Better ARM support for mplayer/ffmpeg, ported from atty fork
[libav.git] / libavcodec / armv4l / dsputil_iwmmxt_rnd.h
1 void DEF(put, pixels8)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
2 {
3 int stride = line_size;
4 __asm__ __volatile__ (
5 "and r12, %[pixels], #7 \n\t"
6 "bic %[pixels], %[pixels], #7 \n\t"
7 "tmcr wcgr1, r12 \n\t"
8 "add r4, %[pixels], %[line_size] \n\t"
9 "add r5, %[block], %[line_size] \n\t"
10 "mov %[line_size], %[line_size], lsl #1 \n\t"
11 "1: \n\t"
12 "wldrd wr0, [%[pixels]] \n\t"
13 "subs %[h], %[h], #2 \n\t"
14 "wldrd wr1, [%[pixels], #8] \n\t"
15 "add %[pixels], %[pixels], %[line_size] \n\t"
16 "wldrd wr3, [r4] \n\t"
17 "pld [%[pixels]] \n\t"
18 "pld [%[pixels], #32] \n\t"
19 "wldrd wr4, [r4, #8] \n\t"
20 "add r4, r4, %[line_size] \n\t"
21 "walignr1 wr8, wr0, wr1 \n\t"
22 "pld [r4] \n\t"
23 "pld [r4, #32] \n\t"
24 "walignr1 wr10, wr3, wr4 \n\t"
25 "wstrd wr8, [%[block]] \n\t"
26 "add %[block], %[block], %[line_size] \n\t"
27 "wstrd wr10, [r5] \n\t"
28 "add r5, r5, %[line_size] \n\t"
29 "bne 1b \n\t"
30 : [block]"+r"(block), [pixels]"+r"(pixels), [line_size]"+r"(stride), [h]"+r"(h)
31 :
32 : "memory", "r4", "r5", "r12");
33 }
34
35 void DEF(avg, pixels8)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
36 {
37 int stride = line_size;
38 __asm__ __volatile__ (
39 "and r12, %[pixels], #7 \n\t"
40 "bic %[pixels], %[pixels], #7 \n\t"
41 "tmcr wcgr1, r12 \n\t"
42 "add r4, %[pixels], %[line_size] \n\t"
43 "add r5, %[block], %[line_size] \n\t"
44 "mov %[line_size], %[line_size], lsl #1 \n\t"
45 "1: \n\t"
46 "wldrd wr0, [%[pixels]] \n\t"
47 "subs %[h], %[h], #2 \n\t"
48 "wldrd wr1, [%[pixels], #8] \n\t"
49 "add %[pixels], %[pixels], %[line_size] \n\t"
50 "wldrd wr3, [r4] \n\t"
51 "pld [%[pixels]] \n\t"
52 "pld [%[pixels], #32] \n\t"
53 "wldrd wr4, [r4, #8] \n\t"
54 "add r4, r4, %[line_size] \n\t"
55 "walignr1 wr8, wr0, wr1 \n\t"
56 "wldrd wr0, [%[block]] \n\t"
57 "wldrd wr2, [r5] \n\t"
58 "pld [r4] \n\t"
59 "pld [r4, #32] \n\t"
60 "walignr1 wr10, wr3, wr4 \n\t"
61 WAVG2B" wr8, wr8, wr0 \n\t"
62 WAVG2B" wr10, wr10, wr2 \n\t"
63 "wstrd wr8, [%[block]] \n\t"
64 "add %[block], %[block], %[line_size] \n\t"
65 "wstrd wr10, [r5] \n\t"
66 "pld [%[block]] \n\t"
67 "pld [%[block], #32] \n\t"
68 "add r5, r5, %[line_size] \n\t"
69 "pld [r5] \n\t"
70 "pld [r5, #32] \n\t"
71 "bne 1b \n\t"
72 : [block]"+r"(block), [pixels]"+r"(pixels), [line_size]"+r"(stride), [h]"+r"(h)
73 :
74 : "memory", "r4", "r5", "r12");
75 }
76
77 void DEF(put, pixels16)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
78 {
79 int stride = line_size;
80 __asm__ __volatile__ (
81 "and r12, %[pixels], #7 \n\t"
82 "bic %[pixels], %[pixels], #7 \n\t"
83 "tmcr wcgr1, r12 \n\t"
84 "add r4, %[pixels], %[line_size] \n\t"
85 "add r5, %[block], %[line_size] \n\t"
86 "mov %[line_size], %[line_size], lsl #1 \n\t"
87 "1: \n\t"
88 "wldrd wr0, [%[pixels]] \n\t"
89 "wldrd wr1, [%[pixels], #8] \n\t"
90 "subs %[h], %[h], #2 \n\t"
91 "wldrd wr2, [%[pixels], #16] \n\t"
92 "add %[pixels], %[pixels], %[line_size] \n\t"
93 "wldrd wr3, [r4] \n\t"
94 "pld [%[pixels]] \n\t"
95 "pld [%[pixels], #32] \n\t"
96 "walignr1 wr8, wr0, wr1 \n\t"
97 "wldrd wr4, [r4, #8] \n\t"
98 "walignr1 wr9, wr1, wr2 \n\t"
99 "wldrd wr5, [r4, #16] \n\t"
100 "add r4, r4, %[line_size] \n\t"
101 "pld [r4] \n\t"
102 "pld [r4, #32] \n\t"
103 "walignr1 wr10, wr3, wr4 \n\t"
104 "wstrd wr8, [%[block]] \n\t"
105 "walignr1 wr11, wr4, wr5 \n\t"
106 "wstrd wr9, [%[block], #8] \n\t"
107 "add %[block], %[block], %[line_size] \n\t"
108 "wstrd wr10, [r5] \n\t"
109 "wstrd wr11, [r5, #8] \n\t"
110 "add r5, r5, %[line_size] \n\t"
111 "bne 1b \n\t"
112 : [block]"+r"(block), [pixels]"+r"(pixels), [line_size]"+r"(stride), [h]"+r"(h)
113 :
114 : "memory", "r4", "r5", "r12");
115 }
116
117 void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
118 {
119 int stride = line_size;
120 __asm__ __volatile__ (
121 "pld [%[pixels]] \n\t"
122 "pld [%[pixels], #32] \n\t"
123 "pld [%[block]] \n\t"
124 "pld [%[block], #32] \n\t"
125 "and r12, %[pixels], #7 \n\t"
126 "bic %[pixels], %[pixels], #7 \n\t"
127 "tmcr wcgr1, r12 \n\t"
128 "add r4, %[pixels], %[line_size]\n\t"
129 "add r5, %[block], %[line_size] \n\t"
130 "mov %[line_size], %[line_size], lsl #1 \n\t"
131 "1: \n\t"
132 "wldrd wr0, [%[pixels]] \n\t"
133 "wldrd wr1, [%[pixels], #8] \n\t"
134 "subs %[h], %[h], #2 \n\t"
135 "wldrd wr2, [%[pixels], #16] \n\t"
