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 FFMPEG_CABAC_H
28 #define FFMPEG_CABAC_H
30 #include "bitstream.h"
35 #include "libavutil/x86_cpu.h"
39 #define CABAC_MASK ((1<<CABAC_BITS)-1)
40 #define BRANCHLESS_CABAC_DECODER 1
41 //#define ARCH_X86_DISABLED 1
43 typedef struct CABACContext{
46 int outstanding_count;
50 const uint8_t *bytestream_start;
51 const uint8_t *bytestream;
52 const uint8_t *bytestream_end;
56 extern uint8_t ff_h264_mlps_state[4*64];
57 extern uint8_t ff_h264_lps_range[4*2*64]; ///< rangeTabLPS
58 extern uint8_t ff_h264_mps_state[2*64]; ///< transIdxMPS
59 extern uint8_t ff_h264_lps_state[2*64]; ///< transIdxLPS
60 extern const uint8_t ff_h264_norm_shift[512];
63 void ff_init_cabac_encoder(CABACContext *c, uint8_t *buf, int buf_size);
64 void ff_init_cabac_decoder(CABACContext *c, const uint8_t *buf, int buf_size);
65 void ff_init_cabac_states(CABACContext *c);
68 static inline void put_cabac_bit(CABACContext *c, int b){
69 put_bits(&c->pb, 1, b);
70 for(;c->outstanding_count; c->outstanding_count--){
71 put_bits(&c->pb, 1, 1-b);
75 static inline void renorm_cabac_encoder(CABACContext *c){
76 while(c->range < 0x100){
80 }else if(c->low<0x200){
81 c->outstanding_count++;
94 static void put_cabac(CABACContext *c, uint8_t * const state, int bit){
95 int RangeLPS= ff_h264_lps_range[2*(c->range&0xC0) + *state];
97 if(bit == ((*state)&1)){
99 *state= ff_h264_mps_state[*state];
101 c->low += c->range - RangeLPS;
103 *state= ff_h264_lps_state[*state];
106 renorm_cabac_encoder(c);
113 static void put_cabac_static(CABACContext *c, int RangeLPS, int bit){
114 assert(c->range > RangeLPS);
117 c->range -= RangeLPS;
119 c->low += c->range - RangeLPS;
123 renorm_cabac_encoder(c);
131 * @param bit 0 -> write zero bit, !=0 write one bit
133 static void put_cabac_bypass(CABACContext *c, int bit){
142 }else if(c->low<0x400){
143 c->outstanding_count++;
157 * @return the number of bytes written
159 static int put_cabac_terminate(CABACContext *c, int bit){
163 renorm_cabac_encoder(c);
168 renorm_cabac_encoder(c);
170 assert(c->low <= 0x1FF);
171 put_cabac_bit(c, c->low>>9);
172 put_bits(&c->pb, 2, ((c->low>>7)&3)|1);
174 flush_put_bits(&c->pb); //FIXME FIXME FIXME XXX wrong
181 return (put_bits_count(&c->pb)+7)>>3;
185 * put (truncated) unary binarization.
187 static void put_cabac_u(CABACContext *c, uint8_t * state, int v, int max, int max_index, int truncated){
194 put_cabac(c, state, 1);
195 if(i < max_index) state++;
197 if(truncated==0 || v<max)
198 put_cabac(c, state, 0);
202 put_cabac(c, state+i, 1);
204 if(truncated==0 || v<max)
205 put_cabac(c, state+i, 0);
207 for(i=0; i<=max_index; i++){
208 put_cabac(c, state+i, 1);
211 put_cabac(c, state+max_index, 1);
213 if(truncated==0 || v<max)
214 put_cabac(c, state+max_index, 0);
220 * put unary exp golomb k-th order binarization.
