Originally committed as revision 423 to svn://svn.ffmpeg.org/ffmpeg/trunk
Michael Niedermayer authored on 2002/04/27 21:30:26... | ... |
@@ -904,8 +904,26 @@ void h263_encode_init(MpegEncContext *s) |
904 | 904 |
s->mv_penalty= mv_penalty; //FIXME exact table for msmpeg4 & h263p |
905 | 905 |
|
906 | 906 |
// use fcodes >1 only for mpeg4 & h263 & h263p FIXME |
907 |
- if(s->h263_plus) s->fcode_tab= umv_fcode_tab; |
|
908 |
- else if(s->h263_pred && !s->h263_msmpeg4) s->fcode_tab= fcode_tab; |
|
907 |
+ switch(s->codec_id){ |
|
908 |
+ case CODEC_ID_MPEG4: |
|
909 |
+ s->fcode_tab= fcode_tab; |
|
910 |
+ s->min_qcoeff= -2048; |
|
911 |
+ s->max_qcoeff= 2047; |
|
912 |
+ break; |
|
913 |
+ case CODEC_ID_H263P: |
|
914 |
+ s->fcode_tab= umv_fcode_tab; |
|
915 |
+ s->min_qcoeff= -128; |
|
916 |
+ s->max_qcoeff= 127; |
|
917 |
+ break; |
|
918 |
+ default: //nothing needed default table allready set in mpegvideo.c |
|
919 |
+ s->min_qcoeff= -128; |
|
920 |
+ s->max_qcoeff= 127; |
|
921 |
+ } |
|
922 |
+ |
|
923 |
+ /* h263 type bias */ |
|
924 |
+ //FIXME mpeg4 mpeg quantizer |
|
925 |
+ s->intra_quant_bias=0; |
|
926 |
+ s->inter_quant_bias=-(1<<(QUANT_BIAS_SHIFT-2)); //(a - x/4)/x |
|
909 | 927 |
} |
910 | 928 |
|
911 | 929 |
static void h263_encode_block(MpegEncContext * s, DCTELEM * block, int n) |
... | ... |
@@ -2702,8 +2720,8 @@ int mpeg4_decode_picture_header(MpegEncContext * s) |
2702 | 2702 |
s->chroma_intra_matrix[i]= v; |
2703 | 2703 |
|
2704 | 2704 |
v= ff_mpeg4_default_non_intra_matrix[i]; |
2705 |
- s->non_intra_matrix[i]= v; |
|
2706 |
- s->chroma_non_intra_matrix[i]= v; |
|
2705 |
+ s->inter_matrix[i]= v; |
|
2706 |
+ s->chroma_inter_matrix[i]= v; |
|
2707 | 2707 |
} |
2708 | 2708 |
|
2709 | 2709 |
/* load custom intra matrix */ |
... | ... |
@@ -2725,15 +2743,15 @@ int mpeg4_decode_picture_header(MpegEncContext * s) |
2725 | 2725 |
if(v==0) break; |
2726 | 2726 |
|
2727 | 2727 |
j= zigzag_direct[i]; |
2728 |
- s->non_intra_matrix[j]= v; |
|
2729 |
- s->chroma_non_intra_matrix[j]= v; |
|
2728 |
+ s->inter_matrix[j]= v; |
|
2729 |
+ s->chroma_inter_matrix[j]= v; |
|
2730 | 2730 |
} |
2731 | 2731 |
|
2732 | 2732 |
/* replicate last value */ |
2733 | 2733 |
for(; i<64; i++){ |
2734 | 2734 |
j= zigzag_direct[i]; |
2735 |
- s->non_intra_matrix[j]= v; |
|
2736 |
- s->chroma_non_intra_matrix[j]= v; |
|
2735 |
+ s->inter_matrix[j]= v; |
|
2736 |
+ s->chroma_inter_matrix[j]= v; |
|
2737 | 2737 |
} |
2738 | 2738 |
} |
2739 | 2739 |
|
... | ... |
@@ -73,17 +73,13 @@ static int h263_decode_init(AVCodecContext *avctx) |
73 | 73 |
default: |
74 | 74 |
return -1; |
75 | 75 |
} |
76 |
- |
|
76 |
+ s->codec_id= avctx->codec->id; |
|
77 |
+ |
|
77 | 78 |
/* for h263, we allocate the images after having read the header */ |
78 | 79 |
if (avctx->codec->id != CODEC_ID_H263 && avctx->codec->id != CODEC_ID_MPEG4) |
79 | 80 |
if (MPV_common_init(s) < 0) |
80 | 81 |
return -1; |
81 | 82 |
|
82 |
- /* XXX: suppress this matrix init, only needed because using mpeg1 |
|
83 |
- dequantize in mmx case */ |
|
84 |
- for(i=0;i<64;i++) |
|
85 |
- s->non_intra_matrix[i] = default_non_intra_matrix[i]; |
|
86 |
- |
|
87 | 83 |
if (s->h263_msmpeg4) |
88 | 84 |
msmpeg4_decode_init_vlc(s); |
89 | 85 |
else |
... | ... |
@@ -251,7 +247,7 @@ static int h263_decode_frame(AVCodecContext *avctx, |
251 | 251 |
if(msmpeg4_decode_ext_header(s, buf_size) < 0) return -1; |
252 | 252 |
|
253 | 253 |
/* divx 5.01+ bistream reorder stuff */ |
254 |
- if(s->h263_pred && s->bitstream_buffer_size==0){ |
|
254 |
+ if(s->codec_id==CODEC_ID_MPEG4 && s->bitstream_buffer_size==0){ |
|
255 | 255 |
int current_pos= get_bits_count(&s->gb)/8; |
256 | 256 |
if( buf_size - current_pos > 5 |
257 | 257 |
&& buf_size - current_pos < BITSTREAM_BUFFER_SIZE){ |
... | ... |
@@ -26,8 +26,6 @@ |
26 | 26 |
#include "../mangle.h" |
27 | 27 |
|
28 | 28 |
extern UINT8 zigzag_end[64]; |
29 |
-extern void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w); |
|
30 |
-extern int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale); |
|
31 | 29 |
|
32 | 30 |
extern UINT8 zigzag_direct_noperm[64]; |
33 | 31 |
extern UINT16 inv_zigzag_direct16[64]; |
... | ... |
@@ -260,7 +258,7 @@ asm volatile( |
260 | 260 |
block[0]= block0; |
261 | 261 |
|
262 | 262 |
} else { |
263 |
- quant_matrix = s->non_intra_matrix; |
|
263 |
+ quant_matrix = s->inter_matrix; |
|
264 | 264 |
asm volatile( |
265 | 265 |
"pcmpeqw %%mm7, %%mm7 \n\t" |
266 | 266 |
"psrlw $15, %%mm7 \n\t" |
... | ... |
@@ -382,7 +380,7 @@ asm volatile( |
382 | 382 |
//Note, we dont do mismatch control for intra as errors cannot accumulate |
383 | 383 |
|
384 | 384 |
} else { |
385 |
- quant_matrix = s->non_intra_matrix; |
|
385 |
