1 /*
2 * Copyright (c) 2021 Loongson Technology Corporation Limited
3 * Contributed by Hecai Yuan <yuanhecai@loongson.cn>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "libavcodec/vp8dsp.h"
23 #include "vp8dsp_loongarch.h"
24 #include "libavutil/loongarch/loongson_intrinsics.h"
25
26 #define VP8_LPF_FILTER4_4W(p1_in_out, p0_in_out, q0_in_out, q1_in_out, \
27 mask_in, hev_in) \
28 { \
29 __m128i p1_m, p0_m, q0_m, q1_m, q0_sub_p0, filt_sign; \
30 __m128i filt, filt1, filt2, cnst4b, cnst3b; \
31 __m128i q0_sub_p0_l, q0_sub_p0_h, filt_h, filt_l, cnst3h; \
32 \
33 p1_m = __lsx_vxori_b(p1_in_out, 0x80); \
34 p0_m = __lsx_vxori_b(p0_in_out, 0x80); \
35 q0_m = __lsx_vxori_b(q0_in_out, 0x80); \
36 q1_m = __lsx_vxori_b(q1_in_out, 0x80); \
37 filt = __lsx_vssub_b(p1_m, q1_m); \
38 filt = filt & hev_in; \
39 \
40 q0_sub_p0 = __lsx_vsub_b(q0_m, p0_m); \
41 filt_sign = __lsx_vslti_b(filt, 0); \
42 \
43 cnst3h = __lsx_vreplgr2vr_h(3); \
44 q0_sub_p0_l = __lsx_vilvl_b(q0_sub_p0, q0_sub_p0); \
45 q0_sub_p0_l = __lsx_vdp2_h_b(q0_sub_p0_l, cnst3h); \
46 filt_l = __lsx_vilvl_b(filt_sign, filt); \
47 filt_l = __lsx_vadd_h(filt_l, q0_sub_p0_l); \
48 filt_l = __lsx_vsat_h(filt_l, 7); \
49 \
50 q0_sub_p0_h = __lsx_vilvh_b(q0_sub_p0, q0_sub_p0); \
51 q0_sub_p0_h = __lsx_vdp2_h_b(q0_sub_p0_h, cnst3h); \
52 filt_h = __lsx_vilvh_b(filt_sign, filt); \
53 filt_h = __lsx_vadd_h(filt_h, q0_sub_p0_h); \
54 filt_h = __lsx_vsat_h(filt_h, 7); \
55 \
56 filt = __lsx_vpickev_b(filt_h, filt_l); \
57 filt = filt & mask_in; \
58 cnst4b = __lsx_vreplgr2vr_b(4); \
59 filt1 = __lsx_vsadd_b(filt, cnst4b); \
60 filt1 = __lsx_vsrai_b(filt1, 3); \
61 \
62 cnst3b = __lsx_vreplgr2vr_b(3); \
63 filt2 = __lsx_vsadd_b(filt, cnst3b); \
64 filt2 = __lsx_vsrai_b(filt2, 3); \
65 \
66 q0_m = __lsx_vssub_b(q0_m, filt1); \
67 q0_in_out = __lsx_vxori_b(q0_m, 0x80); \
68 p0_m = __lsx_vsadd_b(p0_m, filt2); \
69 p0_in_out = __lsx_vxori_b(p0_m, 0x80); \
70 \
71 filt = __lsx_vsrari_b(filt1, 1); \
72 hev_in = __lsx_vxori_b(hev_in, 0xff); \
73 filt = filt & hev_in; \
74 \
75 q1_m = __lsx_vssub_b(q1_m, filt); \
76 q1_in_out = __lsx_vxori_b(q1_m, 0x80); \
77 p1_m = __lsx_vsadd_b(p1_m, filt); \
78 p1_in_out = __lsx_vxori_b(p1_m, 0x80); \
79 }
80
81 #define VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev) \
82 { \
83 __m128i p2_m, p1_m, p0_m, q2_m, q1_m, q0_m; \
84 __m128i filt, q0_sub_p0, cnst4b, cnst3b; \
85 __m128i u, filt1, filt2, filt_sign, q0_sub_p0_sign; \
86 __m128i q0_sub_p0_l, q0_sub_p0_h, filt_l, u_l, u_h, filt_h; \
87 __m128i cnst3h, cnst27h, cnst18h, cnst63h; \
88 \
89 cnst3h = __lsx_vreplgr2vr_h(3); \
90 \
91 p2_m = __lsx_vxori_b(p2, 0x80); \
92 p1_m = __lsx_vxori_b(p1, 0x80); \
