1 /*
2 * Loongson SIMD optimized vc1dsp
3 *
4 * Copyright (c) 2019 Loongson Technology Corporation Limited
5 * gxw <guxiwei-hf@loongson.cn>
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24 #include "vc1dsp_mips.h"
25 #include "constants.h"
26 #include "libavutil/mips/generic_macros_msa.h"
27
ff_vc1_inv_trans_8x8_msa(int16_t block[64])28 void ff_vc1_inv_trans_8x8_msa(int16_t block[64])
29 {
30 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
31 v4i32 in_r0, in_r1, in_r2, in_r3, in_r4, in_r5, in_r6, in_r7;
32 v4i32 in_l0, in_l1, in_l2, in_l3, in_l4, in_l5, in_l6, in_l7;
33 v4i32 t_r1, t_r2, t_r3, t_r4, t_r5, t_r6, t_r7, t_r8;
34 v4i32 t_l1, t_l2, t_l3, t_l4, t_l5, t_l6, t_l7, t_l8;
35 v4i32 cnst_12 = {12, 12, 12, 12};
36 v4i32 cnst_4 = {4, 4, 4, 4};
37 v4i32 cnst_16 = {16, 16, 16, 16};
38 v4i32 cnst_6 = {6, 6, 6, 6};
39 v4i32 cnst_15 = {15, 15, 15, 15};
40 v4i32 cnst_9 = {9, 9, 9, 9};
41 v4i32 cnst_1 = {1, 1, 1, 1};
42 v4i32 cnst_64 = {64, 64, 64, 64};
43
44 LD_SH8(block, 8, in0, in1, in2, in3, in4, in5, in6, in7);
45 UNPCK_SH_SW(in0, in_r0, in_l0);
46 UNPCK_SH_SW(in1, in_r1, in_l1);
47 UNPCK_SH_SW(in2, in_r2, in_l2);
48 UNPCK_SH_SW(in3, in_r3, in_l3);
49 UNPCK_SH_SW(in4, in_r4, in_l4);
50 UNPCK_SH_SW(in5, in_r5, in_l5);
51 UNPCK_SH_SW(in6, in_r6, in_l6);
52 UNPCK_SH_SW(in7, in_r7, in_l7);
53 // First loop
54 t_r1 = cnst_12 * (in_r0 + in_r4) + cnst_4;
55 t_l1 = cnst_12 * (in_l0 + in_l4) + cnst_4;
56 t_r2 = cnst_12 * (in_r0 - in_r4) + cnst_4;
57 t_l2 = cnst_12 * (in_l0 - in_l4) + cnst_4;
58 t_r3 = cnst_16 * in_r2 + cnst_6 * in_r6;
59 t_l3 = cnst_16 * in_l2 + cnst_6 * in_l6;
60 t_r4 = cnst_6 * in_r2 - cnst_16 * in_r6;
61 t_l4 = cnst_6 * in_l2 - cnst_16 * in_l6;
62
63 ADD4(t_r1, t_r3, t_l1, t_l3, t_r2, t_r4, t_l2, t_l4, t_r5, t_l5, t_r6, t_l6);
64 SUB4(t_r2, t_r4, t_l2, t_l4, t_r1, t_r3, t_l1, t_l3, t_r7, t_l7, t_r8, t_l8);
65 t_r1 = cnst_16 * in_r1 + cnst_15 * in_r3 + cnst_9 * in_r5 + cnst_4 * in_r7;
66 t_l1 = cnst_16 * in_l1 + cnst_15 * in_l3 + cnst_9 * in_l5 + cnst_4 * in_l7;
67 t_r2 = cnst_15 * in_r1 - cnst_4 * in_r3 - cnst_16 * in_r5 - cnst_9 * in_r7;
68 t_l2 = cnst_15 * in_l1 - cnst_4 * in_l3 - cnst_16 * in_l5 - cnst_9 * in_l7;
69 t_r3 = cnst_9 * in_r1 - cnst_16 * in_r3 + cnst_4 * in_r5 + cnst_15 * in_r7;
70 t_l3 = cnst_9 * in_l1 - cnst_16 * in_l3 + cnst_4 * in_l5 + cnst_15 * in_l7;
71 t_r4 = cnst_4 * in_r1 - cnst_9 * in_r3 + cnst_15 * in_r5 - cnst_16 * in_r7;
