1 /*
2 * Copyright 2012 The LibYuv Project Authors. All rights reserved.
3 *
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.
9 */
10
11 #include "libyuv/basic_types.h"
12 #include "libyuv/row.h"
13
14 #ifdef __cplusplus
15 namespace libyuv {
16 extern "C" {
17 #endif
18
19 // This module is for GCC MIPS DSPR2
20 #if !defined(LIBYUV_DISABLE_MIPS) && \
21 defined(__mips_dsp) && (__mips_dsp_rev >= 2)
22
ScaleRowDown2_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * dst,int dst_width)23 void ScaleRowDown2_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
24 uint8* dst, int dst_width) {
25 __asm__ __volatile__(
26 ".set push \n"
27 ".set noreorder \n"
28
29 "srl $t9, %[dst_width], 4 \n" // iterations -> by 16
30 "beqz $t9, 2f \n"
31 " nop \n"
32
33 ".p2align 2 \n"
34 "1: \n"
35 "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
36 "lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
37 "lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
38 "lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
39 "lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
40 "lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
41 "lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
42 "lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
43 // TODO(fbarchard): Use odd pixels instead of even.
44 "precr.qb.ph $t8, $t1, $t0 \n" // |6|4|2|0|
45 "precr.qb.ph $t0, $t3, $t2 \n" // |14|12|10|8|
46 "precr.qb.ph $t1, $t5, $t4 \n" // |22|20|18|16|
47 "precr.qb.ph $t2, $t7, $t6 \n" // |30|28|26|24|
48 "addiu %[src_ptr], %[src_ptr], 32 \n"
49 "addiu $t9, $t9, -1 \n"
50 "sw $t8, 0(%[dst]) \n"
51 "sw $t0, 4(%[dst]) \n"
52 "sw $t1, 8(%[dst]) \n"
53 "sw $t2, 12(%[dst]) \n"
54 "bgtz $t9, 1b \n"
55 " addiu %[dst], %[dst], 16 \n"
56
57 "2: \n"
58 "andi $t9, %[dst_width], 0xf \n" // residue
59 "beqz $t9, 3f \n"
60 " nop \n"
61
62 "21: \n"
63 "lbu $t0, 0(%[src_ptr]) \n"
64 "addiu %[src_ptr], %[src_ptr], 2 \n"
65 "addiu $t9, $t9, -1 \n"
66 "sb $t0, 0(%[dst]) \n"
67 "bgtz $t9, 21b \n"
68 " addiu %[dst], %[dst], 1 \n"
69
70 "3: \n"
71 ".set pop \n"
72 : [src_ptr] "+r" (src_ptr),
73 [dst] "+r" (dst)
74 : [dst_width] "r" (dst_width)
75 : "t0", "t1", "t2", "t3", "t4", "t5",
76 "t6", "t7", "t8", "t9"
77 );
78 }
79
ScaleRowDown2Box_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * dst,int dst_width)80 void ScaleRowDown2Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
81 uint8* dst, int dst_width) {
82 const uint8* t = src_ptr + src_stride;
83
84 __asm__ __volatile__ (
85 ".set push \n"
86 ".set noreorder \n"
87
88 "srl $t9, %[dst_width], 3 \n" // iterations -> step 8
89 "bltz $t9, 2f \n"
90 " nop \n"
91
92 ".p2align 2 \n"
93 "1: \n"
94 "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
