1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/arm/armv5te/boolhuff_armv5te.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,310 @@
1.4 +;
1.5 +; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 +;
1.7 +; Use of this source code is governed by a BSD-style license
1.8 +; that can be found in the LICENSE file in the root of the source
1.9 +; tree. An additional intellectual property rights grant can be found
1.10 +; in the file PATENTS. All contributing project authors may
1.11 +; be found in the AUTHORS file in the root of the source tree.
1.12 +;
1.13 +
1.14 +
1.15 + EXPORT |vp8_start_encode|
1.16 + EXPORT |vp8_encode_bool|
1.17 + EXPORT |vp8_stop_encode|
1.18 + EXPORT |vp8_encode_value|
1.19 + IMPORT |vp8_validate_buffer_arm|
1.20 +
1.21 + INCLUDE vp8_asm_enc_offsets.asm
1.22 +
1.23 + ARM
1.24 + REQUIRE8
1.25 + PRESERVE8
1.26 +
1.27 + AREA |.text|, CODE, READONLY
1.28 +
1.29 + ; macro for validating write buffer position
1.30 + ; needs vp8_writer in r0
1.31 + ; start shall not be in r1
1.32 + MACRO
1.33 + VALIDATE_POS $start, $pos
1.34 + push {r0-r3, r12, lr} ; rest of regs are preserved by subroutine call
1.35 + ldr r2, [r0, #vp8_writer_buffer_end]
1.36 + ldr r3, [r0, #vp8_writer_error]
1.37 + mov r1, $pos
1.38 + mov r0, $start
1.39 + bl vp8_validate_buffer_arm
1.40 + pop {r0-r3, r12, lr}
1.41 + MEND
1.42 +
1.43 +; r0 BOOL_CODER *br
1.44 +; r1 unsigned char *source
1.45 +; r2 unsigned char *source_end
1.46 +|vp8_start_encode| PROC
1.47 + str r2, [r0, #vp8_writer_buffer_end]
1.48 + mov r12, #0
1.49 + mov r3, #255
1.50 + mvn r2, #23
1.51 + str r12, [r0, #vp8_writer_lowvalue]
1.52 + str r3, [r0, #vp8_writer_range]
1.53 + str r2, [r0, #vp8_writer_count]
1.54 + str r12, [r0, #vp8_writer_pos]
1.55 + str r1, [r0, #vp8_writer_buffer]
1.56 + bx lr
1.57 + ENDP
1.58 +
1.59 +; r0 BOOL_CODER *br
1.60 +; r1 int bit
1.61 +; r2 int probability
1.62 +|vp8_encode_bool| PROC
1.63 + push {r4-r10, lr}
1.64 +
1.65 + mov r4, r2
1.66 +
1.67 + ldr r2, [r0, #vp8_writer_lowvalue]
1.68 + ldr r5, [r0, #vp8_writer_range]
1.69 + ldr r3, [r0, #vp8_writer_count]
1.70 +
1.71 + sub r7, r5, #1 ; range-1
1.72 +
1.73 + cmp r1, #0
1.74 + mul r6, r4, r7 ; ((range-1) * probability)
1.75 +
1.76 + mov r7, #1
1.77 + add r4, r7, r6, lsr #8 ; 1 + (((range-1) * probability) >> 8)
1.78 +
1.79 + addne r2, r2, r4 ; if (bit) lowvalue += split
1.80 + subne r4, r5, r4 ; if (bit) range = range-split
1.81 +
1.82 + ; Counting the leading zeros is used to normalize range.
1.83 + clz r6, r4
1.84 + sub r6, r6, #24 ; shift
1.85 +
1.86 + ; Flag is set on the sum of count. This flag is used later
1.87 + ; to determine if count >= 0
1.88 + adds r3, r3, r6 ; count += shift
1.89 + lsl r5, r4, r6 ; range <<= shift
1.90 + bmi token_count_lt_zero ; if(count >= 0)
1.91 +
1.92 + sub r6, r6, r3 ; offset = shift - count
1.93 + sub r4, r6, #1 ; offset-1
1.94 + lsls r4, r2, r4 ; if((lowvalue<<(offset-1)) & 0x80000000 )
1.95 + bpl token_high_bit_not_set
1.96 +
1.97 + ldr r4, [r0, #vp8_writer_pos] ; x
1.98 + sub r4, r4, #1 ; x = w->pos-1
1.99 + b token_zero_while_start
1.100 +token_zero_while_loop
1.101 + mov r9, #0
1.102 + strb r9, [r7, r4] ; w->buffer[x] =(unsigned char)0
1.103 + sub r4, r4, #1 ; x--
1.104 +token_zero_while_start
1.105 + cmp r4, #0
1.106 + ldrge r7, [r0, #vp8_writer_buffer]
1.107 + ldrb r1, [r7, r4]
1.108 + cmpge r1, #0xff
1.109 + beq token_zero_while_loop
1.110 +
1.111 + ldr r7, [r0, #vp8_writer_buffer]
1.112 + ldrb r9, [r7, r4] ; w->buffer[x]
1.113 + add r9, r9, #1
1.114 + strb r9, [r7, r4] ; w->buffer[x] + 1
1.115 +token_high_bit_not_set
1.116 + rsb r4, r6, #24 ; 24-offset
1.117 + ldr r9, [r0, #vp8_writer_buffer]
1.118 + lsr r7, r2, r4 ; lowvalue >> (24-offset)
1.119 + ldr r4, [r0, #vp8_writer_pos] ; w->pos