+ quant_matrix = s->inter_matrix; |
|
386 | 386 |
asm volatile( |
387 | 387 |
"pcmpeqw %%mm7, %%mm7 \n\t" |
388 | 388 |
"psrlq $48, %%mm7 \n\t" |
... | ... |
@@ -33,149 +33,160 @@ |
33 | 33 |
|
34 | 34 |
static int RENAME(dct_quantize)(MpegEncContext *s, |
35 | 35 |
DCTELEM *block, int n, |
36 |
- int qscale) |
|
36 |
+ int qscale, int *overflow) |
|
37 | 37 |
{ |
38 |
- int i, level, last_non_zero_p1, q; |
|
39 |
- const UINT16 *qmat; |
|
38 |
+ int level=0, last_non_zero_p1, q; //=0 is cuz gcc says uninitalized ... |
|
39 |
+ const UINT16 *qmat, *bias; |
|
40 | 40 |
static __align8 INT16 temp_block[64]; |
41 |
- int minLevel, maxLevel; |
|
42 |
- |
|
43 |
- if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ |
|
44 |
- /* mpeg4 */ |
|
45 |
- minLevel= -2048; |
|
46 |
- maxLevel= 2047; |
|
47 |
- }else if(s->out_format==FMT_MPEG1){ |
|
48 |
- /* mpeg1 */ |
|
49 |
- minLevel= -255; |
|
50 |
- maxLevel= 255; |
|
51 |
- }else if(s->out_format==FMT_MJPEG){ |
|
52 |
- /* (m)jpeg */ |
|
53 |
- minLevel= -1023; |
|
54 |
- maxLevel= 1023; |
|
55 |
- }else{ |
|
56 |
- /* h263 / msmpeg4 */ |
|
57 |
- minLevel= -128; |
|
58 |
- maxLevel= 127; |
|
59 |
- } |
|
60 | 41 |
|
61 | 42 |
av_fdct (block); |
62 |
- |
|
43 |
+ |
|
63 | 44 |
if (s->mb_intra) { |
64 | 45 |
int dummy; |
65 | 46 |
if (n < 4) |
66 | 47 |
q = s->y_dc_scale; |
67 | 48 |
else |
68 | 49 |
q = s->c_dc_scale; |
69 |
- |
|
70 | 50 |
/* note: block[0] is assumed to be positive */ |
71 | 51 |
#if 1 |
72 |
- asm volatile ( |
|
73 |
- "xorl %%edx, %%edx \n\t" |
|
74 |
- "mul %%ecx \n\t" |
|
75 |
- : "=d" (temp_block[0]), "=a"(dummy) |
|
76 |
- : "a" (block[0] + (q >> 1)), "c" (inverse[q]) |
|
77 |
- ); |
|
52 |
+ asm volatile ( |
|
53 |
+ "xorl %%edx, %%edx \n\t" |
|
54 |
+ "mul %%ecx \n\t" |
|
55 |
+ : "=d" (level), "=a"(dummy) |
|
56 |
+ : "a" (block[0] + (q >> 1)), "c" (inverse[q]) |
|
57 |
+ ); |
|
78 | 58 |
#else |
79 |
- asm volatile ( |
|
80 |
- "xorl %%edx, %%edx \n\t" |
|
81 |
- "divw %%cx \n\t" |
|
82 |
- "movzwl %%ax, %%eax \n\t" |
|
83 |
- : "=a" (temp_block[0]) |
|
84 |
- : "a" (block[0] + (q >> 1)), "c" (q) |
|
85 |
- : "%edx" |
|
86 |
- ); |
|
59 |
+ asm volatile ( |
|
60 |
+ "xorl %%edx, %%edx \n\t" |
|
61 |
+ "divw %%cx \n\t" |
|
62 |
+ "movzwl %%ax, %%eax \n\t" |
|
63 |
+ : "=a" (level) |
|
64 |
+ : "a" (block[0] + (q >> 1)), "c" (q) |
|
65 |
+ : "%edx" |
|
66 |
+ ); |
|
87 | 67 |
#endif |
68 |
+ block[0]=0; //avoid fake overflow |
|
88 | 69 |
// temp_block[0] = (block[0] + (q >> 1)) / q; |
89 |
- i = 1; |
|
90 | 70 |
last_non_zero_p1 = 1; |
91 |
- if (s->out_format == FMT_H263) { |
|
92 |
- qmat = s->q_non_intra_matrix16; |
|
93 |
- } else { |
|
94 |
- qmat = s->q_intra_matrix16; |
|
95 |
- } |
|
96 |
- for(i=1;i<4;i++) { |
|
97 |
- level = block[i] * qmat[i]; |
|
98 |
- level = level / (1 << (QMAT_SHIFT_MMX - 3)); |
|
99 |
- /* XXX: currently, this code is not optimal. the range should be: |
|
100 |
- mpeg1: -255..255 |
|
101 |
- mpeg2: -2048..2047 |
|
102 |
- h263: -128..127 |
|
103 |
- mpeg4: -2048..2047 |
|
104 |
- */ |
|
105 |
- if (level > maxLevel) |
|
106 |
- level = maxLevel; |
|
107 |
- else if (level < minLevel) |
|
108 |
- level = minLevel; |
|
109 |
- temp_block[i] = level; |
|
110 |
- |
|
111 |
- if(level) |
|
112 |
- if(last_non_zero_p1 < inv_zigzag_direct16[i]) last_non_zero_p1= inv_zigzag_direct16[i]; |
|
113 |
- block[i]=0; |
|
114 |
- } |
|
71 |
+ bias = s->q_intra_matrix16_bias[qscale]; |
|
72 |
+ qmat = s->q_intra_matrix16[qscale]; |
|
115 | 73 |
} else { |
116 |
- i = 0; |
|
117 | 74 |
last_non_zero_p1 = 0; |
118 |
- qmat = s->q_non_intra_matrix16; |
|
75 |
+ bias = s->q_inter_matrix16_bias[qscale]; |
|
76 |
+ qmat = s->q_inter_matrix16[qscale]; |
|
119 | 77 |
} |
120 | 78 |
|
121 |
- asm volatile( /* XXX: small rounding bug, but it shouldnt matter */ |
|
122 |
- "movd %3, %%mm3 \n\t" |
|
123 |
- SPREADW(%%mm3) |
|
124 |
- "movd %4, %%mm4 \n\t" |
|
125 |
- SPREADW(%%mm4) |
|
126 |
-#ifndef HAVE_MMX2 |
|
127 |
- "movd %5, %%mm5 \n\t" |
|
128 |
- SPREADW(%%mm5) |
|
129 |
-#endif |
|
130 |
- "pxor %%mm7, %%mm7 \n\t" |
|
131 |
- "movd %%eax, %%mm2 \n\t" |
|
132 |
- SPREADW(%%mm2) |
|
133 |
- "movl %6, %%eax \n\t" |
|
134 |
- ".balign 16 \n\t" |
|
135 |
- "1: \n\t" |
|
136 |
- "movq (%1, %%eax), %%mm0 \n\t" |
|
137 |
- "movq (%2, %%eax), %%mm1 \n\t" |
|
138 |
- "movq %%mm0, %%mm6 \n\t" |
|
139 |
- "psraw $15, %%mm6 \n\t" |
|
140 |
- "pmulhw %%mm0, %%mm1 \n\t" |
|
141 |
- "psubsw %%mm6, %%mm1 \n\t" |
|
142 |
-#ifdef HAVE_MMX2 |
|
143 |
- "pminsw %%mm3, %%mm1 \n\t" |
|
144 |
- "pmaxsw %%mm4, %%mm1 \n\t" |
|
145 |
-#else |
|
146 |
- "paddsw %%mm3, %%mm1 \n\t" |
|
147 |
- "psubusw %%mm4, %%mm1 \n\t" |
|
148 |
- "paddsw %%mm5, %%mm1 \n\t" |
