| 1 | /* |
| 2 | * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder |
| 3 | * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at> |
| 4 | * |
| 5 | * This file is part of FFmpeg. |
| 6 | * |
| 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. |
| 11 | * |
| 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. |
| 16 | * |
| 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 |
| 20 | */ |
| 21 | |
| 22 | /** |
| 23 | * @file |
| 24 | * H.264 / AVC / MPEG4 part10 codec. |
| 25 | * non-MMX i386-specific optimizations for H.264 |
| 26 | * @author Michael Niedermayer <michaelni@gmx.at> |
| 27 | */ |
| 28 | |
| 29 | #ifndef AVCODEC_X86_H264_I386_H |
| 30 | #define AVCODEC_X86_H264_I386_H |
| 31 | |
| 32 | #include <stddef.h> |
| 33 | |
| 34 | #include "libavcodec/cabac.h" |
| 35 | #include "cabac.h" |
| 36 | |
| 37 | #if HAVE_INLINE_ASM |
| 38 | |
| 39 | #if ARCH_X86_64 |
| 40 | #define REG64 "r" |
| 41 | #else |
| 42 | #define REG64 "m" |
| 43 | #endif |
| 44 | |
| 45 | //FIXME use some macros to avoid duplicating get_cabac (cannot be done yet |
| 46 | //as that would make optimization work hard) |
| 47 | #if HAVE_7REGS && !BROKEN_COMPILER |
| 48 | #define decode_significance decode_significance_x86 |
| 49 | static int decode_significance_x86(CABACContext *c, int max_coeff, |
| 50 | uint8_t *significant_coeff_ctx_base, |
| 51 | int *index, x86_reg last_off){ |
| 52 | void *end= significant_coeff_ctx_base + max_coeff - 1; |
| 53 | int minusstart= -(intptr_t)significant_coeff_ctx_base; |
| 54 | int minusindex= 4-(intptr_t)index; |
| 55 | int bit; |
| 56 | x86_reg coeff_count; |
| 57 | |
| 58 | #ifdef BROKEN_RELOCATIONS |
| 59 | void *tables; |
| 60 | |
| 61 | __asm__ volatile( |
| 62 | "lea "MANGLE(ff_h264_cabac_tables)", %0 \n\t" |
| 63 | : "=&r"(tables) |
| 64 | : NAMED_CONSTRAINTS_ARRAY(ff_h264_cabac_tables) |
| 65 | ); |
| 66 | #endif |
| 67 | |
| 68 | __asm__ volatile( |
| 69 | "3: \n\t" |
| 70 | |
| 71 | BRANCHLESS_GET_CABAC("%4", "%q4", "(%1)", "%3", "%w3", |
| 72 | "%5", "%q5", "%k0", "%b0", |
| 73 | "%c11(%6)", "%c12(%6)", |
| 74 | AV_STRINGIFY(H264_NORM_SHIFT_OFFSET), |
| 75 | AV_STRINGIFY(H264_LPS_RANGE_OFFSET), |
| 76 | AV_STRINGIFY(H264_MLPS_STATE_OFFSET), |
| 77 | "%13") |
| 78 | |
| 79 | "test $1, %4 \n\t" |
| 80 | " jz 4f \n\t" |
| 81 | "add %10, %1 \n\t" |
| 82 | |
| 83 | BRANCHLESS_GET_CABAC("%4", "%q4", "(%1)", "%3", "%w3", |
| 84 | "%5", "%q5", "%k0", "%b0", |
| 85 | "%c11(%6)", "%c12(%6)", |
| 86 | AV_STRINGIFY(H264_NORM_SHIFT_OFFSET), |
| 87 | AV_STRINGIFY(H264_LPS_RANGE_OFFSET), |
| 88 | AV_STRINGIFY(H264_MLPS_STATE_OFFSET), |
| 89 | "%13") |
| 90 | |
| 91 | "sub %10, %1 \n\t" |
| 92 | "mov %2, %0 \n\t" |
| 93 | "movl %7, %%ecx \n\t" |
| 94 | "add %1, %%"REG_c" \n\t" |
| 95 | "movl %%ecx, (%0) \n\t" |
| 96 | |
| 97 | "test $1, %4 \n\t" |
| 98 | " jnz 5f \n\t" |
| 99 | |
| 100 | "add"OPSIZE" $4, %2 \n\t" |
| 101 | |
| 102 | "4: \n\t" |
| 103 | "add $1, %1 \n\t" |
| 104 | "cmp %8, %1 \n\t" |
| 105 | " jb 3b \n\t" |
| 106 | "mov %2, %0 \n\t" |
| 107 | "movl %7, %%ecx \n\t" |
| 108 | "add %1, %%"REG_c" \n\t" |
| 109 | "movl %%ecx, (%0) \n\t" |
| 110 | "5: \n\t" |
| 111 | "add %9, %k0 \n\t" |
| 112 | "shr $2, %k0 \n\t" |
| 113 | : "=&q"(coeff_count), "+r"(significant_coeff_ctx_base), "+m"(index), |
