1/*
2 * Copyright (c) 2015 Shivraj Patil (Shivraj.Patil@imgtec.com)
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21#ifndef AVCODEC_MIPS_VP9DSP_MIPS_H
22#define AVCODEC_MIPS_VP9DSP_MIPS_H
23
24#include <stddef.h>
25#include <stdint.h>
26
27#define VP9_8TAP_MIPS_MSA_FUNC(SIZE, type, type_idx)                         \
28void ff_put_8tap_##type##_##SIZE##h_msa(uint8_t *dst, ptrdiff_t dststride,   \
29                                        const uint8_t *src,                  \
30                                        ptrdiff_t srcstride,                 \
31                                        int h, int mx, int my);              \
32                                                                             \
33void ff_put_8tap_##type##_##SIZE##v_msa(uint8_t *dst, ptrdiff_t dststride,   \
34                                        const uint8_t *src,                  \
35                                        ptrdiff_t srcstride,                 \
36                                        int h, int mx, int my);              \
37                                                                             \
38void ff_put_8tap_##type##_##SIZE##hv_msa(uint8_t *dst, ptrdiff_t dststride,  \
39                                         const uint8_t *src,                 \
40                                         ptrdiff_t srcstride,                \
41                                         int h, int mx, int my);             \
42                                                                             \
43void ff_avg_8tap_##type##_##SIZE##h_msa(uint8_t *dst, ptrdiff_t dststride,   \
44                                        const uint8_t *src,                  \
45                                        ptrdiff_t srcstride,                 \
46                                        int h, int mx, int my);              \
47                                                                             \
48void ff_avg_8tap_##type##_##SIZE##v_msa(uint8_t *dst, ptrdiff_t dststride,   \
49                                        const uint8_t *src,                  \
50                                        ptrdiff_t srcstride,                 \
51                                        int h, int mx, int my);              \
52                                                                             \
53void ff_avg_8tap_##type##_##SIZE##hv_msa(uint8_t *dst, ptrdiff_t dststride,  \
54                                         const uint8_t *src,                 \
55                                         ptrdiff_t srcstride,                \
56                                         int h, int mx, int my);
57
58#define VP9_BILINEAR_MIPS_MSA_FUNC(SIZE)                                   \
59void ff_put_bilin_##SIZE##h_msa(uint8_t *dst, ptrdiff_t dststride,         \
60                                const uint8_t *src, ptrdiff_t srcstride,   \
61                                int h, int mx, int my);                    \
62                                                                           \
63void ff_put_bilin_##SIZE##v_msa(uint8_t *dst, ptrdiff_t dststride,         \
64                                const uint8_t *src, ptrdiff_t srcstride,   \
65                                int h, int mx, int my);                    \
66                                                                           \
67void ff_put_bilin_##SIZE##hv_msa(uint8_t *dst, ptrdiff_t dststride,        \
68                                 const uint8_t *src, ptrdiff_t srcstride,  \
69                                 int h, int mx, int my);                   \
70                                                                           \
71void ff_avg_bilin_##SIZE##h_msa(uint8_t *dst, ptrdiff_t dststride,         \
72                                const uint8_t *src, ptrdiff_t srcstride,   \
73                                int h, int mx, int my);                    \
74                                                                           \
75void ff_avg_bilin_##SIZE##v_msa(uint8_t *dst, ptrdiff_t dststride,         \
76                                const uint8_t *src, ptrdiff_t srcstride,   \
77                                int h, int mx, int my);                    \
