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_dsp_rtcd.h"
15 #include "vpx_dsp/mips/convolve_common_dspr2.h"
16 #include "vpx_dsp/vpx_dsp_common.h"
17 #include "vpx_dsp/vpx_filter.h"
18 #include "vpx_ports/mem.h"
19
20 #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)21 static void convolve_vert_4_dspr2(const uint8_t *src, int32_t src_stride,
22 uint8_t *dst, int32_t dst_stride,
23 const int16_t *filter_y, int32_t w,
24 int32_t h) {
25 int32_t x, y;
26 const uint8_t *src_ptr;
27 uint8_t *dst_ptr;
28 uint8_t *cm = vpx_ff_cropTbl;
29 uint32_t vector4a = 64;
30 uint32_t load1, load2, load3, load4;
31 uint32_t p1, p2;
32 uint32_t n1, n2;
33 uint32_t scratch1, scratch2;
34 uint32_t store1, store2;
35 int32_t vector1b, vector2b, vector3b, vector4b;
36 int32_t Temp1, Temp2;
37
38 vector1b = ((const int32_t *)filter_y)[0];
39 vector2b = ((const int32_t *)filter_y)[1];
40 vector3b = ((const int32_t *)filter_y)[2];
41 vector4b = ((const int32_t *)filter_y)[3];
42
43 src -= 3 * src_stride;
44
45 for (y = h; y--;) {
46 /* prefetch data to cache memory */
47 prefetch_store(dst + dst_stride);
48
49 for (x = 0; x < w; x += 4) {
50 src_ptr = src + x;
51 dst_ptr = dst + x;
52
53 __asm__ __volatile__(
54 "ulw %[load1], 0(%[src_ptr]) \n\t"
55 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
56 "ulw %[load2], 0(%[src_ptr]) \n\t"
57 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
58 "ulw %[load3], 0(%[src_ptr]) \n\t"
59 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
60 "ulw %[load4], 0(%[src_ptr]) \n\t"
61
62 "mtlo %[vector4a], $ac0 \n\t"
63 "mtlo %[vector4a], $ac1 \n\t"
64 "mtlo %[vector4a], $ac2 \n\t"
65 "mtlo %[vector4a], $ac3 \n\t"
66 "mthi $zero, $ac0 \n\t"
67 "mthi $zero, $ac1 \n\t"
68 "mthi $zero, $ac2 \n\t"
69 "mthi $zero, $ac3 \n\t"
70
71 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
72 "preceu.ph.qbr %[p1], %[load2] \n\t"
73 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
74 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
75 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
76 "preceu.ph.qbr %[p2], %[load4] \n\t"
77 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
78 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
79
80 "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
81 "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
82 "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
83 "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
84
85 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
86 "preceu.ph.qbl %[p1], %[load2] \n\t"
87 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
88 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
89 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
90 "preceu.ph.qbl %[p2], %[load4] \n\t"
91 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
92 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
93
94 "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
95 "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
96 "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
97 "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
98
99 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
100 "ulw %[load1], 0(%[src_ptr]) \n\t"
101 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
102 "ulw %[load2], 0(%[src_ptr]) \n\t"
103 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
104 "ulw %[load3], 0(%[src_ptr]) \n\t"
105 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
106 "ulw %[load4], 0(%[src_ptr]) \n\t"
107
108 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
109 "preceu.ph.qbr %[p1], %[load2] \n\t"
110 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
111 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
112 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
113 "preceu.ph.qbr %[p2], %[load4] \n\t"
114 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
115 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
116
117 "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
118 "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
119 "extp %[Temp1], $ac0, 31 \n\t"
120 "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
121 "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