222 static void put_cabac_ueg(CABACContext *c, uint8_t * state, int v, int max, int is_signed, int k, int max_index){
226 put_cabac(c, state, 0);
228 const int sign= v < 0;
230 if(is_signed) v= FFABS(v);
234 put_cabac(c, state, 1);
235 if(i < max_index) state++;
238 put_cabac(c, state, 0);
242 for(i=0; i<max; i++){
243 put_cabac(c, state, 1);
244 if(i < max_index) state++;
248 while(v >= m){ //FIXME optimize
249 put_cabac_bypass(c, 1);
253 put_cabac_bypass(c, 0);
255 put_cabac_bypass(c, v&m);
260 put_cabac_bypass(c, sign);
265 static void refill(CABACContext *c){
267 c->low+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
269 c->low+= c->bytestream[0]<<1;
271 c->low -= CABAC_MASK;
272 c->bytestream+= CABAC_BITS/8;
275 #if ! ( defined(ARCH_X86) && defined(HAVE_7REGS) && defined(HAVE_EBX_AVAILABLE) && !defined(BROKEN_RELOCATIONS) )
276 static void refill2(CABACContext *c){
279 x= c->low ^ (c->low-1);
280 i= 7 - ff_h264_norm_shift[x>>(CABAC_BITS-1)];
285 x+= (c->bytestream[0]<<9) + (c->bytestream[1]<<1);
287 x+= c->bytestream[0]<<1;
291 c->bytestream+= CABAC_BITS/8;
295 static inline void renorm_cabac_decoder(CABACContext *c){
296 while(c->range < 0x100){
299 if(!(c->low & CABAC_MASK))
304 static inline void renorm_cabac_decoder_once(CABACContext *c){
305 #ifdef ARCH_X86_DISABLED
310 "lea -0x100(%0), %2 \n\t"
311 "shr $31, %2 \n\t" //FIXME 31->63 for x86-64
314 : "+r"(c->range), "+r"(c->low), "+c"(temp)
319 "cmp $0x100, %0 \n\t"
320 "setb %%cl \n\t" //FIXME 31->63 for x86-64
323 : "+r"(c->range), "+r"(c->low), "+c"(temp)
329 "lea -0x100(%0), %%eax \n\t"
336 : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
342 "cmp $0x100, %0 \n\t"
343 "sbb %%edx, %%edx \n\t"
349 : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
355 "cmp $0x100, %0 \n\t"
356 "lea (%0, %0), %%eax \n\t"
357 "lea (%1, %1), %%edx \n\t"
358 "cmovb %%eax, %0 \n\t"
359 "cmovb %%edx, %1 \n\t"
360 : "+r"(c->range), "+r"(c->low), "+a"(temp), "+d"(temp2)
365 int shift= (uint32_t)(c->range - 0x100)>>31;
369 if(!(c->low & CABAC_MASK))
373 static av_always_inline int get_cabac_inline(CABACContext *c, uint8_t * const state){
374 //FIXME gcc generates duplicate load/stores for c->low and c->range
378 #define BYTESTART "16"
382 #define BYTESTART "12"
386 #if defined(ARCH_X86) && defined(HAVE_7REGS) && defined(HAVE_EBX_AVAILABLE) && !defined(BROKEN_RELOCATIONS)
389 #ifndef BRANCHLESS_CABAC_DECODER
391 "movzbl (%1), %0 \n\t"
392 "movl "RANGE "(%2), %%ebx \n\t"
393 "movl "RANGE "(%2), %%edx \n\t"
394 "andl $0xC0, %%ebx \n\t"
395 "movzbl "MANGLE(ff_h264_lps_range)"(%0, %%ebx, 2), %%esi\n\t"
396 "movl "LOW "(%2), %%ebx \n\t"
