1 |
/***************************************************************************** |
/****************************************************************************** |
2 |
* |
* * |
3 |
* XVID MPEG-4 VIDEO CODEC |
* This file is part of XviD, a free MPEG-4 video encoder/decoder * |
4 |
* - MB Transfert/Quantization functions - |
* * |
5 |
* |
* XviD is an implementation of a part of one or more MPEG-4 Video tools * |
6 |
* Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> |
* as specified in ISO/IEC 14496-2 standard. Those intending to use this * |
7 |
* 2001-2003 Michael Militzer <isibaar@xvid.org> |
* software module in hardware or software products are advised that its * |
8 |
* 2003 Edouard Gomez <ed.gomez@free.fr> |
* use may infringe existing patents or copyrights, and any such use * |
9 |
* |
* would be at such party's own risk. The original developer of this * |
10 |
* This program is free software ; you can redistribute it and/or modify |
* software module and his/her company, and subsequent editors and their * |
11 |
* it under the terms of the GNU General Public License as published by |
* companies, will have no liability for use of this software or * |
12 |
* the Free Software Foundation ; either version 2 of the License, or |
* modifications or derivatives thereof. * |
13 |
* (at your option) any later version. |
* * |
14 |
* |
* XviD is free software; you can redistribute it and/or modify it * |
15 |
* This program is distributed in the hope that it will be useful, |
* under the terms of the GNU General Public License as published by * |
16 |
* but WITHOUT ANY WARRANTY ; without even the implied warranty of |
* the Free Software Foundation; either version 2 of the License, or * |
17 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
* (at your option) any later version. * |
18 |
* GNU General Public License for more details. |
* * |
19 |
* |
* XviD is distributed in the hope that it will be useful, but * |
20 |
* You should have received a copy of the GNU General Public License |
* WITHOUT ANY WARRANTY; without even the implied warranty of * |
21 |
* along with this program ; if not, write to the Free Software |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
22 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
* GNU General Public License for more details. * |
23 |
* |
* * |
24 |
* $Id$ |
* You should have received a copy of the GNU General Public License * |
25 |
* |
* along with this program; if not, write to the Free Software * |
26 |
****************************************************************************/ |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * |
27 |
|
* * |
28 |
|
******************************************************************************/ |
29 |
|
|
30 |
|
/****************************************************************************** |
31 |
|
* * |
32 |
|
* mbtransquant.c * |
33 |
|
* * |
34 |
|
* Copyright (C) 2001 - Peter Ross <pross@cs.rmit.edu.au> * |
35 |
|
* Copyright (C) 2001 - Michael Militzer <isibaar@xvid.org> * |
36 |
|
* * |
37 |
|
* For more information visit the XviD homepage: http://www.xvid.org * |
38 |
|
* * |
39 |
|
******************************************************************************/ |
40 |
|
|
41 |
|
/****************************************************************************** |
42 |
|
* * |
43 |
|
* Revision history: * |
44 |
|
* * |
45 |
|
* 29.03.2002 interlacing speedup - used transfer strides instead of * |
46 |
|
* manual field-to-frame conversion * |
47 |
