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 "vpx_dsp/mips/loopfilter_msa.h"
12
vpx_lpf_horizontal_8_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit_ptr,const uint8_t * limit_ptr,const uint8_t * thresh_ptr)13 void vpx_lpf_horizontal_8_msa(uint8_t *src, int32_t pitch,
14 const uint8_t *b_limit_ptr,
15 const uint8_t *limit_ptr,
16 const uint8_t *thresh_ptr) {
17 uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
18 v16u8 mask, hev, flat, thresh, b_limit, limit;
19 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
20 v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
21 v8i16 p2_filter8, p1_filter8, p0_filter8, q0_filter8, q1_filter8, q2_filter8;
22 v8u16 p3_r, p2_r, p1_r, p0_r, q3_r, q2_r, q1_r, q0_r;
23 v16i8 zero = { 0 };
24
25 /* load vector elements */
26 LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
27
28 thresh = (v16u8)__msa_fill_b(*thresh_ptr);
29 b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
30 limit = (v16u8)__msa_fill_b(*limit_ptr);
31
32 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
33 mask, flat);
34 VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
35 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
36
37 flat = (v16u8)__msa_ilvr_d((v2i64)zero, (v2i64)flat);
38
39 if (__msa_test_bz_v(flat)) {
40 p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
41 p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
42 q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
43 q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
44 SD4(p1_d, p0_d, q0_d, q1_d, (src - 2 * pitch), pitch);
45 } else {
46 ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1, zero,
47 q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r);
48 VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filter8,
49 p1_filter8, p0_filter8, q0_filter8, q1_filter8, q2_filter8);
50
51 /* convert 16 bit output data into 8 bit */
52 PCKEV_B4_SH(zero, p2_filter8, zero, p1_filter8, zero, p0_filter8, zero,
53 q0_filter8, p2_filter8, p1_filter8, p0_filter8, q0_filter8);
54 PCKEV_B2_SH(zero, q1_filter8, zero, q2_filter8, q1_filter8, q2_filter8);
55
56 /* store pixel values */
57 p2_out = __msa_bmnz_v(p2, (v16u8)p2_filter8, flat);
58 p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filter8, flat);
59 p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filter8, flat);
60 q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filter8, flat);
61 q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filter8, flat);
62 q2_out = __msa_bmnz_v(q2, (v16u8)q2_filter8, flat);
63
64 p2_d = __msa_copy_u_d((v2i64)p2_out, 0);
65 p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
66 p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
67 q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
68 q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
69 q2_d = __msa_copy_u_d((v2i64)q2_out, 0);
70
71 src -= 3 * pitch;
72
73 SD4(p2_d, p1_d, p0_d, q0_d, src, pitch);
74 src += (4 * pitch);
75 SD(q1_d, src);
76 src += pitch;
77 SD(q2_d, src);
78 }
79 }
80
vpx_lpf_horizontal_8_dual_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * b_limit1,const uint8_t * limit1,const uint8_t * thresh1)81 void vpx_lpf_horizontal_8_dual_msa(
82 uint8_t *src, int32_t pitch, const uint8_t *b_limit0, const uint8_t *limit0,
83 const uint8_t *thresh0, const uint8_t *b_limit1, const uint8_t *limit1,
84 const uint8_t *thresh1) {
85 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
86 v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
87 v16u8 flat, mask, hev, tmp, thresh, b_limit, limit;
88 v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
89 v8u16 p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l;
90 v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
91 v8i16 p2_filt8_l, p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l;
92 v16u8 zero = { 0 };
93
94 /* load vector elements */
95 LD_UB8(src - (4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
96
97 thresh = (v16u8)__msa_fill_b(*thresh0);
98 tmp = (v16u8)__msa_fill_b(*thresh1);
99 thresh = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)thresh);
100
101 b_limit = (v16u8)__msa_fill_b(*b_limit0);
102 tmp = (v16u8)__msa_fill_b(*b_limit1);
103 b_limit = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)b_limit);
104
105 limit = (v16u8)__msa_fill_b(*limit0);
106 tmp = (v16u8)__msa_fill_b(*limit1);
107 limit = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)limit);
108
109 /* mask and hev */
