31 |
#include <assert.h> |
#include <assert.h> |
32 |
#include <stdio.h> |
#include <stdio.h> |
33 |
#include <stdlib.h> |
#include <stdlib.h> |
34 |
|
#include <string.h> // memcpy |
35 |
|
|
36 |
#include "../encoder.h" |
#include "../encoder.h" |
37 |
#include "../utils/mbfunctions.h" |
#include "../utils/mbfunctions.h" |
48 |
#define FINAL_SKIP_THRESH (50) |
#define FINAL_SKIP_THRESH (50) |
49 |
#define MAX_SAD00_FOR_SKIP (20) |
#define MAX_SAD00_FOR_SKIP (20) |
50 |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
|
#define SKIP_THRESH_B (25) |
|
51 |
|
|
52 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
53 |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
54 |
|
|
|
#define iDiamondSize 2 |
|
|
|
|
55 |
static __inline int |
static __inline int |
56 |
d_mv_bits(int x, int y, const uint32_t iFcode) |
d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv) |
57 |
{ |
{ |
58 |
int xb, yb; |
int xb, yb; |
59 |
|
if (qpel) { x *= 2; y *= 2;} |
60 |
|
else if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
61 |
|
x -= pred.x; |
62 |
|
y -= pred.y; |
63 |
|
|
64 |
if (x == 0) xb = 1; |
if (x) { |
|
else { |
|
65 |
if (x < 0) x = -x; |
if (x < 0) x = -x; |
66 |
x += (1 << (iFcode - 1)) - 1; |
x += (1 << (iFcode - 1)) - 1; |
67 |
x >>= (iFcode - 1); |
x >>= (iFcode - 1); |
68 |
if (x > 32) x = 32; |
if (x > 32) x = 32; |
69 |
xb = mvtab[x] + iFcode; |
xb = mvtab[x] + iFcode; |
70 |
} |
} else xb = 1; |
71 |
|
|
72 |
if (y == 0) yb = 1; |
if (y) { |
|
else { |
|
73 |
if (y < 0) y = -y; |
if (y < 0) y = -y; |
74 |
y += (1 << (iFcode - 1)) - 1; |
y += (1 << (iFcode - 1)) - 1; |
75 |
y >>= (iFcode - 1); |
y >>= (iFcode - 1); |
76 |
if (y > 32) y = 32; |
if (y > 32) y = 32; |
77 |
yb = mvtab[y] + iFcode; |
yb = mvtab[y] + iFcode; |
78 |
} |
} else yb = 1; |
79 |
return xb + yb; |
return xb + yb; |
80 |
} |
} |
81 |
|
|
83 |
ChromaSAD(int dx, int dy, const SearchData * const data) |
ChromaSAD(int dx, int dy, const SearchData * const data) |
84 |
{ |
{ |
85 |
int sad; |
int sad; |
86 |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
const uint32_t stride = data->iEdgedWidth/2; |
87 |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
|
88 |
|
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently |
89 |
|
data->temp[5] = dx; data->temp[6] = dy; // backup |
90 |
|
|
91 |
switch (((dx & 1) << 1) + (dy & 1)) { // ((dx%2)?2:0)+((dy%2)?1:0) |
switch (((dx & 1) << 1) | (dy & 1)) { |
92 |
case 0: |
case 0: |
93 |
sad = sad8(data->CurU, data->RefCU + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2); |
dx = dx / 2; dy = dy / 2; |
94 |
sad += sad8(data->CurV, data->RefCV + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2); |
sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride); |
95 |
|
sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride); |
96 |
break; |
break; |
97 |
case 1: |
case 1: |
98 |
dx = dx / 2; dy = (dy - 1) / 2; |
dx = dx / 2; dy = (dy - 1) / 2; |
99 |
sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2); |
sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride); |
100 |
sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2); |
sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride); |
101 |
break; |
break; |
102 |
case 2: |
case 2: |
103 |
dx = (dx - 1) / 2; dy = dy / 2; |
dx = (dx - 1) / 2; dy = dy / 2; |
104 |
sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2); |
sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride); |
105 |
sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2); |
sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride); |
106 |
break; |
break; |
107 |
default: |
default: |
108 |
dx = (dx - 1) / 2; dy = (dy - 1) / 2; |
dx = (dx - 1) / 2; dy = (dy - 1) / 2; |
109 |
interpolate8x8_halfpel_hv(data->RefQ, |
interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding); |
110 |
data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2, |
sad = sad8(data->CurU, data->RefQ, stride); |
111 |
data->rounding); |
|
112 |
sad = sad8(data->CurU, data->RefQ, data->iEdgedWidth/2); |
interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding); |
113 |
interpolate8x8_halfpel_hv(data->RefQ, |
sad += sad8(data->CurV, data->RefQ, stride); |
|
data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2, |
|
|
data->rounding); |
|
|
sad += sad8(data->CurV, data->RefQ, data->iEdgedWidth/2); |
|
114 |
break; |
break; |
115 |
} |
} |
116 |
|
data->temp[7] = sad; //backup, part 2 |
117 |
return sad; |
return sad; |
118 |
} |
} |
119 |
|
|
120 |
static __inline const uint8_t * |
static __inline const uint8_t * |
121 |
GetReference(const int x, const int y, const int dir, const SearchData * const data) |
GetReferenceB(const int x, const int y, const int dir, const SearchData * const data) |
122 |
{ |
{ |
123 |
// dir : 0 = forward, 1 = backward |
// dir : 0 = forward, 1 = backward |
124 |
switch ( (dir << 2) | ((x&1)<<1) | (y&1) ) { |
switch ( (dir << 2) | ((x&1)<<1) | (y&1) ) { |
130 |
case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
131 |
case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
132 |
default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
133 |
|
} |
134 |
|
} |
135 |
|
|
136 |
|
// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate |
137 |
|
static __inline const uint8_t * |
138 |
|
GetReference(const int x, const int y, const SearchData * const data) |
139 |
|
{ |
140 |
|
switch ( ((x&1)<<1) | (y&1) ) { |
141 |
|
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
142 |
|
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
143 |
|
case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
144 |
|
default : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
145 |
} |
} |
146 |
} |
} |
147 |
|
|
149 |
Interpolate8x8qpel(const int x, const int y, const int block, const int dir, const SearchData * const data) |
Interpolate8x8qpel(const int x, const int y, const int block, const int dir, const SearchData * const data) |
150 |
{ |
{ |
151 |
// create or find a qpel-precision reference picture; return pointer to it |
// create or find a qpel-precision reference picture; return pointer to it |
152 |
uint8_t * Reference = (uint8_t *)data->RefQ; |
uint8_t * Reference = (uint8_t *)data->RefQ + 16*dir; |
153 |
const int32_t iEdgedWidth = data->iEdgedWidth; |
const int32_t iEdgedWidth = data->iEdgedWidth; |
154 |
const uint32_t rounding = data->rounding; |
const uint32_t rounding = data->rounding; |
155 |
const int halfpel_x = x/2; |
const int halfpel_x = x/2; |
156 |
const int halfpel_y = y/2; |
const int halfpel_y = y/2; |
157 |
const uint8_t *ref1, *ref2, *ref3, *ref4; |
const uint8_t *ref1, *ref2, *ref3, *ref4; |
158 |
|
|
159 |
ref1 = GetReference(halfpel_x, halfpel_y, dir, data); // this reference is used in all cases |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
160 |
ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
161 |
switch( ((x&1)<<1) + (y&1) ) { |
switch( ((x&1)<<1) + (y&1) ) { |
162 |
case 0: // pure halfpel position |
case 0: // pure halfpel position |
163 |
Reference = (uint8_t *) GetReference(halfpel_x, halfpel_y, dir, data); |
return (uint8_t *) ref1; |
|
Reference += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
|
164 |
break; |
break; |
165 |
|
|
166 |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
167 |
ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
168 |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
169 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
170 |
break; |
break; |
171 |
|
|
172 |
case 2: // x qpel, y halfpel - left or right during qpel refinement |
case 2: // x qpel, y halfpel - left or right during qpel refinement |
173 |
ref2 = GetReference(x - halfpel_x, halfpel_y, dir, data); |
ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
174 |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
175 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
176 |
break; |
break; |
177 |
|
|
178 |
default: // x and y in qpel resolution - the "corners" (top left/right and |
default: // x and y in qpel resolution - the "corners" (top left/right and |
179 |
// bottom left/right) during qpel refinement |
// bottom left/right) during qpel refinement |
180 |
ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
181 |
ref3 = GetReference(x - halfpel_x, halfpel_y, dir, data); |
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
182 |
ref4 = GetReference(x - halfpel_x, y - halfpel_y, dir, data); |
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
183 |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
184 |
ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
185 |
ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
193 |
Interpolate16x16qpel(const int x, const int y, const int dir, const SearchData * const data) |
Interpolate16x16qpel(const int x, const int y, const int dir, const SearchData * const data) |
194 |
{ |
{ |
195 |
// create or find a qpel-precision reference picture; return pointer to it |
// create or find a qpel-precision reference picture; return pointer to it |
196 |
uint8_t * Reference = (uint8_t *)data->RefQ; |
uint8_t * Reference = (uint8_t *)data->RefQ + 16*dir; |
197 |
const int32_t iEdgedWidth = data->iEdgedWidth; |
const int32_t iEdgedWidth = data->iEdgedWidth; |
198 |
const uint32_t rounding = data->rounding; |
const uint32_t rounding = data->rounding; |
199 |
const int halfpel_x = x/2; |
const int halfpel_x = x/2; |
200 |
const int halfpel_y = y/2; |
const int halfpel_y = y/2; |
201 |
const uint8_t *ref1, *ref2, *ref3, *ref4; |
const uint8_t *ref1, *ref2, *ref3, *ref4; |
202 |
|
|
203 |
ref1 = GetReference(halfpel_x, halfpel_y, dir, data); // this reference is used in all cases |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
204 |