95 "lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
96 "lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
97 "lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
98 "lw $t4, 0(%[t]) \n" // |19|18|17|16|
99 "lw $t5, 4(%[t]) \n" // |23|22|21|20|
100 "lw $t6, 8(%[t]) \n" // |27|26|25|24|
101 "lw $t7, 12(%[t]) \n" // |31|30|29|28|
102 "addiu $t9, $t9, -1 \n"
103 "srl $t8, $t0, 16 \n" // |X|X|3|2|
104 "ins $t0, $t4, 16, 16 \n" // |17|16|1|0|
105 "ins $t4, $t8, 0, 16 \n" // |19|18|3|2|
106 "raddu.w.qb $t0, $t0 \n" // |17+16+1+0|
107 "raddu.w.qb $t4, $t4 \n" // |19+18+3+2|
108 "shra_r.w $t0, $t0, 2 \n" // |t0+2|>>2
109 "shra_r.w $t4, $t4, 2 \n" // |t4+2|>>2
110 "srl $t8, $t1, 16 \n" // |X|X|7|6|
111 "ins $t1, $t5, 16, 16 \n" // |21|20|5|4|
112 "ins $t5, $t8, 0, 16 \n" // |22|23|7|6|
113 "raddu.w.qb $t1, $t1 \n" // |21+20+5+4|
114 "raddu.w.qb $t5, $t5 \n" // |23+22+7+6|
115 "shra_r.w $t1, $t1, 2 \n" // |t1+2|>>2
116 "shra_r.w $t5, $t5, 2 \n" // |t5+2|>>2
117 "srl $t8, $t2, 16 \n" // |X|X|11|10|
118 "ins $t2, $t6, 16, 16 \n" // |25|24|9|8|
119 "ins $t6, $t8, 0, 16 \n" // |27|26|11|10|
120 "raddu.w.qb $t2, $t2 \n" // |25+24+9+8|
121 "raddu.w.qb $t6, $t6 \n" // |27+26+11+10|
122 "shra_r.w $t2, $t2, 2 \n" // |t2+2|>>2
123 "shra_r.w $t6, $t6, 2 \n" // |t5+2|>>2
124 "srl $t8, $t3, 16 \n" // |X|X|15|14|
125 "ins $t3, $t7, 16, 16 \n" // |29|28|13|12|
126 "ins $t7, $t8, 0, 16 \n" // |31|30|15|14|
127 "raddu.w.qb $t3, $t3 \n" // |29+28+13+12|
128 "raddu.w.qb $t7, $t7 \n" // |31+30+15+14|
129 "shra_r.w $t3, $t3, 2 \n" // |t3+2|>>2
130 "shra_r.w $t7, $t7, 2 \n" // |t7+2|>>2
131 "addiu %[src_ptr], %[src_ptr], 16 \n"
132 "addiu %[t], %[t], 16 \n"
133 "sb $t0, 0(%[dst]) \n"
134 "sb $t4, 1(%[dst]) \n"
135 "sb $t1, 2(%[dst]) \n"
136 "sb $t5, 3(%[dst]) \n"
137 "sb $t2, 4(%[dst]) \n"
138 "sb $t6, 5(%[dst]) \n"
139 "sb $t3, 6(%[dst]) \n"
140 "sb $t7, 7(%[dst]) \n"
141 "bgtz $t9, 1b \n"
142 " addiu %[dst], %[dst], 8 \n"
143
144 "2: \n"
145 "andi $t9, %[dst_width], 0x7 \n" // x = residue
146 "beqz $t9, 3f \n"
147 " nop \n"
148
149 "21: \n"
150 "lwr $t1, 0(%[src_ptr]) \n"
151 "lwl $t1, 3(%[src_ptr]) \n"
152 "lwr $t2, 0(%[t]) \n"
153 "lwl $t2, 3(%[t]) \n"
154 "srl $t8, $t1, 16 \n"
155 "ins $t1, $t2, 16, 16 \n"
156 "ins $t2, $t8, 0, 16 \n"
157 "raddu.w.qb $t1, $t1 \n"
158 "raddu.w.qb $t2, $t2 \n"
159 "shra_r.w $t1, $t1, 2 \n"
160 "shra_r.w $t2, $t2, 2 \n"
161 "sb $t1, 0(%[dst]) \n"
162 "sb $t2, 1(%[dst]) \n"
163 "addiu %[src_ptr], %[src_ptr], 4 \n"
164 "addiu $t9, $t9, -2 \n"
165 "addiu %[t], %[t], 4 \n"
166 "bgtz $t9, 21b \n"
167 " addiu %[dst], %[dst], 2 \n"
168
169 "3: \n"
170 ".set pop \n"
171
172 : [src_ptr] "+r" (src_ptr),
173 [dst] "+r" (dst), [t] "+r" (t)
174 : [dst_width] "r" (dst_width)
