1 |
/************************************************************************** |
/***************************************************************************** |
2 |
* |
* |
3 |
* Modifications: |
* XVID MPEG-4 VIDEO CODEC |
4 |
|
* - Motion Estimation module - |
5 |
* |
* |
6 |
* 01.05.2002 updated MotionEstimationBVOP |
* Copyright(C) 2002 Christoph Lampert <gruel@web.de> |
7 |
* 25.04.2002 partial prevMB conversion |
* 2002 Michael Militzer <michael@xvid.org> |
|
* 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
|
|
* 14.04.2002 added MotionEstimationBVOP() |
|
|
* 02.04.2002 add EPZS(^2) as ME algorithm, use PMV_USESQUARES to choose between |
|
|
* EPZS and EPZS^2 |
|
|
* 08.02.2002 split up PMVfast into three routines: PMVFast, PMVFast_MainLoop |
|
|
* PMVFast_Refine to support multiple searches with different start points |
|
|
* 07.01.2002 uv-block-based interpolation |
|
|
* 06.01.2002 INTER/INTRA-decision is now done before any SEARCH8 (speedup) |
|
|
* changed INTER_BIAS to 150 (as suggested by suxen_drol) |
|
|
* removed halfpel refinement step in PMVfastSearch8 + quality=5 |
|
|
* added new quality mode = 6 which performs halfpel refinement |
|
|
* filesize difference between quality 5 and 6 is smaller than 1% |
|
|
* (Isibaar) |
|
|
* 31.12.2001 PMVfastSearch16 and PMVfastSearch8 (gruel) |
|
|
* 30.12.2001 get_range/MotionSearchX simplified; blue/green bug fix |
|
|
* 22.12.2001 commented best_point==99 check |
|
|
* 19.12.2001 modified get_range (purple bug fix) |
|
|
* 15.12.2001 moved pmv displacement from mbprediction |
|
|
* 02.12.2001 motion estimation/compensation split (Isibaar) |
|
|
* 16.11.2001 rewrote/tweaked search algorithms; pross@cs.rmit.edu.au |
|
|
* 10.11.2001 support for sad16/sad8 functions |
|
|
* 28.08.2001 reactivated MODE_INTER4V for EXT_MODE |
|
|
* 24.08.2001 removed MODE_INTER4V_Q, disabled MODE_INTER4V for EXT_MODE |
|
|
* 22.08.2001 added MODE_INTER4V_Q |
|
|
* 20.08.2001 added pragma to get rid of internal compiler error with VC6 |
|
|
* idea by Cyril. Thanks. |
|
8 |
* |
* |
9 |
* Michael Militzer <isibaar@videocoding.de> |
* This program is an implementation of a part of one or more MPEG-4 |
10 |
|
* Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
11 |
|
* to use this software module in hardware or software products are |
12 |
|
* advised that its use may infringe existing patents or copyrights, and |
13 |
|
* any such use would be at such party's own risk. The original |
14 |
|
* developer of this software module and his/her company, and subsequent |
15 |
|
* editors and their companies, will have no liability for use of this |
16 |
|
* software or modifications or derivatives thereof. |
17 |
* |
* |
18 |
**************************************************************************/ |
* This program is free software; you can redistribute it and/or modify |
19 |
|
* it under the terms of the GNU General Public License as published by |
20 |
|
* the Free Software Foundation; either version 2 of the License, or |
21 |
|
* (at your option) any later version. |
22 |
|
* |
23 |
|
* This program is distributed in the hope that it will be useful, |
24 |
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of |
25 |
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
26 |
|
* GNU General Public License for more details. |
27 |
|
* |
28 |
|
* You should have received a copy of the GNU General Public License |
29 |
|
* along with this program; if not, write to the Free Software |
30 |
|
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
31 |
|
* |
32 |
|
*************************************************************************/ |
33 |
|
|
34 |
#include <assert.h> |
#include <assert.h> |
35 |
#include <stdio.h> |
#include <stdio.h> |
43 |
#include "motion.h" |
#include "motion.h" |
44 |
#include "sad.h" |
#include "sad.h" |
45 |
|
|
|
// very large value |
|
|
#define MV_MAX_ERROR (4096 * 256) |
|
|
|
|
|
// stop search if sdelta < THRESHOLD |
|
|
#define MV16_THRESHOLD 192 |
|
|
#define MV8_THRESHOLD 56 |
|
|
|
|
|
/* sad16(0,0) bias; mpeg4 spec suggests nb/2+1 */ |
|
|
/* nb = vop pixels * 2^(bpp-8) */ |
|
|
#define MV16_00_BIAS (128+1) |
|
|
#define MV8_00_BIAS (0) |
|
|
|
|
|
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
|
|
#define INTER_BIAS 512 |
|
46 |
|
|
|
/* Parameters which control inter/inter4v decision */ |
|
|
#define IMV16X16 5 |
|
|
|
|
|
/* vector map (vlc delta size) smoother parameters */ |
|
|
#define NEIGH_TEND_16X16 2 |
|
|
#define NEIGH_TEND_8X8 2 |
|
|
|
|
|
|
|
|
// fast ((A)/2)*2 |
|
|
#define EVEN(A) (((A)<0?(A)+1:(A)) & ~1) |
|
|
|
|
|
|
|
|
int32_t PMVfastSearch16( |
|
|
const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, const int y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV); |
|
|
|
|
|
int32_t EPZSSearch16( |
|
|
const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, const int y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV); |
|
|
|
|
|
|
|
|
int32_t PMVfastSearch8( |
|
|
const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, const int y, |
|
|
const int start_x, const int start_y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV); |
|
|
|
|
|
int32_t EPZSSearch8( |
|
|
const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, const int y, |
|
|
const int start_x, const int start_y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV); |
|
|
|
|
|
|
|
|
typedef int32_t (MainSearch16Func)( |
|
|
const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const uint8_t * const cur, |
|
|
const int x, const int y, |
|
|
int32_t startx, int32_t starty, |
|
|
int32_t iMinSAD, |
|
|
VECTOR * const currMV, |
|
|
const VECTOR * const pmv, |
|
|
const int32_t min_dx, const int32_t max_dx, |
|
|
const int32_t min_dy, const int32_t max_dy, |
|
|
const int32_t iEdgedWidth, |
|
|
const int32_t iDiamondSize, |
|
|
const int32_t iFcode, |
|
|
const int32_t iQuant, |
|
|
int iFound); |
|
|
|
|
|
typedef MainSearch16Func* MainSearch16FuncPtr; |
|
|
|
|
|
|
|
|
typedef int32_t (MainSearch8Func)( |
|
|
const uint8_t * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const uint8_t * const cur, |
|
|
const int x, const int y, |
|
|
int32_t startx, int32_t starty, |
|
|
int32_t iMinSAD, |
|
|
VECTOR * const currMV, |
|
|
const VECTOR * const pmv, |
|
|
const int32_t min_dx, const int32_t max_dx, |
|
|
const int32_t min_dy, const int32_t max_dy, |
|
|
const int32_t iEdgedWidth, |
|
|
const int32_t iDiamondSize, |
|
|
const int32_t iFcode, |
|
|
const int32_t iQuant, |
|
|
int iFound); |
|
|
|
|
|
typedef MainSearch8Func* MainSearch8FuncPtr; |
|
47 |
|
|
48 |
static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ |
static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ |
49 |
{ 0 ,(int)(1.00235+0.5), (int)(1.15582+0.5), (int)(1.31976+0.5), (int)(1.49591+0.5), (int)(1.68601+0.5), |
{ 0, (int) (1.00235 + 0.5), (int) (1.15582 + 0.5), (int) (1.31976 + 0.5), |
50 |
(int)(1.89187+0.5), (int)(2.11542+0.5), (int)(2.35878+0.5), (int)(2.62429+0.5), (int)(2.91455+0.5), |
(int) (1.49591 + 0.5), (int) (1.68601 + 0.5), |
51 |
(int)(3.23253+0.5), (int)(3.58158+0.5), (int)(3.96555+0.5), (int)(4.38887+0.5), (int)(4.85673+0.5), |
(int) (1.89187 + 0.5), (int) (2.11542 + 0.5), (int) (2.35878 + 0.5), |
52 |
(int)(5.37519+0.5), (int)(5.95144+0.5), (int)(6.59408+0.5), (int)(7.31349+0.5), (int)(8.12242+0.5), |
(int) (2.62429 + 0.5), (int) (2.91455 + 0.5), |
53 |
(int)(9.03669+0.5), (int)(10.0763+0.5), (int)(11.2669+0.5), (int)(12.6426+0.5), (int)(14.2493+0.5), |
(int) (3.23253 + 0.5), (int) (3.58158 + 0.5), (int) (3.96555 + 0.5), |
54 |
(int)(16.1512+0.5), (int)(18.442+0.5), (int)(21.2656+0.5), (int)(24.8580+0.5), (int)(29.6436+0.5), |
(int) (4.38887 + 0.5), (int) (4.85673 + 0.5), |
55 |
(int)(36.4949+0.5) }; |
(int) (5.37519 + 0.5), (int) (5.95144 + 0.5), (int) (6.59408 + 0.5), |
56 |
|
(int) (7.31349 + 0.5), (int) (8.12242 + 0.5), |
57 |
|
(int) (9.03669 + 0.5), (int) (10.0763 + 0.5), (int) (11.2669 + 0.5), |
58 |
|
(int) (12.6426 + 0.5), (int) (14.2493 + 0.5), |
59 |
|
(int) (16.1512 + 0.5), (int) (18.442 + 0.5), (int) (21.2656 + 0.5), |
60 |
|
(int) (24.8580 + 0.5), (int) (29.6436 + 0.5), |
61 |
|
(int) (36.4949 + 0.5) |
62 |
|
}; |
63 |
|
|
64 |
static int32_t *lambda_vec8 = lambda_vec16; /* same table for INTER and INTER4V for now*/ |
static int32_t *lambda_vec8 = lambda_vec16; /* same table for INTER and INTER4V for now*/ |
65 |
|
|
74 |
}; |
}; |
75 |
|
|
76 |
|
|
77 |
static __inline uint32_t mv_bits(int32_t component, const uint32_t iFcode) |
static __inline uint32_t |
78 |
|
mv_bits(int32_t component, |
79 |
|
const uint32_t iFcode) |
80 |
{ |
{ |
81 |
if (component == 0) |
if (component == 0) |
82 |
return 1; |
return 1; |
84 |
if (component < 0) |
if (component < 0) |
85 |
component = -component; |
component = -component; |
86 |
|
|
87 |
if (iFcode == 1) |
if (iFcode == 1) { |
|
{ |
|
88 |
if (component > 32) |
if (component > 32) |
89 |
component = 32; |
component = 32; |
90 |
|
|
101 |
} |
} |
102 |
|
|
103 |
|
|
104 |
static __inline uint32_t calc_delta_16(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) |
static __inline uint32_t |
105 |
|
calc_delta_16(const int32_t dx, |
106 |
|
const int32_t dy, |
107 |
|
const uint32_t iFcode, |
108 |
|
const uint32_t iQuant) |
109 |
{ |
{ |
110 |
return NEIGH_TEND_16X16 * lambda_vec16[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
return NEIGH_TEND_16X16 * lambda_vec16[iQuant] * (mv_bits(dx, iFcode) + |
111 |
|
mv_bits(dy, iFcode)); |
112 |
} |
} |
113 |
|
|
114 |
static __inline uint32_t calc_delta_8(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) |
static __inline uint32_t |
115 |
|
calc_delta_8(const int32_t dx, |
116 |
|
const int32_t dy, |
117 |
|
const uint32_t iFcode, |
118 |
|
const uint32_t iQuant) |
119 |
{ |
{ |
120 |
return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + |
121 |
|
mv_bits(dy, iFcode)); |
122 |
} |
} |
123 |
|
|
124 |
|
bool |
125 |
|
MotionEstimation(MBParam * const pParam, |
|
|
|
|
|
|
|
#ifndef SEARCH16 |
|
|
#define SEARCH16 PMVfastSearch16 |
|
|
//#define SEARCH16 FullSearch16 |
|
|
//#define SEARCH16 EPZSSearch16 |
|
|
#endif |
|
|
|
|
|
#ifndef SEARCH8 |
|
|
#define SEARCH8 PMVfastSearch8 |
|
|
//#define SEARCH8 EPZSSearch8 |
|
|
#endif |
|
|
|
|
|
bool MotionEstimation( |
|
|
MBParam * const pParam, |
|
126 |
FRAMEINFO * const current, |
FRAMEINFO * const current, |
127 |
FRAMEINFO * const reference, |
FRAMEINFO * const reference, |
128 |
const IMAGE * const pRefH, |
const IMAGE * const pRefH, |
129 |
const IMAGE * const pRefV, |
const IMAGE * const pRefV, |
130 |
const IMAGE * const pRefHV, |
const IMAGE * const pRefHV, |
131 |
const uint32_t iLimit) |
const uint32_t iLimit) |
|
|
|
132 |
{ |
{ |
133 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
134 |
const uint32_t iHcount = pParam->mb_height; |
const uint32_t iHcount = pParam->mb_height; |
135 |
MACROBLOCK * pMBs = current->mbs; |
MACROBLOCK *const pMBs = current->mbs; |
136 |
IMAGE * pCurrent = ¤t->image; |
MACROBLOCK *const prevMBs = reference->mbs; |
137 |
|
const IMAGE *const pCurrent = ¤t->image; |
138 |
MACROBLOCK * prevMBs = reference->mbs; // previous frame |
const IMAGE *const pRef = &reference->image; |
139 |
IMAGE * pRef = &reference->image; |
|
140 |
|
static const VECTOR zeroMV = { 0, 0 }; |
141 |
|
VECTOR predMV; |
142 |
uint32_t i, j, iIntra = 0; |
|
143 |
|
uint32_t x, y; |
144 |
VECTOR mv16; |
uint32_t iIntra = 0; |
145 |
VECTOR pmv16; |
VECTOR pmv; |
|
|
|
|
int32_t sad8 = 0; |
|
|
int32_t sad16; |
|
|
int32_t deviation; |
|
146 |
|
|
147 |
if (sadInit) |
if (sadInit) |
148 |
(*sadInit)(); |
(*sadInit)(); |
149 |
|
|
150 |
|
for (y = 0; y < iHcount; y++) { |
151 |
|
for (x = 0; x < iWcount; x ++) { |
152 |
|
|
153 |
/* eventhough we have a seperate prevMBs, |
MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
|
pmvfast/epsz does something "funny" with the previous frames data */ |
|
154 |
|
|
155 |
/* for (i = 0; i < iHcount; i++) |
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
for (j = 0; j < iWcount; j++) |
|
|
{ |
|
|
pMBs[j + i * iWcount].mvs[0] = prevMBs[j + i * iWcount].mvs[0]; |
|
|
pMBs[j + i * iWcount].mvs[1] = prevMBs[j + i * iWcount].mvs[1]; |
|
|
pMBs[j + i * iWcount].mvs[2] = prevMBs[j + i * iWcount].mvs[2]; |
|
|
pMBs[j + i * iWcount].mvs[3] = prevMBs[j + i * iWcount].mvs[3]; |
|
|
} |
|
|
*/ |
|
|
/*dprintf("*** BEFORE ***"); |
|
|
for (i = 0; i < iHcount; i++) |
|
|
for (j = 0; j < iWcount; j++) |
|
|
{ |
|
|
dprintf(" [%i,%i] mode=%i dquant=%i mvs=(%i %i %i %i) sad8=(%i %i %i %i) sad16=(%i)", j,i, |
|
|
pMBs[j + i * iWcount].mode, |
|
|
pMBs[j + i * iWcount].dquant, |
|
|
pMBs[j + i * iWcount].mvs[0], |
|
|
pMBs[j + i * iWcount].mvs[1], |
|
|
pMBs[j + i * iWcount].mvs[2], |
|
|
pMBs[j + i * iWcount].mvs[3], |
|
|
prevMBs[j + i * iWcount].sad8[0], |
|
|
prevMBs[j + i * iWcount].sad8[1], |
|
|
prevMBs[j + i * iWcount].sad8[2], |
|
|
prevMBs[j + i * iWcount].sad8[3], |
|
|
prevMBs[j + i * iWcount].sad16); |
|
|
} |
|
|
*/ |
|
|
|
|
|
// note: i==horizontal, j==vertical |
|
|
for (i = 0; i < iHcount; i++) |
|
|
for (j = 0; j < iWcount; j++) |
|
|
{ |
|
|
MACROBLOCK *pMB = &pMBs[j + i * iWcount]; |
|
|
MACROBLOCK *prevMB = &prevMBs[j + i * iWcount]; |
|
|
|
|
|
sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
j, i, current->motion_flags, current->quant, current->fcode, |
|
|
pParam, pMBs, prevMBs, &mv16, &pmv16); |
|
|
pMB->sad16=sad16; |
|
156 |
|
|
157 |
|
pMB->sad16 = |
158 |
|
SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
159 |
|
x, y, predMV.x, predMV.y, predMV.x, predMV.y, |
160 |
|
current->motion_flags, current->quant, |
161 |
|
current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, |
162 |
|
&pMB->pmvs[0]); |
163 |
|
|
164 |
/* decide: MODE_INTER or MODE_INTRA |
if (0 < (pMB->sad16 - MV16_INTER_BIAS)) { |
165 |
if (dev_intra < sad_inter - 2 * nb) use_intra |
int32_t deviation; |
|
*/ |
|
166 |
|
|
167 |