93 p0_m = __lsx_vxori_b(p0, 0x80); \
94 q0_m = __lsx_vxori_b(q0, 0x80); \
95 q1_m = __lsx_vxori_b(q1, 0x80); \
96 q2_m = __lsx_vxori_b(q2, 0x80); \
97 \
98 filt = __lsx_vssub_b(p1_m, q1_m); \
99 q0_sub_p0 = __lsx_vsub_b(q0_m, p0_m); \
100 q0_sub_p0_sign = __lsx_vslti_b(q0_sub_p0, 0); \
101 filt_sign = __lsx_vslti_b(filt, 0); \
102 \
103 /* right part */ \
104 q0_sub_p0_l = __lsx_vilvl_b(q0_sub_p0_sign, q0_sub_p0); \
105 q0_sub_p0_l = __lsx_vmul_h(q0_sub_p0_l, cnst3h); \
106 filt_l = __lsx_vilvl_b(filt_sign, filt); \
107 filt_l = __lsx_vadd_h(filt_l, q0_sub_p0_l); \
108 filt_l = __lsx_vsat_h(filt_l, 7); \
109 \
110 /* left part */ \
111 q0_sub_p0_h = __lsx_vilvh_b(q0_sub_p0_sign, q0_sub_p0); \
112 q0_sub_p0_h = __lsx_vmul_h(q0_sub_p0_h, cnst3h); \
113 filt_h = __lsx_vilvh_b(filt_sign, filt); \
114 filt_h = __lsx_vadd_h(filt_h, q0_sub_p0_h); \
115 filt_h = __lsx_vsat_h(filt_h, 7); \
116 \
117 /* combine left and right part */ \
118 filt = __lsx_vpickev_b(filt_h, filt_l); \
119 filt = filt & mask; \
120 filt2 = filt & hev; \
121 /* filt_val &= ~hev */ \
122 hev = __lsx_vxori_b(hev, 0xff); \
123 filt = filt & hev; \
124 cnst4b = __lsx_vreplgr2vr_b(4); \
125 filt1 = __lsx_vsadd_b(filt2, cnst4b); \
126 filt1 = __lsx_vsrai_b(filt1, 3); \
127 cnst3b = __lsx_vreplgr2vr_b(3); \
128 filt2 = __lsx_vsadd_b(filt2, cnst3b); \
129 filt2 = __lsx_vsrai_b(filt2, 3); \
130 q0_m = __lsx_vssub_b(q0_m, filt1); \
131 p0_m = __lsx_vsadd_b(p0_m, filt2); \
132 \
133 filt_sign = __lsx_vslti_b(filt, 0); \
134 filt_l = __lsx_vilvl_b(filt_sign, filt); \
135 filt_h = __lsx_vilvh_b(filt_sign, filt); \
136 \
137 cnst27h = __lsx_vreplgr2vr_h(27); \
138 cnst63h = __lsx_vreplgr2vr_h(63); \
139 \
140 /* right part */ \
141 u_l = __lsx_vmul_h(filt_l, cnst27h); \
142 u_l = __lsx_vadd_h(u_l, cnst63h); \
143 u_l = __lsx_vsrai_h(u_l, 7); \
144 u_l = __lsx_vsat_h(u_l, 7); \
145 /* left part */ \
146 u_h = __lsx_vmul_h(filt_h, cnst27h); \
147 u_h = __lsx_vadd_h(u_h, cnst63h); \
148 u_h = __lsx_vsrai_h(u_h, 7); \
149 u_h = __lsx_vsat_h(u_h, 7); \
150 /* combine left and right part */ \
151 u = __lsx_vpickev_b(u_h, u_l); \
152 q0_m = __lsx_vssub_b(q0_m, u); \
153 q0 = __lsx_vxori_b(q0_m, 0x80); \
154 p0_m = __lsx_vsadd_b(p0_m, u); \
155 p0 = __lsx_vxori_b(p0_m, 0x80); \
156 cnst18h = __lsx_vreplgr2vr_h(18); \
157 u_l = __lsx_vmul_h(filt_l, cnst18h); \
158 u_l = __lsx_vadd_h(u_l, cnst63h); \
159 u_l = __lsx_vsrai_h(u_l, 7); \
160 u_l = __lsx_vsat_h(u_l, 7); \
161 \
162 /* left part */ \
163 u_h = __lsx_vmul_h(filt_h, cnst18h); \
164 u_h = __lsx_vadd_h(u_h, cnst63h); \
165 u_h = __lsx_vsrai_h(u_h, 7); \
166 u_h = __lsx_vsat_h(u_h, 7); \
167 /* combine left and right part */ \
168 u = __lsx_vpickev_b(u_h, u_l); \