|
* 26.03.2002 interlacing support - moved transfers outside loops * |
48 |
|
* 22.12.2001 get_dc_scaler() moved to common.h * |
49 |
|
* 19.11.2001 introduced coefficient thresholding (Isibaar) * |
50 |
|
* 17.11.2001 initial version * |
51 |
|
* * |
52 |
|
******************************************************************************/ |
53 |
|
|
54 |
#include <string.h> |
#include <string.h> |
|
#include <stdlib.h> |
|
55 |
|
|
56 |
#include "../portab.h" |
#include "../portab.h" |
57 |
#include "mbfunctions.h" |
#include "mbfunctions.h" |
69 |
|
|
70 |
MBFIELDTEST_PTR MBFieldTest; |
MBFIELDTEST_PTR MBFieldTest; |
71 |
|
|
72 |
/* |
#define TOOSMALL_LIMIT 1 /* skip blocks having a coefficient sum below this value */ |
|
* Skip blocks having a coefficient sum below this value. This value will be |
|
|
* corrected according to the MB quantizer to avoid artifacts for quant==1 |
|
|
*/ |
|
|
#define PVOP_TOOSMALL_LIMIT 1 |
|
|
#define BVOP_TOOSMALL_LIMIT 3 |
|
|
|
|
|
/***************************************************************************** |
|
|
* Local functions |
|
|
****************************************************************************/ |
|
|
|
|
|
/* permute block and return field dct choice */ |
|
|
static __inline uint32_t |
|
|
MBDecideFieldDCT(int16_t data[6 * 64]) |
|
|
{ |
|
|
uint32_t field = MBFieldTest(data); |
|
|
|
|
|
if (field) |
|
|
MBFrameToField(data); |
|
73 |
|
|
|
return field; |
|
|
} |
|
|
|
|
|
/* Performs Forward DCT on all blocks */ |
|
74 |
static __inline void |
static __inline void |
75 |
MBfDCT(const MBParam * pParam, |
MBfDCT(int16_t data[6 * 64]) |
|
FRAMEINFO * frame, |
|
|
MACROBLOCK * pMB, |
|
|
uint32_t x_pos, |
|
|
uint32_t y_pos, |
|
|
int16_t data[6 * 64]) |
|
76 |
{ |
{ |
|
/* Handles interlacing */ |
|
|
start_timer(); |
|
|
pMB->field_dct = 0; |
|
|
if ((frame->vol_flags & XVID_VOL_INTERLACING) && |
|
|
(x_pos>0) && (x_pos<pParam->mb_width-1) && |
|
|
(y_pos>0) && (y_pos<pParam->mb_height-1)) { |
|
|
pMB->field_dct = MBDecideFieldDCT(data); |
|
|
} |
|
|
stop_interlacing_timer(); |
|
|
|
|
|
/* Perform DCT */ |
|
77 |
start_timer(); |
start_timer(); |
78 |
fdct(&data[0 * 64]); |
fdct(&data[0 * 64]); |
79 |
fdct(&data[1 * 64]); |
fdct(&data[1 * 64]); |
84 |
stop_dct_timer(); |
stop_dct_timer(); |
85 |
} |
} |
86 |
|
|
|
/* Performs Inverse DCT on all blocks */ |
|
|
static __inline void |
|
|
MBiDCT(int16_t data[6 * 64], |
|
|
const uint8_t cbp) |
|
|
{ |
|
|
start_timer(); |
|
|
if(cbp & (1 << (5 - 0))) idct(&data[0 * 64]); |
|
|
if(cbp & (1 << (5 - 1))) idct(&data[1 * 64]); |
|
|
if(cbp & (1 << (5 - 2))) idct(&data[2 * 64]); |
|
|
if(cbp & (1 << (5 - 3))) idct(&data[3 * 64]); |
|
|
if(cbp & (1 << (5 - 4))) idct(&data[4 * 64]); |
|
|
if(cbp & (1 << (5 - 5))) idct(&data[5 * 64]); |
|
|
stop_idct_timer(); |
|
|
} |
|
87 |
|
|
88 |
/* Quantize all blocks -- Intra mode */ |
static __inline uint32_t |
89 |
static __inline void |
QuantizeInterBlock( int16_t qcoeff[64], |
90 |
MBQuantIntra(const MBParam * pParam, |
const int16_t data[64], |
91 |
const MACROBLOCK * pMB, |
const uint32_t iQuant, |
92 |
int16_t qcoeff[6 * 64], |
const uint32_t quant_type) |
|
int16_t data[6*64]) |
|
93 |
{ |
{ |
94 |
int i; |
uint32_t sum; |
95 |
|
|
|
for (i = 0; i < 6; i++) { |
|
|
uint32_t iDcScaler = get_dc_scaler(pMB->quant, i < 4); |
|
|
|
|
|
/* Quantize the block */ |
|
96 |
start_timer(); |
start_timer(); |
97 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) |
if (quant_type == H263_QUANT) |
98 |