deviation = dev16(pCurrent->y + j*16 + i*16*pParam->edged_width, pParam->edged_width); |
deviation = |
168 |
|
dev16(pCurrent->y + x * 16 + y * 16 * pParam->edged_width, |
169 |
|
pParam->edged_width); |
170 |
|
|
171 |
if (deviation < (sad16 - INTER_BIAS)) |
if (deviation < (pMB->sad16 - MV16_INTER_BIAS)) { |
|
{ |
|
172 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
173 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
174 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
pMB->mvs[3] = zeroMV; |
175 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
176 |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
pMB->sad8[3] = 0; |
177 |
|
|
178 |
iIntra++; |
iIntra++; |
179 |
if(iIntra >= iLimit) |
if(iIntra >= iLimit) |
181 |
|
|
182 |
continue; |
continue; |
183 |
} |
} |
184 |
|
} |
185 |
|
|
186 |
|
pmv = pMB->pmvs[0]; |
187 |
if (current->global_flags & XVID_INTER4V) |
if (current->global_flags & XVID_INTER4V) |
188 |
{ |
if ((!(current->global_flags & XVID_LUMIMASKING) || |
189 |
pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
pMB->dquant == NO_CHANGE)) { |
190 |
2 * j, 2 * i, mv16.x, mv16.y, |
int32_t sad8 = IMV16X16 * current->quant; |
191 |
current->motion_flags, current->quant, current->fcode, |
|
192 |
pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
if (sad8 < pMB->sad16) { |
193 |
|
sad8 += pMB->sad8[0] = |
194 |
pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
195 |
2 * j + 1, 2 * i, mv16.x, mv16.y, |
pCurrent, 2 * x, 2 * y, |
196 |
current->motion_flags, current->quant, current->fcode, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
197 |
pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
current->motion_flags, |
198 |
|
current->quant, current->fcode, pParam, |
199 |
pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
pMBs, prevMBs, &pMB->mvs[0], |
200 |
2 * j, 2 * i + 1, mv16.x, mv16.y, |
&pMB->pmvs[0]); |
201 |
current->motion_flags, current->quant, current->fcode, |
} |
202 |
pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
if (sad8 < pMB->sad16) { |
203 |
|
|
204 |
pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 1); |
205 |
2 * j + 1, 2 * i + 1, mv16.x, mv16.y, |
sad8 += pMB->sad8[1] = |
206 |
current->motion_flags, current->quant, current->fcode, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
207 |
pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
pCurrent, 2 * x + 1, 2 * y, |
208 |
|
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
209 |
sad8 = pMB->sad8[0] + pMB->sad8[1] + pMB->sad8[2] + pMB->sad8[3]; |
current->motion_flags, |
210 |
|
current->quant, current->fcode, pParam, |
211 |
|
pMBs, prevMBs, &pMB->mvs[1], |
212 |
|
&pMB->pmvs[1]); |
213 |
|
} |
214 |
|
if (sad8 < pMB->sad16) { |
215 |
|
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 2); |
216 |
|
sad8 += pMB->sad8[2] = |
217 |
|
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
218 |
|
pCurrent, 2 * x, 2 * y + 1, |
219 |
|
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
220 |
|
current->motion_flags, |
221 |
|
current->quant, current->fcode, pParam, |
222 |
|
pMBs, prevMBs, &pMB->mvs[2], |
223 |
|
&pMB->pmvs[2]); |
224 |
|
} |
225 |
|
if (sad8 < pMB->sad16) { |
226 |
|
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 3); |
227 |
|
sad8 += pMB->sad8[3] = |
228 |
|
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
229 |
|
pCurrent, 2 * x + 1, 2 * y + 1, |
230 |
|
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
231 |
|
current->motion_flags, |
232 |
|
current->quant, current->fcode, pParam, |
233 |
|
pMBs, prevMBs, |
234 |
|
&pMB->mvs[3], |
235 |
|
&pMB->pmvs[3]); |
236 |
} |
} |
237 |
|
|
|
|
|
238 |
/* decide: MODE_INTER or MODE_INTER4V |
/* decide: MODE_INTER or MODE_INTER4V |
239 |
mpeg4: if (sad8 < sad16 - nb/2+1) use_inter4v |
mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
240 |
*/ |
*/ |
241 |
|
|
242 |
if (!(current->global_flags & XVID_LUMIMASKING) || pMB->dquant == NO_CHANGE) |
if (sad8 < pMB->sad16) { |
|
{ |
|
|
if (((current->global_flags & XVID_INTER4V)==0) || |
|
|
(sad16 < (sad8 + (int32_t)(IMV16X16 * current->quant)))) |
|
|
{ |
|
|
|
|
|
sad8 = sad16; |
|
|
pMB->mode = MODE_INTER; |
|
|
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = mv16.x; |
|
|
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = mv16.y; |
|
|
pMB->pmvs[0].x = pmv16.x; |
|
|
pMB->pmvs[0].y = pmv16.y; |
|
|
} |
|
|
else |
|
243 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
244 |
|
pMB->sad8[0] *= 4; |
245 |
|
pMB->sad8[1] *= 4; |
246 |
|
pMB->sad8[2] *= 4; |
247 |
|
pMB->sad8[3] *= 4; |
248 |
|
continue; |
249 |
} |
} |
250 |
else |
|
|
{ |
|
|
sad8 = sad16; |
|
|
pMB->mode = MODE_INTER; |
|
|
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = mv16.x; |
|
|
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = mv16.y; |
|
|
pMB->pmvs[0].x = pmv16.x; |
|
|
pMB->pmvs[0].y = pmv16.y; |
|
|
} |
|
251 |
} |
} |
252 |
|
|
253 |
/* dprintf("*** AFTER ***", pMBs[0].b_mvs[0].x); |
pMB->mode = MODE_INTER; |
254 |
for (i = 0; i < iHcount; i++) |
pMB->pmvs[0] = pmv; /* pMB->pmvs[1] = pMB->pmvs[2] = pMB->pmvs[3] are not needed for INTER */ |
255 |
for (j = 0; j < iWcount; j++) |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; |
256 |
{ |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = |
257 |
dprintf(" [%i,%i] mode=%i dquant=%i mvs=(%i %i %i %i) sad8=(%i %i %i %i) sad16=(%i)", j,i, |
pMB->sad16; |
258 |
pMBs[j + i * iWcount].mode, |
} |
|
pMBs[j + i * iWcount].dquant, |
|
|
pMBs[j + i * iWcount].mvs[0], |
|
|
pMBs[j + i * iWcount].mvs[1], |
|
|
pMBs[j + i * iWcount].mvs[2], |
|
|
pMBs[j + i * iWcount].mvs[3], |
|
|
pMBs[j + i * iWcount].sad8[0], |
|
|
pMBs[j + i * iWcount].sad8[1], |
|
|
pMBs[j + i * iWcount].sad8[2], |
|
|
pMBs[j + i * iWcount].sad8[3], |
|
|
pMBs[j + i * iWcount].sad16); |
|
259 |
} |
} |
|
*/ |
|
260 |
|
|
261 |
return 0; |
return 0; |
262 |
} |
} |
263 |
|
|
|
#define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) |
|
|
|
|
|
#define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) |
|
|
|
|
264 |
|
|
265 |
#define CHECK_MV16_ZERO {\ |
#define CHECK_MV16_ZERO {\ |
266 |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
267 |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
268 |
{ \ |
{ \ |
269 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ |
270 |
iSAD += calc_delta_16(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ |
271 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
272 |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ |
273 |
} |
} |
274 |
|
|
275 |
#define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
#define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
276 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
277 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
278 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
279 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
280 |
} |
} |
284 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
285 |
{ \ |
{ \ |
286 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
287 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
288 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
289 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
290 |
} |
} |
294 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
295 |
{ \ |
{ \ |
296 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
297 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
298 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
299 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
300 |
} |
} |
304 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
305 |
{ \ |
{ \ |
306 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
307 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
308 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
309 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
310 |
} |
} |
312 |
|
|
313 |
#define CHECK_MV8_ZERO {\ |
#define CHECK_MV8_ZERO {\ |
314 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ |
315 |
iSAD += calc_delta_8(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ |
316 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
317 |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ |
318 |
} |
} |
320 |
#define NOCHECK_MV8_CANDIDATE(X,Y) \ |
#define NOCHECK_MV8_CANDIDATE(X,Y) \ |
321 |
{ \ |
{ \ |
322 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
323 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
324 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
325 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
326 |
} |
} |
330 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
331 |
{ \ |
{ \ |
332 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
333 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
334 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
335 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
336 |
} |
} |
340 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
341 |
{ \ |
{ \ |
342 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
343 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
344 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
345 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
346 |
} |
} |
350 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
351 |
{ \ |
{ \ |
352 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
353 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
354 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
355 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
356 |
} |
} |
357 |
|
|
358 |
|
#if 0 |
359 |
/* too slow and not fully functional at the moment */ |
/* too slow and not fully functional at the moment */ |
|
/* |
|
360 |
int32_t ZeroSearch16( |
int32_t ZeroSearch16( |
361 |
const uint8_t * const pRef, |
const uint8_t * const pRef, |
362 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
376 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
377 |
const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; |
const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; |
378 |
int32_t iSAD; |
int32_t iSAD; |
379 |
int32_t pred_x,pred_y; |
VECTOR pred; |
380 |
|
|
381 |
get_pmv(pMBs, x, y, pParam->mb_width, 0, &pred_x, &pred_y); |
|
382 |
|
pred = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
383 |
|
|
384 |
iSAD = sad16( cur, |
iSAD = sad16( cur, |
385 |
get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0,0, iEdgedWidth), |
get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0,0, iEdgedWidth), |
389 |
|
|
390 |
currMV->x = 0; |
currMV->x = 0; |
391 |
currMV->y = 0; |
currMV->y = 0; |
392 |
currPMV->x = -pred_x; |
currPMV->x = -pred.x; |
393 |
currPMV->y = -pred_y; |
currPMV->y = -pred.y; |
394 |
|
|
395 |
return iSAD; |
return iSAD; |
396 |
|
|
397 |
} |
} |
398 |
*/ |
#endif /* 0 */ |
399 |
|
|
400 |
int32_t Diamond16_MainSearch( |
int32_t |
401 |
const uint8_t * const pRef, |
Diamond16_MainSearch(const uint8_t * const pRef, |
402 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
403 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
404 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
405 |
const uint8_t * const cur, |
const uint8_t * const cur, |
406 |
const int x, const int y, |
const int x, |
407 |
int32_t startx, int32_t starty, |
const int y, |
408 |
int32_t iMinSAD, |
const int start_x, |
409 |
|
const int start_y, |
410 |
|
int iMinSAD, |
411 |
VECTOR * const currMV, |
VECTOR * const currMV, |
412 |
const VECTOR * const pmv, |
const int center_x, |
413 |
const int32_t min_dx, const int32_t max_dx, |
const int center_y, |
414 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t min_dx, |
415 |
|
const int32_t max_dx, |
416 |
|
const int32_t min_dy, |
417 |
|
const int32_t max_dy, |
418 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
419 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
420 |
const int32_t iFcode, |
const int32_t iFcode, |
424 |
/* Do a diamond search around given starting point, return SAD of best */ |
/* Do a diamond search around given starting point, return SAD of best */ |
425 |
|
|
426 |
int32_t iDirection=0; |
int32_t iDirection=0; |
427 |
|
int32_t iDirectionBackup; |
428 |
int32_t iSAD; |
int32_t iSAD; |
429 |
VECTOR backupMV; |
VECTOR backupMV; |
430 |
backupMV.x = startx; |
|
431 |
backupMV.y = starty; |
backupMV.x = start_x; |
432 |
|
backupMV.y = start_y; |
433 |
|
|
434 |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
435 |
|
|
438 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
439 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
440 |
|
|
441 |
if (iDirection) |
if (iDirection) { |
442 |
while (!iFound) |
while (!iFound) { |
|
{ |
|
443 |
iFound = 1; |
iFound = 1; |
444 |
backupMV=*currMV; |
backupMV=*currMV; |
445 |
|
iDirectionBackup = iDirection; |
446 |
|
|
447 |
if ( iDirection != 2) |
if (iDirectionBackup != 2) |
448 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
449 |
if ( iDirection != 1) |
backupMV.y, 1); |
450 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x+iDiamondSize,backupMV.y,2); |
if (iDirectionBackup != 1) |
451 |
if ( iDirection != 4) |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
452 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x,backupMV.y-iDiamondSize,3); |
backupMV.y, 2); |
453 |
if ( iDirection != 3) |
if (iDirectionBackup != 4) |
454 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
455 |
|
backupMV.y - iDiamondSize, 3); |
456 |
|
if (iDirectionBackup != 3) |
457 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
458 |
|
backupMV.y + iDiamondSize, 4); |
459 |
} |
} |
460 |
else |
} else { |
461 |
{ |
currMV->x = start_x; |
462 |
currMV->x = startx; |
currMV->y = start_y; |
|
currMV->y = starty; |
|
463 |
} |
} |
464 |
return iMinSAD; |
return iMinSAD; |
465 |
} |
} |
466 |
|
|
467 |
int32_t Square16_MainSearch( |
int32_t |
468 |
const uint8_t * const pRef, |
Square16_MainSearch(const uint8_t * const pRef, |
469 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
