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 "./vp8_rtcd.h"
12 #include "vp8/common/loopfilter.h"
13 #include "vp8/common/mips/msa/vp8_macros_msa.h"
14
15 #define VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask) \
16 { \
17 v16u8 p1_a_sub_q1, p0_a_sub_q0; \
18 \
19 p0_a_sub_q0 = __msa_asub_u_b(p0, q0); \
20 p1_a_sub_q1 = __msa_asub_u_b(p1, q1); \
21 p1_a_sub_q1 = (v16u8)__msa_srli_b((v16i8)p1_a_sub_q1, 1); \
22 p0_a_sub_q0 = __msa_adds_u_b(p0_a_sub_q0, p0_a_sub_q0); \
23 mask = __msa_adds_u_b(p0_a_sub_q0, p1_a_sub_q1); \
24 mask = ((v16u8)mask <= b_limit); \
25 }
26
27 #define VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev) \
28 { \
29 v16i8 p1_m, p0_m, q0_m, q1_m, filt, q0_sub_p0, t1, t2; \
30 const v16i8 cnst4b = __msa_ldi_b(4); \
31 const v16i8 cnst3b = __msa_ldi_b(3); \
32 \
33 p1_m = (v16i8)__msa_xori_b(p1, 0x80); \
34 p0_m = (v16i8)__msa_xori_b(p0, 0x80); \
35 q0_m = (v16i8)__msa_xori_b(q0, 0x80); \
36 q1_m = (v16i8)__msa_xori_b(q1, 0x80); \
37 \
38 filt = __msa_subs_s_b(p1_m, q1_m); \
39 filt &= hev; \
40 q0_sub_p0 = __msa_subs_s_b(q0_m, p0_m); \
41 filt = __msa_adds_s_b(filt, q0_sub_p0); \
42 filt = __msa_adds_s_b(filt, q0_sub_p0); \
43 filt = __msa_adds_s_b(filt, q0_sub_p0); \
44 filt &= mask; \
45 t1 = __msa_adds_s_b(filt, cnst4b); \
46 t1 >>= cnst3b; \
47 t2 = __msa_adds_s_b(filt, cnst3b); \
48 t2 >>= cnst3b; \
49 q0_m = __msa_subs_s_b(q0_m, t1); \
50 q0 = __msa_xori_b((v16u8)q0_m, 0x80); \
51 p0_m = __msa_adds_s_b(p0_m, t2); \
52 p0 = __msa_xori_b((v16u8)p0_m, 0x80); \
53 filt = __msa_srari_b(t1, 1); \
54 hev = __msa_xori_b(hev, 0xff); \
55 filt &= hev; \
56 q1_m = __msa_subs_s_b(q1_m, filt); \
57 q1 = __msa_xori_b((v16u8)q1_m, 0x80); \
58 p1_m = __msa_adds_s_b(p1_m, filt); \
59 p1 = __msa_xori_b((v16u8)p1_m, 0x80); \
60 }
61
62 #define VP8_SIMPLE_FILT(p1_in, p0_in, q0_in, q1_in, mask) \
63 { \
64 v16i8 p1_m, p0_m, q0_m, q1_m, filt, filt1, filt2; \
65 v16i8 q0_sub_p0; \
66 const v16i8 cnst4b = __msa_ldi_b(4); \
67 const v16i8 cnst3b = __msa_ldi_b(3); \
68 \
69 p1_m = (v16i8)__msa_xori_b(p1_in, 0x80); \
70 p0_m = (v16i8)__msa_xori_b(p0_in, 0x80); \
71 q0_m = (v16i8)__msa_xori_b(q0_in, 0x80); \
72 q1_m = (v16i8)__msa_xori_b(q1_in, 0x80); \
73 \
74 filt = __msa_subs_s_b(p1_m, q1_m); \
75 q0_sub_p0 = __msa_subs_s_b(q0_m, p0_m); \
76 filt = __msa_adds_s_b(filt, q0_sub_p0); \
77 filt = __msa_adds_s_b(filt, q0_sub_p0); \
78 filt = __msa_adds_s_b(filt, q0_sub_p0); \
79 filt &= mask; \
80 filt1 = __msa_adds_s_b(filt, cnst4b); \
81 filt1 >>= cnst3b; \
82 filt2 = __msa_adds_s_b(filt, cnst3b); \
83 filt2 >>= cnst3b; \
84 q0_m = __msa_subs_s_b(q0_m, filt1); \
85 p0_m = __msa_adds_s_b(p0_m, filt2); \
86 q0_in = __msa_xori_b((v16u8)q0_m, 0x80); \
87 p0_in = __msa_xori_b((v16u8)p0_m, 0x80); \
88 }
89
90 #define VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev) \
91 { \
92 v16i8 p2_m, p1_m, p0_m, q2_m, q1_m, q0_m; \
93 v16i8 u, filt, t1, t2, filt_sign, q0_sub_p0; \
94 v8i16 filt_r, filt_l, u_r, u_l; \
95 v8i16 temp0, temp1, temp2, temp3; \
96 const v16i8 cnst4b = __msa_ldi_b(4); \
97 const v16i8 cnst3b = __msa_ldi_b(3); \
98 const v8i16 cnst9h = __msa_ldi_h(9); \
99 const v8i16 cnst63h = __msa_ldi_h(63); \
100 \
101 p2_m = (v16i8)__msa_xori_b(p2, 0x80); \
102 p1_m = (v16i8)__msa_xori_b(p1, 0x80); \
103 p0_m = (v16i8)__msa_xori_b(p0, 0x80); \
104 q0_m = (v16i8)__msa_xori_b(q0, 0x80); \
105 q1_m = (v16i8)__msa_xori_b(q1, 0x80); \
106 q2_m = (v16i8)__msa_xori_b(q2, 0x80); \
107 \
108 filt = __msa_subs_s_b(p1_m, q1_m); \
109 q0_sub_p0 = __msa_subs_s_b(q0_m, p0_m); \
110 filt = __msa_adds_s_b(filt, q0_sub_p0); \