1.120 + lsl r2, r2, r6 ; lowvalue <<= offset
1.121 + mov r6, r3 ; shift = count
1.122 + add r1, r4, #1 ; w->pos++
1.123 + bic r2, r2, #0xff000000 ; lowvalue &= 0xffffff
1.124 + str r1, [r0, #vp8_writer_pos]
1.125 + sub r3, r3, #8 ; count -= 8
1.126 +
1.127 + VALIDATE_POS r9, r1 ; validate_buffer at pos
1.128 +
1.129 + strb r7, [r9, r4] ; w->buffer[w->pos++]
1.130 +
1.131 +token_count_lt_zero
1.132 + lsl r2, r2, r6 ; lowvalue <<= shift
1.133 +
1.134 + str r2, [r0, #vp8_writer_lowvalue]
1.135 + str r5, [r0, #vp8_writer_range]
1.136 + str r3, [r0, #vp8_writer_count]
1.137 + pop {r4-r10, pc}
1.138 + ENDP
1.139 +
1.140 +; r0 BOOL_CODER *br
1.141 +|vp8_stop_encode| PROC
1.142 + push {r4-r10, lr}
1.143 +
1.144 + ldr r2, [r0, #vp8_writer_lowvalue]
1.145 + ldr r5, [r0, #vp8_writer_range]
1.146 + ldr r3, [r0, #vp8_writer_count]
1.147 +
1.148 + mov r10, #32
1.149 +
1.150 +stop_encode_loop
1.151 + sub r7, r5, #1 ; range-1
1.152 +
1.153 + mov r4, r7, lsl #7 ; ((range-1) * 128)
1.154 +
1.155 + mov r7, #1
1.156 + add r4, r7, r4, lsr #8 ; 1 + (((range-1) * 128) >> 8)
1.157 +
1.158 + ; Counting the leading zeros is used to normalize range.
1.159 + clz r6, r4
1.160 + sub r6, r6, #24 ; shift
1.161 +
1.162 + ; Flag is set on the sum of count. This flag is used later
1.163 + ; to determine if count >= 0
1.164 + adds r3, r3, r6 ; count += shift
1.165 + lsl r5, r4, r6 ; range <<= shift
1.166 + bmi token_count_lt_zero_se ; if(count >= 0)
1.167 +
1.168 + sub r6, r6, r3 ; offset = shift - count
1.169 + sub r4, r6, #1 ; offset-1
1.170 + lsls r4, r2, r4 ; if((lowvalue<<(offset-1)) & 0x80000000 )
1.171 + bpl token_high_bit_not_set_se
1.172 +
1.173 + ldr r4, [r0, #vp8_writer_pos] ; x
1.174 + sub r4, r4, #1 ; x = w->pos-1
1.175 + b token_zero_while_start_se
1.176 +token_zero_while_loop_se
1.177 + mov r9, #0
1.178 + strb r9, [r7, r4] ; w->buffer[x] =(unsigned char)0
1.179 + sub r4, r4, #1 ; x--
1.180 +token_zero_while_start_se
1.181 + cmp r4, #0
1.182 + ldrge r7, [r0, #vp8_writer_buffer]
1.183 + ldrb r1, [r7, r4]
1.184 + cmpge r1, #0xff
1.185 + beq token_zero_while_loop_se
1.186 +
1.187 + ldr r7, [r0, #vp8_writer_buffer]
1.188 + ldrb r9, [r7, r4] ; w->buffer[x]
1.189 + add r9, r9, #1
1.190 + strb r9, [r7, r4] ; w->buffer[x] + 1
1.191 +token_high_bit_not_set_se
1.192 + rsb r4, r6, #24 ; 24-offset
1.193 + ldr r9, [r0, #vp8_writer_buffer]
1.194 + lsr r7, r2, r4 ; lowvalue >> (24-offset)
1.195 + ldr r4, [r0, #vp8_writer_pos] ; w->pos
1.196 + lsl r2, r2, r6 ; lowvalue <<= offset
1.197 + mov r6, r3 ; shift = count
1.198 + add r1, r4, #1 ; w->pos++
1.199 + bic r2, r2, #0xff000000 ; lowvalue &= 0xffffff
1.200 + str r1, [r0, #vp8_writer_pos]
1.201 + sub r3, r3, #8 ; count -= 8
1.202 +
1.203 + VALIDATE_POS r9, r1 ; validate_buffer at pos
1.204 +
1.205 + strb r7, [r9, r4] ; w->buffer[w->pos++]
1.206 +
1.207 +token_count_lt_zero_se
1.208 + lsl r2, r2, r6 ; lowvalue <<= shift
1.209 +
1.210 + subs r10, r10, #1
1.211 + bne stop_encode_loop
1.212 +
1.213 + str r2, [r0, #vp8_writer_lowvalue]
1.214 + str r5, [r0, #vp8_writer_range]
1.215 + str r3, [r0, #vp8_writer_count]
1.216 + pop {r4-r10, pc}
1.217 +
1.218 + ENDP
1.219 +
1.220 +; r0 BOOL_CODER *br
1.221 +; r1 int data
1.222 +; r2 int bits
1.223 +|vp8_encode_value| PROC
1.224 + push {r4-r12, lr}
1.225 +
1.226 + mov r10, r2
1.227 +
1.228 + ldr r2, [r0, #vp8_writer_lowvalue]
1.229 + ldr r5, [r0, #vp8_writer_range]
1.230 + ldr r3, [r0, #vp8_writer_count]
1.231 +
1.232 + rsb r4, r10, #32 ; 32-n
1.233 +
1.234 + ; v is kept in r1 during the token pack loop
1.235 + lsl r1, r1, r4 ; r1 = v << 32 - n
1.236 +
1.237 +encode_value_loop
1.238 + sub r7, r5, #1 ; range-1
1.239 +
1.240 + ; Decisions are made based on the bit value shifted
1.241 + ; off of v, so set a flag here based on this.