470 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
471 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
472 |
const uint8_t * const cur, |
const uint8_t * const cur, |
473 |
const int x, const int y, |
const int x, |
474 |
int32_t startx, int32_t starty, |
const int y, |
475 |
int32_t iMinSAD, |
const int start_x, |
476 |
|
const int start_y, |
477 |
|
int iMinSAD, |
478 |
VECTOR * const currMV, |
VECTOR * const currMV, |
479 |
const VECTOR * const pmv, |
const int center_x, |
480 |
const int32_t min_dx, const int32_t max_dx, |
const int center_y, |
481 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t min_dx, |
482 |
|
const int32_t max_dx, |
483 |
|
const int32_t min_dy, |
484 |
|
const int32_t max_dy, |
485 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
486 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
487 |
const int32_t iFcode, |
const int32_t iFcode, |
493 |
int32_t iDirection=0; |
int32_t iDirection=0; |
494 |
int32_t iSAD; |
int32_t iSAD; |
495 |
VECTOR backupMV; |
VECTOR backupMV; |
496 |
backupMV.x = startx; |
|
497 |
backupMV.y = starty; |
backupMV.x = start_x; |
498 |
|
backupMV.y = start_y; |
499 |
|
|
500 |
/* It's one search with full square pattern, and new parts for all following diamonds */ |
/* It's one search with full square pattern, and new parts for all following diamonds */ |
501 |
|
|
510 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
511 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
512 |
|
|
513 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
514 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
backupMV.y - iDiamondSize, 5); |
515 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
516 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
backupMV.y + iDiamondSize, 6); |
517 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
518 |
|
backupMV.y - iDiamondSize, 7); |
519 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
520 |
|
backupMV.y + iDiamondSize, 8); |
521 |
|
|
522 |
|
|
523 |
if (iDirection) |
if (iDirection) { |
524 |
while (!iFound) |
while (!iFound) { |
|
{ |
|
525 |
iFound = 1; |
iFound = 1; |
526 |
backupMV=*currMV; |
backupMV=*currMV; |
527 |
|
|
528 |
switch (iDirection) |
switch (iDirection) { |
|
{ |
|
529 |
case 1: |
case 1: |
530 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
531 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
backupMV.y, 1); |
532 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
533 |
|
backupMV.y - iDiamondSize, 5); |
534 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
535 |
|
backupMV.y - iDiamondSize, 7); |
536 |
break; |
break; |
537 |
case 2: |
case 2: |
538 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
539 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
2); |
540 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
541 |
|
backupMV.y + iDiamondSize, 6); |
542 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
543 |
|
backupMV.y + iDiamondSize, 8); |
544 |
break; |
break; |
545 |
|
|
546 |
case 3: |
case 3: |
547 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
548 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
4); |
549 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
550 |
|
backupMV.y - iDiamondSize, 7); |
551 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
552 |
|
backupMV.y + iDiamondSize, 8); |
553 |
break; |
break; |
554 |
|
|
555 |
case 4: |
case 4: |
556 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
557 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
3); |
558 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
559 |
|
backupMV.y - iDiamondSize, 5); |
560 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
561 |
|
backupMV.y + iDiamondSize, 6); |
562 |
break; |
break; |
563 |
|
|
564 |
case 5: |
case 5: |
565 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
566 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
1); |
567 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
568 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
3); |
569 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
570 |
|
backupMV.y - iDiamondSize, 5); |
571 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
572 |
|
backupMV.y + iDiamondSize, 6); |
573 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
574 |
|
backupMV.y - iDiamondSize, 7); |
575 |
break; |
break; |
576 |
|
|
577 |
case 6: |
case 6: |
578 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
579 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
2); |
580 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
581 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
3); |
582 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
|
583 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
584 |
|
backupMV.y - iDiamondSize, 5); |
585 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
586 |
|
backupMV.y + iDiamondSize, 6); |
587 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
588 |
|
backupMV.y + iDiamondSize, 8); |
589 |
|
|
590 |
break; |
break; |
591 |
|
|
592 |
case 7: |
case 7: |
593 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
594 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
backupMV.y, 1); |
595 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
596 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
4); |
597 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
598 |
|
backupMV.y - iDiamondSize, 5); |
599 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
600 |
|
backupMV.y - iDiamondSize, 7); |
601 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
602 |
|
backupMV.y + iDiamondSize, 8); |
603 |
break; |
break; |
604 |
|
|
605 |
case 8: |
case 8: |
606 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
607 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
2); |
608 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
609 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
4); |
610 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
611 |
|
backupMV.y + iDiamondSize, 6); |
612 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
613 |
|
backupMV.y - iDiamondSize, 7); |
614 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
615 |
|
backupMV.y + iDiamondSize, 8); |
616 |
break; |
break; |
617 |
default: |
default: |
618 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
619 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); |
1); |
620 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
621 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
2); |
622 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
623 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
3); |
624 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
625 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
4); |
626 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
|
627 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
628 |
|
backupMV.y - iDiamondSize, 5); |
629 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
630 |
|
backupMV.y + iDiamondSize, 6); |
631 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
632 |
|
backupMV.y - iDiamondSize, 7); |
633 |
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
634 |
|
backupMV.y + iDiamondSize, 8); |
635 |
break; |
break; |
636 |
} |
} |
637 |
} |
} |
638 |
else |
} else { |
639 |
{ |
currMV->x = start_x; |
640 |
currMV->x = startx; |
currMV->y = start_y; |
|
currMV->y = starty; |
|
641 |
} |
} |
642 |
return iMinSAD; |
return iMinSAD; |
643 |
} |
} |
644 |
|
|
645 |
|
|
646 |
int32_t Full16_MainSearch( |
int32_t |
647 |
const uint8_t * const pRef, |
Full16_MainSearch(const uint8_t * const pRef, |
648 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
649 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
650 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
651 |
const uint8_t * const cur, |
const uint8_t * const cur, |
652 |
const int x, const int y, |
const int x, |
653 |
int32_t startx, int32_t starty, |
const int y, |
654 |
int32_t iMinSAD, |
const int start_x, |
655 |
|
const int start_y, |
656 |
|
int iMinSAD, |
657 |
VECTOR * const currMV, |
VECTOR * const currMV, |
658 |
const VECTOR * const pmv, |
const int center_x, |
659 |
const int32_t min_dx, const int32_t max_dx, |
const int center_y, |
660 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t min_dx, |
661 |
|
const int32_t max_dx, |
662 |
|
const int32_t min_dy, |
663 |
|
const int32_t max_dy, |
664 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
665 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
666 |
const int32_t iFcode, |
const int32_t iFcode, |
670 |
int32_t iSAD; |
int32_t iSAD; |
671 |
int32_t dx,dy; |
int32_t dx,dy; |
672 |
VECTOR backupMV; |
VECTOR backupMV; |
673 |
backupMV.x = startx; |
|
674 |
backupMV.y = starty; |
backupMV.x = start_x; |
675 |
|
backupMV.y = start_y; |
676 |
|
|
677 |
for (dx = min_dx; dx<=max_dx; dx+=iDiamondSize) |
for (dx = min_dx; dx<=max_dx; dx+=iDiamondSize) |
678 |
for (dy = min_dy; dy<= max_dy; dy+=iDiamondSize) |
for (dy = min_dy; dy<= max_dy; dy+=iDiamondSize) |
681 |
return iMinSAD; |
return iMinSAD; |
682 |
} |
} |
683 |
|
|
684 |
int32_t Full8_MainSearch( |
int32_t |
685 |
const uint8_t * const pRef, |
AdvDiamond16_MainSearch(const uint8_t * const pRef, |
686 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
687 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
688 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
689 |
const uint8_t * const cur, |
const uint8_t * const cur, |
690 |
const int x, const int y, |
const int x, |
691 |
int32_t startx, int32_t starty, |
const int y, |
692 |
int32_t iMinSAD, |
const int start_xi, |
693 |
|
const int start_yi, |
694 |
|
int iMinSAD, |
695 |
|
VECTOR * const currMV, |
696 |
|
const int center_x, |
697 |
|
const int center_y, |
698 |
|
const int32_t min_dx, |
699 |
|
const int32_t max_dx, |
700 |
|
const int32_t min_dy, |
701 |
|
const int32_t max_dy, |
702 |
|
const int32_t iEdgedWidth, |
703 |
|
const int32_t iDiamondSize, |
704 |
|
const int32_t iFcode, |
705 |
|
const int32_t iQuant, |
706 |
|
int iDirection) |
707 |
|
{ |
708 |
|
|
709 |
|
int32_t iSAD; |
710 |
|
int start_x = start_xi, start_y = start_yi; |
711 |
|
|
712 |
|
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
713 |
|
|
714 |
|
if (iDirection) { |
715 |
|
CHECK_MV16_CANDIDATE(start_x - iDiamondSize, start_y); |
716 |
|
CHECK_MV16_CANDIDATE(start_x + iDiamondSize, start_y); |
717 |
|
CHECK_MV16_CANDIDATE(start_x, start_y - iDiamondSize); |
718 |
|
CHECK_MV16_CANDIDATE(start_x, start_y + iDiamondSize); |
719 |
|
} else { |
720 |
|
int bDirection = 1 + 2 + 4 + 8; |
721 |
|
|
722 |
|
do { |
723 |
|
iDirection = 0; |
724 |
|
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
725 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
726 |
|
|
727 |
|
if (bDirection & 2) |
728 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
729 |
|
|
730 |
|
if (bDirection & 4) |
731 |
|
CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
732 |
|
|
733 |
|
if (bDirection & 8) |
734 |
|
CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
735 |
|
|
736 |
|
/* now we're doing diagonal checks near our candidate */ |
737 |
|
|
738 |
|
if (iDirection) //checking if anything found |
739 |
|
{ |
740 |
|
bDirection = iDirection; |
741 |
|
iDirection = 0; |
742 |
|
start_x = currMV->x; |
743 |
|
start_y = currMV->y; |
744 |
|
if (bDirection & 3) //our candidate is left or right |
745 |
|
{ |
746 |
|
CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
747 |
|
CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
748 |
|
} else // what remains here is up or down |
749 |
|
{ |
750 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
751 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
752 |
|
} |
753 |
|
|
754 |
|
if (iDirection) { |
755 |
|
bDirection += iDirection; |
756 |
|
start_x = currMV->x; |
757 |
|
start_y = currMV->y; |
758 |
|
} |
759 |
|
} else //about to quit, eh? not so fast.... |
760 |
|
{ |
761 |
|
switch (bDirection) { |
762 |
|
case 2: |
763 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
764 |
|
start_y - iDiamondSize, 2 + 4); |
765 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
766 |
|
start_y + iDiamondSize, 2 + 8); |
767 |
|
break; |
768 |
|
case 1: |
769 |
|
|
770 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
771 |
|
start_y - iDiamondSize, 1 + 4); |
772 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
773 |
|
start_y + iDiamondSize, 1 + 8); |
774 |
|
break; |
775 |
|
case 2 + 4: |
776 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
777 |
|
start_y - iDiamondSize, 1 + 4); |
778 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
779 |
|
start_y - iDiamondSize, 2 + 4); |
780 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
781 |
|
start_y + iDiamondSize, 2 + 8); |
782 |
|
break; |
783 |
|
case 4: |
784 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
785 |
|
start_y - iDiamondSize, 2 + 4); |
786 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
787 |
|
start_y - iDiamondSize, 1 + 4); |
788 |
|
break; |
789 |
|
case 8: |
790 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
791 |
|
start_y + iDiamondSize, 2 + 8); |
792 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
793 |
|
start_y + iDiamondSize, 1 + 8); |
794 |
|
break; |
795 |
|
case 1 + 4: |
796 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
797 |
|
start_y + iDiamondSize, 1 + 8); |
798 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
799 |
|
start_y - iDiamondSize, 1 + 4); |
800 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
801 |
|
start_y - iDiamondSize, 2 + 4); |
802 |
|
break; |
803 |
|
case 2 + 8: |
804 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
805 |
|
start_y - iDiamondSize, 1 + 4); |
806 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
807 |
|
start_y + iDiamondSize, 1 + 8); |
808 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
809 |
|
start_y + iDiamondSize, 2 + 8); |
810 |
|