111 filt = __msa_adds_s_b(filt, q0_sub_p0); \
112 filt = __msa_adds_s_b(filt, q0_sub_p0); \
113 filt &= mask; \
114 \
115 t2 = filt & hev; \
116 hev = __msa_xori_b(hev, 0xff); \
117 filt &= hev; \
118 t1 = __msa_adds_s_b(t2, cnst4b); \
119 t1 >>= cnst3b; \
120 t2 = __msa_adds_s_b(t2, cnst3b); \
121 t2 >>= cnst3b; \
122 q0_m = __msa_subs_s_b(q0_m, t1); \
123 p0_m = __msa_adds_s_b(p0_m, t2); \
124 filt_sign = __msa_clti_s_b(filt, 0); \
125 ILVRL_B2_SH(filt_sign, filt, filt_r, filt_l); \
126 temp0 = filt_r * cnst9h; \
127 temp1 = temp0 + cnst63h; \
128 temp2 = filt_l * cnst9h; \
129 temp3 = temp2 + cnst63h; \
130 \
131 u_r = temp1 >> 7; \
132 u_r = __msa_sat_s_h(u_r, 7); \
133 u_l = temp3 >> 7; \
134 u_l = __msa_sat_s_h(u_l, 7); \
135 u = __msa_pckev_b((v16i8)u_l, (v16i8)u_r); \
136 q2_m = __msa_subs_s_b(q2_m, u); \
137 p2_m = __msa_adds_s_b(p2_m, u); \
138 q2 = __msa_xori_b((v16u8)q2_m, 0x80); \
139 p2 = __msa_xori_b((v16u8)p2_m, 0x80); \
140 \
141 temp1 += temp0; \
142 temp3 += temp2; \
143 \
144 u_r = temp1 >> 7; \
145 u_r = __msa_sat_s_h(u_r, 7); \
146 u_l = temp3 >> 7; \
147 u_l = __msa_sat_s_h(u_l, 7); \
148 u = __msa_pckev_b((v16i8)u_l, (v16i8)u_r); \
149 q1_m = __msa_subs_s_b(q1_m, u); \
150 p1_m = __msa_adds_s_b(p1_m, u); \
151 q1 = __msa_xori_b((v16u8)q1_m, 0x80); \
152 p1 = __msa_xori_b((v16u8)p1_m, 0x80); \
153 \
154 temp1 += temp0; \
155 temp3 += temp2; \
156 \
157 u_r = temp1 >> 7; \
158 u_r = __msa_sat_s_h(u_r, 7); \
159 u_l = temp3 >> 7; \
160 u_l = __msa_sat_s_h(u_l, 7); \
161 u = __msa_pckev_b((v16i8)u_l, (v16i8)u_r); \
162 q0_m = __msa_subs_s_b(q0_m, u); \
163 p0_m = __msa_adds_s_b(p0_m, u); \
164 q0 = __msa_xori_b((v16u8)q0_m, 0x80); \
165 p0 = __msa_xori_b((v16u8)p0_m, 0x80); \
166 }
167
168 #define LPF_MASK_HEV(p3_in, p2_in, p1_in, p0_in, q0_in, q1_in, q2_in, q3_in, \
169 limit_in, b_limit_in, thresh_in, hev_out, mask_out, \
170 flat_out) \
171 { \
172 v16u8 p3_asub_p2_m, p2_asub_p1_m, p1_asub_p0_m, q1_asub_q0_m; \
173 v16u8 p1_asub_q1_m, p0_asub_q0_m, q3_asub_q2_m, q2_asub_q1_m; \
174 \
175 p3_asub_p2_m = __msa_asub_u_b((p3_in), (p2_in)); \
176 p2_asub_p1_m = __msa_asub_u_b((p2_in), (p1_in)); \
177 p1_asub_p0_m = __msa_asub_u_b((p1_in), (p0_in)); \
178 q1_asub_q0_m = __msa_asub_u_b((q1_in), (q0_in)); \
179 q2_asub_q1_m = __msa_asub_u_b((q2_in), (q1_in)); \
180 q3_asub_q2_m = __msa_asub_u_b((q3_in), (q2_in)); \
181 p0_asub_q0_m = __msa_asub_u_b((p0_in), (q0_in)); \
182 p1_asub_q1_m = __msa_asub_u_b((p1_in), (q1_in)); \
183 flat_out = __msa_max_u_b(p1_asub_p0_m, q1_asub_q0_m); \
184 hev_out = (thresh_in) < (v16u8)flat_out; \
185 p0_asub_q0_m = __msa_adds_u_b(p0_asub_q0_m, p0_asub_q0_m); \
186 p1_asub_q1_m >>= 1; \
187 p0_asub_q0_m = __msa_adds_u_b(p0_asub_q0_m, p1_asub_q1_m); \
188 mask_out = (b_limit_in) < p0_asub_q0_m; \
189 mask_out = __msa_max_u_b(flat_out, mask_out); \
190 p3_asub_p2_m = __msa_max_u_b(p3_asub_p2_m, p2_asub_p1_m); \
191 mask_out = __msa_max_u_b(p3_asub_p2_m, mask_out); \
192 q2_asub_q1_m = __msa_max_u_b(q2_asub_q1_m, q3_asub_q2_m); \
193 mask_out = __msa_max_u_b(q2_asub_q1_m, mask_out); \
194 mask_out = (limit_in) < (v16u8)mask_out; \
195 mask_out = __msa_xori_b(mask_out, 0xff); \
196 }
197
198 #define VP8_ST6x1_UB(in0, in0_idx, in1, in1_idx, pdst, stride) \
199 { \
200 uint16_t tmp0_h; \
201 uint32_t tmp0_w; \
202 \
203 tmp0_w = __msa_copy_u_w((v4i32)in0, in0_idx); \
204 tmp0_h = __msa_copy_u_h((v8i16)in1, in1_idx); \
205 SW(tmp0_w, pdst); \
206 SH(tmp0_h, pdst + stride); \
207 }
208