quant_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler); |
sum = quant_inter(qcoeff, data, iQuant); |
99 |
else |
else |
100 |
quant4_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler); |
sum = quant4_inter(qcoeff, data, iQuant); |
|
stop_quant_timer(); |
|
|
} |
|
|
} |
|
|
|
|
|
/* DeQuantize all blocks -- Intra mode */ |
|
|
static __inline void |
|
|
MBDeQuantIntra(const MBParam * pParam, |
|
|
const int iQuant, |
|
|
int16_t qcoeff[6 * 64], |
|
|
int16_t data[6*64]) |
|
|
{ |
|
|
int i; |
|
|
|
|
|
for (i = 0; i < 6; i++) { |
|
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
|
101 |
|
|
102 |
start_timer(); |
stop_quant_timer(); |
103 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) |
return sum; |
|
dequant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
|
|
else |
|
|
dequant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
|
|
stop_iquant_timer(); |
|
|
} |
|
104 |
} |
} |
105 |
|
|
106 |
/* Quantize all blocks -- Inter mode */ |
void |
107 |
static __inline uint8_t |
MBTransQuantIntra(const MBParam * const pParam, |
108 |
MBQuantInter(const MBParam * pParam, |
FRAMEINFO * const frame, |
109 |
const MACROBLOCK * pMB, |
MACROBLOCK * const pMB, |
110 |
|
const uint32_t x_pos, |
111 |
|
const uint32_t y_pos, |
112 |
int16_t data[6 * 64], |
int16_t data[6 * 64], |
113 |
int16_t qcoeff[6 * 64], |
int16_t qcoeff[6 * 64]) |
|
int bvop, |
|
|
int limit) |
|
114 |
{ |
{ |
115 |
|
|
116 |
|
uint32_t stride = pParam->edged_width; |
117 |
|
const uint32_t stride2 = stride / 2; |
118 |
|
uint32_t next_block = stride * ((frame->global_flags & XVID_REDUCED)?16:8); |
119 |
int i; |
int i; |
120 |
uint8_t cbp = 0; |
const uint32_t iQuant = pMB->quant; |
121 |
int sum; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
122 |
int code_block; |
const IMAGE * const pCurrent = &frame->image; |
|
|
|
|
for (i = 0; i < 6; i++) { |
|
123 |
|
|
|
/* Quantize the block */ |
|
124 |
start_timer(); |
start_timer(); |
125 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) |
if ((frame->global_flags & XVID_REDUCED)) |
126 |
sum = quant_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant); |
{ |
127 |
else |
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
128 |
sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant); |
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
129 |
stop_quant_timer(); |
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
130 |
|
|
131 |
/* |
filter_18x18_to_8x8(&data[0 * 64], pY_Cur, stride); |
132 |
* We code the block if the sum is higher than the limit and if the first |
filter_18x18_to_8x8(&data[1 * 64], pY_Cur + 16, stride); |
133 |
* two AC coefficients in zig zag order are not zero. |
filter_18x18_to_8x8(&data[2 * 64], pY_Cur + next_block, stride); |
134 |
*/ |
filter_18x18_to_8x8(&data[3 * 64], pY_Cur + next_block + 16, stride); |
135 |
code_block = 0; |
filter_18x18_to_8x8(&data[4 * 64], pU_Cur, stride2); |
136 |
if ((sum >= limit) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) { |
filter_18x18_to_8x8(&data[5 * 64], pV_Cur, stride2); |
|
code_block = 1; |
|
137 |
} else { |
} else { |
138 |
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
139 |
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
140 |
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
141 |
|
|
142 |
if (bvop && (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V)) { |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
143 |
/* dark blocks prevention for direct mode */ |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
