2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
4 ; Use of this source code is governed by a BSD-style license
5 ; that can be found in the LICENSE file in the root of the source
6 ; tree. An additional intellectual property rights grant can be found
7 ; in the file PATENTS. All contributing project authors may
8 ; be found in the AUTHORS file in the root of the source tree.
12 EXPORT |vp8cx_pack_tokens_into_partitions_armv5|
14 INCLUDE vpx_vp8_enc_asm_offsets.asm
20 AREA |.text|, CODE, READONLY
23 ; r1 unsigned char *cx_data
26 ; s0 vp8_coef_encodings
28 ; s2 const vp8_tree_index *,
30 |vp8cx_pack_tokens_into_partitions_armv5| PROC
34 ; Compute address of cpi->common.mb_rows
35 ldr r4, _VP8_COMP_common_
36 ldr r6, _VP8_COMMON_MBrows_
39 ldr r5, [r4, r6] ; load up mb_rows
41 str r5, [sp, #36] ; save mb_rows
42 str r1, [sp, #24] ; save cx_data
43 str r2, [sp, #20] ; save num_part
44 str r3, [sp, #8] ; save *size
46 ; *size = 3*(num_part -1 );
47 sub r2, r2, #1 ; num_part - 1
48 add r2, r2, r2, lsl #1 ; 3*(num_part - 1)
51 add r2, r2, r1 ; cx_data + *size
52 str r2, [sp, #40] ; ptr
54 ldr r4, _VP8_COMP_tplist_
56 ldr r7, [r4, #0] ; dereference cpi->tp_list
57 str r7, [sp, #32] ; store start of cpi->tp_list
59 ldr r11, _VP8_COMP_bc2_ ; load up vp8_writer out of cpi
63 str r11, [sp, #28] ; i
66 ldr r10, [sp, #40] ; ptr
67 ldr r5, [sp, #36] ; move mb_rows to the counting section
70 ; Reset all of the VP8 Writer data for each partition that
73 mov r2, #0 ; vp8_writer_lowvalue
74 mov r5, #255 ; vp8_writer_range
75 mvn r3, #23 ; vp8_writer_count
77 str r2, [r0, #vp8_writer_value]
78 str r2, [r0, #vp8_writer_pos]
79 str r10, [r0, #vp8_writer_buffer]
83 ldr r1, [r7, #tokenlist_start]
84 ldr r9, [r7, #tokenlist_stop]
85 str r9, [sp, #0] ; save stop for later comparison
86 str r7, [sp, #16] ; tokenlist address for next time
90 ; actual work gets done here!
93 ldr r6, [r1, #tokenextra_token] ; t
94 ldr r4, [sp, #80] ; vp8_coef_encodings
96 add r4, r4, r6, lsl #3 ; a = vp8_coef_encodings + t
97 ldr r9, [r1, #tokenextra_context_tree] ; pp
99 ldr r7, [r1, #tokenextra_skip_eob_node]
101 ldr r6, [r4, #vp8_token_value] ; v
102 ldr r8, [r4, #vp8_token_len] ; n
104 ; vp8 specific skip_eob_node
107 subne r8, r8, #1 ; --n
109 rsb r4, r8, #32 ; 32-n
110 ldr r10, [sp, #88] ; vp8_coef_tree
112 ; v is kept in r12 during the token pack loop
113 lsl r12, r6, r4 ; r12 = v << 32 - n
117 ldrb r4, [r9, lr, asr #1] ; pp [i>>1]
118 sub r7, r5, #1 ; range-1
120 ; Decisions are made based on the bit value shifted
121 ; off of v, so set a flag here based on this.
122 ; This value is refered to as "bb"
123 lsls r12, r12, #1 ; bb = v >> n
124 mul r4, r4, r7 ; ((range-1) * pp[i>>1]))
126 ; bb can only be 0 or 1. So only execute this statement
127 ; if bb == 1, otherwise it will act like i + 0
128 addcs lr, lr, #1 ; i + bb
131 ldrsb lr, [r10, lr] ; i = vp8_coef_tree[i+bb]
132 add r4, r7, r4, lsr #8 ; 1 + (((range-1) * pp[i>>1]) >> 8)
134 addcs r2, r2, r4 ; if (bb) lowvalue += split
135 subcs r4, r5, r4 ; if (bb) range = range-split
137 ; Counting the leading zeros is used to normalize range.