loop_filter_horizontal_4_dual_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit0_ptr,const uint8_t * limit0_ptr,const uint8_t * thresh0_ptr,const uint8_t * b_limit1_ptr,const uint8_t * limit1_ptr,const uint8_t * thresh1_ptr)209 static void loop_filter_horizontal_4_dual_msa(uint8_t *src, int32_t pitch,
210 const uint8_t *b_limit0_ptr,
211 const uint8_t *limit0_ptr,
212 const uint8_t *thresh0_ptr,
213 const uint8_t *b_limit1_ptr,
214 const uint8_t *limit1_ptr,
215 const uint8_t *thresh1_ptr) {
216 v16u8 mask, hev, flat;
217 v16u8 thresh0, b_limit0, limit0, thresh1, b_limit1, limit1;
218 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
219
220 LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
221 thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr);
222 thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr);
223 thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0);
224
225 b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr);
226 b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr);
227 b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0);
228
229 limit0 = (v16u8)__msa_fill_b(*limit0_ptr);
230 limit1 = (v16u8)__msa_fill_b(*limit1_ptr);
231 limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0);
232
233 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0, hev,
234 mask, flat);
235 VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
236
237 ST_UB4(p1, p0, q0, q1, (src - 2 * pitch), pitch);
238 }
239
loop_filter_vertical_4_dual_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit0_ptr,const uint8_t * limit0_ptr,const uint8_t * thresh0_ptr,const uint8_t * b_limit1_ptr,const uint8_t * limit1_ptr,const uint8_t * thresh1_ptr)240 static void loop_filter_vertical_4_dual_msa(uint8_t *src, int32_t pitch,
241 const uint8_t *b_limit0_ptr,
242 const uint8_t *limit0_ptr,
243 const uint8_t *thresh0_ptr,
244 const uint8_t *b_limit1_ptr,
245 const uint8_t *limit1_ptr,
246 const uint8_t *thresh1_ptr) {
247 v16u8 mask, hev, flat;
248 v16u8 thresh0, b_limit0, limit0, thresh1, b_limit1, limit1;
249 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
250 v16u8 row0, row1, row2, row3, row4, row5, row6, row7;
251 v16u8 row8, row9, row10, row11, row12, row13, row14, row15;
252 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
253
254 LD_UB8(src - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
255 LD_UB8(src - 4 + (8 * pitch), pitch, row8, row9, row10, row11, row12, row13,
256 row14, row15);
257 TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, row8,
258 row9, row10, row11, row12, row13, row14, row15, p3, p2,
259 p1, p0, q0, q1, q2, q3);
260
261 thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr);
262 thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr);
263 thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0);
264
265 b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr);
266 b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr);
267 b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0);
268
269 limit0 = (v16u8)__msa_fill_b(*limit0_ptr);
270 limit1 = (v16u8)__msa_fill_b(*limit1_ptr);
271 limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0);
272
273 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0, hev,
274 mask, flat);
275 VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
276 ILVR_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
277 ILVRL_H2_SH(tmp1, tmp0, tmp2, tmp3);
278 ILVL_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
279 ILVRL_H2_SH(tmp1, tmp0, tmp4, tmp5);
280
281 src -= 2;
282 ST4x8_UB(tmp2, tmp3, src, pitch);
283 src += (8 * pitch);
284 ST4x8_UB(tmp4, tmp5, src, pitch);
285 }
286
