ffmpeg / libavcodec / arm / vp8_armv6.S @ 4ae3ee4a
History | View | Annotate | Download (7.61 KB)
| 1 |
/** |
|---|---|
| 2 |
* Copyright (C) 2010 Mans Rullgard |
| 3 |
* |
| 4 |
* This file is part of FFmpeg. |
| 5 |
* |
| 6 |
* FFmpeg 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 |
* FFmpeg 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 FFmpeg; if not, write to the Free Software |
| 18 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 |
*/ |
| 20 |
|
| 21 |
#include "asm.S" |
| 22 |
|
| 23 |
.syntax unified |
| 24 |
|
| 25 |
.macro rac_get_prob h, bs, buf, cw, pr, t0, t1 |
| 26 |
adds \bs, \bs, \t0 |
| 27 |
lsl \cw, \cw, \t0 |
| 28 |
lsl \t0, \h, \t0 |
| 29 |
rsb \h, \pr, #256 |
| 30 |
ldrhcs \t1, [\buf], #2 |
| 31 |
smlabb \h, \t0, \pr, \h |
| 32 |
rev16cs \t1, \t1 |
| 33 |
orrcs \cw, \cw, \t1, lsl \bs |
| 34 |
subcs \bs, \bs, #16 |
| 35 |
lsr \h, \h, #8 |
| 36 |
cmp \cw, \h, lsl #16 |
| 37 |
subge \cw, \cw, \h, lsl #16 |
| 38 |
subge \h, \t0, \h |
| 39 |
.endm |
| 40 |
|
| 41 |
.macro rac_get_128 h, bs, buf, cw, t0, t1 |
| 42 |
adds \bs, \bs, \t0 |
| 43 |
lsl \cw, \cw, \t0 |
| 44 |
lsl \t0, \h, \t0 |
| 45 |
ldrhcs \t1, [\buf], #2 |
| 46 |
mov \h, #128 |
| 47 |
rev16cs \t1, \t1 |
| 48 |
add \h, \h, \t0, lsl #7 |
| 49 |
orrcs \cw, \cw, \t1, lsl \bs |
| 50 |
subcs \bs, \bs, #16 |
| 51 |
lsr \h, \h, #8 |
| 52 |
cmp \cw, \h, lsl #16 |
| 53 |
subge \cw, \cw, \h, lsl #16 |
| 54 |
subge \h, \t0, \h |
| 55 |
.endm |
| 56 |
|
| 57 |
function ff_decode_block_coeffs_armv6, export=1 |
| 58 |
push {r0,r1,r4-r11,lr}
|
| 59 |
movrel lr, ff_vp56_norm_shift |
| 60 |
ldrd r4, r5, [sp, #44] @ token_prob, qmul |
| 61 |
cmp r3, #0 |
| 62 |
ldr r11, [r5] |
| 63 |
ldm r0, {r5-r7} @ high, bits, buf
|
| 64 |
pkhtbne r11, r11, r11, asr #16 |
| 65 |
ldr r8, [r0, #16] @ code_word |
| 66 |
0: |
| 67 |
ldrb r9, [lr, r5] |
| 68 |
add r3, r3, #1 |
| 69 |
ldrb r0, [r4, #1] |
| 70 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 71 |
blt 2f |
| 72 |
|
| 73 |
ldrb r9, [lr, r5] |
| 74 |
ldrb r0, [r4, #2] |
| 75 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 76 |
ldrb r9, [lr, r5] |
| 77 |
bge 3f |
| 78 |
|
| 79 |
add r4, r3, r3, lsl #5 |
| 80 |
sxth r12, r11 |
| 81 |
add r4, r2, r4 |
| 82 |
adds r6, r6, r9 |
| 83 |
add r4, r4, #11 |
| 84 |
lsl r8, r8, r9 |
| 85 |
ldrhcs r10, [r7], #2 |
| 86 |
lsl r9, r5, r9 |
| 87 |
mov r5, #128 |
| 88 |
rev16cs r10, r10 |
| 89 |
add r5, r5, r9, lsl #7 |
| 90 |
orrcs r8, r8, r10, lsl r6 |
| 91 |
subcs r6, r6, #16 |
| 92 |
lsr r5, r5, #8 |
| 93 |
cmp r8, r5, lsl #16 |
| 94 |
movrel r10, zigzag_scan-1 |
| 95 |
subge r8, r8, r5, lsl #16 |
| 96 |
subge r5, r9, r5 |
| 97 |
ldrb r10, [r10, r3] |
| 98 |
rsbge r12, r12, #0 |
| 99 |
cmp r3, #16 |
| 100 |
strh r12, [r1, r10] |
| 101 |
bge 6f |
| 102 |
5: |
| 103 |
ldrb r9, [lr, r5] |
| 104 |
ldrb r0, [r4] |
| 105 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 106 |
pkhtb r11, r11, r11, asr #16 |
| 107 |
bge 0b |
| 108 |
|
| 109 |
6: |
| 110 |
ldr r0, [sp] |
| 111 |
ldr r9, [r0, #12] |