139 sub r6, r6, #24 ; shift
141 ; Flag is set on the sum of count. This flag is used later
142 ; to determine if count >= 0
143 adds r3, r3, r6 ; count += shift
144 lsl r5, r4, r6 ; range <<= shift
145 bmi token_count_lt_zero ; if(count >= 0)
147 sub r6, r6, r3 ; offset = shift - count
148 sub r4, r6, #1 ; offset-1
149 lsls r4, r2, r4 ; if((lowvalue<<(offset-1)) & 0x80000000 )
150 bpl token_high_bit_not_set
152 ldr r4, [r0, #vp8_writer_pos] ; x
153 sub r4, r4, #1 ; x = w->pos-1
154 b token_zero_while_start
155 token_zero_while_loop
157 strb r10, [r7, r4] ; w->buffer[x] =(unsigned char)0
159 token_zero_while_start
161 ldrge r7, [r0, #vp8_writer_buffer]
164 beq token_zero_while_loop
166 ldr r7, [r0, #vp8_writer_buffer]
167 ldrb r10, [r7, r4] ; w->buffer[x]
169 strb r10, [r7, r4] ; w->buffer[x] + 1
170 token_high_bit_not_set
171 rsb r4, r6, #24 ; 24-offset
172 ldr r10, [r0, #vp8_writer_buffer]
173 lsr r7, r2, r4 ; lowvalue >> (24-offset)
174 ldr r4, [r0, #vp8_writer_pos] ; w->pos
175 lsl r2, r2, r6 ; lowvalue <<= offset
176 mov r6, r3 ; shift = count
177 add r11, r4, #1 ; w->pos++
178 bic r2, r2, #0xff000000 ; lowvalue &= 0xffffff
179 str r11, [r0, #vp8_writer_pos]
180 sub r3, r3, #8 ; count -= 8
181 strb r7, [r10, r4] ; w->buffer[w->pos++]
183 ; r10 is used earlier in the loop, but r10 is used as
184 ; temp variable here. So after r10 is used, reload
185 ; vp8_coef_tree_dcd into r10
186 ldr r10, [sp, #88] ; vp8_coef_tree
189 lsl r2, r2, r6 ; lowvalue <<= shift
191 subs r8, r8, #1 ; --n
194 ldr r6, [r1, #tokenextra_token] ; t
195 ldr r7, [sp, #84] ; vp8_extra_bits
196 ; Add t * sizeof (vp8_extra_bit_struct) to get the desired
197 ; element. Here vp8_extra_bit_struct == 20
198 add r6, r6, r6, lsl #2 ; b = vp8_extra_bits + t
199 add r12, r7, r6, lsl #2 ; b = vp8_extra_bits + t
201 ldr r4, [r12, #vp8_extra_bit_struct_base_val]
206 ldr r8, [r12, #vp8_extra_bit_struct_len] ; L
207 ldr lr, [r1, #tokenextra_extra] ; e = p->Extra
211 ldr r9, [r12, #vp8_extra_bit_struct_prob]
212 asr r7, lr, #1 ; v=e>>1
214 ldr r10, [r12, #vp8_extra_bit_struct_tree]
215 str r10, [sp, #4] ; b->tree
223 ldrb r4, [r9, lr, asr #1] ; pp[i>>1]
224 sub r7, r5, #1 ; range-1
225 lsls r12, r12, #1 ; v >> n
226 mul r4, r4, r7 ; (range-1) * pp[i>>1]
227 addcs lr, lr, #1 ; i + bb
230 ldrsb lr, [r10, lr] ; i = b->tree[i+bb]
231 add r4, r7, r4, lsr #8 ; split = 1 + (((range-1) * pp[i>>1]) >> 8)
233 addcs r2, r2, r4 ; if (bb) lowvalue += split
234 subcs r4, r5, r4 ; if (bb) range = range-split
239 adds r3, r3, r6 ; count += shift
240 lsl r5, r4, r6 ; range <<= shift
241 bmi extra_count_lt_zero ; if(count >= 0)
243 sub r6, r6, r3 ; offset= shift - count
244 sub r4, r6, #1 ; offset-1
245 lsls r4, r2, r4 ; if((lowvalue<<(offset-1)) & 0x80000000 )
246 bpl extra_high_bit_not_set
248 ldr r4, [r0, #vp8_writer_pos] ; x
249 sub r4, r4, #1 ; x = w->pos - 1
250 b extra_zero_while_start
251 extra_zero_while_loop
253 strb r10, [r7, r4] ; w->buffer[x] =(unsigned char)0
255 extra_zero_while_start
257 ldrge r7, [r0, #vp8_writer_buffer]
260 beq extra_zero_while_loop
262 ldr r7, [r0, #vp8_writer_buffer]
266 extra_high_bit_not_set
267 rsb r4, r6, #24 ; 24-offset
268 ldr r10, [r0, #vp8_writer_buffer]
269 lsr r7, r2, r4 ; lowvalue >> (24-offset)
270 ldr r4, [r0, #vp8_writer_pos]
271 lsl r2, r2, r6 ; lowvalue <<= offset
272 mov r6, r3 ; shift = count
273 add r11, r4, #1 ; w->pos++
274 bic r2, r2, #0xff000000 ; lowvalue &= 0xffffff
