falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
33883 Commits
32 Branches
2.4GB
Tree: 9ee3334840
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '9ee3334840'
${ noResults }
FFmpeg/libavcodec/x86/cabac.h

229 lines
11KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>

  3.  *

  4.  * This file is part of Libav.

  5.  *

  6.  * Libav is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2.1 of the License, or (at your option) any later version.

  10.  *

  11.  * Libav is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with Libav; if not, write to the Free Software

  18.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  19.  */

  20. 

  21. #ifndef AVCODEC_X86_CABAC_H

  22. #define AVCODEC_X86_CABAC_H

  23. 

  24. #include "libavcodec/cabac.h"

  25. #include "libavutil/attributes.h"

  26. #include "libavutil/x86_cpu.h"

  27. #include "libavutil/internal.h"

  28. #include "config.h"

  29. 

  30. #ifdef BROKEN_RELOCATIONS

  31. #define TABLES_ARG , "r"(tables)

  32. 

  33. #if HAVE_FAST_CMOV

  34. #define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \

  35.         "cmp    "low"       , "tmp"                        \n\t"\

  36.         "cmova  %%ecx       , "range"                      \n\t"\

  37.         "sbb    %%rcx       , %%rcx                        \n\t"\

  38.         "and    %%ecx       , "tmp"                        \n\t"\

  39.         "xor    %%rcx       , "retq"                       \n\t"\

  40.         "sub    "tmp"       , "low"                        \n\t"

  41. #else /* HAVE_FAST_CMOV */

  42. #define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \

  43. /* P4 Prescott has crappy cmov,sbb,64bit shift so avoid them */ \

  44.         "sub    "low"       , "tmp"                        \n\t"\

  45.         "sar    $31         , "tmp"                        \n\t"\

  46.         "sub    %%ecx       , "range"                      \n\t"\

  47.         "and    "tmp"       , "range"                      \n\t"\

  48.         "add    %%ecx       , "range"                      \n\t"\

  49.         "shl    $17         , %%ecx                        \n\t"\

  50.         "and    "tmp"       , %%ecx                        \n\t"\

  51.         "sub    %%ecx       , "low"                        \n\t"\

  52.         "xor    "tmp"       , "ret"                        \n\t"\

  53.         "movslq "ret"       , "retq"                       \n\t"

  54. #endif /* HAVE_FAST_CMOV */

  55. 

  56. #define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \

  57.         "movzbl "statep"    , "ret"                                     \n\t"\

  58.         "mov    "range"     , "tmp"                                     \n\t"\

  59.         "and    $0xC0       , "range"                                   \n\t"\

  60.         "lea    ("ret", "range", 2), %%ecx                              \n\t"\

  61.         "movzbl "lps_off"("tables", %%rcx), "range"                     \n\t"\

  62.         "sub    "range"     , "tmp"                                     \n\t"\

  63.         "mov    "tmp"       , %%ecx                                     \n\t"\

  64.         "shl    $17         , "tmp"                                     \n\t"\

  65.         BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp)              \

  66.         "movzbl "norm_off"("tables", "rangeq"), %%ecx                   \n\t"\

  67.         "shl    %%cl        , "range"                                   \n\t"\

  68.         "movzbl "mlps_off"+128("tables", "retq"), "tmp"                 \n\t"\

  69.         "shl    %%cl        , "low"                                     \n\t"\

  70.         "mov    "tmpbyte"   , "statep"                                  \n\t"\

  71.         "test   "lowword"   , "lowword"                                 \n\t"\

  72.         "jnz    2f                                                      \n\t"\

  73.         "mov    "byte"      , %%"REG_c"                                 \n\t"\

  74.         "cmp    "end"       , %%"REG_c"                                 \n\t"\

  75.         "jge    1f                                                      \n\t"\

  76.         "add"OPSIZE" $2     , "byte"                                    \n\t"\

  77.         "1:                                                             \n\t"\

  78.         "movzwl (%%"REG_c") , "tmp"                                     \n\t"\

  79.         "lea    -1("low")   , %%ecx                                     \n\t"\

  80.         "xor    "low"       , %%ecx                                     \n\t"\

  81.         "shr    $15         , %%ecx                                     \n\t"\