169 q1_m = __lsx_vssub_b(q1_m, u); \
170 q1 = __lsx_vxori_b(q1_m, 0x80); \
171 p1_m = __lsx_vsadd_b(p1_m, u); \
172 p1 = __lsx_vxori_b(p1_m, 0x80); \
173 u_l = __lsx_vslli_h(filt_l, 3); \
174 u_l = __lsx_vadd_h(u_l, filt_l); \
175 u_l = __lsx_vadd_h(u_l, cnst63h); \
176 u_l = __lsx_vsrai_h(u_l, 7); \
177 u_l = __lsx_vsat_h(u_l, 7); \
178 \
179 /* left part */ \
180 u_h = __lsx_vslli_h(filt_h, 3); \
181 u_h = __lsx_vadd_h(u_h, filt_h); \
182 u_h = __lsx_vadd_h(u_h, cnst63h); \
183 u_h = __lsx_vsrai_h(u_h, 7); \
184 u_h = __lsx_vsat_h(u_h, 7); \
185 /* combine left and right part */ \
186 u = __lsx_vpickev_b(u_h, u_l); \
187 q2_m = __lsx_vssub_b(q2_m, u); \
188 q2 = __lsx_vxori_b(q2_m, 0x80); \
189 p2_m = __lsx_vsadd_b(p2_m, u); \
190 p2 = __lsx_vxori_b(p2_m, 0x80); \
191 }
192
193 #define LPF_MASK_HEV(p3_src, p2_src, p1_src, p0_src, \
194 q0_src, q1_src, q2_src, q3_src, \
195 limit_src, b_limit_src, thresh_src, \
196 hev_dst, mask_dst, flat_dst) \
197 { \
198 __m128i p3_asub_p2_m, p2_asub_p1_m, p1_asub_p0_m, q1_asub_q0_m; \
199 __m128i p1_asub_q1_m, p0_asub_q0_m, q3_asub_q2_m, q2_asub_q1_m; \
200 \
201 /* absolute subtraction of pixel values */ \
202 p3_asub_p2_m = __lsx_vabsd_bu(p3_src, p2_src); \
203 p2_asub_p1_m = __lsx_vabsd_bu(p2_src, p1_src); \
204 p1_asub_p0_m = __lsx_vabsd_bu(p1_src, p0_src); \
205 q1_asub_q0_m = __lsx_vabsd_bu(q1_src, q0_src); \
206 q2_asub_q1_m = __lsx_vabsd_bu(q2_src, q1_src); \
207 q3_asub_q2_m = __lsx_vabsd_bu(q3_src, q2_src); \
208 p0_asub_q0_m = __lsx_vabsd_bu(p0_src, q0_src); \
209 p1_asub_q1_m = __lsx_vabsd_bu(p1_src, q1_src); \
210 \
211 /* calculation of hev */ \
212 flat_dst = __lsx_vmax_bu(p1_asub_p0_m, q1_asub_q0_m); \
213 hev_dst = __lsx_vslt_bu(thresh_src, flat_dst); \
214 /* calculation of mask */ \
215 p0_asub_q0_m = __lsx_vsadd_bu(p0_asub_q0_m, p0_asub_q0_m); \
216 p1_asub_q1_m = __lsx_vsrli_b(p1_asub_q1_m, 1); \
217 p0_asub_q0_m = __lsx_vsadd_bu(p0_asub_q0_m, p1_asub_q1_m); \
218 mask_dst = __lsx_vslt_bu(b_limit_src, p0_asub_q0_m); \
219 mask_dst = __lsx_vmax_bu(flat_dst, mask_dst); \
220 p3_asub_p2_m = __lsx_vmax_bu(p3_asub_p2_m, p2_asub_p1_m); \
221 mask_dst = __lsx_vmax_bu(p3_asub_p2_m, mask_dst); \
222 q2_asub_q1_m = __lsx_vmax_bu(q2_asub_q1_m, q3_asub_q2_m); \
223 mask_dst = __lsx_vmax_bu(q2_asub_q1_m, mask_dst); \
224 mask_dst = __lsx_vslt_bu(limit_src, mask_dst); \
225 mask_dst = __lsx_vxori_b(mask_dst, 0xff); \
226 }
227
228 #define VP8_ST6x1_UB(in0, in0_idx, in1, in1_idx, pdst, stride) \
229 { \
230 __lsx_vstelm_w(in0, pdst, 0, in0_idx); \
231 __lsx_vstelm_h(in1, pdst + stride, 0, in1_idx); \
232 }
233
234 #define ST_W4(in, idx0, idx1, idx2, idx3, pdst, stride) \
235 { \
236 __lsx_vstelm_w(in, pdst, 0, idx0); \
