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_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;
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
68          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
69          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
70
71          "dpa.w.ph         $ac0,         %[p1],          %[filter45]     \n\t"
72          "dpa.w.ph         $ac1,         %[p2],          %[filter45]     \n\t"
73
74          "preceu.ph.qbl    %[scratch1],  %[load1]                        \n\t"
75          "preceu.ph.qbl    %[p1],        %[load2]                        \n\t"
76
77          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
78          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
79
80          "dpa.w.ph         $ac2,         %[p1],          %[filter45]     \n\t"
81          "dpa.w.ph         $ac3,         %[p2],          %[filter45]     \n\t"
82
83          "extp             %[Temp1],     $ac0,           31              \n\t"
84          "extp             %[Temp2],     $ac1,           31              \n\t"
85
86          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
87          "extp             %[Temp1],     $ac2,           31              \n\t"
88
89          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
90          "extp             %[Temp2],     $ac3,           31              \n\t"
91
92          "sb               %[store1],    0(%[dst_ptr])                   \n\t"
93          "sb               %[store2],    1(%[dst_ptr])                   \n\t"
94
95          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
96          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
97
98          "sb               %[store1],    2(%[dst_ptr])                   \n\t"
99          "sb               %[store2],    3(%[dst_ptr])                   \n\t"
100
101          : [load1] "=&r" (load1), [load2] "=&r" (load2),
102            [p1] "=&r" (p1), [p2] "=&r" (p2),
103            [scratch1] "=&r" (scratch1),
104            [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
105            [store1] "=&r" (store1), [store2] "=&r" (store2),
106            [src_ptr] "+r" (src_ptr)
107          : [filter45] "r" (filter45),[vector4a] "r" (vector4a),
108            [src_stride] "r" (src_stride),
109            [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
110      );
111    }
112
113    /* Next row... */
114    src += src_stride;
115    dst += dst_stride;
116  }
117}
118
119static void convolve_bi_vert_64_dspr2(const uint8_t *src,
120                                      int32_t src_stride,
121                                      uint8_t *dst,
122                                      int32_t dst_stride,
123                                      const int16_t *filter_y,
124                                      int32_t h) {
125  int32_t       x, y;
126  const uint8_t *src_ptr;
127  uint8_t       *dst_ptr;
128  uint8_t       *cm = vpx_ff_cropTbl;
129  uint32_t      vector4a = 64;
130  uint32_t      load1, load2;
131  uint32_t      p1, p2;
132  uint32_t      scratch1;
133  uint32_t      store1, store2;
134  int32_t       Temp1, Temp2;
135  const int16_t *filter = &filter_y[3];
136  uint32_t      filter45;
137
138  filter45 = ((const int32_t *)filter)[0];
139
140  for (y = h; y--;) {
141    /* prefetch data to cache memory */
142    prefetch_store(dst + dst_stride);
143
144    for (x = 0; x < 64; x += 4) {
145      src_ptr = src + x;
146      dst_ptr = dst + x;
147
148      __asm__ __volatile__ (
149          "ulw              %[load1],     0(%[src_ptr])                   \n\t"
150          "add              %[src_ptr],   %[src_ptr],     %[src_stride]   \n\t"
151          "ulw              %[load2],     0(%[src_ptr])                   \n\t"
152
153          "mtlo             %[vector4a],  $ac0                            \n\t"
154          "mtlo             %[vector4a],  $ac1                            \n\t"
155          "mtlo             %[vector4a],  $ac2                            \n\t"
156          "mtlo             %[vector4a],  $ac3                            \n\t"
157          "mthi             $zero,        $ac0                            \n\t"
158          "mthi             $zero,        $ac1                            \n\t"
159          "mthi             $zero,        $ac2                            \n\t"
160          "mthi             $zero,        $ac3                            \n\t"
161
162          "preceu.ph.qbr    %[scratch1],  %[load1]                        \n\t"
163          "preceu.ph.qbr    %[p1],        %[load2]                        \n\t"
164
165          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
166          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
167
168          "dpa.w.ph         $ac0,         %[p1],          %[filter45]     \n\t"
169          "dpa.w.ph         $ac1,         %[p2],          %[filter45]     \n\t"
170
171          "preceu.ph.qbl    %[scratch1],  %[load1]                        \n\t"
172          "preceu.ph.qbl    %[p1],        %[load2]                        \n\t"
173
174          "precrq.ph.w      %[p2],        %[p1],          %[scratch1]     \n\t" /* pixel 2 */
175          "append           %[p1],        %[scratch1],    16              \n\t" /* pixel 1 */
176
177          "dpa.w.ph         $ac2,         %[p1],          %[filter45]     \n\t"
178          "dpa.w.ph         $ac3,         %[p2],          %[filter45]     \n\t"
179
180          "extp             %[Temp1],     $ac0,           31              \n\t"
181          "extp             %[Temp2],     $ac1,           31              \n\t"
182
183          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
184          "extp             %[Temp1],     $ac2,           31              \n\t"
185
186          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
187          "extp             %[Temp2],     $ac3,           31              \n\t"
188
189          "sb               %[store1],    0(%[dst_ptr])                   \n\t"
190          "sb               %[store2],    1(%[dst_ptr])                   \n\t"
191
192          "lbux             %[store1],    %[Temp1](%[cm])                 \n\t"
193          "lbux             %[store2],    %[Temp2](%[cm])                 \n\t"
194
195          "sb               %[store1],    2(%[dst_ptr])                   \n\t"
196          "sb               %[store2],    3(%[dst_ptr])                   \n\t"
197
198          : [load1] "=&r" (load1), [load2] "=&r" (load2),
199            [p1] "=&r" (p1), [p2] "=&r" (p2),
200            [scratch1] "=&r" (scratch1),
201            [Temp1] "=&r" (Temp1), [Temp2] "=&r" (Temp2),
202            [store1] "=&r" (store1), [store2] "=&r" (store2),
203            [src_ptr] "+r" (src_ptr)
204          : [filter45] "r" (filter45),[vector4a] "r" (vector4a),
205            [src_stride] "r" (src_stride),
206            [cm] "r" (cm), [dst_ptr] "r" (dst_ptr)
207      );
208    }
209
210    /* Next row... */
211    src += src_stride;
212    dst += dst_stride;
213  }
214}
215
216void vpx_convolve2_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
217                              uint8_t *dst, ptrdiff_t dst_stride,
218                              const int16_t *filter_x, int x_step_q4,
219                              const int16_t *filter_y, int y_step_q4,
220                              int w, int h) {
221  uint32_t pos = 38;
222
223  assert(y_step_q4 == 16);
224
225  /* bit positon for extract from acc */
226  __asm__ __volatile__ (
227    "wrdsp      %[pos],     1           \n\t"
228    :
229    : [pos] "r" (pos)
230  );
231
232  prefetch_store(dst);
233
234  switch (w) {
235    case 4 :
236    case 8 :
237    case 16 :
238    case 32 :
239      convolve_bi_vert_4_dspr2(src, src_stride,
240                               dst, dst_stride,
241                               filter_y, w, h);
242      break;
243    case 64 :
244      prefetch_store(dst + 32);
245      convolve_bi_vert_64_dspr2(src, src_stride,
246                                dst, dst_stride,
247                                filter_y, h);
248      break;
249    default:
250      vpx_convolve8_vert_c(src, src_stride,
251                           dst, dst_stride,
252                           filter_x, x_step_q4,
253                           filter_y, y_step_q4,
254                           w, h);
255      break;
256  }
257}
258#endif
259