2 * Copyright (c) 2012 The LibYuv 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.
11 #include "libyuv/row.h"
18 // The following are available on Mips platforms:
19 #if !defined(LIBYUV_DISABLE_MIPS) && defined(__mips__) && \
20 (_MIPS_SIM == _MIPS_SIM_ABI32)
22 #ifdef HAS_COPYROW_MIPS
23 void CopyRow_MIPS(const uint8* src, uint8* dst, int count) {
24 __asm__ __volatile__ (
27 "slti $at, %[count], 8 \n"
28 "bne $at ,$zero, $last8 \n"
29 "xor $t8, %[src], %[dst] \n"
30 "andi $t8, $t8, 0x3 \n"
32 "bne $t8, $zero, unaligned \n"
34 // make dst/src aligned
35 "andi $a3, $a3, 0x3 \n"
36 "beq $a3, $zero, $chk16w \n"
37 // word-aligned now count is the remining bytes count
38 "subu %[count], %[count], $a3 \n"
40 "lwr $t8, 0(%[src]) \n"
41 "addu %[src], %[src], $a3 \n"
42 "swr $t8, 0(%[dst]) \n"
43 "addu %[dst], %[dst], $a3 \n"
45 // Now the dst/src are mutually word-aligned with word-aligned addresses
47 "andi $t8, %[count], 0x3f \n" // whole 64-B chunks?
48 // t8 is the byte count after 64-byte chunks
49 "beq %[count], $t8, chk8w \n"
50 // There will be at most 1 32-byte chunk after it
51 "subu $a3, %[count], $t8 \n" // the reminder
52 // Here a3 counts bytes in 16w chunks
53 "addu $a3, %[dst], $a3 \n"
54 // Now a3 is the final dst after 64-byte chunks
55 "addu $t0, %[dst], %[count] \n"
56 // t0 is the "past the end" address
58 // When in the loop we exercise "pref 30,x(a1)", the a1+x should not be past
59 // the "t0-32" address
60 // This means: for x=128 the last "safe" a1 address is "t0-160"
61 // Alternatively, for x=64 the last "safe" a1 address is "t0-96"
62 // we will use "pref 30,128(a1)", so "t0-160" is the limit
63 "subu $t9, $t0, 160 \n"
64 // t9 is the "last safe pref 30,128(a1)" address
65 "pref 0, 0(%[src]) \n" // first line of src
66 "pref 0, 32(%[src]) \n" // second line of src
67 "pref 0, 64(%[src]) \n"
68 "pref 30, 32(%[dst]) \n"
69 // In case the a1 > t9 don't use "pref 30" at all
70 "sgtu $v1, %[dst], $t9 \n"
71 "bgtz $v1, $loop16w \n"
73 // otherwise, start with using pref30
74 "pref 30, 64(%[dst]) \n"
76 "pref 0, 96(%[src]) \n"
77 "lw $t0, 0(%[src]) \n"
78 "bgtz $v1, $skip_pref30_96 \n" // skip
79 "lw $t1, 4(%[src]) \n"
80 "pref 30, 96(%[dst]) \n" // continue
82 "lw $t2, 8(%[src]) \n"
83 "lw $t3, 12(%[src]) \n"
84 "lw $t4, 16(%[src]) \n"
85 "lw $t5, 20(%[src]) \n"
86 "lw $t6, 24(%[src]) \n"
87 "lw $t7, 28(%[src]) \n"
88 "pref 0, 128(%[src]) \n"
89 // bring the next lines of src, addr 128
90 "sw $t0, 0(%[dst]) \n"
91 "sw $t1, 4(%[dst]) \n"
92 "sw $t2, 8(%[dst]) \n"
93 "sw $t3, 12(%[dst]) \n"
94 "sw $t4, 16(%[dst]) \n"
95 "sw $t5, 20(%[dst]) \n"
96 "sw $t6, 24(%[dst]) \n"
97 "sw $t7, 28(%[dst]) \n"
98 "lw $t0, 32(%[src]) \n"
99 "bgtz $v1, $skip_pref30_128 \n" // skip pref 30,128(a1)
100 "lw $t1, 36(%[src]) \n"
101 "pref 30, 128(%[dst]) \n" // set dest, addr 128
102 "$skip_pref30_128: \n"
103 "lw $t2, 40(%[src]) \n"
104 "lw $t3, 44(%[src]) \n"
105 "lw $t4, 48(%[src]) \n"
106 "lw $t5, 52(%[src]) \n"
107 "lw $t6, 56(%[src]) \n"
108 "lw $t7, 60(%[src]) \n"
109 "pref 0, 160(%[src]) \n"
110 // bring the next lines of src, addr 160
111 "sw $t0, 32(%[dst]) \n"
112 "sw $t1, 36(%[dst]) \n"
113 "sw $t2, 40(%[dst]) \n"
114 "sw $t3, 44(%[dst]) \n"
115 "sw $t4, 48(%[dst]) \n"
116 "sw $t5, 52(%[dst]) \n"
117 "sw $t6, 56(%[dst]) \n"
118 "sw $t7, 60(%[dst]) \n"
120 "addiu %[dst], %[dst], 64 \n" // adding 64 to dest
121 "sgtu $v1, %[dst], $t9 \n"
122 "bne %[dst], $a3, $loop16w \n"
123 " addiu %[src], %[src], 64 \n" // adding 64 to src
124 "move %[count], $t8 \n"
126 // Here we have src and dest word-aligned but less than 64-bytes to go
129 "pref 0, 0x0(%[src]) \n"
130 "andi $t8, %[count], 0x1f \n" // 32-byte chunk?