switch( ((x&1)<<1) + (y&1) ) { |
switch( ((x&1)<<1) + (y&1) ) { |
205 |
case 0: // pure halfpel position |
case 0: // pure halfpel position |
206 |
return (uint8_t *) GetReference(halfpel_x, halfpel_y, dir, data); |
return (uint8_t *) ref1; |
207 |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
208 |
ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
209 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
210 |
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
211 |
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8); |
213 |
break; |
break; |
214 |
|
|
215 |
case 2: // x qpel, y halfpel - left or right during qpel refinement |
case 2: // x qpel, y halfpel - left or right during qpel refinement |
216 |
ref2 = GetReference(x - halfpel_x, halfpel_y, dir, data); |
ref2 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
217 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
218 |
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
219 |
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8); |
222 |
|
|
223 |
default: // x and y in qpel resolution - the "corners" (top left/right and |
default: // x and y in qpel resolution - the "corners" (top left/right and |
224 |
// bottom left/right) during qpel refinement |
// bottom left/right) during qpel refinement |
225 |
ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
226 |
ref3 = GetReference(x - halfpel_x, halfpel_y, dir, data); |
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
227 |
ref4 = GetReference(x - halfpel_x, y - halfpel_y, dir, data); |
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
228 |
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
229 |
interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); |
interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); |
230 |
interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); |
interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); |
248 |
|
|
249 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (data->qpel_precision) { // x and y are in 1/4 precision |
250 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
251 |
xc = x/2; yc = y/2; //for chroma sad |
xc = x/2; yc = y/2; //for chroma sad |
252 |
current = data->currentQMV; |
current = data->currentQMV; |
253 |
} else { |
} else { |
254 |
switch ( ((x&1)<<1) + (y&1) ) { |
Reference = GetReference(x, y, data); |
|
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
} |
|
|
if (data->qpel) t = d_mv_bits(2*x - data->predMV.x, 2*y - data->predMV.y, data->iFcode); |
|
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
255 |
current = data->currentMV; |
current = data->currentMV; |
256 |
xc = x; yc = y; |
xc = x; yc = y; |
257 |
} |
} |
258 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0); |
259 |
|
|
260 |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
261 |
|
|
262 |
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
263 |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100; |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100; |
264 |
|
|
265 |
if (data->chroma) data->temp[0] += ChromaSAD(xc, yc, data); |
if (data->chroma) data->temp[0] += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
266 |
|
(yc >> 1) + roundtab_79[yc & 0x3], data); |
267 |
|
|
268 |
if (data->temp[0] < data->iMinSAD[0]) { |
if (data->temp[0] < data->iMinSAD[0]) { |
269 |
data->iMinSAD[0] = data->temp[0]; |
data->iMinSAD[0] = data->temp[0]; |
282 |
} |
} |
283 |
|
|
284 |
static void |
static void |
285 |
|
CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
286 |
|
{ |
287 |
|
int t; |
288 |
|
const uint8_t * Reference; |
289 |
|
|
290 |
|
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || //non-zero integer value |
291 |
|
( x > data->max_dx) || ( x < data->min_dx) |
292 |
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
293 |
|
|
294 |
|
Reference = GetReference(x, y, data); |
295 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
296 |
|
|
297 |
|
data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
298 |
|
|
299 |
|
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
300 |
|
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100; |
301 |
|
|
302 |
|
if (data->temp[0] < data->iMinSAD[0]) { |
303 |
|
data->iMinSAD[0] = data->temp[0]; |
304 |
|
data->currentMV[0].x = x; data->currentMV[0].y = y; |
305 |
|
*dir = Direction; } |
306 |
|
|
307 |
|
if (data->temp[1] < data->iMinSAD[1]) { |
308 |
|
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
309 |
|
if (data->temp[2] < data->iMinSAD[2]) { |
310 |
|
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
311 |
|
if (data->temp[3] < data->iMinSAD[3]) { |
312 |
|
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
313 |
|
if (data->temp[4] < data->iMinSAD[4]) { |
314 |
|
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
315 |
|
} |
316 |
|
|
317 |
|
static void |
318 |
CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
319 |
{ |
{ |
320 |
int32_t sad; |
int32_t sad; |
325 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if (( x > data->max_dx) || ( x < data->min_dx) |
326 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
327 |
|
|
328 |
|
if (data->rrv) |
329 |
|
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) ) return; //non-zero integer value |
330 |
|
|
331 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (data->qpel_precision) { // x and y are in 1/4 precision |
332 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
333 |
current = data->currentQMV; |
current = data->currentQMV; |
334 |
} else { |
} else { |
335 |
switch ( ((x&1)<<1) + (y&1) ) { |
Reference = GetReference(x, y, data); |
|
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
} |
|
|
if (data->qpel) t = d_mv_bits(2*x - data->predMV.x, 2*y - data->predMV.y, data->iFcode); |
|
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
336 |
current = data->currentMV; |
current = data->currentMV; |
337 |
} |
} |
338 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, |
339 |
|
data->qpel && !data->qpel_precision, data->rrv); |
340 |
|
|
341 |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
342 |
sad += (data->lambda16 * t * sad)/1000; |
sad += (data->lambda16 * t * sad)/1000; |
348 |
} |
} |
349 |
|
|
350 |
static void |
static void |
351 |
CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
352 |
{ |
{ |
353 |
// maximum speed - for P/B/I decision |
// maximum speed - for P/B/I decision |
|
int32_t sad; |
|
354 |
|
|
355 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if (( x > data->max_dx) || ( x < data->min_dx) |
356 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
357 |
|
|
358 |
sad = sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
data->temp[0] = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
359 |
data->iEdgedWidth, 256*4096); |
data->iEdgedWidth, data->temp+1); |
360 |
|
if (data->temp[0] < *(data->iMinSAD)) { |
361 |
if (sad < *(data->iMinSAD)) { |
*(data->iMinSAD) = data->temp[0]; |
|
*(data->iMinSAD) = sad; |
|
362 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
363 |
*dir = Direction; } |
*dir = Direction; } |
364 |
} |
if (data->temp[1] < data->iMinSAD[1]) { |
365 |
|
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
366 |
|
if (data->temp[2] < data->iMinSAD[2]) { |
367 |
|
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
368 |
|
if (data->temp[3] < data->iMinSAD[3]) { |
369 |
|
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
370 |
|
if (data->temp[4] < data->iMinSAD[4]) { |
371 |
|
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
372 |
|
|
373 |
|
} |
374 |
|
|
375 |
static void |
static void |
376 |
CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data) |
388 |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
389 |
current = data->currentQMV; |
current = data->currentQMV; |
390 |
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
|
t = d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) |
|
|
+ d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode); |
|
391 |
} else { |
} else { |
392 |
ReferenceF = Interpolate16x16qpel(2*xf, 2*yf, 0, data); |
ReferenceF = GetReference(xf, yf, data); |
393 |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
394 |
ReferenceB = Interpolate16x16qpel(2*xb, 2*yb, 1, data); |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
395 |
current = data->currentMV; |
current = data->currentMV; |
|
if (data->qpel) |
|
|
t = d_mv_bits(2*xf - data->predMV.x, 2*yf - data->predMV.y, data->iFcode) |
|
|
+ d_mv_bits(2*xb - data->bpredMV.x, 2*yb - data->bpredMV.y, data->iFcode); |
|
|
else |
|
|
t = d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) |
|
|
+ d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode); |
|
396 |
} |
} |
397 |
|
|
398 |
|
t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0) |
399 |
|
+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel && !data->qpel_precision, 0); |
400 |
|
|
401 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
402 |
sad += (data->lambda16 * t * sad)/1000; |
sad += (data->lambda16 * t * sad)/1000; |
403 |
|
|
415 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
416 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
417 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
418 |
|
const VECTOR zeroMV={0,0}; |
419 |
|
|
420 |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
421 |
|
|
448 |
if (sad > *(data->iMinSAD)) return; |
if (sad > *(data->iMinSAD)) return; |
449 |
} |
} |
450 |
|
|
451 |
sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000; |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000; |
452 |
|
|
453 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
454 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
463 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
