105 |
return bits; |
return bits; |
106 |
} |
} |
107 |
|
|
108 |
static int32_t ChromaSAD2(int fx, int fy, int bx, int by, const SearchData * const data) |
static int32_t ChromaSAD2(const int fx, const int fy, const int bx, const int by, |
109 |
|
const SearchData * const data) |
110 |
{ |
{ |
111 |
int sad; |
int sad; |
112 |
const uint32_t stride = data->iEdgedWidth/2; |
const uint32_t stride = data->iEdgedWidth/2; |
114 |
* f_refv = data->RefQ + 8, |
* f_refv = data->RefQ + 8, |
115 |
* b_refu = data->RefQ + 16, |
* b_refu = data->RefQ + 16, |
116 |
* b_refv = data->RefQ + 24; |
* b_refv = data->RefQ + 24; |
117 |
|
int offset = (fx>>1) + (fy>>1)*stride; |
118 |
|
|
119 |
switch (((fx & 1) << 1) | (fy & 1)) { |
switch (((fx & 1) << 1) | (fy & 1)) { |
120 |
case 0: |
case 0: |
121 |
fx = fx / 2; fy = fy / 2; |
f_refu = (uint8_t*)data->RefP[4] + offset; |
122 |
f_refu = (uint8_t*)data->RefCU + fy * stride + fx, stride; |
f_refv = (uint8_t*)data->RefP[5] + offset; |
|
f_refv = (uint8_t*)data->RefCV + fy * stride + fx, stride; |
|
123 |
break; |
break; |
124 |
case 1: |
case 1: |
125 |
fx = fx / 2; fy = (fy - 1) / 2; |
interpolate8x8_halfpel_v(f_refu, data->RefP[4] + offset, stride, data->rounding); |
126 |
interpolate8x8_halfpel_v(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
interpolate8x8_halfpel_v(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_v(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
|
127 |
break; |
break; |
128 |
case 2: |
case 2: |
129 |
fx = (fx - 1) / 2; fy = fy / 2; |
interpolate8x8_halfpel_h(f_refu, data->RefP[4] + offset, stride, data->rounding); |
130 |
interpolate8x8_halfpel_h(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
interpolate8x8_halfpel_h(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_h(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
|
131 |
break; |
break; |
132 |
default: |
default: |
133 |
fx = (fx - 1) / 2; fy = (fy - 1) / 2; |
interpolate8x8_halfpel_hv(f_refu, data->RefP[4] + offset, stride, data->rounding); |
134 |
interpolate8x8_halfpel_hv(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
interpolate8x8_halfpel_hv(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_hv(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
|
135 |
break; |
break; |
136 |
} |
} |
137 |
|
|
138 |
|
offset = (bx>>1) + (by>>1)*stride; |
139 |
switch (((bx & 1) << 1) | (by & 1)) { |
switch (((bx & 1) << 1) | (by & 1)) { |
140 |
case 0: |
case 0: |
141 |
bx = bx / 2; by = by / 2; |
b_refu = (uint8_t*)data->b_RefP[4] + offset; |
142 |
b_refu = (uint8_t*)data->b_RefCU + by * stride + bx, stride; |
b_refv = (uint8_t*)data->b_RefP[5] + offset; |
|
b_refv = (uint8_t*)data->b_RefCV + by * stride + bx, stride; |
|
143 |
break; |
break; |
144 |
case 1: |
case 1: |
145 |
bx = bx / 2; by = (by - 1) / 2; |
interpolate8x8_halfpel_v(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
146 |
interpolate8x8_halfpel_v(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
interpolate8x8_halfpel_v(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_v(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
|
147 |
break; |
break; |
148 |
case 2: |
case 2: |
149 |
bx = (bx - 1) / 2; by = by / 2; |
interpolate8x8_halfpel_h(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
150 |
interpolate8x8_halfpel_h(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
interpolate8x8_halfpel_h(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_h(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
|
151 |
break; |
break; |
152 |
default: |
default: |
153 |
bx = (bx - 1) / 2; by = (by - 1) / 2; |
interpolate8x8_halfpel_hv(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
154 |
interpolate8x8_halfpel_hv(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
interpolate8x8_halfpel_hv(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_hv(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
|
155 |
break; |
break; |
156 |
} |
} |
157 |
|
|
161 |
return sad; |
return sad; |
162 |
} |
} |
163 |
|
|
|
|
|
164 |
static int32_t |
static int32_t |
165 |
ChromaSAD(int dx, int dy, const SearchData * const data) |
ChromaSAD(const int dx, const int dy, const SearchData * const data) |
166 |
{ |
{ |
167 |
int sad; |
int sad; |
168 |
const uint32_t stride = data->iEdgedWidth/2; |