110 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
111 mask, flat);
112 VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
113 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
114
115 if (__msa_test_bz_v(flat)) {
116 ST_UB4(p1_out, p0_out, q0_out, q1_out, (src - 2 * pitch), pitch);
117 } else {
118 ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1, zero,
119 q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r);
120 VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
121 p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
122
123 ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l, p2_l, p1_l, p0_l);
124 ILVL_B4_UH(zero, q0, zero, q1, zero, q2, zero, q3, q0_l, q1_l, q2_l, q3_l);
125 VP9_FILTER8(p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l, p2_filt8_l,
126 p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l);
127
128 /* convert 16 bit output data into 8 bit */
129 PCKEV_B4_SH(p2_filt8_l, p2_filt8_r, p1_filt8_l, p1_filt8_r, p0_filt8_l,
130 p0_filt8_r, q0_filt8_l, q0_filt8_r, p2_filt8_r, p1_filt8_r,
131 p0_filt8_r, q0_filt8_r);
132 PCKEV_B2_SH(q1_filt8_l, q1_filt8_r, q2_filt8_l, q2_filt8_r, q1_filt8_r,
133 q2_filt8_r);
134
135 /* store pixel values */
136 p2_out = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
137 p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
138 p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
139 q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
140 q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
141 q2_out = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
142
143 src -= 3 * pitch;
144
145 ST_UB4(p2_out, p1_out, p0_out, q0_out, src, pitch);
146 src += (4 * pitch);
147 ST_UB2(q1_out, q2_out, src, pitch);
148 src += (2 * pitch);
149 }
150 }
151
vpx_lpf_vertical_8_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit_ptr,const uint8_t * limit_ptr,const uint8_t * thresh_ptr)152 void vpx_lpf_vertical_8_msa(uint8_t *src, int32_t pitch,
153 const uint8_t *b_limit_ptr,
154 const uint8_t *limit_ptr,
155 const uint8_t *thresh_ptr) {
156 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
157 v16u8 p1_out, p0_out, q0_out, q1_out;
158 v16u8 flat, mask, hev, thresh, b_limit, limit;
159 v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
160 v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
161 v16u8 zero = { 0 };
162 v8i16 vec0, vec1, vec2, vec3, vec4;
163
164 /* load vector elements */
165 LD_UB8(src - 4, pitch, p3, p2, p1, p0, q0, q1, q2, q3);
166
167 TRANSPOSE8x8_UB_UB(p3, p2, p1, p0, q0, q1, q2, q3, p3, p2, p1, p0, q0, q1, q2,
168 q3);
169
170 thresh = (v16u8)__msa_fill_b(*thresh_ptr);
171 b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
172 limit = (v16u8)__msa_fill_b(*limit_ptr);
173
174 /* mask and hev */
175 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
176 mask, flat);
177 /* flat4 */
178 VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
179 /* filter4 */
180 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
181
182 flat = (v16u8)__msa_ilvr_d((v2i64)zero, (v2i64)flat);
183
184 if (__msa_test_bz_v(flat)) {
185 /* Store 4 pixels p1-_q1 */
186 ILVR_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
187 ILVRL_H2_SH(vec1, vec0, vec2, vec3);
188
189 src -= 2;
190 ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
191 src += 4 * pitch;
192 ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
193 } else {
194 ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1, zero,
195 q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r);
196 VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
197 p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
198 /* convert 16 bit output data into 8 bit */
199 PCKEV_B4_SH(p2_filt8_r, p2_filt8_r, p1_filt8_r, p1_filt8_r, p0_filt8_r,
200 p0_filt8_r, q0_filt8_r, q0_filt8_r, p2_filt8_r, p1_filt8_r,
201 p0_filt8_r, q0_filt8_r);
202 PCKEV_B2_SH(q1_filt8_r, q1_filt8_r, q2_filt8_r, q2_filt8_r, q1_filt8_r,
203 q2_filt8_r);
204
205 /* store pixel values */
206 p2 = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
207 p1 = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
208 p0 = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
209 q0 = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
210 q1 = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
211 q2 = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
212
213 /* Store 6 pixels p2-_q2 */
214 ILVR_B2_SH(p1, p2, q0, p0, vec0, vec1);
215 ILVRL_H2_SH(vec1, vec0, vec2, vec3);
