ba3a5ddfad0608dd0d57cbf229daf8ae2eb8c92a
[libav.git] / libavcodec / x86 / h264_i386.h
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 Libav.
6 *
7 * Libav 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 * Libav 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 Libav; 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
36 //FIXME use some macros to avoid duplicating get_cabac (cannot be done yet
37 //as that would make optimization work hard)
38 #if ARCH_X86 && HAVE_7REGS && !defined(BROKEN_RELOCATIONS)
39 static int decode_significance_x86(CABACContext *c, int max_coeff,
40 uint8_t *significant_coeff_ctx_base,
41 int *index, x86_reg last_off){
42 void *end= significant_coeff_ctx_base + max_coeff - 1;
43 int minusstart= -(int)significant_coeff_ctx_base;
44 int minusindex= 4-(int)index;
45 int coeff_count;
46 int low;
47 __asm__ volatile(
48 "movl %a9(%4), %%esi \n\t"
49 "movl %a10(%4), %3 \n\t"
50
51 "2: \n\t"
52
53 BRANCHLESS_GET_CABAC("%%edx", "%4", "(%1)", "%3",
54 "%w3", "%%esi", "%%eax", "%%al", "%a11")
55
56 "test $1, %%edx \n\t"
57 " jz 3f \n\t"
58 "add %8, %1 \n\t"
59
60 BRANCHLESS_GET_CABAC("%%edx", "%4", "(%1)", "%3",
61 "%w3", "%%esi", "%%eax", "%%al", "%a11")
62
63 "sub %8, %1 \n\t"
64 "mov %2, %%"REG_a" \n\t"
65 "movl %5, %%ecx \n\t"
66 "add %1, %%"REG_c" \n\t"
67 "movl %%ecx, (%%"REG_a") \n\t"
68
69 "test $1, %%edx \n\t"
70 " jnz 4f \n\t"
71
72 "add $4, %%"REG_a" \n\t"
73 "mov %%"REG_a", %2 \n\t"
74
75 "3: \n\t"
76 "add $1, %1 \n\t"
77 "cmp %6, %1 \n\t"
78 " jb 2b \n\t"
79 "mov %2, %%"REG_a" \n\t"
80 "movl %5, %%ecx \n\t"
81 "add %1, %%"REG_c" \n\t"
82 "movl %%ecx, (%%"REG_a") \n\t"
83 "4: \n\t"
84 "add %7, %%eax \n\t"
85 "shr $2, %%eax \n\t"
86
87 "movl %%esi, %a9(%4) \n\t"
88 "movl %3, %a10(%4) \n\t"
89 :"=&a"(coeff_count), "+r"(significant_coeff_ctx_base), "+m"(index),
90 "=&r"(low)
91 :"r"(c), "m"(minusstart), "m"(end), "m"(minusindex), "m"(last_off),
92 "i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
93 "i"(offsetof(CABACContext, bytestream))
94 : "%"REG_c, "%edx", "%esi", "memory"
95 );
96 return coeff_count;
97 }
98
99 static int decode_significance_8x8_x86(CABACContext *c,
100 uint8_t *significant_coeff_ctx_base,
101 int *index, x86_reg last_off, const uint8_t *sig_off){
102 int minusindex= 4-(int)index;
103 int coeff_count;
104 int low;
105 x86_reg last=0;
106 __asm__ volatile(
107 "movl %a9(%4), %%esi \n\t"
108 "movl %a10(%4), %3 \n\t"
109
110 "mov %1, %%"REG_D" \n\t"
111 "2: \n\t"
112
113 "mov %7, %%"REG_a" \n\t"
114 "movzbl (%%"REG_a", %%"REG_D"), %%edi \n\t"
115 "add %6, %%"REG_D" \n\t"
116
117 BRANCHLESS_GET_CABAC("%%edx", "%4", "(%%"REG_D")", "%3",
118 "%w3", "%%esi", "%%eax", "%%al", "%a11")
119
120 "mov %1, %%edi \n\t"
121 "test $1, %%edx \n\t"
122 " jz 3f \n\t"
123
124 "movzbl "MANGLE(last_coeff_flag_offset_8x8)"(%%edi), %%edi\n\t"
125 "add %6, %%"REG_D" \n\t"
126 "add %8, %%"REG_D" \n\t"
127
128 BRANCHLESS_GET_CABAC("%%edx", "%4", "(%%"REG_D")", "%3",
129 "%w3", "%%esi", "%%eax", "%%al", "%a11")
130
131 "mov %2, %%"REG_a" \n\t"
132 "mov %1, %%edi \n\t"
133 "movl %%edi, (%%"REG_a") \n\t"
134
135 "test $1, %%edx \n\t"
136 " jnz 4f \n\t"
137
138 "add $4, %%"REG_a" \n\t"
139 "mov %%"REG_a", %2 \n\t"
140
141 "3: \n\t"
142 "addl $1, %%edi \n\t"
143 "mov %%edi, %1 \n\t"
144 "cmpl $63, %%edi \n\t"
145 " jb 2b \n\t"
146 "mov %2, %%"REG_a" \n\t"
147 "movl %%edi, (%%"REG_a") \n\t"
148 "4: \n\t"
149 "addl %5, %%eax \n\t"
150 "shr $2, %%eax \n\t"
151
152 "movl %%esi, %a9(%4) \n\t"
153 "movl %3, %a10(%4) \n\t"
154 :"=&a"(coeff_count),"+m"(last), "+m"(index), "=&r"(low)
155 :"r"(c), "m"(minusindex), "m"(significant_coeff_ctx_base), "m"(sig_off), "m"(last_off),
156 "i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
157 "i"(offsetof(CABACContext, bytestream))
158 : "%"REG_c, "%edx", "%esi", "%"REG_D, "memory"
159 );
160 return coeff_count;
161 }
162 #endif /* ARCH_X86 && HAVE_7REGS && !defined(BROKEN_RELOCATIONS) */
163
164 #endif /* AVCODEC_X86_H264_I386_H */