break; |
811 |
|
case 1 + 8: |
812 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
813 |
|
start_y - iDiamondSize, 2 + 4); |
814 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
815 |
|
start_y + iDiamondSize, 2 + 8); |
816 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
817 |
|
start_y + iDiamondSize, 1 + 8); |
818 |
|
break; |
819 |
|
default: //1+2+4+8 == we didn't find anything at all |
820 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
821 |
|
start_y - iDiamondSize, 1 + 4); |
822 |
|
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
823 |
|
start_y + iDiamondSize, 1 + 8); |
824 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
825 |
|
start_y - iDiamondSize, 2 + 4); |
826 |
|
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
827 |
|
start_y + iDiamondSize, 2 + 8); |
828 |
|
break; |
829 |
|
} |
830 |
|
if (!iDirection) |
831 |
|
break; //ok, the end. really |
832 |
|
else { |
833 |
|
bDirection = iDirection; |
834 |
|
start_x = currMV->x; |
835 |
|
start_y = currMV->y; |
836 |
|
} |
837 |
|
} |
838 |
|
} |
839 |
|
while (1); //forever |
840 |
|
} |
841 |
|
|
842 |
|
return iMinSAD; |
843 |
|
} |
844 |
|
|
845 |
|
int32_t |
846 |
|
AdvDiamond8_MainSearch(const uint8_t * const pRef, |
847 |
|
const uint8_t * const pRefH, |
848 |
|
const uint8_t * const pRefV, |
849 |
|
const uint8_t * const pRefHV, |
850 |
|
const uint8_t * const cur, |
851 |
|
const int x, |
852 |
|
const int y, |
853 |
|
const int start_xi, |
854 |
|
const int start_yi, |
855 |
|
int iMinSAD, |
856 |
|
VECTOR * const currMV, |
857 |
|
const int center_x, |
858 |
|
const int center_y, |
859 |
|
const int32_t min_dx, |
860 |
|
const int32_t max_dx, |
861 |
|
const int32_t min_dy, |
862 |
|
const int32_t max_dy, |
863 |
|
const int32_t iEdgedWidth, |
864 |
|
const int32_t iDiamondSize, |
865 |
|
const int32_t iFcode, |
866 |
|
const int32_t iQuant, |
867 |
|
int iDirection) |
868 |
|
{ |
869 |
|
|
870 |
|
int32_t iSAD; |
871 |
|
int start_x = start_xi, start_y = start_yi; |
872 |
|
|
873 |
|
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
874 |
|
|
875 |
|
if (iDirection) { |
876 |
|
CHECK_MV8_CANDIDATE(start_x - iDiamondSize, start_y); |
877 |
|
CHECK_MV8_CANDIDATE(start_x + iDiamondSize, start_y); |
878 |
|
CHECK_MV8_CANDIDATE(start_x, start_y - iDiamondSize); |
879 |
|
CHECK_MV8_CANDIDATE(start_x, start_y + iDiamondSize); |
880 |
|
} else { |
881 |
|
int bDirection = 1 + 2 + 4 + 8; |
882 |
|
|
883 |
|
do { |
884 |
|
iDirection = 0; |
885 |
|
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
886 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
887 |
|
|
888 |
|
if (bDirection & 2) |
889 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
890 |
|
|
891 |
|
if (bDirection & 4) |
892 |
|
CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
893 |
|
|
894 |
|
if (bDirection & 8) |
895 |
|
CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
896 |
|
|
897 |
|
/* now we're doing diagonal checks near our candidate */ |
898 |
|
|
899 |
|
if (iDirection) //checking if anything found |
900 |
|
{ |
901 |
|
bDirection = iDirection; |
902 |
|
iDirection = 0; |
903 |
|
start_x = currMV->x; |
904 |
|
start_y = currMV->y; |
905 |
|
if (bDirection & 3) //our candidate is left or right |
906 |
|
{ |
907 |
|
CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
908 |
|
CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
909 |
|
} else // what remains here is up or down |
910 |
|
{ |
911 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
912 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
913 |
|
} |
914 |
|
|
915 |
|
if (iDirection) { |
916 |
|
bDirection += iDirection; |
917 |
|
start_x = currMV->x; |
918 |
|
start_y = currMV->y; |
919 |
|
} |
920 |
|
} else //about to quit, eh? not so fast.... |
921 |
|
{ |
922 |
|
switch (bDirection) { |
923 |
|
case 2: |
924 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
925 |
|
start_y - iDiamondSize, 2 + 4); |
926 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
927 |
|
start_y + iDiamondSize, 2 + 8); |
928 |
|
break; |
929 |
|
case 1: |
930 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
931 |
|
start_y - iDiamondSize, 1 + 4); |
932 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
933 |
|
start_y + iDiamondSize, 1 + 8); |
934 |
|
break; |
935 |
|
case 2 + 4: |
936 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
937 |
|
start_y - iDiamondSize, 1 + 4); |
938 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
939 |
|
start_y - iDiamondSize, 2 + 4); |
940 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
941 |
|
start_y + iDiamondSize, 2 + 8); |
942 |
|
break; |
943 |
|
case 4: |
944 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
945 |
|
start_y - iDiamondSize, 2 + 4); |
946 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
947 |
|
start_y - iDiamondSize, 1 + 4); |
948 |
|
break; |
949 |
|
case 8: |
950 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
951 |
|
start_y + iDiamondSize, 2 + 8); |
952 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
953 |
|
start_y + iDiamondSize, 1 + 8); |
954 |
|
break; |
955 |
|
case 1 + 4: |
956 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
957 |
|
start_y + iDiamondSize, 1 + 8); |
958 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
959 |
|
start_y - iDiamondSize, 1 + 4); |
960 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
961 |
|
start_y - iDiamondSize, 2 + 4); |
962 |
|
break; |
963 |
|
case 2 + 8: |
964 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
965 |
|
start_y - iDiamondSize, 1 + 4); |
966 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
967 |
|
start_y + iDiamondSize, 1 + 8); |
968 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
969 |
|
start_y + iDiamondSize, 2 + 8); |
970 |
|
break; |
971 |
|
case 1 + 8: |
972 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
973 |
|
start_y - iDiamondSize, 2 + 4); |
974 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
975 |
|
start_y + iDiamondSize, 2 + 8); |
976 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
977 |
|
start_y + iDiamondSize, 1 + 8); |
978 |
|
break; |
979 |
|
default: //1+2+4+8 == we didn't find anything at all |
980 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
981 |
|
start_y - iDiamondSize, 1 + 4); |
982 |
|
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
983 |
|
start_y + iDiamondSize, 1 + 8); |
984 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
985 |
|
start_y - iDiamondSize, 2 + 4); |
986 |
|
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
987 |
|
start_y + iDiamondSize, 2 + 8); |
988 |
|
break; |
989 |
|
} |
990 |
|
if (!(iDirection)) |
991 |
|
break; //ok, the end. really |
992 |
|
else { |
993 |
|
bDirection = iDirection; |
994 |
|
start_x = currMV->x; |
995 |
|
start_y = currMV->y; |
996 |
|
} |
997 |
|
} |
998 |
|
} |
999 |
|
while (1); //forever |
1000 |
|
} |
1001 |
|
return iMinSAD; |
1002 |
|
} |
1003 |
|
|
1004 |
|
|
1005 |
|
int32_t |
1006 |
|
Full8_MainSearch(const uint8_t * const pRef, |
1007 |
|
const uint8_t * const pRefH, |
1008 |
|
const uint8_t * const pRefV, |
1009 |
|
const uint8_t * const pRefHV, |
1010 |
|
const uint8_t * const cur, |
1011 |
|
const int x, |
1012 |
|
const int y, |
1013 |
|
const int start_x, |
1014 |
|
const int start_y, |
1015 |
|
int iMinSAD, |
1016 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1017 |
const VECTOR * const pmv, |
const int center_x, |
1018 |
const int32_t min_dx, const int32_t max_dx, |
const int center_y, |
1019 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t min_dx, |
1020 |
|
const int32_t max_dx, |
1021 |
|
const int32_t min_dy, |
1022 |
|
const int32_t max_dy, |
1023 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
1024 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
1025 |
const int32_t iFcode, |
const int32_t iFcode, |
1029 |
int32_t iSAD; |
int32_t iSAD; |
1030 |
int32_t dx,dy; |
int32_t dx,dy; |
1031 |
VECTOR backupMV; |
VECTOR backupMV; |
1032 |
backupMV.x = startx; |
|
1033 |
backupMV.y = starty; |
backupMV.x = start_x; |
1034 |
|
backupMV.y = start_y; |
1035 |
|
|
1036 |
for (dx = min_dx; dx<=max_dx; dx+=iDiamondSize) |
for (dx = min_dx; dx<=max_dx; dx+=iDiamondSize) |
1037 |
for (dy = min_dy; dy<= max_dy; dy+=iDiamondSize) |
for (dy = min_dy; dy<= max_dy; dy+=iDiamondSize) |
1040 |
return iMinSAD; |
return iMinSAD; |
1041 |
} |
} |
1042 |
|
|
1043 |
|
Halfpel8_RefineFuncPtr Halfpel8_Refine; |
1044 |
|
|
1045 |
|
int32_t |
1046 |
int32_t Halfpel16_Refine( |
Halfpel16_Refine(const uint8_t * const pRef, |
|
const uint8_t * const pRef, |
|
1047 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1048 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1049 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1050 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1051 |
const int x, const int y, |
const int x, |
1052 |
|
const int y, |
1053 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1054 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1055 |
const VECTOR * const pmv, |
const int center_x, |
1056 |
const int32_t min_dx, const int32_t max_dx, |
const int center_y, |
1057 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t min_dx, |
1058 |
|
const int32_t max_dx, |
1059 |
|
const int32_t min_dy, |
1060 |
|
const int32_t max_dy, |
1061 |
const int32_t iFcode, |
const int32_t iFcode, |
1062 |
const int32_t iQuant, |
const int32_t iQuant, |
1063 |
const int32_t iEdgedWidth) |
const int32_t iEdgedWidth) |
1082 |
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
1083 |
|
|
1084 |
|
|
1085 |
int32_t PMVfastSearch16( |
|
1086 |
const uint8_t * const pRef, |
int32_t |
1087 |
|
PMVfastSearch16(const uint8_t * const pRef, |
1088 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1089 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1090 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1091 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1092 |
const int x, const int y, |
const int x, |
1093 |
|
const int y, |
1094 |
|
const int start_x, /* start is searched first, so it should contain the most */ |
1095 |
|
const int start_y, /* likely motion vector for this block */ |
1096 |
|
const int center_x, /* center is from where length of MVs is measured */ |
1097 |
|
const int center_y, |
1098 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1099 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1100 |
const uint32_t iFcode, |
const uint32_t iFcode, |
1126 |
VECTOR pmv[4]; |
VECTOR pmv[4]; |
1127 |
int32_t psad[4]; |
int32_t psad[4]; |
1128 |
|
|
1129 |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
MainSearch16FuncPtr MainSearchPtr; |
1130 |
|
|
1131 |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1132 |
|
|
1133 |
static int32_t threshA,threshB; |
int32_t threshA, threshB; |
1134 |
int32_t bPredEq; |
int32_t bPredEq; |
1135 |
int32_t iMinSAD,iSAD; |
int32_t iMinSAD,iSAD; |
1136 |
|
|
1137 |
/* Get maximum range */ |
/* Get maximum range */ |
1138 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, |
1139 |
x, y, 16, iWidth, iHeight, iFcode); |
iFcode); |
1140 |
|
|
1141 |
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
1142 |
|
|
1143 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16)) { |
1144 |
{ min_dx = EVEN(min_dx); |
min_dx = EVEN(min_dx); |
1145 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
1146 |
min_dy = EVEN(min_dy); |
min_dy = EVEN(min_dy); |
1147 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
1148 |
} /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
} |
|
|
|
1149 |
|
|
1150 |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
1151 |
|
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
1152 |
|
|
1153 |
if ((x==0) && (y==0) ) |
if ((x == 0) && (y == 0)) { |
|
{ |
|
1154 |
threshA = 512; |
threshA = 512; |
1155 |
threshB = 1024; |
threshB = 1024; |
1156 |
|
} else { |
|
} |
|
|
else |
|
|
{ |
|
1157 |
threshA = psad[0]; |
threshA = psad[0]; |
1158 |
threshB = threshA+256; |
threshB = threshA+256; |
1159 |
if (threshA< 512) threshA = 512; |
if (threshA < 512) |
1160 |
if (threshA>1024) threshA = 1024; |
threshA = 512; |
1161 |
if (threshB>1792) threshB = 1792; |
if (threshA > 1024) |
1162 |
|
threshA = 1024; |
1163 |
|
if (threshB > 1792) |
1164 |
|
threshB = 1792; |
1165 |
} |
} |
1166 |
|
|
1167 |
iFound=0; |
iFound=0; |
1168 |
|
|
|
/* Step 2: Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
|
|
vector of the median. |
|
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
|
|
*/ |
|
|
|
|
|
if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[0]) ) ) |
|
|
iFound=2; |
|
|
|
|
|
/* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
|
|
Otherwise select large Diamond Search. |
|
|
*/ |
|
|
|
|
|
if ( (pmv[0].x != 0) || (pmv[0].y != 0) || (threshB<1536) || (bPredEq) ) |
|
|
iDiamondSize=1; // halfpel! |
|
|
else |
|
|
iDiamondSize=2; // halfpel! |
|
|
|
|
|
if (!(MotionFlags & PMV_HALFPELDIAMOND16) ) |
|
|
iDiamondSize*=2; |
|
|
|
|
1169 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
1170 |
MinSAD=SAD |
MinSAD=SAD |
1171 |
If Motion Vector equal to Previous frame motion vector |
If Motion Vector equal to Previous frame motion vector |
1173 |
If SAD<=256 goto Step 10. |
If SAD<=256 goto Step 10. |
1174 |
*/ |
*/ |
1175 |
|
|
1176 |
|
currMV->x = start_x; |
1177 |
|
currMV->y = start_y; |
1178 |
|
|
1179 |
// Prepare for main loop |
if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ |
|
|
|
|
*currMV=pmv[0]; /* current best := prediction */ |
|
|
if (!(MotionFlags & PMV_HALFPEL16 )) |
|
|
{ /* This should NOT be necessary! */ |
|
1180 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
1181 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
1182 |
} |
} |
1183 |
|
|
1184 |
if (currMV->x > max_dx) |
if (currMV->x > max_dx) { |
|
{ |
|
1185 |
currMV->x=max_dx; |
currMV->x=max_dx; |
1186 |
} |
} |
1187 |
if (currMV->x < min_dx) |
if (currMV->x < min_dx) { |
|
{ |
|
1188 |
currMV->x=min_dx; |
