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_convolve.h"
17#include "vpx_dsp/vpx_dsp_common.h"
18#include "vpx_ports/mem.h"
19
20#if HAVE_DSPR2
21static void convolve_bi_avg_vert_4_dspr2(const uint8_t *src,
22                                         int32_t src_stride,
23                                         uint8_t *dst,
24                                         int32_t dst_stride,
25                                         const int16_t *filter_y,
26                                         int32_t w,
27                                         int32_t h) {
28  int32_t       x, y;
29  const uint8_t *src_ptr;
30  uint8_t       *dst_ptr;
31  uint8_t       *cm = vpx_ff_cropTbl;
32  uint32_t      vector4a = 64;
33  uint32_t      load1, load2;
34  uint32_t      p1, p2;
35  uint32_t      scratch1, scratch2;
36  uint32_t      store1, store2;
37  int32_t       Temp1, Temp2;
38  const int16_t *filter = &filter_y[3];
39  uint32_t      filter45;
40
41  filter45 = ((const int32_t *)filter)[0];
42
43  for (y = h; y--;) {
44    /* prefetch data to cache memory */
45    prefetch_store(dst + dst_stride);
46
47    for (x = 0; x < w; x += 4) {
48      src_ptr = src + x;
49      dst_ptr = dst + x;
50
51      __asm__ __volatile__ (
52          "ulw              %[load1],     0(%[src_ptr])                   \n\t"
53          "add              %[src_ptr],   %[src_ptr],     %[src_stride]   \n\t"
54          "ulw              %[load2],     0(%[src_ptr])                   \n\t"
55
56          "mtlo             %[vector4a],  $ac0                            \n\t"
57          "mtlo             %[vector4a],  $ac1                            \n\t"
58          "mtlo             %[vector4a],  $ac2                            \n\t"
59          "mtlo             %[vector4a],  $ac3                            \n\t"
60          "mthi             $zero,        $ac0                            \n\t"
61          "mthi             $zero,        $ac1                            \n\t"
62          "mthi             $zero,        $ac2                            \n\t"
63          "mthi             $zero,        $ac3                            \n\t"
64
65          "preceu.ph.qbr    %[scratch1],  %[load1]                        \n\t"
66          "preceu.ph.qbr    %[p1],        %[load2]                        \n\t"
67          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
68          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
69
70          "dpa.w.ph         $ac0,         %[p1],          %[filter45]     \n\t"
71          "dpa.w.ph         $ac1,         %[p2],          %[filter45]     \n\t"
72
73          "preceu.ph.qbl    %[scratch1],  %[load1]                        \n\t"
74          "preceu.ph.qbl    %[p1],        %[load2]                        \n\t"
75          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
76          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
77
78          "dpa.w.ph         $ac2,         %[p1],          %[filter45]     \n\t"
79          "dpa.w.ph         $ac3,         %[p2],          %[filter45]     \n\t"
80
81          "extp             %[Temp1],     $ac0,           31              \n\t"
82          "extp             %[Temp2],     $ac1,           31              \n\t"
83
84          "lbu              %[scratch1],  0(%[dst_ptr])                   \n\t"
85          "lbu              %[scratch2],  1(%[dst_ptr])                   \n\t"
86
87          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
88          "addqh_r.w        %[store1],    %[store1],      %[scratch1]     \n\t" /* pixel 1 */
89          "extp             %[Temp1],     $ac2,           31              \n\t"
90
91          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
92          "addqh_r.w        %[store2],    %[store2],      %[scratch2]     \n\t" /* pixel 2 */
93          "extp             %[Temp2],     $ac3,           31              \n\t"
94          "lbu              %[scratch1],  2(%[dst_ptr])                   \n\t"
95
96          "sb               %[store1],    0(%[dst_ptr])                   \n\t"
97          "sb               %[store2],    1(%[dst_ptr])                   \n\t"
98          "lbu              %[scratch2],  3(%[dst_ptr])                   \n\t"
99
100          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
101          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
102          "addqh_r.w        %[store1],    %[store1],      %[scratch1]     \n\t" /* pixel 3 */
103          "addqh_r.w        %[store2],    %[store2],      %[scratch2]     \n\t" /* pixel 4 */
104
105          "sb               %[store1],    2(%[dst_ptr])                   \n\t"
106          "sb               %[store2],    3(%[dst_ptr])                   \n\t"
107
108          : [load1] "=&r" (load1), [load2] "=&r" (load2),
109            [p1] "=&r" (p1), [p2] "=&r" (p2),
110            [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
111            [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
112            [store1] "=&r" (store1), [store2] "=&r" (store2),
113            [src_ptr] "+r" (src_ptr)
114          : [filter45] "r" (filter45), [vector4a] "r" (vector4a),
115            [src_stride] "r" (src_stride), [cm] "r" (cm),
116            [dst_ptr] "r" (dst_ptr)
117      );
118    }
119
120    /* Next row... */
121    src += src_stride;
122    dst += dst_stride;
123  }
124}
125
126static void convolve_bi_avg_vert_64_dspr2(const uint8_t *src,
127                                          int32_t src_stride,
128                                          uint8_t *dst,
129                                          int32_t dst_stride,
130                                          const int16_t *filter_y,
131                                          int32_t h) {
132  int32_t       x, y;
133  const uint8_t *src_ptr;
134  uint8_t       *dst_ptr;
135  uint8_t       *cm = vpx_ff_cropTbl;
136  uint32_t      vector4a = 64;
137  uint32_t      load1, load2;
138  uint32_t      p1, p2;
139  uint32_t      scratch1, scratch2;
140  uint32_t      store1, store2;