131 // the t8 is the reminder count past 32-bytes
132 "beq %[count], $t8, chk1w \n"
133 // count=t8,no 32-byte chunk
136 "lw $t0, 0(%[src]) \n"
137 "lw $t1, 4(%[src]) \n"
138 "lw $t2, 8(%[src]) \n"
139 "lw $t3, 12(%[src]) \n"
140 "lw $t4, 16(%[src]) \n"
141 "lw $t5, 20(%[src]) \n"
142 "lw $t6, 24(%[src]) \n"
143 "lw $t7, 28(%[src]) \n"
144 "addiu %[src], %[src], 32 \n"
146 "sw $t0, 0(%[dst]) \n"
147 "sw $t1, 4(%[dst]) \n"
148 "sw $t2, 8(%[dst]) \n"
149 "sw $t3, 12(%[dst]) \n"
150 "sw $t4, 16(%[dst]) \n"
151 "sw $t5, 20(%[dst]) \n"
152 "sw $t6, 24(%[dst]) \n"
153 "sw $t7, 28(%[dst]) \n"
154 "addiu %[dst], %[dst], 32 \n"
157 "andi %[count], $t8, 0x3 \n"
158 // now count is the reminder past 1w chunks
159 "beq %[count], $t8, $last8 \n"
160 " subu $a3, $t8, %[count] \n"
161 // a3 is count of bytes in 1w chunks
162 "addu $a3, %[dst], $a3 \n"
163 // now a3 is the dst address past the 1w chunks
164 // copying in words (4-byte chunks)
166 "lw $t3, 0(%[src]) \n"
167 // the first t3 may be equal t0 ... optimize?
168 "addiu %[src], %[src],4 \n"
169 "addiu %[dst], %[dst],4 \n"
170 "bne %[dst], $a3,$wordCopy_loop \n"
171 " sw $t3, -4(%[dst]) \n"
173 // For the last (<8) bytes
175 "blez %[count], leave \n"
176 " addu $a3, %[dst], %[count] \n" // a3 -last dst address
178 "lb $v1, 0(%[src]) \n"
179 "addiu %[src], %[src], 1 \n"
180 "addiu %[dst], %[dst], 1 \n"
181 "bne %[dst], $a3, $last8loop \n"
182 " sb $v1, -1(%[dst]) \n"
193 // got here with a3="negu a1"
194 "andi $a3, $a3, 0x3 \n" // a1 is word aligned?
195 "beqz $a3, $ua_chk16w \n"
196 " subu %[count], %[count], $a3 \n"
197 // bytes left after initial a3 bytes
198 "lwr $v1, 0(%[src]) \n"
199 "lwl $v1, 3(%[src]) \n"
200 "addu %[src], %[src], $a3 \n" // a3 may be 1, 2 or 3
201 "swr $v1, 0(%[dst]) \n"
202 "addu %[dst], %[dst], $a3 \n"
203 // below the dst will be word aligned (NOTE1)
205 "andi $t8, %[count], 0x3f \n" // whole 64-B chunks?
206 // t8 is the byte count after 64-byte chunks
207 "beq %[count], $t8, ua_chk8w \n"
208 // if a2==t8, no 64-byte chunks
209 // There will be at most 1 32-byte chunk after it
210 "subu $a3, %[count], $t8 \n" // the reminder
211 // Here a3 counts bytes in 16w chunks
212 "addu $a3, %[dst], $a3 \n"
213 // Now a3 is the final dst after 64-byte chunks
214 "addu $t0, %[dst], %[count] \n" // t0 "past the end"
215 "subu $t9, $t0, 160 \n"
216 // t9 is the "last safe pref 30,128(a1)" address
217 "pref 0, 0(%[src]) \n" // first line of src
218 "pref 0, 32(%[src]) \n" // second line addr 32
219 "pref 0, 64(%[src]) \n"
220 "pref 30, 32(%[dst]) \n"
221 // safe, as we have at least 64 bytes ahead
222 // In case the a1 > t9 don't use "pref 30" at all
223 "sgtu $v1, %[dst], $t9 \n"
224 "bgtz $v1, $ua_loop16w \n"
225 // skip "pref 30,64(a1)" for too short arrays
227 // otherwise, start with using pref30
228 "pref 30, 64(%[dst]) \n"
230 "pref 0, 96(%[src]) \n"
231 "lwr $t0, 0(%[src]) \n"
232 "lwl $t0, 3(%[src]) \n"
233 "lwr $t1, 4(%[src]) \n"
234 "bgtz $v1, $ua_skip_pref30_96 \n"
235 " lwl $t1, 7(%[src]) \n"
236 "pref 30, 96(%[dst]) \n"
237 // continue setting up the dest, addr 96
238 "$ua_skip_pref30_96: \n"
239 "lwr $t2, 8(%[src]) \n"
240 "lwl $t2, 11(%[src]) \n"
241 "lwr $t3, 12(%[src]) \n"
242 "lwl $t3, 15(%[src]) \n"
243 "lwr $t4, 16(%[src]) \n"
244 "lwl $t4, 19(%[src]) \n"
245 "lwr $t5, 20(%[src]) \n"
246 "lwl $t5, 23(%[src]) \n"
247 "lwr $t6, 24(%[src]) \n"
248 "lwl $t6, 27(%[src]) \n"
249 "lwr $t7, 28(%[src]) \n"
250 "lwl $t7, 31(%[src]) \n"
251 "pref 0, 128(%[src]) \n"
252 // bring the next lines of src, addr 128
253 "sw $t0, 0(%[dst]) \n"
254 "sw $t1, 4(%[dst]) \n"
255 "sw $t2, 8(%[dst]) \n"
256 "sw $t3, 12(%[dst]) \n"
257 "sw $t4, 16(%[dst]) \n"
258 "sw $t5, 20(%[dst]) \n"
259 "sw $t6, 24(%[dst]) \n"
260 "sw $t7, 28(%[dst]) \n"
261 "lwr $t0, 32(%[src]) \n"
262 "lwl $t0, 35(%[src]) \n"
263 "lwr $t1, 36(%[src]) \n"
264 "bgtz $v1, ua_skip_pref30_128 \n"
265 " lwl $t1, 39(%[src]) \n"
266 "pref 30, 128(%[dst]) \n"
267 // continue setting up the dest, addr 128
268 "ua_skip_pref30_128: \n"
270 "lwr $t2, 40(%[src]) \n"
271 "lwl $t2, 43(%[src]) \n"
272 "lwr $t3, 44(%[src]) \n"
273 "lwl $t3, 47(%[src]) \n"
274 "lwr $t4, 48(%[src]) \n"
275 "lwl $t4, 51(%[src]) \n"
276 "lwr $t5, 52(%[src]) \n"
277 "lwl $t5, 55(%[src]) \n"
278 "lwr $t6, 56(%[src]) \n"
279 "lwl $t6, 59(%[src]) \n"
280 "lwr $t7, 60(%[src]) \n"
281 "lwl $t7, 63(%[src]) \n"
282 "pref 0, 160(%[src]) \n"
283 // bring the next lines of src, addr 160
284 "sw $t0, 32(%[dst]) \n"
285 "sw $t1, 36(%[dst]) \n"
286 "sw $t2, 40(%[dst]) \n"
287 "sw $t3, 44(%[dst]) \n"
288 "sw $t4, 48(%[dst]) \n"
289 "sw $t5, 52(%[dst]) \n"
290 "sw $t6, 56(%[dst]) \n"
291 "sw $t7, 60(%[dst]) \n"
293 "addiu %[dst],%[dst],64 \n" // adding 64 to dest
294 "sgtu $v1,%[dst],$t9 \n"
295 "bne %[dst],$a3,$ua_loop16w \n"
296 " addiu %[src],%[src],64 \n" // adding 64 to src
297 "move %[count],$t8 \n"
299 // Here we have src and dest word-aligned but less than 64-bytes to go
302 "pref 0, 0x0(%[src]) \n"
303 "andi $t8, %[count], 0x1f \n" // 32-byte chunk?