78                                                                           \
79void ff_avg_bilin_##SIZE##hv_msa(uint8_t *dst, ptrdiff_t dststride,        \
80                                 const uint8_t *src, ptrdiff_t srcstride,  \
81                                 int h, int mx, int my);
82
83#define VP9_COPY_AVG_MIPS_MSA_FUNC(SIZE)                           \
84void ff_copy##SIZE##_msa(uint8_t *dst, ptrdiff_t dststride,        \
85                         const uint8_t *src, ptrdiff_t srcstride,  \
86                         int h, int mx, int my);                   \
87                                                                   \
88void ff_avg##SIZE##_msa(uint8_t *dst, ptrdiff_t dststride,         \
89                        const uint8_t *src, ptrdiff_t srcstride,   \
90                        int h, int mx, int my);
91
92VP9_8TAP_MIPS_MSA_FUNC(64, regular, FILTER_8TAP_REGULAR);
93VP9_8TAP_MIPS_MSA_FUNC(32, regular, FILTER_8TAP_REGULAR);
94VP9_8TAP_MIPS_MSA_FUNC(16, regular, FILTER_8TAP_REGULAR);
95VP9_8TAP_MIPS_MSA_FUNC(8, regular, FILTER_8TAP_REGULAR);
96VP9_8TAP_MIPS_MSA_FUNC(4, regular, FILTER_8TAP_REGULAR);
97
98VP9_8TAP_MIPS_MSA_FUNC(64, sharp, FILTER_8TAP_SHARP);
99VP9_8TAP_MIPS_MSA_FUNC(32, sharp, FILTER_8TAP_SHARP);
100VP9_8TAP_MIPS_MSA_FUNC(16, sharp, FILTER_8TAP_SHARP);
101VP9_8TAP_MIPS_MSA_FUNC(8, sharp, FILTER_8TAP_SHARP);
102VP9_8TAP_MIPS_MSA_FUNC(4, sharp, FILTER_8TAP_SHARP);
103
104VP9_8TAP_MIPS_MSA_FUNC(64, smooth, FILTER_8TAP_SMOOTH);
105VP9_8TAP_MIPS_MSA_FUNC(32, smooth, FILTER_8TAP_SMOOTH);
106VP9_8TAP_MIPS_MSA_FUNC(16, smooth, FILTER_8TAP_SMOOTH);
107VP9_8TAP_MIPS_MSA_FUNC(8, smooth, FILTER_8TAP_SMOOTH);
108VP9_8TAP_MIPS_MSA_FUNC(4, smooth, FILTER_8TAP_SMOOTH);
109
110VP9_BILINEAR_MIPS_MSA_FUNC(64);
111VP9_BILINEAR_MIPS_MSA_FUNC(32);
112VP9_BILINEAR_MIPS_MSA_FUNC(16);
113VP9_BILINEAR_MIPS_MSA_FUNC(8);
114VP9_BILINEAR_MIPS_MSA_FUNC(4);
115
116VP9_COPY_AVG_MIPS_MSA_FUNC(64);
117VP9_COPY_AVG_MIPS_MSA_FUNC(32);
118VP9_COPY_AVG_MIPS_MSA_FUNC(16);
119VP9_COPY_AVG_MIPS_MSA_FUNC(8);
120VP9_COPY_AVG_MIPS_MSA_FUNC(4);
121
122#undef VP9_8TAP_MIPS_MSA_FUNC
123#undef VP9_BILINEAR_MIPS_MSA_FUNC
124#undef VP9_COPY_AVG_MIPS_MSA_FUNC
125
126void ff_loop_filter_h_4_8_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
127                              int32_t i, int32_t h);
128void ff_loop_filter_h_8_8_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
129                              int32_t i, int32_t h);
130void ff_loop_filter_h_16_8_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
131                               int32_t i, int32_t h);
132void ff_loop_filter_v_4_8_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
133                              int32_t i, int32_t h);
134void ff_loop_filter_v_8_8_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
135                              int32_t i, int32_t h);
136void ff_loop_filter_v_16_8_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
137                               int32_t i, int32_t h);
138void ff_loop_filter_h_44_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
139                                int32_t i, int32_t h);
140void ff_loop_filter_h_88_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
141                                int32_t i, int32_t h);
142void ff_loop_filter_h_16_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
143                                int32_t i, int32_t h);
144void ff_loop_filter_v_44_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
145                                int32_t i, int32_t h);
146void ff_loop_filter_v_88_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
147                                int32_t i, int32_t h);
148void ff_loop_filter_v_16_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
149                                int32_t i, int32_t h);
150void ff_loop_filter_h_48_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
151                                int32_t i, int32_t h);
152void ff_loop_filter_h_84_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
153                                int32_t i, int32_t h);
154void ff_loop_filter_v_48_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
155                                int32_t i, int32_t h);