141  int32_t       Temp1, Temp2;
142  const int16_t *filter = &filter_y[3];
143  uint32_t filter45;;
144
145  filter45 = ((const int32_t *)filter)[0];
146
147  for (y = h; y--;) {
148    /* prefetch data to cache memory */
149    prefetch_store(dst + dst_stride);
150    prefetch_store(dst + dst_stride + 32);
151
152    for (x = 0; x < 64; x += 4) {
153      src_ptr = src + x;
154      dst_ptr = dst + x;
155
156      __asm__ __volatile__ (
157          "ulw              %[load1],     0(%[src_ptr])                   \n\t"
158          "add              %[src_ptr],   %[src_ptr],     %[src_stride]   \n\t"
159          "ulw              %[load2],     0(%[src_ptr])                   \n\t"
160
161          "mtlo             %[vector4a],  $ac0                            \n\t"
162          "mtlo             %[vector4a],  $ac1                            \n\t"
163          "mtlo             %[vector4a],  $ac2                            \n\t"
164          "mtlo             %[vector4a],  $ac3                            \n\t"
165          "mthi             $zero,        $ac0                            \n\t"
166          "mthi             $zero,        $ac1                            \n\t"
167          "mthi             $zero,        $ac2                            \n\t"
168          "mthi             $zero,        $ac3                            \n\t"
169
170          "preceu.ph.qbr    %[scratch1],  %[load1]                        \n\t"
171          "preceu.ph.qbr    %[p1],        %[load2]                        \n\t"
172          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
173          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
174
175          "dpa.w.ph         $ac0,         %[p1],          %[filter45]     \n\t"
176          "dpa.w.ph         $ac1,         %[p2],          %[filter45]     \n\t"
177
178          "preceu.ph.qbl    %[scratch1],  %[load1]                        \n\t"
179          "preceu.ph.qbl    %[p1],        %[load2]                        \n\t"
180          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
181          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
182
183          "dpa.w.ph         $ac2,         %[p1],          %[filter45]     \n\t"
184          "dpa.w.ph         $ac3,         %[p2],          %[filter45]     \n\t"
185
186          "extp             %[Temp1],     $ac0,           31              \n\t"
187          "extp             %[Temp2],     $ac1,           31              \n\t"
188
189          "lbu              %[scratch1],  0(%[dst_ptr])                   \n\t"
190          "lbu              %[scratch2],  1(%[dst_ptr])                   \n\t"
191
192          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
193          "addqh_r.w        %[store1],    %[store1],      %[scratch1]     \n\t" /* pixel 1 */
194          "extp             %[Temp1],     $ac2,           31              \n\t"
195
196          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
197          "addqh_r.w        %[store2],    %[store2],      %[scratch2]     \n\t" /* pixel 2 */
198          "extp             %[Temp2],     $ac3,           31              \n\t"
199          "lbu              %[scratch1],  2(%[dst_ptr])                   \n\t"
200
201          "sb               %[store1],    0(%[dst_ptr])                   \n\t"
202          "sb               %[store2],    1(%[dst_ptr])                   \n\t"
203          "lbu              %[scratch2],  3(%[dst_ptr])                   \n\t"
204
205          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
206          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
207          "addqh_r.w        %[store1],    %[store1],      %[scratch1]     \n\t" /* pixel 3 */
208          "addqh_r.w        %[store2],    %[store2],      %[scratch2]     \n\t" /* pixel 4 */
209
210          "sb               %[store1],    2(%[dst_ptr])                   \n\t"
211          "sb               %[store2],    3(%[dst_ptr])                   \n\t"
212
213          : [load1] "=&r" (load1), [load2] "=&r" (load2),
214            [p1] "=&r" (p1), [p2] "=&r" (p2),
215            [scratch1] "=&r" (scratch1), [scratch2] "=&r" (scratch2),
216            [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
217            [store1] "=&r" (store1), [store2] "=&r" (store2),
218            [src_ptr] "+r" (src_ptr)
219          : [filter45] "r" (filter45), [vector4a] "r" (vector4a),
220            [src_stride] "r" (src_stride), [cm] "r" (cm),
221            [dst_ptr] "r" (dst_ptr)
222      );
223    }
224
225    /* Next row... */
226    src += src_stride;
227    dst += dst_stride;
228  }
229}
230
231void vpx_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
232                                  uint8_t *dst, ptrdiff_t dst_stride,
233                                  const int16_t *filter_x, int x_step_q4,
234                                  const int16_t *filter_y, int y_step_q4,
235                                  int w, int h) {
236  uint32_t pos = 38;
237
238  assert(y_step_q4 == 16);
239
240  /* bit positon for extract from acc */
241  __asm__ __volatile__ (
242    "wrdsp      %[pos],     1           \n\t"
243    :
244    : [pos] "r" (pos)
245  );
246
247  prefetch_store(dst);
248
249  switch (w) {
250    case 4:
251    case 8:
252    case 16:
253    case 32:
254      convolve_bi_avg_vert_4_dspr2(src, src_stride,
255                                   dst, dst_stride,
256                                   filter_y, w, h);
257      break;
258    case 64:
259      prefetch_store(dst + 32);
260      convolve_bi_avg_vert_64_dspr2(src, src_stride,
261                                    dst, dst_stride,
262                                    filter_y, h);
263      break;
264    default:
265      vpx_convolve8_avg_vert_c(src, src_stride,
266                               dst, dst_stride,
267                               filter_x, x_step_q4,
268                               filter_y, y_step_q4,
269                               w, h);
270      break;
271  }
272}
273#endif
274