237 pdst += stride; \
238 __lsx_vstelm_w(in, pdst, 0, idx1); \
239 pdst += stride; \
240 __lsx_vstelm_w(in, pdst, 0, idx2); \
241 pdst += stride; \
242 __lsx_vstelm_w(in, pdst, 0, idx3); \
243 pdst += stride; \
244 }
245
ff_vp8_v_loop_filter16_lsx(uint8_t * dst,ptrdiff_t stride,int b_limit_in,int limit_in,int thresh_in)246 void ff_vp8_v_loop_filter16_lsx(uint8_t *dst, ptrdiff_t stride, int b_limit_in,
247 int limit_in, int thresh_in)
248 {
249 __m128i p3, p2, p1, p0, q3, q2, q1, q0;
250 __m128i mask, hev, flat, thresh, limit, b_limit;
251
252 ptrdiff_t stride2 = stride << 1;
253 ptrdiff_t stride3 = stride2 + stride;
254 ptrdiff_t stride4 = stride2 << 1;
255
256 b_limit = __lsx_vreplgr2vr_b(b_limit_in);
257 limit = __lsx_vreplgr2vr_b(limit_in);
258 thresh = __lsx_vreplgr2vr_b(thresh_in);
259
260 /*load vector elements*/
261 DUP4_ARG2(__lsx_vld, dst - stride4, 0, dst - stride3, 0, dst - stride2, 0,
262 dst - stride, 0, p3, p2, p1, p0);
263 DUP4_ARG2(__lsx_vld, dst, 0, dst + stride, 0, dst + stride2, 0, dst + stride3, 0,
264 q0, q1, q2, q3);
265 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, mask, flat);
266 VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
267
268 /*store vector elements*/
269 __lsx_vst(p2, dst - stride3, 0);
270 __lsx_vst(p1, dst - stride2, 0);
271 __lsx_vst(p0, dst - stride, 0);
272 __lsx_vst(q0, dst, 0);
273
274 __lsx_vst(q1, dst + stride, 0);
275 __lsx_vst(q2, dst + stride2, 0);
276 }
277
ff_vp8_v_loop_filter8uv_lsx(uint8_t * dst_u,uint8_t * dst_v,ptrdiff_t stride,int b_limit_in,int limit_in,int thresh_in)278 void ff_vp8_v_loop_filter8uv_lsx(uint8_t *dst_u, uint8_t *dst_v,
279 ptrdiff_t stride, int b_limit_in,
280 int limit_in, int thresh_in)
281 {
282 __m128i p3, p2, p1, p0, q3, q2, q1, q0;
283 __m128i mask, hev, flat, thresh, limit, b_limit;
284 __m128i p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u;
285 __m128i p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v;
286
287 ptrdiff_t stride2 = stride << 1;
288 ptrdiff_t stride3 = stride2 + stride;
289 ptrdiff_t stride4 = stride2 << 1;
290
291 b_limit = __lsx_vreplgr2vr_b(b_limit_in);
292 limit = __lsx_vreplgr2vr_b(limit_in);
293 thresh = __lsx_vreplgr2vr_b(thresh_in);
294
295 DUP4_ARG2(__lsx_vld, dst_u - stride4, 0, dst_u - stride3, 0, dst_u - stride2, 0,
296 dst_u - stride, 0, p3_u, p2_u, p1_u, p0_u);
297 DUP4_ARG2(__lsx_vld, dst_u, 0, dst_u + stride, 0, dst_u + stride2, 0,
298 dst_u + stride3, 0, q0_u, q1_u, q2_u, q3_u);
299
300 DUP4_ARG2(__lsx_vld, dst_v - stride4, 0, dst_v - stride3, 0, dst_v - stride2, 0,
301 dst_v - stride, 0, p3_v, p2_v, p1_v, p0_v);
302 DUP4_ARG2(__lsx_vld, dst_v, 0, dst_v + stride, 0, dst_v + stride2, 0,