mbloop_filter_horizontal_edge_y_msa(uint8_t * src,int32_t pitch,const uint8_t b_limit_in,const uint8_t limit_in,const uint8_t thresh_in)287 static void mbloop_filter_horizontal_edge_y_msa(uint8_t *src, int32_t pitch,
288 const uint8_t b_limit_in,
289 const uint8_t limit_in,
290 const uint8_t thresh_in) {
291 uint8_t *temp_src;
292 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
293 v16u8 mask, hev, flat, thresh, limit, b_limit;
294
295 b_limit = (v16u8)__msa_fill_b(b_limit_in);
296 limit = (v16u8)__msa_fill_b(limit_in);
297 thresh = (v16u8)__msa_fill_b(thresh_in);
298 temp_src = src - (pitch << 2);
299 LD_UB8(temp_src, pitch, p3, p2, p1, p0, q0, q1, q2, q3);
300 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
301 mask, flat);
302 VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
303 temp_src = src - 3 * pitch;
304 ST_UB4(p2, p1, p0, q0, temp_src, pitch);
305 temp_src += (4 * pitch);
306 ST_UB2(q1, q2, temp_src, pitch);
307 }
308
mbloop_filter_horizontal_edge_uv_msa(uint8_t * src_u,uint8_t * src_v,int32_t pitch,const uint8_t b_limit_in,const uint8_t limit_in,const uint8_t thresh_in)309 static void mbloop_filter_horizontal_edge_uv_msa(uint8_t *src_u, uint8_t *src_v,
310 int32_t pitch,
311 const uint8_t b_limit_in,
312 const uint8_t limit_in,
313 const uint8_t thresh_in) {
314 uint8_t *temp_src;
315 uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
316 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
317 v16u8 mask, hev, flat, thresh, limit, b_limit;
318 v16u8 p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u;
319 v16u8 p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v;
320
321 b_limit = (v16u8)__msa_fill_b(b_limit_in);
322 limit = (v16u8)__msa_fill_b(limit_in);
323 thresh = (v16u8)__msa_fill_b(thresh_in);
324
325 temp_src = src_u - (pitch << 2);
326 LD_UB8(temp_src, pitch, p3_u, p2_u, p1_u, p0_u, q0_u, q1_u, q2_u, q3_u);
327 temp_src = src_v - (pitch << 2);
328 LD_UB8(temp_src, pitch, p3_v, p2_v, p1_v, p0_v, q0_v, q1_v, q2_v, q3_v);
329
330 ILVR_D4_UB(p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, p2, p1, p0);
331 ILVR_D4_UB(q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, q0, q1, q2, q3);
332 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
333 mask, flat);
334 VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
335
336 p2_d = __msa_copy_u_d((v2i64)p2, 0);
337 p1_d = __msa_copy_u_d((v2i64)p1, 0);
338 p0_d = __msa_copy_u_d((v2i64)p0, 0);
339 q0_d = __msa_copy_u_d((v2i64)q0, 0);
340 q1_d = __msa_copy_u_d((v2i64)q1, 0);
341 q2_d = __msa_copy_u_d((v2i64)q2, 0);
342 src_u -= (pitch * 3);
343 SD4(p2_d, p1_d, p0_d, q0_d, src_u, pitch);
344 src_u += 4 * pitch;
345 SD(q1_d, src_u);
346 src_u += pitch;
347 SD(q2_d, src_u);
348
349 p2_d = __msa_copy_u_d((v2i64)p2, 1);
350 p1_d = __msa_copy_u_d((v2i64)p1, 1);
351 p0_d = __msa_copy_u_d((v2i64)p0, 1);
352 q0_d = __msa_copy_u_d((v2i64)q0, 1);
353 q1_d = __msa_copy_u_d((v2i64)q1, 1);
354 q2_d = __msa_copy_u_d((v2i64)q2, 1);
355 src_v -= (pitch * 3);
356 SD4(p2_d, p1_d, p0_d, q0_d, src_v, pitch);
357 src_v += 4 * pitch;
358 SD(q1_d, src_v);
359 src_v += pitch;
360 SD(q2_d, src_v);
361 }
362
mbloop_filter_vertical_edge_y_msa(uint8_t * src,int32_t pitch,const uint8_t b_limit_in,const uint8_t limit_in,const uint8_t thresh_in)363 static void mbloop_filter_vertical_edge_y_msa(uint8_t *src, int32_t pitch,
364 const uint8_t b_limit_in,
365 const uint8_t limit_in,
366 const uint8_t thresh_in) {
367 uint8_t *temp_src;
368 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
369 v16u8 mask, hev, flat, thresh, limit, b_limit;
370 v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
371 v16u8 row9, row10, row11, row12, row13, row14, row15;
