42 |
#include "../quant/quant.h" |
#include "../quant/quant.h" |
43 |
#include "../encoder.h" |
#include "../encoder.h" |
44 |
|
|
|
#include "../image/reduced.h" |
|
45 |
#include "../quant/quant_matrix.h" |
#include "../quant/quant_matrix.h" |
46 |
|
|
47 |
MBFIELDTEST_PTR MBFieldTest; |
MBFIELDTEST_PTR MBFieldTest; |
90 |
|
|
91 |
/* Perform DCT */ |
/* Perform DCT */ |
92 |
start_timer(); |
start_timer(); |
93 |
fdct(&data[0 * 64]); |
fdct((short * const)&data[0 * 64]); |
94 |
fdct(&data[1 * 64]); |
fdct((short * const)&data[1 * 64]); |
95 |
fdct(&data[2 * 64]); |
fdct((short * const)&data[2 * 64]); |
96 |
fdct(&data[3 * 64]); |
fdct((short * const)&data[3 * 64]); |
97 |
fdct(&data[4 * 64]); |
fdct((short * const)&data[4 * 64]); |
98 |
fdct(&data[5 * 64]); |
fdct((short * const)&data[5 * 64]); |
99 |
stop_dct_timer(); |
stop_dct_timer(); |
100 |
} |
} |
101 |
|
|
105 |
const uint8_t cbp) |
const uint8_t cbp) |
106 |
{ |
{ |
107 |
start_timer(); |
start_timer(); |
108 |
if(cbp & (1 << (5 - 0))) idct(&data[0 * 64]); |
if(cbp & (1 << (5 - 0))) idct((short * const)&data[0 * 64]); |
109 |
if(cbp & (1 << (5 - 1))) idct(&data[1 * 64]); |
if(cbp & (1 << (5 - 1))) idct((short * const)&data[1 * 64]); |
110 |
if(cbp & (1 << (5 - 2))) idct(&data[2 * 64]); |
if(cbp & (1 << (5 - 2))) idct((short * const)&data[2 * 64]); |
111 |
if(cbp & (1 << (5 - 3))) idct(&data[3 * 64]); |
if(cbp & (1 << (5 - 3))) idct((short * const)&data[3 * 64]); |
112 |
if(cbp & (1 << (5 - 4))) idct(&data[4 * 64]); |
if(cbp & (1 << (5 - 4))) idct((short * const)&data[4 * 64]); |
113 |
if(cbp & (1 << (5 - 5))) idct(&data[5 * 64]); |
if(cbp & (1 << (5 - 5))) idct((short * const)&data[5 * 64]); |
114 |
stop_idct_timer(); |
stop_idct_timer(); |
115 |
} |
} |
116 |
|
|
182 |
int Q, |
int Q, |
183 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
184 |
const uint16_t * const QuantMatrix, |
const uint16_t * const QuantMatrix, |
185 |
int Non_Zero); |
int Non_Zero, |
186 |
|
int Sum); |
187 |
|
|
188 |
/* Quantize all blocks -- Inter mode */ |
/* Quantize all blocks -- Inter mode */ |
189 |
static __inline uint8_t |
static __inline uint8_t |
216 |
|
|
217 |
sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, pParam->mpeg_quant_matrices); |
sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, pParam->mpeg_quant_matrices); |
218 |
|
|
219 |
if(sum && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
if(sum && (pMB->quant > 2) && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
220 |
const uint16_t *matrix; |
const uint16_t *matrix; |
221 |
const static uint16_t h263matrix[] = |
const static uint16_t h263matrix[] = |
222 |
{ |
{ |
234 |
sum = dct_quantize_trellis_c(&qcoeff[i*64], &data[i*64], |
sum = dct_quantize_trellis_c(&qcoeff[i*64], &data[i*64], |
235 |
pMB->quant, &scan_tables[0][0], |
pMB->quant, &scan_tables[0][0], |
236 |
matrix, |
matrix, |
237 |
63); |
63, |
238 |
|
sum); |
239 |
} |
} |
240 |
stop_quant_timer(); |
stop_quant_timer(); |
241 |
|
|
309 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
310 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
311 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * 8; |
|
int32_t cst; |
|
|
int vop_reduced; |
|
312 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
313 |
const IMAGE * const pCurrent = &frame->image; |
const IMAGE * const pCurrent = &frame->image; |
|
transfer_operation_8to16_t * const functions[2] = |
|
|
{ |
|
|
(transfer_operation_8to16_t *)transfer_8to16copy, |
|
|
(transfer_operation_8to16_t *)filter_18x18_to_8x8 |
|
|
}; |
|
|
transfer_operation_8to16_t *transfer_op = NULL; |
|
|
|
|
|
vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED); |
|
314 |
|
|
315 |
/* Image pointers */ |
/* Image pointers */ |
316 |
pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride + (x_pos << (4+vop_reduced)); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
317 |
pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced)); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
318 |
pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced)); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
|
|
|
|
/* Block size */ |
|
|
cst = 8<<vop_reduced; |
|
|
|
|
|
/* Operation function */ |
|
|
transfer_op = functions[vop_reduced]; |
|
319 |
|
|
320 |
/* Do the transfer */ |
/* Do the transfer */ |
321 |
start_timer(); |
start_timer(); |
322 |
transfer_op(&data[0 * 64], pY_Cur, stride); |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
323 |
transfer_op(&data[1 * 64], pY_Cur + cst, stride); |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
324 |
transfer_op(&data[2 * 64], pY_Cur + next_block, stride); |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
325 |
transfer_op(&data[3 * 64], pY_Cur + next_block + cst, stride); |
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
326 |
transfer_op(&data[4 * 64], pU_Cur, stride2); |
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
327 |
transfer_op(&data[5 * 64], pV_Cur, stride2); |
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
328 |
stop_transfer_timer(); |
stop_transfer_timer(); |
329 |
} |
} |
330 |
|
|
342 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
343 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
344 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * 8; |
|
uint32_t cst; |
|
|
int vop_reduced; |
|
345 |
const IMAGE * const pCurrent = &frame->image; |
const IMAGE * const pCurrent = &frame->image; |
346 |
|
|
347 |
/* Array of function pointers, indexed by [vop_reduced<<1+add] */ |
/* Array of function pointers, indexed by [add] */ |
348 |
transfer_operation_16to8_t * const functions[4] = |
transfer_operation_16to8_t * const functions[2] = |
349 |
{ |
{ |
350 |
(transfer_operation_16to8_t*)transfer_16to8copy, |
(transfer_operation_16to8_t*)transfer_16to8copy, |
351 |
(transfer_operation_16to8_t*)transfer_16to8add, |
(transfer_operation_16to8_t*)transfer_16to8add, |
|
(transfer_operation_16to8_t*)copy_upsampled_8x8_16to8, |
|
|
(transfer_operation_16to8_t*)add_upsampled_8x8_16to8 |
|
352 |
}; |
}; |
353 |
|
|
354 |
transfer_operation_16to8_t *transfer_op = NULL; |
transfer_operation_16to8_t *transfer_op = NULL; |
355 |
|
|
356 |
|
/* Image pointers */ |
357 |
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
358 |
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
359 |
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
360 |
|
|
361 |
if (pMB->field_dct) { |
if (pMB->field_dct) { |
362 |
next_block = stride; |
next_block = stride; |
363 |
stride *= 2; |
stride *= 2; |
364 |
} |
} |
365 |
|
|
|
/* Makes this vars booleans */ |
|
|
vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED); |
|
|
|
|
|
/* Image pointers */ |
|
|
pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride + (x_pos << (4+vop_reduced)); |
|
|
pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced)); |
|
|
pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced)); |
|
|
|
|
|
/* Block size */ |
|
|
cst = 8<<vop_reduced; |
|
|
|
|
366 |
/* Operation function */ |
/* Operation function */ |
367 |
transfer_op = functions[(vop_reduced<<1) + add]; |
transfer_op = functions[add]; |
368 |
|
|
369 |
/* Do the operation */ |
/* Do the operation */ |
370 |
start_timer(); |
start_timer(); |
371 |
