cabac.h 13.8 KB
Newer Older
1 2 3
/*
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7 8 9 10
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12 13 14 15 16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with FFmpeg; if not, write to the Free Software
18 19 20 21 22 23 24 25
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#ifndef AVCODEC_X86_CABAC_H
#define AVCODEC_X86_CABAC_H

#include "libavcodec/cabac.h"
#include "libavutil/attributes.h"
26
#include "libavutil/macros.h"
27
#include "libavutil/x86/asm.h"
28 29
#include "config.h"

30 31 32 33 34 35 36
#if   (defined(__i386) && defined(__clang__) && (__clang_major__<2 || (__clang_major__==2 && __clang_minor__<10)))\
   || (                  !defined(__clang__) && defined(__llvm__) && __GNUC__==4 && __GNUC_MINOR__==2 && __GNUC_PATCHLEVEL__<=1)
#       define BROKEN_COMPILER 1
#else
#       define BROKEN_COMPILER 0
#endif

37 38
#if HAVE_INLINE_ASM

39 40 41 42 43 44 45 46 47 48 49 50
#ifndef UNCHECKED_BITSTREAM_READER
#define UNCHECKED_BITSTREAM_READER !CONFIG_SAFE_BITSTREAM_READER
#endif

#if UNCHECKED_BITSTREAM_READER
#define END_CHECK(end) ""
#else
#define END_CHECK(end) \
        "cmp    "end"       , %%"REG_c"                                 \n\t"\
        "jge    1f                                                      \n\t"
#endif

51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
#ifdef BROKEN_RELOCATIONS
#define TABLES_ARG , "r"(tables)

#if HAVE_FAST_CMOV
#define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
        "cmp    "low"       , "tmp"                        \n\t"\
        "cmova  %%ecx       , "range"                      \n\t"\
        "sbb    %%rcx       , %%rcx                        \n\t"\
        "and    %%ecx       , "tmp"                        \n\t"\
        "xor    %%rcx       , "retq"                       \n\t"\
        "sub    "tmp"       , "low"                        \n\t"
#else /* HAVE_FAST_CMOV */
#define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
/* P4 Prescott has crappy cmov,sbb,64bit shift so avoid them */ \
        "sub    "low"       , "tmp"                        \n\t"\
        "sar    $31         , "tmp"                        \n\t"\
        "sub    %%ecx       , "range"                      \n\t"\
        "and    "tmp"       , "range"                      \n\t"\
        "add    %%ecx       , "range"                      \n\t"\
        "shl    $17         , %%ecx                        \n\t"\
        "and    "tmp"       , %%ecx                        \n\t"\
        "sub    %%ecx       , "low"                        \n\t"\
        "xor    "tmp"       , "ret"                        \n\t"\
        "movslq "ret"       , "retq"                       \n\t"
#endif /* HAVE_FAST_CMOV */

#define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \
        "movzbl "statep"    , "ret"                                     \n\t"\
        "mov    "range"     , "tmp"                                     \n\t"\
        "and    $0xC0       , "range"                                   \n\t"\
        "lea    ("ret", "range", 2), %%ecx                              \n\t"\
        "movzbl "lps_off"("tables", %%rcx), "range"                     \n\t"\
        "sub    "range"     , "tmp"                                     \n\t"\
        "mov    "tmp"       , %%ecx                                     \n\t"\
        "shl    $17         , "tmp"                                     \n\t"\
        BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp)              \
        "movzbl "norm_off"("tables", "rangeq"), %%ecx                   \n\t"\
        "shl    %%cl        , "range"                                   \n\t"\
        "movzbl "mlps_off"+128("tables", "retq"), "tmp"                 \n\t"\
        "shl    %%cl        , "low"                                     \n\t"\
        "mov    "tmpbyte"   , "statep"                                  \n\t"\
        "test   "lowword"   , "lowword"                                 \n\t"\
        "jnz    2f                                                      \n\t"\
        "mov    "byte"      , %%"REG_c"                                 \n\t"\
95
        END_CHECK(end)\
96
        "add"OPSIZE" $2     , "byte"                                    \n\t"\
97
        "1:                                                             \n\t"\
98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
        "movzwl (%%"REG_c") , "tmp"                                     \n\t"\
        "lea    -1("low")   , %%ecx                                     \n\t"\
        "xor    "low"       , %%ecx                                     \n\t"\
        "shr    $15         , %%ecx                                     \n\t"\
        "bswap  "tmp"                                                   \n\t"\
        "shr    $15         , "tmp"                                     \n\t"\
        "movzbl "norm_off"("tables", %%rcx), %%ecx                      \n\t"\
        "sub    $0xFFFF     , "tmp"                                     \n\t"\
        "neg    %%ecx                                                   \n\t"\
        "add    $7          , %%ecx                                     \n\t"\
        "shl    %%cl        , "tmp"                                     \n\t"\
        "add    "tmp"       , "low"                                     \n\t"\
        "2:                                                             \n\t"