175 : "t0", "t1", "t2", "t3", "t4", "t5",
176 "t6", "t7", "t8", "t9"
177 );
178 }
179
ScaleRowDown4_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * dst,int dst_width)180 void ScaleRowDown4_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
181 uint8* dst, int dst_width) {
182 __asm__ __volatile__ (
183 ".set push \n"
184 ".set noreorder \n"
185
186 "srl $t9, %[dst_width], 3 \n"
187 "beqz $t9, 2f \n"
188 " nop \n"
189
190 ".p2align 2 \n"
191 "1: \n"
192 "lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
193 "lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
194 "lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
195 "lw $t4, 12(%[src_ptr]) \n" // |15|14|13|12|
196 "lw $t5, 16(%[src_ptr]) \n" // |19|18|17|16|
197 "lw $t6, 20(%[src_ptr]) \n" // |23|22|21|20|
198 "lw $t7, 24(%[src_ptr]) \n" // |27|26|25|24|
199 "lw $t8, 28(%[src_ptr]) \n" // |31|30|29|28|
200 "precr.qb.ph $t1, $t2, $t1 \n" // |6|4|2|0|
201 "precr.qb.ph $t2, $t4, $t3 \n" // |14|12|10|8|
202 "precr.qb.ph $t5, $t6, $t5 \n" // |22|20|18|16|
203 "precr.qb.ph $t6, $t8, $t7 \n" // |30|28|26|24|
204 "precr.qb.ph $t1, $t2, $t1 \n" // |12|8|4|0|
205 "precr.qb.ph $t5, $t6, $t5 \n" // |28|24|20|16|
206 "addiu %[src_ptr], %[src_ptr], 32 \n"
207 "addiu $t9, $t9, -1 \n"
208 "sw $t1, 0(%[dst]) \n"
209 "sw $t5, 4(%[dst]) \n"
210 "bgtz $t9, 1b \n"
211 " addiu %[dst], %[dst], 8 \n"
212
213 "2: \n"
214 "andi $t9, %[dst_width], 7 \n" // residue
215 "beqz $t9, 3f \n"
216 " nop \n"
217
218 "21: \n"
219 "lbu $t1, 0(%[src_ptr]) \n"
220 "addiu %[src_ptr], %[src_ptr], 4 \n"
221 "addiu $t9, $t9, -1 \n"
222 "sb $t1, 0(%[dst]) \n"
223 "bgtz $t9, 21b \n"
224 " addiu %[dst], %[dst], 1 \n"
225
226 "3: \n"
227 ".set pop \n"
228 : [src_ptr] "+r" (src_ptr),
229 [dst] "+r" (dst)
230 : [dst_width] "r" (dst_width)
231 : "t1", "t2", "t3", "t4", "t5",
232 "t6", "t7", "t8", "t9"
233 );
234 }
235
ScaleRowDown4Box_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * dst,int dst_width)236 void ScaleRowDown4Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
237 uint8* dst, int dst_width) {
238 intptr_t stride = src_stride;
239 const uint8* s1 = src_ptr + stride;
240 const uint8* s2 = s1 + stride;
241 const uint8* s3 = s2 + stride;
242
243 __asm__ __volatile__ (
244 ".set push \n"
245 ".set noreorder \n"
246
247 "srl $t9, %[dst_width], 1 \n"
248 "andi $t8, %[dst_width], 1 \n"
249
250 ".p2align 2 \n"
251 "1: \n"
252 "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
253 "lw $t1, 0(%[s1]) \n" // |7|6|5|4|
254 "lw $t2, 0(%[s2]) \n" // |11|10|9|8|
255 "lw $t3, 0(%[s3]) \n" // |15|14|13|12|
256 "lw $t4, 4(%[src_ptr]) \n" // |19|18|17|16|
257 "lw $t5, 4(%[s1]) \n" // |23|22|21|20|
258 "lw $t6, 4(%[s2]) \n" // |27|26|25|24|