372 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
373
374 b_limit = (v16u8)__msa_fill_b(b_limit_in);
375 limit = (v16u8)__msa_fill_b(limit_in);
376 thresh = (v16u8)__msa_fill_b(thresh_in);
377 temp_src = src - 4;
378 LD_UB8(temp_src, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
379 temp_src += (8 * pitch);
380 LD_UB8(temp_src, pitch, row8, row9, row10, row11, row12, row13, row14, row15);
381 TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, row8,
382 row9, row10, row11, row12, row13, row14, row15, p3, p2,
383 p1, p0, q0, q1, q2, q3);
384
385 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
386 mask, flat);
387 VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
388 ILVR_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
389 ILVRL_H2_SH(tmp1, tmp0, tmp3, tmp4);
390 ILVL_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
391 ILVRL_H2_SH(tmp1, tmp0, tmp6, tmp7);
392 ILVRL_B2_SH(q2, q1, tmp2, tmp5);
393
394 temp_src = src - 3;
395 VP8_ST6x1_UB(tmp3, 0, tmp2, 0, temp_src, 4);
396 temp_src += pitch;
397 VP8_ST6x1_UB(tmp3, 1, tmp2, 1, temp_src, 4);
398 temp_src += pitch;
399 VP8_ST6x1_UB(tmp3, 2, tmp2, 2, temp_src, 4);
400 temp_src += pitch;
401 VP8_ST6x1_UB(tmp3, 3, tmp2, 3, temp_src, 4);
402 temp_src += pitch;
403 VP8_ST6x1_UB(tmp4, 0, tmp2, 4, temp_src, 4);
404 temp_src += pitch;
405 VP8_ST6x1_UB(tmp4, 1, tmp2, 5, temp_src, 4);
406 temp_src += pitch;
407 VP8_ST6x1_UB(tmp4, 2, tmp2, 6, temp_src, 4);
408 temp_src += pitch;
409 VP8_ST6x1_UB(tmp4, 3, tmp2, 7, temp_src, 4);
410 temp_src += pitch;
411 VP8_ST6x1_UB(tmp6, 0, tmp5, 0, temp_src, 4);
412 temp_src += pitch;
413 VP8_ST6x1_UB(tmp6, 1, tmp5, 1, temp_src, 4);
414 temp_src += pitch;
415 VP8_ST6x1_UB(tmp6, 2, tmp5, 2, temp_src, 4);
416 temp_src += pitch;
417 VP8_ST6x1_UB(tmp6, 3, tmp5, 3, temp_src, 4);
418 temp_src += pitch;
419 VP8_ST6x1_UB(tmp7, 0, tmp5, 4, temp_src, 4);
420 temp_src += pitch;
421 VP8_ST6x1_UB(tmp7, 1, tmp5, 5, temp_src, 4);
422 temp_src += pitch;
423 VP8_ST6x1_UB(tmp7, 2, tmp5, 6, temp_src, 4);
424 temp_src += pitch;
425 VP8_ST6x1_UB(tmp7, 3, tmp5, 7, temp_src, 4);
426 }
427
mbloop_filter_vertical_edge_uv_msa(uint8_t * src_u,uint8_t * src_v,int32_t pitch,const uint8_t b_limit_in,const uint8_t limit_in,const uint8_t thresh_in)428 static void mbloop_filter_vertical_edge_uv_msa(uint8_t *src_u, uint8_t *src_v,
429 int32_t pitch,
430 const uint8_t b_limit_in,
431 const uint8_t limit_in,
432 const uint8_t thresh_in) {
433 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
434 v16u8 mask, hev, flat, thresh, limit, b_limit;
435 v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
436 v16u8 row9, row10, row11, row12, row13, row14, row15;
437 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
438
439 b_limit = (v16u8)__msa_fill_b(b_limit_in);
440 limit = (v16u8)__msa_fill_b(limit_in);
441 thresh = (v16u8)__msa_fill_b(thresh_in);
442
443 LD_UB8(src_u - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
444 LD_UB8(src_v - 4, pitch, row8, row9, row10, row11, row12, row13, row14,
445 row15);
446 TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, row8,
447 row9, row10, row11, row12, row13, row14, row15, p3, p2,
448 p1, p0, q0, q1, q2, q3);
449
450 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
451 mask, flat);
452 VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
453
454 ILVR_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
455 ILVRL_H2_SH(tmp1, tmp0, tmp3, tmp4);
456 ILVL_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
457 ILVRL_H2_SH(tmp1, tmp0, tmp6, tmp7);
458 ILVRL_B2_SH(q2, q1, tmp2, tmp5);
459
460 src_u -= 3;
