1 /*
2 * Copyright (c) 2013 The WebM 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 <assert.h>
12 #include <stdio.h>
13
14 #include "./vpx_config.h"
15 #include "./vp9_rtcd.h"
16 #include "vp9/common/vp9_common.h"
17 #include "vpx/vpx_integer.h"
18 #include "vpx_ports/mem.h"
19 #include "vp9/common/vp9_convolve.h"
20 #include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
21
22 #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)23 static void convolve_vert_4_dspr2(const uint8_t *src,
24 int32_t src_stride,
25 uint8_t *dst,
26 int32_t dst_stride,
27 const int16_t *filter_y,
28 int32_t w,
29 int32_t h) {
30 int32_t x, y;
31 const uint8_t *src_ptr;
32 uint8_t *dst_ptr;
33 uint8_t *cm = vp9_ff_cropTbl;
34 uint32_t vector4a = 64;
35 uint32_t load1, load2, load3, load4;
36 uint32_t p1, p2;
37 uint32_t n1, n2;
38 uint32_t scratch1, scratch2;
39 uint32_t store1, store2;
40 int32_t vector1b, vector2b, vector3b, vector4b;
41 int32_t Temp1, Temp2;
42
43 vector1b = ((const int32_t *)filter_y)[0];
44 vector2b = ((const int32_t *)filter_y)[1];
45 vector3b = ((const int32_t *)filter_y)[2];
46 vector4b = ((const int32_t *)filter_y)[3];
47
48 src -= 3 * src_stride;
49
50 for (y = h; y--;) {
51 /* prefetch data to cache memory */
52 vp9_prefetch_store(dst + dst_stride);
53
54 for (x = 0; x < w; x += 4) {
55 src_ptr = src + x;
56 dst_ptr = dst + x;
57
58 __asm__ __volatile__ (
59 "ulw %[load1], 0(%[src_ptr]) \n\t"
60 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
61 "ulw %[load2], 0(%[src_ptr]) \n\t"
62 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
63 "ulw %[load3], 0(%[src_ptr]) \n\t"
64 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
65 "ulw %[load4], 0(%[src_ptr]) \n\t"
66
67 "mtlo %[vector4a], $ac0 \n\t"
68 "mtlo %[vector4a], $ac1 \n\t"
69 "mtlo %[vector4a], $ac2 \n\t"
70 "mtlo %[vector4a], $ac3 \n\t"
71 "mthi $zero, $ac0 \n\t"
72 "mthi $zero, $ac1 \n\t"
73 "mthi $zero, $ac2 \n\t"
74 "mthi $zero, $ac3 \n\t"
75
76 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
77 "preceu.ph.qbr %[p1], %[load2] \n\t"
78 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
79 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
80 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
81 "preceu.ph.qbr %[p2], %[load4] \n\t"
82 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
83 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
84
85 "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
86 "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
87 "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
88 "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
89
90 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
91 "preceu.ph.qbl %[p1], %[load2] \n\t"
92 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
93 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
94 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
95 "preceu.ph.qbl %[p2], %[load4] \n\t"
96 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
97 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
98
99 "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
100 "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
101 "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
102 "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
103
104 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
105 "ulw %[load1], 0(%[src_ptr]) \n\t"
106 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
107 "ulw %[load2], 0(%[src_ptr]) \n\t"
108 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
109 "ulw %[load3], 0(%[src_ptr]) \n\t"
110 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
111 "ulw %[load4], 0(%[src_ptr]) \n\t"
112
113 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
114 "preceu.ph.qbr %[p1], %[load2] \n\t"
115 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
116 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
117 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
118 "preceu.ph.qbr %[p2], %[load4] \n\t"
119 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
120 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
121
122 "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
123 "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
124 "extp %[Temp1], $ac0, 31 \n\t"
125 "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
126 "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
127 "extp %[Temp2], $ac1, 31 \n\t"
128
129 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
130 "preceu.ph.qbl %[p1], %[load2] \n\t"
131 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