259 "lw $t7, 4(%[s3]) \n" // |31|30|29|28|
260 "raddu.w.qb $t0, $t0 \n" // |3 + 2 + 1 + 0|
261 "raddu.w.qb $t1, $t1 \n" // |7 + 6 + 5 + 4|
262 "raddu.w.qb $t2, $t2 \n" // |11 + 10 + 9 + 8|
263 "raddu.w.qb $t3, $t3 \n" // |15 + 14 + 13 + 12|
264 "raddu.w.qb $t4, $t4 \n" // |19 + 18 + 17 + 16|
265 "raddu.w.qb $t5, $t5 \n" // |23 + 22 + 21 + 20|
266 "raddu.w.qb $t6, $t6 \n" // |27 + 26 + 25 + 24|
267 "raddu.w.qb $t7, $t7 \n" // |31 + 30 + 29 + 28|
268 "add $t0, $t0, $t1 \n"
269 "add $t1, $t2, $t3 \n"
270 "add $t0, $t0, $t1 \n"
271 "add $t4, $t4, $t5 \n"
272 "add $t6, $t6, $t7 \n"
273 "add $t4, $t4, $t6 \n"
274 "shra_r.w $t0, $t0, 4 \n"
275 "shra_r.w $t4, $t4, 4 \n"
276 "sb $t0, 0(%[dst]) \n"
277 "sb $t4, 1(%[dst]) \n"
278 "addiu %[src_ptr], %[src_ptr], 8 \n"
279 "addiu %[s1], %[s1], 8 \n"
280 "addiu %[s2], %[s2], 8 \n"
281 "addiu %[s3], %[s3], 8 \n"
282 "addiu $t9, $t9, -1 \n"
283 "bgtz $t9, 1b \n"
284 " addiu %[dst], %[dst], 2 \n"
285 "beqz $t8, 2f \n"
286 " nop \n"
287
288 "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
289 "lw $t1, 0(%[s1]) \n" // |7|6|5|4|
290 "lw $t2, 0(%[s2]) \n" // |11|10|9|8|
291 "lw $t3, 0(%[s3]) \n" // |15|14|13|12|
292 "raddu.w.qb $t0, $t0 \n" // |3 + 2 + 1 + 0|
293 "raddu.w.qb $t1, $t1 \n" // |7 + 6 + 5 + 4|
294 "raddu.w.qb $t2, $t2 \n" // |11 + 10 + 9 + 8|
295 "raddu.w.qb $t3, $t3 \n" // |15 + 14 + 13 + 12|
296 "add $t0, $t0, $t1 \n"
297 "add $t1, $t2, $t3 \n"
298 "add $t0, $t0, $t1 \n"
299 "shra_r.w $t0, $t0, 4 \n"
300 "sb $t0, 0(%[dst]) \n"
301
302 "2: \n"
303 ".set pop \n"
304
305 : [src_ptr] "+r" (src_ptr),
306 [dst] "+r" (dst),
307 [s1] "+r" (s1),
308 [s2] "+r" (s2),
309 [s3] "+r" (s3)
310 : [dst_width] "r" (dst_width)
311 : "t0", "t1", "t2", "t3", "t4", "t5",
312 "t6","t7", "t8", "t9"
313 );
314 }
315
ScaleRowDown34_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * dst,int dst_width)316 void ScaleRowDown34_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
317 uint8* dst, int dst_width) {
318 __asm__ __volatile__ (
319 ".set push \n"
320 ".set noreorder \n"
321 ".p2align 2 \n"
322 "1: \n"
323 "lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
324 "lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
325 "lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
326 "lw $t4, 12(%[src_ptr]) \n" // |15|14|13|12|
327 "lw $t5, 16(%[src_ptr]) \n" // |19|18|17|16|
328 "lw $t6, 20(%[src_ptr]) \n" // |23|22|21|20|
329 "lw $t7, 24(%[src_ptr]) \n" // |27|26|25|24|
330 "lw $t8, 28(%[src_ptr]) \n" // |31|30|29|28|
331 "precrq.qb.ph $t0, $t2, $t4 \n" // |7|5|15|13|
332 "precrq.qb.ph $t9, $t6, $t8 \n" // |23|21|31|30|
333 "addiu %[dst_width], %[dst_width], -24 \n"
334 "ins $t1, $t1, 8, 16 \n" // |3|1|0|X|
335 "ins $t4, $t0, 8, 16 \n" // |X|15|13|12|