304 // the t8 is the reminder count
305 "beq %[count], $t8, $ua_chk1w \n"
306 // when count==t8, no 32-byte chunk
308 "lwr $t0, 0(%[src]) \n"
309 "lwl $t0, 3(%[src]) \n"
310 "lwr $t1, 4(%[src]) \n"
311 "lwl $t1, 7(%[src]) \n"
312 "lwr $t2, 8(%[src]) \n"
313 "lwl $t2, 11(%[src]) \n"
314 "lwr $t3, 12(%[src]) \n"
315 "lwl $t3, 15(%[src]) \n"
316 "lwr $t4, 16(%[src]) \n"
317 "lwl $t4, 19(%[src]) \n"
318 "lwr $t5, 20(%[src]) \n"
319 "lwl $t5, 23(%[src]) \n"
320 "lwr $t6, 24(%[src]) \n"
321 "lwl $t6, 27(%[src]) \n"
322 "lwr $t7, 28(%[src]) \n"
323 "lwl $t7, 31(%[src]) \n"
324 "addiu %[src], %[src], 32 \n"
326 "sw $t0, 0(%[dst]) \n"
327 "sw $t1, 4(%[dst]) \n"
328 "sw $t2, 8(%[dst]) \n"
329 "sw $t3, 12(%[dst]) \n"
330 "sw $t4, 16(%[dst]) \n"
331 "sw $t5, 20(%[dst]) \n"
332 "sw $t6, 24(%[dst]) \n"
333 "sw $t7, 28(%[dst]) \n"
334 "addiu %[dst], %[dst], 32 \n"
337 "andi %[count], $t8, 0x3 \n"
338 // now count is the reminder past 1w chunks
339 "beq %[count], $t8, ua_smallCopy \n"
340 "subu $a3, $t8, %[count] \n"
341 // a3 is count of bytes in 1w chunks
342 "addu $a3, %[dst], $a3 \n"
343 // now a3 is the dst address past the 1w chunks
345 // copying in words (4-byte chunks)
346 "$ua_wordCopy_loop: \n"
347 "lwr $v1, 0(%[src]) \n"
348 "lwl $v1, 3(%[src]) \n"
349 "addiu %[src], %[src], 4 \n"
350 "addiu %[dst], %[dst], 4 \n"
351 // note: dst=a1 is word aligned here, see NOTE1
352 "bne %[dst], $a3, $ua_wordCopy_loop \n"
353 " sw $v1,-4(%[dst]) \n"
355 // Now less than 4 bytes (value in count) left to copy
357 "beqz %[count], leave \n"
358 " addu $a3, %[dst], %[count] \n" // a3 = last dst address
359 "$ua_smallCopy_loop: \n"
360 "lb $v1, 0(%[src]) \n"
361 "addiu %[src], %[src], 1 \n"
362 "addiu %[dst], %[dst], 1 \n"
363 "bne %[dst],$a3,$ua_smallCopy_loop \n"
364 " sb $v1, -1(%[dst]) \n"
370 : [dst] "+r" (dst), [src] "+r" (src)
371 : [count] "r" (count)
372 : "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7",
373 "t8", "t9", "a3", "v1", "at"
376 #endif // HAS_COPYROW_MIPS
378 // MIPS DSPR2 functions
379 #if !defined(LIBYUV_DISABLE_MIPS) && defined(__mips_dsp) && \
380 (__mips_dsp_rev >= 2) && \
381 (_MIPS_SIM == _MIPS_SIM_ABI32)
383 void SplitUVRow_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
385 __asm__ __volatile__ (
388 "srl $t4, %[width], 4 \n" // multiplies of 16
390 " andi %[width], %[width], 0xf \n" // residual
394 "addiu $t4, $t4, -1 \n"
395 "lw $t0, 0(%[src_uv]) \n" // V1 | U1 | V0 | U0
396 "lw $t1, 4(%[src_uv]) \n" // V3 | U3 | V2 | U2
397 "lw $t2, 8(%[src_uv]) \n" // V5 | U5 | V4 | U4
398 "lw $t3, 12(%[src_uv]) \n" // V7 | U7 | V6 | U6
399 "lw $t5, 16(%[src_uv]) \n" // V9 | U9 | V8 | U8
400 "lw $t6, 20(%[src_uv]) \n" // V11 | U11 | V10 | U10
401 "lw $t7, 24(%[src_uv]) \n" // V13 | U13 | V12 | U12
402 "lw $t8, 28(%[src_uv]) \n" // V15 | U15 | V14 | U14
403 "addiu %[src_uv], %[src_uv], 32 \n"
404 "precrq.qb.ph $t9, $t1, $t0 \n" // V3 | V2 | V1 | V0
405 "precr.qb.ph $t0, $t1, $t0 \n" // U3 | U2 | U1 | U0
406 "precrq.qb.ph $t1, $t3, $t2 \n" // V7 | V6 | V5 | V4
407 "precr.qb.ph $t2, $t3, $t2 \n" // U7 | U6 | U5 | U4
408 "precrq.qb.ph $t3, $t6, $t5 \n" // V11 | V10 | V9 | V8
409 "precr.qb.ph $t5, $t6, $t5 \n" // U11 | U10 | U9 | U8
410 "precrq.qb.ph $t6, $t8, $t7 \n" // V15 | V14 | V13 | V12
