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 |
|
|
122 |
int16_t qcoeff[6 * 64], |
int16_t qcoeff[6 * 64], |
123 |
int16_t data[6*64]) |
int16_t data[6*64]) |
124 |
{ |
{ |
|
int mpeg; |
|
125 |
int scaler_lum, scaler_chr; |
int scaler_lum, scaler_chr; |
126 |
|
quant_intraFuncPtr quant; |
127 |
|
|
128 |
quant_intraFuncPtr const quant[2] = |
/* check if quant matrices need to be re-initialized with new quant */ |
129 |
{ |
if (pParam->vol_flags & XVID_VOL_MPEGQUANT) { |
130 |
quant_h263_intra, |
if (pParam->last_quant_initialized_intra != pMB->quant) { |
131 |
quant_mpeg_intra |
init_intra_matrix(pParam->mpeg_quant_matrices, pMB->quant); |
132 |
}; |
} |
133 |
|
quant = quant_mpeg_intra; |
134 |
|
} else { |
135 |
|
quant = quant_h263_intra; |
136 |
|
} |
137 |
|
|
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
|
138 |
scaler_lum = get_dc_scaler(pMB->quant, 1); |
scaler_lum = get_dc_scaler(pMB->quant, 1); |
139 |
scaler_chr = get_dc_scaler(pMB->quant, 0); |
scaler_chr = get_dc_scaler(pMB->quant, 0); |
140 |
|
|
141 |
/* Quantize the block */ |
/* Quantize the block */ |
142 |
start_timer(); |
start_timer(); |
143 |
quant[mpeg](&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
quant(&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
144 |
quant[mpeg](&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
quant(&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
145 |
quant[mpeg](&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
quant(&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
146 |
quant[mpeg](&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
quant(&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
147 |
quant[mpeg](&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
quant(&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
148 |
quant[mpeg](&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
quant(&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
149 |
stop_quant_timer(); |
stop_quant_timer(); |
150 |
} |
} |
151 |
|
|
186 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
187 |
const uint16_t * const QuantMatrix, |
const uint16_t * const QuantMatrix, |
188 |
int Non_Zero, |
int Non_Zero, |
189 |
int Sum); |
int Sum, |
190 |
|
int Lambda_Mod); |
191 |
|
|
192 |
/* Quantize all blocks -- Inter mode */ |
/* Quantize all blocks -- Inter mode */ |
193 |
static __inline uint8_t |
static __inline uint8_t |
220 |
|
|
221 |
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); |
222 |
|
|
223 |
if(sum && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
if(sum && (pMB->quant > 2) && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
224 |
const uint16_t *matrix; |
const uint16_t *matrix; |
225 |
const static uint16_t h263matrix[] = |
const static uint16_t h263matrix[] = |
226 |
{ |
{ |
239 |
pMB->quant, &scan_tables[0][0], |
pMB->quant, &scan_tables[0][0], |
240 |
matrix, |
matrix, |
241 |
63, |
63, |
242 |
sum); |
sum, |
243 |
|
pMB->lambda[i]); |
244 |
} |
} |
245 |
stop_quant_timer(); |
stop_quant_timer(); |
246 |
|
|
314 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
315 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
316 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * 8; |
|
int32_t cst; |
|
|
int vop_reduced; |
|
317 |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
318 |
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); |
|
319 |
|
|
320 |
/* Image pointers */ |
/* Image pointers */ |
321 |
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); |
322 |
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); |
323 |
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]; |
|
324 |
|
|
325 |
/* Do the transfer */ |
/* Do the transfer */ |
326 |
start_timer(); |
start_timer(); |
327 |
transfer_op(&data[0 * 64], pY_Cur, stride); |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
328 |
transfer_op(&data[1 * 64], pY_Cur + cst, stride); |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
329 |
transfer_op(&data[2 * 64], pY_Cur + next_block, stride); |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
330 |
transfer_op(&data[3 * 64], pY_Cur + next_block + cst, stride); |
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
331 |
transfer_op(&data[4 * 64], pU_Cur, stride2); |
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
332 |
transfer_op(&data[5 * 64], pV_Cur, stride2); |
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
333 |
stop_transfer_timer(); |
stop_transfer_timer(); |
334 |
} |
} |
335 |
|
|
347 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
348 |
uint32_t stride2 = stride / 2; |
uint32_t stride2 = stride / 2; |