461 VP8_ST6x1_UB(tmp3, 0, tmp2, 0, src_u, 4);
462 src_u += pitch;
463 VP8_ST6x1_UB(tmp3, 1, tmp2, 1, src_u, 4);
464 src_u += pitch;
465 VP8_ST6x1_UB(tmp3, 2, tmp2, 2, src_u, 4);
466 src_u += pitch;
467 VP8_ST6x1_UB(tmp3, 3, tmp2, 3, src_u, 4);
468 src_u += pitch;
469 VP8_ST6x1_UB(tmp4, 0, tmp2, 4, src_u, 4);
470 src_u += pitch;
471 VP8_ST6x1_UB(tmp4, 1, tmp2, 5, src_u, 4);
472 src_u += pitch;
473 VP8_ST6x1_UB(tmp4, 2, tmp2, 6, src_u, 4);
474 src_u += pitch;
475 VP8_ST6x1_UB(tmp4, 3, tmp2, 7, src_u, 4);
476
477 src_v -= 3;
478 VP8_ST6x1_UB(tmp6, 0, tmp5, 0, src_v, 4);
479 src_v += pitch;
480 VP8_ST6x1_UB(tmp6, 1, tmp5, 1, src_v, 4);
481 src_v += pitch;
482 VP8_ST6x1_UB(tmp6, 2, tmp5, 2, src_v, 4);
483 src_v += pitch;
484 VP8_ST6x1_UB(tmp6, 3, tmp5, 3, src_v, 4);
485 src_v += pitch;
486 VP8_ST6x1_UB(tmp7, 0, tmp5, 4, src_v, 4);
487 src_v += pitch;
488 VP8_ST6x1_UB(tmp7, 1, tmp5, 5, src_v, 4);
489 src_v += pitch;
490 VP8_ST6x1_UB(tmp7, 2, tmp5, 6, src_v, 4);
491 src_v += pitch;
492 VP8_ST6x1_UB(tmp7, 3, tmp5, 7, src_v, 4);
493 }
494
vp8_loop_filter_simple_horizontal_edge_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit_ptr)495 void vp8_loop_filter_simple_horizontal_edge_msa(uint8_t *src, int32_t pitch,
496 const uint8_t *b_limit_ptr) {
497 v16u8 p1, p0, q1, q0;
498 v16u8 mask, b_limit;
499
500 b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
501 LD_UB4(src - (pitch << 1), pitch, p1, p0, q0, q1);
502 VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask);
503 VP8_SIMPLE_FILT(p1, p0, q0, q1, mask);
504 ST_UB2(p0, q0, (src - pitch), pitch);
505 }
506
vp8_loop_filter_simple_vertical_edge_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit_ptr)507 void vp8_loop_filter_simple_vertical_edge_msa(uint8_t *src, int32_t pitch,
508 const uint8_t *b_limit_ptr) {
509 uint8_t *temp_src;
510 v16u8 p1, p0, q1, q0;
511 v16u8 mask, b_limit;
512 v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
513 v16u8 row9, row10, row11, row12, row13, row14, row15;
514 v8i16 tmp0, tmp1;
515
516 b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
517 temp_src = src - 2;
518 LD_UB8(temp_src, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
519 temp_src += (8 * pitch);
520 LD_UB8(temp_src, pitch, row8, row9, row10, row11, row12, row13, row14, row15);
521 TRANSPOSE16x4_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, row8,
522 row9, row10, row11, row12, row13, row14, row15, p1, p0,
523 q0, q1);
524 VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask);
525 VP8_SIMPLE_FILT(p1, p0, q0, q1, mask);
526 ILVRL_B2_SH(q0, p0, tmp1, tmp0);
527
528 src -= 1;
529 ST2x4_UB(tmp1, 0, src, pitch);
530 src += 4 * pitch;
531 ST2x4_UB(tmp1, 4, src, pitch);
532 src += 4 * pitch;
533 ST2x4_UB(tmp0, 0, src, pitch);
534 src += 4 * pitch;
535 ST2x4_UB(tmp0, 4, src, pitch);
536 src += 4 * pitch;
537 }
538
loop_filter_horizontal_edge_uv_msa(uint8_t * src_u,uint8_t * src_v,int32_t pitch,const uint8_t b_limit_in,const uint8_t limit_in,const uint8_t thresh_in)539 static void loop_filter_horizontal_edge_uv_msa(uint8_t *src_u, uint8_t *src_v,
540 int32_t pitch,
541 const uint8_t b_limit_in,
542 const uint8_t limit_in,
543 const uint8_t thresh_in) {
544 uint64_t p1_d, p0_d, q0_d, q1_d;
545 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
546 v16u8 mask, hev, flat, thresh, limit, b_limit;
547 v16u8 p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u;
548 v16u8 p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v;
549
550 thresh = (v16u8)__msa_fill_b(thresh_in);
551 limit = (v16u8)__msa_fill_b(limit_in);
552 b_limit = (v16u8)__msa_fill_b(b_limit_in);
553
554 src_u = src_u - (pitch << 2);
