2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * Context Adaptive Binary Arithmetic Coder.
27 #ifndef AVCODEC_CABAC_H
28 #define AVCODEC_CABAC_H
30 #include "bitstream.h"
34 #include "libavutil/x86_cpu.h"
37 #define CABAC_MASK ((1<<CABAC_BITS)-1)
38 #define BRANCHLESS_CABAC_DECODER 1
39 //#define ARCH_X86_DISABLED 1
41 typedef struct CABACContext{
44 int outstanding_count;
48 const uint8_t *bytestream_start;
49 const uint8_t *bytestream;
50 const uint8_t *bytestream_end;
54 extern uint8_t ff_h264_mlps_state[4*64];
55 extern uint8_t ff_h264_lps_range[4*2*64]; ///< rangeTabLPS
56 extern uint8_t ff_h264_mps_state[2*64]; ///< transIdxMPS
57 extern uint8_t ff_h264_lps_state[2*64]; ///< transIdxLPS
58 extern const uint8_t ff_h264_norm_shift[512];
61 void ff_init_cabac_encoder(CABACContext *c, uint8_t *buf, int buf_size);
62 void ff_init_cabac_decoder(CABACContext *c, const uint8_t *buf, int buf_size);
63 void ff_init_cabac_states(CABACContext *c);
66 static inline void put_cabac_bit(CABACContext *c, int b){
67 put_bits(&c->pb, 1, b);
68 for(;c->outstanding_count; c->outstanding_count--){
69 put_bits(&c->pb, 1, 1-b);
73 static inline void renorm_cabac_encoder(CABACContext *c){
74 while(c->range < 0x100){
78 }else if(c->low<0x200){
79 c->outstanding_count++;
92 static void put_cabac(CABACContext *c, uint8_t * const state, int bit){
93 int RangeLPS= ff_h264_lps_range[2*(c->range&0xC0) + *state];
95 if(bit == ((*state)&1)){
97 *state= ff_h264_mps_state[*state];
99 c->low += c->range - RangeLPS;
101 *state= ff_h264_lps_state[*state];
104 renorm_cabac_encoder(c);
111 static void put_cabac_static(CABACContext *c, int RangeLPS, int bit){
112 assert(c->range > RangeLPS);
115 c->range -= RangeLPS;
117 c->low += c->range - RangeLPS;
121 renorm_cabac_encoder(c);
129 * @param bit 0 -> write zero bit, !=0 write one bit
131 static void put_cabac_bypass(CABACContext *c, int bit){
140 }else if(c->low<0x400){
141 c->outstanding_count++;
155 * @return the number of bytes written
157 static int put_cabac_terminate(CABACContext *c, int bit){
161 renorm_cabac_encoder(c);
166 renorm_cabac_encoder(c);
168 assert(c->low <= 0x1FF);
169 put_cabac_bit(c, c->low>>9);
170 put_bits(&c->pb, 2, ((c->low>>7)&3)|1);
172 flush_put_bits(&c->pb); //FIXME FIXME FIXME XXX wrong
179 return (put_bits_count(&c->pb)+7)>>3;
183 * put (truncated) unary binarization.
185 static void put_cabac_u(CABACContext *c, uint8_t * state, int v, int max, int max_index, int truncated){
192 put_cabac(c, state, 1);
193 if(i < max_index) state++;
195 if(truncated==0 || v<max)
196 put_cabac(c, state, 0);
200 put_cabac(c, state+i, 1);
202 if(truncated==0 || v<max)
203 put_cabac(c, state+i, 0);
205 for(i=0; i<=max_index; i++){
206 put_cabac(c, state+i, 1);
209 put_cabac(c, state+max_index, 1);
211 if(truncated==0 || v<max)
212 put_cabac(c, state+max_index, 0);
218 * put unary exp golomb k-th order binarization.
