FFmpeg: libavcodec/x86/cabac.h Source File

00001 /*
00002  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
00003  *
00004  * This file is part of Libav.
00005  *
00006  * Libav is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * Libav is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with Libav; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #ifndef AVCODEC_X86_CABAC_H
00022 #define AVCODEC_X86_CABAC_H
00023 
00024 #include "libavcodec/cabac.h"
00025 #include "libavutil/attributes.h"
00026 #include "libavutil/x86/asm.h"
00027 #include "libavutil/internal.h"
00028 #include "config.h"
00029 
00030 #if HAVE_INLINE_ASM
00031 
00032 #ifdef BROKEN_RELOCATIONS
00033 #define TABLES_ARG , "r"(tables)
00034 
00035 #if HAVE_FAST_CMOV
00036 #define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
00037         "cmp    "low"       , "tmp"                        \n\t"\
00038         "cmova  %%ecx       , "range"                      \n\t"\
00039         "sbb    %%rcx       , %%rcx                        \n\t"\
00040         "and    %%ecx       , "tmp"                        \n\t"\
00041         "xor    %%rcx       , "retq"                       \n\t"\
00042         "sub    "tmp"       , "low"                        \n\t"
00043 #else /* HAVE_FAST_CMOV */
00044 #define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
00045 /* P4 Prescott has crappy cmov,sbb,64bit shift so avoid them */ \
00046         "sub    "low"       , "tmp"                        \n\t"\
00047         "sar    $31         , "tmp"                        \n\t"\
00048         "sub    %%ecx       , "range"                      \n\t"\
00049         "and    "tmp"       , "range"                      \n\t"\
00050         "add    %%ecx       , "range"                      \n\t"\
00051         "shl    $17         , %%ecx                        \n\t"\
00052         "and    "tmp"       , %%ecx                        \n\t"\
00053         "sub    %%ecx       , "low"                        \n\t"\
00054         "xor    "tmp"       , "ret"                        \n\t"\
00055         "movslq "ret"       , "retq"                       \n\t"
00056 #endif /* HAVE_FAST_CMOV */
00057 
00058 #define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \
00059         "movzbl "statep"    , "ret"                                     \n\t"\
00060         "mov    "range"     , "tmp"                                     \n\t"\
00061         "and    $0xC0       , "range"                                   \n\t"\
00062         "lea    ("ret", "range", 2), %%ecx                              \n\t"\
00063         "movzbl "lps_off"("tables", %%rcx), "range"                     \n\t"\
00064         "sub    "range"     , "tmp"                                     \n\t"\
00065         "mov    "tmp"       , %%ecx                                     \n\t"\
00066         "shl    $17         , "tmp"                                     \n\t"\
00067         BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp)              \
00068         "movzbl "norm_off"("tables", "rangeq"), %%ecx                   \n\t"\
00069         "shl    %%cl        , "range"                                   \n\t"\
00070         "movzbl "mlps_off"+128("tables", "retq"), "tmp"                 \n\t"\
00071         "shl    %%cl        , "low"                                     \n\t"\
00072         "mov    "tmpbyte"   , "statep"                                  \n\t"\
00073         "test   "lowword"   , "lowword"                                 \n\t"\
00074         "jnz    2f                                                      \n\t"\
00075         "mov    "byte"      , %%"REG_c"                                 \n\t"\
00076         "add"OPSIZE" $2     , "byte"                                    \n\t"\
00077         "movzwl (%%"REG_c") , "tmp"                                     \n\t"\
00078         "lea    -1("low")   , %%ecx                                     \n\t"\
00079         "xor    "low"       , %%ecx                                     \n\t"\
00080         "shr    $15         , %%ecx                                     \n\t"\