336 "ins $t5, $t5, 8, 16 \n" // |19|17|16|X|
337 "ins $t8, $t9, 8, 16 \n" // |X|31|29|28|
338 "addiu %[src_ptr], %[src_ptr], 32 \n"
339 "packrl.ph $t0, $t3, $t0 \n" // |9|8|7|5|
340 "packrl.ph $t9, $t7, $t9 \n" // |25|24|23|21|
341 "prepend $t1, $t2, 8 \n" // |4|3|1|0|
342 "prepend $t3, $t4, 24 \n" // |15|13|12|11|
343 "prepend $t5, $t6, 8 \n" // |20|19|17|16|
344 "prepend $t7, $t8, 24 \n" // |31|29|28|27|
345 "sw $t1, 0(%[dst]) \n"
346 "sw $t0, 4(%[dst]) \n"
347 "sw $t3, 8(%[dst]) \n"
348 "sw $t5, 12(%[dst]) \n"
349 "sw $t9, 16(%[dst]) \n"
350 "sw $t7, 20(%[dst]) \n"
351 "bnez %[dst_width], 1b \n"
352 " addiu %[dst], %[dst], 24 \n"
353 ".set pop \n"
354 : [src_ptr] "+r" (src_ptr),
355 [dst] "+r" (dst),
356 [dst_width] "+r" (dst_width)
357 :
358 : "t0", "t1", "t2", "t3", "t4", "t5",
359 "t6","t7", "t8", "t9"
360 );
361 }
362
ScaleRowDown34_0_Box_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * d,int dst_width)363 void ScaleRowDown34_0_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
364 uint8* d, int dst_width) {
365 __asm__ __volatile__ (
366 ".set push \n"
367 ".set noreorder \n"
368 "repl.ph $t3, 3 \n" // 0x00030003
369
370 ".p2align 2 \n"
371 "1: \n"
372 "lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
373 "lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
374 "rotr $t2, $t0, 8 \n" // |S0|S3|S2|S1|
375 "rotr $t6, $t1, 8 \n" // |T0|T3|T2|T1|
376 "muleu_s.ph.qbl $t4, $t2, $t3 \n" // |S0*3|S3*3|
377 "muleu_s.ph.qbl $t5, $t6, $t3 \n" // |T0*3|T3*3|
378 "andi $t0, $t2, 0xFFFF \n" // |0|0|S2|S1|
379 "andi $t1, $t6, 0xFFFF \n" // |0|0|T2|T1|
380 "raddu.w.qb $t0, $t0 \n"
381 "raddu.w.qb $t1, $t1 \n"
382 "shra_r.w $t0, $t0, 1 \n"
383 "shra_r.w $t1, $t1, 1 \n"
384 "preceu.ph.qbr $t2, $t2 \n" // |0|S2|0|S1|
385 "preceu.ph.qbr $t6, $t6 \n" // |0|T2|0|T1|
386 "rotr $t2, $t2, 16 \n" // |0|S1|0|S2|
387 "rotr $t6, $t6, 16 \n" // |0|T1|0|T2|
388 "addu.ph $t2, $t2, $t4 \n"
389 "addu.ph $t6, $t6, $t5 \n"
390 "sll $t5, $t0, 1 \n"
391 "add $t0, $t5, $t0 \n"
392 "shra_r.ph $t2, $t2, 2 \n"
393 "shra_r.ph $t6, $t6, 2 \n"
394 "shll.ph $t4, $t2, 1 \n"
395 "addq.ph $t4, $t4, $t2 \n"
396 "addu $t0, $t0, $t1 \n"
397 "addiu %[src_ptr], %[src_ptr], 4 \n"
398 "shra_r.w $t0, $t0, 2 \n"
399 "addu.ph $t6, $t6, $t4 \n"
400 "shra_r.ph $t6, $t6, 2 \n"
401 "srl $t1, $t6, 16 \n"
402 "addiu %[dst_width], %[dst_width], -3 \n"
403 "sb $t1, 0(%[d]) \n"
404 "sb $t0, 1(%[d]) \n"
405 "sb $t6, 2(%[d]) \n"
406 "bgtz %[dst_width], 1b \n"
407 " addiu %[d], %[d], 3 \n"
408 "3: \n"
409 ".set pop \n"
410 : [src_ptr] "+r" (src_ptr),
411 [src_stride] "+r" (src_stride),
412 [d] "+r" (d),
413 [dst_width] "+r" (dst_width)
414 :
415 : "t0", "t1", "t2", "t3",
416 "t4", "t5", "t6"
417 );
418 }
419