136 "add %[pixels], %[pixels], %[line_size] \n\t"
137 "wldrd wr3, [r4] \n\t"
138 "pld [%[pixels]] \n\t"
139 "pld [%[pixels], #32] \n\t"
140 "walignr1 wr8, wr0, wr1 \n\t"
141 "wldrd wr4, [r4, #8] \n\t"
142 "walignr1 wr9, wr1, wr2 \n\t"
143 "wldrd wr5, [r4, #16] \n\t"
144 "add r4, r4, %[line_size] \n\t"
145 "wldrd wr0, [%[block]] \n\t"
146 "pld [r4] \n\t"
147 "wldrd wr1, [%[block], #8] \n\t"
148 "pld [r4, #32] \n\t"
149 "wldrd wr2, [r5] \n\t"
150 "walignr1 wr10, wr3, wr4 \n\t"
151 "wldrd wr3, [r5, #8] \n\t"
152 WAVG2B" wr8, wr8, wr0 \n\t"
153 WAVG2B" wr9, wr9, wr1 \n\t"
154 WAVG2B" wr10, wr10, wr2 \n\t"
155 "wstrd wr8, [%[block]] \n\t"
156 "walignr1 wr11, wr4, wr5 \n\t"
157 WAVG2B" wr11, wr11, wr3 \n\t"
158 "wstrd wr9, [%[block], #8] \n\t"
159 "add %[block], %[block], %[line_size] \n\t"
160 "wstrd wr10, [r5] \n\t"
161 "pld [%[block]] \n\t"
162 "pld [%[block], #32] \n\t"
163 "wstrd wr11, [r5, #8] \n\t"
164 "add r5, r5, %[line_size] \n\t"
165 "pld [r5] \n\t"
166 "pld [r5, #32] \n\t"
167 "bne 1b \n\t"
168 : [block]"+r"(block), [pixels]"+r"(pixels), [line_size]"+r"(stride), [h]"+r"(h)
169 :
170 : "memory", "r4", "r5", "r12");
171 }
172
173 void DEF(put, pixels8_x2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
174 {
175 int stride = line_size;
176 // [wr0 wr1 wr2 wr3] for previous line
177 // [wr4 wr5 wr6 wr7] for current line
178 SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
179 __asm__ __volatile__(
180 "pld [%[pixels]] \n\t"
181 "pld [%[pixels], #32] \n\t"
182 "and r12, %[pixels], #7 \n\t"
183 "bic %[pixels], %[pixels], #7 \n\t"
184 "tmcr wcgr1, r12 \n\t"
185 "add r12, r12, #1 \n\t"
186 "add r4, %[pixels], %[line_size]\n\t"
187 "tmcr wcgr2, r12 \n\t"
188 "add r5, %[block], %[line_size] \n\t"
189 "mov %[line_size], %[line_size], lsl #1 \n\t"
190
191 "1: \n\t"
192 "wldrd wr10, [%[pixels]] \n\t"
193 "cmp r12, #8 \n\t"
194 "wldrd wr11, [%[pixels], #8] \n\t"
195 "add %[pixels], %[pixels], %[line_size] \n\t"
196 "wldrd wr13, [r4] \n\t"
197 "pld [%[pixels]] \n\t"
198 "wldrd wr14, [r4, #8] \n\t"
199 "pld [%[pixels], #32] \n\t"
200 "add r4, r4, %[line_size] \n\t"
201 "walignr1 wr0, wr10, wr11 \n\t"
202 "pld [r4] \n\t"
203 "pld [r4, #32] \n\t"
204 "walignr1 wr2, wr13, wr14 \n\t"
205 "wmoveq wr4, wr11 \n\t"
206 "wmoveq wr6, wr14 \n\t"
207 "walignr2ne wr4, wr10, wr11 \n\t"
208 "walignr2ne wr6, wr13, wr14 \n\t"
209 WAVG2B" wr0, wr0, wr4 \n\t"
210 WAVG2B" wr2, wr2, wr6 \n\t"
211 "wstrd wr0, [%[block]] \n\t"
212 "subs %[h], %[h], #2 \n\t"
213 "wstrd wr2, [r5] \n\t"
214 "add %[block], %[block], %[line_size] \n\t"
215 "add r5, r5, %[line_size] \n\t"
216 "bne 1b \n\t"
217 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
218 :
219 : "r4", "r5", "r12", "memory");
220 }
221
222 void DEF(put, pixels16_x2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
223 {
224 int stride = line_size;
225 // [wr0 wr1 wr2 wr3] for previous line
226 // [wr4 wr5 wr6 wr7] for current line
227 SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
228 __asm__ __volatile__(
229 "pld [%[pixels]] \n\t"
230 "pld [%[pixels], #32] \n\t"
231 "and r12, %[pixels], #7 \n\t"
232 "bic %[pixels], %[pixels], #7 \n\t"
233 "tmcr wcgr1, r12 \n\t"
234 "add r12, r12, #1 \n\t"
235 "add r4, %[pixels], %[line_size]\n\t"
236 "tmcr wcgr2, r12 \n\t"
237 "add r5, %[block], %[line_size] \n\t"
238 "mov %[line_size], %[line_size], lsl #1 \n\t"
239
240 "1: \n\t"
241 "wldrd wr10, [%[pixels]] \n\t"
242 "cmp r12, #8 \n\t"
243 "wldrd wr11, [%[pixels], #8] \n\t"
244 "wldrd wr12, [%[pixels], #16] \n\t"
245 "add %[pixels], %[pixels], %[line_size] \n\t"
246 "wldrd wr13, [r4] \n\t"
247 "pld [%[pixels]] \n\t"
248 "wldrd wr14, [r4, #8] \n\t"
249 "pld [%[pixels], #32] \n\t"
250 "wldrd wr15, [r4, #16] \n\t"
251 "add r4, r4, %[line_size] \n\t"
252 "walignr1 wr0, wr10, wr11 \n\t"
253 "pld [r4] \n\t"
254 "pld [r4, #32] \n\t"
255 "walignr1 wr1, wr11, wr12 \n\t"
256 "walignr1 wr2, wr13, wr14 \n\t"
257 "walignr1 wr3, wr14, wr15 \n\t"
258 "wmoveq wr4, wr11 \n\t"
259 "wmoveq wr5, wr12 \n\t"
260 "wmoveq wr6, wr14 \n\t"
261 "wmoveq wr7, wr15 \n\t"
262 "walignr2ne wr4, wr10, wr11 \n\t"
263 "walignr2ne wr5, wr11, wr12 \n\t"
264 "walignr2ne wr6, wr13, wr14 \n\t"
265 "walignr2ne wr7, wr14, wr15 \n\t"
266 WAVG2B" wr0, wr0, wr4 \n\t"
267 WAVG2B" wr1, wr1, wr5 \n\t"
268 "wstrd wr0, [%[block]] \n\t"
269 WAVG2B" wr2, wr2, wr6 \n\t"
270 "wstrd wr1, [%[block], #8] \n\t"
271 WAVG2B" wr3, wr3, wr7 \n\t"