220 static void put_cabac_ueg(CABACContext *c, uint8_t * state, int v, int max, int is_signed, int k, int max_index){
224 put_cabac(c, state, 0);
226 const int sign= v < 0;
228 if(is_signed) v= FFABS(v);
232 put_cabac(c, state, 1);
233 if(i < max_index) state++;
236 put_cabac(c, state, 0);
240 for(i=0; i<max; i++){
241 put_cabac(c, state, 1);
242 if(i < max_index) state++;
246 while(v >= m){ //FIXME optimize
247 put_cabac_bypass(c, 1);
251 put_cabac_bypass(c, 0);
253 put_cabac_bypass(c, v&m);
258 put_cabac_bypass(c, sign);
263 static void refill(CABACContext *c){
265 c->low+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
267 c->low+= c->bytestream[0]<<1;
269 c->low -= CABAC_MASK;
270 c->bytestream+= CABAC_BITS/8;
273 #if ! ( ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS) )
274 static void refill2(CABACContext *c){
277 x= c->low ^ (c->low-1);
278 i= 7 - ff_h264_norm_shift[x>>(CABAC_BITS-1)];
283 x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
285 x+= c->bytestream[0]<<1;
289 c->bytestream+= CABAC_BITS/8;
293 static inline void renorm_cabac_decoder(CABACContext *c){
294 while(c->range < 0x100){
297 if(!(c->low & CABAC_MASK))
302 static inline void renorm_cabac_decoder_once(CABACContext *c){
303 #ifdef ARCH_X86_DISABLED
308 "lea -0x100(%0), %2 \n\t"
309 "shr $31, %2 \n\t" //FIXME 31->63 for x86-64
312 : "+r"(c->range), "+r"(c->low), "+c"(temp)
317 "cmp $0x100, %0 \n\t"
318 "setb %%cl \n\t" //FIXME 31->63 for x86-64
321 : "+r"(c->range), "+r"(c->low), "+c"(temp)
327 "lea -0x100(%0), %%eax \n\t"
334 : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
340 "cmp $0x100, %0 \n\t"
341 "sbb %%edx, %%edx \n\t"
347 : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
353 "cmp $0x100, %0 \n\t"
354 "lea (%0, %0), %%eax \n\t"
355 "lea (%1, %1), %%edx \n\t"
356 "cmovb %%eax, %0 \n\t"
357 "cmovb %%edx, %1 \n\t"
358 : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
363 int shift= (uint32_t)(c->range - 0x100)>>31;
367 if(!(c->low & CABAC_MASK))
371 static av_always_inline int get_cabac_inline(CABACContext *c, uint8_t * const state){
372 //FIXME gcc generates duplicate load/stores for c->low and c->range
376 #define BYTESTART "16"
380 #define BYTESTART "12"
384 #if ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS)
387 #ifndef BRANCHLESS_CABAC_DECODER
389 "movzbl (%1), %0 \n\t"
390 "movl "RANGE "(%2), %%ebx \n\t"
391 "movl "RANGE "(%2), %%edx \n\t"
392 "andl $0xC0, %%ebx \n\t"
393 "movzbl "MANGLE(ff_h264_lps_range)"(%0, %%ebx, 2), %%esi\n\t"
394 "movl "LOW "(%2), %%ebx \n\t"
395 //eax:state ebx:low, edx:range, esi:RangeLPS
396 "subl %%esi, %%edx \n\t"
397 "movl %%edx, %%ecx \n\t"
398 "shll $17, %%ecx \n\t"
399 "cmpl %%ecx, %%ebx \n\t"
403 //athlon:4067 P3:4110
404 "lea -0x100(%%edx), %%ecx \n\t"
405 "shr $31, %%ecx \n\t"
406 "shl %%cl, %%edx \n\t"
407 "shl %%cl, %%ebx \n\t"
409 //athlon:4057 P3:4130
410 "cmp $0x100, %%edx \n\t" //FIXME avoidable
412 "shl %%cl, %%edx \n\t"
413 "shl %%cl, %%ebx \n\t"
415 "movzbl "MANGLE(ff_h264_mps_state)"(%0), %%ecx \n\t"
416 "movb %%cl, (%1) \n\t"
417 //eax:state ebx:low, edx:range, esi:RangeLPS
418 "test %%bx, %%bx \n\t"
420 "mov "BYTE "(%2), %%"REG_S" \n\t"
421 "subl $0xFFFF, %%ebx \n\t"
422 "movzwl (%%"REG_S"), %%ecx \n\t"
424 "shrl $15, %%ecx \n\t"
425 "add $2, %%"REG_S" \n\t"
426 "addl %%ecx, %%ebx \n\t"
427 "mov %%"REG_S", "BYTE "(%2) \n\t"
430 //eax:state ebx:low, edx:range, esi:RangeLPS
431 "subl %%ecx, %%ebx \n\t"
432 "movl %%esi, %%edx \n\t"
