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
23static void convolve_bi_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;
36  uint32_t      p1, p2;
37  uint32_t      scratch1;
38  uint32_t      store1, store2;
39  int32_t       Temp1, Temp2;
40  const int16_t *filter = &filter_y[3];
41  uint32_t      filter45;
42
43  filter45 = ((const int32_t *)filter)[0];
44
45  for (y = h; y--;) {
46    /* prefetch data to cache memory */
47    vp9_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
58          "mtlo             %[vector4a],  $ac0                            \n\t"
59          "mtlo             %[vector4a],  $ac1                            \n\t"
60          "mtlo             %[vector4a],  $ac2                            \n\t"
61          "mtlo             %[vector4a],  $ac3                            \n\t"
62          "mthi             $zero,        $ac0                            \n\t"
63          "mthi             $zero,        $ac1                            \n\t"
64          "mthi             $zero,        $ac2                            \n\t"
65          "mthi             $zero,        $ac3                            \n\t"
66
67          "preceu.ph.qbr    %[scratch1],  %[load1]                        \n\t"
68          "preceu.ph.qbr    %[p1],        %[load2]                        \n\t"
69
70          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
71          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
72
73          "dpa.w.ph         $ac0,         %[p1],          %[filter45]     \n\t"
74          "dpa.w.ph         $ac1,         %[p2],          %[filter45]     \n\t"
75
76          "preceu.ph.qbl    %[scratch1],  %[load1]                        \n\t"
77          "preceu.ph.qbl    %[p1],        %[load2]                        \n\t"
78
79          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
80          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
81
82          "dpa.w.ph         $ac2,         %[p1],          %[filter45]     \n\t"
83          "dpa.w.ph         $ac3,         %[p2],          %[filter45]     \n\t"
84
85          "extp             %[Temp1],     $ac0,           31              \n\t"
86          "extp             %[Temp2],     $ac1,           31              \n\t"
87
88          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
89          "extp             %[Temp1],     $ac2,           31              \n\t"
90
91          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
92          "extp             %[Temp2],     $ac3,           31              \n\t"
93
94          "sb               %[store1],    0(%[dst_ptr])                   \n\t"
95          "sb               %[store2],    1(%[dst_ptr])                   \n\t"
96
97          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
98          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
99
100          "sb               %[store1],    2(%[dst_ptr])                   \n\t"
101          "sb               %[store2],    3(%[dst_ptr])                   \n\t"
102
103          : [load1] "=&r" (load1), [load2] "=&r" (load2),
104            [p1] "=&r" (p1), [p2] "=&r" (p2),
105            [scratch1] "=&r" (scratch1),
106            [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
107            [store1] "=&r" (store1), [store2] "=&r" (store2),
108            [src_ptr] "+r" (src_ptr)
109          : [filter45] "r" (filter45),[vector4a] "r" (vector4a),
110            [src_stride] "r" (src_stride),
111            [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
112      );
113    }
114
115    /* Next row... */
116    src += src_stride;
117    dst += dst_stride;
118  }
119}
120
121static void convolve_bi_vert_64_dspr2(const uint8_t *src,
122                                      int32_t src_stride,
123                                      uint8_t *dst,
124                                      int32_t dst_stride,
125                                      const int16_t *filter_y,
126                                      int32_t h) {
127  int32_t       x, y;
128  const uint8_t *src_ptr;
129  uint8_t       *dst_ptr;
130  uint8_t       *cm = vp9_ff_cropTbl;
131  uint32_t      vector4a = 64;
132  uint32_t      load1, load2;
133  uint32_t      p1, p2;
134  uint32_t      scratch1;
135  uint32_t      store1, store2;
136  int32_t       Temp1, Temp2;
137  const int16_t *filter = &filter_y[3];
138  uint32_t      filter45;
139
140  filter45 = ((const int32_t *)filter)[0];
141
142  for (y = h; y--;) {
143    /* prefetch data to cache memory */
144    vp9_prefetch_store(dst + dst_stride);
145
146    for (x = 0; x < 64; x += 4) {
147      src_ptr = src + x;
148      dst_ptr = dst + x;
149
150      __asm__ __volatile__ (
151          "ulw              %[load1],     0(%[src_ptr])                   \n\t"
152          "add              %[src_ptr],   %[src_ptr],     %[src_stride]   \n\t"