464 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
465 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
466 |
|
const VECTOR zeroMV = {0,0}; |
467 |
|
|
468 |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
469 |
|
|
490 |
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
491 |
|
|
492 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
493 |
sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000; |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000; |
494 |
|
|
495 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
496 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
508 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
509 |
|
|
510 |
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
511 |
else Reference = Interpolate16x16qpel(2*x, 2*y, 0, data); |
else Reference = GetReference(x, y, data); |
512 |
|
|
513 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
514 |
if (data->qpel) t = d_mv_bits(2 * x - data->predMV.x, 2 * y - data->predMV.y, data->iFcode); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0); |
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
515 |
|
|
516 |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100; |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100; |
517 |
|
|
533 |
|
|
534 |
int iDirection; |
int iDirection; |
535 |
|
|
536 |
do { |
for(;;) { //forever |
537 |
iDirection = 0; |
iDirection = 0; |
538 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
539 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
551 |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
552 |
} else { // what remains here is up or down |
} else { // what remains here is up or down |
553 |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
554 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); } |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
555 |
|
} |
556 |
|
|
557 |
if (iDirection) { |
if (iDirection) { |
558 |
bDirection += iDirection; |
bDirection += iDirection; |
559 |
x = data->currentMV->x; y = data->currentMV->y; } |
x = data->currentMV->x; y = data->currentMV->y; |
560 |
|
} |
561 |
} else { //about to quit, eh? not so fast.... |
} else { //about to quit, eh? not so fast.... |
562 |
switch (bDirection) { |
switch (bDirection) { |
563 |
case 2: |
case 2: |
608 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
609 |
} |
} |
610 |
} |
} |
|
while (1); //forever |
|
611 |
} |
} |
612 |
|
|
613 |
static void |
static void |
657 |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
658 |
} else { // what remains here is up or down |
} else { // what remains here is up or down |
659 |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
660 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); } |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
661 |
|
} |
662 |
bDirection += iDirection; |
bDirection += iDirection; |
663 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
664 |
} |
} |
681 |
backupMV = *(data->currentQMV); |
backupMV = *(data->currentQMV); |
682 |
else backupMV = *(data->currentMV); |
else backupMV = *(data->currentMV); |
683 |
|
|
684 |
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
685 |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); |
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0); |
|
|
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0); |
|
|
|
|
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0); |
|
686 |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0); |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0); |
687 |
|
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0); |
688 |
CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); |
CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); |
689 |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0); |
690 |
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0); |
691 |
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); |
692 |
} |
} |
693 |
|
|
694 |
static __inline int |
static __inline int |
695 |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
696 |
const int x, const int y, |
const int x, const int y, |
697 |
const uint32_t iEdgedWidth, const uint32_t iQuant) |
const uint32_t iEdgedWidth, const uint32_t iQuant, int rrv) |
698 |
|
|
699 |
{ |
{ |
700 |
/* keep repeating checks for all b-frames before this P frame, |
/* keep repeating checks for all b-frames before this P frame, |
701 |
to make sure that SKIP is possible (todo) |
to make sure that SKIP is possible (todo) |
702 |
how: if skip is not possible set sad00 to a very high value */ |
how: if skip is not possible set sad00 to a very high value */ |
703 |
|
if(rrv) { |
704 |
|
uint32_t sadC = sad16(current->u + x*16 + y*(iEdgedWidth/2)*16, |
705 |
|
reference->u + x*16 + y*(iEdgedWidth/2)*16, iEdgedWidth/2, 256*4096); |
706 |
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
707 |
|
sadC += sad16(current->v + (x + y*(iEdgedWidth/2))*16, |
708 |
|
reference->v + (x + y*(iEdgedWidth/2))*16, iEdgedWidth/2, 256*4096); |
709 |
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
710 |
|
return 1; |
711 |
|
} else { |
712 |
uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8, |
uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8, |
713 |
reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); |
reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); |
714 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
715 |
sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8, |
sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8, |
716 |
reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
717 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
|
|
|
718 |
return 1; |
return 1; |
719 |
} |
} |
720 |
|
} |
721 |
|
|
722 |
static __inline void |
static __inline void |
723 |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
747 |
|
|
748 |
const VECTOR zeroMV = { 0, 0 }; |
const VECTOR zeroMV = { 0, 0 }; |
749 |
|
|
750 |
|
uint32_t mb_width = pParam->mb_width; |
751 |
|
uint32_t mb_height = pParam->mb_height; |
752 |
|
|
753 |
uint32_t x, y; |
uint32_t x, y; |
754 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
755 |
int32_t InterBias, quant = current->quant, sad00; |
int32_t InterBias, quant = current->quant, sad00; |
|
uint8_t *qimage; |
|
756 |
|
|
757 |
// some pre-initialized thingies for SearchP |
// some pre-initialized thingies for SearchP |
758 |
int32_t temp[5]; |
int32_t temp[8]; |
759 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
760 |
VECTOR currentQMV[5]; |
VECTOR currentQMV[5]; |
761 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
762 |
SearchData Data; |
SearchData Data; |
763 |
|
memset(&Data, 0, sizeof(SearchData)); |
764 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
765 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
766 |
Data.currentQMV = currentQMV; |
Data.currentQMV = currentQMV; |
770 |
Data.rounding = pParam->m_rounding_type; |
Data.rounding = pParam->m_rounding_type; |
771 |
Data.qpel = pParam->m_quarterpel; |
Data.qpel = pParam->m_quarterpel; |
772 |
Data.chroma = current->global_flags & XVID_ME_COLOUR; |
Data.chroma = current->global_flags & XVID_ME_COLOUR; |
773 |
|
Data.rrv = current->global_flags & XVID_REDUCED; |
774 |
|
|
775 |
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
if ((current->global_flags & XVID_REDUCED)) { |
776 |
return 1; // allocate some mem for qpel interpolated blocks |
mb_width = (pParam->width + 31) / 32; |
777 |
// somehow this is dirty since I think we shouldn't use malloc outside |
mb_height = (pParam->height + 31) / 32; |
778 |
// encoder_create() - so please fix me! |
Data.qpel = Data.chroma = 0; |
779 |
Data.RefQ = qimage; |
} |
780 |
|
|
781 |
|
Data.RefQ = pRefV->u; // a good place, also used in MC (for similar purpose) |
782 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
783 |
|
|
784 |
for (y = 0; y < pParam->mb_height; y++) { |
for (y = 0; y < mb_height; y++) { |
785 |
for (x = 0; x < pParam->mb_width; x++) { |
for (x = 0; x < mb_width; x++) { |
786 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
787 |
|
|
788 |
pMB->sad16 |
if (Data.rrv) pMB->sad16 = |
789 |
= sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16, |
sad32v_c(pCurrent->y + (x + y * pParam->edged_width) * 32, |
790 |
|
pRef->y + (x + y * pParam->edged_width) * 32, |
791 |
|
pParam->edged_width, pMB->sad8 ); |
792 |
|
|
793 |
|
else pMB->sad16 = |
794 |
|
sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16, |
795 |
pRef->y + (x + y * pParam->edged_width) * 16, |
pRef->y + (x + y * pParam->edged_width) * 16, |
796 |
pParam->edged_width, pMB->sad8 ); |
pParam->edged_width, pMB->sad8 ); |
797 |
|
|
820 |
//initial skip decision |
//initial skip decision |
821 |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
822 |
if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
823 |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH) |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
824 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) { |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) { |
825 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
826 |
continue; |
continue; |
827 |
} |
} |
835 |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
836 |
if (current->coding_type == P_VOP) { |
if (current->coding_type == P_VOP) { |
837 |
if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
838 |
&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) ) |
&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1)) ) |
839 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) { |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) { |
840 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
841 |
continue; |