72 t_l4 = cnst_4 * in_l1 - cnst_9 * in_l3 + cnst_15 * in_l5 - cnst_16 * in_l7;
73
74 in_r0 = (t_r5 + t_r1) >> 3;
75 in_l0 = (t_l5 + t_l1) >> 3;
76 in_r1 = (t_r6 + t_r2) >> 3;
77 in_l1 = (t_l6 + t_l2) >> 3;
78 in_r2 = (t_r7 + t_r3) >> 3;
79 in_l2 = (t_l7 + t_l3) >> 3;
80 in_r3 = (t_r8 + t_r4) >> 3;
81 in_l3 = (t_l8 + t_l4) >> 3;
82
83 in_r4 = (t_r8 - t_r4) >> 3;
84 in_l4 = (t_l8 - t_l4) >> 3;
85 in_r5 = (t_r7 - t_r3) >> 3;
86 in_l5 = (t_l7 - t_l3) >> 3;
87 in_r6 = (t_r6 - t_r2) >> 3;
88 in_l6 = (t_l6 - t_l2) >> 3;
89 in_r7 = (t_r5 - t_r1) >> 3;
90 in_l7 = (t_l5 - t_l1) >> 3;
91 TRANSPOSE4x4_SW_SW(in_r0, in_r1, in_r2, in_r3, in_r0, in_r1, in_r2, in_r3);
92 TRANSPOSE4x4_SW_SW(in_l0, in_l1, in_l2, in_l3, in_l0, in_l1, in_l2, in_l3);
93 TRANSPOSE4x4_SW_SW(in_r4, in_r5, in_r6, in_r7, in_r4, in_r5, in_r6, in_r7);
94 TRANSPOSE4x4_SW_SW(in_l4, in_l5, in_l6, in_l7, in_l4, in_l5, in_l6, in_l7);
95 // Second loop
96 t_r1 = cnst_12 * (in_r0 + in_l0) + cnst_64;
97 t_l1 = cnst_12 * (in_r4 + in_l4) + cnst_64;
98 t_r2 = cnst_12 * (in_r0 - in_l0) + cnst_64;
99 t_l2 = cnst_12 * (in_r4 - in_l4) + cnst_64;
100 t_r3 = cnst_16 * in_r2 + cnst_6 * in_l2;
101 t_l3 = cnst_16 * in_r6 + cnst_6 * in_l6;
102 t_r4 = cnst_6 * in_r2 - cnst_16 * in_l2;
103 t_l4 = cnst_6 * in_r6 - cnst_16 * in_l6;
104
105 ADD4(t_r1, t_r3, t_l1, t_l3, t_r2, t_r4, t_l2, t_l4, t_r5, t_l5, t_r6, t_l6);
106 SUB4(t_r2, t_r4, t_l2, t_l4, t_r1, t_r3, t_l1, t_l3, t_r7, t_l7, t_r8, t_l8);
107 t_r1 = cnst_16 * in_r1 + cnst_15 * in_r3 + cnst_9 * in_l1 + cnst_4 * in_l3;
108 t_l1 = cnst_16 * in_r5 + cnst_15 * in_r7 + cnst_9 * in_l5 + cnst_4 * in_l7;
109 t_r2 = cnst_15 * in_r1 - cnst_4 * in_r3 - cnst_16 * in_l1 - cnst_9 * in_l3;
110 t_l2 = cnst_15 * in_r5 - cnst_4 * in_r7 - cnst_16 * in_l5 - cnst_9 * in_l7;
111 t_r3 = cnst_9 * in_r1 - cnst_16 * in_r3 + cnst_4 * in_l1 + cnst_15 * in_l3;
112 t_l3 = cnst_9 * in_r5 - cnst_16 * in_r7 + cnst_4 * in_l5 + cnst_15 * in_l7;
113 t_r4 = cnst_4 * in_r1 - cnst_9 * in_r3 + cnst_15 * in_l1 - cnst_16 * in_l3;
114 t_l4 = cnst_4 * in_r5 - cnst_9 * in_r7 + cnst_15 * in_l5 - cnst_16 * in_l7;
115
116 in_r0 = (t_r5 + t_r1) >> 7;
117 in_l0 = (t_l5 + t_l1) >> 7;
118 in_r1 = (t_r6 + t_r2) >> 7;
119 in_l1 = (t_l6 + t_l2) >> 7;
120 in_r2 = (t_r7 + t_r3) >> 7;
121 in_l2 = (t_l7 + t_l3) >> 7;
122 in_r3 = (t_r8 + t_r4) >> 7;
123 in_l3 = (t_l8 + t_l4) >> 7;
124
125 in_r4 = (t_r8 - t_r4 + cnst_1) >> 7;
126 in_l4 = (t_l8 - t_l4 + cnst_1) >> 7;
