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
|
152 |
break; |
break; |
153 |
} |
} |
154 |
|
|
158 |
return sad; |
return sad; |
159 |
} |
} |
160 |
|
|
|
|
|
161 |
static int32_t |
static int32_t |
162 |
ChromaSAD(int dx, int dy, const SearchData * const data) |
ChromaSAD(const int dx, const int dy, const SearchData * const data) |
163 |
{ |
{ |
164 |
int sad; |
int sad; |
165 |
const uint32_t stride = data->iEdgedWidth/2; |
const uint32_t stride = data->iEdgedWidth/2; |
166 |
|
int offset = (dx>>1) + (dy>>1)*stride; |
167 |
|
|
168 |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; /* it has been checked recently */ |
169 |
data->temp[5] = dx; data->temp[6] = dy; // backup |
data->temp[5] = dx; data->temp[6] = dy; /* backup */ |
170 |
|
|
171 |
switch (((dx & 1) << 1) | (dy & 1)) { |
switch (((dx & 1) << 1) | (dy & 1)) { |
172 |
case 0: |
case 0: |
173 |
dx = dx / 2; dy = dy / 2; |
sad = sad8(data->CurU, data->RefP[4] + offset, stride); |
174 |
sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride); |
sad += sad8(data->CurV, data->RefP[5] + offset, stride); |
|
sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride); |
|
175 |
break; |
break; |
176 |
case 1: |
case 1: |
177 |
dx = dx / 2; dy = (dy - 1) / 2; |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + stride, stride); |
178 |
sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride); |
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + stride, stride); |
|
sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride); |
|
179 |
break; |
break; |
180 |
case 2: |
case 2: |
181 |
dx = (dx - 1) / 2; dy = dy / 2; |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + 1, stride); |
182 |
sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride); |
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + 1, stride); |
|
sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride); |
|
183 |
break; |
break; |
184 |
default: |
default: |
185 |
dx = (dx - 1) / 2; dy = (dy - 1) / 2; |
interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding); |
|
186 |
sad = sad8(data->CurU, data->RefQ, stride); |
sad = sad8(data->CurU, data->RefQ, stride); |
187 |
|
|
188 |
interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding); |
interpolate8x8_halfpel_hv(data->RefQ, data->RefP[5] + offset, stride, data->rounding); |
189 |
sad += sad8(data->CurV, data->RefQ, stride); |
sad += sad8(data->CurV, data->RefQ, stride); |
190 |
break; |
break; |
191 |
} |
} |
192 |
data->temp[7] = sad; //backup, part 2 |
data->temp[7] = sad; /* backup, part 2 */ |
193 |
return sad; |
return sad; |
194 |
} |
} |
195 |
|
|
196 |
static __inline const uint8_t * |
static __inline const uint8_t * |
197 |
GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data) |
GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data) |
198 |
{ |
{ |
199 |
// dir : 0 = forward, 1 = backward |
/* dir : 0 = forward, 1 = backward */ |
200 |
switch ( (dir << 2) | ((x&1)<<1) | (y&1) ) { |
const uint8_t *const *const direction = ( dir == 0 ? data->RefP : data->b_RefP ); |
201 |
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
const int picture = ((x&1)<<1) | (y&1); |
202 |
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
const int offset = (x>>1) + (y>>1)*data->iEdgedWidth; |
203 |
case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
return direction[picture] + offset; |
|
case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth); |
|
|
case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
|
|
default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
} |
|
204 |
} |
} |
205 |
|
|
206 |
// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate |
/* this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate */ |
207 |
static __inline const uint8_t * |
static __inline const uint8_t * |
208 |
GetReference(const int x, const int y, const SearchData * const data) |
GetReference(const int x, const int y, const SearchData * const data) |
209 |
{ |
{ |
210 |
switch ( ((x&1)<<1) | (y&1) ) { |
const int picture = ((x&1)<<1) | (y&1); |
211 |
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
const int offset = (x>>1) + (y>>1)*data->iEdgedWidth; |
212 |
case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
return data->RefP[picture] + offset; |
|
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
default : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); //case 2 |
|
|
} |
|
213 |
} |
} |
214 |
|
|
215 |
static uint8_t * |
static uint8_t * |
216 |
Interpolate8x8qpel(const int x, const int y, const uint32_t block, const uint32_t dir, const SearchData * const data) |
Interpolate8x8qpel(const int x, const int y, const uint32_t block, const uint32_t dir, const SearchData * const data) |
217 |
{ |
{ |
218 |
// create or find a qpel-precision reference picture; return pointer to it |
/* create or find a qpel-precision reference picture; return pointer to it */ |
219 |
uint8_t * Reference = data->RefQ + 16*dir; |
uint8_t * Reference = data->RefQ + 16*dir; |
220 |
const uint32_t iEdgedWidth = data->iEdgedWidth; |
const uint32_t iEdgedWidth = data->iEdgedWidth; |
221 |
const uint32_t rounding = data->rounding; |
const uint32_t rounding = data->rounding; |
226 |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
227 |
ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
228 |
switch( ((x&1)<<1) + (y&1) ) { |
switch( ((x&1)<<1) + (y&1) ) { |
229 |
case 0: // pure halfpel position |
case 3: /* x and y in qpel resolution - the "corners" (top left/right and */ |
230 |
return (uint8_t *) ref1; |
/* bottom left/right) during qpel refinement */ |
231 |
|
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
232 |
|
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
233 |
|
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
234 |
|
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
235 |
|
ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
236 |
|
ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
237 |
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
238 |
break; |
break; |
239 |
|
|
240 |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
case 1: /* x halfpel, y qpel - top or bottom during qpel refinement */ |
241 |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
242 |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
243 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
244 |
break; |
break; |
245 |
|
|
246 |
case 2: // x qpel, y halfpel - left or right during qpel refinement |
case 2: /* x qpel, y halfpel - left or right during qpel refinement */ |
247 |
ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
248 |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
249 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
250 |
break; |
break; |
251 |
|
|
252 |
default: // x and y in qpel resolution - the "corners" (top left/right and |
default: /* pure halfpel position */ |
253 |
// bottom left/right) during qpel refinement |
return (uint8_t *) ref1; |
254 |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
|
|
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
|
|
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
|
|
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
|
|
ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
|
|
ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
|
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
|
|
break; |
|
255 |
} |
} |
256 |
return Reference; |
return Reference; |
257 |
} |
} |
259 |
static uint8_t * |
static uint8_t * |
260 |
Interpolate16x16qpel(const int x, const int y, const uint32_t dir, const SearchData * const data) |
Interpolate16x16qpel(const int x, const int y, const uint32_t dir, const SearchData * const data) |
261 |
{ |
{ |
262 |
// create or find a qpel-precision reference picture; return pointer to it |
/* create or find a qpel-precision reference picture; return pointer to it */ |
263 |
uint8_t * Reference = data->RefQ + 16*dir; |
uint8_t * Reference = data->RefQ + 16*dir; |
264 |
const uint32_t iEdgedWidth = data->iEdgedWidth; |
const uint32_t iEdgedWidth = data->iEdgedWidth; |
265 |
const uint32_t rounding = data->rounding; |
const uint32_t rounding = data->rounding; |
269 |
|
|
270 |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
271 |
switch( ((x&1)<<1) + (y&1) ) { |
switch( ((x&1)<<1) + (y&1) ) { |
272 |
case 3: // x and y in qpel resolution - the "corners" (top left/right and |
case 3: |
273 |
// bottom left/right) during qpel refinement |
/* |
274 |
|
* x and y in qpel resolution - the "corners" (top left/right and |
275 |
|
* bottom left/right) during qpel refinement |
276 |
|
*/ |
277 |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
278 |
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
279 |
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
283 |
interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); |
interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); |
284 |
break; |
break; |
285 |
|
|
286 |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
case 1: /* x halfpel, y qpel - top or bottom during qpel refinement */ |
287 |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
288 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
289 |
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
291 |
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
292 |
break; |
break; |
293 |
|
|
294 |
case 2: // x qpel, y halfpel - left or right during qpel refinement |
case 2: /* x qpel, y halfpel - left or right during qpel refinement */ |
295 |
ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
296 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
297 |
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
299 |
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
300 |
break; |
break; |
301 |
|
|
302 |
case 0: // pure halfpel position |
default: /* pure halfpel position */ |
303 |
return (uint8_t *) ref1; |
return (uint8_t *) ref1; |
304 |
} |
} |
305 |
return Reference; |
return Reference; |
322 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
323 |
current = data->currentMV; |
current = data->currentMV; |
324 |
xc = x; yc = y; |
xc = x; yc = y; |
325 |
} else { // x and y are in 1/4 precision |
} else { /* x and y are in 1/4 precision */ |
326 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
327 |
xc = x/2; yc = y/2; //for chroma sad |
xc = x/2; yc = y/2; /* for chroma sad */ |
328 |
current = data->currentQMV; |
current = data->currentQMV; |
329 |
} |
} |
330 |
|
|
351 |
data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; } |
data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; } |
352 |
if (data->temp[4] < data->iMinSAD[4]) { |
if (data->temp[4] < data->iMinSAD[4]) { |
353 |
data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; } |
data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; } |
|
|
|
354 |
} |
} |
355 |
|
|
356 |
static void |
static void |
358 |
{ |
{ |
359 |
int32_t sad; uint32_t t; |
int32_t sad; uint32_t t; |
360 |
const uint8_t * Reference; |
const uint8_t * Reference; |
361 |
|
VECTOR * current; |
362 |
|
|
363 |
if ( (x > data->max_dx) || (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
364 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
365 |
|
|
366 |
if (!data->qpel_precision) Reference = GetReference(x, y, data); |
if (!data->qpel_precision) { |
367 |
else Reference = Interpolate8x8qpel(x, y, 0, 0, data); |
Reference = GetReference(x, y, data); |
368 |
|
current = data->currentMV; |
369 |
|
} else { /* x and y are in 1/4 precision */ |
370 |
|
Reference = Interpolate8x8qpel(x, y, 0, 0, data); |
371 |
|