continue; |
842 |
} |
} |
852 |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
853 |
|
|
854 |
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
855 |
|
if (Data.rrv) InterBias *= 4; //?? |
856 |
|
|
857 |
if (InterBias < pMB->sad16) { |
if (InterBias < pMB->sad16) { |
858 |
const int32_t deviation = |
int32_t deviation; |
859 |
dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
if (Data.rrv) { |
860 |
|
deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 32, |
861 |
|
pParam->edged_width) |
862 |
|
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16, |
863 |
|
pParam->edged_width) |
864 |
|
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * pParam->edged_width, |
865 |
|
pParam->edged_width) |
866 |
|
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * (pParam->edged_width+1), |
867 |
|
pParam->edged_width); |
868 |
|
} else |
869 |
|
deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
870 |
pParam->edged_width); |
pParam->edged_width); |
871 |
|
|
872 |
if (deviation < (pMB->sad16 - InterBias)) { |
if (deviation < (pMB->sad16 - InterBias)) { |
873 |
if (++iIntra >= iLimit) { free(qimage); return 1; } |
if (++iIntra >= iLimit) return 1; |
874 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
875 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
876 |
pMB->mvs[3] = zeroMV; |
pMB->mvs[3] = zeroMV; |
882 |
} |
} |
883 |
} |
} |
884 |
} |
} |
|
free(qimage); |
|
885 |
|
|
886 |
if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */ |
if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */ |
887 |
current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode ); |
current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode ); |
913 |
} |
} |
914 |
|
|
915 |
static __inline void |
static __inline void |
916 |
PreparePredictionsP(VECTOR * const pmv, int x, int y, const int iWcount, |
PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount, |
917 |
const int iHcount, const MACROBLOCK * const prevMB) |
int iHcount, const MACROBLOCK * const prevMB, int rrv) |
918 |
{ |
{ |
919 |
|
|
920 |
//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself |
//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself |
921 |
|
if (rrv) { iWcount /= 2; iHcount /= 2; } |
922 |
|
|
923 |
if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour |
if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour |
924 |
pmv[5].x = EVEN(pmv[3].x); |
pmv[5].x = EVEN(pmv[3].x); |
925 |
pmv[5].y = EVEN(pmv[3].y); |
pmv[5].y = EVEN(pmv[3].y); |
926 |
} else pmv[5].x = pmv[5].y = 0; |
} else pmv[5].x = pmv[5].y = 0; |
939 |
pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame |
pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame |
940 |
pmv[2].y = EVEN(prevMB->mvs[0].y); |
pmv[2].y = EVEN(prevMB->mvs[0].y); |
941 |
|
|
942 |
if ((x != iWcount-1) && (y != iHcount-1)) { |
if ((x < iWcount-1) && (y < iHcount-1)) { |
943 |
pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame |
pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame |
944 |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
945 |
} else pmv[6].x = pmv[6].y = 0; |
} else pmv[6].x = pmv[6].y = 0; |
946 |
|
|
947 |
|
if (rrv) { |
948 |
|
int i; |
949 |
|
for (i = 0; i < 7; i++) { |
950 |
|
pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); // halfzero->halfpel |
951 |
|
pmv[i].y = RRV_MV_SCALEUP(pmv[i].y); |
952 |
|
} |
953 |
|
} |
954 |
} |
} |
955 |
|
|
956 |
static void |
static void |
974 |
int i, iDirection = 255, mask, threshA; |
int i, iDirection = 255, mask, threshA; |
975 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
976 |
|
|
|
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
|
977 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
978 |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
979 |
|
|
980 |
|
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
981 |
|
|
982 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
Data->temp[5] = Data->temp[7] = 256*4096; // to reset chroma-sad cache |
983 |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
if (Data->rrv) i = 2; else i = 1; |
984 |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i; |
985 |
|
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
986 |
Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
987 |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
|
988 |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16*i; |
989 |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16*i; |
990 |
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16*i; |
991 |
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16*i; |
992 |
|
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
993 |
|
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
994 |
|
|
995 |
Data->lambda16 = lambda_vec16[iQuant]; |
Data->lambda16 = lambda_vec16[iQuant]; |
996 |
Data->lambda8 = lambda_vec8[iQuant]; |
Data->lambda8 = lambda_vec8[iQuant]; |
997 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
998 |
|
|
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
|
|
Data->min_dx = EVEN(Data->min_dx); |
|
|
Data->max_dx = EVEN(Data->max_dx); |
|
|
Data->min_dy = EVEN(Data->min_dy); |
|
|
Data->max_dy = EVEN(Data->max_dy); } |
|
|
|
|
999 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
1000 |
|
|
1001 |
for(i = 0; i < 5; i++) |
for(i = 0; i < 5; i++) |
1004 |
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1005 |
else Data->predMV = pmv[0]; |
else Data->predMV = pmv[0]; |
1006 |
|
|
1007 |
i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); |
i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0); |
1008 |
Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000; |
Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000; |
1009 |
Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100; |
Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100; |
1010 |
Data->iMinSAD[2] = pMB->sad8[1]; |
Data->iMinSAD[2] = pMB->sad8[1]; |
1018 |
if (threshA > 1024) threshA = 1024; } |
if (threshA > 1024) threshA = 1024; } |
1019 |
|
|
1020 |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
1021 |
prevMBs + x + y * pParam->mb_width); |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
1022 |
|
|
1023 |
if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
if (Data->rrv) CheckCandidate = CheckCandidate32; |
1024 |
|
else if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
1025 |
else CheckCandidate = CheckCandidate16no4v; //for extra speed |
else CheckCandidate = CheckCandidate16no4v; //for extra speed |
1026 |
|
|
1027 |
/* main loop. checking all predictions */ |
/* main loop. checking all predictions */ |
1052 |
if (MotionFlags & PMV_EXTSEARCH16) { |
if (MotionFlags & PMV_EXTSEARCH16) { |
1053 |
int32_t bSAD; |
int32_t bSAD; |
1054 |
VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0]; |
VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0]; |
1055 |
|
if (Data->rrv) { |
1056 |
|
startMV.x = RRV_MV_SCALEUP(startMV.x); |
1057 |
|
startMV.y = RRV_MV_SCALEUP(startMV.y); |
1058 |
|
} else |
1059 |
if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel? |
if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel? |
1060 |
startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y); |
startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y); |
1061 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1069 |
} |
} |
1070 |
|
|
1071 |
backupMV = Data->currentMV[0]; |
backupMV = Data->currentMV[0]; |
1072 |
if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1; |
if (!MotionFlags & PMV_HALFPELREFINE16 || Data->rrv) startMV.x = startMV.y = 0; |
1073 |
else startMV.x = startMV.y = 0; |
else startMV.x = startMV.y = 1; |
1074 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1075 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1076 |
|
|
1090 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
1091 |
} |
} |
1092 |
|
|
1093 |
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
if((Data->qpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
1094 |
|
|
1095 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1096 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1097 |
pParam->width, pParam->height, Data->iFcode, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
1098 |
|
|
1099 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1100 |
} |
} |
1107 |
Data8.iEdgedWidth = Data->iEdgedWidth; |
Data8.iEdgedWidth = Data->iEdgedWidth; |
1108 |
Data8.RefQ = Data->RefQ; |
Data8.RefQ = Data->RefQ; |
1109 |
Data8.qpel = Data->qpel; |
Data8.qpel = Data->qpel; |
1110 |
|
Data8.rrv = Data->rrv; |
1111 |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
1112 |
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
1113 |
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
1114 |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
1115 |
|
|
1116 |
if (Data->chroma) { |
if (Data->chroma) { |
1117 |
int sumx, sumy, dx, dy; |
int sumx, sumy; |
1118 |
|
|
1119 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1120 |
sumx= pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2; |
sumx= pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2; |
1123 |
sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
1124 |
sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
1125 |
} |
} |
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
|
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
|
1126 |
|
|
1127 |
Data->iMinSAD[1] += ChromaSAD(dx, dy, Data); |
Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf], |
1128 |
|
(sumy >> 3) + roundtab_76[sumy & 0xf], Data); |
1129 |
} |
} |
1130 |
} |
} |
1131 |
|
|
1132 |
|
if (Data->rrv) { |
1133 |
|
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
1134 |
|
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