411 "precr.qb.ph $t7, $t8, $t7 \n" // U15 | U14 | U13 | U12
412 "sw $t9, 0(%[dst_v]) \n"
413 "sw $t0, 0(%[dst_u]) \n"
414 "sw $t1, 4(%[dst_v]) \n"
415 "sw $t2, 4(%[dst_u]) \n"
416 "sw $t3, 8(%[dst_v]) \n"
417 "sw $t5, 8(%[dst_u]) \n"
418 "sw $t6, 12(%[dst_v]) \n"
419 "sw $t7, 12(%[dst_u]) \n"
420 "addiu %[dst_v], %[dst_v], 16 \n"
422 " addiu %[dst_u], %[dst_u], 16 \n"
424 "beqz %[width], 3f \n"
428 "lbu $t0, 0(%[src_uv]) \n"
429 "lbu $t1, 1(%[src_uv]) \n"
430 "addiu %[src_uv], %[src_uv], 2 \n"
431 "addiu %[width], %[width], -1 \n"
432 "sb $t0, 0(%[dst_u]) \n"
433 "sb $t1, 0(%[dst_v]) \n"
434 "addiu %[dst_u], %[dst_u], 1 \n"
435 "bgtz %[width], 2b \n"
436 " addiu %[dst_v], %[dst_v], 1 \n"
440 : [src_uv] "+r" (src_uv),
441 [width] "+r" (width),
442 [dst_u] "+r" (dst_u),
445 : "t0", "t1", "t2", "t3",
446 "t4", "t5", "t6", "t7", "t8", "t9"
450 void SplitUVRow_Unaligned_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u,
451 uint8* dst_v, int width) {
452 __asm__ __volatile__ (
455 "srl $t4, %[width], 4 \n" // multiplies of 16
457 " andi %[width], %[width], 0xf \n" // residual
461 "addiu $t4, $t4, -1 \n"
462 "lwr $t0, 0(%[src_uv]) \n"
463 "lwl $t0, 3(%[src_uv]) \n" // V1 | U1 | V0 | U0
464 "lwr $t1, 4(%[src_uv]) \n"
465 "lwl $t1, 7(%[src_uv]) \n" // V3 | U3 | V2 | U2
466 "lwr $t2, 8(%[src_uv]) \n"
467 "lwl $t2, 11(%[src_uv]) \n" // V5 | U5 | V4 | U4
468 "lwr $t3, 12(%[src_uv]) \n"
469 "lwl $t3, 15(%[src_uv]) \n" // V7 | U7 | V6 | U6
470 "lwr $t5, 16(%[src_uv]) \n"
471 "lwl $t5, 19(%[src_uv]) \n" // V9 | U9 | V8 | U8
472 "lwr $t6, 20(%[src_uv]) \n"
473 "lwl $t6, 23(%[src_uv]) \n" // V11 | U11 | V10 | U10
474 "lwr $t7, 24(%[src_uv]) \n"
475 "lwl $t7, 27(%[src_uv]) \n" // V13 | U13 | V12 | U12
476 "lwr $t8, 28(%[src_uv]) \n"
477 "lwl $t8, 31(%[src_uv]) \n" // V15 | U15 | V14 | U14
478 "precrq.qb.ph $t9, $t1, $t0 \n" // V3 | V2 | V1 | V0
479 "precr.qb.ph $t0, $t1, $t0 \n" // U3 | U2 | U1 | U0
480 "precrq.qb.ph $t1, $t3, $t2 \n" // V7 | V6 | V5 | V4
481 "precr.qb.ph $t2, $t3, $t2 \n" // U7 | U6 | U5 | U4
482 "precrq.qb.ph $t3, $t6, $t5 \n" // V11 | V10 | V9 | V8
483 "precr.qb.ph $t5, $t6, $t5 \n" // U11 | U10 | U9 | U8
484 "precrq.qb.ph $t6, $t8, $t7 \n" // V15 | V14 | V13 | V12
485 "precr.qb.ph $t7, $t8, $t7 \n" // U15 | U14 | U13 | U12
486 "addiu %[src_uv], %[src_uv], 32 \n"
487 "swr $t9, 0(%[dst_v]) \n"
488 "swl $t9, 3(%[dst_v]) \n"
489 "swr $t0, 0(%[dst_u]) \n"
490 "swl $t0, 3(%[dst_u]) \n"
491 "swr $t1, 4(%[dst_v]) \n"
492 "swl $t1, 7(%[dst_v]) \n"
493 "swr $t2, 4(%[dst_u]) \n"
494 "swl $t2, 7(%[dst_u]) \n"
495 "swr $t3, 8(%[dst_v]) \n"
496 "swl $t3, 11(%[dst_v]) \n"
497 "swr $t5, 8(%[dst_u]) \n"
498 "swl $t5, 11(%[dst_u]) \n"
499 "swr $t6, 12(%[dst_v]) \n"
500 "swl $t6, 15(%[dst_v]) \n"
501 "swr $t7, 12(%[dst_u]) \n"
502 "swl $t7, 15(%[dst_u]) \n"
503 "addiu %[dst_u], %[dst_u], 16 \n"
505 " addiu %[dst_v], %[dst_v], 16 \n"
507 "beqz %[width], 3f \n"
511 "lbu $t0, 0(%[src_uv]) \n"
512 "lbu $t1, 1(%[src_uv]) \n"
513 "addiu %[src_uv], %[src_uv], 2 \n"
514 "addiu %[width], %[width], -1 \n"
515 "sb $t0, 0(%[dst_u]) \n"
516 "sb $t1, 0(%[dst_v]) \n"
517 "addiu %[dst_u], %[dst_u], 1 \n"
518 "bgtz %[width], 2b \n"
519 " addiu %[dst_v], %[dst_v], 1 \n"
523 : [src_uv] "+r" (src_uv),