|
149 |
-#endif |
|
150 |
- "movq %%mm1, (%8, %%eax) \n\t" |
|
151 |
- "pcmpeqw %%mm7, %%mm1 \n\t" |
|
152 |
- "movq (%7, %%eax), %%mm0 \n\t" |
|
153 |
- "movq %%mm7, (%1, %%eax) \n\t" |
|
154 |
- "pandn %%mm0, %%mm1 \n\t" |
|
155 |
- PMAXW(%%mm1, %%mm2) |
|
156 |
- "addl $8, %%eax \n\t" |
|
157 |
- " js 1b \n\t" |
|
158 |
- "movq %%mm2, %%mm0 \n\t" |
|
159 |
- "psrlq $32, %%mm2 \n\t" |
|
160 |
- PMAXW(%%mm0, %%mm2) |
|
161 |
- "movq %%mm2, %%mm0 \n\t" |
|
162 |
- "psrlq $16, %%mm2 \n\t" |
|
163 |
- PMAXW(%%mm0, %%mm2) |
|
164 |
- "movd %%mm2, %%eax \n\t" |
|
165 |
- "movzbl %%al, %%eax \n\t" |
|
166 |
- : "+a" (last_non_zero_p1) |
|
167 |
- : "r" (block+64), "r" (qmat+64), |
|
168 |
-#ifdef HAVE_MMX2 |
|
169 |
- "m" (maxLevel), "m" (minLevel), "m" (minLevel /* dummy */), "g" (2*i - 128), |
|
170 |
-#else |
|
171 |
- "m" (0x7FFF - maxLevel), "m" (0x7FFF -maxLevel + minLevel), "m" (minLevel), "g" (2*i - 128), |
|
172 |
-#endif |
|
173 |
- "r" (inv_zigzag_direct16+64), "r" (temp_block+64) |
|
174 |
- ); |
|
79 |
+ if(s->out_format == FMT_H263){ |
|
80 |
+ |
|
81 |
+ asm volatile( |
|
82 |
+ "movd %%eax, %%mm3 \n\t" // last_non_zero_p1 |
|
83 |
+ SPREADW(%%mm3) |
|
84 |
+ "pxor %%mm7, %%mm7 \n\t" // 0 |
|
85 |
+ "pxor %%mm4, %%mm4 \n\t" // 0 |
|
86 |
+ "movq (%2), %%mm5 \n\t" // qmat[0] |
|
87 |
+ "pxor %%mm6, %%mm6 \n\t" |
|
88 |
+ "psubw (%3), %%mm6 \n\t" // -bias[0] |
|
89 |
+ "movl $-128, %%eax \n\t" |
|
90 |
+ ".balign 16 \n\t" |
|
91 |
+ "1: \n\t" |
|
92 |
+ "pxor %%mm1, %%mm1 \n\t" // 0 |
|
93 |
+ "movq (%1, %%eax), %%mm0 \n\t" // block[i] |
|
94 |
+ "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00 |
|
95 |
+ "pxor %%mm1, %%mm0 \n\t" |
|
96 |
+ "psubw %%mm1, %%mm0 \n\t" // ABS(block[i]) |
|
97 |
+ "psubusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0] |
|
98 |
+ "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16 |
|
99 |
+ "por %%mm0, %%mm4 \n\t" |
|
100 |
+ "pxor %%mm1, %%mm0 \n\t" |
|
101 |
+ "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) |
|
102 |
+ "movq %%mm0, (%5, %%eax) \n\t" |
|
103 |
+ "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00 |
|
104 |
+ "movq (%4, %%eax), %%mm1 \n\t" |
|
105 |
+ "movq %%mm7, (%1, %%eax) \n\t" // 0 |
|
106 |
+ "pandn %%mm1, %%mm0 \n\t" |
|
107 |
+ PMAXW(%%mm0, %%mm3) |
|
108 |
+ "addl $8, %%eax \n\t" |
|
109 |
+ " js 1b \n\t" |
|
110 |
+ "movq %%mm3, %%mm0 \n\t" |
|
111 |
+ "psrlq $32, %%mm3 \n\t" |
|
112 |
+ PMAXW(%%mm0, %%mm3) |
|
113 |
+ "movq %%mm3, %%mm0 \n\t" |
|
114 |
+ "psrlq $16, %%mm3 \n\t" |
|
115 |
+ PMAXW(%%mm0, %%mm3) |
|
116 |
+ "movd %%mm3, %%eax \n\t" |
|
117 |
+ "movzbl %%al, %%eax \n\t" // last_non_zero_p1 |
|
118 |
+ : "+a" (last_non_zero_p1) |
|
119 |
+ : "r" (block+64), "r" (qmat), "r" (bias), |
|
120 |
+ "r" (inv_zigzag_direct16+64), "r" (temp_block+64) |
|
121 |
+ ); |
|
122 |
+ // note the asm is split cuz gcc doesnt like that many operands ... |
|
123 |
+ asm volatile( |
|
124 |
+ "movd %1, %%mm1 \n\t" // max_qcoeff |
|
125 |
+ SPREADW(%%mm1) |
|
126 |
+ "psubusw %%mm1, %%mm4 \n\t" |
|
127 |
+ "packuswb %%mm4, %%mm4 \n\t" |
|
128 |
+ "movd %%mm4, %0 \n\t" // *overflow |
|
129 |
+ : "=g" (*overflow) |
|
130 |
+ : "g" (s->max_qcoeff) |
|
131 |
+ ); |
|
132 |
+ }else{ // FMT_H263 |
|
133 |
+ asm volatile( |
|
134 |
+ "movd %%eax, %%mm3 \n\t" // last_non_zero_p1 |
|
135 |
+ SPREADW(%%mm3) |
|
136 |
+ "pxor %%mm7, %%mm7 \n\t" // 0 |
|
137 |
+ "pxor %%mm4, %%mm4 \n\t" // 0 |
|
138 |
+ "movl $-128, %%eax \n\t" |
|
139 |
+ ".balign 16 \n\t" |
|
140 |
+ "1: \n\t" |
|
141 |
+ "pxor %%mm1, %%mm1 \n\t" // 0 |
|
142 |
+ "movq (%1, %%eax), %%mm0 \n\t" // block[i] |
|
143 |
+ "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00 |
|
144 |
+ "pxor %%mm1, %%mm0 \n\t" |
|
145 |
+ "psubw %%mm1, %%mm0 \n\t" // ABS(block[i]) |
|
146 |
+ "movq (%3, %%eax), %%mm6 \n\t" // bias[0] |
|
147 |
+ "paddusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0] |
|
148 |
+ "movq (%2, %%eax), %%mm5 \n\t" // qmat[i] |
|
149 |
+ "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16 |
|
150 |
+ "por %%mm0, %%mm4 \n\t" |
|
151 |
+ "pxor %%mm1, %%mm0 \n\t" |
|
152 |
+ "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) |
|
153 |
+ "movq %%mm0, (%5, %%eax) \n\t" |
|
154 |
+ "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00 |
|
155 |
+ "movq (%4, %%eax), %%mm1 \n\t" |
|
156 |
+ "movq %%mm7, (%1, %%eax) \n\t" // 0 |
|
157 |
+ "pandn %%mm1, %%mm0 \n\t" |
|
158 |
+ PMAXW(%%mm0, %%mm3) |
|
159 |
+ "addl $8, %%eax \n\t" |
|
160 |
+ " js 1b \n\t" |
|
161 |
+ "movq %%mm3, %%mm0 \n\t" |
|
162 |
+ "psrlq $32, %%mm3 \n\t" |
|
163 |
+ PMAXW(%%mm0, %%mm3) |
|
164 |
+ "movq %%mm3, %%mm0 \n\t" |
|
165 |
+ "psrlq $16, %%mm3 \n\t" |
|
166 |