1135 |
|
} |
1136 |
|
|
1137 |
if (!(inter4v) || |
if (!(inter4v) || |
1138 |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
1139 |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
1172 |
const int block, |
const int block, |
1173 |
SearchData * const Data) |
SearchData * const Data) |
1174 |
{ |
{ |
1175 |
|
int i = 0; |
1176 |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
1177 |
Data->currentMV = OldData->currentMV + 1 + block; |
Data->currentMV = OldData->currentMV + 1 + block; |
1178 |
Data->currentQMV = OldData->currentQMV + 1 + block; |
Data->currentQMV = OldData->currentQMV + 1 + block; |
1179 |
|
|
1180 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1181 |
Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1182 |
if (block != 0) *(Data->iMinSAD) += (Data->lambda8 * |
if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y, |
1183 |
d_mv_bits( Data->currentQMV->x - Data->predMV.x, |
Data->predMV, Data->iFcode, 0, 0); |
1184 |
Data->currentQMV->y - Data->predMV.y, |
|
|
Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
|
1185 |
} else { |
} else { |
1186 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1187 |
if (block != 0) *(Data->iMinSAD) += (Data->lambda8 * |
if (block != 0) { |
1188 |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y, |
1189 |
Data->currentMV->y - Data->predMV.y, |
Data->predMV, Data->iFcode, 0, Data->rrv); |
|
Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
|
1190 |
} |
} |
1191 |
|
} |
1192 |
|
|
1193 |
|
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
1194 |
|
|
1195 |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
1196 |
|
if (Data->rrv) i = 2; else i = 1; |
1197 |
|
|
1198 |
Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Ref = OldData->Ref + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1199 |
Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefH = OldData->RefH + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1200 |
Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefV = OldData->RefV + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1201 |
Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefHV = OldData->RefHV + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1202 |
|
|
1203 |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Cur = OldData->Cur + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1204 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1205 |
|
|
1206 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1207 |
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, OldData->iFcode - Data->qpel, 0, Data->rrv); |
1208 |
CheckCandidate = CheckCandidate8; |
|
1209 |
|
if (Data->rrv) CheckCandidate = CheckCandidate16no4v; |
1210 |
|
else CheckCandidate = CheckCandidate8; |
1211 |
|
|
1212 |
if (MotionFlags & PMV_EXTSEARCH8) { |
if (MotionFlags & PMV_EXTSEARCH8) { |
1213 |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1236 |
} |
} |
1237 |
} |
} |
1238 |
|
|
1239 |
if(pParam->m_quarterpel) { |
if(Data->qpel) { |
1240 |
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
1241 |
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
1242 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1243 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1244 |
pParam->width, pParam->height, OldData->iFcode, 0); |
pParam->width, pParam->height, OldData->iFcode, 1, 0); |
1245 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1246 |
} |
} |
1247 |
} |
} |
1248 |
} |
} |
1249 |
|
|
1250 |
if(pParam->m_quarterpel) { |
if (Data->rrv) { |
1251 |
|
Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x); |
1252 |
|
Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y); |
1253 |
|
} |
1254 |
|
|
1255 |
|
if(Data->qpel) { |
1256 |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
1257 |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
1258 |
pMB->qmvs[block] = *(Data->currentQMV); |
pMB->qmvs[block] = *(Data->currentQMV); |
1259 |
} |
} else { |
|
else { |
|
1260 |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
1261 |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
1262 |
} |
} |
1307 |
|
|
1308 |
if ((x != 0)&&(y != 0)) { |
if ((x != 0)&&(y != 0)) { |
1309 |
pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr); |
pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr); |
1310 |
pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y); |
pmv[6].x = EVEN(pmv[6].x); pmv[6].y = EVEN(pmv[6].y); |
1311 |
} else pmv[6].x = pmv[6].y = 0; |
} else pmv[6].x = pmv[6].y = 0; |
1312 |
|
|
1313 |
// more? |
// more? |
1334 |
|
|
1335 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1336 |
|
|
1337 |
int i, iDirection, mask; |
int i, iDirection = 255, mask; |
1338 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
1339 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1340 |
*Data->iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
1349 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1350 |
|
|
1351 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1352 |
pParam->width, pParam->height, iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, iFcode - Data->qpel, 0, 0); |
1353 |
|
|
1354 |
pmv[0] = Data->predMV; |
pmv[0] = Data->predMV; |
1355 |
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
1359 |
CheckCandidate = CheckCandidate16no4v; |
CheckCandidate = CheckCandidate16no4v; |
1360 |
|
|
1361 |
// main loop. checking all predictions |
// main loop. checking all predictions |
1362 |
for (i = 0; i < 8; i++) { |
for (i = 0; i < 7; i++) { |
1363 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!(mask = make_mask(pmv, i)) ) continue; |
1364 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
1365 |
} |
} |
1370 |
MainSearchPtr = AdvDiamondSearch; |
MainSearchPtr = AdvDiamondSearch; |
1371 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1372 |
|
|
1373 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
1374 |
|
|
1375 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1376 |
|
|
1377 |
if (Data->qpel) { |
if (Data->qpel && *Data->iMinSAD < *best_sad + 300) { |
1378 |
Data->currentQMV->x = 2*Data->currentMV->x; |
Data->currentQMV->x = 2*Data->currentMV->x; |
1379 |
Data->currentQMV->y = 2*Data->currentMV->y; |
Data->currentQMV->y = 2*Data->currentMV->y; |
1380 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1381 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1382 |
pParam->width, pParam->height, iFcode, 0); |
pParam->width, pParam->height, iFcode, 1, 0); |
1383 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1384 |
} |
} |
1385 |
|
|
1386 |
// three bits are needed to code backward mode. four for forward |
// three bits are needed to code backward mode. four for forward |
1387 |
// we treat the bits just like they were vector's |
|
1388 |
if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * Data->lambda16; |
if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * Data->lambda16; |
1389 |
else *Data->iMinSAD += 3 * Data->lambda16; |
else *Data->iMinSAD += 3 * Data->lambda16; |
1390 |
|
|
1402 |
pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
1403 |
pMB->pmvs[0].y = Data->currentMV->y - predMV->y; |
pMB->pmvs[0].y = Data->currentMV->y - predMV->y; |
1404 |
} |
} |
1405 |
if (mode_current == MODE_FORWARD) |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV; |
1406 |
pMB->mvs[0] = *(Data->currentMV+2) = *Data->currentMV; |
else pMB->b_mvs[0] = *Data->currentMV; |
1407 |
else |
} |
1408 |
pMB->b_mvs[0] = *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search |
if (mode_current == MODE_FORWARD) *(Data->currentMV+2) = *Data->currentMV; |
1409 |
|
else *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search |
1410 |
|
|
1411 |
} |
} |
1412 |
|
|
1413 |
|
static void |
1414 |
|
SkipDecisionB(const IMAGE * const pCur, |
1415 |
|
const IMAGE * const f_Ref, |
1416 |
|
const IMAGE * const b_Ref, |
1417 |
|
MACROBLOCK * const pMB, |
1418 |
|
const uint32_t x, const uint32_t y, |
1419 |
|
const SearchData * const Data) |
1420 |
|
{ |
1421 |
|
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
1422 |
|
uint32_t sum; |
1423 |
|
const int div = 1 + Data->qpel; |
1424 |
|
int k; |
1425 |
|
const uint32_t quant = pMB->quant; |
1426 |
|
//this is not full chroma compensation, only it's fullpel approximation. should work though |
1427 |
|
|
1428 |
|
for (k = 0; k < 4; k++) { |
1429 |
|
dy += Data->directmvF[k].y / div; |
1430 |
|
dx += Data->directmvF[0].x / div; |
1431 |
|
b_dy += Data->directmvB[0].y / div; |
1432 |
|
b_dx += Data->directmvB[0].x / div; |
1433 |
} |
} |
1434 |
|
|
1435 |
static int32_t |
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
1436 |
|
dx = (dx >> 3) + roundtab_76[dx & 0xf]; |
1437 |
|
b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf]; |
1438 |
|
b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf]; |
1439 |
|
|
1440 |
|
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
1441 |
|
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1442 |
|
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1443 |
|
Data->iEdgedWidth/2); |
1444 |
|
|
1445 |
|
if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * quant) return; //no skip |
1446 |
|
|
1447 |
|
sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2), |
1448 |
|
f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1449 |
|
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1450 |
|
Data->iEdgedWidth/2); |
1451 |
|
|
1452 |
|
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped |
1453 |
|
} |
1454 |
|
|
1455 |
|
|
1456 |
|
|
1457 |
|
static __inline uint32_t |
1458 |
SearchDirect(const IMAGE * const f_Ref, |
SearchDirect(const IMAGE * const f_Ref, |
1459 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1460 |