303 dst_v + stride3, 0, q0_v, q1_v, q2_v, q3_v);
304
305 /* rht 8 element of p3 are u pixel and left 8 element of p3 are v pixei */
306 DUP4_ARG2(__lsx_vilvl_d, p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, p2, p1, p0);
307 DUP4_ARG2(__lsx_vilvl_d, q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, q0, q1, q2, q3);
308 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
309 hev, mask, flat);
310 VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
311
312 __lsx_vstelm_d(p2, dst_u - stride3, 0, 0);
313 __lsx_vstelm_d(p1, dst_u - stride2, 0, 0);
314 __lsx_vstelm_d(p0, dst_u - stride , 0, 0);
315 __lsx_vstelm_d(q0, dst_u, 0, 0);
316
317 __lsx_vstelm_d(q1, dst_u + stride, 0, 0);
318 __lsx_vstelm_d(q2, dst_u + stride2, 0, 0);
319
320 __lsx_vstelm_d(p2, dst_v - stride3, 0, 1);
321 __lsx_vstelm_d(p1, dst_v - stride2, 0, 1);
322 __lsx_vstelm_d(p0, dst_v - stride , 0, 1);
323 __lsx_vstelm_d(q0, dst_v, 0, 1);
324
325 __lsx_vstelm_d(q1, dst_v + stride, 0, 1);
326 __lsx_vstelm_d(q2, dst_v + stride2, 0, 1);
327 }
328
ff_vp8_h_loop_filter16_lsx(uint8_t * dst,ptrdiff_t stride,int b_limit_in,int limit_in,int thresh_in)329 void ff_vp8_h_loop_filter16_lsx(uint8_t *dst, ptrdiff_t stride, int b_limit_in,
330 int limit_in, int thresh_in)
331 {
332 uint8_t *temp_src;
333 __m128i p3, p2, p1, p0, q3, q2, q1, q0;
334 __m128i mask, hev, flat, thresh, limit, b_limit;
335 __m128i row0, row1, row2, row3, row4, row5, row6, row7, row8;
336 __m128i row9, row10, row11, row12, row13, row14, row15;
337 __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
338
339 ptrdiff_t stride2 = stride << 1;
340 ptrdiff_t stride3 = stride2 + stride;
341 ptrdiff_t stride4 = stride2 << 1;
342
343 b_limit = __lsx_vreplgr2vr_b(b_limit_in);
344 limit = __lsx_vreplgr2vr_b(limit_in);
345 thresh = __lsx_vreplgr2vr_b(thresh_in);
346
347 temp_src = dst - 4;
348 DUP4_ARG2(__lsx_vld, temp_src, 0, temp_src + stride, 0, temp_src + stride2, 0,
349 temp_src + stride3, 0, row0, row1, row2, row3);
350 temp_src += stride4;
351 DUP4_ARG2(__lsx_vld, temp_src, 0, temp_src + stride, 0, temp_src + stride2, 0,
352 temp_src + stride3, 0, row4, row5, row6, row7);
353
354 temp_src += stride4;
355 DUP4_ARG2(__lsx_vld, temp_src, 0, temp_src + stride, 0, temp_src + stride2, 0,
356 temp_src + stride3, 0, row8, row9, row10, row11);
357 temp_src += stride4;
358 DUP4_ARG2(__lsx_vld, temp_src, 0, temp_src + stride, 0, temp_src + stride2, 0,
359 temp_src + stride3, 0, row12, row13, row14, row15);
360 LSX_TRANSPOSE16x8_B(row0, row1, row2, row3, row4, row5, row6, row7, row8, row9, row10,
361 row11, row12, row13, row14, row15, p3, p2, p1, p0, q0, q1, q2, q3);
362
363 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, mask, flat);
