1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #include <assert.h>
13 #include <stdio.h>
14
15 #include "./aom_dsp_rtcd.h"
16 #include "aom_dsp/mips/convolve_common_dspr2.h"
17 #include "aom_dsp/aom_dsp_common.h"
18 #include "aom_dsp/aom_filter.h"
19 #include "aom_ports/mem.h"
20
21 #if HAVE_DSPR2
convolve_vert_4_dspr2(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int16_t * filter_y,int32_t w,int32_t h)22 static void convolve_vert_4_dspr2(const uint8_t *src, int32_t src_stride,
23 uint8_t *dst, int32_t dst_stride,
24 const int16_t *filter_y, int32_t w,
25 int32_t h) {
26 int32_t x, y;
27 const uint8_t *src_ptr;
28 uint8_t *dst_ptr;
29 uint8_t *cm = aom_ff_cropTbl;
30 uint32_t vector4a = 64;
31 uint32_t load1, load2, load3, load4;
32 uint32_t p1, p2;
33 uint32_t n1, n2;
34 uint32_t scratch1, scratch2;
35 uint32_t store1, store2;
36 int32_t vector1b, vector2b, vector3b, vector4b;
37 int32_t Temp1, Temp2;
38
39 vector1b = ((const int32_t *)filter_y)[0];
40 vector2b = ((const int32_t *)filter_y)[1];
41 vector3b = ((const int32_t *)filter_y)[2];
42 vector4b = ((const int32_t *)filter_y)[3];
43
44 src -= 3 * src_stride;
45
46 for (y = h; y--;) {
47 /* prefetch data to cache memory */
48 prefetch_store(dst + dst_stride);
49
50 for (x = 0; x < w; x += 4) {
51 src_ptr = src + x;
52 dst_ptr = dst + x;
53
54 __asm__ __volatile__(
55 "ulw %[load1], 0(%[src_ptr]) \n\t"
56 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
57 "ulw %[load2], 0(%[src_ptr]) \n\t"
58 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
59 "ulw %[load3], 0(%[src_ptr]) \n\t"
60 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
61 "ulw %[load4], 0(%[src_ptr]) \n\t"
62
63 "mtlo %[vector4a], $ac0 \n\t"
64 "mtlo %[vector4a], $ac1 \n\t"
65 "mtlo %[vector4a], $ac2 \n\t"
66 "mtlo %[vector4a], $ac3 \n\t"
67 "mthi $zero, $ac0 \n\t"
68 "mthi $zero, $ac1 \n\t"
69 "mthi $zero, $ac2 \n\t"
70 "mthi $zero, $ac3 \n\t"
71
72 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
73 "preceu.ph.qbr %[p1], %[load2] \n\t"
74 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
75 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
76 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
77 "preceu.ph.qbr %[p2], %[load4] \n\t"
78 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
79 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
80
81 "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
82 "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
83 "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
84 "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
85
86 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
87 "preceu.ph.qbl %[p1], %[load2] \n\t"
88 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
89 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
90 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
91 "preceu.ph.qbl %[p2], %[load4] \n\t"
92 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
93 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
94
95 "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
96 "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
97 "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
98 "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
99
100 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
101 "ulw %[load1], 0(%[src_ptr]) \n\t"
102 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
103 "ulw %[load2], 0(%[src_ptr]) \n\t"
104 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
105 "ulw %[load3], 0(%[src_ptr]) \n\t"
106 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
107 "ulw %[load4], 0(%[src_ptr]) \n\t"
108
109 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
110 "preceu.ph.qbr %[p1], %[load2] \n\t"
111 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
112 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
113 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
114 "preceu.ph.qbr %[p2], %[load4] \n\t"
115 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
116 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
117
118 "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
119 "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
120 "extp %[Temp1], $ac0, 31 \n\t"
121 "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
122 "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
123 "extp %[Temp2], $ac1, 31 \n\t"
124
125 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
126 "preceu.ph.qbl %[p1], %[load2] \n\t"