524 [width] "+r" (width),
525 [dst_u] "+r" (dst_u),
528 : "t0", "t1", "t2", "t3",
529 "t4", "t5", "t6", "t7", "t8", "t9"
533 void MirrorRow_MIPS_DSPR2(const uint8* src, uint8* dst, int width) {
534 __asm__ __volatile__ (
538 "srl $t4, %[width], 4 \n" // multiplies of 16
539 "andi $t5, %[width], 0xf \n"
541 " addu %[src], %[src], %[width] \n" // src += width
545 "lw $t0, -16(%[src]) \n" // |3|2|1|0|
546 "lw $t1, -12(%[src]) \n" // |7|6|5|4|
547 "lw $t2, -8(%[src]) \n" // |11|10|9|8|
548 "lw $t3, -4(%[src]) \n" // |15|14|13|12|
549 "wsbh $t0, $t0 \n" // |2|3|0|1|
550 "wsbh $t1, $t1 \n" // |6|7|4|5|
551 "wsbh $t2, $t2 \n" // |10|11|8|9|
552 "wsbh $t3, $t3 \n" // |14|15|12|13|
553 "rotr $t0, $t0, 16 \n" // |0|1|2|3|
554 "rotr $t1, $t1, 16 \n" // |4|5|6|7|
555 "rotr $t2, $t2, 16 \n" // |8|9|10|11|
556 "rotr $t3, $t3, 16 \n" // |12|13|14|15|
557 "addiu %[src], %[src], -16 \n"
558 "addiu $t4, $t4, -1 \n"
559 "sw $t3, 0(%[dst]) \n" // |15|14|13|12|
560 "sw $t2, 4(%[dst]) \n" // |11|10|9|8|
561 "sw $t1, 8(%[dst]) \n" // |7|6|5|4|
562 "sw $t0, 12(%[dst]) \n" // |3|2|1|0|
564 " addiu %[dst], %[dst], 16 \n"
569 "lbu $t0, -1(%[src]) \n"
570 "addiu $t5, $t5, -1 \n"
571 "addiu %[src], %[src], -1 \n"
572 "sb $t0, 0(%[dst]) \n"
574 " addiu %[dst], %[dst], 1 \n"
578 : [src] "+r" (src), [dst] "+r" (dst)
579 : [width] "r" (width)
580 : "t0", "t1", "t2", "t3", "t4", "t5"
584 void MirrorUVRow_MIPS_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
588 __asm__ __volatile__ (
592 "addu $t4, %[width], %[width] \n"
593 "srl %[x], %[width], 4 \n"
594 "andi %[y], %[width], 0xf \n"
596 " addu %[src_uv], %[src_uv], $t4 \n"
600 "lw $t0, -32(%[src_uv]) \n" // |3|2|1|0|
601 "lw $t1, -28(%[src_uv]) \n" // |7|6|5|4|
602 "lw $t2, -24(%[src_uv]) \n" // |11|10|9|8|
603 "lw $t3, -20(%[src_uv]) \n" // |15|14|13|12|
604 "lw $t4, -16(%[src_uv]) \n" // |19|18|17|16|
605 "lw $t6, -12(%[src_uv]) \n" // |23|22|21|20|
606 "lw $t7, -8(%[src_uv]) \n" // |27|26|25|24|
607 "lw $t8, -4(%[src_uv]) \n" // |31|30|29|28|
609 "rotr $t0, $t0, 16 \n" // |1|0|3|2|
610 "rotr $t1, $t1, 16 \n" // |5|4|7|6|
611 "rotr $t2, $t2, 16 \n" // |9|8|11|10|
612 "rotr $t3, $t3, 16 \n" // |13|12|15|14|
613 "rotr $t4, $t4, 16 \n" // |17|16|19|18|
614 "rotr $t6, $t6, 16 \n" // |21|20|23|22|
615 "rotr $t7, $t7, 16 \n" // |25|24|27|26|
616 "rotr $t8, $t8, 16 \n" // |29|28|31|30|
617 "precr.qb.ph $t9, $t0, $t1 \n" // |0|2|4|6|
618 "precrq.qb.ph $t5, $t0, $t1 \n" // |1|3|5|7|
619 "precr.qb.ph $t0, $t2, $t3 \n" // |8|10|12|14|
620 "precrq.qb.ph $t1, $t2, $t3 \n" // |9|11|13|15|
621 "precr.qb.ph $t2, $t4, $t6 \n" // |16|18|20|22|
622 "precrq.qb.ph $t3, $t4, $t6 \n" // |17|19|21|23|
623 "precr.qb.ph $t4, $t7, $t8 \n" // |24|26|28|30|
624 "precrq.qb.ph $t6, $t7, $t8 \n" // |25|27|29|31|
625 "addiu %[src_uv], %[src_uv], -32 \n"
626 "addiu %[x], %[x], -1 \n"
627 "swr $t4, 0(%[dst_u]) \n"
628 "swl $t4, 3(%[dst_u]) \n" // |30|28|26|24|
629 "swr $t6, 0(%[dst_v]) \n"
630 "swl $t6, 3(%[dst_v]) \n" // |31|29|27|25|
631 "swr $t2, 4(%[dst_u]) \n"
632 "swl $t2, 7(%[dst_u]) \n" // |22|20|18|16|
633 "swr $t3, 4(%[dst_v]) \n"
634 "swl $t3, 7(%[dst_v]) \n" // |23|21|19|17|
635 "swr $t0, 8(%[dst_u]) \n"
636 "swl $t0, 11(%[dst_u]) \n" // |14|12|10|8|
637 "swr $t1, 8(%[dst_v]) \n"