#else /* BROKEN_RELOCATIONS */
113
#define TABLES_ARG NAMED_CONSTRAINTS_ADD(ff_h264_cabac_tables)
114 115
#define RIP_ARG

116
#if HAVE_FAST_CMOV
117
#define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\
118 119 120 121 122 123
        "mov    "tmp"       , %%ecx     \n\t"\
        "shl    $17         , "tmp"     \n\t"\
        "cmp    "low"       , "tmp"     \n\t"\
        "cmova  %%ecx       , "range"   \n\t"\
        "sbb    %%ecx       , %%ecx     \n\t"\
        "and    %%ecx       , "tmp"     \n\t"\
124 125
        "xor    %%ecx       , "ret"     \n\t"\
        "sub    "tmp"       , "low"     \n\t"
126
#else /* HAVE_FAST_CMOV */
127
#define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\
128 129 130 131 132 133 134 135 136 137 138 139 140
        "mov    "tmp"       , %%ecx     \n\t"\
        "shl    $17         , "tmp"     \n\t"\
        "sub    "low"       , "tmp"     \n\t"\
        "sar    $31         , "tmp"     \n\t" /*lps_mask*/\
        "sub    %%ecx       , "range"   \n\t" /*RangeLPS - range*/\
        "and    "tmp"       , "range"   \n\t" /*(RangeLPS - range)&lps_mask*/\
        "add    %%ecx       , "range"   \n\t" /*new range*/\
        "shl    $17         , %%ecx     \n\t"\
        "and    "tmp"       , %%ecx     \n\t"\
        "sub    %%ecx       , "low"     \n\t"\
        "xor    "tmp"       , "ret"     \n\t"
#endif /* HAVE_FAST_CMOV */

141
#define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \
142 143 144
        "movzbl "statep"    , "ret"                                     \n\t"\
        "mov    "range"     , "tmp"                                     \n\t"\
        "and    $0xC0       , "range"                                   \n\t"\
145
        "movzbl "MANGLE(ff_h264_cabac_tables)"+"lps_off"("ret", "range", 2), "range" \n\t"\
146
        "sub    "range"     , "tmp"                                     \n\t"\
147
        BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)                    \
148
        "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"("range"), %%ecx    \n\t"\
149
        "shl    %%cl        , "range"                                   \n\t"\
150
        "movzbl "MANGLE(ff_h264_cabac_tables)"+"mlps_off"+128("ret"), "tmp"  \n\t"\
151
        "shl    %%cl        , "low"                                     \n\t"\
152
        "mov    "tmpbyte"   , "statep"                                  \n\t"\
153
        "test   "lowword"   , "lowword"                                 \n\t"\
154
        " jnz   2f                                                      \n\t"\
155
        "mov    "byte"      , %%"REG_c"                                 \n\t"\
156
        END_CHECK(end)\
157
        "add"OPSIZE" $2     , "byte"                                    \n\t"\
158
        "1:                                                             \n\t"\
159 160 161 162
        "movzwl (%%"REG_c")     , "tmp"                                 \n\t"\
        "lea    -1("low")   , %%ecx                                     \n\t"\
        "xor    "low"       , %%ecx                                     \n\t"\
        "shr    $15         , %%ecx                                     \n\t"\
163 164
        "bswap  "tmp"                                                   \n\t"\
        "shr    $15         , "tmp"                                     \n\t"\
165
        "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"(%%ecx), %%ecx \n\t"\
166
        "sub    $0xFFFF     , "tmp"                                     \n\t"\
167 168 169 170
        "neg    %%ecx                                                   \n\t"\
        "add    $7          , %%ecx                                     \n\t"\
        "shl    %%cl        , "tmp"                                     \n\t"\
        "add    "tmp"       , "low"                                     \n\t"\
171
        "2:                                                             \n\t"
172

173
#endif /* BROKEN_RELOCATIONS */
174

175
#if HAVE_7REGS && !BROKEN_COMPILER
176 177 178 179
#define get_cabac_inline get_cabac_inline_x86
static av_always_inline int get_cabac_inline_x86(CABACContext *c,
                                                 uint8_t *const state)
{
180
    int bit, tmp;
181 182 183 184 185 186
#ifdef BROKEN_RELOCATIONS
    void *tables;

    __asm__ volatile(
        "lea    "MANGLE(ff_h264_cabac_tables)", %0      \n\t"
        : "=&r"(tables)
187
        : NAMED_CONSTRAINTS(ff_h264_cabac_tables)
188 189
    );
#endif
190 191