397 //eax:state ebx:low, edx:range, esi:RangeLPS
398 "subl %%esi, %%edx \n\t"
399 "movl %%edx, %%ecx \n\t"
400 "shll $17, %%ecx \n\t"
401 "cmpl %%ecx, %%ebx \n\t"
405 //athlon:4067 P3:4110
406 "lea -0x100(%%edx), %%ecx \n\t"
407 "shr $31, %%ecx \n\t"
408 "shl %%cl, %%edx \n\t"
409 "shl %%cl, %%ebx \n\t"
411 //athlon:4057 P3:4130
412 "cmp $0x100, %%edx \n\t" //FIXME avoidable
414 "shl %%cl, %%edx \n\t"
415 "shl %%cl, %%ebx \n\t"
417 "movzbl "MANGLE(ff_h264_mps_state)"(%0), %%ecx \n\t"
418 "movb %%cl, (%1) \n\t"
419 //eax:state ebx:low, edx:range, esi:RangeLPS
420 "test %%bx, %%bx \n\t"
422 "mov "BYTE "(%2), %%"REG_S" \n\t"
423 "subl $0xFFFF, %%ebx \n\t"
424 "movzwl (%%"REG_S"), %%ecx \n\t"
426 "shrl $15, %%ecx \n\t"
427 "add $2, %%"REG_S" \n\t"
428 "addl %%ecx, %%ebx \n\t"
429 "mov %%"REG_S", "BYTE "(%2) \n\t"
432 //eax:state ebx:low, edx:range, esi:RangeLPS
433 "subl %%ecx, %%ebx \n\t"
434 "movl %%esi, %%edx \n\t"
435 "movzbl " MANGLE(ff_h264_norm_shift) "(%%esi), %%ecx \n\t"
436 "shll %%cl, %%ebx \n\t"
437 "shll %%cl, %%edx \n\t"
438 "movzbl "MANGLE(ff_h264_lps_state)"(%0), %%ecx \n\t"
439 "movb %%cl, (%1) \n\t"
441 "test %%bx, %%bx \n\t"
444 "mov "BYTE "(%2), %%"REG_c" \n\t"
445 "movzwl (%%"REG_c"), %%esi \n\t"
447 "shrl $15, %%esi \n\t"
448 "subl $0xFFFF, %%esi \n\t"
449 "add $2, %%"REG_c" \n\t"
450 "mov %%"REG_c", "BYTE "(%2) \n\t"
452 "leal -1(%%ebx), %%ecx \n\t"
453 "xorl %%ebx, %%ecx \n\t"
454 "shrl $15, %%ecx \n\t"
455 "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"
459 "shll %%cl , %%esi \n\t"
460 "addl %%esi, %%ebx \n\t"
462 "movl %%edx, "RANGE "(%2) \n\t"
463 "movl %%ebx, "LOW "(%2) \n\t"
464 :"=&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
466 : "%"REG_c, "%ebx", "%edx", "%"REG_S, "memory"
469 #else /* BRANCHLESS_CABAC_DECODER */
472 #if defined HAVE_FAST_CMOV
473 #define BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
474 "mov "tmp" , %%ecx \n\t"\
475 "shl $17 , "tmp" \n\t"\
476 "cmp "low" , "tmp" \n\t"\
477 "cmova %%ecx , "range" \n\t"\
478 "sbb %%ecx , %%ecx \n\t"\
479 "and %%ecx , "tmp" \n\t"\
480 "sub "tmp" , "low" \n\t"\
481 "xor %%ecx , "ret" \n\t"
482 #else /* HAVE_FAST_CMOV */
483 #define BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
484 "mov "tmp" , %%ecx \n\t"\
485 "shl $17 , "tmp" \n\t"\
486 "sub "low" , "tmp" \n\t"\
487 "sar $31 , "tmp" \n\t" /*lps_mask*/\
488 "sub %%ecx , "range" \n\t" /*RangeLPS - range*/\
489 "and "tmp" , "range" \n\t" /*(RangeLPS - range)&lps_mask*/\
490 "add %%ecx , "range" \n\t" /*new range*/\
491 "shl $17 , %%ecx \n\t"\
492 "and "tmp" , %%ecx \n\t"\
493 "sub %%ecx , "low" \n\t"\
494 "xor "tmp" , "ret" \n\t"