638 "swl $t1, 11(%[dst_v]) \n" // |15|13|11|9|
639 "swr $t9, 12(%[dst_u]) \n"
640 "swl $t9, 15(%[dst_u]) \n" // |6|4|2|0|
641 "swr $t5, 12(%[dst_v]) \n"
642 "swl $t5, 15(%[dst_v]) \n" // |7|5|3|1|
643 "addiu %[dst_v], %[dst_v], 16 \n"
645 " addiu %[dst_u], %[dst_u], 16 \n"
652 "lbu $t0, -2(%[src_uv]) \n"
653 "lbu $t1, -1(%[src_uv]) \n"
654 "addiu %[src_uv], %[src_uv], -2 \n"
655 "addiu %[y], %[y], -1 \n"
656 "sb $t0, 0(%[dst_u]) \n"
657 "sb $t1, 0(%[dst_v]) \n"
658 "addiu %[dst_u], %[dst_u], 1 \n"
660 " addiu %[dst_v], %[dst_v], 1 \n"
664 : [src_uv] "+r" (src_uv),
665 [dst_u] "+r" (dst_u),
666 [dst_v] "+r" (dst_v),
669 : [width] "r" (width)
670 : "t0", "t1", "t2", "t3", "t4",
671 "t5", "t7", "t8", "t9"
675 // Convert (4 Y and 2 VU) I422 and arrange RGB values into
676 // t5 = | 0 | B0 | 0 | b0 |
677 // t4 = | 0 | B1 | 0 | b1 |
678 // t9 = | 0 | G0 | 0 | g0 |
679 // t8 = | 0 | G1 | 0 | g1 |
680 // t2 = | 0 | R0 | 0 | r0 |
681 // t1 = | 0 | R1 | 0 | r1 |
682 #define I422ToTransientMipsRGB \
683 "lw $t0, 0(%[y_buf]) \n" \
684 "lhu $t1, 0(%[u_buf]) \n" \
685 "lhu $t2, 0(%[v_buf]) \n" \
686 "preceu.ph.qbr $t1, $t1 \n" \
687 "preceu.ph.qbr $t2, $t2 \n" \
688 "preceu.ph.qbra $t3, $t0 \n" \
689 "preceu.ph.qbla $t0, $t0 \n" \
690 "subu.ph $t1, $t1, $s5 \n" \
691 "subu.ph $t2, $t2, $s5 \n" \
692 "subu.ph $t3, $t3, $s4 \n" \
693 "subu.ph $t0, $t0, $s4 \n" \
694 "mul.ph $t3, $t3, $s0 \n" \
695 "mul.ph $t0, $t0, $s0 \n" \
696 "shll.ph $t4, $t1, 0x7 \n" \
697 "subu.ph $t4, $t4, $t1 \n" \
698 "mul.ph $t6, $t1, $s1 \n" \
699 "mul.ph $t1, $t2, $s2 \n" \
700 "addq_s.ph $t5, $t4, $t3 \n" \
701 "addq_s.ph $t4, $t4, $t0 \n" \
702 "shra.ph $t5, $t5, 6 \n" \
703 "shra.ph $t4, $t4, 6 \n" \
704 "addiu %[u_buf], 2 \n" \
705 "addiu %[v_buf], 2 \n" \
706 "addu.ph $t6, $t6, $t1 \n" \
707 "mul.ph $t1, $t2, $s3 \n" \
708 "addu.ph $t9, $t6, $t3 \n" \
709 "addu.ph $t8, $t6, $t0 \n" \
710 "shra.ph $t9, $t9, 6 \n" \
711 "shra.ph $t8, $t8, 6 \n" \
712 "addu.ph $t2, $t1, $t3 \n" \
713 "addu.ph $t1, $t1, $t0 \n" \
714 "shra.ph $t2, $t2, 6 \n" \
715 "shra.ph $t1, $t1, 6 \n" \
716 "subu.ph $t5, $t5, $s5 \n" \
717 "subu.ph $t4, $t4, $s5 \n" \
718 "subu.ph $t9, $t9, $s5 \n" \
719 "subu.ph $t8, $t8, $s5 \n" \
720 "subu.ph $t2, $t2, $s5 \n" \
721 "subu.ph $t1, $t1, $s5 \n" \
722 "shll_s.ph $t5, $t5, 8 \n" \
723 "shll_s.ph $t4, $t4, 8 \n" \
724 "shll_s.ph $t9, $t9, 8 \n" \
725 "shll_s.ph $t8, $t8, 8 \n" \
726 "shll_s.ph $t2, $t2, 8 \n" \
727 "shll_s.ph $t1, $t1, 8 \n" \
728 "shra.ph $t5, $t5, 8 \n" \
729 "shra.ph $t4, $t4, 8 \n" \
730 "shra.ph $t9, $t9, 8 \n" \
731 "shra.ph $t8, $t8, 8 \n" \
732 "shra.ph $t2, $t2, 8 \n" \
733 "shra.ph $t1, $t1, 8 \n" \
734 "addu.ph $t5, $t5, $s5 \n" \
735 "addu.ph $t4, $t4, $s5 \n" \
736 "addu.ph $t9, $t9, $s5 \n" \
737 "addu.ph $t8, $t8, $s5 \n" \
738 "addu.ph $t2, $t2, $s5 \n" \
739 "addu.ph $t1, $t1, $s5 \n"
741 void I422ToARGBRow_MIPS_DSPR2(const uint8* y_buf,
746 __asm__ __volatile__ (
749 "beqz %[width], 2f \n"
750 " repl.ph $s0, 74 \n" // |YG|YG| = |74|74|
751 "repl.ph $s1, -25 \n" // |UG|UG| = |-25|-25|
752 "repl.ph $s2, -52 \n" // |VG|VG| = |-52|-52|
753 "repl.ph $s3, 102 \n" // |VR|VR| = |102|102|
754 "repl.ph $s4, 16 \n" // |0|16|0|16|
755 "repl.ph $s5, 128 \n" // |128|128| // clipping