+ PMAXW(%%mm0, %%mm3) |
|
167 |
+ "movd %%mm3, %%eax \n\t" |
|
168 |
+ "movzbl %%al, %%eax \n\t" // last_non_zero_p1 |
|
169 |
+ : "+a" (last_non_zero_p1) |
|
170 |
+ : "r" (block+64), "r" (qmat+64), "r" (bias+64), |
|
171 |
+ "r" (inv_zigzag_direct16+64), "r" (temp_block+64) |
|
172 |
+ ); |
|
173 |
+ // note the asm is split cuz gcc doesnt like that many operands ... |
|
174 |
+ asm volatile( |
|
175 |
+ "movd %1, %%mm1 \n\t" // max_qcoeff |
|
176 |
+ SPREADW(%%mm1) |
|
177 |
+ "psubusw %%mm1, %%mm4 \n\t" |
|
178 |
+ "packuswb %%mm4, %%mm4 \n\t" |
|
179 |
+ "movd %%mm4, %0 \n\t" // *overflow |
|
180 |
+ : "=g" (*overflow) |
|
181 |
+ : "g" (s->max_qcoeff) |
|
182 |
+ ); |
|
183 |
+ } |
|
184 |
+ |
|
185 |
+ if(s->mb_intra) temp_block[0]= level; //FIXME move afer permute |
|
175 | 186 |
// last_non_zero_p1=64; |
176 | 187 |
/* permute for IDCT */ |
177 | 188 |
asm volatile( |
178 |
- "movl %0, %%eax \n\t" |
|
189 |
+ "movl %0, %%eax \n\t" |
|
179 | 190 |
"pushl %%ebp \n\t" |
180 | 191 |
"movl %%esp, " MANGLE(esp_temp) "\n\t" |
181 | 192 |
"1: \n\t" |
... | ... |
@@ -203,5 +214,6 @@ static int RENAME(dct_quantize)(MpegEncContext *s, |
203 | 203 |
} |
204 | 204 |
*/ |
205 | 205 |
//block_permute(block); |
206 |
+ |
|
206 | 207 |
return last_non_zero_p1 - 1; |
207 | 208 |
} |
... | ... |
@@ -160,6 +160,9 @@ int mjpeg_init(MpegEncContext *s) |
160 | 160 |
m = malloc(sizeof(MJpegContext)); |
161 | 161 |
if (!m) |
162 | 162 |
return -1; |
163 |
+ |
|
164 |
+ s->min_qcoeff=-1023; |
|
165 |
+ s->max_qcoeff= 1023; |
|
163 | 166 |
|
164 | 167 |
/* build all the huffman tables */ |
165 | 168 |
build_huffman_codes(m->huff_size_dc_luminance, |
... | ... |
@@ -399,8 +399,11 @@ void mpeg1_encode_init(MpegEncContext *s) |
399 | 399 |
} |
400 | 400 |
} |
401 | 401 |
s->mv_penalty= mv_penalty; |
402 |
- |
|
403 | 402 |
s->fcode_tab= fcode_tab; |
403 |
+ s->min_qcoeff=-255; |
|
404 |
+ s->max_qcoeff= 255; |
|
405 |
+ s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x |
|
406 |
+ s->inter_quant_bias= 0; |
|
404 | 407 |
} |
405 | 408 |
|
406 | 409 |
static inline void encode_dc(MpegEncContext *s, int diff, int component) |
... | ... |
@@ -1027,9 +1030,9 @@ static int mpeg2_decode_block_non_intra(MpegEncContext *s, |
1027 | 1027 |
UINT8 *buf_ptr; |
1028 | 1028 |
i = 0; |
1029 | 1029 |
if (n < 4) |
1030 |
- matrix = s->non_intra_matrix; |
|
1030 |
+ matrix = s->inter_matrix; |
|
1031 | 1031 |
else |
1032 |
- matrix = s->chroma_non_intra_matrix; |
|
1032 |
+ matrix = s->chroma_inter_matrix; |
|
1033 | 1033 |
|
1034 | 1034 |
/* special case for the first coef. no need to add a second vlc table */ |
1035 | 1035 |
SAVE_BITS(&s->gb); |
... | ... |
@@ -1183,6 +1186,7 @@ static int mpeg_decode_init(AVCodecContext *avctx) |
1183 | 1183 |
s->buf_ptr = s->buffer; |
1184 | 1184 |
s->mpeg_enc_ctx.picture_number = 0; |
1185 | 1185 |
s->repeat_field = 0; |
1186 |
+ s->mpeg_enc_ctx.codec_id= avctx->codec->id; |
|
1186 | 1187 |
return 0; |
1187 | 1188 |
} |
1188 | 1189 |
|
... | ... |
@@ -1292,8 +1296,8 @@ static void mpeg_decode_quant_matrix_extension(MpegEncContext *s) |
1292 | 1292 |
for(i=0;i<64;i++) { |
1293 | 1293 |
v = get_bits(&s->gb, 8); |
1294 | 1294 |
j = zigzag_direct[i]; |
1295 |
- s->non_intra_matrix[j] = v; |
|
1296 |
- s->chroma_non_intra_matrix[j] = v; |
|
1295 |
+ s->inter_matrix[j] = v; |
|
1296 |
+ s->chroma_inter_matrix[j] = v; |
|
1297 | 1297 |
} |
1298 | 1298 |
} |
1299 | 1299 |
if (get_bits1(&s->gb)) { |
... | ... |
@@ -1307,7 +1311,7 @@ static void mpeg_decode_quant_matrix_extension(MpegEncContext *s) |
1307 | 1307 |
for(i=0;i<64;i++) { |
1308 | 1308 |
v = get_bits(&s->gb, 8); |
1309 | 1309 |
j = zigzag_direct[i]; |
1310 |
- s->chroma_non_intra_matrix[j] = v; |
|
1310 |
+ s->chroma_inter_matrix[j] = v; |
|
1311 | 1311 |
} |
1312 | 1312 |
} |
1313 | 1313 |
} |
... | ... |
@@ -1386,7 +1390,6 @@ static int mpeg_decode_slice(AVCodecContext *avctx, |
1386 | 1386 |
s->mb_x = -1; |
1387 | 1387 |
s->mb_y = start_code; |
1388 | 1388 |
s->mb_incr = 0; |
1389 |
- |
|
1390 | 1389 |
/* start frame decoding */ |
1391 | 1390 |
if (s->first_slice) { |
1392 | 1391 |
s->first_slice = 0; |
... | ... |
@@ -1526,20 +1529,20 @@ static int mpeg1_decode_sequence(AVCodecContext *avctx, |
1526 | 1526 |
for(i=0;i<64;i++) { |
1527 | 1527 |
v = get_bits(&s->gb, 8); |
1528 | 1528 |
j = zigzag_direct[i]; |
1529 |
- s->non_intra_matrix[j] = v; |
|
1530 |
- s->chroma_non_intra_matrix[j] = v; |
|
1529 |
+ s->inter_matrix[j] = v; |
|
1530 |
+ s->chroma_inter_matrix[j] = v; |
|
1531 | 1531 |
} |
1532 | 1532 |
#ifdef DEBUG |
1533 | 1533 |
dprintf("non intra matrix present\n"); |
1534 | 1534 |
for(i=0;i<64;i++) |
1535 |
- dprintf(" %d", s->non_intra_matrix[zigzag_direct[i]]); |