const uint8_t * const f_RefV, |
const uint8_t * const f_RefV, |
1501 |
Data->min_dy *= 2; |
Data->min_dy *= 2; |
1502 |
Data->referencemv = b_mb->qmvs; |
Data->referencemv = b_mb->qmvs; |
1503 |
} else Data->referencemv = b_mb->mvs; |
} else Data->referencemv = b_mb->mvs; |
1504 |
Data->qpel_precision = 0; // it'm a trick. it's 1 not 0, but we need 0 here |
Data->qpel_precision = 0; // it's a trick. it's 1 not 0, but we need 0 here |
1505 |
|
|
1506 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1507 |
pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD); |
pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD); |
1515 |
*best_sad = 256*4096; // in that case, we won't use direct mode |
*best_sad = 256*4096; // in that case, we won't use direct mode |
1516 |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
1517 |
pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; |
pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; |
1518 |
return 0; |
return 256*4096; |
1519 |
} |
} |
1520 |
if (b_mb->mode != MODE_INTER4V) { |
if (b_mb->mode != MODE_INTER4V) { |
1521 |
pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; |
pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; |
1526 |
} |
} |
1527 |
} |
} |
1528 |
|
|
|
|
|
1529 |
if (b_mb->mode == MODE_INTER4V) CheckCandidate = CheckCandidateDirect; |
if (b_mb->mode == MODE_INTER4V) CheckCandidate = CheckCandidateDirect; |
1530 |
else CheckCandidate = CheckCandidateDirectno4v; |
else CheckCandidate = CheckCandidateDirectno4v; |
1531 |
|
|
1532 |
(*CheckCandidate)(0, 0, 255, &k, Data); |
(*CheckCandidate)(0, 0, 255, &k, Data); |
1533 |
|
|
1534 |
// skip decision |
// initial (fast) skip decision |
1535 |
if (*Data->iMinSAD < pMB->quant * SKIP_THRESH_B) { |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH*2) { |
1536 |
//possible skip - checking chroma. everything copied from MC |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); //possible skip - checking chroma |
1537 |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip. |
|
int sum, dx, dy, b_dx, b_dy; |
|
|
|
|
|
if (Data->qpel) { |
|
|
sum = pMB->mvs[0].y/2 + pMB->mvs[1].y/2 + pMB->mvs[2].y/2 + pMB->mvs[3].y/2; |
|
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
sum = pMB->mvs[0].x/2 + pMB->mvs[1].x/2 + pMB->mvs[2].x/2 + pMB->mvs[3].x/2; |
|
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
|
|
|
sum = pMB->b_mvs[0].y/2 + pMB->b_mvs[1].y/2 + pMB->b_mvs[2].y/2 + pMB->b_mvs[3].y/2; |
|
|
b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
sum = pMB->b_mvs[0].x/2 + pMB->b_mvs[1].x/2 + pMB->b_mvs[2].x/2 + pMB->b_mvs[3].x/2; |
|
|
b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
|
|
|
} else { |
|
|
sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
|
|
dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
|
|
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
|
|
|
sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; |
|
|
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
|
|
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
} |
|
|
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
|
|
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
|
|
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
|
|
Data->iEdgedWidth/2); |
|
|
sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2), |
|
|
f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
|
|
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
|
|
Data->iEdgedWidth/2); |
|
|
|
|
|
if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
|
|
pMB->mode = MODE_DIRECT_NONE_MV; |
|
|
return *Data->iMinSAD; |
|
|
} |
|
1538 |
} |
} |
1539 |
|
|
1540 |
skip_sad = *Data->iMinSAD; |
skip_sad = *Data->iMinSAD; |
1550 |
|
|
1551 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1552 |
|
|
|
*Data->iMinSAD += 1 * Data->lambda16; // one bit is needed to code direct mode |
|
1553 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1554 |
|
|
1555 |
// if (b_mb->mode == MODE_INTER4V) |
if (b_mb->mode == MODE_INTER4V || Data->qpel) pMB->mode = MODE_DIRECT; |
1556 |
pMB->mode = MODE_DIRECT; |
else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
|
// else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
|
1557 |
|
|
1558 |
pMB->pmvs[3] = *Data->currentMV; |
pMB->pmvs[3] = *Data->currentMV; |
1559 |
|
|
1585 |
} |
} |
1586 |
|
|
1587 |
|
|
1588 |
static __inline void |
static void |
1589 |
SearchInterpolate(const uint8_t * const f_Ref, |
SearchInterpolate(const uint8_t * const f_Ref, |
1590 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1591 |
const uint8_t * const f_RefV, |
const uint8_t * const f_RefV, |
1608 |
|
|
1609 |
{ |
{ |
1610 |
|
|
|
const int32_t iEdgedWidth = pParam->edged_width; |
|
1611 |
int iDirection, i, j; |
int iDirection, i, j; |
1612 |
SearchData bData; |
SearchData bData; |
1613 |
|
|
1614 |
*(bData.iMinSAD = fData->iMinSAD) = 4096*256; |
fData->qpel_precision = 0; |
1615 |
bData.Cur = fData->Cur; |
memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data |
1616 |
fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
*fData->iMinSAD = 4096*256; |
1617 |
bData.currentMV = fData->currentMV + 1; bData.currentQMV = fData->currentQMV + 1; |
bData.currentMV++; bData.currentQMV++; |
|
bData.lambda16 = fData->lambda16; |
|
1618 |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1619 |
|
|
1620 |
bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
i = (x + y * fData->iEdgedWidth) * 16; |
1621 |
bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16; |
bData.bRef = fData->Ref = f_Ref + i; |
1622 |
bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16; |
bData.bRefH = fData->RefH = f_RefH + i; |
1623 |
bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16; |
bData.bRefV = fData->RefV = f_RefV + i; |
1624 |
bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16; |
bData.bRefHV = fData->RefHV = f_RefHV + i; |
1625 |
bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16; |
bData.Ref = fData->bRef = b_Ref + i; |
1626 |
bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16; |
bData.RefH = fData->bRefH = b_RefH + i; |
1627 |
bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; |
bData.RefV = fData->bRefV = b_RefV + i; |
1628 |
bData.RefQ = fData->RefQ; |
bData.RefHV = fData->bRefHV = b_RefHV + i; |
|
fData->qpel_precision = bData.qpel_precision = 0; |
|
|
bData.rounding = 0; |
|
1629 |
|
|
1630 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
1631 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
|
|
|
1632 |
fData->currentMV[0] = fData->currentMV[2]; |
fData->currentMV[0] = fData->currentMV[2]; |
1633 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, pParam->m_quarterpel); |
|
1634 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, pParam->m_quarterpel); |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - fData->qpel, 0, 0); |
1635 |
|
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode - fData->qpel, 0, 0); |
1636 |
|
|
1637 |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
1638 |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy; |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx; |
1639 |
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx; |
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy; |
1640 |
if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
1641 |
|
|
1642 |
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
1643 |
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy; |
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx; |
1644 |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx; |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy; |
1645 |
if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
1646 |
|
|
1647 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
1648 |
|
|
1668 |
|
|
1669 |
} while (!(iDirection)); |
} while (!(iDirection)); |
1670 |
|
|
|
*fData->iMinSAD += 2 * fData->lambda16; // two bits are needed to code interpolate mode. |
|
|
|
|
1671 |
if (fData->qpel) { |
if (fData->qpel) { |
1672 |
|
if (*fData->iMinSAD > *best_sad + 500) return; |
1673 |
|
CheckCandidate = CheckCandidateInt; |
1674 |
fData->qpel_precision = bData.qpel_precision = 1; |
fData->qpel_precision = bData.qpel_precision = 1; |
1675 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 0); |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 1, 0); |
1676 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 0); |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0); |
1677 |
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
1678 |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
1679 |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
1680 |
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
1681 |
SubpelRefine(fData); |
SubpelRefine(fData); |
1682 |
|
if (*fData->iMinSAD > *best_sad + 300) return; |
1683 |
fData->currentQMV[2] = fData->currentQMV[0]; |
fData->currentQMV[2] = fData->currentQMV[0]; |
1684 |
SubpelRefine(&bData); |
SubpelRefine(&bData); |
1685 |
} |
} |
1686 |
|
|
1687 |
|
*fData->iMinSAD += (2+3) * fData->lambda16; // two bits are needed to code interpolate mode. |
1688 |
|
|
1689 |
if (*fData->iMinSAD < *best_sad) { |
if (*fData->iMinSAD < *best_sad) { |
1690 |
*best_sad = *fData->iMinSAD; |
*best_sad = *fData->iMinSAD; |
1691 |
pMB->mvs[0] = fData->currentMV[0]; |
pMB->mvs[0] = fData->currentMV[0]; |
1726 |
const IMAGE * const b_refHV) |
const IMAGE * const b_refHV) |
1727 |
{ |
{ |
1728 |
uint32_t i, j; |
uint32_t i, j; |
1729 |
int32_t best_sad, skip_sad; |
int32_t best_sad; |
1730 |
|
uint32_t skip_sad; |
1731 |
int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0; |
int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0; |
1732 |
static const VECTOR zeroMV={0,0}; |