const uint32_t stride = data->iEdgedWidth/2; |
169 |
|
int offset = (dx>>1) + (dy>>1)*stride; |
170 |
|
|
171 |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently |
172 |
data->temp[5] = dx; data->temp[6] = dy; // backup |
data->temp[5] = dx; data->temp[6] = dy; // backup |
173 |
|
|
174 |
switch (((dx & 1) << 1) | (dy & 1)) { |
switch (((dx & 1) << 1) | (dy & 1)) { |
175 |
case 0: |
case 0: |
176 |
dx = dx / 2; dy = dy / 2; |
sad = sad8(data->CurU, data->RefP[4] + offset, stride); |
177 |
sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride); |
sad += sad8(data->CurV, data->RefP[5] + offset, stride); |
|
sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride); |
|
178 |
break; |
break; |
179 |
case 1: |
case 1: |
180 |
dx = dx / 2; dy = (dy - 1) / 2; |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + stride, stride); |
181 |
sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride); |
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + stride, stride); |
|
sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride); |
|
182 |
break; |
break; |
183 |
case 2: |
case 2: |
184 |
dx = (dx - 1) / 2; dy = dy / 2; |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + 1, stride); |
185 |
sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride); |
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + 1, stride); |
|
sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride); |
|
186 |
break; |
break; |
187 |
default: |
default: |
188 |
dx = (dx - 1) / 2; dy = (dy - 1) / 2; |
interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding); |
|
interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding); |
|
189 |
sad = sad8(data->CurU, data->RefQ, stride); |
sad = sad8(data->CurU, data->RefQ, stride); |
190 |
|
|
191 |
interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding); |
interpolate8x8_halfpel_hv(data->RefQ, data->RefP[5] + offset, stride, data->rounding); |
192 |
sad += sad8(data->CurV, data->RefQ, stride); |
sad += sad8(data->CurV, data->RefQ, stride); |
193 |
break; |
break; |
194 |
} |
} |
200 |
GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data) |
GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data) |
201 |
{ |
{ |
202 |
// dir : 0 = forward, 1 = backward |
// dir : 0 = forward, 1 = backward |
203 |
switch ( (dir << 2) | ((x&1)<<1) | (y&1) ) { |
const uint8_t* const *direction = ( dir == 0 ? data->RefP : data->b_RefP ); |
204 |
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
const int picture = ((x&1)<<1) | (y&1); |
205 |
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
const int offset = (x>>1) + (y>>1)*data->iEdgedWidth; |
206 |
case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
return direction[picture] + offset; |
|
case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth); |
|
|
case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
|
|
default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
} |
|
207 |
} |
} |
208 |
|
|
209 |
// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate |
// this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate |
210 |
static __inline const uint8_t * |
static __inline const uint8_t * |
211 |
GetReference(const int x, const int y, const SearchData * const data) |
GetReference(const int x, const int y, const SearchData * const data) |
212 |
{ |
{ |
213 |
switch ( ((x&1)<<1) | (y&1) ) { |
const int picture = ((x&1)<<1) | (y&1); |
214 |
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
const int offset = (x>>1) + (y>>1)*data->iEdgedWidth; |
215 |
case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
return data->RefP[picture] + offset; |
|
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
default : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); //case 2 |
|
|
} |
|
216 |
} |
} |
217 |
|
|
218 |
static uint8_t * |
static uint8_t * |
465 |
if ( (x > data->max_dx) || (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
466 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
467 |
|
|
468 |
sad = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
sad = sad32v_c(data->Cur, data->RefP[0] + x/2 + (y/2)*(data->iEdgedWidth), |
469 |
data->iEdgedWidth, data->temp+1); |
data->iEdgedWidth, data->temp+1); |
470 |
|
|
471 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
