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_rtcd.h"
13#include "vpx_dsp/mips/macros_msa.h"
14
15static void copy_width8_msa(const uint8_t *src, int32_t src_stride,
16                            uint8_t *dst, int32_t dst_stride, int32_t height) {
17  int32_t cnt;
18  uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
19  v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
20
21  if (0 == height % 12) {
22    for (cnt = (height / 12); cnt--;) {
23      LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
24      src += (8 * src_stride);
25
26      out0 = __msa_copy_u_d((v2i64)src0, 0);
27      out1 = __msa_copy_u_d((v2i64)src1, 0);
28      out2 = __msa_copy_u_d((v2i64)src2, 0);
29      out3 = __msa_copy_u_d((v2i64)src3, 0);
30      out4 = __msa_copy_u_d((v2i64)src4, 0);
31      out5 = __msa_copy_u_d((v2i64)src5, 0);
32      out6 = __msa_copy_u_d((v2i64)src6, 0);
33      out7 = __msa_copy_u_d((v2i64)src7, 0);
34
35      SD4(out0, out1, out2, out3, dst, dst_stride);
36      dst += (4 * dst_stride);
37      SD4(out4, out5, out6, out7, dst, dst_stride);
38      dst += (4 * dst_stride);
39
40      LD_UB4(src, src_stride, src0, src1, src2, src3);
41      src += (4 * src_stride);
42
43      out0 = __msa_copy_u_d((v2i64)src0, 0);
44      out1 = __msa_copy_u_d((v2i64)src1, 0);
45      out2 = __msa_copy_u_d((v2i64)src2, 0);
46      out3 = __msa_copy_u_d((v2i64)src3, 0);
47      SD4(out0, out1, out2, out3, dst, dst_stride);
48      dst += (4 * dst_stride);
49    }
50  } else if (0 == height % 8) {
51    for (cnt = height >> 3; cnt--;) {
52      LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
53      src += (8 * src_stride);
54
55      out0 = __msa_copy_u_d((v2i64)src0, 0);
56      out1 = __msa_copy_u_d((v2i64)src1, 0);
57      out2 = __msa_copy_u_d((v2i64)src2, 0);
58      out3 = __msa_copy_u_d((v2i64)src3, 0);
59      out4 = __msa_copy_u_d((v2i64)src4, 0);
60      out5 = __msa_copy_u_d((v2i64)src5, 0);
61      out6 = __msa_copy_u_d((v2i64)src6, 0);
62      out7 = __msa_copy_u_d((v2i64)src7, 0);
63
64      SD4(out0, out1, out2, out3, dst, dst_stride);
65      dst += (4 * dst_stride);
66      SD4(out4, out5, out6, out7, dst, dst_stride);
67      dst += (4 * dst_stride);
68    }
69  } else if (0 == height % 4) {
70    for (cnt = (height / 4); cnt--;) {
71      LD_UB4(src, src_stride, src0, src1, src2, src3);
72      src += (4 * src_stride);
73      out0 = __msa_copy_u_d((v2i64)src0, 0);
74      out1 = __msa_copy_u_d((v2i64)src1, 0);
75      out2 = __msa_copy_u_d((v2i64)src2, 0);
76      out3 = __msa_copy_u_d((v2i64)src3, 0);
77
78      SD4(out0, out1, out2, out3, dst, dst_stride);
79      dst += (4 * dst_stride);
80    }
81  } else if (0 == height % 2) {
82    for (cnt = (height / 2); cnt--;) {
83      LD_UB2(src, src_stride, src0, src1);
84      src += (2 * src_stride);
85      out0 = __msa_copy_u_d((v2i64)src0, 0);
86      out1 = __msa_copy_u_d((v2i64)src1, 0);
87
88      SD(out0, dst);
89      dst += dst_stride;
90      SD(out1, dst);
91      dst += dst_stride;
92    }
93  }
94}
95
96static void copy_16multx8mult_msa(const uint8_t *src, int32_t src_stride,
97                                  uint8_t *dst, int32_t dst_stride,
98                                  int32_t height, int32_t width) {
99  int32_t cnt, loop_cnt;
100  const uint8_t *src_tmp;
101  uint8_t *dst_tmp;
102  v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
103
104  for (cnt = (width >> 4); cnt--;) {
105    src_tmp = src;
106    dst_tmp = dst;
107
108    for (loop_cnt = (height >> 3); loop_cnt--;) {
109      LD_UB8(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6,
110             src7);
111      src_tmp += (8 * src_stride);
112
113      ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, dst_tmp,
114             dst_stride);
115      dst_tmp += (8 * dst_stride);
116    }
117
118    src += 16;
119    dst += 16;
120  }
121}
122
123static void copy_width16_msa(const uint8_t *src, int32_t src_stride,
124                             uint8_t *dst, int32_t dst_stride, int32_t height) {
125  int32_t cnt;
126  v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
127
128  if (0 == height % 12) {
129    for (cnt = (height / 12); cnt--;) {