433 "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx \n\t"
434 "shll %%cl, %%ebx \n\t"
435 "shll %%cl, %%edx \n\t"
436 "movzbl "MANGLE(ff_h264_lps_state)"(%0), %%ecx \n\t"
437 "movb %%cl, (%1) \n\t"
439 "test %%bx, %%bx \n\t"
442 "mov "BYTE "(%2), %%"REG_c" \n\t"
443 "movzwl (%%"REG_c"), %%esi \n\t"
445 "shrl $15, %%esi \n\t"
446 "subl $0xFFFF, %%esi \n\t"
447 "add $2, %%"REG_c" \n\t"
448 "mov %%"REG_c", "BYTE "(%2) \n\t"
450 "leal -1(%%ebx), %%ecx \n\t"
451 "xorl %%ebx, %%ecx \n\t"
452 "shrl $15, %%ecx \n\t"
453 "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"
457 "shll %%cl , %%esi \n\t"
458 "addl %%esi, %%ebx \n\t"
460 "movl %%edx, "RANGE "(%2) \n\t"
461 "movl %%ebx, "LOW "(%2) \n\t"
462 :"=&a"(bit) //FIXME this is fragile gcc either runs out of registers or miscompiles it (for example if "+a"(bit) or "+m"(*state) is used
464 : "%"REG_c, "%ebx", "%edx", "%"REG_S, "memory"
467 #else /* BRANCHLESS_CABAC_DECODER */
471 #define BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
472 "mov "tmp" , %%ecx \n\t"\
473 "shl $17 , "tmp" \n\t"\
474 "cmp "low" , "tmp" \n\t"\
475 "cmova %%ecx , "range" \n\t"\
476 "sbb %%ecx , %%ecx \n\t"\
477 "and %%ecx , "tmp" \n\t"\
478 "sub "tmp" , "low" \n\t"\
479 "xor %%ecx , "ret" \n\t"
480 #else /* HAVE_FAST_CMOV */
481 #define BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
482 "mov "tmp" , %%ecx \n\t"\
483 "shl $17 , "tmp" \n\t"\
484 "sub "low" , "tmp" \n\t"\
485 "sar $31 , "tmp" \n\t" /*lps_mask*/\
486 "sub %%ecx , "range" \n\t" /*RangeLPS - range*/\
487 "and "tmp" , "range" \n\t" /*(RangeLPS - range)&lps_mask*/\
488 "add %%ecx , "range" \n\t" /*new range*/\
489 "shl $17 , %%ecx \n\t"\
490 "and "tmp" , %%ecx \n\t"\
491 "sub %%ecx , "low" \n\t"\
492 "xor "tmp" , "ret" \n\t"
493 #endif /* HAVE_FAST_CMOV */
496 #define BRANCHLESS_GET_CABAC(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
497 "movzbl "statep" , "ret" \n\t"\
498 "mov "range" , "tmp" \n\t"\
499 "and $0xC0 , "range" \n\t"\
500 "movzbl "MANGLE(ff_h264_lps_range)"("ret", "range", 2), "range" \n\t"\
501 "sub "range" , "tmp" \n\t"\
502 BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
503 "movzbl " MANGLE(ff_h264_norm_shift) "("range"), %%ecx \n\t"\
504 "shl %%cl , "range" \n\t"\
505 "movzbl "MANGLE(ff_h264_mlps_state)"+128("ret"), "tmp" \n\t"\
506 "mov "tmpbyte" , "statep" \n\t"\
507 "shl %%cl , "low" \n\t"\
508 "test "lowword" , "lowword" \n\t"\
510 "mov "BYTE"("cabac"), %%"REG_c" \n\t"\
511 "movzwl (%%"REG_c") , "tmp" \n\t"\
513 "shr $15 , "tmp" \n\t"\
514 "sub $0xFFFF , "tmp" \n\t"\
515 "add $2 , %%"REG_c" \n\t"\
516 "mov %%"REG_c" , "BYTE "("cabac") \n\t"\
517 "lea -1("low") , %%ecx \n\t"\
518 "xor "low" , %%ecx \n\t"\
519 "shr $15 , %%ecx \n\t"\
520 "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"\
522 "add $7 , %%ecx \n\t"\
523 "shl %%cl , "tmp" \n\t"\
524 "add "tmp" , "low" \n\t"\
528 "movl "RANGE "(%2), %%esi \n\t"
529 "movl "LOW "(%2), %%ebx \n\t"
530 BRANCHLESS_GET_CABAC("%0", "%2", "(%1)", "%%ebx", "%%bx", "%%esi", "%%edx", "%%dl")
531 "movl %%esi, "RANGE "(%2) \n\t"
532 "movl %%ebx, "LOW "(%2) \n\t"
536 : "%"REG_c, "%ebx", "%edx", "%esi", "memory"
539 #endif /* BRANCHLESS_CABAC_DECODER */
540 #else /* ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS) */
542 int RangeLPS= ff_h264_lps_range[2*(c->range&0xC0) + s];
543 int bit, lps_mask av_unused;
545 c->range -= RangeLPS;
546 #ifndef BRANCHLESS_CABAC_DECODER
547 if(c->low < (c->range<<(CABAC_BITS+1))){