685 |
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
686 |
|
|
687 |
//chroma U |
//chroma U |
688 |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCU, 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
689 |
transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2); |
transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2); |
690 |
fdct(in); |
fdct(in); |
691 |
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
697 |
|
|
698 |
if (bits < data->iMinSAD[0]) { |
if (bits < data->iMinSAD[0]) { |
699 |
//chroma V |
//chroma V |
700 |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCV, 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
701 |
transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2); |
transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2); |
702 |
fdct(in); |
fdct(in); |
703 |
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
1265 |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
1266 |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
1267 |
|
|
1268 |
Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16*i; |
1269 |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16*i; |
1270 |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16*i; |
1271 |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16*i; |
1272 |
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
1273 |
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
1274 |
|
|
1275 |
Data->lambda16 = lambda_vec16[iQuant]; |
Data->lambda16 = lambda_vec16[iQuant]; |
1276 |
Data->lambda8 = lambda_vec8[iQuant]; |
Data->lambda8 = lambda_vec8[iQuant]; |
1472 |
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10; |
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10; |
1473 |
|
|
1474 |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8|PMV_QUARTERPELREFINE8)) { |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8|PMV_QUARTERPELREFINE8)) { |
|
if (Data->rrv) i = 2; else i = 1; |
|
1475 |
|
|
1476 |
Data->Ref = OldData->Ref + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
if (Data->rrv) i = 16; else i = 8; |
1477 |
Data->RefH = OldData->RefH + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
|
1478 |
Data->RefV = OldData->RefV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->RefP[0] = OldData->RefP[0] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1479 |
Data->RefHV = OldData->RefHV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->RefP[1] = OldData->RefP[1] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1480 |
|
Data->RefP[2] = OldData->RefP[2] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1481 |
|
Data->RefP[3] = OldData->RefP[3] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1482 |
|
|
1483 |
Data->Cur = OldData->Cur + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->Cur = OldData->Cur + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1484 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1485 |
|
|
1486 |
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, |
1614 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1615 |
Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; // reset chroma-sad cache |
Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; // reset chroma-sad cache |
1616 |
|
|
1617 |
Data->Ref = pRef->y + (x + y * Data->iEdgedWidth) * 16; |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
1618 |
Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16; |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16; |
1619 |
Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16; |
Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16; |
1620 |
Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16; |
Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
1621 |
Data->RefCU = pRef->u + (x + y * Data->iEdgedWidth/2) * 8; |
Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
1622 |
Data->RefCV = pRef->v + (x + y * Data->iEdgedWidth/2) * 8; |
Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
1623 |
|
|
1624 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1625 |
|
|
1757 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1758 |
|
|
1759 |
*Data->iMinSAD = 256*4096; |
*Data->iMinSAD = 256*4096; |
1760 |
Data->Ref = f_Ref->y + k; |
Data->RefP[0] = f_Ref->y + k; |