130      LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
131      src += (8 * src_stride);
132      ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, dst, dst_stride);
133      dst += (8 * dst_stride);
134
135      LD_UB4(src, src_stride, src0, src1, src2, src3);
136      src += (4 * src_stride);
137      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
138      dst += (4 * dst_stride);
139    }
140  } else if (0 == height % 8) {
141    copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 16);
142  } else if (0 == height % 4) {
143    for (cnt = (height >> 2); cnt--;) {
144      LD_UB4(src, src_stride, src0, src1, src2, src3);
145      src += (4 * src_stride);
146
147      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
148      dst += (4 * dst_stride);
149    }
150  }
151}
152
153static void copy_width32_msa(const uint8_t *src, int32_t src_stride,
154                             uint8_t *dst, int32_t dst_stride, int32_t height) {
155  int32_t cnt;
156  v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
157
158  if (0 == height % 12) {
159    for (cnt = (height / 12); cnt--;) {
160      LD_UB4(src, src_stride, src0, src1, src2, src3);
161      LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
162      src += (4 * src_stride);
163      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
164      ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
165      dst += (4 * dst_stride);
166
167      LD_UB4(src, src_stride, src0, src1, src2, src3);
168      LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
169      src += (4 * src_stride);
170      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
171      ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
172      dst += (4 * dst_stride);
173
174      LD_UB4(src, src_stride, src0, src1, src2, src3);
175      LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
176      src += (4 * src_stride);
177      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
178      ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
179      dst += (4 * dst_stride);
180    }
181  } else if (0 == height % 8) {
182    copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 32);
183  } else if (0 == height % 4) {
184    for (cnt = (height >> 2); cnt--;) {
185      LD_UB4(src, src_stride, src0, src1, src2, src3);
186      LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
187      src += (4 * src_stride);
188      ST_UB4(src0, src1, src2, src3, dst, dst_stride);
189      ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
190      dst += (4 * dst_stride);
191    }
192  }
193}
194
195static void copy_width64_msa(const uint8_t *src, int32_t src_stride,
196                             uint8_t *dst, int32_t dst_stride, int32_t height) {
197  copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 64);
198}
199
200void vpx_convolve_copy_msa(const uint8_t *src, ptrdiff_t src_stride,
201                           uint8_t *dst, ptrdiff_t dst_stride,
202                           const InterpKernel *filter, int x0_q4,
203                           int32_t x_step_q4, int y0_q4, int32_t y_step_q4,
204                           int32_t w, int32_t h) {
205  (void)filter;
206  (void)x0_q4;
207  (void)x_step_q4;
208  (void)y0_q4;
209  (void)y_step_q4;
210
211  switch (w) {
212    case 4: {
213      uint32_t cnt, tmp;
214      /* 1 word storage */
215      for (cnt = h; cnt--;) {
216        tmp = LW(src);
217        SW(tmp, dst);
218        src += src_stride;
219        dst += dst_stride;
220      }
221      break;
222    }
223    case 8: {
224      copy_width8_msa(src, src_stride, dst, dst_stride, h);
225      break;
226    }
227    case 16: {
228      copy_width16_msa(src, src_stride, dst, dst_stride, h);
229      break;
230    }
231    case 32: {
232      copy_width32_msa(src, src_stride, dst, dst_stride, h);
233      break;
234    }
235    case 64: {
236      copy_width64_msa(src, src_stride, dst, dst_stride, h);
237      break;
238    }
239    default: {
240      uint32_t cnt;
241      for (cnt = h; cnt--;) {
242        memcpy(dst, src, w);
243        src += src_stride;
244        dst += dst_stride;
245      }
246      break;
247    }
248  }
249}
250