122 "extp %[Temp2], $ac1, 31 \n\t"
123
124 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
125 "preceu.ph.qbl %[p1], %[load2] \n\t"
126 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
127 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
128 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
129 "preceu.ph.qbl %[p2], %[load4] \n\t"
130 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
131 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
132
133 "lbux %[store1], %[Temp1](%[cm]) \n\t"
134 "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
135 "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
136 "extp %[Temp1], $ac2, 31 \n\t"
137
138 "lbux %[store2], %[Temp2](%[cm]) \n\t"
139 "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
140 "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
141 "extp %[Temp2], $ac3, 31 \n\t"
142
143 "sb %[store1], 0(%[dst_ptr]) \n\t"
144 "sb %[store2], 1(%[dst_ptr]) \n\t"
145
146 "lbux %[store1], %[Temp1](%[cm]) \n\t"
147 "lbux %[store2], %[Temp2](%[cm]) \n\t"
148
149 "sb %[store1], 2(%[dst_ptr]) \n\t"
150 "sb %[store2], 3(%[dst_ptr]) \n\t"
151
152 : [load1] "=&r"(load1), [load2] "=&r"(load2), [load3] "=&r"(load3),
153 [load4] "=&r"(load4), [p1] "=&r"(p1), [p2] "=&r"(p2),
154 [n1] "=&r"(n1), [n2] "=&r"(n2), [scratch1] "=&r"(scratch1),
155 [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
156 [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
157 [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
158 : [vector1b] "r"(vector1b), [vector2b] "r"(vector2b),
159 [vector3b] "r"(vector3b), [vector4b] "r"(vector4b),
160 [vector4a] "r"(vector4a), [src_stride] "r"(src_stride),
161 [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
162 }
163
164 /* Next row... */
165 src += src_stride;
166 dst += dst_stride;
167 }
168 }
169
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)170 static void convolve_vert_64_dspr2(const uint8_t *src, int32_t src_stride,
171 uint8_t *dst, int32_t dst_stride,
172 const int16_t *filter_y, int32_t h) {
173 int32_t x, y;
174 const uint8_t *src_ptr;
175 uint8_t *dst_ptr;
176 uint8_t *cm = vpx_ff_cropTbl;
177 uint32_t vector4a = 64;
178 uint32_t load1, load2, load3, load4;
179 uint32_t p1, p2;
180 uint32_t n1, n2;
181 uint32_t scratch1, scratch2;
182 uint32_t store1, store2;
183 int32_t vector1b, vector2b, vector3b, vector4b;
184 int32_t Temp1, Temp2;
185
186 vector1b = ((const int32_t *)filter_y)[0];
187 vector2b = ((const int32_t *)filter_y)[1];
188 vector3b = ((const int32_t *)filter_y)[2];
189 vector4b = ((const int32_t *)filter_y)[3];
190
191 src -= 3 * src_stride;
192
193 for (y = h; y--;) {
194 /* prefetch data to cache memory */
195 prefetch_store(dst + dst_stride);
196 prefetch_store(dst + dst_stride + 32);
197
198 for (x = 0; x < 64; x += 4) {
199 src_ptr = src + x;
200 dst_ptr = dst + x;
201
202 __asm__ __volatile__(
203 "ulw %[load1], 0(%[src_ptr]) \n\t"
204 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
205 "ulw %[load2], 0(%[src_ptr]) \n\t"
206 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
207 "ulw %[load3], 0(%[src_ptr]) \n\t"
208 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
209 "ulw %[load4], 0(%[src_ptr]) \n\t"
210
211 "mtlo %[vector4a], $ac0 \n\t"
212 "mtlo %[vector4a], $ac1 \n\t"
213 "mtlo %[vector4a], $ac2 \n\t"
214 "mtlo %[vector4a], $ac3 \n\t"
215 "mthi $zero, $ac0 \n\t"
216 "mthi $zero, $ac1 \n\t"
217 "mthi $zero, $ac2 \n\t"
218 "mthi $zero, $ac3 \n\t"
219
220 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
221 "preceu.ph.qbr %[p1], %[load2] \n\t"
222 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
223 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
224 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
225 "preceu.ph.qbr %[p2], %[load4] \n\t"
226 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
227 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
228
229 "dpa.w.ph $ac0, %[p1], %[vector1b] \n\t"
230 "dpa.w.ph $ac0, %[p2], %[vector2b] \n\t"
231 "dpa.w.ph $ac1, %[n1], %[vector1b] \n\t"
232 "dpa.w.ph $ac1, %[n2], %[vector2b] \n\t"
233
234 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
235 "preceu.ph.qbl %[p1], %[load2] \n\t"