00081         "bswap  "tmp"                                                   \n\t"\
00082         "shr    $15         , "tmp"                                     \n\t"\
00083         "movzbl "norm_off"("tables", %%rcx), %%ecx                      \n\t"\
00084         "sub    $0xFFFF     , "tmp"                                     \n\t"\
00085         "neg    %%ecx                                                   \n\t"\
00086         "add    $7          , %%ecx                                     \n\t"\
00087         "shl    %%cl        , "tmp"                                     \n\t"\
00088         "add    "tmp"       , "low"                                     \n\t"\
00089         "2:                                                             \n\t"
00090 
00091 #else /* BROKEN_RELOCATIONS */
00092 #define TABLES_ARG
00093 #define RIP_ARG
00094 
00095 #if HAVE_FAST_CMOV
00096 #define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\
00097         "mov    "tmp"       , %%ecx     \n\t"\
00098         "shl    $17         , "tmp"     \n\t"\
00099         "cmp    "low"       , "tmp"     \n\t"\
00100         "cmova  %%ecx       , "range"   \n\t"\
00101         "sbb    %%ecx       , %%ecx     \n\t"\
00102         "and    %%ecx       , "tmp"     \n\t"\
00103         "xor    %%ecx       , "ret"     \n\t"\
00104         "sub    "tmp"       , "low"     \n\t"
00105 #else /* HAVE_FAST_CMOV */
00106 #define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\
00107         "mov    "tmp"       , %%ecx     \n\t"\
00108         "shl    $17         , "tmp"     \n\t"\
00109         "sub    "low"       , "tmp"     \n\t"\
00110         "sar    $31         , "tmp"     \n\t" /*lps_mask*/\
00111         "sub    %%ecx       , "range"   \n\t" /*RangeLPS - range*/\
00112         "and    "tmp"       , "range"   \n\t" /*(RangeLPS - range)&lps_mask*/\
00113         "add    %%ecx       , "range"   \n\t" /*new range*/\
00114         "shl    $17         , %%ecx     \n\t"\
00115         "and    "tmp"       , %%ecx     \n\t"\
00116         "sub    %%ecx       , "low"     \n\t"\
00117         "xor    "tmp"       , "ret"     \n\t"
00118 #endif /* HAVE_FAST_CMOV */
00119 
00120 #define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, norm_off, lps_off, mlps_off, tables) \
00121         "movzbl "statep"    , "ret"                                     \n\t"\
00122         "mov    "range"     , "tmp"                                     \n\t"\
00123         "and    $0xC0       , "range"                                   \n\t"\
00124         "movzbl "MANGLE(ff_h264_cabac_tables)"+"lps_off"("ret", "range", 2), "range" \n\t"\
00125         "sub    "range"     , "tmp"                                     \n\t"\
00126         BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)                    \
00127         "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"("range"), %%ecx    \n\t"\
00128         "shl    %%cl        , "range"                                   \n\t"\
00129         "movzbl "MANGLE(ff_h264_cabac_tables)"+"mlps_off"+128("ret"), "tmp"  \n\t"\
00130         "shl    %%cl        , "low"                                     \n\t"\
00131         "mov    "tmpbyte"   , "statep"                                  \n\t"\
00132         "test   "lowword"   , "lowword"                                 \n\t"\
00133         " jnz   2f                                                      \n\t"\
00134         "mov    "byte"      , %%"REG_c"                                 \n\t"\
00135         "add"OPSIZE" $2     , "byte"                                    \n\t"\
00136         "movzwl (%%"REG_c")     , "tmp"                                 \n\t"\
00137         "lea    -1("low")   , %%ecx                                     \n\t"\
00138         "xor    "low"       , %%ecx                                     \n\t"\
00139         "shr    $15         , %%ecx                                     \n\t"\
00140         "bswap  "tmp"                                                   \n\t"\
00141         "shr    $15         , "tmp"                                     \n\t"\