current = data->currentQMV; |
372 |
|
} |
373 |
|
|
374 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
375 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
378 |
|
|
379 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
380 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
381 |
data->currentMV->x = x; data->currentMV->y = y; |
current->x = x; current->y = y; |
382 |
*dir = Direction; |
*dir = Direction; |
383 |
} |
} |
384 |
} |
} |
385 |
|
|
|
|
|
386 |
static void |
static void |
387 |
CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
388 |
{ |
{ |
389 |
uint32_t t; |
uint32_t t; |
390 |
const uint8_t * Reference; |
const uint8_t * Reference; |
391 |
|
|
392 |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || //non-zero integer value |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || /* non-zero even value */ |
393 |
(x > data->max_dx) || (x < data->min_dx) |
(x > data->max_dx) || (x < data->min_dx) |
394 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
395 |
|
|
424 |
uint32_t t; |
uint32_t t; |
425 |
VECTOR * current; |
VECTOR * current; |
426 |
|
|
427 |
if ( (x > data->max_dx) | ( x < data->min_dx) |
if ( (x > data->max_dx) || ( x < data->min_dx) |
428 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
429 |
|
|
430 |
if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; //non-zero even value |
if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; /* non-zero even value */ |
431 |
|
|
432 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (data->qpel_precision) { /* x and y are in 1/4 precision */ |
433 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
434 |
current = data->currentQMV; |
current = data->currentQMV; |
435 |
xc = x/2; yc = y/2; |
xc = x/2; yc = y/2; |
457 |
static void |
static void |
458 |
CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
459 |
{ |
{ |
460 |
// maximum speed - for P/B/I decision |
/* maximum speed - for P/B/I decision */ |
461 |
int32_t sad; |
int32_t sad; |
462 |
|
|
463 |
if ( (x > data->max_dx) || (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
464 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
465 |
|
|
466 |
sad = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
sad = sad32v_c(data->Cur, data->RefP[0] + (x>>1) + (y>>1)*(data->iEdgedWidth), |
467 |
data->iEdgedWidth, data->temp+1); |
data->iEdgedWidth, data->temp+1); |
468 |
|
|
469 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
562 |
} else { |
} else { |
563 |
xcf += mvs.x; ycf += mvs.y; |
xcf += mvs.x; ycf += mvs.y; |
564 |
xcb += b_mvs.x; ycb += b_mvs.y; |
xcb += b_mvs.x; ycb += b_mvs.y; |
565 |
mvs.x *= 2; mvs.y *= 2; //we move to qpel precision anyway |
mvs.x *= 2; mvs.y *= 2; /* we move to qpel precision anyway */ |
566 |
b_mvs.x *= 2; b_mvs.y *= 2; |
b_mvs.x *= 2; b_mvs.y *= 2; |
567 |
} |
} |
568 |
|
|
596 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
597 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
598 |
|
|
599 |
if (( x > 31) | ( x < -32) | ( y > 31) | (y < -32)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
600 |
|
|
601 |
mvs.x = data->directmvF[0].x + x; |
mvs.x = data->directmvF[0].x + x; |
602 |
b_mvs.x = ((x == 0) ? |
b_mvs.x = ((x == 0) ? |
608 |
data->directmvB[0].y |
data->directmvB[0].y |
609 |
: mvs.y - data->referencemv[0].y); |
: mvs.y - data->referencemv[0].y); |
610 |
|
|
611 |
if ( (mvs.x > data->max_dx) | (mvs.x < data->min_dx) |
if ( (mvs.x > data->max_dx) || (mvs.x < data->min_dx) |
612 |
| (mvs.y > data->max_dy) | (mvs.y < data->min_dy) |
|| (mvs.y > data->max_dy) || (mvs.y < data->min_dy) |
613 |
| (b_mvs.x > data->max_dx) | (b_mvs.x < data->min_dx) |
|| (b_mvs.x > data->max_dx) || (b_mvs.x < data->min_dx) |
614 |
| (b_mvs.y > data->max_dy) | (b_mvs.y < data->min_dy) ) return; |
|| (b_mvs.y > data->max_dy) || (b_mvs.y < data->min_dy) ) return; |
615 |
|
|
616 |
if (data->qpel) { |
if (data->qpel) { |
617 |
xcf = 4*(mvs.x/2); ycf = 4*(mvs.y/2); |
xcf = 4*(mvs.x/2); ycf = 4*(mvs.y/2); |
645 |
CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
646 |
{ |
{ |
647 |
|
|
648 |
static int16_t in[64], coeff[64]; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
649 |
int32_t bits = 0, sum; |
int32_t bits = 0; |
650 |
VECTOR * current; |
VECTOR * current; |
651 |
const uint8_t * ptr; |
const uint8_t * ptr; |
652 |
int i, cbp = 0, t, xc, yc; |
int i, cbp = 0, t, xc, yc; |
658 |
ptr = GetReference(x, y, data); |
ptr = GetReference(x, y, data); |
659 |
current = data->currentMV; |
current = data->currentMV; |
660 |
xc = x; yc = y; |
xc = x; yc = y; |
661 |
} else { // x and y are in 1/4 precision |
} else { /* x and y are in 1/4 precision */ |
662 |
ptr = Interpolate16x16qpel(x, y, 0, data); |
ptr = Interpolate16x16qpel(x, y, 0, data); |
663 |
current = data->currentQMV; |
current = data->currentQMV; |
664 |
xc = x/2; yc = y/2; |
xc = x/2; yc = y/2; |
667 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
668 |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
669 |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
670 |
fdct(in); |
bits += data->temp[i] = Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, i); |
|
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
|
|
else sum = quant4_inter(coeff, in, data->lambda16); |
|
|
if (sum > 0) { |
|
|
cbp |= 1 << (5 - i); |
|
|
bits += data->temp[i] = CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
|
|
} else data->temp[i] = 0; |
|
671 |
} |
} |
672 |
|
|
673 |
bits += t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
bits += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
674 |
|
|
675 |
if (bits < data->iMinSAD[0]) { // there is still a chance, adding chroma |
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
676 |
|
|
677 |
|
if (bits >= data->iMinSAD[0]) return; |
678 |
|
|
679 |
|
/* chroma */ |
680 |
xc = (xc >> 1) + roundtab_79[xc & 0x3]; |
xc = (xc >> 1) + roundtab_79[xc & 0x3]; |
681 |
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
682 |
|
|
683 |
//chroma U |
/* chroma U */ |
684 |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCU, 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
685 |
transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2); |
transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2); |
686 |
fdct(in); |
bits += Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 4); |
687 |
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
if (bits >= data->iMinSAD[0]) return; |
|
else sum = quant4_inter(coeff, in, data->lambda16); |
|
|
if (sum > 0) { |
|
|
cbp |= 1 << (5 - 4); |
|
|
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
|
|
} |
|
688 |
|
|
689 |
if (bits < data->iMinSAD[0]) { |
/* chroma V */ |
690 |
//chroma V |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
|
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCV, 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
|
691 |
transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2); |
transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2); |
692 |
fdct(in); |
bits += Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 5); |
|
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
|
|
else sum = quant4_inter(coeff, in, data->lambda16); |
|
|
if (sum > 0) { |
|
|
cbp |= 1 << (5 - 5); |
|
|
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
|
|
} |
|
|
} |
|
|
} |
|
693 |
|
|
694 |
bits += xvid_cbpy_tab[15-(cbp>>2)].len; |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
|
bits += mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
|
695 |
|
|
696 |
if (bits < data->iMinSAD[0]) { |
if (bits < data->iMinSAD[0]) { |
697 |
data->iMinSAD[0] = bits; |
data->iMinSAD[0] = bits; |
713 |
CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
714 |
{ |
{ |
715 |
|
|
716 |
static int16_t in[64], coeff[64]; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
717 |
int32_t sum, bits; |
int32_t bits; |
718 |
VECTOR * current; |
VECTOR * current; |
719 |
const uint8_t * ptr; |
const uint8_t * ptr; |
720 |
int cbp; |
int cbp = 0; |
721 |
|
|
722 |
if ( (x > data->max_dx) || (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
723 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
725 |
if (!data->qpel_precision) { |
if (!data->qpel_precision) { |
726 |
ptr = GetReference(x, y, data); |
ptr = GetReference(x, y, data); |
727 |
current = data->currentMV; |
current = data->currentMV; |
728 |
} else { // x and y are in 1/4 precision |
} else { /* x and y are in 1/4 precision */ |
729 |
ptr = Interpolate8x8qpel(x, y, 0, 0, data); |
ptr = Interpolate8x8qpel(x, y, 0, 0, data); |
730 |
current = data->currentQMV; |
current = data->currentQMV; |
731 |
} |
} |
732 |
|
|
733 |
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
734 |
fdct(in); |
bits = Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 5); |
735 |
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
bits += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
|
else sum = quant4_inter(coeff, in, data->lambda16); |
|
|
if (sum > 0) { |
|
|
bits = CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
|
|
cbp = 1; |
|
|
} else cbp = bits = 0; |
|
|
|
|
|
bits += sum = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
|
736 |
|
|
737 |
if (bits < data->iMinSAD[0]) { |
if (bits < data->iMinSAD[0]) { |
738 |
data->temp[0] = cbp; |
data->temp[0] = cbp; |
754 |
|
|
755 |
int iDirection; |
int iDirection; |
756 |
|
|
757 |
for(;;) { //forever |
for(;;) { /* forever */ |
758 |
iDirection = 0; |
iDirection = 0; |
759 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
760 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
763 |
|
|
764 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
765 |
|
|
766 |
if (iDirection) { //if anything found |
if (iDirection) { /* if anything found */ |
767 |
bDirection = iDirection; |
bDirection = iDirection; |
768 |
iDirection = 0; |
iDirection = 0; |
769 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
770 |
if (bDirection & 3) { //our candidate is left or right |
if (bDirection & 3) { /* our candidate is left or right */ |
771 |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
772 |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
773 |
} else { // what remains here is up or down |
} else { /* what remains here is up or down */ |
774 |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
775 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
776 |
} |
} |
779 |
bDirection += iDirection; |
bDirection += iDirection; |
780 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
781 |
} |
} |
782 |
} else { //about to quit, eh? not so fast.... |
} else { /* about to quit, eh? not so fast.... */ |
783 |
switch (bDirection) { |
switch (bDirection) { |
784 |
case 2: |
case 2: |
785 |
CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); |
CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); |
817 |
CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
818 |
CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); |
CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); |
819 |
break; |
break; |
820 |
default: //1+2+4+8 == we didn't find anything at all |
default: /* 1+2+4+8 == we didn't find anything at all */ |
821 |
CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4); |
CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4); |
822 |
CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); |
CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8); |
823 |
CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); |
CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4); |
824 |
CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
825 |