236 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
237 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
238 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
239 "preceu.ph.qbl %[p2], %[load4] \n\t"
240 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
241 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
242
243 "dpa.w.ph $ac2, %[p1], %[vector1b] \n\t"
244 "dpa.w.ph $ac2, %[p2], %[vector2b] \n\t"
245 "dpa.w.ph $ac3, %[n1], %[vector1b] \n\t"
246 "dpa.w.ph $ac3, %[n2], %[vector2b] \n\t"
247
248 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
249 "ulw %[load1], 0(%[src_ptr]) \n\t"
250 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
251 "ulw %[load2], 0(%[src_ptr]) \n\t"
252 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
253 "ulw %[load3], 0(%[src_ptr]) \n\t"
254 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
255 "ulw %[load4], 0(%[src_ptr]) \n\t"
256
257 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
258 "preceu.ph.qbr %[p1], %[load2] \n\t"
259 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
260 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
261 "preceu.ph.qbr %[scratch2], %[load3] \n\t"
262 "preceu.ph.qbr %[p2], %[load4] \n\t"
263 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
264 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
265
266 "dpa.w.ph $ac0, %[p1], %[vector3b] \n\t"
267 "dpa.w.ph $ac0, %[p2], %[vector4b] \n\t"
268 "extp %[Temp1], $ac0, 31 \n\t"
269 "dpa.w.ph $ac1, %[n1], %[vector3b] \n\t"
270 "dpa.w.ph $ac1, %[n2], %[vector4b] \n\t"
271 "extp %[Temp2], $ac1, 31 \n\t"
272
273 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
274 "preceu.ph.qbl %[p1], %[load2] \n\t"
275 "precrq.ph.w %[n1], %[p1], %[scratch1] \n\t" /* pixel 2 */
276 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
277 "preceu.ph.qbl %[scratch2], %[load3] \n\t"
278 "preceu.ph.qbl %[p2], %[load4] \n\t"
279 "precrq.ph.w %[n2], %[p2], %[scratch2] \n\t" /* pixel 2 */
280 "append %[p2], %[scratch2], 16 \n\t" /* pixel 1 */
281
282 "lbux %[store1], %[Temp1](%[cm]) \n\t"
283 "dpa.w.ph $ac2, %[p1], %[vector3b] \n\t"
284 "dpa.w.ph $ac2, %[p2], %[vector4b] \n\t"
285 "extp %[Temp1], $ac2, 31 \n\t"
286
287 "lbux %[store2], %[Temp2](%[cm]) \n\t"
288 "dpa.w.ph $ac3, %[n1], %[vector3b] \n\t"
289 "dpa.w.ph $ac3, %[n2], %[vector4b] \n\t"
290 "extp %[Temp2], $ac3, 31 \n\t"
291
292 "sb %[store1], 0(%[dst_ptr]) \n\t"
293 "sb %[store2], 1(%[dst_ptr]) \n\t"
294
295 "lbux %[store1], %[Temp1](%[cm]) \n\t"
296 "lbux %[store2], %[Temp2](%[cm]) \n\t"
297
298 "sb %[store1], 2(%[dst_ptr]) \n\t"
299 "sb %[store2], 3(%[dst_ptr]) \n\t"
300
301 : [load1] "=&r"(load1), [load2] "=&r"(load2), [load3] "=&r"(load3),
302 [load4] "=&r"(load4), [p1] "=&r"(p1), [p2] "=&r"(p2),
303 [n1] "=&r"(n1), [n2] "=&r"(n2), [scratch1] "=&r"(scratch1),
304 [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
305 [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
306 [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
307 : [vector1b] "r"(vector1b), [vector2b] "r"(vector2b),
308 [vector3b] "r"(vector3b), [vector4b] "r"(vector4b),
309 [vector4a] "r"(vector4a), [src_stride] "r"(src_stride),
310 [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
311 }
312
313 /* Next row... */
314 src += src_stride;
315 dst += dst_stride;
316 }
317 }
318
vpx_convolve8_vert_dspr2(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const InterpKernel * filter,int x0_q4,int x_step_q4,int y0_q4,int y_step_q4,int w,int h)319 void vpx_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
320 uint8_t *dst, ptrdiff_t dst_stride,
321 const InterpKernel *filter, int x0_q4,
322 int x_step_q4, int y0_q4, int y_step_q4, int w,
323 int h) {
324 const int16_t *const filter_y = filter[y0_q4];
325 assert(y_step_q4 == 16);
326 assert(((const int32_t *)filter_y)[1] != 0x800000);
327
328 if (vpx_get_filter_taps(filter_y) == 2) {
329 vpx_convolve2_vert_dspr2(src, src_stride, dst, dst_stride, filter, x0_q4,
330 x_step_q4, y0_q4, 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 vpx_convolve8_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
354 x_step_q4, y0_q4, y_step_q4, w, h);
355 break;
356 }
357 }
358 }
359
360 #endif
361