549 *state= ff_h264_mps_state[s];
550 renorm_cabac_decoder_once(c);
552 bit= ff_h264_norm_shift[RangeLPS];
553 c->low -= (c->range<<(CABAC_BITS+1));
554 *state= ff_h264_lps_state[s];
555 c->range = RangeLPS<<bit;
559 if(!(c->low & CABAC_MASK)){
563 #else /* BRANCHLESS_CABAC_DECODER */
564 lps_mask= ((c->range<<(CABAC_BITS+1)) - c->low)>>31;
566 c->low -= (c->range<<(CABAC_BITS+1)) & lps_mask;
567 c->range += (RangeLPS - c->range) & lps_mask;
570 *state= (ff_h264_mlps_state+128)[s];
573 lps_mask= ff_h264_norm_shift[c->range];
574 c->range<<= lps_mask;
576 if(!(c->low & CABAC_MASK))
578 #endif /* BRANCHLESS_CABAC_DECODER */
579 #endif /* ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS) */
583 static int av_noinline av_unused get_cabac_noinline(CABACContext *c, uint8_t * const state){
584 return get_cabac_inline(c,state);
587 static int av_unused get_cabac(CABACContext *c, uint8_t * const state){
588 return get_cabac_inline(c,state);
591 static int av_unused get_cabac_bypass(CABACContext *c){
595 "movl "RANGE "(%1), %%ebx \n\t"
596 "movl "LOW "(%1), %%eax \n\t"
597 "shl $17, %%ebx \n\t"
598 "add %%eax, %%eax \n\t"
599 "sub %%ebx, %%eax \n\t"
601 "and %%edx, %%ebx \n\t"
602 "add %%ebx, %%eax \n\t"
603 "test %%ax, %%ax \n\t"
605 "movl "BYTE "(%1), %%"REG_b" \n\t"
606 "subl $0xFFFF, %%eax \n\t"
607 "movzwl (%%"REG_b"), %%ecx \n\t"
609 "shrl $15, %%ecx \n\t"
610 "addl $2, %%"REG_b" \n\t"
611 "addl %%ecx, %%eax \n\t"
612 "movl %%"REG_b", "BYTE "(%1) \n\t"
614 "movl %%eax, "LOW "(%1) \n\t"
618 : "%eax", "%"REG_b, "%ecx", "memory"
625 if(!(c->low & CABAC_MASK))
628 range= c->range<<(CABAC_BITS+1);
639 static av_always_inline int get_cabac_bypass_sign(CABACContext *c, int val){
640 #if ARCH_X86 && !(defined(PIC) && defined(__GNUC__))
642 "movl "RANGE "(%1), %%ebx \n\t"
643 "movl "LOW "(%1), %%eax \n\t"
644 "shl $17, %%ebx \n\t"
645 "add %%eax, %%eax \n\t"
646 "sub %%ebx, %%eax \n\t"
648 "and %%edx, %%ebx \n\t"
649 "add %%ebx, %%eax \n\t"
650 "xor %%edx, %%ecx \n\t"
651 "sub %%edx, %%ecx \n\t"
652 "test %%ax, %%ax \n\t"
654 "mov "BYTE "(%1), %%"REG_b" \n\t"
655 "subl $0xFFFF, %%eax \n\t"
656 "movzwl (%%"REG_b"), %%edx \n\t"
658 "shrl $15, %%edx \n\t"
659 "add $2, %%"REG_b" \n\t"
660 "addl %%edx, %%eax \n\t"
661 "mov %%"REG_b", "BYTE "(%1) \n\t"
663 "movl %%eax, "LOW "(%1) \n\t"
667 : "%eax", "%"REG_b, "%edx", "memory"
674 if(!(c->low & CABAC_MASK))
677 range= c->range<<(CABAC_BITS+1);
682 return (val^mask)-mask;
688 * @return the number of bytes read or 0 if no end
690 static int av_unused get_cabac_terminate(CABACContext *c){
692 if(c->low < c->range<<(CABAC_BITS+1)){
693 renorm_cabac_decoder_once(c);
696 return c->bytestream - c->bytestream_start;
702 * Get (truncated) unary binarization.
704 static int get_cabac_u(CABACContext *c, uint8_t * state, int max, int max_index, int truncated){
707 for(i=0; i<max; i++){
708 if(get_cabac(c, state)==0)
711 if(i< max_index) state++;
714 return truncated ? max : -1;
718 * get unary exp golomb k-th order binarization.
720 static int get_cabac_ueg(CABACContext *c, uint8_t * state, int max, int is_signed, int k, int max_index){
724 if(get_cabac(c, state)==0)
727 if(0 < max_index) state++;
729 for(i=1; i<max; i++){
730 if(get_cabac(c, state)==0){
731 if(is_signed && get_cabac_bypass(c)){
737 if(i < max_index) state++;
740 while(get_cabac_bypass(c)){
747 v+= v + get_cabac_bypass(c);
751 if(is_signed && get_cabac_bypass(c)){
758 #endif /* AVCODEC_CABAC_H */