break; |
break; |
826 |
} |
} |
827 |
if (!iDirection) break; //ok, the end. really |
if (!iDirection) break; /* ok, the end. really */ |
828 |
bDirection = iDirection; |
bDirection = iDirection; |
829 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
830 |
} |
} |
869 |
|
|
870 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
871 |
|
|
872 |
if (iDirection) { //checking if anything found |
if (iDirection) { /* checking if anything found */ |
873 |
bDirection = iDirection; |
bDirection = iDirection; |
874 |
iDirection = 0; |
iDirection = 0; |
875 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
876 |
if (bDirection & 3) { //our candidate is left or right |
if (bDirection & 3) { /* our candidate is left or right */ |
877 |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
878 |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
879 |
} else { // what remains here is up or down |
} else { /* what remains here is up or down */ |
880 |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
881 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
882 |
} |
} |
894 |
{ |
{ |
895 |
/* Do a half-pel or q-pel refinement */ |
/* Do a half-pel or q-pel refinement */ |
896 |
const VECTOR centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV; |
const VECTOR centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV; |
897 |
int iDirection; //only needed because macro expects it |
int iDirection; /* only needed because macro expects it */ |
898 |
|
|
899 |
CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0); |
CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0); |
900 |
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0); |
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0); |
912 |
const uint32_t stride, const uint32_t iQuant, int rrv) |
const uint32_t stride, const uint32_t iQuant, int rrv) |
913 |
|
|
914 |
{ |
{ |
915 |
|
int offset = (x + y*stride)*8; |
916 |
if(!rrv) { |
if(!rrv) { |
917 |
uint32_t sadC = sad8(current->u + x*8 + y*stride*8, |
uint32_t sadC = sad8(current->u + offset, |
918 |
reference->u + x*8 + y*stride*8, stride); |
reference->u + offset, stride); |
919 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
920 |
sadC += sad8(current->v + (x + y*stride)*8, |
sadC += sad8(current->v + offset, |
921 |
reference->v + (x + y*stride)*8, stride); |
reference->v + offset, stride); |
922 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
923 |
return 1; |
return 1; |
924 |
|
|
925 |
} else { |
} else { |
926 |
uint32_t sadC = sad16(current->u + x*16 + y*stride*16, |
uint32_t sadC = sad16(current->u + 2*offset, |
927 |
reference->u + x*16 + y*stride*16, stride, 256*4096); |
reference->u + 2*offset, stride, 256*4096); |
928 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
929 |
sadC += sad16(current->v + (x + y*stride)*16, |
sadC += sad16(current->v + 2*offset, |
930 |
reference->v + (x + y*stride)*16, stride, 256*4096); |
reference->v + 2*offset, stride, 256*4096); |
931 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
932 |
return 1; |
return 1; |
933 |
} |
} |
942 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
943 |
} |
} |
944 |
|
|
945 |
|
static __inline void |
946 |
|
ModeDecision(SearchData * const Data, |
947 |
|
MACROBLOCK * const pMB, |
948 |
|
const MACROBLOCK * const pMBs, |
949 |
|
const int x, const int y, |
950 |
|
const MBParam * const pParam, |
951 |
|
const uint32_t MotionFlags, |
952 |
|
const uint32_t VopFlags, |
953 |
|
const uint32_t VolFlags, |
954 |
|
const IMAGE * const pCurrent, |
955 |
|
const IMAGE * const pRef) |
956 |
|
{ |
957 |
|
int mode = MODE_INTER; |
958 |
|
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
959 |
|
const uint32_t iQuant = pMB->quant; |
960 |
|
|
961 |
|
const int skip_possible = (!(VolFlags & XVID_VOL_GMC)) && (pMB->dquant == 0); |
962 |
|
|
963 |
|
if (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) { /* normal, fast, SAD-based mode decision */ |
964 |
|
int sad; |
965 |
|
int InterBias = MV16_INTER_BIAS; |
966 |
|
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
967 |
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) { |
968 |
|
mode = MODE_INTER; |
969 |
|
sad = Data->iMinSAD[0]; |
970 |
|
} else { |
971 |
|
mode = MODE_INTER4V; |
972 |
|
sad = Data->iMinSAD[1] + Data->iMinSAD[2] + |
973 |
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant; |
974 |
|
Data->iMinSAD[0] = sad; |
975 |
|
} |
976 |
|
|
977 |
|
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
978 |
|
if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
979 |
|
if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
980 |
|
if (Data->chroma || SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) { |
981 |
|
mode = MODE_NOT_CODED; |
982 |
|
sad = 0; |
983 |
|
} |
984 |
|
|
985 |
|
/* intra decision */ |
986 |
|
|
987 |
|
if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */ |
988 |
|
if (y != 0) |
989 |
|
if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80; |
990 |
|
if (x != 0) |
991 |
|
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
992 |
|
|
993 |
|
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? */ |
994 |
|
if (Data->rrv) InterBias *= 4; |
995 |
|
|
996 |
|
if (InterBias < pMB->sad16) { |
997 |
|
int32_t deviation; |
998 |
|
if (!Data->rrv) deviation = dev16(Data->Cur, Data->iEdgedWidth); |
999 |
|
else deviation = dev16(Data->Cur, Data->iEdgedWidth) + |
1000 |
|
dev16(Data->Cur+16, Data->iEdgedWidth) + |
1001 |
|
dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + |
1002 |
|
dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); |
1003 |
|
|
1004 |
|
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
1005 |
|
} |
1006 |
|
|
1007 |
|
} else { /* BITS */ |
1008 |
|
|
1009 |
|
int bits, intra, i; |
1010 |
|
VECTOR backup[5], *v; |
1011 |
|
Data->iQuant = iQuant; |
1012 |
|
|
1013 |
|
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
1014 |
|
for (i = 0; i < 5; i++) { |
1015 |
|
Data->iMinSAD[i] = 256*4096; |
1016 |
|
backup[i] = v[i]; |
1017 |
|
} |
1018 |
|
|
1019 |
|
bits = CountMBBitsInter(Data, pMBs, x, y, pParam, MotionFlags); |
1020 |
|
if (bits == 0) |
1021 |
|
mode = MODE_INTER; /* quick stop */ |
1022 |
|
else { |
1023 |
|
if (inter4v) { |
1024 |
|
int bits_inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
1025 |
|
if (bits_inter4v < bits) { Data->iMinSAD[0] = bits = bits_inter4v; mode = MODE_INTER4V; } |
1026 |
|
} |
1027 |
|
|
1028 |
|
intra = CountMBBitsIntra(Data); |
1029 |
|
|
1030 |
|
if (intra < bits) { *Data->iMinSAD = bits = intra; mode = MODE_INTRA; } |
1031 |
|
} |
1032 |
|
} |
1033 |
|
|
1034 |
|
if (Data->rrv) { |
1035 |
|
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
1036 |
|
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
1037 |
|
} |
1038 |
|
|
1039 |
|
if (mode == MODE_INTER) { |
1040 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1041 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1042 |
|
|
1043 |
|
if(Data->qpel) { |
1044 |
|
pMB->qmvs[0] = pMB->qmvs[1] |
1045 |
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
1046 |
|
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
1047 |
|
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; |
1048 |
|
} else { |
1049 |
|
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1050 |
|
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1051 |
|
} |
1052 |
|
|
1053 |
|
} else if (mode == MODE_INTER4V) |
1054 |
|
pMB->sad16 = Data->iMinSAD[0]; |
1055 |
|
else /* INTRA, NOT_CODED */ |
1056 |
|
SkipMacroblockP(pMB, 0); |
1057 |
|
|
1058 |
|
pMB->mode = mode; |
1059 |
|
} |
1060 |
|
|
1061 |
bool |
bool |
1062 |
MotionEstimation(MBParam * const pParam, |
MotionEstimation(MBParam * const pParam, |
1063 |
FRAMEINFO * const current, |
FRAMEINFO * const current, |
1084 |
(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \ |
(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \ |
1085 |
(current->vop_flags & XVID_VOP_MODEDECISION_BITS ? 2:1); |
(current->vop_flags & XVID_VOP_MODEDECISION_BITS ? 2:1); |
1086 |
|
|
1087 |
// some pre-initialized thingies for SearchP |
/* some pre-initialized thingies for SearchP */ |
1088 |
int32_t temp[8]; |
int32_t temp[8]; |
1089 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
1090 |
VECTOR currentQMV[5]; |
VECTOR currentQMV[5]; |
1091 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
1092 |
|
DECLARE_ALIGNED_MATRIX(dct_space, 2, 64, int16_t, CACHE_LINE); |
1093 |
SearchData Data; |
SearchData Data; |
1094 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
1095 |
Data.iEdgedWidth = iEdgedWidth; |
Data.iEdgedWidth = iEdgedWidth; |
1099 |
Data.temp = temp; |
Data.temp = temp; |
1100 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
1101 |
Data.rounding = pParam->m_rounding_type; |
Data.rounding = pParam->m_rounding_type; |
1102 |
Data.qpel = current->vol_flags & XVID_VOL_QUARTERPEL; |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
1103 |
Data.chroma = MotionFlags & XVID_ME_CHROMA16; |
Data.chroma = MotionFlags & XVID_ME_CHROMA16; |
1104 |
Data.rrv = current->vop_flags & XVID_VOP_REDUCED; |
Data.rrv = (current->vop_flags & XVID_VOP_REDUCED ? 1:0); |
1105 |
|
Data.dctSpace = dct_space; |
1106 |
|
Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
1107 |
|
|
1108 |
if ((current->vop_flags & XVID_VOP_REDUCED)) { |
if ((current->vop_flags & XVID_VOP_REDUCED)) { |
1109 |
mb_width = (pParam->width + 31) / 32; |
mb_width = (pParam->width + 31) / 32; |
1111 |
Data.qpel = 0; |
Data.qpel = 0; |
1112 |
} |
} |
1113 |
|
|
1114 |
Data.RefQ = pRefV->u; // a good place, also used in MC (for similar purpose) |
Data.RefQ = pRefV->u; /* a good place, also used in MC (for similar purpose) */ |
1115 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
1116 |
|
|
1117 |
for (y = 0; y < mb_height; y++) { |
for (y = 0; y < mb_height; y++) { |
1143 |
if (quant > 31) quant = 31; |
if (quant > 31) quant = 31; |
1144 |
else if (quant < 1) quant = 1; |
else if (quant < 1) quant = 1; |
1145 |
} |
} |
1146 |
|
pMB->quant = quant; |
1147 |
|
|
1148 |
pMB->quant = current->quant; |
/* initial skip decision */ |
|
|
|
|
//initial skip decision |
|
1149 |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
1150 |
if (!(current->vol_flags & XVID_VOL_GMC)) { /* no fast SKIP for S(GMC)-VOPs */ |
if (!(current->vol_flags & XVID_VOL_GMC)) { /* no fast SKIP for S(GMC)-VOPs */ |
1151 |
if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh) |
if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh) |
1156 |
} |
} |
1157 |
|
|
1158 |
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
1159 |
y, MotionFlags, current->vol_flags, pMB->quant, |
y, MotionFlags, current->vop_flags, current->vol_flags, |
1160 |
&Data, pParam, pMBs, reference->mbs, |
&Data, pParam, pMBs, reference->mbs, pMB); |
1161 |
current->vop_flags & XVID_VOP_INTER4V, pMB); |
|
1162 |
|
ModeDecision(&Data, pMB, pMBs, x, y, pParam, |
1163 |
|
MotionFlags, current->vop_flags, current->vol_flags, |
1164 |
|
pCurrent, pRef); |
1165 |
|
|
|
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
|
|
if (!(current->vol_flags & XVID_VOL_GMC || current->vop_flags & XVID_VOP_MODEDECISION_BITS)) { |
|
|
if ( pMB->dquant == 0 && sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) { |
|
|
if ( (100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
|
|
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) |
|
|
SkipMacroblockP(pMB, sad00); |
|
|
} |
|
|
} |
|
1166 |
if (pMB->mode == MODE_INTRA) |
if (pMB->mode == MODE_INTRA) |
1167 |
if (++iIntra > iLimit) return 1; |
if (++iIntra > iLimit) return 1; |
1168 |
} |
} |
1181 |
{ |
{ |
1182 |
int mask = 255, j; |