| 114 | "+&r"(c->low), "=&r"(bit), "+&r"(c->range) |
| 115 | : "r"(c), "m"(minusstart), "m"(end), "m"(minusindex), "m"(last_off), |
| 116 | "i"(offsetof(CABACContext, bytestream)), |
| 117 | "i"(offsetof(CABACContext, bytestream_end)) |
| 118 | TABLES_ARG |
| 119 | : "%"REG_c, "memory" |
| 120 | ); |
| 121 | return coeff_count; |
| 122 | } |
| 123 | |
| 124 | #define decode_significance_8x8 decode_significance_8x8_x86 |
| 125 | static int decode_significance_8x8_x86(CABACContext *c, |
| 126 | uint8_t *significant_coeff_ctx_base, |
| 127 | int *index, uint8_t *last_coeff_ctx_base, const uint8_t *sig_off){ |
| 128 | int minusindex= 4-(intptr_t)index; |
| 129 | int bit; |
| 130 | x86_reg coeff_count; |
| 131 | x86_reg last=0; |
| 132 | x86_reg state; |
| 133 | |
| 134 | #ifdef BROKEN_RELOCATIONS |
| 135 | void *tables; |
| 136 | |
| 137 | __asm__ volatile( |
| 138 | "lea "MANGLE(ff_h264_cabac_tables)", %0 \n\t" |
| 139 | : "=&r"(tables) |
| 140 | : NAMED_CONSTRAINTS_ARRAY(ff_h264_cabac_tables) |
| 141 | ); |
| 142 | #endif |
| 143 | |
| 144 | __asm__ volatile( |
| 145 | "mov %1, %6 \n\t" |
| 146 | "3: \n\t" |
| 147 | |
| 148 | "mov %10, %0 \n\t" |
| 149 | "movzb (%0, %6), %6 \n\t" |
| 150 | "add %9, %6 \n\t" |
| 151 | |
| 152 | BRANCHLESS_GET_CABAC("%4", "%q4", "(%6)", "%3", "%w3", |
| 153 | "%5", "%q5", "%k0", "%b0", |
| 154 | "%c12(%7)", "%c13(%7)", |
| 155 | AV_STRINGIFY(H264_NORM_SHIFT_OFFSET), |
| 156 | AV_STRINGIFY(H264_LPS_RANGE_OFFSET), |
| 157 | AV_STRINGIFY(H264_MLPS_STATE_OFFSET), |
| 158 | "%15") |
| 159 | |
| 160 | "mov %1, %6 \n\t" |
| 161 | "test $1, %4 \n\t" |
| 162 | " jz 4f \n\t" |
| 163 | |
| 164 | #ifdef BROKEN_RELOCATIONS |
| 165 | "movzb %c14(%15, %q6), %6\n\t" |
| 166 | #else |
| 167 | "movzb "MANGLE(ff_h264_cabac_tables)"+%c14(%6), %6\n\t" |
| 168 | #endif |
| 169 | "add %11, %6 \n\t" |
| 170 | |
| 171 | BRANCHLESS_GET_CABAC("%4", "%q4", "(%6)", "%3", "%w3", |
| 172 | "%5", "%q5", "%k0", "%b0", |
| 173 | "%c12(%7)", "%c13(%7)", |
| 174 | AV_STRINGIFY(H264_NORM_SHIFT_OFFSET), |
| 175 | AV_STRINGIFY(H264_LPS_RANGE_OFFSET), |
| 176 | AV_STRINGIFY(H264_MLPS_STATE_OFFSET), |
| 177 | "%15") |
| 178 | |
| 179 | "mov %2, %0 \n\t" |
| 180 | "mov %1, %6 \n\t" |
| 181 | "mov %k6, (%0) \n\t" |
| 182 | |
| 183 | "test $1, %4 \n\t" |
| 184 | " jnz 5f \n\t" |
| 185 | |
| 186 | "add"OPSIZE" $4, %2 \n\t" |
| 187 | |
| 188 | "4: \n\t" |
| 189 | "add $1, %6 \n\t" |
| 190 | "mov %6, %1 \n\t" |
| 191 | "cmp $63, %6 \n\t" |
| 192 | " jb 3b \n\t" |
| 193 | "mov %2, %0 \n\t" |
| 194 | "mov %k6, (%0) \n\t" |
| 195 | "5: \n\t" |
| 196 | "addl %8, %k0 \n\t" |
| 197 | "shr $2, %k0 \n\t" |
| 198 | : "=&q"(coeff_count), "+"REG64(last), "+"REG64(index), "+&r"(c->low), |
| 199 | "=&r"(bit), "+&r"(c->range), "=&r"(state) |
| 200 | : "r"(c), "m"(minusindex), "m"(significant_coeff_ctx_base), |
| 201 | REG64(sig_off), REG64(last_coeff_ctx_base), |
| 202 | "i"(offsetof(CABACContext, bytestream)), |
| 203 | "i"(offsetof(CABACContext, bytestream_end)), |
| 204 | "i"(H264_LAST_COEFF_FLAG_OFFSET_8x8_OFFSET) TABLES_ARG |
| 205 | : "%"REG_c, "memory" |
| 206 | ); |
| 207 | return coeff_count; |
| 208 | } |
| 209 | #endif /* HAVE_7REGS && BROKEN_COMPILER */ |
| 210 | |
| 211 | #endif /* HAVE_INLINE_ASM */ |
| 212 | #endif /* AVCODEC_X86_H264_I386_H */ |