static const VECTOR zeroMV={0,0}; |
1733 |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
1736 |
|
|
1737 |
const int32_t TRB = time_pp - time_bp; |
const int32_t TRB = time_pp - time_bp; |
1738 |
const int32_t TRD = time_pp; |
const int32_t TRD = time_pp; |
|
uint8_t * qimage; |
|
1739 |
|
|
1740 |
// some pre-inintialized data for the rest of the search |
// some pre-inintialized data for the rest of the search |
1741 |
|
|
1743 |
int32_t iMinSAD; |
int32_t iMinSAD; |
1744 |
VECTOR currentMV[3]; |
VECTOR currentMV[3]; |
1745 |
VECTOR currentQMV[3]; |
VECTOR currentQMV[3]; |
1746 |
|
memset(&Data, 0, sizeof(SearchData)); |
1747 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
1748 |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
1749 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
1750 |
Data.lambda16 = lambda_vec16[frame->quant]; |
Data.lambda16 = lambda_vec16[frame->quant]; |
1751 |
|
Data.chroma = frame->quant; |
1752 |
Data.qpel = pParam->m_quarterpel; |
Data.qpel = pParam->m_quarterpel; |
1753 |
Data.rounding = 0; |
Data.rounding = 0; |
1754 |
|
|
1755 |
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose) |
|
return; // allocate some mem for qpel interpolated blocks |
|
|
// somehow this is dirty since I think we shouldn't use malloc outside |
|
|
// encoder_create() - so please fix me! |
|
|
Data.RefQ = qimage; |
|
|
|
|
1756 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
1757 |
for (j = 0; j < pParam->mb_height; j++) { |
for (j = 0; j < pParam->mb_height; j++) { |
1758 |
|
|
1804 |
MODE_BACKWARD, &Data); |
MODE_BACKWARD, &Data); |
1805 |
|
|
1806 |
// interpolate search comes last, because it uses data from forward and backward as prediction |
// interpolate search comes last, because it uses data from forward and backward as prediction |
|
|
|
1807 |
SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
1808 |
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
1809 |
&frame->image, |
&frame->image, |
1815 |
pMB, &best_sad, |
pMB, &best_sad, |
1816 |
&Data); |
&Data); |
1817 |
|
|
1818 |
|
// final skip decision |
1819 |
|
if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP*2) |
1820 |
|
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
1821 |
|
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
1822 |
|
|
1823 |
switch (pMB->mode) { |
switch (pMB->mode) { |
1824 |
case MODE_FORWARD: |
case MODE_FORWARD: |
1825 |
f_count++; |
f_count++; |
1826 |
if (pParam->m_quarterpel) f_predMV = pMB->qmvs[0]; |
if (Data.qpel) f_predMV = pMB->qmvs[0]; |
1827 |
else f_predMV = pMB->mvs[0]; |
else f_predMV = pMB->mvs[0]; |
1828 |
break; |
break; |
1829 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
1830 |
b_count++; |
b_count++; |
1831 |
if (pParam->m_quarterpel) b_predMV = pMB->b_qmvs[0]; |
if (Data.qpel) b_predMV = pMB->b_qmvs[0]; |
1832 |
else b_predMV = pMB->b_mvs[0]; |
else b_predMV = pMB->b_mvs[0]; |
1833 |
break; |
break; |
1834 |
case MODE_INTERPOLATE: |
case MODE_INTERPOLATE: |
1835 |
i_count++; |
i_count++; |
1836 |
if (pParam->m_quarterpel) { |
if (Data.qpel) { |
1837 |
f_predMV = pMB->qmvs[0]; |
f_predMV = pMB->qmvs[0]; |
1838 |
b_predMV = pMB->b_qmvs[0]; |
b_predMV = pMB->b_qmvs[0]; |
1839 |
} else { |
} else { |
1844 |
case MODE_DIRECT: |
case MODE_DIRECT: |
1845 |
case MODE_DIRECT_NO4V: |
case MODE_DIRECT_NO4V: |
1846 |
d_count++; |
d_count++; |
|
break; |
|
1847 |
default: |
default: |
1848 |
break; |
break; |
1849 |
} |
} |
1850 |
} |
} |
1851 |
} |
} |
|
free(qimage); |
|
1852 |
} |
} |
1853 |
|
|
1854 |
/* Hinted ME starts here */ |
static __inline void |
|
|
|
|
static void |
|
|
SearchPhinted ( const IMAGE * 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 MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
int inter4v, |
|
|
MACROBLOCK * const pMB, |
|
|
SearchData * const Data) |
|
|
{ |
|
|
|
|
|
int i, t; |
|
|
MainSearchFunc * MainSearchPtr; |
|
|
|
|
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
|
|
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
|
|
|
|
|
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
|
|
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
|
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
|
|
|
|
Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
|
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
|
Data->qpel_precision = 0; |
|
|
|
|
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
|
|
Data->min_dx = EVEN(Data->min_dx); |
|
|
Data->max_dx = EVEN(Data->max_dx); |
|
|
Data->min_dy = EVEN(Data->min_dy); |
|
|
Data->max_dy = EVEN(Data->max_dy); |
|
|
} |
|
|
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
|
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
|
|
|
|
for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
|
|
|
|
|
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
|
|
|
|
|
if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
|
|
else CheckCandidate = CheckCandidate16no4v; |
|
|
|
|
|
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
|
|
pMB->mvs[0].y = EVEN(pMB->mvs[0].y); |
|
|
if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed |
|
|
if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx; |
|
|
if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy; |
|
|
if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy; |
|
|
|
|
|
(*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data); |
|
|
|
|
|
if (pMB->mode == MODE_INTER4V) |
|
|
for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search |
|
|
pMB->mvs[i].x = EVEN(pMB->mvs[i].x); |
|
|
pMB->mvs[i].y = EVEN(pMB->mvs[i].y); |
|
|
if (!(make_mask(pMB->mvs, i))) |
|
|
(*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data); |
|
|
} |
|
|
|
|
|
if (MotionFlags & PMV_USESQUARES16) |
|
|
MainSearchPtr = SquareSearch; |
|
|
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
|
|
MainSearchPtr = AdvDiamondSearch; |
|
|
else MainSearchPtr = DiamondSearch; |
|
|
|
|
|
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
|
|
|
|
|
if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data); |
|
|
|
|
|
for(i = 0; i < 5; i++) { |
|
|
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
|
|
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
|
|
} |
|
|
|
|
|
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
|
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
|
|
pParam->width, pParam->height, Data->iFcode, 0); |
|
|
Data->qpel_precision = 1; |
|
|
SubpelRefine(Data); |
|
|
} |
|
|
|
|
|
if (inter4v) { |
|
|
SearchData Data8; |
|
|
Data8.iFcode = Data->iFcode; |
|
|
Data8.lambda8 = Data->lambda8; |
|
|
Data8.iEdgedWidth = Data->iEdgedWidth; |
|
|
Data8.RefQ = Data->RefQ; |
|
|
Data8.qpel = Data->qpel; |
|
|
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
|
|
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
|
|
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
|
|
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
|
|
|
|
|
if (Data->chroma) { |
|
|
int sumx, sumy, dx, dy; |
|
|
|
|
|
if(pParam->m_quarterpel) { |
|
|
sumx= pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2; |
|
|
sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2; |
|
|
} else { |
|
|
sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
|
|
sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
|
|
} |
|
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
|
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
|
|
|
|
|
Data->iMinSAD[1] += ChromaSAD(dx, dy, Data); |
|
|
} |
|
|
} |
|
|
|
|
|
if (!(inter4v) || |
|
|
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + |
|
|
Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
|
|
// INTER MODE |
|
|
pMB->mode = MODE_INTER; |
|
|
pMB->mvs[0] = pMB->mvs[1] |
|
|
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
|
|
|
|
|
pMB->qmvs[0] = pMB->qmvs[1] |
|
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
|
|
|
|
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
|
|
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
|
|
|
|
|
if(pParam->m_quarterpel) { |
|
|
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
|
|
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; |
|
|
} else { |
|
|
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
|
|
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
|
|
} |
|
|
} else { |
|
|
// INTER4V MODE; all other things are already set in Search8 |
|
|
pMB->mode = MODE_INTER4V; |
|
|
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] |
|
|
+ Data->iMinSAD[4] + IMV16X16 * iQuant; |
|
|
} |
|
|
|
|
|
} |
|
|
|
|
|
void |
|
|
MotionEstimationHinted( MBParam * const pParam, |
|
|
FRAMEINFO * const current, |
|
|
FRAMEINFO * const reference, |
|
|
const IMAGE * const pRefH, |
|
|
const IMAGE * const pRefV, |
|
|
const IMAGE * const pRefHV) |
|
|
{ |
|
|
MACROBLOCK *const pMBs = current->mbs; |
|
|
const IMAGE *const pCurrent = ¤t->image; |
|
|
const IMAGE *const pRef = &reference->image; |
|
|
|
|
|
uint32_t x, y; |
|
|
uint8_t * qimage; |
|
|
int32_t temp[5], quant = current->quant; |
|
|
int32_t iMinSAD[5]; |
|
|
VECTOR currentMV[5], currentQMV[5]; |
|
|
SearchData Data; |
|
|
Data.iEdgedWidth = pParam->edged_width; |
|
|
Data.currentMV = currentMV; |
|
|
Data.currentQMV = currentQMV; |
|
|
Data.iMinSAD = iMinSAD; |
|
|
Data.temp = temp; |
|
|
Data.iFcode = current->fcode; |
|
|
Data.rounding = pParam->m_rounding_type; |
|
|
Data.qpel = pParam->m_quarterpel; |
|
|
Data.chroma = current->global_flags & XVID_ME_COLOUR; |
|
|
|
|
|
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
|
|
return; // allocate some mem for qpel interpolated blocks |
|
|
// somehow this is dirty since I think we shouldn't use malloc outside |
|
|
// encoder_create() - so please fix me! |
|
|
|
|
|
Data.RefQ = qimage; |
|
|
|
|
|