132 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
133 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
134 "preceu.ph.qbl %[p2], %[load4] \n\t"
135 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
136 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
137
138 "lbux %[store1], %[Temp1](%[cm]) \n\t"
139 "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
140 "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
141 "extp %[Temp1], $ac2, 31 \n\t"
142
143 "lbux %[store2], %[Temp2](%[cm]) \n\t"
144 "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
145 "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
146 "extp %[Temp2], $ac3, 31 \n\t"
147
148 "sb %[store1], 0(%[dst_ptr]) \n\t"
149 "sb %[store2], 1(%[dst_ptr]) \n\t"
150
151 "lbux %[store1], %[Temp1](%[cm]) \n\t"
152 "lbux %[store2], %[Temp2](%[cm]) \n\t"
153
154 "sb %[store1], 2(%[dst_ptr]) \n\t"
155 "sb %[store2], 3(%[dst_ptr]) \n\t"
156
157 : [load1] "=&r" (load1), [load2] "=&r" (load2),
158 [load3] "=&r" (load3), [load4] "=&r" (load4),
159 [p1] "=&r" (p1), [p2] "=&r" (p2),
160 [n1] "=&r" (n1), [n2] "=&r" (n2),
161 [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
162 [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
163 [store1] "=&r" (store1), [store2] "=&r" (store2),
164 [src_ptr] "+r" (src_ptr)
165 : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
166 [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
167 [vector4a] "r" (vector4a), [src_stride] "r" (src_stride),
168 [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
169 );
170 }
171
172 /* Next row... */
173 src += src_stride;
174 dst += dst_stride;
175 }
176 }
177
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)178 static void convolve_vert_64_dspr2(const uint8_t *src,
179 int32_t src_stride,
180 uint8_t *dst,
181 int32_t dst_stride,
182 const int16_t *filter_y,
183 int32_t h) {
184 int32_t x, y;
185 const uint8_t *src_ptr;
186 uint8_t *dst_ptr;
187 uint8_t *cm = vp9_ff_cropTbl;
188 uint32_t vector4a = 64;
189 uint32_t load1, load2, load3, load4;
190 uint32_t p1, p2;
191 uint32_t n1, n2;
192 uint32_t scratch1, scratch2;
193 uint32_t store1, store2;
194 int32_t vector1b, vector2b, vector3b, vector4b;
195 int32_t Temp1, Temp2;
196
197 vector1b = ((const int32_t *)filter_y)[0];
198 vector2b = ((const int32_t *)filter_y)[1];
199 vector3b = ((const int32_t *)filter_y)[2];
200 vector4b = ((const int32_t *)filter_y)[3];
201
202 src -= 3 * src_stride;
203
204 for (y = h; y--;) {
205 /* prefetch data to cache memory */
206 vp9_prefetch_store(dst + dst_stride);
207 vp9_prefetch_store(dst + dst_stride + 32);
208
209 for (x = 0; x < 64; x += 4) {
210 src_ptr = src + x;
211 dst_ptr = dst + x;
212
213 __asm__ __volatile__ (
214 "ulw %[load1], 0(%[src_ptr]) \n\t"
215 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
216 "ulw %[load2], 0(%[src_ptr]) \n\t"
217 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
218 "ulw %[load3], 0(%[src_ptr]) \n\t"
219 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
220 "ulw %[load4], 0(%[src_ptr]) \n\t"
221
222 "mtlo %[vector4a], $ac0 \n\t"
223 "mtlo %[vector4a], $ac1 \n\t"
224 "mtlo %[vector4a], $ac2 \n\t"
225 "mtlo %[vector4a], $ac3 \n\t"
226 "mthi $zero, $ac0 \n\t"
227 "mthi $zero, $ac1 \n\t"
228 "mthi $zero, $ac2 \n\t"
229 "mthi $zero, $ac3 \n\t"
230
231 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
232 "preceu.ph.qbr %[p1], %[load2] \n\t"
233 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
234 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
235 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
236 "preceu.ph.qbr %[p2], %[load4] \n\t"
237 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
238 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
239
240 "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
241 "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
242 "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
243 "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
244
245 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
246 "preceu.ph.qbl %[p1], %[load2] \n\t"
247 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
248 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
249 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
250 "preceu.ph.qbl %[p2], %[load4] \n\t"
251 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
252 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
253
254 "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
255 "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