1761 |
Data->RefH = f_RefH + k; |
Data->RefP[2] = f_RefH + k; |
1762 |
Data->RefV = f_RefV + k; |
Data->RefP[1] = f_RefV + k; |
1763 |
Data->RefHV = f_RefHV + k; |
Data->RefP[3] = f_RefHV + k; |
1764 |
Data->bRef = b_Ref->y + k; |
Data->b_RefP[0] = b_Ref->y + k; |
1765 |
Data->bRefH = b_RefH + k; |
Data->b_RefP[2] = b_RefH + k; |
1766 |
Data->bRefV = b_RefV + k; |
Data->b_RefP[1] = b_RefV + k; |
1767 |
Data->bRefHV = b_RefHV + k; |
Data->b_RefP[3] = b_RefHV + k; |
1768 |
Data->RefCU = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
Data->RefP[4] = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
1769 |
Data->RefCV = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
Data->RefP[5] = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
1770 |
Data->b_RefCU = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
Data->b_RefP[4] = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
1771 |
Data->b_RefCV = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
Data->b_RefP[5] = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
1772 |
|
|
1773 |
k = Data->qpel ? 4 : 2; |
k = Data->qpel ? 4 : 2; |
1774 |
Data->max_dx = k * (pParam->width - x * 16); |
Data->max_dx = k * (pParam->width - x * 16); |
1899 |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1900 |
|
|
1901 |
i = (x + y * fData->iEdgedWidth) * 16; |
i = (x + y * fData->iEdgedWidth) * 16; |
|
bData.bRef = fData->Ref = f_Ref->y + i; |
|
|
bData.bRefH = fData->RefH = f_RefH + i; |
|
|
bData.bRefV = fData->RefV = f_RefV + i; |
|
|
bData.bRefHV = fData->RefHV = f_RefHV + i; |
|
|
bData.Ref = fData->bRef = b_Ref->y + i; |
|
|
bData.RefH = fData->bRefH = b_RefH + i; |
|
|
bData.RefV = fData->bRefV = b_RefV + i; |
|
|
bData.RefHV = fData->bRefHV = b_RefHV + i; |
|
|
bData.b_RefCU = fData->RefCU = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.b_RefCV = fData->RefCV = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.RefCU = fData->b_RefCU = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.RefCV = fData->b_RefCV = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
1902 |
|
|
1903 |
|
bData.b_RefP[0] = fData->RefP[0] = f_Ref->y + i; |
1904 |
|
bData.b_RefP[2] = fData->RefP[2] = f_RefH + i; |
1905 |
|
bData.b_RefP[1] = fData->RefP[1] = f_RefV + i; |
1906 |
|
bData.b_RefP[3] = fData->RefP[3] = f_RefHV + i; |
1907 |
|
bData.RefP[0] = fData->b_RefP[0] = b_Ref->y + i; |
1908 |
|
bData.RefP[2] = fData->b_RefP[2] = b_RefH + i; |
1909 |
|
bData.RefP[1] = fData->b_RefP[1] = b_RefV + i; |
1910 |
|
bData.RefP[3] = fData->b_RefP[3] = b_RefHV + i; |
1911 |
|
bData.b_RefP[4] = fData->RefP[4] = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
1912 |
|
bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
1913 |
|
bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
1914 |
|
bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
1915 |
|
|
1916 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
1917 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
2162 |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, 0); |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, 0); |
2163 |
|
|
2164 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
2165 |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16; |
2166 |
|
|
2167 |
pmv[1].x = EVEN(pMB->mvs[0].x); |
pmv[1].x = EVEN(pMB->mvs[0].x); |
2168 |
pmv[1].y = EVEN(pMB->mvs[0].y); |
pmv[1].y = EVEN(pMB->mvs[0].y); |
2171 |
pmv[0].x = pmv[0].y = 0; |
pmv[0].x = pmv[0].y = 0; |
2172 |
|
|
2173 |
CheckCandidate32I(0, 0, 255, &i, Data); |
CheckCandidate32I(0, 0, 255, &i, Data); |
|
Data->iMinSAD[1] -= 50; |
|
|
Data->iMinSAD[2] -= 50; |
|
|
Data->iMinSAD[3] -= 50; |
|
|
Data->iMinSAD[4] -= 50; |
|
2174 |
|
|
2175 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { |
2176 |
|
|
2192 |
} |
} |
2193 |
|
|
2194 |
#define INTRA_THRESH 2400 |
#define INTRA_THRESH 2400 |
2195 |
#define INTER_THRESH 1100 |
#define INTER_THRESH 1300 |
2196 |
|
|
2197 |
int |
int |
2198 |
MEanalysis( const IMAGE * const pRef, |
MEanalysis( const IMAGE * const pRef, |
2267 |
sSAD /= blocks; |
sSAD /= blocks; |
2268 |
s = (10*s) / blocks; |
s = (10*s) / blocks; |
2269 |
|
|
2270 |
if (s > 5) sSAD += (s - 4) * (180 - 2*b_thresh); //static block - looks bad when in bframe... |