127 in_r5 = (t_r7 - t_r3 + cnst_1) >> 7;
128 in_l5 = (t_l7 - t_l3 + cnst_1) >> 7;
129 in_r6 = (t_r6 - t_r2 + cnst_1) >> 7;
130 in_l6 = (t_l6 - t_l2 + cnst_1) >> 7;
131 in_r7 = (t_r5 - t_r1 + cnst_1) >> 7;
132 in_l7 = (t_l5 - t_l1 + cnst_1) >> 7;
133 PCKEV_H4_SH(in_l0, in_r0, in_l1, in_r1, in_l2, in_r2, in_l3, in_r3,
134 in0, in1, in2, in3);
135 PCKEV_H4_SH(in_l4, in_r4, in_l5, in_r5, in_l6, in_r6, in_l7, in_r7,
136 in4, in5, in6, in7);
137 ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, block, 8);
138 }
139
ff_vc1_inv_trans_4x8_msa(uint8_t * dest,ptrdiff_t linesize,int16_t * block)140 void ff_vc1_inv_trans_4x8_msa(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
141 {
142 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
143 v4i32 in_r0, in_r1, in_r2, in_r3, in_r4, in_r5, in_r6, in_r7;
144 v4i32 t1, t2, t3, t4, t5, t6, t7, t8;
145 v4i32 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
146 v16i8 zero_m = { 0 };
147 v4i32 cnst_17 = {17, 17, 17, 17};
148 v4i32 cnst_22 = {22, 22, 22, 22};
149 v4i32 cnst_10 = {10, 10, 10, 10};
150 v4i32 cnst_12 = {12, 12, 12, 12};
151 v4i32 cnst_64 = {64, 64, 64, 64};
152 v4i32 cnst_16 = {16, 16, 16, 16};
153 v4i32 cnst_15 = {15, 15, 15, 15};
154 v4i32 cnst_4 = {4, 4, 4, 4};
155 v4i32 cnst_6 = {6, 6, 6, 6};
156 v4i32 cnst_9 = {9, 9, 9, 9};
157 v4i32 cnst_1 = {1, 1, 1, 1};
158
159 LD_SH8(block, 8, in0, in1, in2, in3, in4, in5, in6, in7);
160 UNPCK_R_SH_SW(in0, in_r0);
161 UNPCK_R_SH_SW(in1, in_r1);
162 UNPCK_R_SH_SW(in2, in_r2);
163 UNPCK_R_SH_SW(in3, in_r3);
164 UNPCK_R_SH_SW(in4, in_r4);
165 UNPCK_R_SH_SW(in5, in_r5);
166 UNPCK_R_SH_SW(in6, in_r6);
167 UNPCK_R_SH_SW(in7, in_r7);
168 // First loop
169 TRANSPOSE4x4_SW_SW(in_r0, in_r1, in_r2, in_r3, in_r0, in_r1, in_r2, in_r3);
170 TRANSPOSE4x4_SW_SW(in_r4, in_r5, in_r6, in_r7, in_r4, in_r5, in_r6, in_r7);
171 t1 = cnst_17 * (in_r0 + in_r2) + cnst_4;
172 t5 = cnst_17 * (in_r4 + in_r6) + cnst_4;
173 t2 = cnst_17 * (in_r0 - in_r2) + cnst_4;
174 t6 = cnst_17 * (in_r4 - in_r6) + cnst_4;
175 t3 = cnst_22 * in_r1 + cnst_10 * in_r3;
176 t7 = cnst_22 * in_r5 + cnst_10 * in_r7;
177 t4 = cnst_22 * in_r3 - cnst_10 * in_r1;
178 t8 = cnst_22 * in_r7 - cnst_10 * in_r5;
179
180 in_r0 = (t1 + t3) >> 3;
181 in_r4 = (t5 + t7) >> 3;
182 in_r1 = (t2 - t4) >> 3;
183 in_r5 = (t6 - t8) >> 3;
184 in_r2 = (t2 + t4) >> 3;
185 in_r6 = (t6 + t8) >> 3;
186 in_r3 = (t1 - t3) >> 3;