495 #endif /* HAVE_FAST_CMOV */
498 #define BRANCHLESS_GET_CABAC(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
499 "movzbl "statep" , "ret" \n\t"\
500 "mov "range" , "tmp" \n\t"\
501 "and $0xC0 , "range" \n\t"\
502 "movzbl "MANGLE(ff_h264_lps_range)"("ret", "range", 2), "range" \n\t"\
503 "sub "range" , "tmp" \n\t"\
504 BRANCHLESS_GET_CABAC_UPDATE(ret, cabac, statep, low, lowword, range, tmp, tmpbyte)\
505 "movzbl " MANGLE(ff_h264_norm_shift) "("range"), %%ecx \n\t"\
506 "shl %%cl , "range" \n\t"\
507 "movzbl "MANGLE(ff_h264_mlps_state)"+128("ret"), "tmp" \n\t"\
508 "mov "tmpbyte" , "statep" \n\t"\
509 "shl %%cl , "low" \n\t"\
510 "test "lowword" , "lowword" \n\t"\
512 "mov "BYTE"("cabac"), %%"REG_c" \n\t"\
513 "movzwl (%%"REG_c") , "tmp" \n\t"\
515 "shr $15 , "tmp" \n\t"\
516 "sub $0xFFFF , "tmp" \n\t"\
517 "add $2 , %%"REG_c" \n\t"\
518 "mov %%"REG_c" , "BYTE "("cabac") \n\t"\
519 "lea -1("low") , %%ecx \n\t"\
520 "xor "low" , %%ecx \n\t"\
521 "shr $15 , %%ecx \n\t"\
522 "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"\
524 "add $7 , %%ecx \n\t"\
525 "shl %%cl , "tmp" \n\t"\
526 "add "tmp" , "low" \n\t"\
530 "movl "RANGE "(%2), %%esi \n\t"
531 "movl "LOW "(%2), %%ebx \n\t"
532 BRANCHLESS_GET_CABAC("%0", "%2", "(%1)", "%%ebx", "%%bx", "%%esi", "%%edx", "%%dl")
533 "movl %%esi, "RANGE "(%2) \n\t"
534 "movl %%ebx, "LOW "(%2) \n\t"
538 : "%"REG_c, "%ebx", "%edx", "%esi", "memory"
541 #endif /* BRANCHLESS_CABAC_DECODER */
542 #else /* defined(ARCH_X86) && defined(HAVE_7REGS) && defined(HAVE_EBX_AVAILABLE) && !defined(BROKEN_RELOCATIONS) */
544 int RangeLPS= ff_h264_lps_range[2*(c->range&0xC0) + s];
545 int bit, lps_mask av_unused;
547 c->range -= RangeLPS;
548 #ifndef BRANCHLESS_CABAC_DECODER
549 if(c->low < (c->range<<(CABAC_BITS+1))){
551 *state= ff_h264_mps_state[s];
552 renorm_cabac_decoder_once(c);
554 bit= ff_h264_norm_shift[RangeLPS];
555 c->low -= (c->range<<(CABAC_BITS+1));
556 *state= ff_h264_lps_state[s];
557 c->range = RangeLPS<<bit;
561 if(!(c->low & CABAC_MASK)){
565 #else /* BRANCHLESS_CABAC_DECODER */
566 lps_mask= ((c->range<<(CABAC_BITS+1)) - c->low)>>31;
568 c->low -= (c->range<<(CABAC_BITS+1)) & lps_mask;
569 c->range += (RangeLPS - c->range) & lps_mask;
572 *state= (ff_h264_mlps_state+128)[s];
575 lps_mask= ff_h264_norm_shift[c->range];
576 c->range<<= lps_mask;
578 if(!(c->low & CABAC_MASK))
580 #endif /* BRANCHLESS_CABAC_DECODER */
581 #endif /* defined(ARCH_X86) && defined(HAVE_7REGS) && defined(HAVE_EBX_AVAILABLE) && !defined(BROKEN_RELOCATIONS) */
585 static int av_noinline get_cabac_noinline(CABACContext *c, uint8_t * const state){