ScaleRowDown34_1_Box_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * d,int dst_width)420 void ScaleRowDown34_1_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
421 uint8* d, int dst_width) {
422 __asm__ __volatile__ (
423 ".set push \n"
424 ".set noreorder \n"
425 "repl.ph $t2, 3 \n" // 0x00030003
426
427 ".p2align 2 \n"
428 "1: \n"
429 "lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
430 "lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
431 "rotr $t4, $t0, 8 \n" // |S0|S3|S2|S1|
432 "rotr $t6, $t1, 8 \n" // |T0|T3|T2|T1|
433 "muleu_s.ph.qbl $t3, $t4, $t2 \n" // |S0*3|S3*3|
434 "muleu_s.ph.qbl $t5, $t6, $t2 \n" // |T0*3|T3*3|
435 "andi $t0, $t4, 0xFFFF \n" // |0|0|S2|S1|
436 "andi $t1, $t6, 0xFFFF \n" // |0|0|T2|T1|
437 "raddu.w.qb $t0, $t0 \n"
438 "raddu.w.qb $t1, $t1 \n"
439 "shra_r.w $t0, $t0, 1 \n"
440 "shra_r.w $t1, $t1, 1 \n"
441 "preceu.ph.qbr $t4, $t4 \n" // |0|S2|0|S1|
442 "preceu.ph.qbr $t6, $t6 \n" // |0|T2|0|T1|
443 "rotr $t4, $t4, 16 \n" // |0|S1|0|S2|
444 "rotr $t6, $t6, 16 \n" // |0|T1|0|T2|
445 "addu.ph $t4, $t4, $t3 \n"
446 "addu.ph $t6, $t6, $t5 \n"
447 "shra_r.ph $t6, $t6, 2 \n"
448 "shra_r.ph $t4, $t4, 2 \n"
449 "addu.ph $t6, $t6, $t4 \n"
450 "addiu %[src_ptr], %[src_ptr], 4 \n"
451 "shra_r.ph $t6, $t6, 1 \n"
452 "addu $t0, $t0, $t1 \n"
453 "addiu %[dst_width], %[dst_width], -3 \n"
454 "shra_r.w $t0, $t0, 1 \n"
455 "srl $t1, $t6, 16 \n"
456 "sb $t1, 0(%[d]) \n"
457 "sb $t0, 1(%[d]) \n"
458 "sb $t6, 2(%[d]) \n"
459 "bgtz %[dst_width], 1b \n"
460 " addiu %[d], %[d], 3 \n"
461 "3: \n"
462 ".set pop \n"
463 : [src_ptr] "+r" (src_ptr),
464 [src_stride] "+r" (src_stride),
465 [d] "+r" (d),
466 [dst_width] "+r" (dst_width)
467 :
468 : "t0", "t1", "t2", "t3",
469 "t4", "t5", "t6"
470 );
471 }
472
ScaleRowDown38_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * dst,int dst_width)473 void ScaleRowDown38_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
474 uint8* dst, int dst_width) {
475 __asm__ __volatile__ (
476 ".set push \n"
477 ".set noreorder \n"
478
479 ".p2align 2 \n"
480 "1: \n"
481 "lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
482 "lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
483 "lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
484 "lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
485 "lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
486 "lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
487 "lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
488 "lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
489 "wsbh $t0, $t0 \n" // |2|3|0|1|
490 "wsbh $t6, $t6 \n" // |26|27|24|25|
491 "srl $t0, $t0, 8 \n" // |X|2|3|0|
492 "srl $t3, $t3, 16 \n" // |X|X|15|14|
493 "srl $t5, $t5, 16 \n" // |X|X|23|22|