153          "ulw              %[load2],     0(%[src_ptr])                   \n\t"
154
155          "mtlo             %[vector4a],  $ac0                            \n\t"
156          "mtlo             %[vector4a],  $ac1                            \n\t"
157          "mtlo             %[vector4a],  $ac2                            \n\t"
158          "mtlo             %[vector4a],  $ac3                            \n\t"
159          "mthi             $zero,        $ac0                            \n\t"
160          "mthi             $zero,        $ac1                            \n\t"
161          "mthi             $zero,        $ac2                            \n\t"
162          "mthi             $zero,        $ac3                            \n\t"
163
164          "preceu.ph.qbr    %[scratch1],  %[load1]                        \n\t"
165          "preceu.ph.qbr    %[p1],        %[load2]                        \n\t"
166
167          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
168          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
169
170          "dpa.w.ph         $ac0,         %[p1],          %[filter45]     \n\t"
171          "dpa.w.ph         $ac1,         %[p2],          %[filter45]     \n\t"
172
173          "preceu.ph.qbl    %[scratch1],  %[load1]                        \n\t"
174          "preceu.ph.qbl    %[p1],        %[load2]                        \n\t"
175
176          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
177          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
178
179          "dpa.w.ph         $ac2,         %[p1],          %[filter45]     \n\t"
180          "dpa.w.ph         $ac3,         %[p2],          %[filter45]     \n\t"
181
182          "extp             %[Temp1],     $ac0,           31              \n\t"
183          "extp             %[Temp2],     $ac1,           31              \n\t"
184
185          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
186          "extp             %[Temp1],     $ac2,           31              \n\t"
187
188          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
189          "extp             %[Temp2],     $ac3,           31              \n\t"
190
191          "sb               %[store1],    0(%[dst_ptr])                   \n\t"
192          "sb               %[store2],    1(%[dst_ptr])                   \n\t"
193
194          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
195          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
196
197          "sb               %[store1],    2(%[dst_ptr])                   \n\t"
198          "sb               %[store2],    3(%[dst_ptr])                   \n\t"
199
200          : [load1] "=&r" (load1), [load2] "=&r" (load2),
201            [p1] "=&r" (p1), [p2] "=&r" (p2),
202            [scratch1] "=&r" (scratch1),
203            [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
204            [store1] "=&r" (store1), [store2] "=&r" (store2),
205            [src_ptr] "+r" (src_ptr)
206          : [filter45] "r" (filter45),[vector4a] "r" (vector4a),
207            [src_stride] "r" (src_stride),
208            [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
209      );
210    }
211
212    /* Next row... */
213    src += src_stride;
214    dst += dst_stride;
215  }
216}
217
218void vp9_convolve2_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
219                              uint8_t *dst, ptrdiff_t dst_stride,
220                              const int16_t *filter_x, int x_step_q4,
221                              const int16_t *filter_y, int y_step_q4,
222                              int w, int h) {
223  if (16 == y_step_q4) {
224    uint32_t pos = 38;
225
226    /* bit positon for extract from acc */
227    __asm__ __volatile__ (
228      "wrdsp      %[pos],     1           \n\t"
229      :
230      : [pos] "r" (pos)
231    );
232
233    vp9_prefetch_store(dst);
234
235    switch (w) {
236      case 4 :
237      case 8 :
238      case 16 :
239      case 32 :
240        convolve_bi_vert_4_dspr2(src, src_stride,
241                                 dst, dst_stride,
242                                 filter_y, w, h);
243        break;
244      case 64 :
245        vp9_prefetch_store(dst + 32);
246        convolve_bi_vert_64_dspr2(src, src_stride,
247                                  dst, dst_stride,
248                                  filter_y, h);
249        break;
250      default:
251        vp9_convolve8_vert_c(src, src_stride,
252                             dst, dst_stride,
253                             filter_x, x_step_q4,
254                             filter_y, y_step_q4,
255                             w, h);
256        break;
257    }
258  } else {
259    vp9_convolve8_vert_c(src, src_stride,
260                         dst, dst_stride,
261                         filter_x, x_step_q4,
262                         filter_y, y_step_q4,
263                         w, h);
264  }
265}
266#endif
267