272 "add %[block], %[block], %[line_size] \n\t"
273 "wstrd wr2, [r5] \n\t"
274 "subs %[h], %[h], #2 \n\t"
275 "wstrd wr3, [r5, #8] \n\t"
276 "add r5, r5, %[line_size] \n\t"
277 "bne 1b \n\t"
278 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
279 :
280 : "r4", "r5", "r12", "memory");
281 }
282
283 void DEF(avg, pixels8_x2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
284 {
285 int stride = line_size;
286 // [wr0 wr1 wr2 wr3] for previous line
287 // [wr4 wr5 wr6 wr7] for current line
288 SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
289 __asm__ __volatile__(
290 "pld [%[pixels]] \n\t"
291 "pld [%[pixels], #32] \n\t"
292 "pld [%[block]] \n\t"
293 "pld [%[block], #32] \n\t"
294 "and r12, %[pixels], #7 \n\t"
295 "bic %[pixels], %[pixels], #7 \n\t"
296 "tmcr wcgr1, r12 \n\t"
297 "add r12, r12, #1 \n\t"
298 "add r4, %[pixels], %[line_size]\n\t"
299 "tmcr wcgr2, r12 \n\t"
300 "add r5, %[block], %[line_size] \n\t"
301 "mov %[line_size], %[line_size], lsl #1 \n\t"
302 "pld [r5] \n\t"
303 "pld [r5, #32] \n\t"
304
305 "1: \n\t"
306 "wldrd wr10, [%[pixels]] \n\t"
307 "cmp r12, #8 \n\t"
308 "wldrd wr11, [%[pixels], #8] \n\t"
309 "add %[pixels], %[pixels], %[line_size] \n\t"
310 "wldrd wr13, [r4] \n\t"
311 "pld [%[pixels]] \n\t"
312 "wldrd wr14, [r4, #8] \n\t"
313 "pld [%[pixels], #32] \n\t"
314 "add r4, r4, %[line_size] \n\t"
315 "walignr1 wr0, wr10, wr11 \n\t"
316 "pld [r4] \n\t"
317 "pld [r4, #32] \n\t"
318 "walignr1 wr2, wr13, wr14 \n\t"
319 "wmoveq wr4, wr11 \n\t"
320 "wmoveq wr6, wr14 \n\t"
321 "walignr2ne wr4, wr10, wr11 \n\t"
322 "wldrd wr10, [%[block]] \n\t"
323 "walignr2ne wr6, wr13, wr14 \n\t"
324 "wldrd wr12, [r5] \n\t"
325 WAVG2B" wr0, wr0, wr4 \n\t"
326 WAVG2B" wr2, wr2, wr6 \n\t"
327 WAVG2B" wr0, wr0, wr10 \n\t"
328 WAVG2B" wr2, wr2, wr12 \n\t"
329 "wstrd wr0, [%[block]] \n\t"
330 "subs %[h], %[h], #2 \n\t"
331 "wstrd wr2, [r5] \n\t"
332 "add %[block], %[block], %[line_size] \n\t"
333 "add r5, r5, %[line_size] \n\t"
334 "pld [%[block]] \n\t"
335 "pld [%[block], #32] \n\t"
336 "pld [r5] \n\t"
337 "pld [r5, #32] \n\t"
338 "bne 1b \n\t"
339 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
340 :
341 : "r4", "r5", "r12", "memory");
342 }
343
344 void DEF(avg, pixels16_x2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
345 {
346 int stride = line_size;
347 // [wr0 wr1 wr2 wr3] for previous line
348 // [wr4 wr5 wr6 wr7] for current line
349 SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
350 __asm__ __volatile__(
351 "pld [%[pixels]] \n\t"
352 "pld [%[pixels], #32] \n\t"
353 "pld [%[block]] \n\t"
354 "pld [%[block], #32] \n\t"
355 "and r12, %[pixels], #7 \n\t"
356 "bic %[pixels], %[pixels], #7 \n\t"
357 "tmcr wcgr1, r12 \n\t"
358 "add r12, r12, #1 \n\t"
359 "add r4, %[pixels], %[line_size]\n\t"
360 "tmcr wcgr2, r12 \n\t"
361 "add r5, %[block], %[line_size] \n\t"
362 "mov %[line_size], %[line_size], lsl #1 \n\t"
363 "pld [r5] \n\t"
364 "pld [r5, #32] \n\t"
365
366 "1: \n\t"
367 "wldrd wr10, [%[pixels]] \n\t"
368 "cmp r12, #8 \n\t"
369 "wldrd wr11, [%[pixels], #8] \n\t"
370 "wldrd wr12, [%[pixels], #16] \n\t"
371 "add %[pixels], %[pixels], %[line_size] \n\t"
372 "wldrd wr13, [r4] \n\t"
373 "pld [%[pixels]] \n\t"
374 "wldrd wr14, [r4, #8] \n\t"
375 "pld [%[pixels], #32] \n\t"
376 "wldrd wr15, [r4, #16] \n\t"
377 "add r4, r4, %[line_size] \n\t"
378 "walignr1 wr0, wr10, wr11 \n\t"
379 "pld [r4] \n\t"
380 "pld [r4, #32] \n\t"
381 "walignr1 wr1, wr11, wr12 \n\t"
382 "walignr1 wr2, wr13, wr14 \n\t"
383 "walignr1 wr3, wr14, wr15 \n\t"
384 "wmoveq wr4, wr11 \n\t"
385 "wmoveq wr5, wr12 \n\t"
386 "wmoveq wr6, wr14 \n\t"
387 "wmoveq wr7, wr15 \n\t"
388 "walignr2ne wr4, wr10, wr11 \n\t"
389 "walignr2ne wr5, wr11, wr12 \n\t"
390 "walignr2ne wr6, wr13, wr14 \n\t"
391 "walignr2ne wr7, wr14, wr15 \n\t"
392 "wldrd wr10, [%[block]] \n\t"
393 WAVG2B" wr0, wr0, wr4 \n\t"
394 "wldrd wr11, [%[block], #8] \n\t"
395 WAVG2B" wr1, wr1, wr5 \n\t"
396 "wldrd wr12, [r5] \n\t"
397 WAVG2B" wr2, wr2, wr6 \n\t"
398 "wldrd wr13, [r5, #8] \n\t"
399 WAVG2B" wr3, wr3, wr7 \n\t"
400 WAVG2B" wr0, wr0, wr10 \n\t"
401 WAVG2B" wr1, wr1, wr11 \n\t"
402 WAVG2B" wr2, wr2, wr12 \n\t"
403 WAVG2B" wr3, wr3, wr13 \n\t"
404 "wstrd wr0, [%[block]] \n\t"
405 "subs %[h], %[h], #2 \n\t"
406 "wstrd wr1, [%[block], #8] \n\t"
407 "add %[block], %[block], %[line_size] \n\t"
408 "wstrd wr2, [r5] \n\t"
409 "pld [%[block]] \n\t"
410 "wstrd wr3, [r5, #8] \n\t"