|
1535 |
+ dprintf(" %d", s->inter_matrix[zigzag_direct[i]]); |
|
1536 | 1536 |
printf("\n"); |
1537 | 1537 |
#endif |
1538 | 1538 |
} else { |
1539 | 1539 |
for(i=0;i<64;i++) { |
1540 | 1540 |
v = default_non_intra_matrix[i]; |
1541 |
- s->non_intra_matrix[i] = v; |
|
1542 |
- s->chroma_non_intra_matrix[i] = v; |
|
1541 |
+ s->inter_matrix[i] = v; |
|
1542 |
+ s->chroma_inter_matrix[i] = v; |
|
1543 | 1543 |
} |
1544 | 1544 |
} |
1545 | 1545 |
|
... | ... |
@@ -1566,7 +1569,7 @@ static int mpeg_decode_frame(AVCodecContext *avctx, |
1566 | 1566 |
dprintf("fill_buffer\n"); |
1567 | 1567 |
|
1568 | 1568 |
*data_size = 0; |
1569 |
- |
|
1569 |
+ |
|
1570 | 1570 |
/* special case for last picture */ |
1571 | 1571 |
if (buf_size == 0) { |
1572 | 1572 |
if (s2->picture_number > 0) { |
... | ... |
@@ -1591,7 +1594,7 @@ static int mpeg_decode_frame(AVCodecContext *avctx, |
1591 | 1591 |
*data_size = sizeof(AVPicture); |
1592 | 1592 |
goto the_end; |
1593 | 1593 |
} |
1594 |
- |
|
1594 |
+ |
|
1595 | 1595 |
while (buf_ptr < buf_end) { |
1596 | 1596 |
buf_start = buf_ptr; |
1597 | 1597 |
/* find start next code */ |
... | ... |
@@ -38,9 +38,9 @@ static void dct_unquantize_mpeg2_c(MpegEncContext *s, |
38 | 38 |
static void dct_unquantize_h263_c(MpegEncContext *s, |
39 | 39 |
DCTELEM *block, int n, int qscale); |
40 | 40 |
static void draw_edges_c(UINT8 *buf, int wrap, int width, int height, int w); |
41 |
-static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale); |
|
41 |
+static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow); |
|
42 | 42 |
|
43 |
-int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale)= dct_quantize_c; |
|
43 |
+int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow)= dct_quantize_c; |
|
44 | 44 |
void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w)= draw_edges_c; |
45 | 45 |
|
46 | 46 |
#define EDGE_WIDTH 16 |
... | ... |
@@ -78,29 +78,38 @@ extern UINT8 zigzag_end[64]; |
78 | 78 |
/* default motion estimation */ |
79 | 79 |
int motion_estimation_method = ME_EPZS; |
80 | 80 |
|
81 |
-static void convert_matrix(int *qmat, UINT16 *qmat16, const UINT16 *quant_matrix, int qscale) |
|
81 |
+static void convert_matrix(int (*qmat)[64], uint16_t (*qmat16)[64], uint16_t (*qmat16_bias)[64], |
|
82 |
+ const UINT16 *quant_matrix, int bias) |
|
82 | 83 |
{ |
83 |
- int i; |
|
84 |
- |
|
85 |
- if (av_fdct == jpeg_fdct_ifast) { |
|
86 |
- for(i=0;i<64;i++) { |
|
87 |
- /* 16 <= qscale * quant_matrix[i] <= 7905 */ |
|
88 |
- /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
|
89 |
- /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ |
|
90 |
- /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ |
|
91 |
- |
|
92 |
- qmat[block_permute_op(i)] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
|
93 |
- (aanscales[i] * qscale * quant_matrix[block_permute_op(i)])); |
|
94 |
- } |
|
95 |
- } else { |
|
96 |
- for(i=0;i<64;i++) { |
|
97 |
- /* We can safely suppose that 16 <= quant_matrix[i] <= 255 |
|
98 |
- So 16 <= qscale * quant_matrix[i] <= 7905 |
|
99 |
- so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 |
|
100 |
- so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 |
|
101 |
- */ |
|
102 |
- qmat[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
|
103 |
- qmat16[i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); |
|
84 |
+ int qscale; |
|
85 |
+ |
|
86 |
+ for(qscale=1; qscale<32; qscale++){ |
|
87 |
+ int i; |
|
88 |
+ if (av_fdct == jpeg_fdct_ifast) { |
|
89 |
+ for(i=0;i<64;i++) { |
|
90 |
+ const int j= block_permute_op(i); |
|
91 |
+ /* 16 <= qscale * quant_matrix[i] <= 7905 */ |
|
92 |
+ /* 19952 <= aanscales[i] * qscale * quant_matrix[i] <= 249205026 */ |
|
93 |
+ /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */ |
|
94 |
+ /* 3444240 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= 275 */ |
|
95 |
+ |
|
96 |
+ qmat[qscale][j] = (int)((UINT64_C(1) << (QMAT_SHIFT + 11)) / |
|
97 |
+ (aanscales[i] * qscale * quant_matrix[j])); |
|
98 |
+ } |
|
99 |
+ } else { |
|
100 |
+ for(i=0;i<64;i++) { |
|
101 |
+ /* We can safely suppose that 16 <= quant_matrix[i] <= 255 |
|
102 |
+ So 16 <= qscale * quant_matrix[i] <= 7905 |
|
103 |
+ so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905 |
|
104 |
+ so 32768 >= (1<<19) / (qscale * quant_matrix[i]) >= 67 |
|
105 |
+ */ |
|
106 |
+ qmat [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]); |
|
107 |
+ qmat16[qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[block_permute_op(i)]); |
|
108 |
+ |
|
109 |
+ if(qmat16[qscale][i]==0 || qmat16[qscale][i]==128*256) qmat16[qscale][i]=128*256-1; |
|
110 |
+ |
|
111 |
+ qmat16_bias[qscale][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][i]); |