currMV->x=min_dx; |
1189 |
} |
} |
1190 |
if (currMV->y > max_dy) |
if (currMV->y > max_dy) { |
|
{ |
|
1191 |
currMV->y=max_dy; |
currMV->y=max_dy; |
1192 |
} |
} |
1193 |
if (currMV->y < min_dy) |
if (currMV->y < min_dy) { |
|
{ |
|
1194 |
currMV->y=min_dy; |
currMV->y=min_dy; |
1195 |
} |
} |
1196 |
|
|
1197 |
iMinSAD = sad16( cur, |
iMinSAD = |
1198 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
sad16(cur, |
1199 |
iEdgedWidth, MV_MAX_ERROR); |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
1200 |
iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
1201 |
|
iMinSAD += |
1202 |
if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,prevMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
1203 |
{ |
(uint8_t) iFcode, iQuant); |
1204 |
|
|
1205 |
|
if ((iMinSAD < 256) || |
1206 |
|
((MVequal(*currMV, prevMB->mvs[0])) && |
1207 |
|
((int32_t) iMinSAD < prevMB->sad16))) { |
1208 |
|
if (iMinSAD < (int)(2 * iQuant)) // high chances for SKIP-mode |
1209 |
|
{ |
1210 |
|
if (!MVzero(*currMV)) { |
1211 |
|
iMinSAD += MV16_00_BIAS; |
1212 |
|
CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures |
1213 |
|
iMinSAD -= MV16_00_BIAS; |
1214 |
|
} |
1215 |
|
} |
1216 |
|
|
1217 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1218 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1220 |
goto PMVfast16_Terminate_with_Refine; |
goto PMVfast16_Terminate_with_Refine; |
1221 |
} |
} |
1222 |
|
|
1223 |
|
|
1224 |
|
/* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
1225 |
|
vector of the median. |
1226 |
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
1227 |
|
*/ |
1228 |
|
|
1229 |
|
if ((bPredEq) && (MVequal(pmv[0], prevMB->mvs[0]))) |
1230 |
|
iFound = 2; |
1231 |
|
|
1232 |
|
/* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
1233 |
|
Otherwise select large Diamond Search. |
1234 |
|
*/ |
1235 |
|
|
1236 |
|
if ((!MVzero(pmv[0])) || (threshB < 1536) || (bPredEq)) |
1237 |
|
iDiamondSize = 1; // halfpel! |
1238 |
|
else |
1239 |
|
iDiamondSize = 2; // halfpel! |
1240 |
|
|
1241 |
|
if (!(MotionFlags & PMV_HALFPELDIAMOND16)) |
1242 |
|
iDiamondSize *= 2; |
1243 |
|
|
1244 |
/* |
/* |
1245 |
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1246 |
Also calculate (0,0) but do not subtract offset. |
Also calculate (0,0) but do not subtract offset. |
1250 |
|
|
1251 |
// (0,0) is always possible |
// (0,0) is always possible |
1252 |
|
|
1253 |
|
if (!MVzero(pmv[0])) |
1254 |
CHECK_MV16_ZERO; |
CHECK_MV16_ZERO; |
1255 |
|
|
1256 |
// previous frame MV is always possible |
// previous frame MV is always possible |
1257 |
|
|
1258 |
|
if (!MVzero(prevMB->mvs[0])) |
1259 |
|
if (!MVequal(prevMB->mvs[0], pmv[0])) |
1260 |
CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
1261 |
|
|
1262 |
// left neighbour, if allowed |
// left neighbour, if allowed |
1263 |
if (x != 0) |
|
1264 |
{ |
if (!MVzero(pmv[1])) |
1265 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!MVequal(pmv[1], prevMB->mvs[0])) |
1266 |
{ pmv[1].x = EVEN(pmv[1].x); |
if (!MVequal(pmv[1], pmv[0])) { |
1267 |
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
1268 |
|
pmv[1].x = EVEN(pmv[1].x); |
1269 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
1270 |
} |
} |
1271 |
|
|
1272 |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
1273 |
} |
} |
1274 |
|
// top neighbour, if allowed |
1275 |
// top neighbour, if allowed |
if (!MVzero(pmv[2])) |
1276 |
if (y != 0) |
if (!MVequal(pmv[2], prevMB->mvs[0])) |
1277 |
{ |
if (!MVequal(pmv[2], pmv[0])) |
1278 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!MVequal(pmv[2], pmv[1])) { |
1279 |
{ pmv[2].x = EVEN(pmv[2].x); |
if (!(MotionFlags & PMV_HALFPEL16)) { |
1280 |
|
pmv[2].x = EVEN(pmv[2].x); |
1281 |
pmv[2].y = EVEN(pmv[2].y); |
pmv[2].y = EVEN(pmv[2].y); |
1282 |
} |
} |
1283 |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
1284 |
|
|
1285 |
// top right neighbour, if allowed |
// top right neighbour, if allowed |
1286 |
if ((uint32_t)x != (iWcount-1)) |
if (!MVzero(pmv[3])) |
1287 |
{ |
if (!MVequal(pmv[3], prevMB->mvs[0])) |
1288 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!MVequal(pmv[3], pmv[0])) |
1289 |
{ pmv[3].x = EVEN(pmv[3].x); |
if (!MVequal(pmv[3], pmv[1])) |
1290 |
|
if (!MVequal(pmv[3], pmv[2])) { |
1291 |
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
1292 |
|
pmv[3].x = EVEN(pmv[3].x); |
1293 |
pmv[3].y = EVEN(pmv[3].y); |
pmv[3].y = EVEN(pmv[3].y); |
1294 |
} |
} |
1295 |
CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); |
CHECK_MV16_CANDIDATE(pmv[3].x, |
1296 |
|
pmv[3].y); |
1297 |
} |
} |
1298 |
} |
} |
1299 |
|
|
1300 |
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96)*/ ) |
if ((MVzero(*currMV)) && |
1301 |
|
(!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
1302 |
iMinSAD -= MV16_00_BIAS; |
iMinSAD -= MV16_00_BIAS; |
1303 |
|
|
1304 |
|
|
1306 |
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
1307 |
*/ |
*/ |
1308 |
|
|
1309 |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
if ((iMinSAD <= threshA) || |
1310 |
{ |
(MVequal(*currMV, prevMB->mvs[0]) && |
1311 |
|
((int32_t) iMinSAD < prevMB->sad16))) { |
1312 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1313 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1314 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1326 |
Refine by using small diamond and goto step 10. |
Refine by using small diamond and goto step 10. |
1327 |
*/ |
*/ |
1328 |
|
|
1329 |
|
if (MotionFlags & PMV_USESQUARES16) |
1330 |
|
MainSearchPtr = Square16_MainSearch; |
1331 |
|
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
1332 |
|
MainSearchPtr = AdvDiamond16_MainSearch; |
1333 |
|
else |
1334 |
|
MainSearchPtr = Diamond16_MainSearch; |
1335 |
|
|
1336 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1337 |
|
|
1338 |
|
|
1339 |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
1340 |
iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1341 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1342 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->x, currMV->y, iMinSAD, &newMV, center_x, center_y, |
1343 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dx, max_dx, |
1344 |
|
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1345 |
|
iQuant, iFound); |
1346 |
|
|
1347 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1348 |
*currMV = newMV; |
*currMV = newMV; |
1349 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1350 |
} |
} |
1351 |
|
|
1352 |
if (MotionFlags & PMV_EXTSEARCH16) |
if (MotionFlags & PMV_EXTSEARCH16) { |
|
{ |
|
1353 |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
1354 |
|
|
1355 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0], backupMV))) { |
1356 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1357 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1358 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
center_x, center_y, iMinSAD, &newMV, center_x, center_y, |
1359 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1360 |
|
iDiamondSize, iFcode, iQuant, iFound); |
1361 |
|
|
1362 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1363 |
*currMV = newMV; |
*currMV = newMV; |
1364 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1365 |
} |
} |
1366 |
} |
} |
1367 |
|
|
1368 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1369 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1370 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
1371 |
0, 0, iMinSAD, &newMV, |
iMinSAD, &newMV, center_x, center_y, |
1372 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dx, max_dx, min_dy, max_dy, |
1373 |
|
iEdgedWidth, iDiamondSize, iFcode, |
1374 |
|
iQuant, iFound); |
1375 |
|
|
1376 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1377 |
*currMV = newMV; |
*currMV = newMV; |
1378 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1379 |
} |
} |
1386 |
|
|
1387 |
PMVfast16_Terminate_with_Refine: |
PMVfast16_Terminate_with_Refine: |
1388 |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
1389 |
iMinSAD = Halfpel16_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
iMinSAD = |
1390 |
x, y, |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
1391 |
currMV, iMinSAD, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
1392 |
pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
1393 |
|
|
1394 |
PMVfast16_Terminate_without_Refine: |
PMVfast16_Terminate_without_Refine: |
1395 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
1396 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
1397 |
return iMinSAD; |
return iMinSAD; |
1398 |
} |
} |
1399 |
|
|
1402 |
|
|
1403 |
|
|
1404 |
|
|
1405 |
int32_t Diamond8_MainSearch( |
int32_t |
1406 |
const uint8_t * const pRef, |
Diamond8_MainSearch(const uint8_t * const pRef, |
1407 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1408 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1409 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1410 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1411 |
const int x, const int y, |
const int x, |
1412 |
int32_t startx, int32_t starty, |
const int y, |
1413 |
|
const int32_t start_x, |
1414 |
|
const int32_t start_y, |
1415 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1416 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1417 |
const VECTOR * const pmv, |
const int center_x, |
1418 |
const int32_t min_dx, const int32_t max_dx, |
const int center_y, |
1419 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t min_dx, |
1420 |
|
const int32_t max_dx, |
1421 |
|
const int32_t min_dy, |
1422 |
|
const int32_t max_dy, |
1423 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
1424 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
1425 |
const int32_t iFcode, |
const int32_t iFcode, |
1429 |
/* Do a diamond search around given starting point, return SAD of best */ |
/* Do a diamond search around given starting point, return SAD of best */ |
1430 |
|
|
1431 |
int32_t iDirection=0; |
int32_t iDirection=0; |
1432 |
|
int32_t iDirectionBackup; |
1433 |
int32_t iSAD; |
int32_t iSAD; |
1434 |
VECTOR backupMV; |
VECTOR backupMV; |
1435 |
backupMV.x = startx; |
|
1436 |
backupMV.y = starty; |
backupMV.x = start_x; |
1437 |
|
backupMV.y = start_y; |
1438 |
|
|
1439 |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
1440 |
|
|
1443 |
CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
1444 |
CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
1445 |
|
|
1446 |
if (iDirection) |
if (iDirection) { |
1447 |
while (!iFound) |
while (!iFound) { |
|
{ |
|
1448 |
iFound = 1; |
iFound = 1; |
1449 |
backupMV=*currMV; // since iDirection!=0, this is well defined! |
backupMV=*currMV; // since iDirection!=0, this is well defined! |
1450 |
|
iDirectionBackup = iDirection; |
1451 |
|
|
1452 |
if ( iDirection != 2) |
if (iDirectionBackup != 2) |
1453 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1454 |
if ( iDirection != 1) |
backupMV.y, 1); |
1455 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x+iDiamondSize,backupMV.y,2); |
if (iDirectionBackup != 1) |
1456 |
if ( iDirection != 4) |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1457 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x,backupMV.y-iDiamondSize,3); |
backupMV.y, 2); |
1458 |
if ( iDirection != 3) |
if (iDirectionBackup != 4) |
1459 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1460 |
|
backupMV.y - iDiamondSize, 3); |
1461 |
|
if (iDirectionBackup != 3) |
1462 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1463 |
|
backupMV.y + iDiamondSize, 4); |
1464 |
} |
} |
1465 |
else |
} else { |
1466 |
|
currMV->x = start_x; |
1467 |
|
currMV->y = start_y; |
1468 |
|
} |
1469 |
|
return iMinSAD; |
1470 |
|
} |
1471 |
|
|
1472 |
|
|
1473 |
|
|
1474 |
|
|
1475 |
|
int32_t |
1476 |
|
Square8_MainSearch(const uint8_t * const pRef, |
1477 |
|
const uint8_t * const pRefH, |
1478 |
|
const uint8_t * const pRefV, |
1479 |
|
const uint8_t * const pRefHV, |
1480 |
|
const uint8_t * const cur, |
1481 |
|
const int x, |
1482 |
|
const int y, |
1483 |
|
const int32_t start_x, |
1484 |
|
const int32_t start_y, |
1485 |
|
int32_t iMinSAD, |
1486 |
|
VECTOR * const currMV, |
1487 |
|
const int center_x, |
1488 |
|
const int center_y, |
1489 |
|
const int32_t min_dx, |
1490 |
|
const int32_t max_dx, |
1491 |
|
const int32_t min_dy, |
1492 |
|
const int32_t max_dy, |
1493 |
|
const int32_t iEdgedWidth, |
1494 |
|
const int32_t iDiamondSize, |
1495 |
|
const int32_t iFcode, |
1496 |
|
const int32_t iQuant, |
1497 |
|
int iFound) |
1498 |
{ |
{ |
1499 |
currMV->x = startx; |
/* Do a square search around given starting point, return SAD of best */ |
1500 |
currMV->y = starty; |
|
1501 |
|
int32_t iDirection = 0; |
1502 |
|
int32_t iSAD; |
1503 |
|
VECTOR backupMV; |
1504 |
|
|
1505 |
|
backupMV.x = start_x; |
1506 |
|
backupMV.y = start_y; |
1507 |
|
|
1508 |
|
/* It's one search with full square pattern, and new parts for all following diamonds */ |
1509 |
|
|
1510 |
|
/* new direction are extra, so 1-4 is normal diamond |
1511 |
|
537 |
1512 |
|
1*2 |
1513 |
|
648 |
1514 |
|
*/ |
1515 |
|
|
1516 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); |
1517 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); |
1518 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
1519 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
1520 |
|
|
1521 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
1522 |
|
backupMV.y - iDiamondSize, 5); |
1523 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
1524 |
|
backupMV.y + iDiamondSize, 6); |
1525 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
1526 |
|
backupMV.y - iDiamondSize, 7); |
1527 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
1528 |
|
backupMV.y + iDiamondSize, 8); |
1529 |
|
|
1530 |
|
|
1531 |
|
if (iDirection) { |
1532 |
|
while (!iFound) { |
1533 |
|
iFound = 1; |
1534 |
|
backupMV = *currMV; |
1535 |
|
|
1536 |
|
switch (iDirection) { |
1537 |
|
case 1: |