187 in_r7 = (t5 - t7) >> 3;
188 TRANSPOSE4x4_SW_SW(in_r0, in_r1, in_r2, in_r3, in_r0, in_r1, in_r2, in_r3);
189 TRANSPOSE4x4_SW_SW(in_r4, in_r5, in_r6, in_r7, in_r4, in_r5, in_r6, in_r7);
190 PCKEV_H4_SH(in_r1, in_r0, in_r3, in_r2, in_r5, in_r4, in_r7, in_r6,
191 in0, in1, in2, in3);
192 ST_D8(in0, in1, in2, in3, 0, 1, 0, 1, 0, 1, 0, 1, block, 8);
193 // Second loop
194 t1 = cnst_12 * (in_r0 + in_r4) + cnst_64;
195 t2 = cnst_12 * (in_r0 - in_r4) + cnst_64;
196 t3 = cnst_16 * in_r2 + cnst_6 * in_r6;
197 t4 = cnst_6 * in_r2 - cnst_16 * in_r6;
198 t5 = t1 + t3, t6 = t2 + t4;
199 t7 = t2 - t4, t8 = t1 - t3;
200 t1 = cnst_16 * in_r1 + cnst_15 * in_r3 + cnst_9 * in_r5 + cnst_4 * in_r7;
201 t2 = cnst_15 * in_r1 - cnst_4 * in_r3 - cnst_16 * in_r5 - cnst_9 * in_r7;
202 t3 = cnst_9 * in_r1 - cnst_16 * in_r3 + cnst_4 * in_r5 + cnst_15 * in_r7;
203 t4 = cnst_4 * in_r1 - cnst_9 * in_r3 + cnst_15 * in_r5 - cnst_16 * in_r7;
204 LD_SW8(dest, linesize, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
205 ILVR_B8_SW(zero_m, dst0, zero_m, dst1, zero_m, dst2, zero_m, dst3,
206 zero_m, dst4, zero_m, dst5, zero_m, dst6, zero_m, dst7,
207 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
208 ILVR_H4_SW(zero_m, dst0, zero_m, dst1, zero_m, dst2, zero_m, dst3,
209 dst0, dst1, dst2, dst3);
210 ILVR_H4_SW(zero_m, dst4, zero_m, dst5, zero_m, dst6, zero_m, dst7,
211 dst4, dst5, dst6, dst7);
212 in_r0 = (t5 + t1) >> 7;
213 in_r1 = (t6 + t2) >> 7;
214 in_r2 = (t7 + t3) >> 7;
215 in_r3 = (t8 + t4) >> 7;
216 in_r4 = (t8 - t4 + cnst_1) >> 7;
217 in_r5 = (t7 - t3 + cnst_1) >> 7;
218 in_r6 = (t6 - t2 + cnst_1) >> 7;
219 in_r7 = (t5 - t1 + cnst_1) >> 7;
220 ADD4(in_r0, dst0, in_r1, dst1, in_r2, dst2, in_r3, dst3,
221 in_r0, in_r1, in_r2, in_r3);
222 ADD4(in_r4, dst4, in_r5, dst5, in_r6, dst6, in_r7, dst7,
223 in_r4, in_r5, in_r6, in_r7);
224 CLIP_SW8_0_255(in_r0, in_r1, in_r2, in_r3, in_r4, in_r5, in_r6, in_r7);
225 PCKEV_H4_SH(in_r1, in_r0, in_r3, in_r2, in_r5, in_r4, in_r7, in_r6,
226 in0, in1, in2, in3);
227 PCKEV_B2_SH(in1, in0, in3, in2, in0, in1);
228 ST_W8(in0, in1, 0, 1, 2, 3, 0, 1, 2, 3, dest, linesize);
229 }
230
ff_vc1_inv_trans_8x4_msa(uint8_t * dest,ptrdiff_t linesize,int16_t * block)231 void ff_vc1_inv_trans_8x4_msa(uint8_t *dest, ptrdiff_t linesize, int16_t *block)
232 {
233 v4i32 in0, in1, in2, in3, in4, in5, in6, in7;
234 v4i32 t1, t2, t3, t4, t5, t6, t7, t8;
235 v4i32 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