if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); |
if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); |
372 |
if (cbp&16) transfer_op(pY_Cur + cst, &data[1 * 64], stride); |
if (cbp&16) transfer_op(pY_Cur + 8, &data[1 * 64], stride); |
373 |
if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride); |
if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride); |
374 |
if (cbp& 4) transfer_op(pY_Cur + next_block + cst, &data[3 * 64], stride); |
if (cbp& 4) transfer_op(pY_Cur + next_block + 8, &data[3 * 64], stride); |
375 |
if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); |
if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); |
376 |
if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); |
if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); |
377 |
stop_transfer_timer(); |
stop_transfer_timer(); |
756 |
return -1; |
return -1; |
757 |
} |
} |
758 |
|
|
|
static int __inline |
|
|
Compute_Sum(const int16_t *C, int last) |
|
|
{ |
|
|
int sum = 0; |
|
|
|
|
|
while(last--) |
|
|
sum += abs(C[last]); |
|
|
|
|
|
return(sum); |
|
|
} |
|
|
|
|
759 |
/* this routine has been strippen of all debug code */ |
/* this routine has been strippen of all debug code */ |
760 |
static int |
static int |
761 |
dct_quantize_trellis_c(int16_t *const Out, |
dct_quantize_trellis_c(int16_t *const Out, |
763 |
int Q, |
int Q, |
764 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
765 |
const uint16_t * const QuantMatrix, |
const uint16_t * const QuantMatrix, |
766 |
int Non_Zero) |
int Non_Zero, |
767 |
|
int Sum) |
768 |
{ |
{ |
769 |
|
|
770 |
/* Note: We should search last non-zero coeffs on *real* DCT input coeffs |
/* Note: We should search last non-zero coeffs on *real* DCT input coeffs |
774 |
* helps. */ |
* helps. */ |
775 |
typedef struct { int16_t Run, Level; } NODE; |
typedef struct { int16_t Run, Level; } NODE; |
776 |
|
|
777 |
NODE Nodes[65], Last; |
NODE Nodes[65], Last = { 0, 0}; |
778 |
uint32_t Run_Costs0[64+1]; |
uint32_t Run_Costs0[64+1]; |
779 |
uint32_t * const Run_Costs = Run_Costs0 + 1; |
uint32_t * const Run_Costs = Run_Costs0 + 1; |
780 |
|
|
787 |
int Last_Node = -1; |
int Last_Node = -1; |
788 |
uint32_t Last_Cost = 0; |
uint32_t Last_Cost = 0; |
789 |
|
|
790 |
int i, j, sum; |
int i, j; |
791 |
|
|
792 |
/* source (w/ CBP penalty) */ |
/* source (w/ CBP penalty) */ |
793 |
Run_Costs[-1] = 2<<TL_SHIFT; |
Run_Costs[-1] = 2<<TL_SHIFT; |
967 |
} |
} |
968 |
} |
} |
969 |
|
|
970 |
/* It seems trellis doesn't give good results... just compute the Out sum |
/* It seems trellis doesn't give good results... just leave the block untouched |
971 |
* and quit */ |
* and return the original sum value */ |
972 |
if (Last_Node<0) |
if (Last_Node<0) |
973 |
return Compute_Sum(Out, Non_Zero); |
return Sum; |
974 |
|
|
975 |
/* reconstruct optimal sequence backward with surviving paths */ |
/* reconstruct optimal sequence backward with surviving paths */ |
976 |
memset(Out, 0x00, 64*sizeof(*Out)); |
memset(Out, 0x00, 64*sizeof(*Out)); |
977 |
Out[Zigzag[Last_Node]] = Last.Level; |
Out[Zigzag[Last_Node]] = Last.Level; |
978 |
i = Last_Node - Last.Run; |
i = Last_Node - Last.Run; |
979 |
sum = 0; |
Sum = abs(Last.Level); |
980 |
while(i>=0) { |
while(i>=0) { |
981 |
Out[Zigzag[i]] = Nodes[i].Level; |
Out[Zigzag[i]] = Nodes[i].Level; |
982 |
sum += abs(Nodes[i].Level); |
Sum += abs(Nodes[i].Level); |
983 |
i -= Nodes[i].Run; |
i -= Nodes[i].Run; |
984 |
} |
} |
985 |
|
|
986 |
return sum; |
return Sum; |
987 |
} |
} |
988 |
|
|
989 |
/* original version including heavy debugging info */ |
/* original version including heavy debugging info */ |