144 |
if ((qcoeff[i*64] < -1) || (qcoeff[i*64] > 0)) |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
145 |
code_block = 1; |
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
146 |
} else { |
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
147 |
/* not direct mode */ |
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
|
if (qcoeff[i*64] != 0) |
|
|
code_block = 1; |
|
148 |
} |
} |
149 |
|
stop_transfer_timer(); |
150 |
|
|
151 |
|
/* XXX: rrv+interlacing is buggy */ |
152 |
|
start_timer(); |
153 |
|
pMB->field_dct = 0; |
154 |
|
if ((frame->global_flags & XVID_INTERLACING) && |
155 |
|
(x_pos>0) && (x_pos<pParam->mb_width-1) && |
156 |
|
(y_pos>0) && (y_pos<pParam->mb_height-1)) { |
157 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
158 |
} |
} |
159 |
|
stop_interlacing_timer(); |
160 |
|
|
161 |
/* Set the corresponding cbp bit */ |
MBfDCT(data); |
|
cbp |= code_block << (5 - i); |
|
162 |
|
|
163 |
} |
for (i = 0; i < 6; i++) { |
164 |
|
const uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
165 |
|
|
166 |
return(cbp); |
start_timer(); |
167 |
} |
if (pParam->m_quant_type == H263_QUANT) |
168 |
|
quant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
169 |
|
else |
170 |
|
quant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler); |
171 |
|
stop_quant_timer(); |
172 |
|
|
173 |
/* DeQuantize all blocks -- Inter mode */ |
/* speedup: dont decode when encoding only ivops */ |
174 |
static __inline void |
if (pParam->iMaxKeyInterval != 1 || pParam->max_bframes > 0) |
|
MBDeQuantInter(const MBParam * pParam, |
|
|
const int iQuant, |
|
|
int16_t data[6 * 64], |
|
|
int16_t qcoeff[6 * 64], |
|
|
const uint8_t cbp) |
|
175 |
{ |
{ |
|
int i; |
|
|
|
|
|
for (i = 0; i < 6; i++) { |
|
|
if (cbp & (1 << (5 - i))) { |
|
176 |
start_timer(); |
start_timer(); |
177 |
if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) |
if (pParam->m_quant_type == H263_QUANT) |
178 |
dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
dequant_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
179 |
else |
else |
180 |
dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
dequant4_intra(&data[i * 64], &qcoeff[i * 64], iQuant, iDcScaler); |
181 |
stop_iquant_timer(); |
stop_iquant_timer(); |
182 |
} |
|
183 |
|
start_timer(); |
184 |
|
idct(&data[i * 64]); |
185 |
|
stop_idct_timer(); |
186 |
} |
} |
187 |
} |
} |
188 |
|
|
189 |
typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS); |
/* speedup: dont decode when encoding only ivops */ |
190 |
typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS); |
if (pParam->iMaxKeyInterval != 1 || pParam->max_bframes > 0) |
|
|
|
|
|
|
|
static __inline void |
|
|
MBTrans8to16(const MBParam * pParam, |
|
|
FRAMEINFO * frame, |
|
|
MACROBLOCK * pMB, |
|
|
const uint32_t x_pos, |
|
|
const uint32_t y_pos, |
|
|
int16_t data[6 * 64]) |
|
191 |
{ |
{ |
|
uint32_t stride = pParam->edged_width; |
|
|
uint32_t stride2 = stride / 2; |
|
|
uint32_t next_block = stride * 8; |
|
|
int32_t cst; |
|
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
|
|
IMAGE *pCurrent = &frame->image; |
|
|
transfer_operation_8to16_t *transfer_op = NULL; |
|
|
|
|
|
if ((frame->vop_flags & XVID_VOP_REDUCED)) { |
|
|
|
|
|
/* Image pointers */ |
|
|
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
|
|
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
|
|
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
|
|
|
|
|
/* Block size */ |
|
|
cst = 16; |
|
|
|
|
|
/* Operation function */ |
