1/*
2 *  Copyright (c) 2015 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 <string.h>
12#include "vpx_dsp/mips/macros_msa.h"
13
14static void copy_width8_msa(const uint8_t *src, int32_t src_stride,
15                            uint8_t *dst, int32_t dst_stride, int32_t height) {
16  int32_t cnt;
17  uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
18  v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
19
20  if (0 == height % 12) {
21    for (cnt = (height / 12); cnt--;) {
22      LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
23      src += (8 * src_stride);
24
25      out0 = __msa_copy_u_d((v2i64)src0, 0);
26      out1 = __msa_copy_u_d((v2i64)src1, 0);
27      out2 = __msa_copy_u_d((v2i64)src2, 0);
28      out3 = __msa_copy_u_d((v2i64)src3, 0);
29      out4 = __msa_copy_u_d((v2i64)src4, 0);
30      out5 = __msa_copy_u_d((v2i64)src5, 0);
31      out6 = __msa_copy_u_d((v2i64)src6, 0);
32      out7 = __msa_copy_u_d((v2i64)src7, 0);
33
34      SD4(out0, out1, out2, out3, dst, dst_stride);
35      dst += (4 * dst_stride);
36      SD4(out4, out5, out6, out7, dst, dst_stride);
37      dst += (4 * dst_stride);
38
39      LD_UB4(src, src_stride, src0, src1, src2, src3);
40      src += (4 * src_stride);
41
42      out0 = __msa_copy_u_d((v2i64)src0, 0);
43      out1 = __msa_copy_u_d((v2i64)src1, 0);
44      out2 = __msa_copy_u_d((v2i64)src2, 0);
45      out3 = __msa_copy_u_d((v2i64)src3, 0);
46      SD4(out0, out1, out2, out3, dst, dst_stride);
47      dst += (4 * dst_stride);
48    }
49  } else if (0 == height % 8) {
50    for (cnt = height >> 3; cnt--;) {
51      LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
52      src += (8 * src_stride);
53
54      out0 = __msa_copy_u_d((v2i64)src0, 0);
55      out1 = __msa_copy_u_d((v2i64)src1, 0);
56      out2 = __msa_copy_u_d((v2i64)src2, 0);
57      out3 = __msa_copy_u_d((v2i64)src3, 0);
58      out4 = __msa_copy_u_d((v2i64)src4, 0);
59      out5 = __msa_copy_u_d((v2i64)src5, 0);
60      out6 = __msa_copy_u_d((v2i64)src6, 0);
61      out7 = __msa_copy_u_d((v2i64)src7, 0);
62
63      SD4(out0, out1, out2, out3, dst, dst_stride);
64      dst += (4 * dst_stride);
65      SD4(out4, out5, out6, out7, dst, dst_stride);
66      dst += (4 * dst_stride);
67    }
68  } else if (0 == height % 4) {
69    for (cnt = (height / 4); cnt--;) {
70      LD_UB4(src, src_stride, src0, src1, src2, src3);
71      src += (4 * src_stride);
72      out0 = __msa_copy_u_d((v2i64)src0, 0);
73      out1 = __msa_copy_u_d((v2i64)src1, 0);
74      out2 = __msa_copy_u_d((v2i64)src2, 0);
75      out3 = __msa_copy_u_d((v2i64)src3, 0);
76
77      SD4(out0, out1, out2, out3, dst, dst_stride);
78      dst += (4 * dst_stride);
79    }
80  } else if (0 == height % 2) {
81    for (cnt = (height / 2); cnt--;) {
82      LD_UB2(src, src_stride, src0, src1);
83      src += (2 * src_stride);
84      out0 = __msa_copy_u_d((v2i64)src0, 0);
85      out1 = __msa_copy_u_d((v2i64)src1, 0);
86
87      SD(out0, dst);
88      dst += dst_stride;
89      SD(out1, dst);
90      dst += dst_stride;
91    }
92  }
93}
94
95static void copy_16multx8mult_msa(const uint8_t *src, int32_t src_stride,
96                                  uint8_t *dst, int32_t dst_stride,
97                                  int32_t height, int32_t width) {
98  int32_t cnt, loop_cnt;
99  const uint8_t *src_tmp;
100  uint8_t *dst_tmp;
101  v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
102
103  for (cnt = (width >> 4); cnt--;) {
104    src_tmp = src;
105    dst_tmp = dst;
106
107    for (loop_cnt = (height >> 3); loop_cnt--;) {
108      LD_UB8(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6,
109             src7);
110      src_tmp += (8 * src_stride);
111
112      ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, dst_tmp,
113             dst_stride);
114      dst_tmp += (8 * dst_stride);
115    }
116
117    src += 16;
118    dst += 16;
119  }
120}
121
122static void copy_width16_msa(const uint8_t *src, int32_t src_stride,
123                             uint8_t *dst, int32_t dst_stride, int32_t height) {
124  int32_t cnt;
125  v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
126
127  if (0 == height % 12) {
128    for (cnt = (height / 12); cnt--;) {