494 "srl $t7, $t7, 16 \n" // |X|X|31|30|
495 "ins $t1, $t2, 24, 8 \n" // |8|6|5|4|
496 "ins $t6, $t5, 0, 8 \n" // |26|27|24|22|
497 "ins $t1, $t0, 0, 16 \n" // |8|6|3|0|
498 "ins $t6, $t7, 24, 8 \n" // |30|27|24|22|
499 "prepend $t2, $t3, 24 \n" // |X|15|14|11|
500 "ins $t4, $t4, 16, 8 \n" // |19|16|17|X|
501 "ins $t4, $t2, 0, 16 \n" // |19|16|14|11|
502 "addiu %[src_ptr], %[src_ptr], 32 \n"
503 "addiu %[dst_width], %[dst_width], -12 \n"
504 "addiu $t8,%[dst_width], -12 \n"
505 "sw $t1, 0(%[dst]) \n"
506 "sw $t4, 4(%[dst]) \n"
507 "sw $t6, 8(%[dst]) \n"
508 "bgez $t8, 1b \n"
509 " addiu %[dst], %[dst], 12 \n"
510 ".set pop \n"
511 : [src_ptr] "+r" (src_ptr),
512 [dst] "+r" (dst),
513 [dst_width] "+r" (dst_width)
514 :
515 : "t0", "t1", "t2", "t3", "t4",
516 "t5", "t6", "t7", "t8"
517 );
518 }
519
ScaleRowDown38_2_Box_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * dst_ptr,int dst_width)520 void ScaleRowDown38_2_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
521 uint8* dst_ptr, int dst_width) {
522 intptr_t stride = src_stride;
523 const uint8* t = src_ptr + stride;
524 const int c = 0x2AAA;
525
526 __asm__ __volatile__ (
527 ".set push \n"
528 ".set noreorder \n"
529
530 ".p2align 2 \n"
531 "1: \n"
532 "lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
533 "lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
534 "lw $t2, 0(%[t]) \n" // |T3|T2|T1|T0|
535 "lw $t3, 4(%[t]) \n" // |T7|T6|T5|T4|
536 "rotr $t1, $t1, 16 \n" // |S5|S4|S7|S6|
537 "packrl.ph $t4, $t1, $t3 \n" // |S7|S6|T7|T6|
538 "packrl.ph $t5, $t3, $t1 \n" // |T5|T4|S5|S4|
539 "raddu.w.qb $t4, $t4 \n" // S7+S6+T7+T6
540 "raddu.w.qb $t5, $t5 \n" // T5+T4+S5+S4
541 "precrq.qb.ph $t6, $t0, $t2 \n" // |S3|S1|T3|T1|
542 "precrq.qb.ph $t6, $t6, $t6 \n" // |S3|T3|S3|T3|
543 "srl $t4, $t4, 2 \n" // t4 / 4
544 "srl $t6, $t6, 16 \n" // |0|0|S3|T3|
545 "raddu.w.qb $t6, $t6 \n" // 0+0+S3+T3
546 "addu $t6, $t5, $t6 \n"
547 "mul $t6, $t6, %[c] \n" // t6 * 0x2AAA
548 "sll $t0, $t0, 8 \n" // |S2|S1|S0|0|
549 "sll $t2, $t2, 8 \n" // |T2|T1|T0|0|
550 "raddu.w.qb $t0, $t0 \n" // S2+S1+S0+0
551 "raddu.w.qb $t2, $t2 \n" // T2+T1+T0+0
552 "addu $t0, $t0, $t2 \n"
553 "mul $t0, $t0, %[c] \n" // t0 * 0x2AAA
554 "addiu %[src_ptr], %[src_ptr], 8 \n"
555 "addiu %[t], %[t], 8 \n"
556 "addiu %[dst_width], %[dst_width], -3 \n"
557 "addiu %[dst_ptr], %[dst_ptr], 3 \n"
558 "srl $t6, $t6, 16 \n"
559 "srl $t0, $t0, 16 \n"
560 "sb $t4, -1(%[dst_ptr]) \n"
561 "sb $t6, -2(%[dst_ptr]) \n"
562 "bgtz %[dst_width], 1b \n"
563 " sb $t0, -3(%[dst_ptr]) \n"
564 ".set pop \n"
565 : [src_ptr] "+r" (src_ptr),
566 [dst_ptr] "+r" (dst_ptr),
567 [t] "+r" (t),
568 [dst_width] "+r" (dst_width)
569 : [c] "r" (c)
570 : "t0", "t1", "t2", "t3", "t4", "t5", "t6"