|
|
transfer_op = (transfer_operation_8to16_t*)filter_18x18_to_8x8; |
|
|
} else { |
|
192 |
|
|
193 |
/* Image pointers */ |
if (pMB->field_dct) { |
194 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
next_block = stride; |
195 |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
stride *= 2; |
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
|
|
|
|
|
/* Block size */ |
|
|
cst = 8; |
|
|
|
|
|
/* Operation function */ |
|
|
transfer_op = (transfer_operation_8to16_t*)transfer_8to16copy; |
|
196 |
} |
} |
197 |
|
|
|
/* Do the transfer */ |
|
198 |
start_timer(); |
start_timer(); |
199 |
transfer_op(&data[0 * 64], pY_Cur, stride); |
if ((frame->global_flags & XVID_REDUCED)) { |
200 |
transfer_op(&data[1 * 64], pY_Cur + cst, stride); |
copy_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); |
201 |
transfer_op(&data[2 * 64], pY_Cur + next_block, stride); |
copy_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); |
202 |
transfer_op(&data[3 * 64], pY_Cur + next_block + cst, stride); |
copy_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); |
203 |
transfer_op(&data[4 * 64], pU_Cur, stride2); |
copy_upsampled_8x8_16to8(pY_Cur + next_block + 16, &data[3 * 64], stride); |
204 |
transfer_op(&data[5 * 64], pV_Cur, stride2); |
copy_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); |
205 |
|
copy_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); |
206 |
|
} else { |
207 |
|
transfer_16to8copy(pY_Cur, &data[0 * 64], stride); |
208 |
|
transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); |
209 |
|
transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); |
210 |
|
transfer_16to8copy(pY_Cur + next_block + 8, &data[3 * 64], stride); |
211 |
|
transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); |
212 |
|
transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); |
213 |
|
} |
214 |
stop_transfer_timer(); |
stop_transfer_timer(); |
215 |
} |
} |
216 |
|
|
217 |
static __inline void |
} |
218 |
MBTrans16to8(const MBParam * pParam, |
|
219 |
FRAMEINFO * frame, |
uint8_t |
220 |
MACROBLOCK * pMB, |
MBTransQuantInter(const MBParam * const pParam, |
221 |
|
FRAMEINFO * const frame, |
222 |
|
MACROBLOCK * const pMB, |
223 |
const uint32_t x_pos, |
const uint32_t x_pos, |
224 |
const uint32_t y_pos, |
const uint32_t y_pos, |
225 |
int16_t data[6 * 64], |
int16_t data[6 * 64], |
226 |
const uint32_t add, |
int16_t qcoeff[6 * 64]) |
|
const uint8_t cbp) |
|
227 |
{ |
{ |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
|
228 |
uint32_t stride = pParam->edged_width; |
uint32_t stride = pParam->edged_width; |
229 |
uint32_t stride2 = stride / 2; |
const uint32_t stride2 = stride / 2; |
230 |
uint32_t next_block = stride * 8; |
uint32_t next_block = stride * ((frame->global_flags & XVID_REDUCED)?16:8); |
231 |
uint32_t cst; |
int i; |
232 |
IMAGE *pCurrent = &frame->image; |
const uint32_t iQuant = pMB->quant; |
233 |
transfer_operation_16to8_t *transfer_op = NULL; |
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
234 |
|
int cbp = 0; |
235 |
if (pMB->field_dct) { |
uint32_t sum; |
236 |
next_block = stride; |
const IMAGE * const pCurrent = &frame->image; |
|
stride *= 2; |
|
|
} |
|
|
|
|
|
if ((frame->vop_flags & XVID_VOP_REDUCED)) { |
|
237 |
|
|
238 |
/* Image pointers */ |
if ((frame->global_flags & XVID_REDUCED)) { |
239 |
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5); |
240 |
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4); |
241 |
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4); |