411 "add r5, r5, %[line_size] \n\t"
412 "pld [%[block], #32] \n\t"
413 "pld [r5] \n\t"
414 "pld [r5, #32] \n\t"
415 "bne 1b \n\t"
416 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
417 :
418 :"r4", "r5", "r12", "memory");
419 }
420
421 void DEF(avg, pixels8_y2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
422 {
423 int stride = line_size;
424 // [wr0 wr1 wr2 wr3] for previous line
425 // [wr4 wr5 wr6 wr7] for current line
426 __asm__ __volatile__(
427 "pld [%[pixels]] \n\t"
428 "pld [%[pixels], #32] \n\t"
429 "and r12, %[pixels], #7 \n\t"
430 "tmcr wcgr1, r12 \n\t"
431 "bic %[pixels], %[pixels], #7 \n\t"
432
433 "wldrd wr10, [%[pixels]] \n\t"
434 "wldrd wr11, [%[pixels], #8] \n\t"
435 "pld [%[block]] \n\t"
436 "add %[pixels], %[pixels], %[line_size] \n\t"
437 "walignr1 wr0, wr10, wr11 \n\t"
438 "pld [%[pixels]] \n\t"
439 "pld [%[pixels], #32] \n\t"
440
441 "1: \n\t"
442 "wldrd wr10, [%[pixels]] \n\t"
443 "wldrd wr11, [%[pixels], #8] \n\t"
444 "add %[pixels], %[pixels], %[line_size] \n\t"
445 "pld [%[pixels]] \n\t"
446 "pld [%[pixels], #32] \n\t"
447 "walignr1 wr4, wr10, wr11 \n\t"
448 "wldrd wr10, [%[block]] \n\t"
449 WAVG2B" wr8, wr0, wr4 \n\t"
450 WAVG2B" wr8, wr8, wr10 \n\t"
451 "wstrd wr8, [%[block]] \n\t"
452 "add %[block], %[block], %[line_size] \n\t"
453
454 "wldrd wr10, [%[pixels]] \n\t"
455 "wldrd wr11, [%[pixels], #8] \n\t"
456 "pld [%[block]] \n\t"
457 "add %[pixels], %[pixels], %[line_size] \n\t"
458 "pld [%[pixels]] \n\t"
459 "pld [%[pixels], #32] \n\t"
460 "walignr1 wr0, wr10, wr11 \n\t"
461 "wldrd wr10, [%[block]] \n\t"
462 WAVG2B" wr8, wr0, wr4 \n\t"
463 WAVG2B" wr8, wr8, wr10 \n\t"
464 "wstrd wr8, [%[block]] \n\t"
465 "add %[block], %[block], %[line_size] \n\t"
466
467 "subs %[h], %[h], #2 \n\t"
468 "pld [%[block]] \n\t"
469 "bne 1b \n\t"
470 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
471 :
472 : "cc", "memory", "r12");
473 }
474
475 void DEF(put, pixels16_y2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
476 {
477 int stride = line_size;
478 // [wr0 wr1 wr2 wr3] for previous line
479 // [wr4 wr5 wr6 wr7] for current line
480 __asm__ __volatile__(
481 "pld [%[pixels]] \n\t"
482 "pld [%[pixels], #32] \n\t"
483 "and r12, %[pixels], #7 \n\t"
484 "tmcr wcgr1, r12 \n\t"
485 "bic %[pixels], %[pixels], #7 \n\t"
486
487 "wldrd wr10, [%[pixels]] \n\t"
488 "wldrd wr11, [%[pixels], #8] \n\t"
489 "wldrd wr12, [%[pixels], #16] \n\t"
490 "add %[pixels], %[pixels], %[line_size] \n\t"
491 "pld [%[pixels]] \n\t"
492 "pld [%[pixels], #32] \n\t"
493 "walignr1 wr0, wr10, wr11 \n\t"
494 "walignr1 wr1, wr11, wr12 \n\t"
495
496 "1: \n\t"
497 "wldrd wr10, [%[pixels]] \n\t"
498 "wldrd wr11, [%[pixels], #8] \n\t"
499 "wldrd wr12, [%[pixels], #16] \n\t"
500 "add %[pixels], %[pixels], %[line_size] \n\t"
501 "pld [%[pixels]] \n\t"
502 "pld [%[pixels], #32] \n\t"
503 "walignr1 wr4, wr10, wr11 \n\t"
504 "walignr1 wr5, wr11, wr12 \n\t"
505 WAVG2B" wr8, wr0, wr4 \n\t"
506 WAVG2B" wr9, wr1, wr5 \n\t"
507 "wstrd wr8, [%[block]] \n\t"
508 "wstrd wr9, [%[block], #8] \n\t"
509 "add %[block], %[block], %[line_size] \n\t"
510
511 "wldrd wr10, [%[pixels]] \n\t"
512 "wldrd wr11, [%[pixels], #8] \n\t"
513 "wldrd wr12, [%[pixels], #16] \n\t"
514 "add %[pixels], %[pixels], %[line_size] \n\t"
515 "pld [%[pixels]] \n\t"
516 "pld [%[pixels], #32] \n\t"
517 "walignr1 wr0, wr10, wr11 \n\t"
518 "walignr1 wr1, wr11, wr12 \n\t"
519 WAVG2B" wr8, wr0, wr4 \n\t"
520 WAVG2B" wr9, wr1, wr5 \n\t"
521 "wstrd wr8, [%[block]] \n\t"
522 "wstrd wr9, [%[block], #8] \n\t"
523 "add %[block], %[block], %[line_size] \n\t"
524
525 "subs %[h], %[h], #2 \n\t"
526 "bne 1b \n\t"
527 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
528 :
529 : "r4", "r5", "r12", "memory");
530 }
531
532 void DEF(avg, pixels16_y2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
533 {
534 int stride = line_size;
535 // [wr0 wr1 wr2 wr3] for previous line
536 // [wr4 wr5 wr6 wr7] for current line
537 __asm__ __volatile__(
538 "pld [%[pixels]] \n\t"
539 "pld [%[pixels], #32] \n\t"
540 "and r12, %[pixels], #7 \n\t"
541 "tmcr wcgr1, r12 \n\t"
542 "bic %[pixels], %[pixels], #7 \n\t"
543
544 "wldrd wr10, [%[pixels]] \n\t"
545 "wldrd wr11, [%[pixels], #8] \n\t"
546 "pld [%[block]] \n\t"
547 "wldrd wr12, [%[pixels], #16] \n\t"
548 "add %[pixels], %[pixels], %[line_size] \n\t"
549 "pld [%[pixels]] \n\t"
550 "pld [%[pixels], #32] \n\t"
551 "walignr1 wr0, wr10, wr11 \n\t"