| 112 |
cmp r7, r9 |
| 113 |
movhi r7, r9 |
| 114 |
stm r0, {r5-r7} @ high, bits, buf
|
| 115 |
str r8, [r0, #16] @ code_word |
| 116 |
|
| 117 |
add sp, sp, #8 |
| 118 |
mov r0, r3 |
| 119 |
pop {r4-r11,pc}
|
| 120 |
2: |
| 121 |
add r4, r3, r3, lsl #5 |
| 122 |
cmp r3, #16 |
| 123 |
add r4, r2, r4 |
| 124 |
pkhtb r11, r11, r11, asr #16 |
| 125 |
bne 0b |
| 126 |
b 6b |
| 127 |
3: |
| 128 |
ldrb r0, [r4, #3] |
| 129 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 130 |
ldrb r9, [lr, r5] |
| 131 |
bge 1f |
| 132 |
|
| 133 |
mov r12, #2 |
| 134 |
ldrb r0, [r4, #4] |
| 135 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 136 |
addge r12, #1 |
| 137 |
ldrb r9, [lr, r5] |
| 138 |
blt 4f |
| 139 |
ldrb r0, [r4, #5] |
| 140 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 141 |
addge r12, #1 |
| 142 |
ldrb r9, [lr, r5] |
| 143 |
b 4f |
| 144 |
1: |
| 145 |
ldrb r0, [r4, #6] |
| 146 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 147 |
ldrb r9, [lr, r5] |
| 148 |
bge 3f |
| 149 |
|
| 150 |
ldrb r0, [r4, #7] |
| 151 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 152 |
ldrb r9, [lr, r5] |
| 153 |
bge 2f |
| 154 |
|
| 155 |
mov r12, #5 |
| 156 |
mov r0, #159 |
| 157 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 158 |
addge r12, r12, #1 |
| 159 |
ldrb r9, [lr, r5] |
| 160 |
b 4f |
| 161 |
2: |
| 162 |
mov r12, #7 |
| 163 |
mov r0, #165 |
| 164 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 165 |
addge r12, r12, #2 |
| 166 |
ldrb r9, [lr, r5] |
| 167 |
mov r0, #145 |
| 168 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 169 |
addge r12, r12, #1 |
| 170 |
ldrb r9, [lr, r5] |
| 171 |
b 4f |
| 172 |
3: |
| 173 |
ldrb r0, [r4, #8] |
| 174 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 175 |
addge r4, r4, #1 |
| 176 |
ldrb r9, [lr, r5] |
| 177 |
movge r12, #2 |
| 178 |
movlt r12, #0 |
| 179 |
ldrb r0, [r4, #9] |
| 180 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 181 |
mov r9, #8 |
| 182 |
addge r12, r12, #1 |
| 183 |
movrel r4, ff_vp8_dct_cat_prob |
| 184 |
lsl r9, r9, r12 |
| 185 |
ldr r4, [r4, r12, lsl #2] |
| 186 |
add r12, r9, #3 |
| 187 |
mov r1, #0 |
| 188 |
ldrb r0, [r4], #1 |
| 189 |
1: |
| 190 |
ldrb r9, [lr, r5] |
| 191 |
lsl r1, r1, #1 |
| 192 |
rac_get_prob r5, r6, r7, r8, r0, r9, r10 |
| 193 |
ldrb r0, [r4], #1 |
| 194 |
addge r1, r1, #1 |
| 195 |
cmp r0, #0 |
| 196 |
bne 1b |
| 197 |
ldrb r9, [lr, r5] |
| 198 |
add r12, r12, r1 |
| 199 |
ldr r1, [sp, #4] |
| 200 |
4: |
| 201 |
add r4, r3, r3, lsl #5 |
| 202 |
add r4, r2, r4 |
| 203 |
add r4, r4, #22 |
| 204 |
rac_get_128 r5, r6, r7, r8, r9, r10 |
| 205 |
rsbge r12, r12, #0 |
| 206 |
smulbb r12, r12, r11 |
| 207 |
movrel r9, zigzag_scan-1 |
| 208 |
ldrb r9, [r9, r3] |
| 209 |
cmp r3, #16 |
| 210 |
strh r12, [r1, r9] |
| 211 |
bge 6b |
| 212 |
b 5b |
| 213 |
endfunc |
| 214 |
|
| 215 |
.section .rodata |
| 216 |
zigzag_scan: |
| 217 |
.byte 0, 2, 8, 16 |
| 218 |
.byte 10, 4, 6, 12 |
| 219 |
.byte 18, 24, 26, 20 |
| 220 |
.byte 14, 22, 28, 30 |