if (s > 4) sSAD += (s - 3) * (300 - 2*b_thresh); //static block - looks bad when in bframe... |
2271 |
|
|
2272 |
if (sSAD > InterThresh ) return P_VOP; |
if (sSAD > InterThresh ) return P_VOP; |
2273 |
emms(); |
emms(); |
2549 |
Data8->currentMV = Data->currentMV + i + 1; |
Data8->currentMV = Data->currentMV + i + 1; |
2550 |
Data8->currentQMV = Data->currentQMV + i + 1; |
Data8->currentQMV = Data->currentQMV + i + 1; |
2551 |
Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2552 |
Data8->Ref = Data->Ref + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[0] = Data->RefP[0] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2553 |
Data8->RefH = Data->RefH + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2554 |
Data8->RefV = Data->RefV + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2555 |
Data8->RefHV = Data->RefHV + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2556 |
|
|
2557 |
if(Data->qpel) { |
if(Data->qpel) { |
2558 |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
2645 |
sumy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
sumy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
2646 |
|
|
2647 |
//chroma U |
//chroma U |
2648 |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefCU, 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
2649 |
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
2650 |
fdct(in); |
fdct(in); |
2651 |
if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16); |
if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16); |
2657 |
|
|
2658 |
if (bits < *Data->iMinSAD) { // still possible |
if (bits < *Data->iMinSAD) { // still possible |
2659 |
//chroma V |
//chroma V |
2660 |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefCV, 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
2661 |
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
2662 |
fdct(in); |
fdct(in); |
2663 |
if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16); |
if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16); |
2679 |
CountMBBitsIntra(const SearchData * const Data) |
CountMBBitsIntra(const SearchData * const Data) |
2680 |
{ |
{ |
2681 |
int bits = 1; //this one is ac/dc prediction flag. always 1. |
int bits = 1; //this one is ac/dc prediction flag. always 1. |
2682 |
int cbp = 0, i, t, dc = 0, b_dc = 1024; |
int cbp = 0, i, t, dc = 1024, b_dc; |
2683 |
const uint32_t iQuant = Data->lambda16; |
const uint32_t iQuant = Data->lambda16; |
2684 |
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
2685 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, 1);; |
2686 |
|
|
2687 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, 1); |
|
|
|
|
2688 |
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2689 |
transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth); |
transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth); |
2690 |
fdct(in); |
fdct(in); |
2691 |
b_dc = dc; |
b_dc = in[0]; |
2692 |
dc = in[0]; |
in[0] -= dc; |
2693 |
in[0] -= b_dc; |
dc = b_dc; |
2694 |
if (Data->lambda8 == 0) quant_intra_c(coeff, in, iQuant, iDcScaler); |
if (Data->lambda8 == 0) quant_intra(coeff, in, iQuant, iDcScaler); |
2695 |
else quant4_intra_c(coeff, in, iQuant, iDcScaler); |
else quant4_intra(coeff, in, iQuant, iDcScaler); |
|
|
|
|
b_dc = dc; |
|
|
dc = coeff[0]; |
|
|
if (i != 0) coeff[0] -= b_dc; |
|
2696 |
|
|
2697 |
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcy_tab[coeff[0] + 255].len;; |
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcy_tab[coeff[0] + 255].len;; |
2698 |
Data->temp[i] = t; |
Data->temp[i] = t; |
2701 |
} |
} |
2702 |
|
|
2703 |
if (bits < Data->iMinSAD[0]) { // INTRA still looks good, let's add chroma |
if (bits < Data->iMinSAD[0]) { // INTRA still looks good, let's add chroma |
2704 |
uint32_t iDcScaler = get_dc_scaler(iQuant, 0); |
iDcScaler = get_dc_scaler(iQuant, 0); |
2705 |
//chroma U |
//chroma U |
2706 |
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
2707 |
fdct(in); |
fdct(in); |
2713 |
if (t != 0) cbp |= 1 << (5 - 4); |
if (t != 0) cbp |= 1 << (5 - 4); |
2714 |
|
|
2715 |
if (bits < Data->iMinSAD[0]) { |
if (bits < Data->iMinSAD[0]) { |
|
iDcScaler = get_dc_scaler(iQuant, 1); |
|
2716 |
//chroma V |
//chroma V |
2717 |
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
2718 |
fdct(in); |
fdct(in); |