int mask = 255, j; |
1183 |
for (j = 0; j < i; j++) { |
for (j = 0; j < i; j++) { |
1184 |
if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already |
if (MVequal(pmv[i], pmv[j])) return 0; /* same vector has been checked already */ |
1185 |
if (pmv[i].x == pmv[j].x) { |
if (pmv[i].x == pmv[j].x) { |
1186 |
if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4; |
if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4; |
1187 |
else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8; |
else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8; |
1198 |
PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount, |
PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount, |
1199 |
int iHcount, const MACROBLOCK * const prevMB, int rrv) |
int iHcount, const MACROBLOCK * const prevMB, int rrv) |
1200 |
{ |
{ |
1201 |
|
/* this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself */ |
|
//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself |
|
1202 |
if (rrv) { iWcount /= 2; iHcount /= 2; } |
if (rrv) { iWcount /= 2; iHcount /= 2; } |
1203 |
|
|
1204 |
if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour |
if ( (y != 0) && (x < (iWcount-1)) ) { /* [5] top-right neighbour */ |
1205 |
pmv[5].x = EVEN(pmv[3].x); |
pmv[5].x = EVEN(pmv[3].x); |
1206 |
pmv[5].y = EVEN(pmv[3].y); |
pmv[5].y = EVEN(pmv[3].y); |
1207 |
} else pmv[5].x = pmv[5].y = 0; |
} else pmv[5].x = pmv[5].y = 0; |
1208 |
|
|
1209 |
if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour |
if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }/* pmv[3] is left neighbour */ |
1210 |
else pmv[3].x = pmv[3].y = 0; |
else pmv[3].x = pmv[3].y = 0; |
1211 |
|
|
1212 |
if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour |
if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }/* [4] top neighbour */ |
1213 |
else pmv[4].x = pmv[4].y = 0; |
else pmv[4].x = pmv[4].y = 0; |
1214 |
|
|
1215 |
// [1] median prediction |
/* [1] median prediction */ |
1216 |
pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); |
pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); |
1217 |
|
|
1218 |
pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask |
pmv[0].x = pmv[0].y = 0; /* [0] is zero; not used in the loop (checked before) but needed here for make_mask */ |
1219 |
|
|
1220 |
pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame |
pmv[2].x = EVEN(prevMB->mvs[0].x); /* [2] is last frame */ |
1221 |
pmv[2].y = EVEN(prevMB->mvs[0].y); |
pmv[2].y = EVEN(prevMB->mvs[0].y); |
1222 |
|
|
1223 |
if ((x < iWcount-1) && (y < iHcount-1)) { |
if ((x < iWcount-1) && (y < iHcount-1)) { |
1224 |
pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame |
pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); /* [6] right-down neighbour in last frame */ |
1225 |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
1226 |
} else pmv[6].x = pmv[6].y = 0; |
} else pmv[6].x = pmv[6].y = 0; |
1227 |
|
|
1234 |
} |
} |
1235 |
} |
} |
1236 |
|
|
|
static int |
|
|
ModeDecision(const uint32_t iQuant, SearchData * const Data, |
|
|
int inter4v, |
|
|
MACROBLOCK * const pMB, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const int x, const int y, |
|
|
const MBParam * const pParam, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t VopFlags) |
|
|
{ |
|
|
|
|
|
int mode = MODE_INTER; |
|
|
|
|
|
if (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) { //normal, fast, SAD-based mode decision |
|
|
// int intra = 0; |
|
|
int sad; |
|
|
int InterBias = MV16_INTER_BIAS; |
|
|
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
|
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) { |
|
|
mode = 0; //inter |
|
|
sad = Data->iMinSAD[0]; |
|
|
} else { |
|
|
mode = MODE_INTER4V; |
|
|
sad = Data->iMinSAD[1] + Data->iMinSAD[2] + |
|
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant; |
|
|
Data->iMinSAD[0] = sad; |
|
|
} |
|
|
|
|
|
/* intra decision */ |
|
|
|
|
|
if (iQuant > 8) InterBias += 100 * (iQuant - 8); // to make high quants work |
|
|
if (y != 0) |
|
|
if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80; |
|
|
if (x != 0) |
|
|
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
|
|
|
|
|
if (Data->chroma) InterBias += 50; // to compensate bigger SAD |
|
|
if (Data->rrv) InterBias *= 4; |
|
|
|
|
|
if (InterBias < pMB->sad16) { |
|
|
int32_t deviation; |
|
|
if (!Data->rrv) deviation = dev16(Data->Cur, Data->iEdgedWidth); |
|
|
else deviation = dev16(Data->Cur, Data->iEdgedWidth) + |
|
|
dev16(Data->Cur+8, Data->iEdgedWidth) + |
|
|
dev16(Data->Cur + 8*Data->iEdgedWidth, Data->iEdgedWidth) + |
|
|
dev16(Data->Cur+8+8*Data->iEdgedWidth, Data->iEdgedWidth); |
|
|
|
|
|
if (deviation < (sad - InterBias)) return MODE_INTRA;// intra |
|
|
} |
|
|
return mode; |
|
|
|
|
|
} else { |
|
|
|
|
|
int bits, intra, i; |
|
|
VECTOR backup[5], *v; |
|
|
Data->lambda16 = iQuant; |
|
|
Data->lambda8 = (pParam->vol_flags & XVID_VOL_MPEGQUANT)?1:0; |
|
|
|
|
|
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
|
|
for (i = 0; i < 5; i++) { |
|
|
Data->iMinSAD[i] = 256*4096; |
|
|
backup[i] = v[i]; |
|
|
} |
|
|
|
|
|
bits = CountMBBitsInter(Data, pMBs, x, y, pParam, MotionFlags); |
|
|
if (bits == 0) return MODE_INTER; // quick stop |
|
|
|
|
|
if (inter4v) { |
|
|
int inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
|
|
if (inter4v < bits) { Data->iMinSAD[0] = bits = inter4v; mode = MODE_INTER4V; } |
|
|
} |
|
|
|
|
|
|
|
|
intra = CountMBBitsIntra(Data); |
|
|
|
|
|
if (intra < bits) { *Data->iMinSAD = bits = intra; return MODE_INTRA; } |
|
|
|
|
|
return mode; |
|
|
} |
|
|
} |
|
|
|
|
1237 |
static void |
static void |
1238 |
SearchP(const IMAGE * const pRef, |
SearchP(const IMAGE * const pRef, |
1239 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1244 |
const int y, |
const int y, |
1245 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1246 |
const uint32_t VopFlags, |
const uint32_t VopFlags, |
1247 |
const uint32_t iQuant, |
const uint32_t VolFlags, |
1248 |
SearchData * const Data, |
SearchData * const Data, |
1249 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1250 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1251 |
const MACROBLOCK * const prevMBs, |
const MACROBLOCK * const prevMBs, |
|
int inter4v, |
|
1252 |
MACROBLOCK * const pMB) |
MACROBLOCK * const pMB) |
1253 |
{ |
{ |
1254 |
|
|
1255 |
int i, iDirection = 255, mask, threshA; |
int i, iDirection = 255, mask, threshA; |
1256 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
1257 |
|
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
1258 |
|
|
1259 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1260 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
1261 |
|
|
1262 |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); |
1263 |
|
|
1264 |
Data->temp[5] = Data->temp[6] = 0; // chroma-sad cache |
Data->temp[5] = Data->temp[6] = 0; /* chroma-sad cache */ |
1265 |
i = Data->rrv ? 2 : 1; |
i = Data->rrv ? 2 : 1; |
1266 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i; |
1267 |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
1268 |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
1269 |
|
|
1270 |
Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16*i; |
1271 |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16*i; |
1272 |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16*i; |
1273 |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16*i; |
1274 |
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
1275 |
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
1276 |
|
|
1277 |
Data->lambda16 = lambda_vec16[iQuant]; |
Data->lambda16 = lambda_vec16[pMB->quant]; |
1278 |
Data->lambda8 = lambda_vec8[iQuant]; |
Data->lambda8 = lambda_vec8[pMB->quant]; |
1279 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1280 |
|
|
1281 |
if (pMB->dquant != 0) inter4v = 0; |
memset(Data->currentMV, 0, 5*sizeof(VECTOR)); |
|
|
|
|
for(i = 0; i < 5; i++) |
|
|
Data->currentMV[i].x = Data->currentMV[i].y = 0; |
|
1282 |
|
|
1283 |
if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1284 |
else Data->predMV = pmv[0]; |
else Data->predMV = pmv[0]; |
1291 |
Data->iMinSAD[4] = pMB->sad8[3]; |
Data->iMinSAD[4] = pMB->sad8[3]; |
1292 |
|
|
1293 |
if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) || (x | y)) { |
if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) || (x | y)) { |
1294 |
threshA = Data->temp[0]; // that's where we keep this SAD atm |
threshA = Data->temp[0]; /* that's where we keep this SAD atm */ |
1295 |
if (threshA < 512) threshA = 512; |
if (threshA < 512) threshA = 512; |
1296 |
else if (threshA > 1024) threshA = 1024; |
else if (threshA > 1024) threshA = 1024; |
1297 |
} else |
} else |
1302 |
|
|
1303 |
if (!Data->rrv) { |
if (!Data->rrv) { |
1304 |
if (inter4v | Data->chroma) CheckCandidate = CheckCandidate16; |
if (inter4v | Data->chroma) CheckCandidate = CheckCandidate16; |
1305 |
else CheckCandidate = CheckCandidate16no4v; //for extra speed |
else CheckCandidate = CheckCandidate16no4v; /* for extra speed */ |
1306 |
} else CheckCandidate = CheckCandidate32; |
} else CheckCandidate = CheckCandidate32; |
1307 |
|
|
1308 |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
1315 |
|
|
1316 |
if ((Data->iMinSAD[0] <= threshA) || |
if ((Data->iMinSAD[0] <= threshA) || |
1317 |
(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && |
(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && |
1318 |
(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) { |
(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) |
1319 |
if (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) inter4v = 0; } |
inter4v = 0; |
1320 |
else { |
else { |
1321 |
|
|
1322 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
1362 |
} |
} |
1363 |
|
|
1364 |
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
|
if ((!(MotionFlags & XVID_ME_HALFPELREFINE16_BITS)) || Data->iMinSAD[0] < 200*(int)iQuant) |
|
1365 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1366 |
|
|
1367 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
1368 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* initialize qpel vectors */ |
1369 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
1370 |
} |
} |
1371 |
|
|
1372 |
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) { |
if (Data->qpel) { |
|
|
|
1373 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1374 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
|
|
|
|
if ((!(MotionFlags & XVID_ME_QUARTERPELREFINE16_BITS)) || (Data->iMinSAD[0] < 200*(int)iQuant)) { |
|
1375 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1376 |
|
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) |
1377 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1378 |
} |
} |
|
} |
|
|
|
|
|
if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)iQuant * 30)) inter4v = 0; |
|
1379 |
|
|
1380 |
if (inter4v && (!(VopFlags & XVID_VOP_MODEDECISION_BITS) || |
if (Data->iMinSAD[0] < (int32_t)pMB->quant * 30) |
1381 |
(!(MotionFlags & XVID_ME_QUARTERPELREFINE8_BITS)) || (!(MotionFlags & XVID_ME_HALFPELREFINE8_BITS)) || |
inter4v = 0; |
|
((!(MotionFlags & XVID_ME_EXTSEARCH_BITS)) && (!(MotionFlags&XVID_ME_EXTSEARCH8)) ))) { |
|
|
// if decision is BITS-based and all refinement steps will be done in BITS domain, there is no reason to call this loop |
|
1382 |
|
|
1383 |
|
if (inter4v) { |
1384 |
SearchData Data8; |
SearchData Data8; |
1385 |
memcpy(&Data8, Data, sizeof(SearchData)); //quick copy of common data |
memcpy(&Data8, Data, sizeof(SearchData)); /* quick copy of common data */ |
1386 |
|
|
1387 |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