  82.         "bswap  "tmp"                                                   \n\t"\

  83.         "shr    $15         , "tmp"                                     \n\t"\

  84.         "movzbl "norm_off"("tables", %%rcx), %%ecx                      \n\t"\

  85.         "sub    $0xFFFF     , "tmp"                                     \n\t"\

  86.         "neg    %%ecx                                                   \n\t"\

  87.         "add    $7          , %%ecx                                     \n\t"\

  88.         "shl    %%cl        , "tmp"                                     \n\t"\

  89.         "add    "tmp"       , "low"                                     \n\t"\

  90.         "2:                                                             \n\t"

  91. 

  92. #else /* BROKEN_RELOCATIONS */

  93. #define TABLES_ARG

  94. 

  95. #if HAVE_FAST_CMOV

  96. #define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\

  97.         "mov    "tmp"       , %%ecx     \n\t"\

  98.         "shl    $17         , "tmp"     \n\t"\

  99.         "cmp    "low"       , "tmp"     \n\t"\

  100.         "cmova  %%ecx       , "range"   \n\t"\

  101.         "sbb    %%ecx       , %%ecx     \n\t"\

  102.         "and    %%ecx       , "tmp"     \n\t"\

  103.         "xor    %%ecx       , "ret"     \n\t"\

  104.         "sub    "tmp"       , "low"     \n\t"

  105. #else /* HAVE_FAST_CMOV */

  106. #define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\

  107.         "mov    "tmp"       , %%ecx     \n\t"\

  108.         "shl    $17         , "tmp"     \n\t"\

  109.         "sub    "low"       , "tmp"     \n\t"\

  110.         "sar    $31         , "tmp"     \n\t" /*lps_mask*/\

  111.         "sub    %%ecx       , "range"   \n\t" /*RangeLPS - range*/\

  112.         "and    "tmp"       , "range"   \n\t" /*(RangeLPS - range)&lps_mask*/\

  113.         "add    %%ecx       , "range"   \n\t" /*new range*/\

  114.         "shl    $17         , %%ecx     \n\t"\

  115.         "and    "tmp"       , %%ecx     \n\t"\

  116.         "sub    %%ecx       , "low"     \n\t"\

  117.         "xor    "tmp"       , "ret"     \n\t"

  118. #endif /* HAVE_FAST_CMOV */

  119. 

  120. #define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \

  121.         "movzbl "statep"    , "ret"                                     \n\t"\

  122.         "mov    "range"     , "tmp"                                     \n\t"\

  123.         "and    $0xC0       , "range"                                   \n\t"\

  124.         "movzbl "MANGLE(ff_h264_cabac_tables)"+"lps_off"("ret", "range", 2), "range" \n\t"\

  125.         "sub    "range"     , "tmp"                                     \n\t"\

  126.         BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)                    \

  127.         "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"("range"), %%ecx    \n\t"\

  128.         "shl    %%cl        , "range"                                   \n\t"\

  129.         "movzbl "MANGLE(ff_h264_cabac_tables)"+"mlps_off"+128("ret"), "tmp"  \n\t"\

  130.         "shl    %%cl        , "low"                                     \n\t"\

  131.         "mov    "tmpbyte"   , "statep"                                  \n\t"\

  132.         "test   "lowword"   , "lowword"                                 \n\t"\

  133.         " jnz   2f                                                      \n\t"\

  134.         "mov    "byte"      , %%"REG_c"                                 \n\t"\

  135.         "cmp    "end"       , %%"REG_c"                                 \n\t"\

  136.         "jge    1f                                                      \n\t"\

  137.         "add"OPSIZE" $2     , "byte"                                    \n\t"\

  138.         "1:                                                             \n\t"\

  139.         "movzwl (%%"REG_c")     , "tmp"                                 \n\t"\

  140.         "lea    -1("low")   , %%ecx                                     \n\t"\

  141.         "xor    "low"       , %%ecx                                     \n\t"\

  142.         "shr    $15         , %%ecx                                     \n\t"\

  143.         "bswap  "tmp"                                                   \n\t"\

  144.         "shr    $15         , "tmp"                                     \n\t"\

  145.         "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"(%%ecx), %%ecx \n\t"\

  146.         "sub    $0xFFFF     , "tmp"                                     \n\t"\

  147.         "neg    %%ecx                                                   \n\t"\

  148.         "add    $7          , %%ecx                                     \n\t"\

  149.         "shl    %%cl        , "tmp"                                     \n\t"\

  150.         "add    "tmp"       , "low"                                     \n\t"\

  151.         "2:                                                             \n\t"

  152. 