|
112 |
+ } |
|
104 | 113 |
} |
105 | 114 |
} |
106 | 115 |
} |
... | ... |
@@ -388,7 +397,8 @@ int MPV_encode_init(AVCodecContext *avctx) |
388 | 388 |
s->max_b_frames= avctx->max_b_frames; |
389 | 389 |
s->rc_strategy= avctx->rc_strategy; |
390 | 390 |
s->b_frame_strategy= avctx->b_frame_strategy; |
391 |
- |
|
391 |
+ s->codec_id= avctx->codec->id; |
|
392 |
+ |
|
392 | 393 |
if (s->gop_size <= 1) { |
393 | 394 |
s->intra_only = 1; |
394 | 395 |
s->gop_size = 12; |
... | ... |
@@ -523,8 +533,21 @@ int MPV_encode_init(AVCodecContext *avctx) |
523 | 523 |
|
524 | 524 |
/* init default q matrix */ |
525 | 525 |
for(i=0;i<64;i++) { |
526 |
- s->intra_matrix[i] = default_intra_matrix[i]; |
|
527 |
- s->non_intra_matrix[i] = default_non_intra_matrix[i]; |
|
526 |
+ if(s->out_format == FMT_H263) |
|
527 |
+ s->intra_matrix[i] = default_non_intra_matrix[i]; |
|
528 |
+ else |
|
529 |
+ s->intra_matrix[i] = default_intra_matrix[i]; |
|
530 |
+ |
|
531 |
+ s->inter_matrix[i] = default_non_intra_matrix[i]; |
|
532 |
+ } |
|
533 |
+ |
|
534 |
+ /* precompute matrix */ |
|
535 |
+ /* for mjpeg, we do include qscale in the matrix */ |
|
536 |
+ if (s->out_format != FMT_MJPEG) { |
|
537 |
+ convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->q_intra_matrix16_bias, |
|
538 |
+ s->intra_matrix, s->intra_quant_bias); |
|
539 |
+ convert_matrix(s->q_inter_matrix, s->q_inter_matrix16, s->q_inter_matrix16_bias, |
|
540 |
+ s->inter_matrix, s->inter_quant_bias); |
|
528 | 541 |
} |
529 | 542 |
|
530 | 543 |
if(ff_rate_control_init(s) < 0) |
... | ... |
@@ -1307,6 +1330,21 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]) |
1307 | 1307 |
emms_c(); //FIXME remove |
1308 | 1308 |
} |
1309 | 1309 |
|
1310 |
+static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block, int last_index) |
|
1311 |
+{ |
|
1312 |
+ int i; |
|
1313 |
+ const int maxlevel= s->max_qcoeff; |
|
1314 |
+ const int minlevel= s->min_qcoeff; |
|
1315 |
+ |
|
1316 |
+ for(i=0; i<=last_index; i++){ |
|
1317 |
+ const int j = zigzag_direct[i]; |
|
1318 |
+ int level = block[j]; |
|
1319 |
+ |
|
1320 |
+ if (level>maxlevel) level=maxlevel; |
|
1321 |
+ else if(level<minlevel) level=minlevel; |
|
1322 |
+ block[j]= level; |
|
1323 |
+ } |
|
1324 |
+} |
|
1310 | 1325 |
|
1311 | 1326 |
static void encode_mb(MpegEncContext *s, int motion_x, int motion_y) |
1312 | 1327 |
{ |
... | ... |
@@ -1407,8 +1445,19 @@ static void encode_mb(MpegEncContext *s, int motion_x, int motion_y) |
1407 | 1407 |
s->y_dc_scale = 8; |
1408 | 1408 |
s->c_dc_scale = 8; |
1409 | 1409 |
} |
1410 |
- for(i=0;i<6;i++) { |
|
1411 |
- s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale); |
|
1410 |
+ if(s->out_format==FMT_MJPEG){ |
|
1411 |
+ for(i=0;i<6;i++) { |
|
1412 |
+ int overflow; |
|
1413 |
+ s->block_last_index[i] = dct_quantize(s, s->block[i], i, 8, &overflow); |
|
1414 |
+ if(overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]); |
|
1415 |
+ } |
|
1416 |
+ }else{ |
|
1417 |
+ for(i=0;i<6;i++) { |
|
1418 |
+ int overflow; |
|
1419 |
+ s->block_last_index[i] = dct_quantize(s, s->block[i], i, s->qscale, &overflow); |
|
1420 |
+ // FIXME we could decide to change to quantizer instead of clipping |
|
1421 |
+ if(overflow) clip_coeffs(s, s->block[i], s->block_last_index[i]); |
|
1422 |
+ } |
|
1412 | 1423 |
} |
1413 | 1424 |
|
1414 | 1425 |
/* huffman encode */ |
... | ... |
@@ -1596,17 +1645,13 @@ static void encode_picture(MpegEncContext *s, int picture_number) |
1596 | 1596 |
else if (!s->fixed_qscale) |
1597 | 1597 |
s->qscale = ff_rate_estimate_qscale(s); |
1598 | 1598 |
|
1599 |
- |
|
1600 |
- /* precompute matrix */ |
|
1601 | 1599 |
if (s->out_format == FMT_MJPEG) { |
1602 | 1600 |
/* for mjpeg, we do include qscale in the matrix */ |
1603 | 1601 |
s->intra_matrix[0] = default_intra_matrix[0]; |
1604 | 1602 |
for(i=1;i<64;i++) |
1605 | 1603 |
s->intra_matrix[i] = (default_intra_matrix[i] * s->qscale) >> 3; |
1606 |
- convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, 8); |
|
1607 |
- } else { |
|
1608 |
- convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, s->intra_matrix, s->qscale); |
|
1609 |
- convert_matrix(s->q_non_intra_matrix, s->q_non_intra_matrix16, s->non_intra_matrix, s->qscale); |
|
1604 |
+ convert_matrix(s->q_intra_matrix, s->q_intra_matrix16, |
|
1605 |
+ s->q_intra_matrix16_bias, s->intra_matrix, s->intra_quant_bias); |
|
1610 | 1606 |
} |
1611 | 1607 |
|
1612 | 1608 |
s->last_bits= get_bit_count(&s->pb); |
... | ... |
@@ -1957,29 +2002,13 @@ static void encode_picture(MpegEncContext *s, int picture_number) |
1957 | 1957 |
|
1958 | 1958 |
static int dct_quantize_c(MpegEncContext *s, |
1959 | 1959 |
DCTELEM *block, int n, |