1388 |
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
1390 |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
1391 |
|
|
1392 |
if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_BITS))) { |
if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_BITS))) { |
1393 |
// chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, there is no reason to compute it |
/* chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, it will not be used */ |
1394 |
int sumx = 0, sumy = 0; |
int sumx = 0, sumy = 0; |
|
const int div = 1 + Data->qpel; |
|
|
const VECTOR * const mv = Data->qpel ? pMB->qmvs : pMB->mvs; |
|
1395 |
|
|
1396 |
for (i = 0; i < 4; i++) { |
if (Data->qpel) |
1397 |
sumx += mv[i].x / div; |
for (i = 1; i < 5; i++) { |
1398 |
sumy += mv[i].y / div; |
sumx += Data->currentQMV[i].x/2; |
1399 |
|
sumy += Data->currentQMV[i].y/2; |
1400 |
|
} |
1401 |
|
else |
1402 |
|
for (i = 1; i < 5; i++) { |
1403 |
|
sumx += Data->currentMV[i].x; |
1404 |
|
sumy += Data->currentMV[i].y; |
1405 |
} |
} |
1406 |
|
|
1407 |
Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf], |
Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf], |
1408 |
(sumy >> 3) + roundtab_76[sumy & 0xf], Data); |
(sumy >> 3) + roundtab_76[sumy & 0xf], Data); |
1409 |
} |
} |
1410 |
} |
} else Data->iMinSAD[1] = 4096*256; |
|
|
|
|
inter4v = ModeDecision(iQuant, Data, inter4v, pMB, pMBs, x, y, pParam, MotionFlags, VopFlags); |
|
|
|
|
|
if (Data->rrv) { |
|
|
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
|
|
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
|
|
} |
|
|
|
|
|
if (inter4v == MODE_INTER) { |
|
|
pMB->mode = MODE_INTER; |
|
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
|
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
|
|
|
|
|
if(Data->qpel) { |
|
|
pMB->qmvs[0] = pMB->qmvs[1] |
|
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
|
|
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
|
|
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; |
|
|
} else { |
|
|
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
|
|
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
|
|
} |
|
|
|
|
|
} else if (inter4v == MODE_INTER4V) { |
|
|
pMB->mode = MODE_INTER4V; |
|
|
pMB->sad16 = Data->iMinSAD[0]; |
|
|
} else { // INTRA mode |
|
|
SkipMacroblockP(pMB, 0); // not skip, but similar enough |
|
|
pMB->mode = MODE_INTRA; |
|
|
} |
|
|
|
|
1411 |
} |
} |
1412 |
|
|
1413 |
static void |
static void |
1438 |
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10; |
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10; |
1439 |
|
|
1440 |
if (MotionFlags & (XVID_ME_EXTSEARCH8|XVID_ME_HALFPELREFINE8|XVID_ME_QUARTERPELREFINE8)) { |
if (MotionFlags & (XVID_ME_EXTSEARCH8|XVID_ME_HALFPELREFINE8|XVID_ME_QUARTERPELREFINE8)) { |
|
if (Data->rrv) i = 2; else i = 1; |
|
1441 |
|
|
1442 |
Data->Ref = OldData->Ref + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
if (Data->rrv) i = 16; else i = 8; |
1443 |
Data->RefH = OldData->RefH + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
|
1444 |
Data->RefV = OldData->RefV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->RefP[0] = OldData->RefP[0] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1445 |
Data->RefHV = OldData->RefHV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->RefP[1] = OldData->RefP[1] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1446 |
|
Data->RefP[2] = OldData->RefP[2] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1447 |
|
Data->RefP[3] = OldData->RefP[3] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1448 |
|
|
1449 |
Data->Cur = OldData->Cur + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->Cur = OldData->Cur + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1450 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1451 |
|
|
1452 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1456 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
1457 |
|
|
1458 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_BITS))) { |
if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_BITS))) { |
1459 |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
1460 |
|
|
1461 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1462 |
if (MotionFlags & XVID_ME_USESQUARES8) MainSearchPtr = SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES8) MainSearchPtr = SquareSearch; |
1466 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1467 |
|
|
1468 |
if(*(Data->iMinSAD) < temp_sad) { |
if(*(Data->iMinSAD) < temp_sad) { |
1469 |
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
1470 |
Data->currentQMV->y = 2 * Data->currentMV->y; |
Data->currentQMV->y = 2 * Data->currentMV->y; |
1471 |
} |
} |
1472 |
} |
} |
1473 |
|
|
1474 |
if (MotionFlags & XVID_ME_HALFPELREFINE8) { |
if (MotionFlags & XVID_ME_HALFPELREFINE8) { |
1475 |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
1476 |
|
|
1477 |
SubpelRefine(Data); // perform halfpel refine of current best vector |
SubpelRefine(Data); /* perform halfpel refine of current best vector */ |
1478 |
|
|
1479 |
if(*(Data->iMinSAD) < temp_sad) { // we have found a better match |
if(*(Data->iMinSAD) < temp_sad) { /* we have found a better match */ |
1480 |
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
1481 |
Data->currentQMV->y = 2 * Data->currentMV->y; |
Data->currentQMV->y = 2 * Data->currentMV->y; |
1482 |
} |
} |
1483 |
} |
} |
1524 |
const uint32_t mode_curr) |
const uint32_t mode_curr) |
1525 |
{ |
{ |
1526 |
|
|
1527 |
// [0] is prediction |
/* [0] is prediction */ |
1528 |
pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y); |
pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y); |
1529 |
|
|
1530 |
pmv[1].x = pmv[1].y = 0; // [1] is zero |
pmv[1].x = pmv[1].y = 0; /* [1] is zero */ |
1531 |
|
|
1532 |
pmv[2] = ChoosePred(pMB, mode_curr); |
pmv[2] = ChoosePred(pMB, mode_curr); |
1533 |
pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); |
pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); |
1534 |
|
|
1535 |
if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour |
if ((y != 0)&&(x != (int)(iWcount+1))) { /* [3] top-right neighbour */ |
1536 |
pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr); |
pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr); |
1537 |
pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); |
pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); |
1538 |
} else pmv[3].x = pmv[3].y = 0; |
} else pmv[3].x = pmv[3].y = 0; |
1578 |
*Data->iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
1579 |
Data->iFcode = iFcode; |
Data->iFcode = iFcode; |
1580 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1581 |
Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; // reset chroma-sad cache |
Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; /* reset chroma-sad cache */ |
1582 |
|
|
1583 |
Data->Ref = pRef->y + (x + y * Data->iEdgedWidth) * 16; |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
1584 |
Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16; |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16; |
1585 |
Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16; |
Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16; |
1586 |
Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16; |
Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
1587 |
Data->RefCU = pRef->u + (x + y * Data->iEdgedWidth/2) * 8; |
Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
1588 |
Data->RefCV = pRef->v + (x + y * Data->iEdgedWidth/2) * 8; |
Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
1589 |
|
|
1590 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1591 |
|
|
1600 |
Data->currentMV->x = Data->currentMV->y = 0; |
Data->currentMV->x = Data->currentMV->y = 0; |
1601 |
CheckCandidate = CheckCandidate16no4v; |
CheckCandidate = CheckCandidate16no4v; |
1602 |
|
|
1603 |
// main loop. checking all predictions |
/* main loop. checking all predictions */ |
1604 |
for (i = 0; i < 7; i++) { |
for (i = 0; i < 7; i++) { |
1605 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!(mask = make_mask(pmv, i)) ) continue; |
1606 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
1623 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1624 |
} |
} |
1625 |
|
|
1626 |
// three bits are needed to code backward mode. four for forward |
/* three bits are needed to code backward mode. four for forward */ |
1627 |
|
|
1628 |
if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * Data->lambda16; |
if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * Data->lambda16; |
1629 |
else *Data->iMinSAD += 3 * Data->lambda16; |
else *Data->iMinSAD += 3 * Data->lambda16; |
1647 |
} |
} |
1648 |
|
|
1649 |
if (mode_current == MODE_FORWARD) *(Data->currentMV+2) = *Data->currentMV; |
if (mode_current == MODE_FORWARD) *(Data->currentMV+2) = *Data->currentMV; |
1650 |
else *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search |
else *(Data->currentMV+1) = *Data->currentMV; /* we store currmv for interpolate search */ |
1651 |
} |
} |
1652 |
|
|
1653 |
static void |
static void |
1663 |
const int div = 1 + Data->qpel; |
const int div = 1 + Data->qpel; |
1664 |
int k; |
int k; |
1665 |
const uint32_t stride = Data->iEdgedWidth/2; |
const uint32_t stride = Data->iEdgedWidth/2; |
1666 |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
/* this is not full chroma compensation, only it's fullpel approximation. should work though */ |
1667 |
|
|
1668 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1669 |
dy += Data->directmvF[k].y / div; |
dy += Data->directmvF[k].y / div; |
1670 |
dx += Data->directmvF[0].x / div; |
dx += Data->directmvF[k].x / div; |
1671 |
b_dy += Data->directmvB[0].y / div; |
b_dy += Data->directmvB[k].y / div; |
1672 |
b_dx += Data->directmvB[0].x / div; |
b_dx += Data->directmvB[k].x / div; |
1673 |
} |
} |
1674 |
|
|
1675 |
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
1682 |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
1683 |
stride); |
stride); |
1684 |
|
|
1685 |
if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; //no skip |
if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; /* no skip */ |
1686 |
|
|
1687 |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
1688 |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
1689 |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
1690 |
stride); |
stride); |
1691 |
|
|
1692 |
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped |
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
1693 |
|
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
1694 |
|
for (k = 0; k < 4; k++) { |
1695 |
|
pMB->qmvs[k] = pMB->mvs[k]; |
1696 |
|
pMB->b_qmvs[k] = pMB->b_mvs[k]; |
1697 |
|
} |
1698 |
|
} |
1699 |
} |
} |
1700 |
|
|
1701 |
static __inline uint32_t |
static __inline uint32_t |
1723 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1724 |
|
|
1725 |
*Data->iMinSAD = 256*4096; |
*Data->iMinSAD = 256*4096; |
1726 |
Data->Ref = f_Ref->y + k; |
Data->RefP[0] = f_Ref->y + k; |
1727 |
Data->RefH = f_RefH + k; |
Data->RefP[2] = f_RefH + k; |
1728 |
Data->RefV = f_RefV + k; |
Data->RefP[1] = f_RefV + k; |
1729 |
Data->RefHV = f_RefHV + k; |
Data->RefP[3] = f_RefHV + k; |
1730 |
Data->bRef = b_Ref->y + k; |
Data->b_RefP[0] = b_Ref->y + k; |
1731 |
Data->bRefH = b_RefH + k; |
Data->b_RefP[2] = b_RefH + k; |
1732 |
Data->bRefV = b_RefV + k; |
Data->b_RefP[1] = b_RefV + k; |
1733 |
Data->bRefHV = b_RefHV + k; |
Data->b_RefP[3] = b_RefHV + k; |
1734 |
Data->RefCU = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