552 "walignr1 wr1, wr11, wr12 \n\t"
553
554 "1: \n\t"
555 "wldrd wr10, [%[pixels]] \n\t"
556 "wldrd wr11, [%[pixels], #8] \n\t"
557 "wldrd wr12, [%[pixels], #16] \n\t"
558 "add %[pixels], %[pixels], %[line_size] \n\t"
559 "pld [%[pixels]] \n\t"
560 "pld [%[pixels], #32] \n\t"
561 "walignr1 wr4, wr10, wr11 \n\t"
562 "walignr1 wr5, wr11, wr12 \n\t"
563 "wldrd wr10, [%[block]] \n\t"
564 "wldrd wr11, [%[block], #8] \n\t"
565 WAVG2B" wr8, wr0, wr4 \n\t"
566 WAVG2B" wr9, wr1, wr5 \n\t"
567 WAVG2B" wr8, wr8, wr10 \n\t"
568 WAVG2B" wr9, wr9, wr11 \n\t"
569 "wstrd wr8, [%[block]] \n\t"
570 "wstrd wr9, [%[block], #8] \n\t"
571 "add %[block], %[block], %[line_size] \n\t"
572
573 "wldrd wr10, [%[pixels]] \n\t"
574 "wldrd wr11, [%[pixels], #8] \n\t"
575 "pld [%[block]] \n\t"
576 "wldrd wr12, [%[pixels], #16] \n\t"
577 "add %[pixels], %[pixels], %[line_size] \n\t"
578 "pld [%[pixels]] \n\t"
579 "pld [%[pixels], #32] \n\t"
580 "walignr1 wr0, wr10, wr11 \n\t"
581 "walignr1 wr1, wr11, wr12 \n\t"
582 "wldrd wr10, [%[block]] \n\t"
583 "wldrd wr11, [%[block], #8] \n\t"
584 WAVG2B" wr8, wr0, wr4 \n\t"
585 WAVG2B" wr9, wr1, wr5 \n\t"
586 WAVG2B" wr8, wr8, wr10 \n\t"
587 WAVG2B" wr9, wr9, wr11 \n\t"
588 "wstrd wr8, [%[block]] \n\t"
589 "wstrd wr9, [%[block], #8] \n\t"
590 "add %[block], %[block], %[line_size] \n\t"
591
592 "subs %[h], %[h], #2 \n\t"
593 "pld [%[block]] \n\t"
594 "bne 1b \n\t"
595 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block), [line_size]"+r"(stride)
596 :
597 : "r4", "r5", "r12", "memory");
598 }
599
600 void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
601 {
602 // [wr0 wr1 wr2 wr3] for previous line
603 // [wr4 wr5 wr6 wr7] for current line
604 SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
605 __asm__ __volatile__(
606 "pld [%[pixels]] \n\t"
607 "mov r12, #2 \n\t"
608 "pld [%[pixels], #32] \n\t"
609 "tmcr wcgr0, r12 \n\t" /* for shift value */
610 "and r12, %[pixels], #7 \n\t"
611 "bic %[pixels], %[pixels], #7 \n\t"
612 "tmcr wcgr1, r12 \n\t"
613
614 // [wr0 wr1 wr2 wr3] <= *
615 // [wr4 wr5 wr6 wr7]
616 "wldrd wr12, [%[pixels]] \n\t"
617 "add r12, r12, #1 \n\t"
618 "wldrd wr13, [%[pixels], #8] \n\t"
619 "tmcr wcgr2, r12 \n\t"
620 "add %[pixels], %[pixels], %[line_size] \n\t"
621 "cmp r12, #8 \n\t"
622 "pld [%[pixels]] \n\t"
623 "pld [%[pixels], #32] \n\t"
624 "walignr1 wr2, wr12, wr13 \n\t"
625 "wmoveq wr10, wr13 \n\t"
626 "walignr2ne wr10, wr12, wr13 \n\t"
627 "wunpckelub wr0, wr2 \n\t"
628 "wunpckehub wr1, wr2 \n\t"
629 "wunpckelub wr8, wr10 \n\t"
630 "wunpckehub wr9, wr10 \n\t"
631 "waddhus wr0, wr0, wr8 \n\t"
632 "waddhus wr1, wr1, wr9 \n\t"
633
634 "1: \n\t"
635 // [wr0 wr1 wr2 wr3]
636 // [wr4 wr5 wr6 wr7] <= *
637 "wldrd wr12, [%[pixels]] \n\t"
638 "cmp r12, #8 \n\t"
639 "wldrd wr13, [%[pixels], #8] \n\t"
640 "add %[pixels], %[pixels], %[line_size] \n\t"
641 "walignr1 wr6, wr12, wr13 \n\t"
642 "pld [%[pixels]] \n\t"
643 "pld [%[pixels], #32] \n\t"
644 "wmoveq wr10, wr13 \n\t"
645 "walignr2ne wr10, wr12, wr13 \n\t"
646 "wunpckelub wr4, wr6 \n\t"
647 "wunpckehub wr5, wr6 \n\t"
648 "wunpckelub wr8, wr10 \n\t"
649 "wunpckehub wr9, wr10 \n\t"
650 "waddhus wr4, wr4, wr8 \n\t"
651 "waddhus wr5, wr5, wr9 \n\t"
652 "waddhus wr8, wr0, wr4 \n\t"
653 "waddhus wr9, wr1, wr5 \n\t"
654 "waddhus wr8, wr8, wr15 \n\t"
655 "waddhus wr9, wr9, wr15 \n\t"
656 "wsrlhg wr8, wr8, wcgr0 \n\t"
657 "wsrlhg wr9, wr9, wcgr0 \n\t"
658 "wpackhus wr8, wr8, wr9 \n\t"
659 "wstrd wr8, [%[block]] \n\t"
660 "add %[block], %[block], %[line_size] \n\t"
661
662 // [wr0 wr1 wr2 wr3] <= *
663 // [wr4 wr5 wr6 wr7]
664 "wldrd wr12, [%[pixels]] \n\t"
665 "wldrd wr13, [%[pixels], #8] \n\t"
666 "add %[pixels], %[pixels], %[line_size] \n\t"
667 "walignr1 wr2, wr12, wr13 \n\t"
668 "pld [%[pixels]] \n\t"
669 "pld [%[pixels], #32] \n\t"
670 "wmoveq wr10, wr13 \n\t"
671 "walignr2ne wr10, wr12, wr13 \n\t"
672 "wunpckelub wr0, wr2 \n\t"
673 "wunpckehub wr1, wr2 \n\t"
674 "wunpckelub wr8, wr10 \n\t"
675 "wunpckehub wr9, wr10 \n\t"
676 "waddhus wr0, wr0, wr8 \n\t"
677 "waddhus wr1, wr1, wr9 \n\t"
678 "waddhus wr8, wr0, wr4 \n\t"
679 "waddhus wr9, wr1, wr5 \n\t"
680 "waddhus wr8, wr8, wr15 \n\t"
681 "waddhus wr9, wr9, wr15 \n\t"
682 "wsrlhg wr8, wr8, wcgr0 \n\t"
683 "wsrlhg wr9, wr9, wcgr0 \n\t"
684 "wpackhus wr8, wr8, wr9 \n\t"
685 "subs %[h], %[h], #2 \n\t"
686 "wstrd wr8, [%[block]] \n\t"
687 "add %[block], %[block], %[line_size] \n\t"
688 "bne 1b \n\t"