156void ff_loop_filter_v_84_16_msa(uint8_t *dst, ptrdiff_t stride, int32_t e,
157                                int32_t i, int32_t h);
158void ff_idct_idct_4x4_add_msa(uint8_t *dst, ptrdiff_t stride,
159                              int16_t *block, int eob);
160void ff_idct_idct_8x8_add_msa(uint8_t *dst, ptrdiff_t stride,
161                              int16_t *block, int eob);
162void ff_idct_idct_16x16_add_msa(uint8_t *dst, ptrdiff_t stride,
163                                int16_t *block, int eob);
164void ff_idct_idct_32x32_add_msa(uint8_t *dst, ptrdiff_t stride,
165                                int16_t *block, int eob);
166void ff_iadst_iadst_4x4_add_msa(uint8_t *dst, ptrdiff_t stride,
167                                int16_t *block, int eob);
168void ff_iadst_iadst_8x8_add_msa(uint8_t *dst, ptrdiff_t stride,
169                                int16_t *block, int eob);
170void ff_iadst_iadst_16x16_add_msa(uint8_t *dst, ptrdiff_t stride,
171                                  int16_t *block, int eob);
172void ff_iadst_idct_4x4_add_msa(uint8_t *dst, ptrdiff_t stride,
173                               int16_t *block, int eob);
174void ff_iadst_idct_8x8_add_msa(uint8_t *dst, ptrdiff_t stride,
175                               int16_t *block, int eob);
176void ff_iadst_idct_16x16_add_msa(uint8_t *dst, ptrdiff_t stride,
177                                 int16_t *block, int eob);
178void ff_idct_iadst_4x4_add_msa(uint8_t *pu8Dest, ptrdiff_t stride,
179                               int16_t *block, int eob);
180void ff_idct_iadst_8x8_add_msa(uint8_t *pu8Dest, ptrdiff_t stride,
181                               int16_t *block, int eob);
182void ff_idct_iadst_16x16_add_msa(uint8_t *pu8Dest, ptrdiff_t stride,
183                                 int16_t *block, int eob);
184void ff_iwht_iwht_4x4_add_msa(uint8_t *dst, ptrdiff_t stride,
185                              int16_t *block, int eob);
186
187void ff_vert_16x16_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
188                       const uint8_t *top);
189void ff_vert_32x32_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
190                       const uint8_t *top);
191void ff_hor_16x16_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
192                      const uint8_t *top);
193void ff_hor_32x32_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
194                      const uint8_t *top);
195void ff_dc_4x4_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
196                   const uint8_t *top);
197void ff_dc_8x8_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
198                   const uint8_t *top);
199void ff_dc_16x16_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
200                     const uint8_t *top);
201void ff_dc_32x32_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
202                     const uint8_t *top);
203void ff_dc_left_4x4_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
204                        const uint8_t *top);
205void ff_dc_left_8x8_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
206                        const uint8_t *top);
207void ff_dc_left_16x16_msa(uint8_t *dst, ptrdiff_t stride,
208                          const uint8_t *left, const uint8_t *top);
209void ff_dc_left_32x32_msa(uint8_t *dst, ptrdiff_t stride,
210                          const uint8_t *left, const uint8_t *top);
211void ff_dc_top_4x4_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
212                       const uint8_t *top);
213void ff_dc_top_8x8_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
214                       const uint8_t *top);
215void ff_dc_top_16x16_msa(uint8_t *dst, ptrdiff_t stride,
216                         const uint8_t *left, const uint8_t *top);
217void ff_dc_top_32x32_msa(uint8_t *dst, ptrdiff_t stride,
218                         const uint8_t *left, const uint8_t *top);
219void ff_dc_128_16x16_msa(uint8_t *dst, ptrdiff_t stride,
220                         const uint8_t *left, const uint8_t *top);
221void ff_dc_128_32x32_msa(uint8_t *dst, ptrdiff_t stride,
222                         const uint8_t *left, const uint8_t *top);
223void ff_dc_127_16x16_msa(uint8_t *dst, ptrdiff_t stride,
224                         const uint8_t *left, const uint8_t *top);