1960 |
- int qscale) |
|
1960 |
+ int qscale, int *overflow) |
|
1961 | 1961 |
{ |
1962 | 1962 |
int i, j, level, last_non_zero, q; |
1963 | 1963 |
const int *qmat; |
1964 |
- int minLevel, maxLevel; |
|
1965 |
- |
|
1966 |
- if(s->avctx!=NULL && s->avctx->codec->id==CODEC_ID_MPEG4){ |
|
1967 |
- /* mpeg4 */ |
|
1968 |
- minLevel= -2048; |
|
1969 |
- maxLevel= 2047; |
|
1970 |
- }else if(s->out_format==FMT_MPEG1){ |
|
1971 |
- /* mpeg1 */ |
|
1972 |
- minLevel= -255; |
|
1973 |
- maxLevel= 255; |
|
1974 |
- }else if(s->out_format==FMT_MJPEG){ |
|
1975 |
- /* (m)jpeg */ |
|
1976 |
- minLevel= -1023; |
|
1977 |
- maxLevel= 1023; |
|
1978 |
- }else{ |
|
1979 |
- /* h263 / msmpeg4 */ |
|
1980 |
- minLevel= -128; |
|
1981 |
- maxLevel= 127; |
|
1982 |
- } |
|
1964 |
+ int bias; |
|
1965 |
+ int max=0; |
|
1966 |
+ unsigned int threshold1, threshold2; |
|
1983 | 1967 |
|
1984 | 1968 |
av_fdct (block); |
1985 | 1969 |
|
... | ... |
@@ -1998,71 +2027,40 @@ static int dct_quantize_c(MpegEncContext *s, |
1998 | 1998 |
block[0] = (block[0] + (q >> 1)) / q; |
1999 | 1999 |
i = 1; |
2000 | 2000 |
last_non_zero = 0; |
2001 |
- if (s->out_format == FMT_H263) { |
|
2002 |
- qmat = s->q_non_intra_matrix; |
|
2003 |
- } else { |
|
2004 |
- qmat = s->q_intra_matrix; |
|
2005 |
- } |
|
2001 |
+ qmat = s->q_intra_matrix[qscale]; |
|
2002 |
+ bias= s->intra_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT); |
|
2006 | 2003 |
} else { |
2007 | 2004 |
i = 0; |
2008 | 2005 |
last_non_zero = -1; |
2009 |
- qmat = s->q_non_intra_matrix; |
|
2006 |
+ qmat = s->q_inter_matrix[qscale]; |
|
2007 |
+ bias= s->inter_quant_bias<<(QMAT_SHIFT - 3 - QUANT_BIAS_SHIFT); |
|
2010 | 2008 |
} |
2009 |
+ threshold1= (1<<(QMAT_SHIFT - 3)) - bias - 1; |
|
2010 |
+ threshold2= threshold1<<1; |
|
2011 | 2011 |
|
2012 | 2012 |
for(;i<64;i++) { |
2013 | 2013 |
j = zigzag_direct[i]; |
2014 | 2014 |
level = block[j]; |
2015 | 2015 |
level = level * qmat[j]; |
2016 |
-#ifdef PARANOID |
|
2017 |
- { |
|
2018 |
- static int count = 0; |
|
2019 |
- int level1, level2, qmat1; |
|
2020 |
- double val; |
|
2021 |
- if (qmat == s->q_non_intra_matrix) { |
|
2022 |
- qmat1 = default_non_intra_matrix[j] * s->qscale; |
|
2023 |
- } else { |
|
2024 |
- qmat1 = default_intra_matrix[j] * s->qscale; |
|
2025 |
- } |
|
2026 |
- if (av_fdct != jpeg_fdct_ifast) |
|
2027 |
- val = ((double)block[j] * 8.0) / (double)qmat1; |
|
2028 |
- else |
|
2029 |
- val = ((double)block[j] * 8.0 * 2048.0) / |
|
2030 |
- ((double)qmat1 * aanscales[j]); |
|
2031 |
- level1 = (int)val; |
|
2032 |
- level2 = level / (1 << (QMAT_SHIFT - 3)); |
|
2033 |
- if (level1 != level2) { |
|
2034 |
- fprintf(stderr, "%d: quant error qlevel=%d wanted=%d level=%d qmat1=%d qmat=%d wantedf=%0.6f\n", |
|
2035 |
- count, level2, level1, block[j], qmat1, qmat[j], |
|
2036 |
- val); |
|
2037 |
- count++; |
|
2038 |
- } |
|
2039 | 2016 |
|
2040 |
- } |
|
2041 |
-#endif |
|
2042 |
- /* XXX: slight error for the low range. Test should be equivalent to |
|
2043 |
- (level <= -(1 << (QMAT_SHIFT - 3)) || level >= (1 << |
|
2044 |
- (QMAT_SHIFT - 3))) |
|
2045 |
- */ |
|
2046 |
- if (((level << (31 - (QMAT_SHIFT - 3))) >> (31 - (QMAT_SHIFT - 3))) != |
|
2047 |
- level) { |
|
2048 |
- level = level / (1 << (QMAT_SHIFT - 3)); |
|
2049 |
- /* XXX: currently, this code is not optimal. the range should be: |
|
2050 |
- mpeg1: -255..255 |
|
2051 |
- mpeg2: -2048..2047 |
|
2052 |
- h263: -128..127 |
|
2053 |
- mpeg4: -2048..2047 |
|
2054 |
- */ |
|
2055 |
- if (level > maxLevel) |
|
2056 |
- level = maxLevel; |
|
2057 |
- else if (level < minLevel) |
|
2058 |
- level = minLevel; |
|
2059 |
- |
|
2060 |
- block[j] = level; |
|
2017 |
+// if( bias+level >= (1<<(QMAT_SHIFT - 3)) |
|
2018 |
+// || bias-level >= (1<<(QMAT_SHIFT - 3))){ |
|
2019 |
+ if(((unsigned)(level+threshold1))>threshold2){ |
|
2020 |
+ if(level>0){ |
|
2021 |
+ level= (bias + level)>>(QMAT_SHIFT - 3); |
|
2022 |
+ block[j]= level; |
|
2023 |
+ }else{ |
|
2024 |
+ level= (bias - level)>>(QMAT_SHIFT - 3); |
|
2025 |
+ block[j]= -level; |
|
2026 |
+ } |
|
2027 |
+ max |=level; |
|
2061 | 2028 |
last_non_zero = i; |
2062 |
- } else { |
|
2063 |
- block[j] = 0; |
|
2029 |
+ }else{ |
|
2030 |
+ block[j]=0; |
|
2064 | 2031 |
} |
2065 | 2032 |
} |
2033 |
+ *overflow= s->max_qcoeff < max; //overflow might have happend |
|
2034 |
+ |
|
2066 | 2035 |
return last_non_zero; |
2067 | 2036 |
} |
2068 | 2037 |
|
... | ... |
@@ -2104,7 +2102,7 @@ static void dct_unquantize_mpeg1_c(MpegEncContext *s, |
2104 | 2104 |
} |
2105 | 2105 |
} else { |
2106 | 2106 |
i = 0; |
2107 |
- quant_matrix = s->non_intra_matrix; |
|
2107 |
+ quant_matrix = s->inter_matrix; |
|
2108 | 2108 |