1538 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1539 |
|
backupMV.y, 1); |
1540 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1541 |
|
backupMV.y - iDiamondSize, 5); |
1542 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1543 |
|
backupMV.y - iDiamondSize, 7); |
1544 |
|
break; |
1545 |
|
case 2: |
1546 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1547 |
|
2); |
1548 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1549 |
|
backupMV.y + iDiamondSize, 6); |
1550 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1551 |
|
backupMV.y + iDiamondSize, 8); |
1552 |
|
break; |
1553 |
|
|
1554 |
|
case 3: |
1555 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1556 |
|
4); |
1557 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1558 |
|
backupMV.y - iDiamondSize, 7); |
1559 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1560 |
|
backupMV.y + iDiamondSize, 8); |
1561 |
|
break; |
1562 |
|
|
1563 |
|
case 4: |
1564 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1565 |
|
3); |
1566 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1567 |
|
backupMV.y - iDiamondSize, 5); |
1568 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1569 |
|
backupMV.y + iDiamondSize, 6); |
1570 |
|
break; |
1571 |
|
|
1572 |
|
case 5: |
1573 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
1574 |
|
1); |
1575 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1576 |
|
3); |
1577 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1578 |
|
backupMV.y - iDiamondSize, 5); |
1579 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1580 |
|
backupMV.y + iDiamondSize, 6); |
1581 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1582 |
|
backupMV.y - iDiamondSize, 7); |
1583 |
|
break; |
1584 |
|
|
1585 |
|
case 6: |
1586 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1587 |
|
2); |
1588 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1589 |
|
3); |
1590 |
|
|
1591 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1592 |
|
backupMV.y - iDiamondSize, 5); |
1593 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1594 |
|
backupMV.y + iDiamondSize, 6); |
1595 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1596 |
|
backupMV.y + iDiamondSize, 8); |
1597 |
|
|
1598 |
|
break; |
1599 |
|
|
1600 |
|
case 7: |
1601 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1602 |
|
backupMV.y, 1); |
1603 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1604 |
|
4); |
1605 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1606 |
|
backupMV.y - iDiamondSize, 5); |
1607 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1608 |
|
backupMV.y - iDiamondSize, 7); |
1609 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1610 |
|
backupMV.y + iDiamondSize, 8); |
1611 |
|
break; |
1612 |
|
|
1613 |
|
case 8: |
1614 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1615 |
|
2); |
1616 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1617 |
|
4); |
1618 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1619 |
|
backupMV.y + iDiamondSize, 6); |
1620 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1621 |
|
backupMV.y - iDiamondSize, 7); |
1622 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1623 |
|
backupMV.y + iDiamondSize, 8); |
1624 |
|
break; |
1625 |
|
default: |
1626 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
1627 |
|
1); |
1628 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1629 |
|
2); |
1630 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1631 |
|
3); |
1632 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1633 |
|
4); |
1634 |
|
|
1635 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1636 |
|
backupMV.y - iDiamondSize, 5); |
1637 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1638 |
|
backupMV.y + iDiamondSize, 6); |
1639 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1640 |
|
backupMV.y - iDiamondSize, 7); |
1641 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1642 |
|
backupMV.y + iDiamondSize, 8); |
1643 |
|
break; |
1644 |
|
} |
1645 |
|
} |
1646 |
|
} else { |
1647 |
|
currMV->x = start_x; |
1648 |
|
currMV->y = start_y; |
1649 |
} |
} |
1650 |
return iMinSAD; |
return iMinSAD; |
1651 |
} |
} |
1652 |
|
|
1653 |
int32_t Halfpel8_Refine( |
|
1654 |
const uint8_t * const pRef, |
|
1655 |
|
|
1656 |
|
|
1657 |
|
int32_t |
1658 |
|
Halfpel8_Refine_c(const uint8_t * const pRef, |
1659 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1660 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1661 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1662 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1663 |
const int x, const int y, |
const int x, |
1664 |
|
const int y, |
1665 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1666 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1667 |
const VECTOR * const pmv, |
const int center_x, |
1668 |
const int32_t min_dx, const int32_t max_dx, |
const int center_y, |
1669 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t min_dx, |
1670 |
|
const int32_t max_dx, |
1671 |
|
const int32_t min_dy, |
1672 |
|
const int32_t max_dy, |
1673 |
const int32_t iFcode, |
const int32_t iFcode, |
1674 |
const int32_t iQuant, |
const int32_t iQuant, |
1675 |
const int32_t iEdgedWidth) |
const int32_t iEdgedWidth) |
1694 |
|
|
1695 |
#define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) |
#define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) |
1696 |
|
|
1697 |
int32_t PMVfastSearch8( |
int32_t |
1698 |
const uint8_t * const pRef, |
PMVfastSearch8(const uint8_t * const pRef, |
1699 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1700 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1701 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1702 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1703 |
const int x, const int y, |
const int x, |
1704 |
const int start_x, const int start_y, |
const int y, |
1705 |
|
const int start_x, |
1706 |
|
const int start_y, |
1707 |
|
const int center_x, |
1708 |
|
const int center_y, |
1709 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1710 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1711 |
const uint32_t iFcode, |
const uint32_t iFcode, |
1733 |
int32_t psad[4]; |
int32_t psad[4]; |
1734 |
VECTOR newMV; |
VECTOR newMV; |
1735 |
VECTOR backupMV; |
VECTOR backupMV; |
1736 |
|
VECTOR startMV; |
1737 |
|
|
1738 |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1739 |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1740 |
|
|
1741 |
static int32_t threshA,threshB; |
int32_t threshA, threshB; |
1742 |
int32_t iFound,bPredEq; |
int32_t iFound,bPredEq; |
1743 |
int32_t iMinSAD,iSAD; |
int32_t iMinSAD,iSAD; |
1744 |
|
|
1745 |
int32_t iSubBlock = ((y&1)<<1) + (x&1); |
int32_t iSubBlock = (y & 1) + (y & 1) + (x & 1); |
1746 |
|
|
1747 |
|
MainSearch8FuncPtr MainSearchPtr; |
1748 |
|
|
1749 |
|
/* Init variables */ |
1750 |
|
startMV.x = start_x; |
1751 |
|
startMV.y = start_y; |
1752 |
|
|
1753 |
/* Get maximum range */ |
/* Get maximum range */ |
1754 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 8, iWidth, iHeight, |
1755 |
x, y, 8, iWidth, iHeight, iFcode); |
iFcode); |
1756 |
|
|
1757 |
if (!(MotionFlags & PMV_HALFPELDIAMOND8 )) |
if (!(MotionFlags & PMV_HALFPELDIAMOND8)) { |
1758 |
{ min_dx = EVEN(min_dx); |
min_dx = EVEN(min_dx); |
1759 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
1760 |
min_dy = EVEN(min_dy); |
min_dy = EVEN(min_dy); |
1761 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
1762 |
} /* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
} |
|
|
|
1763 |
|
|
1764 |
bPredEq = get_pmvdata(pMBs, (x>>1), (y>>1), iWcount, iSubBlock, pmv, psad); |
/* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
1765 |
|
bPredEq = get_pmvdata2(pMBs, iWcount, 0, (x >> 1), (y >> 1), iSubBlock, pmv, psad); |
1766 |
|
|
1767 |
if ((x==0) && (y==0) ) |
if ((x == 0) && (y == 0)) { |
|
{ |
|
1768 |
threshA = 512/4; |
threshA = 512/4; |
1769 |
threshB = 1024/4; |
threshB = 1024/4; |
1770 |
|
|
1771 |
} |
} else { |
1772 |
else |
threshA = psad[0] / 4; /* good estimate? */ |
|
{ |
|
|
threshA = psad[0]/4; /* good estimate */ |
|
1773 |
threshB = threshA+256/4; |
threshB = threshA+256/4; |
1774 |
if (threshA< 512/4) threshA = 512/4; |
if (threshA < 512 / 4) |
1775 |
if (threshA>1024/4) threshA = 1024/4; |
threshA = 512 / 4; |
1776 |
if (threshB>1792/4) threshB = 1792/4; |
if (threshA > 1024 / 4) |
1777 |
|
threshA = 1024 / 4; |
1778 |
|
if (threshB > 1792 / 4) |
1779 |
|
threshB = 1792 / 4; |
1780 |
} |
} |
1781 |
|
|
1782 |
iFound=0; |
iFound=0; |
1783 |
|
|
|
/* Step 2: Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
|
|
vector of the median. |
|
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
|
|
*/ |
|
|
|
|
|
if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[iSubBlock]) ) ) |
|
|
iFound=2; |
|
|
|
|
|
/* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
|
|
Otherwise select large Diamond Search. |
|
|
*/ |
|
|
|
|
|
if ( (pmv[0].x != 0) || (pmv[0].y != 0) || (threshB<1536/4) || (bPredEq) ) |
|
|
iDiamondSize=1; // 1 halfpel! |
|
|
else |
|
|
iDiamondSize=2; // 2 halfpel = 1 full pixel! |
|
|
|
|
|
if (!(MotionFlags & PMV_HALFPELDIAMOND8) ) |
|
|
iDiamondSize*=2; |
|
|
|
|
1784 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
1785 |
MinSAD=SAD |
MinSAD=SAD |
1786 |
If Motion Vector equal to Previous frame motion vector |
If Motion Vector equal to Previous frame motion vector |
1791 |
|
|
1792 |
// Prepare for main loop |
// Prepare for main loop |
1793 |
|
|
1794 |
currMV->x=start_x; /* start with mv16 */ |
if (MotionFlags & PMV_USESQUARES8) |
1795 |
currMV->y=start_y; |
MainSearchPtr = Square8_MainSearch; |
1796 |
|
else |
1797 |
|
|
1798 |
|
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
1799 |
|
MainSearchPtr = AdvDiamond8_MainSearch; |
1800 |
|
else |
1801 |
|
MainSearchPtr = Diamond8_MainSearch; |
1802 |
|
|
|
iMinSAD = sad8( cur, |
|
|
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
|
|
iEdgedWidth); |
|
|
iMinSAD += calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, (uint8_t)iFcode, iQuant); |
|
1803 |
|
|
1804 |
if ( (iMinSAD < 256/4 ) || ( (MVequal(*currMV,prevMB->mvs[iSubBlock])) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
*currMV = startMV; |
1805 |
{ |
|
1806 |
|
iMinSAD = |
1807 |
|
sad8(cur, |
1808 |
|
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
1809 |
|
iEdgedWidth), iEdgedWidth); |
1810 |
|
iMinSAD += |
1811 |
|
calc_delta_8(currMV->x - center_x, currMV->y - center_y, |
1812 |
|
(uint8_t) iFcode, iQuant); |
1813 |
|
|
1814 |
|
if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) |
1815 |
|
&& ((int32_t) iMinSAD < |
1816 |
|
prevMB->sad8[iSubBlock]))) { |
1817 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1818 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1819 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1820 |
goto PMVfast8_Terminate_with_Refine; |
goto PMVfast8_Terminate_with_Refine; |
1821 |
} |
} |
1822 |
|
|
1823 |
|
/* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
1824 |
|
vector of the median. |
1825 |
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
1826 |
|
*/ |
1827 |
|
|
1828 |
|
if ((bPredEq) && (MVequal(pmv[0], prevMB->mvs[iSubBlock]))) |
1829 |
|
iFound = 2; |
1830 |
|
|
1831 |
|
/* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
1832 |
|
Otherwise select large Diamond Search. |
1833 |
|
*/ |
1834 |
|
|
1835 |
|
if ((!MVzero(pmv[0])) || (threshB < 1536 / 4) || (bPredEq)) |
1836 |
|
iDiamondSize = 1; // 1 halfpel! |
1837 |
|
else |
1838 |
|
iDiamondSize = 2; // 2 halfpel = 1 full pixel! |
1839 |
|
|
1840 |
|
if (!(MotionFlags & PMV_HALFPELDIAMOND8)) |
1841 |
|
iDiamondSize *= 2; |
1842 |
|
|
1843 |
|
|
1844 |
/* |
/* |
1845 |
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1848 |
If MV is (0,0) subtract offset. |
If MV is (0,0) subtract offset. |
1849 |
*/ |
*/ |
1850 |
|
|
1851 |
// the prediction might be even better than mv16 |
// the median prediction might be even better than mv16 |
|
CHECK_MV8_CANDIDATE(pmv[0].x,pmv[0].y); |
|
1852 |
|
|
1853 |
// (0,0) is always possible |
if (!MVequal(pmv[0], startMV)) |
1854 |
|
CHECK_MV8_CANDIDATE(center_x, center_y); |
1855 |
|
|
1856 |
|
// (0,0) if needed |
1857 |
|
if (!MVzero(pmv[0])) |
1858 |
|
if (!MVzero(startMV)) |
1859 |
CHECK_MV8_ZERO; |
CHECK_MV8_ZERO; |
1860 |
|
|
1861 |
// previous frame MV is always possible |
// previous frame MV if needed |
1862 |
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
if (!MVzero(prevMB->mvs[iSubBlock])) |
1863 |
|
if (!MVequal(prevMB->mvs[iSubBlock], startMV)) |
1864 |
|
if (!MVequal(prevMB->mvs[iSubBlock], pmv[0])) |
1865 |
|
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x, |
1866 |
|
prevMB->mvs[iSubBlock].y); |
1867 |
|
|
1868 |
|
if ((iMinSAD <= threshA) || |
1869 |
|
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
1870 |
|
((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
1871 |
|
if (MotionFlags & PMV_QUICKSTOP16) |
1872 |
|
goto PMVfast8_Terminate_without_Refine; |
1873 |
|
if (MotionFlags & PMV_EARLYSTOP16) |
1874 |
|
goto PMVfast8_Terminate_with_Refine; |
1875 |
|
} |
1876 |
|
|
1877 |
// left neighbour, if allowed |
// left neighbour, if allowed and needed |
1878 |
if (psad[1] != MV_MAX_ERROR) |
if (!MVzero(pmv[1])) |
1879 |
{ |
if (!MVequal(pmv[1], startMV)) |
1880 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!MVequal(pmv[1], prevMB->mvs[iSubBlock])) |
1881 |
{ pmv[1].x = EVEN(pmv[1].x); |
if (!MVequal(pmv[1], pmv[0])) { |
1882 |
|
if (!(MotionFlags & PMV_HALFPEL8)) { |
1883 |
|
pmv[1].x = EVEN(pmv[1].x); |
1884 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
1885 |
} |
} |
1886 |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
1887 |
} |
} |
1888 |
|
// top neighbour, if allowed and needed |
1889 |
// top neighbour, if allowed |
if (!MVzero(pmv[2])) |
1890 |
if (psad[2] != MV_MAX_ERROR) |
if (!MVequal(pmv[2], startMV)) |
1891 |
{ |
if (!MVequal(pmv[2], prevMB->mvs[iSubBlock])) |