00142         "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"(%%ecx), %%ecx \n\t"\
00143         "sub    $0xFFFF     , "tmp"                                     \n\t"\
00144         "neg    %%ecx                                                   \n\t"\
00145         "add    $7          , %%ecx                                     \n\t"\
00146         "shl    %%cl        , "tmp"                                     \n\t"\
00147         "add    "tmp"       , "low"                                     \n\t"\
00148         "2:                                                             \n\t"
00149 
00150 #endif /* BROKEN_RELOCATIONS */
00151 
00152 
00153 #if HAVE_7REGS && !(defined(__i386) && defined(__clang__) && (__clang_major__<2 || (__clang_major__==2 && __clang_minor__<10)))\
00154                && !(defined(__i386) && !defined(__clang__) && defined(__llvm__) && __GNUC__==4 && __GNUC_MINOR__==2 && __GNUC_PATCHLEVEL__<=1)
00155 #define get_cabac_inline get_cabac_inline_x86
00156 static av_always_inline int get_cabac_inline_x86(CABACContext *c,
00157                                                  uint8_t *const state)
00158 {
00159     int bit, tmp;
00160 #ifdef BROKEN_RELOCATIONS
00161     void *tables;
00162 
00163     __asm__ volatile(
00164         "lea    "MANGLE(ff_h264_cabac_tables)", %0      \n\t"
00165         : "=&r"(tables)
00166     );
00167 #endif
00168 
00169     __asm__ volatile(
00170         BRANCHLESS_GET_CABAC("%0", "%q0", "(%4)", "%1", "%w1",
00171                              "%2", "%q2", "%3", "%b3",
00172                              "%c6(%5)", "%c7(%5)",
00173                              AV_STRINGIFY(H264_NORM_SHIFT_OFFSET),
00174                              AV_STRINGIFY(H264_LPS_RANGE_OFFSET),
00175                              AV_STRINGIFY(H264_MLPS_STATE_OFFSET),
00176                              "%8")
00177         : "=&r"(bit), "=&r"(c->low), "=&r"(c->range), "=&q"(tmp)
00178         : "r"(state), "r"(c),
00179           "i"(offsetof(CABACContext, bytestream)),
00180           "i"(offsetof(CABACContext, bytestream_end))
00181           TABLES_ARG
00182           ,"1"(c->low), "2"(c->range)
00183         : "%"REG_c, "memory"
00184     );
00185     return bit & 1;
00186 }
00187 #endif /* HAVE_7REGS */
00188 
00189 #define get_cabac_bypass_sign get_cabac_bypass_sign_x86
00190 static av_always_inline int get_cabac_bypass_sign_x86(CABACContext *c, int val)
00191 {
00192     x86_reg tmp;
00193     __asm__ volatile(
00194         "movl        %c6(%2), %k1       \n\t"
00195         "movl        %c3(%2), %%eax     \n\t"
00196         "shl             $17, %k1       \n\t"
00197         "add           %%eax, %%eax     \n\t"
00198         "sub             %k1, %%eax     \n\t"
00199         "cltd                           \n\t"
00200         "and           %%edx, %k1       \n\t"
00201         "add             %k1, %%eax     \n\t"
00202         "xor           %%edx, %%ecx     \n\t"
00203         "sub           %%edx, %%ecx     \n\t"
00204         "test           %%ax, %%ax      \n\t"
00205         "jnz              1f            \n\t"
00206         "mov         %c4(%2), %1        \n\t"
00207         "subl        $0xFFFF, %%eax     \n\t"
00208         "movzwl         (%1), %%edx     \n\t"
00209         "bswap         %%edx            \n\t"
00210         "shrl            $15, %%edx     \n\t"
00211         "add              $2, %1        \n\t"
00212         "addl          %%edx, %%eax     \n\t"
00213         "mov              %1, %c4(%2)   \n\t"
00214         "1:                             \n\t"
00215         "movl          %%eax, %c3(%2)   \n\t"
00216 
00217         : "+c"(val), "=&r"(tmp)
00218         : "r"(c),
00219           "i"(offsetof(CABACContext, low)),
00220           "i"(offsetof(CABACContext, bytestream)),
00221           "i"(offsetof(CABACContext, bytestream_end)),
00222           "i"(offsetof(CABACContext, range))
00223         : "%eax", "%edx", "memory"
00224     );
00225     return val;
00226 }
00227 
00228 #endif /* HAVE_INLINE_ASM */
00229 #endif /* AVCODEC_X86_CABAC_H */