Data->RefP[4] = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
1735 |
Data->RefCV = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
Data->RefP[5] = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
1736 |
Data->b_RefCU = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
Data->b_RefP[4] = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
1737 |
Data->b_RefCV = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
Data->b_RefP[5] = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
1738 |
|
|
1739 |
k = Data->qpel ? 4 : 2; |
k = Data->qpel ? 4 : 2; |
1740 |
Data->max_dx = k * (pParam->width - x * 16); |
Data->max_dx = k * (pParam->width - x * 16); |
1754 |
if ( (pMB->b_mvs[k].x > Data->max_dx) | (pMB->b_mvs[k].x < Data->min_dx) |
if ( (pMB->b_mvs[k].x > Data->max_dx) | (pMB->b_mvs[k].x < Data->min_dx) |
1755 |
| (pMB->b_mvs[k].y > Data->max_dy) | (pMB->b_mvs[k].y < Data->min_dy) ) { |
| (pMB->b_mvs[k].y > Data->max_dy) | (pMB->b_mvs[k].y < Data->min_dy) ) { |
1756 |
|
|
1757 |
*best_sad = 256*4096; // in that case, we won't use direct mode |
*best_sad = 256*4096; /* in that case, we won't use direct mode */ |
1758 |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
pMB->mode = MODE_DIRECT; /* just to make sure it doesn't say "MODE_DIRECT_NONE_MV" */ |
1759 |
pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; |
pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; |
1760 |
return 256*4096; |
return 256*4096; |
1761 |
} |
} |
1772 |
|
|
1773 |
CheckCandidate(0, 0, 255, &k, Data); |
CheckCandidate(0, 0, 255, &k, Data); |
1774 |
|
|
1775 |
// initial (fast) skip decision |
/* initial (fast) skip decision */ |
1776 |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (2 + Data->chroma?1:0)) { |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
1777 |
//possible skip |
/* possible skip */ |
1778 |
if (Data->chroma) { |
if (Data->chroma) { |
1779 |
pMB->mode = MODE_DIRECT_NONE_MV; |
pMB->mode = MODE_DIRECT_NONE_MV; |
1780 |
return *Data->iMinSAD; // skip. |
return *Data->iMinSAD; /* skip. */ |
1781 |
} else { |
} else { |
1782 |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
1783 |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip. |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; /* skip. */ |
1784 |
} |
} |
1785 |
} |
} |
1786 |
|
|
1787 |
*Data->iMinSAD += Data->lambda16; |
*Data->iMinSAD += Data->lambda16; |
1788 |
skip_sad = *Data->iMinSAD; |
skip_sad = *Data->iMinSAD; |
1789 |
|
|
1790 |
// DIRECT MODE DELTA VECTOR SEARCH. |
/* |
1791 |
// This has to be made more effective, but at the moment I'm happy it's running at all |
* DIRECT MODE DELTA VECTOR SEARCH. |
1792 |
|
* This has to be made more effective, but at the moment I'm happy it's running at all |
1793 |
|
*/ |
1794 |
|
|
1795 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
1796 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1803 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1804 |
|
|
1805 |
if (Data->qpel || b_mb->mode == MODE_INTER4V) pMB->mode = MODE_DIRECT; |
if (Data->qpel || b_mb->mode == MODE_INTER4V) pMB->mode = MODE_DIRECT; |
1806 |
else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
else pMB->mode = MODE_DIRECT_NO4V; /* for faster compensation */ |
1807 |
|
|
1808 |
pMB->pmvs[3] = *Data->currentMV; |
pMB->pmvs[3] = *Data->currentMV; |
1809 |
|
|
1861 |
SearchData bData; |
SearchData bData; |
1862 |
|
|
1863 |
fData->qpel_precision = 0; |
fData->qpel_precision = 0; |
1864 |
memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data |
memcpy(&bData, fData, sizeof(SearchData)); /* quick copy of common data */ |
1865 |
*fData->iMinSAD = 4096*256; |
*fData->iMinSAD = 4096*256; |
1866 |
bData.currentMV++; bData.currentQMV++; |
bData.currentMV++; bData.currentQMV++; |
1867 |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1868 |
|
|
1869 |
i = (x + y * fData->iEdgedWidth) * 16; |
i = (x + y * fData->iEdgedWidth) * 16; |
|
bData.bRef = fData->Ref = f_Ref->y + i; |
|
|
bData.bRefH = fData->RefH = f_RefH + i; |
|
|
bData.bRefV = fData->RefV = f_RefV + i; |
|
|
bData.bRefHV = fData->RefHV = f_RefHV + i; |
|
|
bData.Ref = fData->bRef = b_Ref->y + i; |
|
|
bData.RefH = fData->bRefH = b_RefH + i; |
|
|
bData.RefV = fData->bRefV = b_RefV + i; |
|
|
bData.RefHV = fData->bRefHV = b_RefHV + i; |
|
|
bData.b_RefCU = fData->RefCU = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.b_RefCV = fData->RefCV = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.RefCU = fData->b_RefCU = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.RefCV = fData->b_RefCV = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
1870 |
|
|
1871 |
|
bData.b_RefP[0] = fData->RefP[0] = f_Ref->y + i; |
1872 |
|
bData.b_RefP[2] = fData->RefP[2] = f_RefH + i; |
1873 |
|
bData.b_RefP[1] = fData->RefP[1] = f_RefV + i; |
1874 |
|
bData.b_RefP[3] = fData->RefP[3] = f_RefHV + i; |
1875 |
|
bData.RefP[0] = fData->b_RefP[0] = b_Ref->y + i; |
1876 |
|
bData.RefP[2] = fData->b_RefP[2] = b_RefH + i; |
1877 |
|
bData.RefP[1] = fData->b_RefP[1] = b_RefV + i; |
1878 |
|
bData.RefP[3] = fData->b_RefP[3] = b_RefHV + i; |
1879 |
|
bData.b_RefP[4] = fData->RefP[4] = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
1880 |
|
bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
1881 |
|
bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
1882 |
|
bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
1883 |
|
|
1884 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
1885 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
1900 |
|
|
1901 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
1902 |
|
|
1903 |
//diamond |
/* diamond */ |
1904 |
do { |
do { |
1905 |
iDirection = 255; |
iDirection = 255; |
1906 |
// forward MV moves |
/* forward MV moves */ |
1907 |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
1908 |
|
|
1909 |
CheckCandidateInt(i + 1, j, 0, &iDirection, fData); |
CheckCandidateInt(i + 1, j, 0, &iDirection, fData); |
1911 |
CheckCandidateInt(i - 1, j, 0, &iDirection, fData); |
CheckCandidateInt(i - 1, j, 0, &iDirection, fData); |
1912 |
CheckCandidateInt(i, j - 1, 0, &iDirection, fData); |
CheckCandidateInt(i, j - 1, 0, &iDirection, fData); |
1913 |
|
|
1914 |
// backward MV moves |
/* backward MV moves */ |
1915 |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
1916 |
fData->currentMV[2] = fData->currentMV[0]; |
fData->currentMV[2] = fData->currentMV[0]; |
1917 |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
1921 |
|
|
1922 |
} while (!(iDirection)); |
} while (!(iDirection)); |
1923 |
|
|
1924 |
//qpel refinement |
/* qpel refinement */ |
1925 |
if (fData->qpel) { |
if (fData->qpel) { |
1926 |
if (*fData->iMinSAD > *best_sad + 500) return; |
if (*fData->iMinSAD > *best_sad + 500) return; |
1927 |
CheckCandidate = CheckCandidateInt; |
CheckCandidate = CheckCandidateInt; |
1938 |
SubpelRefine(&bData); |
SubpelRefine(&bData); |
1939 |
} |
} |
1940 |
|
|
1941 |
*fData->iMinSAD += (2+3) * fData->lambda16; // two bits are needed to code interpolate mode. |
*fData->iMinSAD += (2+3) * fData->lambda16; /* two bits are needed to code interpolate mode. */ |
1942 |
|
|
1943 |
if (*fData->iMinSAD < *best_sad) { |
if (*fData->iMinSAD < *best_sad) { |
1944 |
*best_sad = *fData->iMinSAD; |
*best_sad = *fData->iMinSAD; |
1966 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
1967 |
const int32_t time_bp, |
const int32_t time_bp, |
1968 |
const int32_t time_pp, |
const int32_t time_pp, |
1969 |
// forward (past) reference |
/* forward (past) reference */ |
1970 |
const MACROBLOCK * const f_mbs, |
const MACROBLOCK * const f_mbs, |
1971 |
const IMAGE * const f_ref, |
const IMAGE * const f_ref, |
1972 |
const IMAGE * const f_refH, |
const IMAGE * const f_refH, |
1973 |
const IMAGE * const f_refV, |
const IMAGE * const f_refV, |
1974 |
const IMAGE * const f_refHV, |
const IMAGE * const f_refHV, |
1975 |
// backward (future) reference |
/* backward (future) reference */ |
1976 |
const FRAMEINFO * const b_reference, |
const FRAMEINFO * const b_reference, |
1977 |
const IMAGE * const b_ref, |
const IMAGE * const b_ref, |
1978 |
const IMAGE * const b_refH, |
const IMAGE * const b_refH, |
1990 |
const int32_t TRB = time_pp - time_bp; |
const int32_t TRB = time_pp - time_bp; |
1991 |
const int32_t TRD = time_pp; |
const int32_t TRD = time_pp; |
1992 |
|
|
1993 |
// some pre-inintialized data for the rest of the search |
/* some pre-inintialized data for the rest of the search */ |
1994 |
|
|
1995 |
SearchData Data; |
SearchData Data; |
1996 |
int32_t iMinSAD; |
int32_t iMinSAD; |
2007 |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA8; |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA8; |
2008 |
Data.temp = temp; |
Data.temp = temp; |
2009 |
|
|
2010 |
Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose) |
Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */ |
2011 |
// note: i==horizontal, j==vertical |
|
2012 |
|
/* note: i==horizontal, j==vertical */ |
2013 |
for (j = 0; j < pParam->mb_height; j++) { |
for (j = 0; j < pParam->mb_height; j++) { |
2014 |
|
|
2015 |
f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */ |
f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */ |
2045 |
|
|
2046 |
if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
2047 |
|
|
2048 |
// forward search |
/* forward search */ |
2049 |
SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
2050 |
&frame->image, i, j, |
&frame->image, i, j, |
2051 |
frame->motion_flags, |
frame->motion_flags, |
2053 |
pMB, &f_predMV, &best_sad, |
pMB, &f_predMV, &best_sad, |
2054 |
MODE_FORWARD, &Data); |
MODE_FORWARD, &Data); |
2055 |
|
|
2056 |
// backward search |
/* backward search */ |
2057 |
SearchBF(b_ref, b_refH->y, b_refV->y, b_refHV->y, |
SearchBF(b_ref, b_refH->y, b_refV->y, b_refHV->y, |
2058 |
&frame->image, i, j, |
&frame->image, i, j, |
2059 |
frame->motion_flags, |
frame->motion_flags, |
2061 |
pMB, &b_predMV, &best_sad, |
pMB, &b_predMV, &best_sad, |
2062 |
MODE_BACKWARD, &Data); |
MODE_BACKWARD, &Data); |
2063 |
|
|
2064 |
// interpolate search comes last, because it uses data from forward and backward as prediction |
/* interpolate search comes last, because it uses data from forward and backward as prediction */ |
2065 |
SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
2066 |
b_ref, b_refH->y, b_refV->y, b_refHV->y, |
b_ref, b_refH->y, b_refV->y, b_refHV->y, |
2067 |
&frame->image, |
&frame->image, |
2073 |
pMB, &best_sad, |
pMB, &best_sad, |
2074 |
&Data); |
&Data); |
2075 |
|
|
2076 |
// final skip decision |
/* final skip decision */ |
2077 |
if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP * 2) |
if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP * 2) |
2078 |
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
2079 |
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
2119 |
|
|
2120 |
for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
2121 |
|
|
2122 |
//median is only used as prediction. it doesn't have to be real |
/* median is only used as prediction. it doesn't have to be real */ |
2123 |
if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0; |
if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0; |
2124 |
else |
else |
2125 |
if (x == 1) //left macroblock does not have any vector now |
if (x == 1) /* left macroblock does not have any vector now */ |
2126 |
Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median |
Data->predMV = (pMB - pParam->mb_width)->mvs[0]; /* top instead of median */ |
2127 |
else if (y == 1) // top macroblock doesn't have it's vector |
else if (y == 1) /* top macroblock doesn't have it's vector */ |
2128 |