127 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
128 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
129 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
130 "preceu.ph.qbl %[p2], %[load4] \n\t"
131 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
132 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
133
134 "lbux %[store1], %[Temp1](%[cm]) \n\t"
135 "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
136 "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
137 "extp %[Temp1], $ac2, 31 \n\t"
138
139 "lbux %[store2], %[Temp2](%[cm]) \n\t"
140 "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
141 "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
142 "extp %[Temp2], $ac3, 31 \n\t"
143
144 "sb %[store1], 0(%[dst_ptr]) \n\t"
145 "sb %[store2], 1(%[dst_ptr]) \n\t"
146
147 "lbux %[store1], %[Temp1](%[cm]) \n\t"
148 "lbux %[store2], %[Temp2](%[cm]) \n\t"
149
150 "sb %[store1], 2(%[dst_ptr]) \n\t"
151 "sb %[store2], 3(%[dst_ptr]) \n\t"
152
153 : [load1] "=&r"(load1), [load2] "=&r"(load2), [load3] "=&r"(load3),
154 [load4] "=&r"(load4), [p1] "=&r"(p1), [p2] "=&r"(p2),
155 [n1] "=&r"(n1), [n2] "=&r"(n2), [scratch1] "=&r"(scratch1),
156 [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
157 [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
158 [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
159 : [vector1b] "r"(vector1b), [vector2b] "r"(vector2b),
160 [vector3b] "r"(vector3b), [vector4b] "r"(vector4b),
161 [vector4a] "r"(vector4a), [src_stride] "r"(src_stride),
162 [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
163 }
164
165 /* Next row... */
166 src += src_stride;
167 dst += dst_stride;
168 }
169 }
170
convolve_vert_64_dspr2(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int16_t * filter_y,int32_t h)171 static void convolve_vert_64_dspr2(const uint8_t *src, int32_t src_stride,
172 uint8_t *dst, int32_t dst_stride,
173 const int16_t *filter_y, int32_t h) {
174 int32_t x, y;
175 const uint8_t *src_ptr;
176 uint8_t *dst_ptr;
177 uint8_t *cm = aom_ff_cropTbl;
178 uint32_t vector4a = 64;
179 uint32_t load1, load2, load3, load4;
180 uint32_t p1, p2;
181 uint32_t n1, n2;
182 uint32_t scratch1, scratch2;
183 uint32_t store1, store2;
184 int32_t vector1b, vector2b, vector3b, vector4b;
185 int32_t Temp1, Temp2;
186
187 vector1b = ((const int32_t *)filter_y)[0];
188 vector2b = ((const int32_t *)filter_y)[1];
189 vector3b = ((const int32_t *)filter_y)[2];
190 vector4b = ((const int32_t *)filter_y)[3];
191
192 src -= 3 * src_stride;
193
194 for (y = h; y--;) {
195 /* prefetch data to cache memory */
196 prefetch_store(dst + dst_stride);
197 prefetch_store(dst + dst_stride + 32);
198
199 for (x = 0; x < 64; x += 4) {
200 src_ptr = src + x;
201 dst_ptr = dst + x;
202
203 __asm__ __volatile__(
204 "ulw %[load1], 0(%[src_ptr]) \n\t"
205 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
206 "ulw %[load2], 0(%[src_ptr]) \n\t"
207 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
208 "ulw %[load3], 0(%[src_ptr]) \n\t"
209 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
210 "ulw %[load4], 0(%[src_ptr]) \n\t"
211
212 "mtlo %[vector4a], $ac0 \n\t"
213 "mtlo %[vector4a], $ac1 \n\t"
214 "mtlo %[vector4a], $ac2 \n\t"
215 "mtlo %[vector4a], $ac3 \n\t"
216 "mthi $zero, $ac0 \n\t"
217 "mthi $zero, $ac1 \n\t"
218 "mthi $zero, $ac2 \n\t"
219 "mthi $zero, $ac3 \n\t"
220
221 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
222 "preceu.ph.qbr %[p1], %[load2] \n\t"
223 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
224 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
225 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
226 "preceu.ph.qbr %[p2], %[load4] \n\t"
227 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
228 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
229
230 "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
231 "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
232 "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
233 "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
234
235 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
236 "preceu.ph.qbl %[p1], %[load2] \n\t"
237 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
238 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
239 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
240 "preceu.ph.qbl %[p2], %[load4] \n\t"
241 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
242 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