349 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * 8; |
|
uint32_t cst; |
|
|
int vop_reduced; |
|
350 |
const IMAGE * const pCurrent = &frame->image; |
const IMAGE * const pCurrent = &frame->image; |
351 |
|
|
352 |
/* Array of function pointers, indexed by [vop_reduced<<1+add] */ |
/* Array of function pointers, indexed by [add] */ |
353 |
transfer_operation_16to8_t * const functions[4] = |
transfer_operation_16to8_t * const functions[2] = |
354 |
{ |
{ |
355 |
(transfer_operation_16to8_t*)transfer_16to8copy, |
(transfer_operation_16to8_t*)transfer_16to8copy, |
356 |
(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 |
|
357 |
}; |
}; |
358 |
|
|
359 |
transfer_operation_16to8_t *transfer_op = NULL; |
transfer_operation_16to8_t *transfer_op = NULL; |
360 |
|
|
|
/* Makes this vars booleans */ |
|
|
vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED); |
|
|
|
|
361 |
/* Image pointers */ |
/* Image pointers */ |
362 |
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); |
363 |
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); |
364 |
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); |
365 |
|
|
366 |
if (pMB->field_dct) { |
if (pMB->field_dct) { |
367 |
next_block = stride; |
next_block = stride; |
368 |
stride *= 2; |
stride *= 2; |
369 |
} |
} |
370 |
|
|
|
/* Block size */ |
|
|
cst = 8<<vop_reduced; |
|
|
|
|
371 |
/* Operation function */ |
/* Operation function */ |
372 |
transfer_op = functions[(vop_reduced<<1) + add]; |
transfer_op = functions[add]; |
373 |
|
|
374 |
/* Do the operation */ |
/* Do the operation */ |
375 |
start_timer(); |
start_timer(); |
376 |
if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); |
if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); |
377 |
if (cbp&16) transfer_op(pY_Cur + cst, &data[1 * 64], stride); |
if (cbp&16) transfer_op(pY_Cur + 8, &data[1 * 64], stride); |
378 |
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); |
379 |
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); |
380 |
if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); |
if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); |
381 |
if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); |
if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); |
382 |
stop_transfer_timer(); |
stop_transfer_timer(); |
761 |
return -1; |
return -1; |
762 |
} |
} |
763 |
|
|
764 |
|
#define TRELLIS_MIN_EFFORT 3 |
765 |
|
|
766 |
/* this routine has been strippen of all debug code */ |
/* this routine has been strippen of all debug code */ |
767 |
static int |
static int |
768 |
dct_quantize_trellis_c(int16_t *const Out, |
dct_quantize_trellis_c(int16_t *const Out, |
771 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
772 |
const uint16_t * const QuantMatrix, |
const uint16_t * const QuantMatrix, |
773 |
int Non_Zero, |
int Non_Zero, |
774 |
int Sum) |
int Sum, |
775 |
|
int Lambda_Mod) |
776 |
{ |
{ |
777 |
|
|
778 |
/* 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 |
782 |
* helps. */ |
* helps. */ |
783 |
typedef struct { int16_t Run, Level; } NODE; |
typedef struct { int16_t Run, Level; } NODE; |
784 |
|
|
785 |
NODE Nodes[65], Last; |
NODE Nodes[65], Last = { 0, 0}; |
786 |
uint32_t Run_Costs0[64+1]; |
uint32_t Run_Costs0[64+1]; |
787 |
uint32_t * const Run_Costs = Run_Costs0 + 1; |
uint32_t * const Run_Costs = Run_Costs0 + 1; |
788 |
|
|
789 |
/* it's 1/lambda, actually */ |
/* it's 1/lambda, actually */ |
790 |
const int Lambda = Trellis_Lambda_Tabs[Q-1]; |
const int Lambda = (Lambda_Mod*Trellis_Lambda_Tabs[Q-1])>>LAMBDA_EXP; |
791 |
|
|
792 |
int Run_Start = -1; |
int Run_Start = -1; |
793 |
uint32_t Min_Cost = 2<<TL_SHIFT; |
uint32_t Min_Cost = 2<<TL_SHIFT; |
801 |
Run_Costs[-1] = 2<<TL_SHIFT; |
Run_Costs[-1] = 2<<TL_SHIFT; |
802 |
|
|
803 |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
804 |
if (Non_Zero<0) |
if (Non_Zero < TRELLIS_MIN_EFFORT) |
805 |
return 0; /* Sum is zero if there are only zero coeffs */ |
Non_Zero = TRELLIS_MIN_EFFORT; |
806 |
|
|
807 |
for(i=0; i<=Non_Zero; i++) { |
for(i=0; i<=Non_Zero; i++) { |
808 |
const int q = ((Q*QuantMatrix[Zigzag[i]])>>4); |
const int q = ((Q*QuantMatrix[Zigzag[i]])>>4); |
984 |
memset(Out, 0x00, 64*sizeof(*Out)); |
memset(Out, 0x00, 64*sizeof(*Out)); |
985 |
Out[Zigzag[Last_Node]] = Last.Level; |
Out[Zigzag[Last_Node]] = Last.Level; |
986 |
i = Last_Node - Last.Run; |
i = Last_Node - Last.Run; |
987 |
Sum = 0; |
Sum = abs(Last.Level); |
988 |
while(i>=0) { |
while(i>=0) { |
989 |
Out[Zigzag[i]] = Nodes[i].Level; |
Out[Zigzag[i]] = Nodes[i].Level; |
990 |
Sum += abs(Nodes[i].Level); |
Sum += abs(Nodes[i].Level); |