x86: Require 7 registers for the cabac asm
[libav.git] / libavcodec / x86 / cabac.h
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 "config.h"
28
29 #if HAVE_FAST_CMOV
30 #define BRANCHLESS_GET_CABAC_UPDATE(ret, statep, low, lowword, range, tmp)\
31 "mov "tmp" , %%ecx \n\t"\
32 "shl $17 , "tmp" \n\t"\
33 "cmp "low" , "tmp" \n\t"\
34 "cmova %%ecx , "range" \n\t"\
35 "sbb %%ecx , %%ecx \n\t"\
36 "and %%ecx , "tmp" \n\t"\
37 "xor %%ecx , "ret" \n\t"\
38 "sub "tmp" , "low" \n\t"
39 #else /* HAVE_FAST_CMOV */
40 #define BRANCHLESS_GET_CABAC_UPDATE(ret, statep, low, lowword, range, tmp)\
41 "mov "tmp" , %%ecx \n\t"\
42 "shl $17 , "tmp" \n\t"\
43 "sub "low" , "tmp" \n\t"\
44 "sar $31 , "tmp" \n\t" /*lps_mask*/\
45 "sub %%ecx , "range" \n\t" /*RangeLPS - range*/\
46 "and "tmp" , "range" \n\t" /*(RangeLPS - range)&lps_mask*/\
47 "add %%ecx , "range" \n\t" /*new range*/\
48 "shl $17 , %%ecx \n\t"\
49 "and "tmp" , %%ecx \n\t"\
50 "sub %%ecx , "low" \n\t"\
51 "xor "tmp" , "ret" \n\t"
52 #endif /* HAVE_FAST_CMOV */
53
54 #define BRANCHLESS_GET_CABAC(ret, statep, low, lowword, range, tmp, tmpbyte, byte) \
55 "movzbl "statep" , "ret" \n\t"\
56 "mov "range" , "tmp" \n\t"\
57 "and $0xC0 , "range" \n\t"\
58 "movzbl "MANGLE(ff_h264_lps_range)"("ret", "range", 2), "range" \n\t"\
59 "sub "range" , "tmp" \n\t"\
60 BRANCHLESS_GET_CABAC_UPDATE(ret, statep, low, lowword, range, tmp) \
61 "movzbl " MANGLE(ff_h264_norm_shift) "("range"), %%ecx \n\t"\
62 "shl %%cl , "range" \n\t"\
63 "movzbl "MANGLE(ff_h264_mlps_state)"+128("ret"), "tmp" \n\t"\
64 "shl %%cl , "low" \n\t"\
65 "mov "tmpbyte" , "statep" \n\t"\
66 "test "lowword" , "lowword" \n\t"\
67 " jnz 1f \n\t"\
68 "mov "byte" , %%"REG_c" \n\t"\
69 "add"OPSIZE" $2 , "byte" \n\t"\
70 "movzwl (%%"REG_c") , "tmp" \n\t"\
71 "lea -1("low") , %%ecx \n\t"\
72 "xor "low" , %%ecx \n\t"\
73 "shr $15 , %%ecx \n\t"\
74 "bswap "tmp" \n\t"\
75 "shr $15 , "tmp" \n\t"\
76 "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"\
77 "sub $0xFFFF , "tmp" \n\t"\
78 "neg %%ecx \n\t"\
79 "add $7 , %%ecx \n\t"\
80 "shl %%cl , "tmp" \n\t"\
81 "add "tmp" , "low" \n\t"\
82 "1: \n\t"
83
84 #if HAVE_7REGS && !defined(BROKEN_RELOCATIONS)
85 #define get_cabac_inline get_cabac_inline_x86
86 static av_always_inline int get_cabac_inline_x86(CABACContext *c,
87 uint8_t *const state)
88 {
89 int bit, tmp;
90
91 __asm__ volatile(
92 BRANCHLESS_GET_CABAC("%0", "(%5)", "%1", "%w1", "%2",
93 "%3", "%b3", "%4")
94 :"=&r"(bit), "+&r"(c->low), "+&r"(c->range), "=&q"(tmp),
95 "+m"(c->bytestream)
96 :"r"(state)
97 : "%"REG_c, "memory"
98 );
99 return bit & 1;
100 }
101 #endif /* HAVE_7REGS && !defined(BROKEN_RELOCATIONS) */
102
103 #define get_cabac_bypass_sign get_cabac_bypass_sign_x86
104 static av_always_inline int get_cabac_bypass_sign_x86(CABACContext *c, int val)
105 {
106 x86_reg tmp;
107 __asm__ volatile(
108 "movl %4, %k1 \n\t"
109 "movl %2, %%eax \n\t"
110 "shl $17, %k1 \n\t"
111 "add %%eax, %%eax \n\t"
112 "sub %k1, %%eax \n\t"
113 "cltd \n\t"
114 "and %%edx, %k1 \n\t"
115 "add %k1, %%eax \n\t"
116 "xor %%edx, %%ecx \n\t"
117 "sub %%edx, %%ecx \n\t"
118 "test %%ax, %%ax \n\t"
119 " jnz 1f \n\t"
120 "mov %3, %1 \n\t"
121 "subl $0xFFFF, %%eax \n\t"
122 "movzwl (%1), %%edx \n\t"
123 "bswap %%edx \n\t"
124 "shrl $15, %%edx \n\t"
125 "add $2, %1 \n\t"
126 "addl %%edx, %%eax \n\t"
127 "mov %1, %3 \n\t"
128 "1: \n\t"
129 "movl %%eax, %2 \n\t"
130
131 :"+c"(val), "=&r"(tmp), "+m"(c->low), "+m"(c->bytestream)
132 :"m"(c->range)
133 : "%eax", "%edx"
134 );
135 return val;
136 }
137
138 #endif /* AVCODEC_X86_CABAC_H */