    __asm__ volatile(
192 193
        BRANCHLESS_GET_CABAC("%0", "%q0", "(%4)", "%1", "%w1",
                             "%2", "%q2", "%3", "%b3",
194 195 196 197 198
                             "%c6(%5)", "%c7(%5)",
                             AV_STRINGIFY(H264_NORM_SHIFT_OFFSET),
                             AV_STRINGIFY(H264_LPS_RANGE_OFFSET),
                             AV_STRINGIFY(H264_MLPS_STATE_OFFSET),
                             "%8")
199
        : "=&r"(bit), "=&r"(c->low), "=&r"(c->range), "=&q"(tmp)
200
        : "r"(state), "r"(c),
201
          "i"(offsetof(CABACContext, bytestream)),
202 203
          "i"(offsetof(CABACContext, bytestream_end))
          TABLES_ARG
204
          ,"1"(c->low), "2"(c->range)
205 206 207 208
        : "%"REG_c, "memory"
    );
    return bit & 1;
}
209
#endif /* HAVE_7REGS */
210

211
#if !BROKEN_COMPILER
212 213 214 215 216
#define get_cabac_bypass_sign get_cabac_bypass_sign_x86
static av_always_inline int get_cabac_bypass_sign_x86(CABACContext *c, int val)
{
    x86_reg tmp;
    __asm__ volatile(
217 218
        "movl        %c6(%2), %k1       \n\t"
        "movl        %c3(%2), %%eax     \n\t"
219 220 221 222 223 224 225 226 227 228
        "shl             $17, %k1       \n\t"
        "add           %%eax, %%eax     \n\t"
        "sub             %k1, %%eax     \n\t"
        "cltd                           \n\t"
        "and           %%edx, %k1       \n\t"
        "add             %k1, %%eax     \n\t"
        "xor           %%edx, %%ecx     \n\t"
        "sub           %%edx, %%ecx     \n\t"
        "test           %%ax, %%ax      \n\t"
        "jnz              1f            \n\t"
229
        "mov         %c4(%2), %1        \n\t"
230 231 232 233
        "subl        $0xFFFF, %%eax     \n\t"
        "movzwl         (%1), %%edx     \n\t"
        "bswap         %%edx            \n\t"
        "shrl            $15, %%edx     \n\t"
234
#if UNCHECKED_BITSTREAM_READER
235
        "add              $2, %1        \n\t"
236
        "addl          %%edx, %%eax     \n\t"
237
        "mov              %1, %c4(%2)   \n\t"
238 239 240 241 242 243
#else
        "addl          %%edx, %%eax     \n\t"
        "cmp         %c5(%2), %1        \n\t"
        "jge              1f            \n\t"
        "add"OPSIZE"      $2, %c4(%2)   \n\t"
#endif
244
        "1:                             \n\t"
245
        "movl          %%eax, %c3(%2)   \n\t"
246

247 248 249 250
        : "+c"(val), "=&r"(tmp)
        : "r"(c),
          "i"(offsetof(CABACContext, low)),
          "i"(offsetof(CABACContext, bytestream)),
251
          "i"(offsetof(CABACContext, bytestream_end)),
252 253
          "i"(offsetof(CABACContext, range))
        : "%eax", "%edx", "memory"
254 255 256 257
    );
    return val;
}

258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296
#define get_cabac_bypass get_cabac_bypass_x86
static av_always_inline int get_cabac_bypass_x86(CABACContext *c)
{
    x86_reg tmp;
    int res;
    __asm__ volatile(
        "movl        %c6(%2), %k1       \n\t"
        "movl        %c3(%2), %%eax     \n\t"
        "shl             $17, %k1       \n\t"
        "add           %%eax, %%eax     \n\t"
        "sub             %k1, %%eax     \n\t"
        "cltd                           \n\t"
        "and           %%edx, %k1       \n\t"
        "add             %k1, %%eax     \n\t"
        "inc           %%edx            \n\t"
        "test           %%ax, %%ax      \n\t"
        "jnz              1f            \n\t"
        "mov         %c4(%2), %1        \n\t"
        "subl        $0xFFFF, %%eax     \n\t"
        "movzwl         (%1), %%ecx     \n\t"
        "bswap         %%ecx            \n\t"
        "shrl            $15, %%ecx     \n\t"
        "addl          %%ecx, %%eax     \n\t"
        "cmp         %c5(%2), %1        \n\t"
        "jge              1f            \n\t"
        "add"OPSIZE"      $2, %c4(%2)   \n\t"
        "1:                             \n\t"
        "movl          %%eax, %c3(%2)   \n\t"

        : "=&d"(res), "=&r"(tmp)
        : "r"(c),
          "i"(offsetof(CABACContext, low)),
          "i"(offsetof(CABACContext, bytestream)),
          "i"(offsetof(CABACContext, bytestream_end)),
          "i"(offsetof(CABACContext, range))
        : "%eax", "%ecx", "memory"
    );
    return res;
}
297
#endif /* !BROKEN_COMPILER */
298

299
#endif /* HAVE_INLINE_ASM */
300
#endif /* AVCODEC_X86_CABAC_H */