1892 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!MVequal(pmv[2], pmv[0])) |
1893 |
{ pmv[2].x = EVEN(pmv[2].x); |
if (!MVequal(pmv[2], pmv[1])) { |
1894 |
|
if (!(MotionFlags & PMV_HALFPEL8)) { |
1895 |
|
pmv[2].x = EVEN(pmv[2].x); |
1896 |
pmv[2].y = EVEN(pmv[2].y); |
pmv[2].y = EVEN(pmv[2].y); |
1897 |
} |
} |
1898 |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
1899 |
|
|
1900 |
// top right neighbour, if allowed |
// top right neighbour, if allowed and needed |
1901 |
if (psad[3] != MV_MAX_ERROR) |
if (!MVzero(pmv[3])) |
1902 |
{ |
if (!MVequal(pmv[3], startMV)) |
1903 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!MVequal(pmv[3], prevMB->mvs[iSubBlock])) |
1904 |
{ pmv[3].x = EVEN(pmv[3].x); |
if (!MVequal(pmv[3], pmv[0])) |
1905 |
|
if (!MVequal(pmv[3], pmv[1])) |
1906 |
|
if (!MVequal(pmv[3], pmv[2])) { |
1907 |
|
if (! |
1908 |
|
(MotionFlags & |
1909 |
|
PMV_HALFPEL8)) { |
1910 |
|
pmv[3].x = EVEN(pmv[3].x); |
1911 |
pmv[3].y = EVEN(pmv[3].y); |
pmv[3].y = EVEN(pmv[3].y); |
1912 |
} |
} |
1913 |
CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
CHECK_MV8_CANDIDATE(pmv[3].x, |
1914 |
|
pmv[3].y); |
1915 |
} |
} |
1916 |
} |
} |
1917 |
|
|
1918 |
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
if ((MVzero(*currMV)) && |
1919 |
|
(!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
1920 |
iMinSAD -= MV8_00_BIAS; |
iMinSAD -= MV8_00_BIAS; |
1921 |
|
|
1922 |
|
|
1924 |
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
1925 |
*/ |
*/ |
1926 |
|
|
1927 |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[iSubBlock]) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
if ((iMinSAD <= threshA) || |
1928 |
{ |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
1929 |
|
((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
1930 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1931 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1932 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1946 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1947 |
|
|
1948 |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
1949 |
iSAD = Diamond8_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1950 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
1951 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
1952 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1953 |
|
iQuant, iFound); |
1954 |
|
|
1955 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1956 |
*currMV = newMV; |
*currMV = newMV; |
1957 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1958 |
} |
} |
1959 |
|
|
1960 |
if (MotionFlags & PMV_EXTSEARCH8) |
if (MotionFlags & PMV_EXTSEARCH8) { |
|
{ |
|
1961 |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
1962 |
|
|
1963 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0], backupMV))) { |
1964 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1965 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1966 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
1967 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1968 |
|
iDiamondSize, iFcode, iQuant, iFound); |
1969 |
|
|
1970 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1971 |
*currMV = newMV; |
*currMV = newMV; |
1972 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1973 |
} |
} |
1974 |
} |
} |
1975 |
|
|
1976 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1977 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1978 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
1979 |
0, 0, iMinSAD, &newMV, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
1980 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1981 |
|
iQuant, iFound); |
1982 |
|
|
1983 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1984 |
*currMV = newMV; |
*currMV = newMV; |
1985 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1986 |
} |
} |
1993 |
|
|
1994 |
PMVfast8_Terminate_with_Refine: |
PMVfast8_Terminate_with_Refine: |
1995 |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
1996 |
iMinSAD = Halfpel8_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
iMinSAD = |
1997 |
x, y, |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
1998 |
currMV, iMinSAD, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
1999 |
pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2000 |
|
|
2001 |
|
|
2002 |
PMVfast8_Terminate_without_Refine: |
PMVfast8_Terminate_without_Refine: |
2003 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2004 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2005 |
|
|
2006 |
return iMinSAD; |
return iMinSAD; |
2007 |
} |
} |
2008 |
|
|
2009 |
int32_t EPZSSearch16( |
int32_t |
2010 |
const uint8_t * const pRef, |
EPZSSearch16(const uint8_t * const pRef, |
2011 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
2012 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
2013 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
2014 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
2015 |
const int x, const int y, |
const int x, |
2016 |
|
const int y, |
2017 |
|
const int start_x, |
2018 |
|
const int start_y, |
2019 |
|
const int center_x, |
2020 |
|
const int center_y, |
2021 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2022 |
const uint32_t iQuant, |
const uint32_t iQuant, |
2023 |
const uint32_t iFcode, |
const uint32_t iFcode, |
2048 |
int32_t psad[8]; |
int32_t psad[8]; |
2049 |
|
|
2050 |
static MACROBLOCK * oldMBs = NULL; |
static MACROBLOCK * oldMBs = NULL; |
2051 |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
|
2052 |
|
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
2053 |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
2054 |
MACROBLOCK * oldMB = NULL; |
MACROBLOCK * oldMB = NULL; |
2055 |
|
|
2056 |
static int32_t thresh2; |
int32_t thresh2; |
2057 |
int32_t bPredEq; |
int32_t bPredEq; |
2058 |
int32_t iMinSAD,iSAD=9999; |
int32_t iMinSAD,iSAD=9999; |
2059 |
|
|
2060 |
MainSearch16FuncPtr EPZSMainSearchPtr; |
MainSearch16FuncPtr MainSearchPtr; |
2061 |
|
|
2062 |
if (oldMBs == NULL) |
if (oldMBs == NULL) { |
2063 |
{ oldMBs = (MACROBLOCK*) calloc(iWcount*iHcount,sizeof(MACROBLOCK)); |
oldMBs = (MACROBLOCK *) calloc(iWcount * iHcount, sizeof(MACROBLOCK)); |
2064 |
// fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
// fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
2065 |
} |
} |
2066 |
oldMB = oldMBs + x + y * iWcount; |
oldMB = oldMBs + x + y * iWcount; |
2067 |
|
|
2068 |
/* Get maximum range */ |
/* Get maximum range */ |
2069 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, |
2070 |
x, y, 16, iWidth, iHeight, iFcode); |
iFcode); |
2071 |
|
|
2072 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2073 |
{ min_dx = EVEN(min_dx); |
min_dx = EVEN(min_dx); |
2074 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
2075 |
min_dy = EVEN(min_dy); |
min_dy = EVEN(min_dy); |
2076 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
2077 |
} /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
} |
2078 |
|
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
2079 |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
2080 |
|
|
2081 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
2082 |
MinSAD=SAD |
MinSAD=SAD |
2087 |
|
|
2088 |
// Prepare for main loop |
// Prepare for main loop |
2089 |
|
|
2090 |
*currMV=pmv[0]; /* current best := median prediction */ |
currMV->x = start_x; |
2091 |
if (!(MotionFlags & PMV_HALFPEL16)) |
currMV->y = start_y; |
2092 |
{ |
|
2093 |
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
2094 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
2095 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
2096 |
} |
} |
2106 |
|
|
2107 |
/***************** This is predictor SET A: only median prediction ******************/ |
/***************** This is predictor SET A: only median prediction ******************/ |
2108 |
|
|
2109 |
iMinSAD = sad16( cur, |
iMinSAD = |
2110 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
sad16(cur, |
2111 |
iEdgedWidth, MV_MAX_ERROR); |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
2112 |
iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
2113 |
|
iMinSAD += |
2114 |
|
calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
2115 |
|
(uint8_t) iFcode, iQuant); |
2116 |
|
|
2117 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
2118 |
if ( (iMinSAD < 256 ) || ( (MVequal(*currMV, prevMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
if ((iMinSAD < 256) || |
2119 |
{ |
((MVequal(*currMV, prevMB->mvs[0])) && |
2120 |
|
((int32_t) iMinSAD < prevMB->sad16))) { |
2121 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2122 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2123 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2132 |
// set threshhold based on Min of Prediction and SAD of collocated block |
// set threshhold based on Min of Prediction and SAD of collocated block |
2133 |
// CHECK_MV16 always uses iSAD for the SAD of last vector to check, so now iSAD is what we want |
// CHECK_MV16 always uses iSAD for the SAD of last vector to check, so now iSAD is what we want |
2134 |
|
|
2135 |
if ((x==0) && (y==0) ) |
if ((x == 0) && (y == 0)) { |
|
{ |
|
2136 |
thresh2 = 512; |
thresh2 = 512; |
2137 |
} |
} else { |
|
else |
|
|
{ |
|
2138 |
/* T_k = 1.2 * MIN(SAD_top,SAD_left,SAD_topleft,SAD_coll) +128; [Tourapis, 2002] */ |
/* T_k = 1.2 * MIN(SAD_top,SAD_left,SAD_topleft,SAD_coll) +128; [Tourapis, 2002] */ |
2139 |
|
|
2140 |
thresh2 = MIN(psad[0],iSAD)*6/5 + 128; |
thresh2 = MIN(psad[0],iSAD)*6/5 + 128; |
2146 |
|
|
2147 |
|
|
2148 |
// left neighbour, if allowed |
// left neighbour, if allowed |
2149 |
if (x != 0) |
if (x != 0) { |
2150 |
{ |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2151 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
pmv[1].x = EVEN(pmv[1].x); |
|
{ pmv[1].x = EVEN(pmv[1].x); |
|
2152 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
2153 |
} |
} |
2154 |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
2155 |
} |
} |
|
|
|
2156 |
// top neighbour, if allowed |
// top neighbour, if allowed |
2157 |
if (y != 0) |
if (y != 0) { |
2158 |
{ |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2159 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
pmv[2].x = EVEN(pmv[2].x); |
|
{ pmv[2].x = EVEN(pmv[2].x); |
|
2160 |
pmv[2].y = EVEN(pmv[2].y); |
pmv[2].y = EVEN(pmv[2].y); |
2161 |
} |
} |
2162 |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
2163 |
|
|
2164 |
// top right neighbour, if allowed |
// top right neighbour, if allowed |
2165 |
if ((uint32_t)x != (iWcount-1)) |
if ((uint32_t) x != (iWcount - 1)) { |
2166 |
{ |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2167 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
pmv[3].x = EVEN(pmv[3].x); |
|
{ pmv[3].x = EVEN(pmv[3].x); |
|
2168 |
pmv[3].y = EVEN(pmv[3].y); |
pmv[3].y = EVEN(pmv[3].y); |
2169 |
} |
} |
2170 |
CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); |
CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); |
2176 |
*/ |
*/ |
2177 |
|
|
2178 |
if ( (iMinSAD <= thresh2) |
if ( (iMinSAD <= thresh2) |
2179 |
|| ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD <= prevMB->sad16) ) ) |
|| (MVequal(*currMV, prevMB->mvs[0]) && |
2180 |
{ |
((int32_t) iMinSAD <= prevMB->sad16))) { |
2181 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2182 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2183 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2198 |
|
|
2199 |
// top neighbour |
// top neighbour |
2200 |
if (y != 0) |
if (y != 0) |
2201 |
CHECK_MV16_CANDIDATE((prevMB-iWcount)->mvs[0].x,(prevMB-iWcount)->mvs[0].y); |
CHECK_MV16_CANDIDATE((prevMB - iWcount)->mvs[0].x, |
2202 |
|
(prevMB - iWcount)->mvs[0].y); |
2203 |
|
|
2204 |
// right neighbour, if allowed (this value is not written yet, so take it from pMB->mvs |
// right neighbour, if allowed (this value is not written yet, so take it from pMB->mvs |
2205 |
|
|
2208 |
|
|
2209 |
// bottom neighbour, dito |
// bottom neighbour, dito |
2210 |
if ((uint32_t)y != iHcount-1) |
if ((uint32_t)y != iHcount-1) |
2211 |
CHECK_MV16_CANDIDATE((prevMB+iWcount)->mvs[0].x,(prevMB+iWcount)->mvs[0].y); |
CHECK_MV16_CANDIDATE((prevMB + iWcount)->mvs[0].x, |
2212 |
|
(prevMB + iWcount)->mvs[0].y); |
2213 |
|
|
2214 |
/* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ |
/* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ |
2215 |
if (iMinSAD <= thresh2) |
if (iMinSAD <= thresh2) { |
|
{ |
|
2216 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2217 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2218 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2223 |
|
|
2224 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
2225 |
|
|
|
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
|
|
|
|
2226 |
if (MotionFlags & PMV_USESQUARES16) |
if (MotionFlags & PMV_USESQUARES16) |
2227 |
EPZSMainSearchPtr = Square16_MainSearch; |
MainSearchPtr = Square16_MainSearch; |
2228 |
else |
else |
2229 |
EPZSMainSearchPtr = Diamond16_MainSearch; |
if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
2230 |
|
MainSearchPtr = AdvDiamond16_MainSearch; |
2231 |
|
else |
2232 |
|
MainSearchPtr = Diamond16_MainSearch; |
2233 |
|
|
2234 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
|
x, y, |
|
|
currMV->x, currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
|
|
2, iFcode, iQuant, 0); |
|
2235 |
|
|
2236 |
if (iSAD < iMinSAD) |
iSAD = |
2237 |
{ |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2238 |
|
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2239 |
|
min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2240 |
|
|
2241 |
|
if (iSAD < iMinSAD) { |
2242 |
*currMV = newMV; |
*currMV = newMV; |
2243 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2244 |
} |
} |
2245 |
|
|
2246 |
|
|
2247 |
if (MotionFlags & PMV_EXTSEARCH16) |
if (MotionFlags & PMV_EXTSEARCH16) { |
|
{ |
|
2248 |
/* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