225void ff_dc_127_32x32_msa(uint8_t *dst, ptrdiff_t stride,
226                         const uint8_t *left, const uint8_t *top);
227void ff_dc_129_16x16_msa(uint8_t *dst, ptrdiff_t stride,
228                         const uint8_t *left, const uint8_t *top);
229void ff_dc_129_32x32_msa(uint8_t *dst, ptrdiff_t stride,
230                         const uint8_t *left, const uint8_t *top);
231void ff_tm_4x4_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
232                   const uint8_t *top);
233void ff_tm_8x8_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
234                   const uint8_t *top);
235void ff_tm_16x16_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
236                     const uint8_t *top);
237void ff_tm_32x32_msa(uint8_t *dst, ptrdiff_t stride, const uint8_t *left,
238                     const uint8_t *top);
239
240#define VP9_8TAP_MIPS_MMI_FUNC(SIZE, type, type_idx)                         \
241void ff_put_8tap_##type##_##SIZE##h_mmi(uint8_t *dst, ptrdiff_t dststride,   \
242                                        const uint8_t *src,                  \
243                                        ptrdiff_t srcstride,                 \
244                                        int h, int mx, int my);              \
245                                                                             \
246void ff_put_8tap_##type##_##SIZE##v_mmi(uint8_t *dst, ptrdiff_t dststride,   \
247                                        const uint8_t *src,                  \
248                                        ptrdiff_t srcstride,                 \
249                                        int h, int mx, int my);              \
250                                                                             \
251void ff_put_8tap_##type##_##SIZE##hv_mmi(uint8_t *dst, ptrdiff_t dststride,  \
252                                         const uint8_t *src,                 \
253                                         ptrdiff_t srcstride,                \
254                                         int h, int mx, int my);             \
255                                                                             \
256void ff_avg_8tap_##type##_##SIZE##h_mmi(uint8_t *dst, ptrdiff_t dststride,   \
257                                        const uint8_t *src,                  \
258                                        ptrdiff_t srcstride,                 \
259                                        int h, int mx, int my);              \
260                                                                             \
261void ff_avg_8tap_##type##_##SIZE##v_mmi(uint8_t *dst, ptrdiff_t dststride,   \
262                                        const uint8_t *src,                  \
263                                        ptrdiff_t srcstride,                 \
264                                        int h, int mx, int my);              \
265                                                                             \
266void ff_avg_8tap_##type##_##SIZE##hv_mmi(uint8_t *dst, ptrdiff_t dststride,  \
267                                         const uint8_t *src,                 \
268                                         ptrdiff_t srcstride,                \
269                                         int h, int mx, int my);
270
271VP9_8TAP_MIPS_MMI_FUNC(64, regular, FILTER_8TAP_REGULAR);
272VP9_8TAP_MIPS_MMI_FUNC(32, regular, FILTER_8TAP_REGULAR);
273VP9_8TAP_MIPS_MMI_FUNC(16, regular, FILTER_8TAP_REGULAR);
274VP9_8TAP_MIPS_MMI_FUNC(8, regular, FILTER_8TAP_REGULAR);
275VP9_8TAP_MIPS_MMI_FUNC(4, regular, FILTER_8TAP_REGULAR);
276
277VP9_8TAP_MIPS_MMI_FUNC(64, sharp, FILTER_8TAP_SHARP);
278VP9_8TAP_MIPS_MMI_FUNC(32, sharp, FILTER_8TAP_SHARP);
279VP9_8TAP_MIPS_MMI_FUNC(16, sharp, FILTER_8TAP_SHARP);
280VP9_8TAP_MIPS_MMI_FUNC(8, sharp, FILTER_8TAP_SHARP);
281VP9_8TAP_MIPS_MMI_FUNC(4, sharp, FILTER_8TAP_SHARP);
282
283VP9_8TAP_MIPS_MMI_FUNC(64, smooth, FILTER_8TAP_SMOOTH);
284VP9_8TAP_MIPS_MMI_FUNC(32, smooth, FILTER_8TAP_SMOOTH);
285VP9_8TAP_MIPS_MMI_FUNC(16, smooth, FILTER_8TAP_SMOOTH);
286VP9_8TAP_MIPS_MMI_FUNC(8, smooth, FILTER_8TAP_SMOOTH);
287VP9_8TAP_MIPS_MMI_FUNC(4, smooth, FILTER_8TAP_SMOOTH);
288#undef VP9_8TAP_MIPS_MMI_FUNC
289
290#endif  // #ifndef AVCODEC_MIPS_VP9DSP_MIPS_H
291