757 "ori $s6, 0xff00 \n" // |ff|00|ff|00|ff|
761 I422ToTransientMipsRGB
762 // Arranging into argb format
763 "precr.qb.ph $t4, $t8, $t4 \n" // |G1|g1|B1|b1|
764 "precr.qb.ph $t5, $t9, $t5 \n" // |G0|g0|B0|b0|
765 "addiu %[width], -4 \n"
766 "precrq.qb.ph $t8, $t4, $t5 \n" // |G1|B1|G0|B0|
767 "precr.qb.ph $t9, $t4, $t5 \n" // |g1|b1|g0|b0|
768 "precr.qb.ph $t2, $t1, $t2 \n" // |R1|r1|R0|r0|
770 "addiu %[y_buf], 4 \n"
771 "preceu.ph.qbla $t1, $t2 \n" // |0 |R1|0 |R0|
772 "preceu.ph.qbra $t2, $t2 \n" // |0 |r1|0 |r0|
773 "or $t1, $t1, $s6 \n" // |ff|R1|ff|R0|
774 "or $t2, $t2, $s6 \n" // |ff|r1|ff|r0|
775 "precrq.ph.w $t0, $t2, $t9 \n" // |ff|r1|g1|b1|
776 "precrq.ph.w $t3, $t1, $t8 \n" // |ff|R1|G1|B1|
777 "sll $t9, $t9, 16 \n"
778 "sll $t8, $t8, 16 \n"
779 "packrl.ph $t2, $t2, $t9 \n" // |ff|r0|g0|b0|
780 "packrl.ph $t1, $t1, $t8 \n" // |ff|R0|G0|B0|
782 "sw $t2, 0(%[rgb_buf]) \n"
783 "sw $t0, 4(%[rgb_buf]) \n"
784 "sw $t1, 8(%[rgb_buf]) \n"
785 "sw $t3, 12(%[rgb_buf]) \n"
786 "bnez %[width], 1b \n"
787 " addiu %[rgb_buf], 16 \n"
790 :[y_buf] "+r" (y_buf),
791 [u_buf] "+r" (u_buf),
792 [v_buf] "+r" (v_buf),
793 [width] "+r" (width),
794 [rgb_buf] "+r" (rgb_buf)
796 : "t0", "t1", "t2", "t3", "t4", "t5",
797 "t6", "t7", "t8", "t9",
798 "s0", "s1", "s2", "s3",
803 void I422ToABGRRow_MIPS_DSPR2(const uint8* y_buf,
808 __asm__ __volatile__ (
811 "beqz %[width], 2f \n"
812 " repl.ph $s0, 74 \n" // |YG|YG| = |74|74|
813 "repl.ph $s1, -25 \n" // |UG|UG| = |-25|-25|
814 "repl.ph $s2, -52 \n" // |VG|VG| = |-52|-52|
815 "repl.ph $s3, 102 \n" // |VR|VR| = |102|102|
816 "repl.ph $s4, 16 \n" // |0|16|0|16|
817 "repl.ph $s5, 128 \n" // |128|128|
819 "ori $s6, 0xff00 \n" // |ff|00|ff|00|
823 I422ToTransientMipsRGB
824 // Arranging into abgr format
825 "precr.qb.ph $t0, $t8, $t1 \n" // |G1|g1|R1|r1|
826 "precr.qb.ph $t3, $t9, $t2 \n" // |G0|g0|R0|r0|
827 "precrq.qb.ph $t8, $t0, $t3 \n" // |G1|R1|G0|R0|
828 "precr.qb.ph $t9, $t0, $t3 \n" // |g1|r1|g0|r0|
830 "precr.qb.ph $t2, $t4, $t5 \n" // |B1|b1|B0|b0|
831 "addiu %[width], -4 \n"
832 "addiu %[y_buf], 4 \n"
833 "preceu.ph.qbla $t1, $t2 \n" // |0 |B1|0 |B0|
834 "preceu.ph.qbra $t2, $t2 \n" // |0 |b1|0 |b0|
835 "or $t1, $t1, $s6 \n" // |ff|B1|ff|B0|
836 "or $t2, $t2, $s6 \n" // |ff|b1|ff|b0|
837 "precrq.ph.w $t0, $t2, $t9 \n" // |ff|b1|g1|r1|
838 "precrq.ph.w $t3, $t1, $t8 \n" // |ff|B1|G1|R1|
839 "sll $t9, $t9, 16 \n"
840 "sll $t8, $t8, 16 \n"
841 "packrl.ph $t2, $t2, $t9 \n" // |ff|b0|g0|r0|
842 "packrl.ph $t1, $t1, $t8 \n" // |ff|B0|G0|R0|
844 "sw $t2, 0(%[rgb_buf]) \n"
845 "sw $t0, 4(%[rgb_buf]) \n"
846 "sw $t1, 8(%[rgb_buf]) \n"
847 "sw $t3, 12(%[rgb_buf]) \n"
848 "bnez %[width], 1b \n"
849 " addiu %[rgb_buf], 16 \n"
852 :[y_buf] "+r" (y_buf),
853 [u_buf] "+r" (u_buf),
854 [v_buf] "+r" (v_buf),
855 [width] "+r" (width),
856 [rgb_buf] "+r" (rgb_buf)
858 : "t0", "t1", "t2", "t3", "t4", "t5",
859 "t6", "t7", "t8", "t9",
860 "s0", "s1", "s2", "s3",
865 void I422ToBGRARow_MIPS_DSPR2(const uint8* y_buf,
870 __asm__ __volatile__ (
873 "beqz %[width], 2f \n"
874 " repl.ph $s0, 74 \n" // |YG|YG| = |74 |74 |
875 "repl.ph $s1, -25 \n" // |UG|UG| = |-25|-25|
876 "repl.ph $s2, -52 \n" // |VG|VG| = |-52|-52|
877 "repl.ph $s3, 102 \n" // |VR|VR| = |102|102|