|
|
|
|
/* Block size */ |
|
|
cst = 16; |
|
|
|
|
|
/* Operation function */ |
|
|
if(add) |
|
|
transfer_op = (transfer_operation_16to8_t*)add_upsampled_8x8_16to8; |
|
|
else |
|
|
transfer_op = (transfer_operation_16to8_t*)copy_upsampled_8x8_16to8; |
|
242 |
} else { |
} else { |
|
|
|
|
/* Image pointers */ |
|
243 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
244 |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
245 |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
|
|
|
|
/* Block size */ |
|
|
cst = 8; |
|
|
|
|
|
/* Operation function */ |
|
|
if(add) |
|
|
transfer_op = (transfer_operation_16to8_t*)transfer_16to8add; |
|
|
else |
|
|
transfer_op = (transfer_operation_16to8_t*)transfer_16to8copy; |
|
246 |
} |
} |
247 |
|
|
|
/* Do the operation */ |
|
248 |
start_timer(); |
start_timer(); |
249 |
if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); |
pMB->field_dct = 0; |
250 |
if (cbp&16) transfer_op(pY_Cur + cst, &data[1 * 64], stride); |
if ((frame->global_flags & XVID_INTERLACING) && |
251 |
if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride); |
(x_pos>0) && (x_pos<pParam->mb_width-1) && |
252 |
if (cbp& 4) transfer_op(pY_Cur + next_block + cst, &data[3 * 64], stride); |
(y_pos>0) && (y_pos<pParam->mb_height-1)) { |
253 |
if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); |
pMB->field_dct = MBDecideFieldDCT(data); |
|
if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); |
|
|
stop_transfer_timer(); |
|
|
} |
|
|
|
|
|
/***************************************************************************** |
|
|
* Module functions |
|
|
****************************************************************************/ |
|
|
|
|
|
void |
|
|
MBTransQuantIntra(const MBParam * pParam, |
|
|
FRAMEINFO * frame, |
|
|
MACROBLOCK * pMB, |
|
|
const uint32_t x_pos, |
|
|
const uint32_t y_pos, |
|
|
int16_t data[6 * 64], |
|
|
int16_t qcoeff[6 * 64]) |
|
|
{ |
|
|
|
|
|
/* Transfer data */ |
|
|
MBTrans8to16(pParam, frame, pMB, x_pos, y_pos, data); |
|
|
|
|
|
/* Perform DCT (and field decision) */ |
|
|
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
|
|
|
|
|
/* Quantize the block */ |
|
|
MBQuantIntra(pParam, pMB, data, qcoeff); |
|
|
|
|
|
/* DeQuantize the block */ |
|
|
MBDeQuantIntra(pParam, pMB->quant, data, qcoeff); |
|
|
|
|
|
/* Perform inverse DCT*/ |
|
|
MBiDCT(data, 0x3F); |
|
|
|
|
|
/* Transfer back the data -- Don't add data */ |
|
|
MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 0, 0x3F); |
|
254 |
} |
} |
255 |
|
stop_interlacing_timer(); |
256 |
|
|
257 |
|
MBfDCT(data); |
258 |
|
|
259 |
uint8_t |
for (i = 0; i < 6; i++) { |
260 |
MBTransQuantInter(const MBParam * pParam, |
const uint32_t limit = TOOSMALL_LIMIT + ((iQuant == 1) ? 1 : 0); |
|
FRAMEINFO * frame, |
|
|
MACROBLOCK * pMB, |
|
|
const uint32_t x_pos, |
|
|
const uint32_t y_pos, |
|
|
int16_t data[6 * 64], |
|
|
int16_t qcoeff[6 * 64]) |
|
|
{ |
|
|
uint8_t cbp; |
|
|
uint32_t limit; |
|
|
|
|
261 |
/* |
/* |
262 |
* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
* no need to transfer 8->16-bit |
263 |
* already |
* (this is performed already in motion compensation) |
264 |
*/ |
*/ |
265 |
|
|
266 |
/* Perform DCT (and field decision) */ |
sum = QuantizeInterBlock(&qcoeff[i * 64], &data[i * 64], iQuant, pParam->m_quant_type); |
|
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
|
267 |
|
|
268 |
/* Set the limit threshold */ |
if (sum >= limit) { |
|
limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0); |