1.242 + ; This value is refered to as "bb"
1.243 + lsls r1, r1, #1 ; bit = v >> n
1.244 + mov r4, r7, lsl #7 ; ((range-1) * 128)
1.245 +
1.246 + mov r7, #1
1.247 + add r4, r7, r4, lsr #8 ; 1 + (((range-1) * 128) >> 8)
1.248 +
1.249 + addcs r2, r2, r4 ; if (bit) lowvalue += split
1.250 + subcs r4, r5, r4 ; if (bit) range = range-split
1.251 +
1.252 + ; Counting the leading zeros is used to normalize range.
1.253 + clz r6, r4
1.254 + sub r6, r6, #24 ; shift
1.255 +
1.256 + ; Flag is set on the sum of count. This flag is used later
1.257 + ; to determine if count >= 0
1.258 + adds r3, r3, r6 ; count += shift
1.259 + lsl r5, r4, r6 ; range <<= shift
1.260 + bmi token_count_lt_zero_ev ; if(count >= 0)
1.261 +
1.262 + sub r6, r6, r3 ; offset = shift - count
1.263 + sub r4, r6, #1 ; offset-1
1.264 + lsls r4, r2, r4 ; if((lowvalue<<(offset-1)) & 0x80000000 )
1.265 + bpl token_high_bit_not_set_ev
1.266 +
1.267 + ldr r4, [r0, #vp8_writer_pos] ; x
1.268 + sub r4, r4, #1 ; x = w->pos-1
1.269 + b token_zero_while_start_ev
1.270 +token_zero_while_loop_ev
1.271 + mov r9, #0
1.272 + strb r9, [r7, r4] ; w->buffer[x] =(unsigned char)0
1.273 + sub r4, r4, #1 ; x--
1.274 +token_zero_while_start_ev
1.275 + cmp r4, #0
1.276 + ldrge r7, [r0, #vp8_writer_buffer]
1.277 + ldrb r11, [r7, r4]
1.278 + cmpge r11, #0xff
1.279 + beq token_zero_while_loop_ev
1.280 +
1.281 + ldr r7, [r0, #vp8_writer_buffer]
1.282 + ldrb r9, [r7, r4] ; w->buffer[x]
1.283 + add r9, r9, #1
1.284 + strb r9, [r7, r4] ; w->buffer[x] + 1
1.285 +token_high_bit_not_set_ev
1.286 + rsb r4, r6, #24 ; 24-offset
1.287 + ldr r9, [r0, #vp8_writer_buffer]
1.288 + lsr r7, r2, r4 ; lowvalue >> (24-offset)
1.289 + ldr r4, [r0, #vp8_writer_pos] ; w->pos
1.290 + lsl r2, r2, r6 ; lowvalue <<= offset
1.291 + mov r6, r3 ; shift = count
1.292 + add r11, r4, #1 ; w->pos++
1.293 + bic r2, r2, #0xff000000 ; lowvalue &= 0xffffff
1.294 + str r11, [r0, #vp8_writer_pos]
1.295 + sub r3, r3, #8 ; count -= 8
1.296 +
1.297 + VALIDATE_POS r9, r11 ; validate_buffer at pos
1.298 +
1.299 + strb r7, [r9, r4] ; w->buffer[w->pos++]
1.300 +
1.301 +token_count_lt_zero_ev
1.302 + lsl r2, r2, r6 ; lowvalue <<= shift
1.303 +
1.304 + subs r10, r10, #1
1.305 + bne encode_value_loop
1.306 +
1.307 + str r2, [r0, #vp8_writer_lowvalue]
1.308 + str r5, [r0, #vp8_writer_range]
1.309 + str r3, [r0, #vp8_writer_count]
1.310 + pop {r4-r12, pc}
1.311 + ENDP
1.312 +
1.313 + END