236 v16i8 zero_m = { 0 };
237 v4i32 cnst_17 = {17, 17, 17, 17};
238 v4i32 cnst_22 = {22, 22, 22, 22};
239 v4i32 cnst_10 = {10, 10, 10, 10};
240 v4i32 cnst_12 = {12, 12, 12, 12};
241 v4i32 cnst_64 = {64, 64, 64, 64};
242 v4i32 cnst_16 = {16, 16, 16, 16};
243 v4i32 cnst_15 = {15, 15, 15, 15};
244 v4i32 cnst_4 = {4, 4, 4, 4};
245 v4i32 cnst_6 = {6, 6, 6, 6};
246 v4i32 cnst_9 = {9, 9, 9, 9};
247
248 LD_SW4(block, 8, t1, t2, t3, t4);
249 UNPCK_SH_SW(t1, in0, in4);
250 UNPCK_SH_SW(t2, in1, in5);
251 UNPCK_SH_SW(t3, in2, in6);
252 UNPCK_SH_SW(t4, in3, in7);
253 TRANSPOSE4x4_SW_SW(in0, in1, in2, in3, in0, in1, in2, in3);
254 TRANSPOSE4x4_SW_SW(in4, in5, in6, in7, in4, in5, in6, in7);
255 // First loop
256 t1 = cnst_12 * (in0 + in4) + cnst_4;
257 t2 = cnst_12 * (in0 - in4) + cnst_4;
258 t3 = cnst_16 * in2 + cnst_6 * in6;
259 t4 = cnst_6 * in2 - cnst_16 * in6;
260 t5 = t1 + t3, t6 = t2 + t4;
261 t7 = t2 - t4, t8 = t1 - t3;
262 t1 = cnst_16 * in1 + cnst_15 * in3 + cnst_9 * in5 + cnst_4 * in7;
263 t2 = cnst_15 * in1 - cnst_4 * in3 - cnst_16 * in5 - cnst_9 * in7;
264 t3 = cnst_9 * in1 - cnst_16 * in3 + cnst_4 * in5 + cnst_15 * in7;
265 t4 = cnst_4 * in1 - cnst_9 * in3 + cnst_15 * in5 - cnst_16 * in7;
266 in0 = (t5 + t1) >> 3;
267 in1 = (t6 + t2) >> 3;
268 in2 = (t7 + t3) >> 3;
269 in3 = (t8 + t4) >> 3;
270 in4 = (t8 - t4) >> 3;
271 in5 = (t7 - t3) >> 3;
272 in6 = (t6 - t2) >> 3;
273 in7 = (t5 - t1) >> 3;
274 TRANSPOSE4x4_SW_SW(in0, in1, in2, in3, in0, in1, in2, in3);
275 TRANSPOSE4x4_SW_SW(in4, in5, in6, in7, in4, in5, in6, in7);
276 PCKEV_H4_SW(in4, in0, in5, in1, in6, in2, in7, in3, t1, t2, t3, t4);
277 ST_SW4(t1, t2, t3, t4, block, 8);
278 // Second loop
279 LD_SW4(dest, linesize, dst0, dst1, dst2, dst3);
280 ILVR_B4_SW(zero_m, dst0, zero_m, dst1, zero_m, dst2, zero_m, dst3,
281 dst0, dst1, dst2, dst3);
282 ILVL_H4_SW(zero_m, dst0, zero_m, dst1, zero_m, dst2, zero_m, dst3,
283 dst4, dst5, dst6, dst7);
284 ILVR_H4_SW(zero_m, dst0, zero_m, dst1, zero_m, dst2, zero_m, dst3,
285 dst0, dst1, dst2, dst3);
286 // Right part
287 t1 = cnst_17 * (in0 + in2) + cnst_64;
288 t2 = cnst_17 * (in0 - in2) + cnst_64;
289 t3 = cnst_22 * in1 + cnst_10 * in3;
290 t4 = cnst_22 * in3 - cnst_10 * in1;
291 in0 = (t1 + t3) >> 7;
292 in1 = (t2 - t4) >> 7;
293 in2 = (t2 + t4) >> 7;
294 in3 = (t1 - t3) >> 7;
295 ADD4(in0, dst0, in1, dst1, in2, dst2, in3, dst3, in0, in1, in2, in3);