2249 |
|
|
2250 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0], backupMV))) { |
2251 |
{ |
iSAD = |
2252 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2253 |
x, y, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2254 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2255 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2, iFcode, iQuant, 0); |
2256 |
} |
} |
2257 |
|
|
2258 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2259 |
*currMV = newMV; |
*currMV = newMV; |
2260 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2261 |
} |
} |
2262 |
|
|
2263 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2264 |
{ |
iSAD = |
2265 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2266 |
x, y, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2267 |
0, 0, iMinSAD, &newMV, |
max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
|
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
|
2268 |
|
|
2269 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2270 |
*currMV = newMV; |
*currMV = newMV; |
2271 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2272 |
} |
} |
2277 |
|
|
2278 |
EPZS16_Terminate_with_Refine: |
EPZS16_Terminate_with_Refine: |
2279 |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
2280 |
iMinSAD = Halfpel16_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
iMinSAD = |
2281 |
x, y, |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2282 |
currMV, iMinSAD, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2283 |
pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2284 |
|
|
2285 |
EPZS16_Terminate_without_Refine: |
EPZS16_Terminate_without_Refine: |
2286 |
|
|
2287 |
*oldMB = *prevMB; |
*oldMB = *prevMB; |
2288 |
|
|
2289 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2290 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2291 |
return iMinSAD; |
return iMinSAD; |
2292 |
} |
} |
2293 |
|
|
2294 |
|
|
2295 |
int32_t EPZSSearch8( |
int32_t |
2296 |
const uint8_t * const pRef, |
EPZSSearch8(const uint8_t * const pRef, |
2297 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
2298 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
2299 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
2300 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
2301 |
const int x, const int y, |
const int x, |
2302 |
const int start_x, const int start_y, |
const int y, |
2303 |
|
const int start_x, |
2304 |
|
const int start_y, |
2305 |
|
const int center_x, |
2306 |
|
const int center_y, |
2307 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2308 |
const uint32_t iQuant, |
const uint32_t iQuant, |
2309 |
const uint32_t iFcode, |
const uint32_t iFcode, |
2337 |
|
|
2338 |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
2339 |
|
|
2340 |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
2341 |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
2342 |
|
|
2343 |
int32_t bPredEq; |
int32_t bPredEq; |
2344 |
int32_t iMinSAD,iSAD=9999; |
int32_t iMinSAD,iSAD=9999; |
2345 |
|
|
2346 |
MainSearch8FuncPtr EPZSMainSearchPtr; |
MainSearch8FuncPtr MainSearchPtr; |
2347 |
|
|
2348 |
/* Get maximum range */ |
/* Get maximum range */ |
2349 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 8, iWidth, iHeight, |
2350 |
x, y, 8, iWidth, iHeight, iFcode); |
iFcode); |
2351 |
|
|
2352 |
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
2353 |
|
|
2354 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!(MotionFlags & PMV_HALFPEL8)) { |
2355 |
{ min_dx = EVEN(min_dx); |
min_dx = EVEN(min_dx); |
2356 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
2357 |
min_dy = EVEN(min_dy); |
min_dy = EVEN(min_dy); |
2358 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
2359 |
} /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
} |
2360 |
|
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
2361 |
bPredEq = get_pmvdata(pMBs, x>>1, y>>1, iWcount, iSubBlock, pmv, psad); |
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x >> 1, y >> 1, iSubBlock, pmv, psad); |
2362 |
|
|
2363 |
|
|
2364 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
2371 |
// Prepare for main loop |
// Prepare for main loop |
2372 |
|
|
2373 |
|
|
2374 |
if (!(MotionFlags & PMV_HALFPEL8)) |
if (!(MotionFlags & PMV_HALFPEL8)) { |
|
{ |
|
2375 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
2376 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
2377 |
} |
} |
2388 |
/***************** This is predictor SET A: only median prediction ******************/ |
/***************** This is predictor SET A: only median prediction ******************/ |
2389 |
|
|
2390 |
|
|
2391 |
iMinSAD = sad8( cur, |
iMinSAD = |
2392 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
sad8(cur, |
2393 |
iEdgedWidth); |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
2394 |
iMinSAD += calc_delta_8(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
iEdgedWidth), iEdgedWidth); |
2395 |
|
iMinSAD += |
2396 |
|
calc_delta_8(currMV->x - center_x, currMV->y - center_y, |
2397 |
|
(uint8_t) iFcode, iQuant); |
2398 |
|
|
2399 |
|
|
2400 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
2401 |
if (iMinSAD < 256/4 ) |
if (iMinSAD < 256 / 4) { |
|
{ |
|
2402 |
if (MotionFlags & PMV_QUICKSTOP8) |
if (MotionFlags & PMV_QUICKSTOP8) |
2403 |
goto EPZS8_Terminate_without_Refine; |
goto EPZS8_Terminate_without_Refine; |
2404 |
if (MotionFlags & PMV_EARLYSTOP8) |
if (MotionFlags & PMV_EARLYSTOP8) |
2415 |
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
2416 |
|
|
2417 |
// left neighbour, if allowed |
// left neighbour, if allowed |
2418 |
if (psad[1] != MV_MAX_ERROR) |
if (psad[1] != MV_MAX_ERROR) { |
2419 |
{ |
if (!(MotionFlags & PMV_HALFPEL8)) { |
2420 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
pmv[1].x = EVEN(pmv[1].x); |
|
{ pmv[1].x = EVEN(pmv[1].x); |
|
2421 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
2422 |
} |
} |
2423 |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
2424 |
} |
} |
|
|
|
2425 |
// top neighbour, if allowed |
// top neighbour, if allowed |
2426 |
if (psad[2] != MV_MAX_ERROR) |
if (psad[2] != MV_MAX_ERROR) { |
2427 |
{ |
if (!(MotionFlags & PMV_HALFPEL8)) { |
2428 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
pmv[2].x = EVEN(pmv[2].x); |
|
{ pmv[2].x = EVEN(pmv[2].x); |
|
2429 |
pmv[2].y = EVEN(pmv[2].y); |
pmv[2].y = EVEN(pmv[2].y); |
2430 |
} |
} |
2431 |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
2432 |
|
|
2433 |
// top right neighbour, if allowed |
// top right neighbour, if allowed |
2434 |
if (psad[3] != MV_MAX_ERROR) |
if (psad[3] != MV_MAX_ERROR) { |
2435 |
{ |
if (!(MotionFlags & PMV_HALFPEL8)) { |
2436 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
pmv[3].x = EVEN(pmv[3].x); |
|
{ pmv[3].x = EVEN(pmv[3].x); |
|
2437 |
pmv[3].y = EVEN(pmv[3].y); |
pmv[3].y = EVEN(pmv[3].y); |
2438 |
} |
} |
2439 |
CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
2451 |
Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] |
Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] |
2452 |
*/ |
*/ |
2453 |
|
|
2454 |
if (iMinSAD < 512/4) /* T_2 == 512/4 hardcoded */ |
if (iMinSAD < 512 / 4) { /* T_2 == 512/4 hardcoded */ |
|
{ |
|
2455 |
if (MotionFlags & PMV_QUICKSTOP8) |
if (MotionFlags & PMV_QUICKSTOP8) |
2456 |
goto EPZS8_Terminate_without_Refine; |
goto EPZS8_Terminate_without_Refine; |
2457 |
if (MotionFlags & PMV_EARLYSTOP8) |
if (MotionFlags & PMV_EARLYSTOP8) |
2467 |
|
|
2468 |
/* default: use best prediction as starting point for one call of EPZS_MainSearch */ |
/* default: use best prediction as starting point for one call of EPZS_MainSearch */ |
2469 |
|
|
2470 |
/* // there is no EPZS^2 for inter4v at the moment |
// there is no EPZS^2 for inter4v at the moment |
2471 |
|
|
2472 |
if (MotionFlags & PMV_USESQUARES8) |
if (MotionFlags & PMV_USESQUARES8) |
2473 |
EPZSMainSearchPtr = Square8_MainSearch; |
MainSearchPtr = Square8_MainSearch; |
2474 |
else |
else |
|
*/ |
|
2475 |
|
|
2476 |
EPZSMainSearchPtr = Diamond8_MainSearch; |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
2477 |
|
MainSearchPtr = AdvDiamond8_MainSearch; |
2478 |
|
else |
2479 |
|
MainSearchPtr = Diamond8_MainSearch; |
2480 |
|
|
2481 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
2482 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2483 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2484 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2485 |
iDiamondSize, iFcode, iQuant, 0); |
iQuant, 0); |
2486 |
|
|
2487 |
|
|
2488 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2489 |
*currMV = newMV; |
*currMV = newMV; |
2490 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2491 |
} |
} |
2492 |
|
|
2493 |
if (MotionFlags & PMV_EXTSEARCH8) |
if (MotionFlags & PMV_EXTSEARCH8) { |
|
{ |
|
2494 |
/* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
2495 |
|
|
2496 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0], backupMV))) { |
2497 |
{ |
iSAD = |
2498 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2499 |
x, y, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2500 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2501 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); |
iDiamondSize, iFcode, iQuant, 0); |
2502 |
|
|
2503 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2504 |
*currMV = newMV; |
*currMV = newMV; |
2505 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2506 |
} |
} |
2507 |
} |
} |
2508 |
|
|
2509 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2510 |
{ |
iSAD = |
2511 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2512 |
x, y, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2513 |
0, 0, iMinSAD, &newMV, |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2514 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); |
iQuant, 0); |
2515 |
|
|
2516 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2517 |
*currMV = newMV; |
*currMV = newMV; |
2518 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2519 |
} |
} |
2524 |
|
|
2525 |
EPZS8_Terminate_with_Refine: |
EPZS8_Terminate_with_Refine: |
2526 |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
2527 |
iMinSAD = Halfpel8_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
iMinSAD = |
2528 |
x, y, |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2529 |
currMV, iMinSAD, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2530 |
pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2531 |
|
|
2532 |
EPZS8_Terminate_without_Refine: |
EPZS8_Terminate_without_Refine: |
2533 |
|
|
2534 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2535 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2536 |
return iMinSAD; |
return iMinSAD; |
2537 |
} |
} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/* *********************************************************** |
|
|
bvop motion estimation |
|
|
// TODO: need to incorporate prediction here (eg. sad += calc_delta_16) |
|
|
***************************************************************/ |
|
|
|
|
|
|
|
|
void MotionEstimationBVOP( |
|
|
MBParam * const pParam, |
|
|
FRAMEINFO * const frame, |
|
|
|
|
|
// forward (past) reference |
|
|
const MACROBLOCK * const f_mbs, |
|
|
const IMAGE * const f_ref, |
|
|
const IMAGE * const f_refH, |
|
|
const IMAGE * const f_refV, |
|
|
const IMAGE * const f_refHV, |
|
|
// backward (future) reference |
|
|
const MACROBLOCK * const b_mbs, |
|
|
const IMAGE * const b_ref, |
|
|
const IMAGE * const b_refH, |
|
|
const IMAGE * const b_refV, |
|
|
const IMAGE * const b_refHV) |
|
|
{ |
|
|
const uint32_t mb_width = pParam->mb_width; |
|
|
const uint32_t mb_height = pParam->mb_height; |
|
|
const int32_t edged_width = pParam->edged_width; |
|
|
|
|
|
uint32_t i,j; |
|
|
|
|
|
int32_t f_sad16; |
|
|
int32_t b_sad16; |
|
|
int32_t i_sad16; |
|
|
int32_t d_sad16; |
|
|
int32_t best_sad; |
|
|
|
|
|
VECTOR pmv_dontcare; |
|
|
|
|
|
// note: i==horizontal, j==vertical |
|
|
for (j = 0; j < mb_height; j++) |
|
|
{ |
|
|
for (i = 0; i < mb_width; i++) |
|
|
{ |
|
|
MACROBLOCK *mb = &frame->mbs[i + j*mb_width]; |
|
|
const MACROBLOCK *f_mb = &f_mbs[i + j*mb_width]; |
|
|
const MACROBLOCK *b_mb = &b_mbs[i + j*mb_width]; |
|
|
|
|
|
if (b_mb->mode == MODE_INTER |
|
|
&& b_mb->cbp == 0 |
|
|
&& b_mb->mvs[0].x == 0 |
|
|
&& b_mb->mvs[0].y == 0) |
|
|
{ |
|
|
mb->mode = MODE_NOT_CODED; |
|
|
mb->mvs[0].x = 0; |
|
|
mb->mvs[0].y = 0; |
|
|
mb->b_mvs[0].x = 0; |
|
|
mb->b_mvs[0].y = 0; |
|
|
continue; |
|
|
} |
|
|
|
|
|
|
|
|
// forward search |
|
|
f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
|
|
&frame->image, |
|
|
i, j, |
|
|
frame->motion_flags, frame->quant, frame->fcode, |
|
|
pParam, |
|
|
f_mbs, f_mbs /* todo */, |
|
|
&mb->mvs[0], &pmv_dontcare); // ignore pmv |
|
|
|
|
|
// backward search |
|
|
b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
|
|
&frame->image, |
|
|
i, j, |
|
|
frame->motion_flags, frame->quant, frame->bcode, |
|
|
pParam, |
|
|
b_mbs, b_mbs, /* todo */ |
|
|
&mb->b_mvs[0], &pmv_dontcare); // ignore pmv |
|
|
|
|
|
// interpolate search (simple, but effective) |
|
|
i_sad16 = sad16bi_c( |
|
|
frame->image.y + i*16 + j*16*edged_width, |
|
|
get_ref(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
|
|
i, j, 16, mb->mvs[0].x, mb->mvs[0].y, edged_width), |
|
|
get_ref(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
|
|
i, j, 16, mb->b_mvs[0].x, mb->b_mvs[0].x, edged_width), |
|
|
edged_width); |
|
|
|
|
|
// TODO: direct search |
|
|
// predictor + range of [-32,32] |
|
|
d_sad16 = 65535; |
|
|
|
|
|
|
|
|
if (f_sad16 < b_sad16) |
|
|
{ |
|
|
best_sad = f_sad16; |
|
|
mb->mode = MODE_FORWARD; |
|
|
} |
|
|
else |
|
|
{ |
|
|
best_sad = b_sad16; |
|
|
mb->mode = MODE_BACKWARD; |
|
|
} |
|
|
|
|
|
if (i_sad16 < best_sad) |
|
|
{ |
|
|
best_sad = i_sad16; |
|
|
mb->mode = MODE_INTERPOLATE; |
|
|
} |
|
|
|
|
|
if (d_sad16 < best_sad) |
|
|
{ |
|
|
best_sad = d_sad16; |
|
|
mb->mode = MODE_DIRECT; |
|
|
} |
|
|
|
|
|
} |
|
|
} |
|
|
} |
|