571 );
572 }
573
ScaleRowDown38_3_Box_MIPS_DSPR2(const uint8 * src_ptr,ptrdiff_t src_stride,uint8 * dst_ptr,int dst_width)574 void ScaleRowDown38_3_Box_MIPS_DSPR2(const uint8* src_ptr,
575 ptrdiff_t src_stride,
576 uint8* dst_ptr, int dst_width) {
577 intptr_t stride = src_stride;
578 const uint8* s1 = src_ptr + stride;
579 stride += stride;
580 const uint8* s2 = src_ptr + stride;
581 const int c1 = 0x1C71;
582 const int c2 = 0x2AAA;
583
584 __asm__ __volatile__ (
585 ".set push \n"
586 ".set noreorder \n"
587
588 ".p2align 2 \n"
589 "1: \n"
590 "lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
591 "lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
592 "lw $t2, 0(%[s1]) \n" // |T3|T2|T1|T0|
593 "lw $t3, 4(%[s1]) \n" // |T7|T6|T5|T4|
594 "lw $t4, 0(%[s2]) \n" // |R3|R2|R1|R0|
595 "lw $t5, 4(%[s2]) \n" // |R7|R6|R5|R4|
596 "rotr $t1, $t1, 16 \n" // |S5|S4|S7|S6|
597 "packrl.ph $t6, $t1, $t3 \n" // |S7|S6|T7|T6|
598 "raddu.w.qb $t6, $t6 \n" // S7+S6+T7+T6
599 "packrl.ph $t7, $t3, $t1 \n" // |T5|T4|S5|S4|
600 "raddu.w.qb $t7, $t7 \n" // T5+T4+S5+S4
601 "sll $t8, $t5, 16 \n" // |R5|R4|0|0|
602 "raddu.w.qb $t8, $t8 \n" // R5+R4
603 "addu $t7, $t7, $t8 \n"
604 "srl $t8, $t5, 16 \n" // |0|0|R7|R6|
605 "raddu.w.qb $t8, $t8 \n" // R7 + R6
606 "addu $t6, $t6, $t8 \n"
607 "mul $t6, $t6, %[c2] \n" // t6 * 0x2AAA
608 "precrq.qb.ph $t8, $t0, $t2 \n" // |S3|S1|T3|T1|
609 "precrq.qb.ph $t8, $t8, $t4 \n" // |S3|T3|R3|R1|
610 "srl $t8, $t8, 8 \n" // |0|S3|T3|R3|
611 "raddu.w.qb $t8, $t8 \n" // S3 + T3 + R3
612 "addu $t7, $t7, $t8 \n"
613 "mul $t7, $t7, %[c1] \n" // t7 * 0x1C71
614 "sll $t0, $t0, 8 \n" // |S2|S1|S0|0|
615 "sll $t2, $t2, 8 \n" // |T2|T1|T0|0|
616 "sll $t4, $t4, 8 \n" // |R2|R1|R0|0|
617 "raddu.w.qb $t0, $t0 \n"
618 "raddu.w.qb $t2, $t2 \n"
619 "raddu.w.qb $t4, $t4 \n"
620 "addu $t0, $t0, $t2 \n"
621 "addu $t0, $t0, $t4 \n"
622 "mul $t0, $t0, %[c1] \n" // t0 * 0x1C71
623 "addiu %[src_ptr], %[src_ptr], 8 \n"
624 "addiu %[s1], %[s1], 8 \n"
625 "addiu %[s2], %[s2], 8 \n"
626 "addiu %[dst_width], %[dst_width], -3 \n"
627 "addiu %[dst_ptr], %[dst_ptr], 3 \n"
628 "srl $t6, $t6, 16 \n"
629 "srl $t7, $t7, 16 \n"
630 "srl $t0, $t0, 16 \n"
631 "sb $t6, -1(%[dst_ptr]) \n"
632 "sb $t7, -2(%[dst_ptr]) \n"
633 "bgtz %[dst_width], 1b \n"
634 " sb $t0, -3(%[dst_ptr]) \n"
635 ".set pop \n"
636 : [src_ptr] "+r" (src_ptr),
637 [dst_ptr] "+r" (dst_ptr),
638 [s1] "+r" (s1),
639 [s2] "+r" (s2),
640 [dst_width] "+r" (dst_width)
641 : [c1] "r" (c1), [c2] "r" (c2)
642 : "t0", "t1", "t2", "t3", "t4",
643 "t5", "t6", "t7", "t8"
644 );
645 }
646
647 #endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
648
649 #ifdef __cplusplus
650 } // extern "C"
651 } // namespace libyuv
652 #endif
653
654