364 VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
365
366 tmp0 = __lsx_vilvl_b(p1, p2);
367 tmp1 = __lsx_vilvl_b(q0, p0);
368
369 tmp3 = __lsx_vilvl_h(tmp1, tmp0);
370 tmp4 = __lsx_vilvh_h(tmp1, tmp0);
371
372 tmp0 = __lsx_vilvh_b(p1, p2);
373 tmp1 = __lsx_vilvh_b(q0, p0);
374
375 tmp6 = __lsx_vilvl_h(tmp1, tmp0);
376 tmp7 = __lsx_vilvh_h(tmp1, tmp0);
377
378 tmp2 = __lsx_vilvl_b(q2, q1);
379 tmp5 = __lsx_vilvh_b(q2, q1);
380
381 temp_src = dst - 3;
382 VP8_ST6x1_UB(tmp3, 0, tmp2, 0, temp_src, 4);
383 temp_src += stride;
384 VP8_ST6x1_UB(tmp3, 1, tmp2, 1, temp_src, 4);
385 temp_src += stride;
386 VP8_ST6x1_UB(tmp3, 2, tmp2, 2, temp_src, 4);
387 temp_src += stride;
388 VP8_ST6x1_UB(tmp3, 3, tmp2, 3, temp_src, 4);
389 temp_src += stride;
390 VP8_ST6x1_UB(tmp4, 0, tmp2, 4, temp_src, 4);
391 temp_src += stride;
392 VP8_ST6x1_UB(tmp4, 1, tmp2, 5, temp_src, 4);
393 temp_src += stride;
394 VP8_ST6x1_UB(tmp4, 2, tmp2, 6, temp_src, 4);
395 temp_src += stride;
396 VP8_ST6x1_UB(tmp4, 3, tmp2, 7, temp_src, 4);
397 temp_src += stride;
398 VP8_ST6x1_UB(tmp6, 0, tmp5, 0, temp_src, 4);
399 temp_src += stride;
400 VP8_ST6x1_UB(tmp6, 1, tmp5, 1, temp_src, 4);
401 temp_src += stride;
402 VP8_ST6x1_UB(tmp6, 2, tmp5, 2, temp_src, 4);
403 temp_src += stride;
404 VP8_ST6x1_UB(tmp6, 3, tmp5, 3, temp_src, 4);
405 temp_src += stride;
406 VP8_ST6x1_UB(tmp7, 0, tmp5, 4, temp_src, 4);
407 temp_src += stride;
408 VP8_ST6x1_UB(tmp7, 1, tmp5, 5, temp_src, 4);
409 temp_src += stride;
410 VP8_ST6x1_UB(tmp7, 2, tmp5, 6, temp_src, 4);
411 temp_src += stride;
412 VP8_ST6x1_UB(tmp7, 3, tmp5, 7, temp_src, 4);
413 }
414
ff_vp8_h_loop_filter8uv_lsx(uint8_t * dst_u,uint8_t * dst_v,ptrdiff_t stride,int b_limit_in,int limit_in,int thresh_in)415 void ff_vp8_h_loop_filter8uv_lsx(uint8_t *dst_u, uint8_t *dst_v,
416 ptrdiff_t stride, int b_limit_in,
417 int limit_in, int thresh_in)
418 {
419 uint8_t *temp_src;
420 __m128i p3, p2, p1, p0, q3, q2, q1, q0;
421 __m128i mask, hev, flat, thresh, limit, b_limit;
422 __m128i row0, row1, row2, row3, row4, row5, row6, row7, row8;
423 __m128i row9, row10, row11, row12, row13, row14, row15;
424 __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
425
426 ptrdiff_t stride2 = stride << 1;
427 ptrdiff_t stride3 = stride2 + stride;
428 ptrdiff_t stride4 = stride2 << 1;
429
430 b_limit = __lsx_vreplgr2vr_b(b_limit_in);
431 limit = __lsx_vreplgr2vr_b(limit_in);
432 thresh = __lsx_vreplgr2vr_b(thresh_in);
433
434 temp_src = dst_u - 4;
435 DUP4_ARG2(__lsx_vld, temp_src, 0, temp_src + stride, 0, temp_src + stride2, 0,
436 temp_src + stride3, 0, row0, row1, row2, row3);
437 temp_src += stride4;
438 DUP4_ARG2(__lsx_vld, temp_src, 0, temp_src + stride, 0, temp_src + stride2, 0,