296 CLIP_SW4_0_255(in0, in1, in2, in3);
297 // Left part
298 t5 = cnst_17 * (in4 + in6) + cnst_64;
299 t6 = cnst_17 * (in4 - in6) + cnst_64;
300 t7 = cnst_22 * in5 + cnst_10 * in7;
301 t8 = cnst_22 * in7 - cnst_10 * in5;
302 in4 = (t5 + t7) >> 7;
303 in5 = (t6 - t8) >> 7;
304 in6 = (t6 + t8) >> 7;
305 in7 = (t5 - t7) >> 7;
306 ADD4(in4, dst4, in5, dst5, in6, dst6, in7, dst7, in4, in5, in6, in7);
307 CLIP_SW4_0_255(in4, in5, in6, in7);
308 PCKEV_H4_SW(in4, in0, in5, in1, in6, in2, in7, in3, in0, in1, in2, in3);
309 PCKEV_B2_SW(in1, in0, in3, in2, in0, in1);
310 ST_D4(in0, in1, 0, 1, 0, 1, dest, linesize);
311 }
312
put_vc1_mspel_mc_h_v_msa(uint8_t * dst,const uint8_t * src,ptrdiff_t stride,int hmode,int vmode,int rnd)313 static void put_vc1_mspel_mc_h_v_msa(uint8_t *dst, const uint8_t *src,
314 ptrdiff_t stride, int hmode, int vmode,
315 int rnd)
316 {
317 v8i16 in_r0, in_r1, in_r2, in_r3, in_l0, in_l1, in_l2, in_l3;
318 v8i16 t0, t1, t2, t3, t4, t5, t6, t7;
319 v8i16 t8, t9, t10, t11, t12, t13, t14, t15;
320 v8i16 cnst_para0, cnst_para1, cnst_para2, cnst_para3, cnst_r;
321 static const int para_value[][4] = {{4, 53, 18, 3},
322 {1, 9, 9, 1},
323 {3, 18, 53, 4}};
324 static const int shift_value[] = {0, 5, 1, 5};
325 int shift = (shift_value[hmode] + shift_value[vmode]) >> 1;
326 int r = (1 << (shift - 1)) + rnd - 1;
327 cnst_r = __msa_fill_h(r);
328 src -= 1, src -= stride;
329 cnst_para0 = __msa_fill_h(para_value[vmode - 1][0]);
330 cnst_para1 = __msa_fill_h(para_value[vmode - 1][1]);
331 cnst_para2 = __msa_fill_h(para_value[vmode - 1][2]);
332 cnst_para3 = __msa_fill_h(para_value[vmode - 1][3]);
333 LD_SH4(src, stride, in_l0, in_l1, in_l2, in_l3);
334 UNPCK_UB_SH(in_l0, in_r0, in_l0);
335 UNPCK_UB_SH(in_l1, in_r1, in_l1);
336 UNPCK_UB_SH(in_l2, in_r2, in_l2);
337 UNPCK_UB_SH(in_l3, in_r3, in_l3);
338 // row 0
339 t0 = cnst_para1 * in_r1 + cnst_para2 * in_r2
340 - cnst_para0 * in_r0 - cnst_para3 * in_r3;
341 t8 = cnst_para1 * in_l1 + cnst_para2 * in_l2
342 - cnst_para0 * in_l0 - cnst_para3 * in_l3;
343 in_l0 = LD_SH(src + 4 * stride);
344 UNPCK_UB_SH(in_l0, in_r0, in_l0);
345 // row 1
346 t1 = cnst_para1 * in_r2 + cnst_para2 * in_r3
347 - cnst_para0 * in_r1 - cnst_para3 * in_r0;
348 t9 = cnst_para1 * in_l2 + cnst_para2 * in_l3
349 - cnst_para0 * in_l1 - cnst_para3 * in_l0;
350 in_l1 = LD_SH(src + 5 * stride);
351 UNPCK_UB_SH(in_l1, in_r1, in_l1);
352 // row 2
353 t2 = cnst_para1 * in_r3 + cnst_para2 * in_r0