689 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block)
690 : [line_size]"r"(line_size)
691 : "r12", "memory");
692 }
693
694 void DEF(put, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
695 {
696 // [wr0 wr1 wr2 wr3] for previous line
697 // [wr4 wr5 wr6 wr7] for current line
698 SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
699 __asm__ __volatile__(
700 "pld [%[pixels]] \n\t"
701 "mov r12, #2 \n\t"
702 "pld [%[pixels], #32] \n\t"
703 "tmcr wcgr0, r12 \n\t" /* for shift value */
704 /* alignment */
705 "and r12, %[pixels], #7 \n\t"
706 "bic %[pixels], %[pixels], #7 \n\t"
707 "tmcr wcgr1, r12 \n\t"
708 "add r12, r12, #1 \n\t"
709 "tmcr wcgr2, r12 \n\t"
710
711 // [wr0 wr1 wr2 wr3] <= *
712 // [wr4 wr5 wr6 wr7]
713 "wldrd wr12, [%[pixels]] \n\t"
714 "cmp r12, #8 \n\t"
715 "wldrd wr13, [%[pixels], #8] \n\t"
716 "wldrd wr14, [%[pixels], #16] \n\t"
717 "add %[pixels], %[pixels], %[line_size] \n\t"
718 "pld [%[pixels]] \n\t"
719 "walignr1 wr2, wr12, wr13 \n\t"
720 "pld [%[pixels], #32] \n\t"
721 "walignr1 wr3, wr13, wr14 \n\t"
722 "wmoveq wr10, wr13 \n\t"
723 "wmoveq wr11, wr14 \n\t"
724 "walignr2ne wr10, wr12, wr13 \n\t"
725 "walignr2ne wr11, wr13, wr14 \n\t"
726 "wunpckelub wr0, wr2 \n\t"
727 "wunpckehub wr1, wr2 \n\t"
728 "wunpckelub wr2, wr3 \n\t"
729 "wunpckehub wr3, wr3 \n\t"
730 "wunpckelub wr8, wr10 \n\t"
731 "wunpckehub wr9, wr10 \n\t"
732 "wunpckelub wr10, wr11 \n\t"
733 "wunpckehub wr11, wr11 \n\t"
734 "waddhus wr0, wr0, wr8 \n\t"
735 "waddhus wr1, wr1, wr9 \n\t"
736 "waddhus wr2, wr2, wr10 \n\t"
737 "waddhus wr3, wr3, wr11 \n\t"
738
739 "1: \n\t"
740 // [wr0 wr1 wr2 wr3]
741 // [wr4 wr5 wr6 wr7] <= *
742 "wldrd wr12, [%[pixels]] \n\t"
743 "cmp r12, #8 \n\t"
744 "wldrd wr13, [%[pixels], #8] \n\t"
745 "wldrd wr14, [%[pixels], #16] \n\t"
746 "add %[pixels], %[pixels], %[line_size] \n\t"
747 "walignr1 wr6, wr12, wr13 \n\t"
748 "pld [%[pixels]] \n\t"
749 "pld [%[pixels], #32] \n\t"
750 "walignr1 wr7, wr13, wr14 \n\t"
751 "wmoveq wr10, wr13 \n\t"
752 "wmoveq wr11, wr14 \n\t"
753 "walignr2ne wr10, wr12, wr13 \n\t"
754 "walignr2ne wr11, wr13, wr14 \n\t"
755 "wunpckelub wr4, wr6 \n\t"
756 "wunpckehub wr5, wr6 \n\t"
757 "wunpckelub wr6, wr7 \n\t"
758 "wunpckehub wr7, wr7 \n\t"
759 "wunpckelub wr8, wr10 \n\t"
760 "wunpckehub wr9, wr10 \n\t"
761 "wunpckelub wr10, wr11 \n\t"
762 "wunpckehub wr11, wr11 \n\t"
763 "waddhus wr4, wr4, wr8 \n\t"
764 "waddhus wr5, wr5, wr9 \n\t"
765 "waddhus wr6, wr6, wr10 \n\t"
766 "waddhus wr7, wr7, wr11 \n\t"
767 "waddhus wr8, wr0, wr4 \n\t"
768 "waddhus wr9, wr1, wr5 \n\t"
769 "waddhus wr10, wr2, wr6 \n\t"
770 "waddhus wr11, wr3, wr7 \n\t"
771 "waddhus wr8, wr8, wr15 \n\t"
772 "waddhus wr9, wr9, wr15 \n\t"
773 "waddhus wr10, wr10, wr15 \n\t"
774 "waddhus wr11, wr11, wr15 \n\t"
775 "wsrlhg wr8, wr8, wcgr0 \n\t"
776 "wsrlhg wr9, wr9, wcgr0 \n\t"
777 "wsrlhg wr10, wr10, wcgr0 \n\t"
778 "wsrlhg wr11, wr11, wcgr0 \n\t"
779 "wpackhus wr8, wr8, wr9 \n\t"
780 "wpackhus wr9, wr10, wr11 \n\t"
781 "wstrd wr8, [%[block]] \n\t"
782 "wstrd wr9, [%[block], #8] \n\t"
783 "add %[block], %[block], %[line_size] \n\t"
784
785 // [wr0 wr1 wr2 wr3] <= *
786 // [wr4 wr5 wr6 wr7]
787 "wldrd wr12, [%[pixels]] \n\t"
788 "wldrd wr13, [%[pixels], #8] \n\t"
789 "wldrd wr14, [%[pixels], #16] \n\t"
790 "add %[pixels], %[pixels], %[line_size] \n\t"
791 "walignr1 wr2, wr12, wr13 \n\t"
792 "pld [%[pixels]] \n\t"
793 "pld [%[pixels], #32] \n\t"
794 "walignr1 wr3, wr13, wr14 \n\t"
795 "wmoveq wr10, wr13 \n\t"
796 "wmoveq wr11, wr14 \n\t"
797 "walignr2ne wr10, wr12, wr13 \n\t"
798 "walignr2ne wr11, wr13, wr14 \n\t"
799 "wunpckelub wr0, wr2 \n\t"
800 "wunpckehub wr1, wr2 \n\t"
801 "wunpckelub wr2, wr3 \n\t"
802 "wunpckehub wr3, wr3 \n\t"
803 "wunpckelub wr8, wr10 \n\t"
804 "wunpckehub wr9, wr10 \n\t"
805 "wunpckelub wr10, wr11 \n\t"
806 "wunpckehub wr11, wr11 \n\t"
807 "waddhus wr0, wr0, wr8 \n\t"
808 "waddhus wr1, wr1, wr9 \n\t"
809 "waddhus wr2, wr2, wr10 \n\t"
810 "waddhus wr3, wr3, wr11 \n\t"
811 "waddhus wr8, wr0, wr4 \n\t"
812 "waddhus wr9, wr1, wr5 \n\t"
813 "waddhus wr10, wr2, wr6 \n\t"
814 "waddhus wr11, wr3, wr7 \n\t"
815 "waddhus wr8, wr8, wr15 \n\t"
816 "waddhus wr9, wr9, wr15 \n\t"
817 "waddhus wr10, wr10, wr15 \n\t"
818 "waddhus wr11, wr11, wr15 \n\t"
819 "wsrlhg wr8, wr8, wcgr0 \n\t"
820 "wsrlhg wr9, wr9, wcgr0 \n\t"
821 "wsrlhg wr10, wr10, wcgr0 \n\t"
822 "wsrlhg wr11, wr11, wcgr0 \n\t"
823 "wpackhus wr8, wr8, wr9 \n\t"