for(;i<nCoeffs;i++) { |
2109 | 2109 |
int j= zigzag_direct[i]; |
2110 | 2110 |
level = block[j]; |
... | ... |
@@ -2166,7 +2164,7 @@ static void dct_unquantize_mpeg2_c(MpegEncContext *s, |
2166 | 2166 |
} else { |
2167 | 2167 |
int sum=-1; |
2168 | 2168 |
i = 0; |
2169 |
- quant_matrix = s->non_intra_matrix; |
|
2169 |
+ quant_matrix = s->inter_matrix; |
|
2170 | 2170 |
for(;i<nCoeffs;i++) { |
2171 | 2171 |
int j= zigzag_direct[i]; |
2172 | 2172 |
level = block[j]; |
... | ... |
@@ -83,11 +83,15 @@ typedef struct MpegEncContext { |
83 | 83 |
int bit_rate; /* wanted bit rate */ |
84 | 84 |
int bit_rate_tolerance; /* amount of +- bits (>0)*/ |
85 | 85 |
enum OutputFormat out_format; /* output format */ |
86 |
+ int h263_pred; /* use mpeg4/h263 ac/dc predictions */ |
|
87 |
+ |
|
88 |
+/* the following codec id fields are deprecated in favor of codec_id */ |
|
86 | 89 |
int h263_plus; /* h263 plus headers */ |
87 | 90 |
int h263_rv10; /* use RV10 variation for H263 */ |
88 |
- int h263_pred; /* use mpeg4/h263 ac/dc predictions */ |
|
89 |
- int h263_msmpeg4; /* generate MSMPEG4 compatible stream */ |
|
91 |
+ int h263_msmpeg4; /* generate MSMPEG4 compatible stream (deprecated, use msmpeg4_version instead)*/ |
|
90 | 92 |
int h263_intel; /* use I263 intel h263 header */ |
93 |
+ |
|
94 |
+ int codec_id; /* see CODEC_ID_xxx */ |
|
91 | 95 |
int fixed_qscale; /* fixed qscale if non zero */ |
92 | 96 |
float qcompress; /* amount of qscale change between easy & hard scenes (0.0-1.0) */ |
93 | 97 |
float qblur; /* amount of qscale smoothing over time (0.0-1.0) */ |
... | ... |
@@ -213,14 +217,21 @@ typedef struct MpegEncContext { |
213 | 213 |
/* matrix transmitted in the bitstream */ |
214 | 214 |
UINT16 intra_matrix[64]; |
215 | 215 |
UINT16 chroma_intra_matrix[64]; |
216 |
- UINT16 non_intra_matrix[64]; |
|
217 |
- UINT16 chroma_non_intra_matrix[64]; |
|
216 |
+ UINT16 inter_matrix[64]; |
|
217 |
+ UINT16 chroma_inter_matrix[64]; |
|
218 |
+#define QUANT_BIAS_SHIFT 4 |
|
219 |
+ int intra_quant_bias; /* bias for the quantizer */ |
|
220 |
+ int inter_quant_bias; /* bias for the quantizer */ |
|
221 |
+ int min_qcoeff; /* minimum encodable coefficient */ |
|
222 |
+ int max_qcoeff; /* maximum encodable coefficient */ |
|
218 | 223 |
/* precomputed matrix (combine qscale and DCT renorm) */ |
219 |
- int q_intra_matrix[64]; |
|
220 |
- int q_non_intra_matrix[64]; |
|
224 |
+ int q_intra_matrix[32][64]; |
|
225 |
+ int q_inter_matrix[32][64]; |
|
221 | 226 |
/* identical to the above but for MMX & these are not permutated */ |
222 |
- UINT16 __align8 q_intra_matrix16[64]; |
|
223 |
- UINT16 __align8 q_non_intra_matrix16[64]; |
|
227 |
+ UINT16 __align8 q_intra_matrix16[32][64]; |
|
228 |
+ UINT16 __align8 q_inter_matrix16[32][64]; |
|
229 |
+ UINT16 __align8 q_intra_matrix16_bias[32][64]; |
|
230 |
+ UINT16 __align8 q_inter_matrix16_bias[32][64]; |
|
224 | 231 |
int block_last_index[6]; /* last non zero coefficient in block */ |
225 | 232 |
|
226 | 233 |
void *opaque; /* private data for the user */ |
... | ... |
@@ -328,7 +339,7 @@ typedef struct MpegEncContext { |
328 | 328 |
int first_slice_line; /* used in mpeg4 too to handle resync markers */ |
329 | 329 |
int flipflop_rounding; |
330 | 330 |
int bitrate; |
331 |
- int msmpeg4_version; /* 1=mp41, 2=mp42, 3=mp43/divx3 */ |
|
331 |
+ int msmpeg4_version; /* 0=not msmpeg4, 1=mp41, 2=mp42, 3=mp43/divx3 */ |
|
332 | 332 |
/* decompression specific */ |
333 | 333 |
GetBitContext gb; |
334 | 334 |
|
... | ... |
@@ -386,6 +397,8 @@ void MPV_frame_end(MpegEncContext *s); |
386 | 386 |
#ifdef HAVE_MMX |
387 | 387 |
void MPV_common_init_mmx(MpegEncContext *s); |
388 | 388 |
#endif |
389 |
+int (*dct_quantize)(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow); |
|
390 |
+void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w); |
|
389 | 391 |
|
390 | 392 |
/* motion_est.c */ |
391 | 393 |
void ff_estimate_p_frame_motion(MpegEncContext * s, |
... | ... |
@@ -340,6 +340,7 @@ static int rv10_decode_init(AVCodecContext *avctx) |
340 | 340 |
int i; |
341 | 341 |
static int done; |
342 | 342 |
|
343 |
+// s->avctx= avctx; |
|
343 | 344 |
s->out_format = FMT_H263; |
344 | 345 |
|
345 | 346 |
s->width = avctx->width; |
... | ... |
@@ -351,11 +352,6 @@ static int rv10_decode_init(AVCodecContext *avctx) |
351 | 351 |
if (MPV_common_init(s) < 0) |
352 | 352 |
return -1; |
353 | 353 |
|
354 |
- /* XXX: suppress this matrix init, only needed because using mpeg1 |
|
355 |
- dequantize in mmx case */ |
|
356 |
- for(i=0;i<64;i++) |
|
357 |
- s->non_intra_matrix[i] = default_non_intra_matrix[i]; |
|
358 |
- |
|
359 | 354 |
h263_decode_init_vlc(s); |
360 | 355 |
|
361 | 356 |
/* init rv vlc */ |