|
269 |
|
|
270 |
/* Quantize the block */ |
start_timer(); |
271 |
cbp = MBQuantInter(pParam, pMB, data, qcoeff, 0, limit); |
if (pParam->m_quant_type == H263_QUANT) |
272 |
|
dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
273 |
|
else |
274 |
|
dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant); |
275 |
|
stop_iquant_timer(); |
276 |
|
|
277 |
/* DeQuantize the block */ |
cbp |= 1 << (5 - i); |
|
MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp); |
|
278 |
|
|
279 |
/* Perform inverse DCT*/ |
start_timer(); |
280 |
MBiDCT(data, cbp); |
idct(&data[i * 64]); |
281 |
|
stop_idct_timer(); |
282 |
|
} |
283 |
|
} |
284 |
|
|
285 |
/* Transfer back the data -- Add the data */ |
if (pMB->field_dct) { |
286 |
MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp); |
next_block = stride; |
287 |
|
stride *= 2; |
288 |
|
} |
289 |
|
|
290 |
return(cbp); |
start_timer(); |
291 |
|
if ((frame->global_flags & XVID_REDUCED)) { |
292 |
|
if (cbp & 32) |
293 |
|
add_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); |
294 |
|
if (cbp & 16) |
295 |
|
add_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); |
296 |
|
if (cbp & 8) |
297 |
|
add_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); |
298 |
|
if (cbp & 4) |
299 |
|
add_upsampled_8x8_16to8(pY_Cur + 16 + next_block, &data[3 * 64], stride); |
300 |
|
if (cbp & 2) |
301 |
|
add_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); |
302 |
|
if (cbp & 1) |
303 |
|
add_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); |
304 |
|
} else { |
305 |
|
if (cbp & 32) |
306 |
|
transfer_16to8add(pY_Cur, &data[0 * 64], stride); |
307 |
|
if (cbp & 16) |
308 |
|
transfer_16to8add(pY_Cur + 8, &data[1 * 64], stride); |
309 |
|
if (cbp & 8) |
310 |
|
transfer_16to8add(pY_Cur + next_block, &data[2 * 64], stride); |
311 |
|
if (cbp & 4) |
312 |
|
transfer_16to8add(pY_Cur + next_block + 8, &data[3 * 64], stride); |
313 |
|
if (cbp & 2) |
314 |
|
transfer_16to8add(pU_Cur, &data[4 * 64], stride2); |
315 |
|
if (cbp & 1) |
316 |
|
transfer_16to8add(pV_Cur, &data[5 * 64], stride2); |
317 |
|
} |
318 |
|
stop_transfer_timer(); |
319 |
|
|
320 |
|
return (uint8_t) cbp; |
321 |
} |
} |
322 |
|
|
323 |
uint8_t |
uint8_t |
324 |
MBTransQuantInterBVOP(const MBParam * pParam, |
MBTransQuantInterBVOP(const MBParam * pParam, |
325 |
FRAMEINFO * frame, |
FRAMEINFO * frame, |
326 |
MACROBLOCK * pMB, |
MACROBLOCK * pMB, |
|
const uint32_t x_pos, |
|
|
const uint32_t y_pos, |
|
327 |
int16_t data[6 * 64], |
int16_t data[6 * 64], |
328 |
int16_t qcoeff[6 * 64]) |
int16_t qcoeff[6 * 64]) |
329 |
{ |
{ |
330 |
uint8_t cbp; |
int cbp = 0; |
331 |
uint32_t limit; |
int i; |
332 |
|
|
333 |
/* |
/* there is no MBTrans for Inter block, that's done in motion compensation already */ |
|
* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
|
|
* already |
|
|
*/ |
|
334 |
|
|
335 |
/* Perform DCT (and field decision) */ |
start_timer(); |
336 |
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
pMB->field_dct = 0; |
337 |
|
if ((frame->global_flags & XVID_INTERLACING)) { |
338 |
|
pMB->field_dct = MBDecideFieldDCT(data); |
339 |
|
} |
340 |
|
stop_interlacing_timer(); |
341 |
|
|
342 |
/* Set the limit threshold */ |
MBfDCT(data); |
|
limit = BVOP_TOOSMALL_LIMIT; |
|
343 |
|
|
344 |
/* Quantize the block */ |
for (i = 0; i < 6; i++) { |
345 |
cbp = MBQuantInter(pParam, pMB, data, qcoeff, 1, limit); |
int codedecision = 0; |
346 |
|
|
347 |
/* |
int sum = QuantizeInterBlock(&qcoeff[i * 64], &data[i * 64], pMB->quant, pParam->m_quant_type); |