Data->predMV = (pMB - 1)->mvs[0]; // left instead of median |
Data->predMV = (pMB - 1)->mvs[0]; /* left instead of median */ |
2129 |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); /* else median */ |
2130 |
|
|
2131 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2132 |
pParam->width, pParam->height, Data->iFcode - quarterpel, 0, 0); |
pParam->width, pParam->height, Data->iFcode - quarterpel, 0, 0); |
2133 |
|
|
2134 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
2135 |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16; |
2136 |
|
|
2137 |
pmv[1].x = EVEN(pMB->mvs[0].x); |
pmv[1].x = EVEN(pMB->mvs[0].x); |
2138 |
pmv[1].y = EVEN(pMB->mvs[0].y); |
pmv[1].y = EVEN(pMB->mvs[0].y); |
2141 |
pmv[0].x = pmv[0].y = 0; |
pmv[0].x = pmv[0].y = 0; |
2142 |
|
|
2143 |
CheckCandidate32I(0, 0, 255, &i, Data); |
CheckCandidate32I(0, 0, 255, &i, Data); |
|
Data->iMinSAD[1] -= 50; |
|
|
Data->iMinSAD[2] -= 50; |
|
|
Data->iMinSAD[3] -= 50; |
|
|
Data->iMinSAD[4] -= 50; |
|
2144 |
|
|
2145 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { |
2146 |
|
|
2149 |
if (!(mask = make_mask(pmv, 2))) |
if (!(mask = make_mask(pmv, 2))) |
2150 |
CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data); |
CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data); |
2151 |
|
|
2152 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) // diamond only if needed |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) /* diamond only if needed */ |
2153 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
2154 |
} |
} |
2155 |
|
|
2161 |
} |
} |
2162 |
} |
} |
2163 |
|
|
2164 |
#define INTRA_THRESH 2400 |
#define INTRA_THRESH 1700 |
2165 |
#define INTER_THRESH 1100 |
#define INTER_THRESH 1200 |
2166 |
|
|
2167 |
int |
int |
2168 |
MEanalysis( const IMAGE * const pRef, |
MEanalysis( const IMAGE * const pRef, |
2169 |
const FRAMEINFO * const Current, |
const FRAMEINFO * const Current, |
2170 |
const MBParam * const pParam, |
const MBParam * const pParam, |
2171 |
const int maxIntra, //maximum number if non-I frames |
const int maxIntra, /* maximum number if non-I frames */ |
2172 |
const int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame |
const int intraCount, /* number of non-I frames after last I frame; 0 if we force P/B frame */ |
2173 |
const int bCount, // number of B frames in a row |
const int bCount, /* number of B frames in a row */ |
2174 |
const int b_thresh) |
const int b_thresh) |
2175 |
{ |
{ |
2176 |
uint32_t x, y, intra = 0; |
uint32_t x, y, intra = 0; |
2190 |
Data.temp = temp; |
Data.temp = temp; |
2191 |
CheckCandidate = CheckCandidate32I; |
CheckCandidate = CheckCandidate32I; |
2192 |
|
|
2193 |
if (intraCount != 0 && intraCount < 10) // we're right after an I frame |
if (intraCount != 0) { |
2194 |
IntraThresh += 8 * (intraCount - 10) * (intraCount - 10); |
if (intraCount < 10) /* we're right after an I frame */ |
2195 |
|
IntraThresh += 15* (intraCount - 10) * (intraCount - 10); |
2196 |
else |
else |
2197 |
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
if ( 5*(maxIntra - intraCount) < maxIntra) /* we're close to maximum. 2 sec when max is 10 sec */ |
2198 |
IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra; |
IntraThresh -= (IntraThresh * (maxIntra - 8*(maxIntra - intraCount)))/maxIntra; |
2199 |
|
} |
2200 |
|
|
2201 |
InterThresh -= (350 - 8*b_thresh) * bCount; |
InterThresh -= (350 - 8*b_thresh) * bCount; |
2202 |
if (InterThresh < 300 + 5*b_thresh) InterThresh = 300 + 5*b_thresh; |
if (InterThresh < 300 + 5*b_thresh) InterThresh = 300 + 5*b_thresh; |
2209 |
blocks += 4; |
blocks += 4; |
2210 |
|
|
2211 |
if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV; |
if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV; |
2212 |
else { //extrapolation of the vector found for last frame |
else { /* extrapolation of the vector found for last frame */ |
2213 |
pMBs[x + y * pParam->mb_width].mvs[0].x = |
pMBs[x + y * pParam->mb_width].mvs[0].x = |
2214 |
(pMBs[x + y * pParam->mb_width].mvs[0].x * (bCount+1) ) / bCount; |
(pMBs[x + y * pParam->mb_width].mvs[0].x * (bCount+1) ) / bCount; |
2215 |
pMBs[x + y * pParam->mb_width].mvs[0].y = |
pMBs[x + y * pParam->mb_width].mvs[0].y = |
2237 |
} |
} |
2238 |
|
|
2239 |
sSAD /= blocks; |
sSAD /= blocks; |
|
s = (10*s) / blocks; |
|
2240 |
|
|
2241 |
if (s > 5) sSAD += (s - 4) * (180 - 2*b_thresh); //static block - looks bad when in bframe... |
if (b_thresh < 20) { |
2242 |
|
s = (10*s) / blocks; |
2243 |
|
if (s > 4) sSAD += (s - 2) * (40 - 2*b_thresh); /* static block - looks bad when in bframe... */ |
2244 |
|
} |
2245 |
|
|
2246 |
if (sSAD > InterThresh ) return P_VOP; |
if (sSAD > InterThresh ) return P_VOP; |
2247 |
emms(); |
emms(); |
2259 |
const IMAGE * const pRefHV ) |
const IMAGE * const pRefHV ) |
2260 |
{ |
{ |
2261 |
|
|
2262 |
const int deltax=8; // upper bound for difference between a MV and it's neighbour MVs |
const int deltax=8; /* upper bound for difference between a MV and it's neighbour MVs */ |
2263 |
const int deltay=8; |
const int deltay=8; |
2264 |
const int grad=512; // lower bound for deviation in MB |
const int grad=512; /* lower bound for deviation in MB */ |
2265 |
|
|
2266 |
WARPPOINTS gmc; |
WARPPOINTS gmc; |
2267 |
|
|
2283 |
gmc.duv[2].x= gmc.duv[2].y = 0; |
gmc.duv[2].x= gmc.duv[2].y = 0; |
2284 |
return gmc; } |
return gmc; } |
2285 |
|
|
2286 |
// filter mask of all blocks |
/* filter mask of all blocks */ |
2287 |
|
|
2288 |
for (my = 1; my < (uint32_t)MBh-1; my++) |
for (my = 1; my < (uint32_t)MBh-1; my++) |
2289 |
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2295 |
if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED) |
if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED) |
2296 |
continue; |
continue; |
2297 |
|
|
2298 |
if ( ( (ABS(mv.x - (pMB-1)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB-1)->mvs[0].y) < deltay) ) |
if ( ( (abs(mv.x - (pMB-1)->mvs[0].x) < deltax) && (abs(mv.y - (pMB-1)->mvs[0].y) < deltay) ) |
2299 |
&& ( (ABS(mv.x - (pMB+1)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB+1)->mvs[0].y) < deltay) ) |
&& ( (abs(mv.x - (pMB+1)->mvs[0].x) < deltax) && (abs(mv.y - (pMB+1)->mvs[0].y) < deltay) ) |
2300 |
&& ( (ABS(mv.x - (pMB-MBw)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB-MBw)->mvs[0].y) < deltay) ) |
&& ( (abs(mv.x - (pMB-MBw)->mvs[0].x) < deltax) && (abs(mv.y - (pMB-MBw)->mvs[0].y) < deltay) ) |
2301 |
&& ( (ABS(mv.x - (pMB+MBw)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) ) |
&& ( (abs(mv.x - (pMB+MBw)->mvs[0].x) < deltax) && (abs(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) ) |
2302 |
MBmask[mbnum]=1; |
MBmask[mbnum]=1; |
2303 |
} |
} |
2304 |
|
|
2373 |
continue; |
continue; |
2374 |
|
|
2375 |
oldnum++; |
oldnum++; |
2376 |
meanx += ABS(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ); |
meanx += fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ); |
2377 |
meany += ABS(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ); |
meany += fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ); |
2378 |
} |
} |
2379 |
|
|
2380 |
if (4*meanx > oldnum) /* better fit than 0.25 is useless */ |
if (4*meanx > oldnum) /* better fit than 0.25 is useless */ |
2401 |
if (!MBmask[mbnum]) |
if (!MBmask[mbnum]) |
2402 |
continue; |
continue; |
2403 |
|
|
2404 |
if ( ( ABS(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ) > meanx ) |
if ( ( fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ) > meanx ) |
2405 |
|| ( ABS(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ) > meany ) ) |
|| ( fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ) > meany ) ) |
2406 |
MBmask[mbnum]=0; |
MBmask[mbnum]=0; |
2407 |
else |
else |
2408 |
num++; |
num++; |
2424 |
gmc.duv[2].x=0; |
gmc.duv[2].x=0; |
2425 |
gmc.duv[2].y=0; |
gmc.duv[2].y=0; |
2426 |
} |
} |
2427 |
// fprintf(stderr,"wp1 = ( %4d, %4d) wp2 = ( %4d, %4d) \n", gmc.duv[0].x, gmc.duv[0].y, gmc.duv[1].x, gmc.duv[1].y); |
/* fprintf(stderr,"wp1 = ( %4d, %4d) wp2 = ( %4d, %4d) \n", gmc.duv[0].x, gmc.duv[0].y, gmc.duv[1].x, gmc.duv[1].y); */ |
2428 |
|
|
2429 |
free(MBmask); |
free(MBmask); |
2430 |
|
|
2431 |
return gmc; |
return gmc; |
2432 |
} |
} |
2433 |
|
|
2434 |
// functions which perform BITS-based search/bitcount |
/* functions which perform BITS-based search/bitcount */ |
2435 |
|
|
2436 |
static int |
static int |
2437 |
CountMBBitsInter(SearchData * const Data, |
CountMBBitsInter(SearchData * const Data, |
2452 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
2453 |
CheckCandidateBits16(Data->currentQMV[0].x, Data->currentQMV[0].y, 255, &iDirection, Data); |
CheckCandidateBits16(Data->currentQMV[0].x, Data->currentQMV[0].y, 255, &iDirection, Data); |
2454 |
|
|
2455 |
//checking if this vector is perfect. if it is, we stop. |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_BITS | XVID_ME_EXTSEARCH_BITS)) { /* we have to prepare for halfpixel-precision search */ |
|
if (Data->temp[0] == 0 && Data->temp[1] == 0 && Data->temp[2] == 0 && Data->temp[3] == 0) |
|
|
return 0; //quick stop |
|
|
|
|
|
if (MotionFlags & (XVID_ME_HALFPELREFINE16_BITS | XVID_ME_EXTSEARCH_BITS)) { //we have to prepare for halfpixel-precision search |
|
2456 |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
2457 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2458 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
2461 |
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
2462 |
} |
} |
2463 |
|
|
2464 |
} else { // not qpel |
} else { /* not qpel */ |
2465 |
|
|
2466 |
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
|
//checking if this vector is perfect. if it is, we stop. |
|
|
if (Data->temp[0] == 0 && Data->temp[1] == 0 && Data->temp[2] == 0 && Data->temp[3] == 0) { |
|
|
return 0; //inter |
|
|
} |
|
2467 |
} |
} |
2468 |
|
|
2469 |
if (MotionFlags&XVID_ME_EXTSEARCH_BITS) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
if (MotionFlags&XVID_ME_EXTSEARCH_BITS) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
2471 |
if (MotionFlags&XVID_ME_HALFPELREFINE16_BITS) SubpelRefine(Data); |
if (MotionFlags&XVID_ME_HALFPELREFINE16_BITS) SubpelRefine(Data); |
2472 |
|
|
2473 |
if (Data->qpel) { |
if (Data->qpel) { |
2474 |
if (MotionFlags&(XVID_ME_EXTSEARCH_BITS | XVID_ME_HALFPELREFINE16_BITS)) { // there was halfpel-precision search |
if (MotionFlags&(XVID_ME_EXTSEARCH_BITS | XVID_ME_HALFPELREFINE16_BITS)) { /* there was halfpel-precision search */ |
2475 |
for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) { |
for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) { |
2476 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // we have found a better match |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */ |
2477 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
2478 |
} |
} |
2479 |
|
|
2480 |
// preparing for qpel-precision search |
/* preparing for qpel-precision search */ |
2481 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
2482 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2483 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
2485 |
if (MotionFlags&XVID_ME_QUARTERPELREFINE16_BITS) SubpelRefine(Data); |
if (MotionFlags&XVID_ME_QUARTERPELREFINE16_BITS) SubpelRefine(Data); |
2486 |
} |
} |
2487 |
|
|
2488 |
if (MotionFlags&XVID_ME_CHECKPREDICTION_BITS) { //let's check vector equal to prediction |
if (MotionFlags&XVID_ME_CHECKPREDICTION_BITS) { /* let's check vector equal to prediction */ |
2489 |
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
2490 |
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
2491 |
CheckCandidateBits16(Data->predMV.x, Data->predMV.y, 255, &iDirection, Data); |