354 - cnst_para0 * in_r2 - cnst_para3 * in_r1;
355 t10 = cnst_para1 * in_l3 + cnst_para2 * in_l0
356 - cnst_para0 * in_l2 - cnst_para3 * in_l1;
357 in_l2 = LD_SH(src + 6 * stride);
358 UNPCK_UB_SH(in_l2, in_r2, in_l2);
359 // row 3
360 t3 = cnst_para1 * in_r0 + cnst_para2 * in_r1
361 - cnst_para0 * in_r3 - cnst_para3 * in_r2;
362 t11 = cnst_para1 * in_l0 + cnst_para2 * in_l1
363 - cnst_para0 * in_l3 - cnst_para3 * in_l2;
364 in_l3 = LD_SH(src + 7 * stride);
365 UNPCK_UB_SH(in_l3, in_r3, in_l3);
366 // row 4
367 t4 = cnst_para1 * in_r1 + cnst_para2 * in_r2
368 - cnst_para0 * in_r0 - cnst_para3 * in_r3;
369 t12 = cnst_para1 * in_l1 + cnst_para2 * in_l2
370 - cnst_para0 * in_l0 - cnst_para3 * in_l3;
371 in_l0 = LD_SH(src + 8 * stride);
372 UNPCK_UB_SH(in_l0, in_r0, in_l0);
373 // row 5
374 t5 = cnst_para1 * in_r2 + cnst_para2 * in_r3
375 - cnst_para0 * in_r1 - cnst_para3 * in_r0;
376 t13 = cnst_para1 * in_l2 + cnst_para2 * in_l3
377 - cnst_para0 * in_l1 - cnst_para3 * in_l0;
378 in_l1 = LD_SH(src + 9 * stride);
379 UNPCK_UB_SH(in_l1, in_r1, in_l1);
380 // row 6
381 t6 = cnst_para1 * in_r3 + cnst_para2 * in_r0
382 - cnst_para0 * in_r2 - cnst_para3 * in_r1;
383 t14 = cnst_para1 * in_l3 + cnst_para2 * in_l0
384 - cnst_para0 * in_l2 - cnst_para3 * in_l1;
385 in_l2 = LD_SH(src + 10 * stride);
386 UNPCK_UB_SH(in_l2, in_r2, in_l2);
387 // row 7
388 t7 = cnst_para1 * in_r0 + cnst_para2 * in_r1
389 - cnst_para0 * in_r3 - cnst_para3 * in_r2;
390 t15 = cnst_para1 * in_l0 + cnst_para2 * in_l1
391 - cnst_para0 * in_l3 - cnst_para3 * in_l2;
392
393 ADD4(t0, cnst_r, t1, cnst_r, t2, cnst_r, t3, cnst_r, t0, t1, t2, t3);
394 ADD4(t4, cnst_r, t5, cnst_r, t6, cnst_r, t7, cnst_r, t4, t5, t6, t7);
395 ADD4(t8, cnst_r, t9, cnst_r, t10, cnst_r, t11, cnst_r,
396 t8, t9, t10, t11);
397 ADD4(t12, cnst_r, t13, cnst_r, t14, cnst_r, t15, cnst_r,
398 t12, t13, t14, t15);
399 t0 >>= shift, t1 >>= shift, t2 >>= shift, t3 >>= shift;
400 t4 >>= shift, t5 >>= shift, t6 >>= shift, t7 >>= shift;
401 t8 >>= shift, t9 >>= shift, t10 >>= shift, t11 >>= shift;
402 t12 >>= shift, t13 >>= shift, t14 >>= shift, t15 >>= shift;
403 TRANSPOSE8x8_SH_SH(t0, t1, t2, t3, t4, t5, t6, t7,
404 t0, t1, t2, t3, t4, t5, t6, t7);
405 TRANSPOSE8x8_SH_SH(t8, t9, t10, t11, t12, t13, t14, t15,
406 t8, t9, t10, t11, t12, t13, t14, t15);
407 cnst_para0 = __msa_fill_h(para_value[hmode - 1][0]);