439 temp_src + stride3, 0, row4, row5, row6, row7);
440
441 temp_src = dst_v - 4;
442 DUP4_ARG2(__lsx_vld, temp_src, 0, temp_src + stride, 0, temp_src + stride2, 0,
443 temp_src + stride3, 0, row8, row9, row10, row11);
444 temp_src += stride4;
445 DUP4_ARG2(__lsx_vld, temp_src, 0, temp_src + stride, 0, temp_src + stride2, 0,
446 temp_src + stride3, 0, row12, row13, row14, row15);
447
448 LSX_TRANSPOSE16x8_B(row0, row1, row2, row3, row4, row5, row6, row7,
449 row8, row9, row10, row11, row12, row13, row14, row15,
450 p3, p2, p1, p0, q0, q1, q2, q3);
451
452 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, mask, flat);
453 VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
454
455 tmp0 = __lsx_vilvl_b(p1, p2);
456 tmp1 = __lsx_vilvl_b(q0, p0);
457
458 tmp3 = __lsx_vilvl_h(tmp1, tmp0);
459 tmp4 = __lsx_vilvh_h(tmp1, tmp0);
460
461 tmp0 = __lsx_vilvh_b(p1, p2);
462 tmp1 = __lsx_vilvh_b(q0, p0);
463
464 tmp6 = __lsx_vilvl_h(tmp1, tmp0);
465 tmp7 = __lsx_vilvh_h(tmp1, tmp0);
466
467 tmp2 = __lsx_vilvl_b(q2, q1);
468 tmp5 = __lsx_vilvh_b(q2, q1);
469
470 dst_u -= 3;
471 VP8_ST6x1_UB(tmp3, 0, tmp2, 0, dst_u, 4);
472 dst_u += stride;
473 VP8_ST6x1_UB(tmp3, 1, tmp2, 1, dst_u, 4);
474 dst_u += stride;
475 VP8_ST6x1_UB(tmp3, 2, tmp2, 2, dst_u, 4);
476 dst_u += stride;
477 VP8_ST6x1_UB(tmp3, 3, tmp2, 3, dst_u, 4);
478 dst_u += stride;
479 VP8_ST6x1_UB(tmp4, 0, tmp2, 4, dst_u, 4);
480 dst_u += stride;
481 VP8_ST6x1_UB(tmp4, 1, tmp2, 5, dst_u, 4);
482 dst_u += stride;
483 VP8_ST6x1_UB(tmp4, 2, tmp2, 6, dst_u, 4);
484 dst_u += stride;
485 VP8_ST6x1_UB(tmp4, 3, tmp2, 7, dst_u, 4);
486
487 dst_v -= 3;
488 VP8_ST6x1_UB(tmp6, 0, tmp5, 0, dst_v, 4);
489 dst_v += stride;
490 VP8_ST6x1_UB(tmp6, 1, tmp5, 1, dst_v, 4);
491 dst_v += stride;
492 VP8_ST6x1_UB(tmp6, 2, tmp5, 2, dst_v, 4);
493 dst_v += stride;
494 VP8_ST6x1_UB(tmp6, 3, tmp5, 3, dst_v, 4);
495 dst_v += stride;
496 VP8_ST6x1_UB(tmp7, 0, tmp5, 4, dst_v, 4);
497 dst_v += stride;
498 VP8_ST6x1_UB(tmp7, 1, tmp5, 5, dst_v, 4);
499 dst_v += stride;
500 VP8_ST6x1_UB(tmp7, 2, tmp5, 6, dst_v, 4);
501 dst_v += stride;
502 VP8_ST6x1_UB(tmp7, 3, tmp5, 7, dst_v, 4);
503 }
504
ff_vp8_v_loop_filter16_inner_lsx(uint8_t * src,ptrdiff_t stride,int32_t e,int32_t i,int32_t h)505 void ff_vp8_v_loop_filter16_inner_lsx(uint8_t *src, ptrdiff_t stride,
506 int32_t e, int32_t i, int32_t h)
507 {
508 __m128i mask, hev, flat;
509 __m128i thresh, b_limit, limit;
510 __m128i p3, p2, p1, p0, q3, q2, q1, q0;
511
512 ptrdiff_t stride2 = stride << 1;
513 ptrdiff_t stride3 = stride2 + stride;
514 ptrdiff_t stride4 = stride2 << 1;
515
516 /* load vector elements */
517 src -= stride4;