878 "repl.ph $s4, 16 \n" // |0|16|0|16|
879 "repl.ph $s5, 128 \n" // |128|128|
881 "ori $s6, 0xff \n" // |00|ff|00|ff|
885 I422ToTransientMipsRGB
886 // Arranging into bgra format
887 "precr.qb.ph $t4, $t4, $t8 \n" // |B1|b1|G1|g1|
888 "precr.qb.ph $t5, $t5, $t9 \n" // |B0|b0|G0|g0|
889 "precrq.qb.ph $t8, $t4, $t5 \n" // |B1|G1|B0|G0|
890 "precr.qb.ph $t9, $t4, $t5 \n" // |b1|g1|b0|g0|
892 "precr.qb.ph $t2, $t1, $t2 \n" // |R1|r1|R0|r0|
893 "addiu %[width], -4 \n"
894 "addiu %[y_buf], 4 \n"
895 "preceu.ph.qbla $t1, $t2 \n" // |0 |R1|0 |R0|
896 "preceu.ph.qbra $t2, $t2 \n" // |0 |r1|0 |r0|
897 "sll $t1, $t1, 8 \n" // |R1|0 |R0|0 |
898 "sll $t2, $t2, 8 \n" // |r1|0 |r0|0 |
899 "or $t1, $t1, $s6 \n" // |R1|ff|R0|ff|
900 "or $t2, $t2, $s6 \n" // |r1|ff|r0|ff|
901 "precrq.ph.w $t0, $t9, $t2 \n" // |b1|g1|r1|ff|
902 "precrq.ph.w $t3, $t8, $t1 \n" // |B1|G1|R1|ff|
903 "sll $t1, $t1, 16 \n"
904 "sll $t2, $t2, 16 \n"
905 "packrl.ph $t2, $t9, $t2 \n" // |b0|g0|r0|ff|
906 "packrl.ph $t1, $t8, $t1 \n" // |B0|G0|R0|ff|
908 "sw $t2, 0(%[rgb_buf]) \n"
909 "sw $t0, 4(%[rgb_buf]) \n"
910 "sw $t1, 8(%[rgb_buf]) \n"
911 "sw $t3, 12(%[rgb_buf]) \n"
912 "bnez %[width], 1b \n"
913 " addiu %[rgb_buf], 16 \n"
916 :[y_buf] "+r" (y_buf),
917 [u_buf] "+r" (u_buf),
918 [v_buf] "+r" (v_buf),
919 [width] "+r" (width),
920 [rgb_buf] "+r" (rgb_buf)
922 : "t0", "t1", "t2", "t3", "t4", "t5",
923 "t6", "t7", "t8", "t9",
924 "s0", "s1", "s2", "s3",
929 // Bilinear filter 8x2 -> 8x1
930 void InterpolateRows_MIPS_DSPR2(uint8* dst_ptr, const uint8* src_ptr,
931 ptrdiff_t src_stride, int dst_width,
932 int source_y_fraction) {
933 int y0_fraction = 256 - source_y_fraction;
934 const uint8* src_ptr1 = src_ptr + src_stride;
936 __asm__ __volatile__ (
940 "replv.ph $t0, %[y0_fraction] \n"
941 "replv.ph $t1, %[source_y_fraction] \n"
945 "lw $t2, 0(%[src_ptr]) \n"
946 "lw $t3, 0(%[src_ptr1]) \n"
947 "lw $t4, 4(%[src_ptr]) \n"
948 "lw $t5, 4(%[src_ptr1]) \n"
949 "muleu_s.ph.qbl $t6, $t2, $t0 \n"
950 "muleu_s.ph.qbr $t7, $t2, $t0 \n"
951 "muleu_s.ph.qbl $t8, $t3, $t1 \n"
952 "muleu_s.ph.qbr $t9, $t3, $t1 \n"
953 "muleu_s.ph.qbl $t2, $t4, $t0 \n"
954 "muleu_s.ph.qbr $t3, $t4, $t0 \n"
955 "muleu_s.ph.qbl $t4, $t5, $t1 \n"
956 "muleu_s.ph.qbr $t5, $t5, $t1 \n"
957 "addq.ph $t6, $t6, $t8 \n"
958 "addq.ph $t7, $t7, $t9 \n"
959 "addq.ph $t2, $t2, $t4 \n"
960 "addq.ph $t3, $t3, $t5 \n"
961 "shra.ph $t6, $t6, 8 \n"
962 "shra.ph $t7, $t7, 8 \n"
963 "shra.ph $t2, $t2, 8 \n"
964 "shra.ph $t3, $t3, 8 \n"
965 "precr.qb.ph $t6, $t6, $t7 \n"
966 "precr.qb.ph $t2, $t2, $t3 \n"
967 "addiu %[src_ptr], %[src_ptr], 8 \n"
968 "addiu %[src_ptr1], %[src_ptr1], 8 \n"
969 "addiu %[dst_width], %[dst_width], -8 \n"
970 "sw $t6, 0(%[dst_ptr]) \n"
971 "sw $t2, 4(%[dst_ptr]) \n"
972 "bgtz %[dst_width], 1b \n"
973 " addiu %[dst_ptr], %[dst_ptr], 8 \n"
976 : [dst_ptr] "+r" (dst_ptr),
977 [src_ptr1] "+r" (src_ptr1),
978 [src_ptr] "+r" (src_ptr),
979 [dst_width] "+r" (dst_width)
980 : [source_y_fraction] "r" (source_y_fraction),
981 [y0_fraction] "r" (y0_fraction),
982 [src_stride] "r" (src_stride)
983 : "t0", "t1", "t2", "t3", "t4", "t5",
984 "t6", "t7", "t8", "t9"
987 #endif // __mips_dsp_rev >= 2
989 #endif // defined(__mips__)
993 } // namespace libyuv