408 cnst_para1 = __msa_fill_h(para_value[hmode - 1][1]);
409 cnst_para2 = __msa_fill_h(para_value[hmode - 1][2]);
410 cnst_para3 = __msa_fill_h(para_value[hmode - 1][3]);
411 r = 64 - rnd;
412 cnst_r = __msa_fill_h(r);
413 // col 0 ~ 7
414 t0 = cnst_para1 * t1 + cnst_para2 * t2 - cnst_para0 * t0 - cnst_para3 * t3;
415 t1 = cnst_para1 * t2 + cnst_para2 * t3 - cnst_para0 * t1 - cnst_para3 * t4;
416 t2 = cnst_para1 * t3 + cnst_para2 * t4 - cnst_para0 * t2 - cnst_para3 * t5;
417 t3 = cnst_para1 * t4 + cnst_para2 * t5 - cnst_para0 * t3 - cnst_para3 * t6;
418 t4 = cnst_para1 * t5 + cnst_para2 * t6 - cnst_para0 * t4 - cnst_para3 * t7;
419 t5 = cnst_para1 * t6 + cnst_para2 * t7 - cnst_para0 * t5 - cnst_para3 * t8;
420 t6 = cnst_para1 * t7 + cnst_para2 * t8 - cnst_para0 * t6 - cnst_para3 * t9;
421 t7 = cnst_para1 * t8 + cnst_para2 * t9 - cnst_para0 * t7 - cnst_para3 * t10;
422 ADD4(t0, cnst_r, t1, cnst_r, t2, cnst_r, t3, cnst_r, t0, t1, t2, t3);
423 ADD4(t4, cnst_r, t5, cnst_r, t6, cnst_r, t7, cnst_r, t4, t5, t6, t7);
424 t0 >>= 7, t1 >>= 7, t2 >>= 7, t3 >>= 7;
425 t4 >>= 7, t5 >>= 7, t6 >>= 7, t7 >>= 7;
426 TRANSPOSE8x8_SH_SH(t0, t1, t2, t3, t4, t5, t6, t7,
427 t0, t1, t2, t3, t4, t5, t6, t7);
428 CLIP_SH8_0_255(t0, t1, t2, t3, t4, t5, t6, t7);
429 PCKEV_B4_SH(t1, t0, t3, t2, t5, t4, t7, t6, t0, t1, t2, t3);
430 ST_D8(t0, t1, t2, t3, 0, 1, 0, 1, 0, 1, 0, 1, dst, stride);
431 }
432
433 #define PUT_VC1_MSPEL_MC_MSA(hmode, vmode) \
434 void ff_put_vc1_mspel_mc ## hmode ## vmode ## _msa(uint8_t *dst, \
435 const uint8_t *src, \
436 ptrdiff_t stride, int rnd) \
437 { \
438 put_vc1_mspel_mc_h_v_msa(dst, src, stride, hmode, vmode, rnd); \
439 } \
440 void ff_put_vc1_mspel_mc ## hmode ## vmode ## _16_msa(uint8_t *dst, \
441 const uint8_t *src, \
442 ptrdiff_t stride, int rnd) \
443 { \
444 put_vc1_mspel_mc_h_v_msa(dst, src, stride, hmode, vmode, rnd); \
445 put_vc1_mspel_mc_h_v_msa(dst + 8, src + 8, stride, hmode, vmode, rnd); \
446 dst += 8 * stride, src += 8 * stride; \
447 put_vc1_mspel_mc_h_v_msa(dst, src, stride, hmode, vmode, rnd); \
448 put_vc1_mspel_mc_h_v_msa(dst + 8, src + 8, stride, hmode, vmode, rnd); \
449 }
450
451 PUT_VC1_MSPEL_MC_MSA(1, 1);
452 PUT_VC1_MSPEL_MC_MSA(1, 2);
453 PUT_VC1_MSPEL_MC_MSA(1, 3);
454
455 PUT_VC1_MSPEL_MC_MSA(2, 1);
456 PUT_VC1_MSPEL_MC_MSA(2, 2);
457 PUT_VC1_MSPEL_MC_MSA(2, 3);
458
459 PUT_VC1_MSPEL_MC_MSA(3, 1);
460 PUT_VC1_MSPEL_MC_MSA(3, 2);
461 PUT_VC1_MSPEL_MC_MSA(3, 3);
462