129      LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
130      src += (8 * src_stride);
131      ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, dst, dst_stride);
132      dst += (8 * dst_stride);
133
134      LD_UB4(src, src_stride, src0, src1, src2, src3);
135      src += (4 * src_stride);
136      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
137      dst += (4 * dst_stride);
138    }
139  } else if (0 == height % 8) {
140    copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 16);
141  } else if (0 == height % 4) {
142    for (cnt = (height >> 2); cnt--;) {
143      LD_UB4(src, src_stride, src0, src1, src2, src3);
144      src += (4 * src_stride);
145
146      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
147      dst += (4 * dst_stride);
148    }
149  }
150}
151
152static void copy_width32_msa(const uint8_t *src, int32_t src_stride,
153                             uint8_t *dst, int32_t dst_stride, int32_t height) {
154  int32_t cnt;
155  v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
156
157  if (0 == height % 12) {
158    for (cnt = (height / 12); cnt--;) {
159      LD_UB4(src, src_stride, src0, src1, src2, src3);
160      LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
161      src += (4 * src_stride);
162      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
163      ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
164      dst += (4 * dst_stride);
165
166      LD_UB4(src, src_stride, src0, src1, src2, src3);
167      LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
168      src += (4 * src_stride);
169      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
170      ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
171      dst += (4 * dst_stride);
172
173      LD_UB4(src, src_stride, src0, src1, src2, src3);
174      LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
175      src += (4 * src_stride);
176      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
177      ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
178      dst += (4 * dst_stride);
179    }
180  } else if (0 == height % 8) {
181    copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 32);
182  } else if (0 == height % 4) {
183    for (cnt = (height >> 2); cnt--;) {
184      LD_UB4(src, src_stride, src0, src1, src2, src3);
185      LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
186      src += (4 * src_stride);
187      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
188      ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
189      dst += (4 * dst_stride);
190    }
191  }
192}
193
194static void copy_width64_msa(const uint8_t *src, int32_t src_stride,
195                             uint8_t *dst, int32_t dst_stride, int32_t height) {
196  copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 64);
197}
198
199void vpx_convolve_copy_msa(const uint8_t *src, ptrdiff_t src_stride,
200                           uint8_t *dst, ptrdiff_t dst_stride,
201                           const int16_t *filter_x, int32_t filter_x_stride,
202                           const int16_t *filter_y, int32_t filter_y_stride,
203                           int32_t w, int32_t h) {
204  (void)filter_x;
205  (void)filter_y;
206  (void)filter_x_stride;
207  (void)filter_y_stride;
208
209  switch (w) {
210    case 4: {
211      uint32_t cnt, tmp;
212      /* 1 word storage */
213      for (cnt = h; cnt--;) {
214        tmp = LW(src);
215        SW(tmp, dst);
216        src += src_stride;
217        dst += dst_stride;
218      }
219      break;
220    }
221    case 8: {
222      copy_width8_msa(src, src_stride, dst, dst_stride, h);
223      break;
224    }
225    case 16: {
226      copy_width16_msa(src, src_stride, dst, dst_stride, h);
227      break;
228    }
229    case 32: {
230      copy_width32_msa(src, src_stride, dst, dst_stride, h);
231      break;
232    }
233    case 64: {
234      copy_width64_msa(src, src_stride, dst, dst_stride, h);
235      break;
236    }
237    default: {
238      uint32_t cnt;
239      for (cnt = h; cnt--;) {
240        memcpy(dst, src, w);
241        src += src_stride;
242        dst += dst_stride;
243      }
244      break;
245    }
246  }
247}
248