586 return get_cabac_inline(c,state);
589 static int get_cabac(CABACContext *c, uint8_t * const state){
590 return get_cabac_inline(c,state);
593 static int get_cabac_bypass(CABACContext *c){
597 "movl "RANGE "(%1), %%ebx \n\t"
598 "movl "LOW "(%1), %%eax \n\t"
599 "shl $17, %%ebx \n\t"
600 "add %%eax, %%eax \n\t"
601 "sub %%ebx, %%eax \n\t"
603 "and %%edx, %%ebx \n\t"
604 "add %%ebx, %%eax \n\t"
605 "test %%ax, %%ax \n\t"
607 "movl "BYTE "(%1), %%"REG_b" \n\t"
608 "subl $0xFFFF, %%eax \n\t"
609 "movzwl (%%"REG_b"), %%ecx \n\t"
611 "shrl $15, %%ecx \n\t"
612 "addl $2, %%"REG_b" \n\t"
613 "addl %%ecx, %%eax \n\t"
614 "movl %%"REG_b", "BYTE "(%1) \n\t"
616 "movl %%eax, "LOW "(%1) \n\t"
620 : "%eax", "%"REG_b, "%ecx", "memory"
627 if(!(c->low & CABAC_MASK))
630 range= c->range<<(CABAC_BITS+1);
641 static av_always_inline int get_cabac_bypass_sign(CABACContext *c, int val){
642 #if defined(ARCH_X86) && !(defined(PIC) && defined(__GNUC__))
644 "movl "RANGE "(%1), %%ebx \n\t"
645 "movl "LOW "(%1), %%eax \n\t"
646 "shl $17, %%ebx \n\t"
647 "add %%eax, %%eax \n\t"
648 "sub %%ebx, %%eax \n\t"
650 "and %%edx, %%ebx \n\t"
651 "add %%ebx, %%eax \n\t"
652 "xor %%edx, %%ecx \n\t"
653 "sub %%edx, %%ecx \n\t"
654 "test %%ax, %%ax \n\t"
656 "mov "BYTE "(%1), %%"REG_b" \n\t"
657 "subl $0xFFFF, %%eax \n\t"
658 "movzwl (%%"REG_b"), %%edx \n\t"
660 "shrl $15, %%edx \n\t"
661 "add $2, %%"REG_b" \n\t"
662 "addl %%edx, %%eax \n\t"
663 "mov %%"REG_b", "BYTE "(%1) \n\t"
665 "movl %%eax, "LOW "(%1) \n\t"
669 : "%eax", "%"REG_b, "%edx", "memory"
676 if(!(c->low & CABAC_MASK))
679 range= c->range<<(CABAC_BITS+1);
684 return (val^mask)-mask;
690 * @return the number of bytes read or 0 if no end
692 static int get_cabac_terminate(CABACContext *c){
694 if(c->low < c->range<<(CABAC_BITS+1)){
695 renorm_cabac_decoder_once(c);
698 return c->bytestream - c->bytestream_start;
704 * Get (truncated) unary binarization.
706 static int get_cabac_u(CABACContext *c, uint8_t * state, int max, int max_index, int truncated){
709 for(i=0; i<max; i++){
710 if(get_cabac(c, state)==0)
713 if(i< max_index) state++;
716 return truncated ? max : -1;
720 * get unary exp golomb k-th order binarization.
722 static int get_cabac_ueg(CABACContext *c, uint8_t * state, int max, int is_signed, int k, int max_index){
726 if(get_cabac(c, state)==0)
729 if(0 < max_index) state++;
731 for(i=1; i<max; i++){
732 if(get_cabac(c, state)==0){
733 if(is_signed && get_cabac_bypass(c)){
739 if(i < max_index) state++;
742 while(get_cabac_bypass(c)){
749 v+= v + get_cabac_bypass(c);
753 if(is_signed && get_cabac_bypass(c)){
760 #endif /* FFMPEG_CABAC_H */