  153. #endif /* BROKEN_RELOCATIONS */

  154. 

  155. 

  156. #if HAVE_7REGS

  157. #define get_cabac_inline get_cabac_inline_x86

  158. static av_always_inline int get_cabac_inline_x86(CABACContext *c,

  159.                                                  uint8_t *const state)

  160. {

  161.     int bit, tmp;

  162. #ifdef BROKEN_RELOCATIONS

  163.     void *tables;

  164. 

  165.     __asm__ volatile(

  166.         "lea    "MANGLE(ff_h264_cabac_tables)", %0      \n\t"

  167.         : "=&r"(tables)

  168.     );

  169. #endif

  170. 

  171.     __asm__ volatile(

  172.         BRANCHLESS_GET_CABAC("%0", "%q0", "(%4)", "%1", "%w1",

  173.                              "%2", "%q2", "%3", "%b3",

  174.                              "%a6(%5)", "%a7(%5)", "%a8", "%a9", "%a10", "%11")

  175.         : "=&r"(bit), "+&r"(c->low), "+&r"(c->range), "=&q"(tmp)

  176.         : "r"(state), "r"(c),

  177.           "i"(offsetof(CABACContext, bytestream)),

  178.           "i"(offsetof(CABACContext, bytestream_end)),

  179.           "i"(H264_NORM_SHIFT_OFFSET),

  180.           "i"(H264_LPS_RANGE_OFFSET),

  181.           "i"(H264_MLPS_STATE_OFFSET) TABLES_ARG

  182.         : "%"REG_c, "memory"

  183.     );

  184.     return bit & 1;

  185. }

  186. #endif /* HAVE_7REGS */

  187. 

  188. #define get_cabac_bypass_sign get_cabac_bypass_sign_x86

  189. static av_always_inline int get_cabac_bypass_sign_x86(CABACContext *c, int val)

  190. {

  191.     x86_reg tmp;

  192.     __asm__ volatile(

  193.         "movl        %a6(%2), %k1       \n\t"

  194.         "movl        %a3(%2), %%eax     \n\t"

  195.         "shl             $17, %k1       \n\t"

  196.         "add           %%eax, %%eax     \n\t"

  197.         "sub             %k1, %%eax     \n\t"

  198.         "cltd                           \n\t"

  199.         "and           %%edx, %k1       \n\t"

  200.         "add             %k1, %%eax     \n\t"

  201.         "xor           %%edx, %%ecx     \n\t"

  202.         "sub           %%edx, %%ecx     \n\t"

  203.         "test           %%ax, %%ax      \n\t"

  204.         "jnz              1f            \n\t"

  205.         "mov         %a4(%2), %1        \n\t"

  206.         "subl        $0xFFFF, %%eax     \n\t"

  207.         "movzwl         (%1), %%edx     \n\t"

  208.         "bswap         %%edx            \n\t"

  209.         "shrl            $15, %%edx     \n\t"

  210.         "addl          %%edx, %%eax     \n\t"

  211.         "cmp         %a5(%2), %1        \n\t"

  212.         "jge              1f            \n\t"

  213.         "add"OPSIZE"      $2, %a4(%2)   \n\t"

  214.         "1:                             \n\t"

  215.         "movl          %%eax, %a3(%2)   \n\t"

  216. 

  217.         : "+c"(val), "=&r"(tmp)

  218.         : "r"(c),

  219.           "i"(offsetof(CABACContext, low)),

  220.           "i"(offsetof(CABACContext, bytestream)),

  221.           "i"(offsetof(CABACContext, bytestream_end)),

  222.           "i"(offsetof(CABACContext, range))

  223.         : "%eax", "%edx", "memory"

  224.     );

  225.     return val;

  226. }

  227. 

  228. #endif /* AVCODEC_X86_CABAC_H */