824 "wpackhus wr9, wr10, wr11 \n\t"
825 "wstrd wr8, [%[block]] \n\t"
826 "wstrd wr9, [%[block], #8] \n\t"
827 "add %[block], %[block], %[line_size] \n\t"
828
829 "subs %[h], %[h], #2 \n\t"
830 "bne 1b \n\t"
831 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block)
832 : [line_size]"r"(line_size)
833 : "r12", "memory");
834 }
835
836 void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
837 {
838 // [wr0 wr1 wr2 wr3] for previous line
839 // [wr4 wr5 wr6 wr7] for current line
840 SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
841 __asm__ __volatile__(
842 "pld [%[block]] \n\t"
843 "pld [%[block], #32] \n\t"
844 "pld [%[pixels]] \n\t"
845 "mov r12, #2 \n\t"
846 "pld [%[pixels], #32] \n\t"
847 "tmcr wcgr0, r12 \n\t" /* for shift value */
848 "and r12, %[pixels], #7 \n\t"
849 "bic %[pixels], %[pixels], #7 \n\t"
850 "tmcr wcgr1, r12 \n\t"
851
852 // [wr0 wr1 wr2 wr3] <= *
853 // [wr4 wr5 wr6 wr7]
854 "wldrd wr12, [%[pixels]] \n\t"
855 "add r12, r12, #1 \n\t"
856 "wldrd wr13, [%[pixels], #8] \n\t"
857 "tmcr wcgr2, r12 \n\t"
858 "add %[pixels], %[pixels], %[line_size] \n\t"
859 "cmp r12, #8 \n\t"
860 "pld [%[pixels]] \n\t"
861 "pld [%[pixels], #32] \n\t"
862 "walignr1 wr2, wr12, wr13 \n\t"
863 "wmoveq wr10, wr13 \n\t"
864 "walignr2ne wr10, wr12, wr13 \n\t"
865 "wunpckelub wr0, wr2 \n\t"
866 "wunpckehub wr1, wr2 \n\t"
867 "wunpckelub wr8, wr10 \n\t"
868 "wunpckehub wr9, wr10 \n\t"
869 "waddhus wr0, wr0, wr8 \n\t"
870 "waddhus wr1, wr1, wr9 \n\t"
871
872 "1: \n\t"
873 // [wr0 wr1 wr2 wr3]
874 // [wr4 wr5 wr6 wr7] <= *
875 "wldrd wr12, [%[pixels]] \n\t"
876 "cmp r12, #8 \n\t"
877 "wldrd wr13, [%[pixels], #8] \n\t"
878 "add %[pixels], %[pixels], %[line_size] \n\t"
879 "walignr1 wr6, wr12, wr13 \n\t"
880 "pld [%[pixels]] \n\t"
881 "pld [%[pixels], #32] \n\t"
882 "wmoveq wr10, wr13 \n\t"
883 "walignr2ne wr10, wr12, wr13 \n\t"
884 "wunpckelub wr4, wr6 \n\t"
885 "wunpckehub wr5, wr6 \n\t"
886 "wunpckelub wr8, wr10 \n\t"
887 "wunpckehub wr9, wr10 \n\t"
888 "waddhus wr4, wr4, wr8 \n\t"
889 "waddhus wr5, wr5, wr9 \n\t"
890 "waddhus wr8, wr0, wr4 \n\t"
891 "waddhus wr9, wr1, wr5 \n\t"
892 "waddhus wr8, wr8, wr15 \n\t"
893 "waddhus wr9, wr9, wr15 \n\t"
894 "wldrd wr12, [%[block]] \n\t"
895 "wsrlhg wr8, wr8, wcgr0 \n\t"
896 "wsrlhg wr9, wr9, wcgr0 \n\t"
897 "wpackhus wr8, wr8, wr9 \n\t"
898 WAVG2B" wr8, wr8, wr12 \n\t"
899 "wstrd wr8, [%[block]] \n\t"
900 "add %[block], %[block], %[line_size] \n\t"
901 "wldrd wr12, [%[pixels]] \n\t"
902 "pld [%[block]] \n\t"
903 "pld [%[block], #32] \n\t"
904
905 // [wr0 wr1 wr2 wr3] <= *
906 // [wr4 wr5 wr6 wr7]
907 "wldrd wr13, [%[pixels], #8] \n\t"
908 "add %[pixels], %[pixels], %[line_size] \n\t"
909 "walignr1 wr2, wr12, wr13 \n\t"
910 "pld [%[pixels]] \n\t"
911 "pld [%[pixels], #32] \n\t"
912 "wmoveq wr10, wr13 \n\t"
913 "walignr2ne wr10, wr12, wr13 \n\t"
914 "wunpckelub wr0, wr2 \n\t"
915 "wunpckehub wr1, wr2 \n\t"
916 "wunpckelub wr8, wr10 \n\t"
917 "wunpckehub wr9, wr10 \n\t"
918 "waddhus wr0, wr0, wr8 \n\t"
919 "waddhus wr1, wr1, wr9 \n\t"
920 "waddhus wr8, wr0, wr4 \n\t"
921 "waddhus wr9, wr1, wr5 \n\t"
922 "waddhus wr8, wr8, wr15 \n\t"
923 "waddhus wr9, wr9, wr15 \n\t"
924 "wldrd wr12, [%[block]] \n\t"
925 "wsrlhg wr8, wr8, wcgr0 \n\t"
926 "wsrlhg wr9, wr9, wcgr0 \n\t"
927 "wpackhus wr8, wr8, wr9 \n\t"
928 "subs %[h], %[h], #2 \n\t"
929 WAVG2B" wr8, wr8, wr12 \n\t"
930 "wstrd wr8, [%[block]] \n\t"
931 "add %[block], %[block], %[line_size] \n\t"
932 "pld [%[block]] \n\t"
933 "pld [%[block], #32] \n\t"
934 "bne 1b \n\t"
935 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block)
936 : [line_size]"r"(line_size)
937 : "r12", "memory");
938 }
939
940 void DEF(avg, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
941 {
942 // [wr0 wr1 wr2 wr3] for previous line
943 // [wr4 wr5 wr6 wr7] for current line
944 SET_RND(wr15); // =2 for rnd and =1 for no_rnd version
945 __asm__ __volatile__(
946 "pld [%[block]] \n\t"
947 "pld [%[block], #32] \n\t"
948 "pld [%[pixels]] \n\t"
949 "mov r12, #2 \n\t"
950 "pld [%[pixels], #32] \n\t"
951 "tmcr wcgr0, r12 \n\t" /* for shift value */
952 /* alignment */
953 "and r12, %[pixels], #7 \n\t"
954 "bic %[pixels], %[pixels], #7 \n\t"
955 "tmcr wcgr1, r12 \n\t"
956 "add r12, r12, #1 \n\t"
957 "tmcr wcgr2, r12 \n\t"
958
959 // [wr0 wr1 wr2 wr3] <= *
960 // [wr4 wr5 wr6 wr7]
961 "wldrd wr12, [%[pixels]] \n\t"
962 "cmp r12, #8 \n\t"
963 "wldrd wr13, [%[pixels], #8] \n\t"