|
* History comment: |
|
|
* We don't have to DeQuant, iDCT and Transfer back data for B-frames. |
|
|
* |
|
|
* BUT some plugins require the original frame to be passed so we have |
|
|
* to take care of that here |
|
|
*/ |
|
|
if((pParam->plugin_flags & XVID_REQORIGINAL)) { |
|
348 |
|
|
349 |
/* DeQuantize the block */ |
if ((sum > 2) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0) ) codedecision = 1; |
350 |
MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp); |
else { |
351 |
|
if (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V) { |
352 |
|
// dark blocks prevention for direct mode |
353 |
|
if ( (qcoeff[i*64] < -1) || (qcoeff[i*64] > 0) ) codedecision = 1; |
354 |
|
} else |
355 |
|
if (qcoeff[i*64] != 0) codedecision = 1; // not direct mode |
356 |
|
} |
357 |
|
|
358 |
/* Perform inverse DCT*/ |
if (codedecision) cbp |= 1 << (5 - i); |
359 |
MBiDCT(data, cbp); |
} |
360 |
|
|
361 |
/* Transfer back the data -- Add the data */ |
/* we don't have to DeQuant, iDCT and Transfer back data for B-frames if we don't reconstruct this frame */ |
362 |
MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp); |
/* warning: reconstruction not supported yet */ |
363 |
|
return (uint8_t) cbp; |
364 |
} |
} |
365 |
|
|
366 |
return(cbp); |
/* permute block and return field dct choice */ |
367 |
|
|
368 |
|
static uint32_t |
369 |
|
MBDecideFieldDCT(int16_t data[6 * 64]) |
370 |
|
{ |
371 |
|
const uint32_t field = MBFieldTest(data); |
372 |
|
if (field) MBFrameToField(data); |
373 |
|
|
374 |
|
return field; |
375 |
} |
} |
376 |
|
|
377 |
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
378 |
|
|
379 |
uint32_t |
uint32_t |
380 |
MBFieldTest_c(int16_t data[6 * 64]) |
MBFieldTest_c(int16_t data[6 * 64]) |
381 |
{ |
{ |
389 |
for (i = 0; i < 7; ++i) { |
for (i = 0; i < 7; ++i) { |
390 |
for (j = 0; j < 8; ++j) { |
for (j = 0; j < 8; ++j) { |
391 |
frame += |
frame += |
392 |
abs(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]); |
ABS(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]); |
393 |
frame += |
frame += |
394 |
abs(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]); |
ABS(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]); |
395 |
frame += |
frame += |
396 |
abs(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]); |
ABS(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]); |
397 |
frame += |
frame += |
398 |
abs(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]); |
ABS(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]); |
399 |
|
|
400 |
field += |
field += |
401 |
abs(data[blocks[i + 1] + lines[i + 1] + j] - |
ABS(data[blocks[i + 1] + lines[i + 1] + j] - |
402 |
data[blocks[i] + lines[i] + j]); |
data[blocks[i] + lines[i] + j]); |
403 |
field += |
field += |
404 |
abs(data[blocks[i + 1] + lines[i + 1] + 8 + j] - |
ABS(data[blocks[i + 1] + lines[i + 1] + 8 + j] - |
405 |
data[blocks[i] + lines[i] + 8 + j]); |
data[blocks[i] + lines[i] + 8 + j]); |
406 |
field += |
field += |
407 |
abs(data[blocks[i + 1] + 64 + lines[i + 1] + j] - |
ABS(data[blocks[i + 1] + 64 + lines[i + 1] + j] - |
408 |
data[blocks[i] + 64 + lines[i] + j]); |
data[blocks[i] + 64 + lines[i] + j]); |
409 |
field += |
field += |
410 |
abs(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] - |
ABS(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] - |
411 |
data[blocks[i] + 64 + lines[i] + 8 + j]); |
data[blocks[i] + 64 + lines[i] + 8 + j]); |
412 |
} |
} |
413 |
} |
} |