275 str r11, [r0, #vp8_writer_pos]
276 sub r3, r3, #8 ; count -= 8
277 strb r7, [r10, r4] ; w->buffer[w->pos++]=(lowvalue >> (24-offset))
278 ldr r10, [sp, #4] ; b->tree
282 subs r8, r8, #1 ; --n
283 bne extra_bits_loop ; while (n)
286 ldr lr, [r1, #4] ; e = p->Extra
287 add r4, r5, #1 ; range + 1
289 lsr r4, r4, #1 ; split = (range + 1) >> 1
290 addne r2, r2, r4 ; lowvalue += split
291 subne r4, r5, r4 ; range = range-split
292 tst r2, #0x80000000 ; lowvalue & 0x80000000
293 lsl r5, r4, #1 ; range <<= 1
294 beq end_high_bit_not_set
296 ldr r4, [r0, #vp8_writer_pos]
299 b end_zero_while_start
305 ldrge r6, [r0, #vp8_writer_buffer]
308 beq end_zero_while_loop
310 ldr r6, [r0, #vp8_writer_buffer]
315 adds r3, r3, #1 ; ++count
316 lsl r2, r2, #1 ; lowvalue <<= 1
319 ldr r4, [r0, #vp8_writer_pos]
321 ldr r7, [r0, #vp8_writer_buffer]
322 lsr r6, r2, #24 ; lowvalue >> 24
323 add r12, r4, #1 ; w->pos++
324 bic r2, r2, #0xff000000 ; lowvalue &= 0xffffff
329 add r1, r1, #TOKENEXTRA_SZ ; ++p
331 ldr r4, [sp, #0] ; stop
332 cmp r1, r4 ; while( p < stop)
335 ldr r10, [sp, #20] ; num_parts
336 mov r1, #TOKENLIST_SZ
339 ldr r6, [sp, #12] ; mb_rows
340 ldr r7, [sp, #16] ; tokenlist address
342 add r7, r7, r1 ; next element in the array
349 sub r7, r5, #1 ; range-1
351 mov r4, r7, lsl #7 ; ((range-1) * 128)
354 add r4, r7, r4, lsr #8 ; 1 + (((range-1) * 128) >> 8)
356 ; Counting the leading zeros is used to normalize range.
358 sub r6, r6, #24 ; shift
360 ; Flag is set on the sum of count. This flag is used later
361 ; to determine if count >= 0
362 adds r3, r3, r6 ; count += shift
363 lsl r5, r4, r6 ; range <<= shift
364 bmi token_count_lt_zero_se ; if(count >= 0)
366 sub r6, r6, r3 ; offset = shift - count
367 sub r4, r6, #1 ; offset-1
368 lsls r4, r2, r4 ; if((lowvalue<<(offset-1)) & 0x80000000 )
369 bpl token_high_bit_not_set_se
371 ldr r4, [r0, #vp8_writer_pos] ; x
372 sub r4, r4, #1 ; x = w->pos-1
373 b token_zero_while_start_se
374 token_zero_while_loop_se
376 strb r10, [r7, r4] ; w->buffer[x] =(unsigned char)0
378 token_zero_while_start_se
380 ldrge r7, [r0, #vp8_writer_buffer]
383 beq token_zero_while_loop_se
385 ldr r7, [r0, #vp8_writer_buffer]
386 ldrb r10, [r7, r4] ; w->buffer[x]
388 strb r10, [r7, r4] ; w->buffer[x] + 1
389 token_high_bit_not_set_se
390 rsb r4, r6, #24 ; 24-offset
391 ldr r10, [r0, #vp8_writer_buffer]
392 lsr r7, r2, r4 ; lowvalue >> (24-offset)
393 ldr r4, [r0, #vp8_writer_pos] ; w->pos
394 lsl r2, r2, r6 ; lowvalue <<= offset
395 mov r6, r3 ; shift = count
396 add r11, r4, #1 ; w->pos++
397 bic r2, r2, #0xff000000 ; lowvalue &= 0xffffff
398 str r11, [r0, #vp8_writer_pos]
399 sub r3, r3, #8 ; count -= 8
400 strb r7, [r10, r4] ; w->buffer[w->pos++]
402 token_count_lt_zero_se
403 lsl r2, r2, r6 ; lowvalue <<= shift
408 ldr r10, [sp, #8] ; *size
410 ldr r4, [r0, #vp8_writer_pos] ; w->pos
411 add r11, r11, r4 ; *size += w->pos
414 ldr r9, [sp, #20] ; num_parts
416 ldr r10, [sp, #28] ; i
417 cmp r10, r9 ; if(i<(num_part - 1))
418 bge skip_write_partition
420 ldr r12, [sp, #40] ; ptr
421 add r12, r12, r4 ; ptr += w->pos
424 ldr r9, [sp, #24] ; cx_data
431 add r9, r9, #3 ; cx_data += 3
436 ldr r11, [sp, #28] ; i
437 ldr r10, [sp, #20] ; num_parts
439 add r11, r11, #1 ; i++
442 ldr r7, [sp, #32] ; cpi->tp_list[i]
443 mov r1, #TOKENLIST_SZ
444 add r7, r7, r1 ; next element in cpi->tp_list
445 str r7, [sp, #32] ; cpi->tp_list[i+1]
458 DCD vp8_common_mb_rows