518 DUP4_ARG2(__lsx_vld, src, 0, src + stride, 0, src + stride2, 0,
519 src + stride3, 0, p3, p2, p1, p0);
520 src += stride4;
521 DUP4_ARG2(__lsx_vld, src, 0, src + stride, 0, src + stride2, 0,
522 src + stride3, 0, q0, q1, q2, q3);
523 thresh = __lsx_vreplgr2vr_b(h);
524 b_limit = __lsx_vreplgr2vr_b(e);
525 limit = __lsx_vreplgr2vr_b(i);
526
527 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
528 hev, mask, flat);
529 VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
530
531 __lsx_vst(p1, src - stride2, 0);
532 __lsx_vst(p0, src - stride, 0);
533 __lsx_vst(q0, src, 0);
534 __lsx_vst(q1, src + stride, 0);
535 }
536
ff_vp8_h_loop_filter16_inner_lsx(uint8_t * src,ptrdiff_t stride,int32_t e,int32_t i,int32_t h)537 void ff_vp8_h_loop_filter16_inner_lsx(uint8_t *src, ptrdiff_t stride,
538 int32_t e, int32_t i, int32_t h)
539 {
540 __m128i mask, hev, flat;
541 __m128i thresh, b_limit, limit;
542 __m128i p3, p2, p1, p0, q3, q2, q1, q0;
543 __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
544 __m128i tmp8, tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
545
546 ptrdiff_t stride2 = stride << 1;
547 ptrdiff_t stride3 = stride2 + stride;
548 ptrdiff_t stride4 = stride2 << 1;
549
550 src -= 4;
551 DUP4_ARG2(__lsx_vld, src, 0, src + stride, 0, src + stride2, 0,
552 src + stride3, 0, tmp0, tmp1, tmp2, tmp3);
553 src += stride4;
554 DUP4_ARG2(__lsx_vld, src, 0, src + stride, 0, src + stride2, 0,
555 src + stride3, 0, tmp4, tmp5, tmp6, tmp7);
556 src += stride4;
557 DUP4_ARG2(__lsx_vld, src, 0, src + stride, 0, src + stride2, 0,
558 src + stride3, 0, tmp8, tmp9, tmp10, tmp11);
559 src += stride4;
560 DUP4_ARG2(__lsx_vld, src, 0, src + stride, 0, src + stride2, 0,
561 src + stride3, 0, tmp12, tmp13, tmp14, tmp15);
562 src -= 3 * stride4;
563
564 LSX_TRANSPOSE16x8_B(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
565 tmp8, tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15,
566 p3, p2, p1, p0, q0, q1, q2, q3);
567
568 thresh = __lsx_vreplgr2vr_b(h);
569 b_limit = __lsx_vreplgr2vr_b(e);
570 limit = __lsx_vreplgr2vr_b(i);
571
572 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
573 hev, mask, flat);
574 VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
575
576 DUP2_ARG2(__lsx_vilvl_b, p0, p1, q1, q0, tmp0, tmp1);
577 tmp2 = __lsx_vilvl_h(tmp1, tmp0);
578 tmp3 = __lsx_vilvh_h(tmp1, tmp0);
579
580 src += 2;
581 ST_W4(tmp2, 0, 1, 2, 3, src, stride);
582 ST_W4(tmp3, 0, 1, 2, 3, src, stride);
583
584 DUP2_ARG2(__lsx_vilvh_b, p0, p1, q1, q0, tmp0, tmp1);
585 tmp2 = __lsx_vilvl_h(tmp1, tmp0);
586 tmp3 = __lsx_vilvh_h(tmp1, tmp0);
587
588 ST_W4(tmp2, 0, 1, 2, 3, src, stride);
589 ST_W4(tmp3, 0, 1, 2, 3, src, stride);
590 src -= 4 * stride4;
591 }
592