256 "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
257 "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
258
259 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
260 "ulw %[load1], 0(%[src_ptr]) \n\t"
261 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
262 "ulw %[load2], 0(%[src_ptr]) \n\t"
263 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
264 "ulw %[load3], 0(%[src_ptr]) \n\t"
265 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
266 "ulw %[load4], 0(%[src_ptr]) \n\t"
267
268 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
269 "preceu.ph.qbr %[p1], %[load2] \n\t"
270 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
271 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
272 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
273 "preceu.ph.qbr %[p2], %[load4] \n\t"
274 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
275 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
276
277 "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
278 "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
279 "extp %[Temp1], $ac0, 31 \n\t"
280 "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
281 "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
282 "extp %[Temp2], $ac1, 31 \n\t"
283
284 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
285 "preceu.ph.qbl %[p1], %[load2] \n\t"
286 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
287 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
288 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
289 "preceu.ph.qbl %[p2], %[load4] \n\t"
290 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
291 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
292
293 "lbux %[store1], %[Temp1](%[cm]) \n\t"
294 "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
295 "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
296 "extp %[Temp1], $ac2, 31 \n\t"
297
298 "lbux %[store2], %[Temp2](%[cm]) \n\t"
299 "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
300 "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
301 "extp %[Temp2], $ac3, 31 \n\t"
302
303 "sb %[store1], 0(%[dst_ptr]) \n\t"
304 "sb %[store2], 1(%[dst_ptr]) \n\t"
305
306 "lbux %[store1], %[Temp1](%[cm]) \n\t"
307 "lbux %[store2], %[Temp2](%[cm]) \n\t"
308
309 "sb %[store1], 2(%[dst_ptr]) \n\t"
310 "sb %[store2], 3(%[dst_ptr]) \n\t"
311
312 : [load1] "=&r" (load1), [load2] "=&r" (load2),
313 [load3] "=&r" (load3), [load4] "=&r" (load4),
314 [p1] "=&r" (p1), [p2] "=&r" (p2),
315 [n1] "=&r" (n1), [n2] "=&r" (n2),
316 [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
317 [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
318 [store1] "=&r" (store1), [store2] "=&r" (store2),
319 [src_ptr] "+r" (src_ptr)
320 : [vector1b] "r" (vector1b), [vector2b] "r" (vector2b),
321 [vector3b] "r" (vector3b), [vector4b] "r" (vector4b),
322 [vector4a] "r" (vector4a), [src_stride] "r" (src_stride),
323 [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
324 );
325 }
326
327 /* Next row... */
328 src += src_stride;
329 dst += dst_stride;
330 }
331 }
332
vp9_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)333 void vp9_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
334 uint8_t *dst, ptrdiff_t dst_stride,
335 const int16_t *filter_x, int x_step_q4,
336 const int16_t *filter_y, int y_step_q4,
337 int w, int h) {
338 if (((const int32_t *)filter_y)[1] == 0x800000) {
339 vp9_convolve_copy(src, src_stride,
340 dst, dst_stride,
341 filter_x, x_step_q4,
342 filter_y, y_step_q4,
343 w, h);
344 } else if (((const int32_t *)filter_y)[0] == 0) {
345 vp9_convolve2_vert_dspr2(src, src_stride,
346 dst, dst_stride,
347 filter_x, x_step_q4,
348 filter_y, y_step_q4,
349 w, h);
350 } else {
351 if (16 == y_step_q4) {
352 uint32_t pos = 38;
353
354 /* bit positon for extract from acc */
355 __asm__ __volatile__ (
356 "wrdsp %[pos], 1 \n\t"
357 :
358 : [pos] "r" (pos)
359 );
360
361 vp9_prefetch_store(dst);
362
363 switch (w) {
364 case 4 :
365 case 8 :
366 case 16 :
367 case 32 :
368 convolve_vert_4_dspr2(src, src_stride,
369 dst, dst_stride,
370 filter_y, w, h);
371 break;
372 case 64 :
373 vp9_prefetch_store(dst + 32);
374 convolve_vert_64_dspr2(src, src_stride,
375 dst, dst_stride,
376 filter_y, h);
377 break;
378 default:
379 vp9_convolve8_vert_c(src, src_stride,
380 dst, dst_stride,
381 filter_x, x_step_q4,
382 filter_y, y_step_q4,
383 w, h);
384 break;
385 }
386 } else {
387 vp9_convolve8_vert_c(src, src_stride,
388 dst, dst_stride,
389 filter_x, x_step_q4,
390 filter_y, y_step_q4,
391 w, h);
392 }
393 }
394 }
395
396 #endif
397