CheckCandidateBits16(Data->predMV.x, Data->predMV.y, 255, &iDirection, Data); |
2493 |
return Data->iMinSAD[0]; |
return Data->iMinSAD[0]; |
2494 |
} |
} |
2495 |
|
|
|
|
|
2496 |
static int |
static int |
2497 |
CountMBBitsInter4v(const SearchData * const Data, |
CountMBBitsInter4v(const SearchData * const Data, |
2498 |
MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, |
MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, |
2504 |
int cbp = 0, bits = 0, t = 0, i, iDirection; |
int cbp = 0, bits = 0, t = 0, i, iDirection; |
2505 |
SearchData Data2, *Data8 = &Data2; |
SearchData Data2, *Data8 = &Data2; |
2506 |
int sumx = 0, sumy = 0; |
int sumx = 0, sumy = 0; |
2507 |
int16_t in[64], coeff[64]; |
int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; |
2508 |
|
uint8_t * ptr; |
2509 |
|
|
2510 |
memcpy(Data8, Data, sizeof(SearchData)); |
memcpy(Data8, Data, sizeof(SearchData)); |
2511 |
CheckCandidate = CheckCandidateBits8; |
CheckCandidate = CheckCandidateBits8; |
2512 |
|
|
2513 |
for (i = 0; i < 4; i++) { |
for (i = 0; i < 4; i++) { /* for all luma blocks */ |
2514 |
|
|
2515 |
Data8->iMinSAD = Data->iMinSAD + i + 1; |
Data8->iMinSAD = Data->iMinSAD + i + 1; |
2516 |
Data8->currentMV = Data->currentMV + i + 1; |
Data8->currentMV = Data->currentMV + i + 1; |
2517 |
Data8->currentQMV = Data->currentQMV + i + 1; |
Data8->currentQMV = Data->currentQMV + i + 1; |
2518 |
Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2519 |
Data8->Ref = Data->Ref + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[0] = Data->RefP[0] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2520 |
Data8->RefH = Data->RefH + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2521 |
Data8->RefV = Data->RefV + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2522 |
Data8->RefHV = Data->RefHV + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2523 |
|
|
2524 |
if(Data->qpel) { |
if(Data->qpel) { |
2525 |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
2534 |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8, |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8, |
2535 |
pParam->width, pParam->height, Data8->iFcode, Data8->qpel, 0); |
pParam->width, pParam->height, Data8->iFcode, Data8->qpel, 0); |
2536 |
|
|
2537 |
*Data8->iMinSAD += t; |
*Data8->iMinSAD += BITS_MULT*t; |
2538 |
|
|
2539 |
Data8->qpel_precision = Data8->qpel; |
Data8->qpel_precision = Data8->qpel; |
2540 |
// checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) |
/* checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) */ |
2541 |
if (Data8->qpel) { |
{ |
2542 |
if (!(Data8->currentQMV->x == backup[i+1].x && Data8->currentQMV->y == backup[i+1].y)) |
VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV; |
2543 |
CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
if (!MVequal (*v, backup[i+1]) ) |
|
} else { |
|
|
if (!(Data8->currentMV->x == backup[i+1].x && Data8->currentMV->y == backup[i+1].y)) |
|
2544 |
CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
2545 |
} |
} |
2546 |
|
|
2547 |
if (Data8->qpel) { |
if (Data8->qpel) { |
2548 |
if (MotionFlags&XVID_ME_HALFPELREFINE8_BITS || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_BITS)) { // halfpixel motion search follows |
if (MotionFlags&XVID_ME_HALFPELREFINE8_BITS || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_BITS)) { /* halfpixel motion search follows */ |
2549 |
int32_t s = *Data8->iMinSAD; |
int32_t s = *Data8->iMinSAD; |
2550 |
Data8->currentMV->x = Data8->currentQMV->x/2; |
Data8->currentMV->x = Data8->currentQMV->x/2; |
2551 |
Data8->currentMV->y = Data8->currentQMV->y/2; |
Data8->currentMV->y = Data8->currentQMV->y/2; |
2559 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS) |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS) |
2560 |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
2561 |
|
|
2562 |
if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) SubpelRefine(Data8); |
if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) |
2563 |
|
SubpelRefine(Data8); |
2564 |
|
|
2565 |
if(s > *Data8->iMinSAD) { //we have found a better match |
if(s > *Data8->iMinSAD) { /* we have found a better match */ |
2566 |
Data8->currentQMV->x = 2*Data8->currentMV->x; |
Data8->currentQMV->x = 2*Data8->currentMV->x; |
2567 |
Data8->currentQMV->y = 2*Data8->currentMV->y; |
Data8->currentQMV->y = 2*Data8->currentMV->y; |
2568 |
} |
} |
2574 |
} |
} |
2575 |
if (MotionFlags & XVID_ME_QUARTERPELREFINE8_BITS) SubpelRefine(Data8); |
if (MotionFlags & XVID_ME_QUARTERPELREFINE8_BITS) SubpelRefine(Data8); |
2576 |
|
|
2577 |
} else // not qpel |
} else { /* not qpel */ |
|
if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) SubpelRefine(Data8); //halfpel mode, halfpel refinement |
|
2578 |
|
|
2579 |
//checking vector equal to predicion |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS) /* extsearch */ |
2580 |
|
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
2581 |
|
|
2582 |
|
if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) |
2583 |
|
SubpelRefine(Data8); /* halfpel refinement */ |
2584 |
|
} |
2585 |
|
|
2586 |
|
/* checking vector equal to predicion */ |
2587 |
if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_BITS) { |
if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_BITS) { |
2588 |
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
2589 |
if (!(Data8->predMV.x == v->x && Data8->predMV.y == v->y)) |
if (!MVequal(*v, Data8->predMV)) |
2590 |
CheckCandidateBits8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8); |
CheckCandidateBits8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8); |
2591 |
} |
} |
2592 |
|
|
2593 |
bits += *Data8->iMinSAD; |
bits += *Data8->iMinSAD; |
2594 |
if (bits >= Data->iMinSAD[0]) break; // no chances for INTER4V |
if (bits >= Data->iMinSAD[0]) return bits; /* no chances for INTER4V */ |
2595 |
|
|
2596 |
// MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else |
/* MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else */ |
2597 |
if(Data->qpel) { |
if(Data->qpel) { |
2598 |
pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x; |
pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x; |
2599 |
pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y; |
pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y; |
2609 |
pMB->mvs[i] = *Data8->currentMV; |
pMB->mvs[i] = *Data8->currentMV; |
2610 |
pMB->sad8[i] = 4 * *Data8->iMinSAD; |
pMB->sad8[i] = 4 * *Data8->iMinSAD; |
2611 |
if (Data8->temp[0]) cbp |= 1 << (5 - i); |
if (Data8->temp[0]) cbp |= 1 << (5 - i); |
|
} |
|
2612 |
|
|
2613 |
if (bits < *Data->iMinSAD) { // there is still a chance for inter4v mode. let's check chroma |
} /* /for all luma blocks */ |
2614 |
const uint8_t * ptr; |
|
2615 |
|
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
2616 |
|
|
2617 |
|
/* let's check chroma */ |
2618 |
sumx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
sumx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
2619 |
sumy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
sumy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
2620 |
|
|
2621 |
//chroma U |
/* chroma U */ |
2622 |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefCU, 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
2623 |
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
2624 |
fdct(in); |
bits += Block_CalcBits(coeff, in, Data->iQuant, Data->quant_type, &cbp, 4); |
2625 |
if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16); |
|
2626 |
else i = quant4_inter(coeff, in, Data->lambda16); |
if (bits >= *Data->iMinSAD) return bits; |
|
if (i > 0) { |
|
|
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
|
|
cbp |= 1 << (5 - 4); |
|
|
} |
|
2627 |
|
|
2628 |
if (bits < *Data->iMinSAD) { // still possible |
/* chroma V */ |
2629 |
//chroma V |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
|
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefCV, 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
|
2630 |
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
2631 |
fdct(in); |
bits += Block_CalcBits(coeff, in, Data->iQuant, Data->quant_type, &cbp, 5); |
2632 |
if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16); |
|
2633 |
else i = quant4_inter(coeff, in, Data->lambda16); |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
|
if (i > 0) { |
|
|
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
|
|
cbp |= 1 << (5 - 5); |
|
|
} |
|
|
bits += xvid_cbpy_tab[15-(cbp>>2)].len; |
|
|
bits += mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
|
|
} |
|
|
} |
|
2634 |
|
|
2635 |
return bits; |
return bits; |
2636 |
} |
} |
2637 |
|
|
|
|
|
2638 |
static int |
static int |
2639 |
CountMBBitsIntra(const SearchData * const Data) |
CountMBBitsIntra(const SearchData * const Data) |
2640 |
{ |
{ |
2641 |
int bits = 1; //this one is ac/dc prediction flag. always 1. |
int bits = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ |
2642 |
int cbp = 0, i, t, dc = 0, b_dc = 1024; |
int cbp = 0, i, dc = 0; |
2643 |
const uint32_t iQuant = Data->lambda16; |
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
|
int16_t in[64], coeff[64]; |
|
2644 |
|
|
2645 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, 1); |
|
|
|
|
2646 |
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2647 |
transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth); |
transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth); |
2648 |
fdct(in); |
bits += Block_CalcBitsIntra(coeff, in, Data->iQuant, Data->quant_type, &cbp, i, &dc); |
2649 |
b_dc = dc; |
|
2650 |
dc = in[0]; |
if (bits >= Data->iMinSAD[0]) return bits; |
2651 |
in[0] -= b_dc; |
} |
2652 |
if (Data->lambda8 == 0) quant_intra_c(coeff, in, iQuant, iDcScaler); |
|
2653 |
else quant4_intra_c(coeff, in, iQuant, iDcScaler); |
bits += BITS_MULT*xvid_cbpy_tab[cbp>>2].len; |
2654 |
|
|
2655 |
b_dc = dc; |
/*chroma U */ |
|
dc = coeff[0]; |
|
|
if (i != 0) coeff[0] -= b_dc; |
|
|
|
|
|
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcy_tab[coeff[0] + 255].len;; |
|
|
Data->temp[i] = t; |
|
|
if (t != 0) cbp |= 1 << (5 - i); |
|
|
if (bits >= Data->iMinSAD[0]) break; |
|
|
} |
|
|
|
|
|
if (bits < Data->iMinSAD[0]) { // INTRA still looks good, let's add chroma |
|
|
uint32_t iDcScaler = get_dc_scaler(iQuant, 0); |
|
|
//chroma U |
|
2656 |
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
2657 |
fdct(in); |
bits += Block_CalcBitsIntra(coeff, in, Data->iQuant, Data->quant_type, &cbp, 4, &dc); |
2658 |
in[0] -= 1024; |
|
2659 |
if (Data->lambda8 == 0) quant_intra(coeff, in, iQuant, iDcScaler); |
if (bits >= Data->iMinSAD[0]) return bits; |
2660 |
else quant4_intra(coeff, in, iQuant, iDcScaler); |
|
2661 |
|
/* chroma V */ |
|
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len; |
|
|
if (t != 0) cbp |= 1 << (5 - 4); |
|
|
|
|
|
if (bits < Data->iMinSAD[0]) { |
|
|
iDcScaler = get_dc_scaler(iQuant, 1); |
|
|
//chroma V |
|
2662 |
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
2663 |
fdct(in); |
bits += Block_CalcBitsIntra(coeff, in, Data->iQuant, Data->quant_type, &cbp, 5, &dc); |
|
in[0] -= 1024; |
|
|
if (Data->lambda8 == 0) quant_intra(coeff, in, iQuant, iDcScaler); |
|
|
else quant4_intra(coeff, in, iQuant, iDcScaler); |
|
2664 |
|
|
2665 |
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len; |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len; |
|
if (t != 0) cbp |= 1 << (5 - 5); |
|
2666 |
|
|
|
bits += xvid_cbpy_tab[cbp>>2].len; |
|
|
bits += mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len; |
|
|
} |
|
|
} |
|
2667 |
return bits; |
return bits; |
2668 |
} |
} |