555 LD_UB8(src_u, pitch, p3_u, p2_u, p1_u, p0_u, q0_u, q1_u, q2_u, q3_u);
556 src_u += (5 * pitch);
557 src_v = src_v - (pitch << 2);
558 LD_UB8(src_v, pitch, p3_v, p2_v, p1_v, p0_v, q0_v, q1_v, q2_v, q3_v);
559 src_v += (5 * pitch);
560
561 /* right 8 element of p3 are u pixel and
562 left 8 element of p3 are v pixel */
563 ILVR_D4_UB(p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, p2, p1, p0);
564 ILVR_D4_UB(q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, q0, q1, q2, q3);
565 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
566 mask, flat);
567 VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
568
569 p1_d = __msa_copy_u_d((v2i64)p1, 0);
570 p0_d = __msa_copy_u_d((v2i64)p0, 0);
571 q0_d = __msa_copy_u_d((v2i64)q0, 0);
572 q1_d = __msa_copy_u_d((v2i64)q1, 0);
573 SD4(q1_d, q0_d, p0_d, p1_d, src_u, (-pitch));
574
575 p1_d = __msa_copy_u_d((v2i64)p1, 1);
576 p0_d = __msa_copy_u_d((v2i64)p0, 1);
577 q0_d = __msa_copy_u_d((v2i64)q0, 1);
578 q1_d = __msa_copy_u_d((v2i64)q1, 1);
579 SD4(q1_d, q0_d, p0_d, p1_d, src_v, (-pitch));
580 }
581
loop_filter_vertical_edge_uv_msa(uint8_t * src_u,uint8_t * src_v,int32_t pitch,const uint8_t b_limit_in,const uint8_t limit_in,const uint8_t thresh_in)582 static void loop_filter_vertical_edge_uv_msa(uint8_t *src_u, uint8_t *src_v,
583 int32_t pitch,
584 const uint8_t b_limit_in,
585 const uint8_t limit_in,
586 const uint8_t thresh_in) {
587 uint8_t *temp_src_u, *temp_src_v;
588 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
589 v16u8 mask, hev, flat, thresh, limit, b_limit;
590 v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
591 v16u8 row9, row10, row11, row12, row13, row14, row15;
592 v4i32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
593
594 thresh = (v16u8)__msa_fill_b(thresh_in);
595 limit = (v16u8)__msa_fill_b(limit_in);
596 b_limit = (v16u8)__msa_fill_b(b_limit_in);
597
598 LD_UB8(src_u - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
599 LD_UB8(src_v - 4, pitch, row8, row9, row10, row11, row12, row13, row14,
600 row15);
601 TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, row8,
602 row9, row10, row11, row12, row13, row14, row15, p3, p2,
603 p1, p0, q0, q1, q2, q3);
604
605 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
606 mask, flat);
607 VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
608 ILVR_B2_SW(p0, p1, q1, q0, tmp0, tmp1);
609 ILVRL_H2_SW(tmp1, tmp0, tmp2, tmp3);
610 tmp0 = (v4i32)__msa_ilvl_b((v16i8)p0, (v16i8)p1);
611 tmp1 = (v4i32)__msa_ilvl_b((v16i8)q1, (v16i8)q0);
612 ILVRL_H2_SW(tmp1, tmp0, tmp4, tmp5);
613
614 temp_src_u = src_u - 2;
615 ST4x4_UB(tmp2, tmp2, 0, 1, 2, 3, temp_src_u, pitch);
616 temp_src_u += 4 * pitch;
617 ST4x4_UB(tmp3, tmp3, 0, 1, 2, 3, temp_src_u, pitch);
618
619 temp_src_v = src_v - 2;
620 ST4x4_UB(tmp4, tmp4, 0, 1, 2, 3, temp_src_v, pitch);
621 temp_src_v += 4 * pitch;
622 ST4x4_UB(tmp5, tmp5, 0, 1, 2, 3, temp_src_v, pitch);
623 }
624
vp8_loop_filter_mbh_msa(uint8_t * src_y,uint8_t * src_u,uint8_t * src_v,int32_t pitch_y,int32_t pitch_u_v,loop_filter_info * lpf_info_ptr)625 void vp8_loop_filter_mbh_msa(uint8_t *src_y, uint8_t *src_u, uint8_t *src_v,
626 int32_t pitch_y, int32_t pitch_u_v,
627 loop_filter_info *lpf_info_ptr) {
628 mbloop_filter_horizontal_edge_y_msa(src_y, pitch_y, *lpf_info_ptr->mblim,
629 *lpf_info_ptr->lim,
630 *lpf_info_ptr->hev_thr);
631 if (src_u) {
632 mbloop_filter_horizontal_edge_uv_msa(
633 src_u, src_v, pitch_u_v, *lpf_info_ptr->mblim, *lpf_info_ptr->lim,
634 *lpf_info_ptr->hev_thr);
635 }
636 }
637