964 "wldrd wr14, [%[pixels], #16] \n\t"
965 "add %[pixels], %[pixels], %[line_size] \n\t"
966 "pld [%[pixels]] \n\t"
967 "walignr1 wr2, wr12, wr13 \n\t"
968 "pld [%[pixels], #32] \n\t"
969 "walignr1 wr3, wr13, wr14 \n\t"
970 "wmoveq wr10, wr13 \n\t"
971 "wmoveq wr11, wr14 \n\t"
972 "walignr2ne wr10, wr12, wr13 \n\t"
973 "walignr2ne wr11, wr13, wr14 \n\t"
974 "wunpckelub wr0, wr2 \n\t"
975 "wunpckehub wr1, wr2 \n\t"
976 "wunpckelub wr2, wr3 \n\t"
977 "wunpckehub wr3, wr3 \n\t"
978 "wunpckelub wr8, wr10 \n\t"
979 "wunpckehub wr9, wr10 \n\t"
980 "wunpckelub wr10, wr11 \n\t"
981 "wunpckehub wr11, wr11 \n\t"
982 "waddhus wr0, wr0, wr8 \n\t"
983 "waddhus wr1, wr1, wr9 \n\t"
984 "waddhus wr2, wr2, wr10 \n\t"
985 "waddhus wr3, wr3, wr11 \n\t"
986
987 "1: \n\t"
988 // [wr0 wr1 wr2 wr3]
989 // [wr4 wr5 wr6 wr7] <= *
990 "wldrd wr12, [%[pixels]] \n\t"
991 "cmp r12, #8 \n\t"
992 "wldrd wr13, [%[pixels], #8] \n\t"
993 "wldrd wr14, [%[pixels], #16] \n\t"
994 "add %[pixels], %[pixels], %[line_size] \n\t"
995 "walignr1 wr6, wr12, wr13 \n\t"
996 "pld [%[pixels]] \n\t"
997 "pld [%[pixels], #32] \n\t"
998 "walignr1 wr7, wr13, wr14 \n\t"
999 "wmoveq wr10, wr13 \n\t"
1000 "wmoveq wr11, wr14 \n\t"
1001 "walignr2ne wr10, wr12, wr13 \n\t"
1002 "walignr2ne wr11, wr13, wr14 \n\t"
1003 "wunpckelub wr4, wr6 \n\t"
1004 "wunpckehub wr5, wr6 \n\t"
1005 "wunpckelub wr6, wr7 \n\t"
1006 "wunpckehub wr7, wr7 \n\t"
1007 "wunpckelub wr8, wr10 \n\t"
1008 "wunpckehub wr9, wr10 \n\t"
1009 "wunpckelub wr10, wr11 \n\t"
1010 "wunpckehub wr11, wr11 \n\t"
1011 "waddhus wr4, wr4, wr8 \n\t"
1012 "waddhus wr5, wr5, wr9 \n\t"
1013 "waddhus wr6, wr6, wr10 \n\t"
1014 "waddhus wr7, wr7, wr11 \n\t"
1015 "waddhus wr8, wr0, wr4 \n\t"
1016 "waddhus wr9, wr1, wr5 \n\t"
1017 "waddhus wr10, wr2, wr6 \n\t"
1018 "waddhus wr11, wr3, wr7 \n\t"
1019 "waddhus wr8, wr8, wr15 \n\t"
1020 "waddhus wr9, wr9, wr15 \n\t"
1021 "waddhus wr10, wr10, wr15 \n\t"
1022 "waddhus wr11, wr11, wr15 \n\t"
1023 "wsrlhg wr8, wr8, wcgr0 \n\t"
1024 "wsrlhg wr9, wr9, wcgr0 \n\t"
1025 "wldrd wr12, [%[block]] \n\t"
1026 "wldrd wr13, [%[block], #8] \n\t"
1027 "wsrlhg wr10, wr10, wcgr0 \n\t"
1028 "wsrlhg wr11, wr11, wcgr0 \n\t"
1029 "wpackhus wr8, wr8, wr9 \n\t"
1030 "wpackhus wr9, wr10, wr11 \n\t"
1031 WAVG2B" wr8, wr8, wr12 \n\t"
1032 WAVG2B" wr9, wr9, wr13 \n\t"
1033 "wstrd wr8, [%[block]] \n\t"
1034 "wstrd wr9, [%[block], #8] \n\t"
1035 "add %[block], %[block], %[line_size] \n\t"
1036
1037 // [wr0 wr1 wr2 wr3] <= *
1038 // [wr4 wr5 wr6 wr7]
1039 "wldrd wr12, [%[pixels]] \n\t"
1040 "pld [%[block]] \n\t"
1041 "wldrd wr13, [%[pixels], #8] \n\t"
1042 "pld [%[block], #32] \n\t"
1043 "wldrd wr14, [%[pixels], #16] \n\t"
1044 "add %[pixels], %[pixels], %[line_size] \n\t"
1045 "walignr1 wr2, wr12, wr13 \n\t"
1046 "pld [%[pixels]] \n\t"
1047 "pld [%[pixels], #32] \n\t"
1048 "walignr1 wr3, wr13, wr14 \n\t"
1049 "wmoveq wr10, wr13 \n\t"
1050 "wmoveq wr11, wr14 \n\t"
1051 "walignr2ne wr10, wr12, wr13 \n\t"
1052 "walignr2ne wr11, wr13, wr14 \n\t"
1053 "wunpckelub wr0, wr2 \n\t"
1054 "wunpckehub wr1, wr2 \n\t"
1055 "wunpckelub wr2, wr3 \n\t"
1056 "wunpckehub wr3, wr3 \n\t"
1057 "wunpckelub wr8, wr10 \n\t"
1058 "wunpckehub wr9, wr10 \n\t"
1059 "wunpckelub wr10, wr11 \n\t"
1060 "wunpckehub wr11, wr11 \n\t"
1061 "waddhus wr0, wr0, wr8 \n\t"
1062 "waddhus wr1, wr1, wr9 \n\t"
1063 "waddhus wr2, wr2, wr10 \n\t"
1064 "waddhus wr3, wr3, wr11 \n\t"
1065 "waddhus wr8, wr0, wr4 \n\t"
1066 "waddhus wr9, wr1, wr5 \n\t"
1067 "waddhus wr10, wr2, wr6 \n\t"
1068 "waddhus wr11, wr3, wr7 \n\t"
1069 "waddhus wr8, wr8, wr15 \n\t"
1070 "waddhus wr9, wr9, wr15 \n\t"
1071 "waddhus wr10, wr10, wr15 \n\t"
1072 "waddhus wr11, wr11, wr15 \n\t"
1073 "wsrlhg wr8, wr8, wcgr0 \n\t"
1074 "wsrlhg wr9, wr9, wcgr0 \n\t"
1075 "wldrd wr12, [%[block]] \n\t"
1076 "wldrd wr13, [%[block], #8] \n\t"
1077 "wsrlhg wr10, wr10, wcgr0 \n\t"
1078 "wsrlhg wr11, wr11, wcgr0 \n\t"
1079 "wpackhus wr8, wr8, wr9 \n\t"
1080 "wpackhus wr9, wr10, wr11 \n\t"
1081 WAVG2B" wr8, wr8, wr12 \n\t"
1082 WAVG2B" wr9, wr9, wr13 \n\t"
1083 "wstrd wr8, [%[block]] \n\t"
1084 "wstrd wr9, [%[block], #8] \n\t"
1085 "add %[block], %[block], %[line_size] \n\t"
1086 "subs %[h], %[h], #2 \n\t"
1087 "pld [%[block]] \n\t"
1088 "pld [%[block], #32] \n\t"
1089 "bne 1b \n\t"
1090 : [h]"+r"(h), [pixels]"+r"(pixels), [block]"+r"(block)
1091 : [line_size]"r"(line_size)
1092 : "r12", "memory");
1093 }