243
244 "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
245 "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
246 "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
247 "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
248
249 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
250 "ulw %[load1], 0(%[src_ptr]) \n\t"
251 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
252 "ulw %[load2], 0(%[src_ptr]) \n\t"
253 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
254 "ulw %[load3], 0(%[src_ptr]) \n\t"
255 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
256 "ulw %[load4], 0(%[src_ptr]) \n\t"
257
258 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
259 "preceu.ph.qbr %[p1], %[load2] \n\t"
260 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
261 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
262 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
263 "preceu.ph.qbr %[p2], %[load4] \n\t"
264 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
265 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
266
267 "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
268 "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
269 "extp %[Temp1], $ac0, 31 \n\t"
270 "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
271 "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
272 "extp %[Temp2], $ac1, 31 \n\t"
273
274 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
275 "preceu.ph.qbl %[p1], %[load2] \n\t"
276 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
277 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
278 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
279 "preceu.ph.qbl %[p2], %[load4] \n\t"
280 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
281 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
282
283 "lbux %[store1], %[Temp1](%[cm]) \n\t"
284 "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
285 "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
286 "extp %[Temp1], $ac2, 31 \n\t"
287
288 "lbux %[store2], %[Temp2](%[cm]) \n\t"
289 "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
290 "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
291 "extp %[Temp2], $ac3, 31 \n\t"
292
293 "sb %[store1], 0(%[dst_ptr]) \n\t"
294 "sb %[store2], 1(%[dst_ptr]) \n\t"
295
296 "lbux %[store1], %[Temp1](%[cm]) \n\t"
297 "lbux %[store2], %[Temp2](%[cm]) \n\t"
298
299 "sb %[store1], 2(%[dst_ptr]) \n\t"
300 "sb %[store2], 3(%[dst_ptr]) \n\t"
301
302 : [load1] "=&r"(load1), [load2] "=&r"(load2), [load3] "=&r"(load3),
303 [load4] "=&r"(load4), [p1] "=&r"(p1), [p2] "=&r"(p2),
304 [n1] "=&r"(n1), [n2] "=&r"(n2), [scratch1] "=&r"(scratch1),
305 [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
306 [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
307 [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
308 : [vector1b] "r"(vector1b), [vector2b] "r"(vector2b),
309 [vector3b] "r"(vector3b), [vector4b] "r"(vector4b),
310 [vector4a] "r"(vector4a), [src_stride] "r"(src_stride),
311 [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
312 }
313
314 /* Next row... */
315 src += src_stride;
316 dst += dst_stride;
317 }
318 }
319
aom_convolve8_vert_dspr2(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const int16_t * filter_x,int x_step_q4,const int16_t * filter_y,int y_step_q4,int w,int h)320 void aom_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
321 uint8_t *dst, ptrdiff_t dst_stride,
322 const int16_t *filter_x, int x_step_q4,
323 const int16_t *filter_y, int y_step_q4, int w,
324 int h) {
325 assert(y_step_q4 == 16);
326 assert(((const int32_t *)filter_y)[1] != 0x800000);
327
328 if (((const int32_t *)filter_y)[0] == 0) {
329 aom_convolve2_vert_dspr2(src, src_stride, dst, dst_stride, filter_x,
330 x_step_q4, filter_y, y_step_q4, w, h);
331 } else {
332 uint32_t pos = 38;
333
334 /* bit positon for extract from acc */
335 __asm__ __volatile__("wrdsp %[pos], 1 \n\t"
336 :
337 : [pos] "r"(pos));
338
339 prefetch_store(dst);
340
341 switch (w) {
342 case 4:
343 case 8:
344 case 16:
345 case 32:
346 convolve_vert_4_dspr2(src, src_stride, dst, dst_stride, filter_y, w, h);
347 break;
348 case 64:
349 prefetch_store(dst + 32);
350 convolve_vert_64_dspr2(src, src_stride, dst, dst_stride, filter_y, h);
351 break;
352 default:
353 aom_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x,
354 x_step_q4, filter_y, y_step_q4, w, h);
355 break;
356 }
357 }
358 }
359
360 #endif
361