vp8_loop_filter_mbv_msa(uint8_t * src_y,uint8_t * src_u,uint8_t * src_v,int32_t pitch_y,int32_t pitch_u_v,loop_filter_info * lpf_info_ptr)638 void vp8_loop_filter_mbv_msa(uint8_t *src_y, uint8_t *src_u, uint8_t *src_v,
639 int32_t pitch_y, int32_t pitch_u_v,
640 loop_filter_info *lpf_info_ptr) {
641 mbloop_filter_vertical_edge_y_msa(src_y, pitch_y, *lpf_info_ptr->mblim,
642 *lpf_info_ptr->lim, *lpf_info_ptr->hev_thr);
643 if (src_u) {
644 mbloop_filter_vertical_edge_uv_msa(src_u, src_v, pitch_u_v,
645 *lpf_info_ptr->mblim, *lpf_info_ptr->lim,
646 *lpf_info_ptr->hev_thr);
647 }
648 }
649
vp8_loop_filter_bh_msa(uint8_t * src_y,uint8_t * src_u,uint8_t * src_v,int32_t pitch_y,int32_t pitch_u_v,loop_filter_info * lpf_info_ptr)650 void vp8_loop_filter_bh_msa(uint8_t *src_y, uint8_t *src_u, uint8_t *src_v,
651 int32_t pitch_y, int32_t pitch_u_v,
652 loop_filter_info *lpf_info_ptr) {
653 loop_filter_horizontal_4_dual_msa(src_y + 4 * pitch_y, pitch_y,
654 lpf_info_ptr->blim, lpf_info_ptr->lim,
655 lpf_info_ptr->hev_thr, lpf_info_ptr->blim,
656 lpf_info_ptr->lim, lpf_info_ptr->hev_thr);
657 loop_filter_horizontal_4_dual_msa(src_y + 8 * pitch_y, pitch_y,
658 lpf_info_ptr->blim, lpf_info_ptr->lim,
659 lpf_info_ptr->hev_thr, lpf_info_ptr->blim,
660 lpf_info_ptr->lim, lpf_info_ptr->hev_thr);
661 loop_filter_horizontal_4_dual_msa(src_y + 12 * pitch_y, pitch_y,
662 lpf_info_ptr->blim, lpf_info_ptr->lim,
663 lpf_info_ptr->hev_thr, lpf_info_ptr->blim,
664 lpf_info_ptr->lim, lpf_info_ptr->hev_thr);
665 if (src_u) {
666 loop_filter_horizontal_edge_uv_msa(
667 src_u + (4 * pitch_u_v), src_v + (4 * pitch_u_v), pitch_u_v,
668 *lpf_info_ptr->blim, *lpf_info_ptr->lim, *lpf_info_ptr->hev_thr);
669 }
670 }
671
vp8_loop_filter_bv_msa(uint8_t * src_y,uint8_t * src_u,uint8_t * src_v,int32_t pitch_y,int32_t pitch_u_v,loop_filter_info * lpf_info_ptr)672 void vp8_loop_filter_bv_msa(uint8_t *src_y, uint8_t *src_u, uint8_t *src_v,
673 int32_t pitch_y, int32_t pitch_u_v,
674 loop_filter_info *lpf_info_ptr) {
675 loop_filter_vertical_4_dual_msa(src_y + 4, pitch_y, lpf_info_ptr->blim,
676 lpf_info_ptr->lim, lpf_info_ptr->hev_thr,
677 lpf_info_ptr->blim, lpf_info_ptr->lim,
678 lpf_info_ptr->hev_thr);
679 loop_filter_vertical_4_dual_msa(src_y + 8, pitch_y, lpf_info_ptr->blim,
680 lpf_info_ptr->lim, lpf_info_ptr->hev_thr,
681 lpf_info_ptr->blim, lpf_info_ptr->lim,
682 lpf_info_ptr->hev_thr);
683 loop_filter_vertical_4_dual_msa(src_y + 12, pitch_y, lpf_info_ptr->blim,
684 lpf_info_ptr->lim, lpf_info_ptr->hev_thr,
685 lpf_info_ptr->blim, lpf_info_ptr->lim,
686 lpf_info_ptr->hev_thr);
687 if (src_u) {
688 loop_filter_vertical_edge_uv_msa(src_u + 4, src_v + 4, pitch_u_v,
689 *lpf_info_ptr->blim, *lpf_info_ptr->lim,
690 *lpf_info_ptr->hev_thr);
691 }
692 }
693
vp8_loop_filter_bhs_msa(uint8_t * src_y,int32_t pitch_y,const uint8_t * b_limit_ptr)694 void vp8_loop_filter_bhs_msa(uint8_t *src_y, int32_t pitch_y,
695 const uint8_t *b_limit_ptr) {
696 vp8_loop_filter_simple_horizontal_edge_msa(src_y + (4 * pitch_y), pitch_y,
697 b_limit_ptr);
698 vp8_loop_filter_simple_horizontal_edge_msa(src_y + (8 * pitch_y), pitch_y,
699 b_limit_ptr);
700 vp8_loop_filter_simple_horizontal_edge_msa(src_y + (12 * pitch_y), pitch_y,
701 b_limit_ptr);
702 }
703
vp8_loop_filter_bvs_msa(uint8_t * src_y,int32_t pitch_y,const uint8_t * b_limit_ptr)704 void vp8_loop_filter_bvs_msa(uint8_t *src_y, int32_t pitch_y,
705 const uint8_t *b_limit_ptr) {
706 vp8_loop_filter_simple_vertical_edge_msa(src_y + 4, pitch_y, b_limit_ptr);
707 vp8_loop_filter_simple_vertical_edge_msa(src_y + 8, pitch_y, b_limit_ptr);
708 vp8_loop_filter_simple_vertical_edge_msa(src_y + 12, pitch_y, b_limit_ptr);
709 }
710