if (sadInit) (*sadInit) (); |
|
|
|
|
|
for (y = 0; y < pParam->mb_height; y++) { |
|
|
for (x = 0; x < pParam->mb_width; x++) { |
|
|
|
|
|
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
|
|
|
|
|
//intra mode is copied from the first pass. At least for the time being |
|
|
if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue; |
|
|
|
|
|
if (!(current->global_flags & XVID_LUMIMASKING)) { |
|
|
pMB->dquant = NO_CHANGE; |
|
|
pMB->quant = current->quant; } |
|
|
else { |
|
|
if (pMB->dquant != NO_CHANGE) { |
|
|
quant += DQtab[pMB->dquant]; |
|
|
if (quant > 31) quant = 31; |
|
|
else if (quant < 1) quant = 1; |
|
|
} |
|
|
pMB->quant = quant; |
|
|
} |
|
|
|
|
|
SearchPhinted(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
|
|
y, current->motion_flags, pMB->quant, |
|
|
pParam, pMBs, current->global_flags & XVID_INTER4V, pMB, |
|
|
&Data); |
|
|
|
|
|
} |
|
|
} |
|
|
free(qimage); |
|
|
} |
|
|
|
|
|
static __inline int |
|
1855 |
MEanalyzeMB ( const uint8_t * const pRef, |
MEanalyzeMB ( const uint8_t * const pRef, |
1856 |
const uint8_t * const pCur, |
const uint8_t * const pCur, |
1857 |
const int x, |
const int x, |
1858 |
const int y, |
const int y, |
1859 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1860 |
const MACROBLOCK * const pMBs, |
MACROBLOCK * const pMBs, |
|
MACROBLOCK * const pMB, |
|
1861 |
SearchData * const Data) |
SearchData * const Data) |
1862 |
{ |
{ |
1863 |
|
|
1864 |
int i = 255, mask; |
int i, mask; |
1865 |
VECTOR pmv[3]; |
VECTOR pmv[3]; |
1866 |
*(Data->iMinSAD) = MV_MAX_ERROR; |
MACROBLOCK * pMB = &pMBs[x + y * pParam->mb_width]; |
1867 |
|
|
1868 |
|
for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
1869 |
|
|
1870 |
//median is only used as prediction. it doesn't have to be real |
//median is only used as prediction. it doesn't have to be real |
1871 |
if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0; |
if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0; |
1872 |
else |
else |
1873 |
if (x == 1) //left macroblock does not have any vector now |
if (x == 1) //left macroblock does not have any vector now |
1874 |
Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median |
Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median |
1875 |
else if (y == 1) // top macroblock don't have it's vector |
else if (y == 1) // top macroblock doesn't have it's vector |
1876 |
Data->predMV = (pMB - 1)->mvs[0]; // left instead of median |
Data->predMV = (pMB - 1)->mvs[0]; // left instead of median |
1877 |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median |
1878 |
|
|
1879 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1880 |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, Data->rrv); |
1881 |
|
|
1882 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
1883 |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
1888 |
pmv[2].y = EVEN(Data->predMV.y); |
pmv[2].y = EVEN(Data->predMV.y); |
1889 |
pmv[0].x = pmv[0].y = 0; |
pmv[0].x = pmv[0].y = 0; |
1890 |
|
|
1891 |
(*CheckCandidate)(0, 0, 255, &i, Data); |
CheckCandidate32I(0, 0, 255, &i, Data); |
1892 |
|
|
1893 |
//early skip for 0,0 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP * 4) { |
|
if (*Data->iMinSAD < MAX_SAD00_FOR_SKIP * 4) { |
|
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
|
|
pMB->mode = MODE_NOT_CODED; |
|
|
return 0; |
|
|
} |
|
1894 |
|
|
1895 |
if (!(mask = make_mask(pmv, 1))) |
if (!(mask = make_mask(pmv, 1))) |
1896 |
(*CheckCandidate)(pmv[1].x, pmv[1].y, mask, &i, Data); |
CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data); |
1897 |
if (!(mask = make_mask(pmv, 2))) |
if (!(mask = make_mask(pmv, 2))) |
1898 |
(*CheckCandidate)(pmv[2].x, pmv[2].y, mask, &i, Data); |
CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data); |
1899 |
|
|
1900 |
if (*Data->iMinSAD > MAX_SAD00_FOR_SKIP * 4) // diamond only if needed |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP * 4) // diamond only if needed |
1901 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
1902 |
|
|
1903 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
for (i = 0; i < 4; i++) { |
1904 |
pMB->mode = MODE_INTER; |
MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1) * pParam->mb_width)]; |
1905 |
return *(Data->iMinSAD); |
MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i]; |
1906 |
|
MB->mode = MODE_INTER; |
1907 |
|
MB->sad16 = Data->iMinSAD[i+1]; |
1908 |
|
} |
1909 |
|
} |
1910 |
} |
} |
1911 |
|
|
1912 |
#define INTRA_THRESH 1350 |
#define INTRA_BIAS 2500 |
1913 |
#define INTER_THRESH 900 |
#define INTRA_THRESH 1500 |
1914 |
|
#define INTER_THRESH 1400 |
1915 |
|
|
1916 |
|
|
1917 |
int |
int |
1927 |
MACROBLOCK * const pMBs = Current->mbs; |
MACROBLOCK * const pMBs = Current->mbs; |
1928 |
const IMAGE * const pCurrent = &Current->image; |
const IMAGE * const pCurrent = &Current->image; |
1929 |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH; |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH; |
1930 |
|
const VECTOR zeroMV = {0,0}; |
1931 |
|
|
1932 |
VECTOR currentMV; |
int32_t iMinSAD[5], temp[5]; |
1933 |
int32_t iMinSAD; |
VECTOR currentMV[5]; |
1934 |
SearchData Data; |
SearchData Data; |
1935 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
1936 |
Data.currentMV = ¤tMV; |
Data.currentMV = currentMV; |
1937 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = iMinSAD; |
1938 |
Data.iFcode = Current->fcode; |
Data.iFcode = Current->fcode; |
1939 |
CheckCandidate = CheckCandidate16no4vI; |
Data.rrv = Current->global_flags & XVID_REDUCED; |
1940 |
|
Data.temp = temp; |
1941 |
|
CheckCandidate = CheckCandidate32I; |
1942 |
|
|
1943 |
if (intraCount < 10) // we're right after an I frame |
if (intraCount < 10) // we're right after an I frame |
1944 |
IntraThresh += 4 * (intraCount - 10) * (intraCount - 10); |
IntraThresh += 4 * (intraCount - 10) * (intraCount - 10); |
1946 |
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
1947 |
IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra; |
IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra; |
1948 |
|
|
1949 |
|
InterThresh += 400 * (1 - bCount); |
1950 |
InterThresh += 300 * (1 - bCount); |
if (InterThresh < 300) InterThresh = 300; |
|
if (InterThresh < 200) InterThresh = 200; |
|
1951 |
|
|
1952 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
1953 |
|
|
1954 |
for (y = 1; y < pParam->mb_height-1; y++) { |
for (y = 1; y < pParam->mb_height-1; y+=2) { |
1955 |
for (x = 1; x < pParam->mb_width-1; x++) { |
for (x = 1; x < pParam->mb_width-1; x+=2) { |
1956 |
int sad, dev; |
int i; |
1957 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
|
1958 |
|
if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV; |
1959 |
sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, |
|
1960 |
pParam, pMBs, pMB, &Data); |
MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data); |
1961 |
|
|
1962 |
if (sad > IntraThresh) { |
for (i = 0; i < 4; i++) { |
1963 |
dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
int dev; |
1964 |
|
MACROBLOCK *pMB = &pMBs[x+(i&1) + y+(i>>1) * pParam->mb_width]; |
1965 |
|
if (pMB->sad16 > IntraThresh) { |
1966 |
|
dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1))* pParam->edged_width) * 16, |
1967 |
pParam->edged_width); |
pParam->edged_width); |
1968 |
if (dev + IntraThresh < sad) { |
if (dev + IntraThresh < pMB->sad16) { |
1969 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
1970 |
if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame |
if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP; |
1971 |
|
} |
1972 |
} |
} |
1973 |
|
sSAD += pMB->sad16; |
1974 |
} |
} |
|
sSAD += sad; |
|
1975 |
} |
} |
1976 |
} |
} |
1977 |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
1978 |
if (sSAD > InterThresh ) return 1; //P frame |
if (sSAD > IntraThresh + INTRA_BIAS ) return I_VOP; |
1979 |
|
if (sSAD > InterThresh ) return P_VOP; |
1980 |
emms(); |
emms(); |
1981 |
return 0; // B frame |
return B_VOP; |
|
|
|
|
} |
|
|
|
|
|
int |
|
|
FindFcode( const MBParam * const pParam, |
|
|
const FRAMEINFO * const current) |
|
|
{ |
|
|
uint32_t x, y; |
|
|
int max = 0, min = 0, i; |
|
1982 |
|
|
|
for (y = 0; y < pParam->mb_height; y++) { |
|
|
for (x = 0; x < pParam->mb_width; x++) { |
|
|
|
|
|
MACROBLOCK *pMB = ¤t->mbs[x + y * pParam->mb_width]; |
|
|
for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) { |
|
|
if (pMB->mvs[i].x > max) max = pMB->mvs[i].x; |
|
|
if (pMB->mvs[i].y > max) max = pMB->mvs[i].y; |
|
|
|
|
|
if (pMB->mvs[i].x < min) min = pMB->mvs[i].x; |
|
|
if (pMB->mvs[i].y < min) min = pMB->mvs[i].y; |
|
|
} |
|
|
} |
|
|
} |
|
|
|
|
|
min = -min; |
|
|
max += 1; |
|
|
if (min > max) max = min; |
|
|
if (pParam->m_quarterpel) max *= 2; |
|
|
|
|
|
for (i = 1; (max > 32 << (i - 1)); i++); |
|
|
return i; |
|
1983 |
} |
} |
1984 |
|
|
1985 |
static void |
static void |
2046 |
max_x = gmc.x + step; |
max_x = gmc.x + step; |
2047 |
min_y = gmc.y - step; |
min_y = gmc.y - step; |
2048 |
max_y = gmc.y + step; |
max_y = gmc.y + step; |
|
|
|
2049 |
} |
} |
2050 |
|
|
2051 |
if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10) |
if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10) |
2052 |
gmc.x = gmc.y = 0; //no camara pan, no GMC |
gmc.x = gmc.y = 0; //no camara pan, no GMC |
2053 |
|
|
2054 |
// step2: let's refine camera panning using gradiend-descent approach. |
// step2: let's refine camera panning using gradiend-descent approach |
2055 |
// TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond) |
// TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond) |
2056 |
bestcount = 0; |
bestcount = 0; |
2057 |
CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam); |
CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam); |