216 vec4 = (v8i16)__msa_ilvr_b((v16i8)q2, (v16i8)q1);
217
218 src -= 3;
219 ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
220 ST2x4_UB(vec4, 0, src + 4, pitch);
221 src += (4 * pitch);
222 ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
223 ST2x4_UB(vec4, 4, src + 4, pitch);
224 }
225 }
226
vpx_lpf_vertical_8_dual_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit0,const uint8_t * limit0,const uint8_t * thresh0,const uint8_t * b_limit1,const uint8_t * limit1,const uint8_t * thresh1)227 void vpx_lpf_vertical_8_dual_msa(uint8_t *src, int32_t pitch,
228 const uint8_t *b_limit0, const uint8_t *limit0,
229 const uint8_t *thresh0,
230 const uint8_t *b_limit1, const uint8_t *limit1,
231 const uint8_t *thresh1) {
232 uint8_t *temp_src;
233 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
234 v16u8 p1_out, p0_out, q0_out, q1_out;
235 v16u8 flat, mask, hev, thresh, b_limit, limit;
236 v16u8 row4, row5, row6, row7, row12, row13, row14, row15;
237 v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
238 v8u16 p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l;
239 v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
240 v8i16 p2_filt8_l, p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l;
241 v16u8 zero = { 0 };
242 v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
243
244 temp_src = src - 4;
245
246 LD_UB8(temp_src, pitch, p0, p1, p2, p3, row4, row5, row6, row7);
247 temp_src += (8 * pitch);
248 LD_UB8(temp_src, pitch, q3, q2, q1, q0, row12, row13, row14, row15);
249
250 /* transpose 16x8 matrix into 8x16 */
251 TRANSPOSE16x8_UB_UB(p0, p1, p2, p3, row4, row5, row6, row7, q3, q2, q1, q0,
252 row12, row13, row14, row15, p3, p2, p1, p0, q0, q1, q2,
253 q3);
254
255 thresh = (v16u8)__msa_fill_b(*thresh0);
256 vec0 = (v8i16)__msa_fill_b(*thresh1);
257 thresh = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)thresh);
258
259 b_limit = (v16u8)__msa_fill_b(*b_limit0);
260 vec0 = (v8i16)__msa_fill_b(*b_limit1);
261 b_limit = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)b_limit);
262
263 limit = (v16u8)__msa_fill_b(*limit0);
264 vec0 = (v8i16)__msa_fill_b(*limit1);
265 limit = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)limit);
266
267 /* mask and hev */
268 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
269 mask, flat);
270 /* flat4 */
271 VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
272 /* filter4 */
273 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
274
275 if (__msa_test_bz_v(flat)) {
276 ILVR_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
277 ILVRL_H2_SH(vec1, vec0, vec2, vec3);
278 ILVL_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
279 ILVRL_H2_SH(vec1, vec0, vec4, vec5);
280
281 src -= 2;
282 ST4x8_UB(vec2, vec3, src, pitch);
283 src += 8 * pitch;
284 ST4x8_UB(vec4, vec5, src, pitch);
285 } else {
286 ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1, zero,
287 q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r);
288 VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
289 p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
290
291 ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l, p2_l, p1_l, p0_l);
292 ILVL_B4_UH(zero, q0, zero, q1, zero, q2, zero, q3, q0_l, q1_l, q2_l, q3_l);
293
294 /* filter8 */
295 VP9_FILTER8(p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l, p2_filt8_l,
296 p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l);
297
298 /* convert 16 bit output data into 8 bit */
299 PCKEV_B4_SH(p2_filt8_l, p2_filt8_r, p1_filt8_l, p1_filt8_r, p0_filt8_l,
300 p0_filt8_r, q0_filt8_l, q0_filt8_r, p2_filt8_r, p1_filt8_r,
301 p0_filt8_r, q0_filt8_r);
302 PCKEV_B2_SH(q1_filt8_l, q1_filt8_r, q2_filt8_l, q2_filt8_r, q1_filt8_r,
303 q2_filt8_r);
304
305 /* store pixel values */
306 p2 = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
307 p1 = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
308 p0 = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
309 q0 = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
310 q1 = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
311 q2 = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
312
313 ILVR_B2_SH(p1, p2, q0, p0, vec0, vec1);
314 ILVRL_H2_SH(vec1, vec0, vec3, vec4);
315 ILVL_B2_SH(p1, p2, q0, p0, vec0, vec1);
316 ILVRL_H2_SH(vec1, vec0, vec6, vec7);
317 ILVRL_B2_SH(q2, q1, vec2, vec5);
318
319 src -= 3;
320 ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
321 ST2x4_UB(vec2, 0, src + 4, pitch);
322 src += (4 * pitch);
323 ST4x4_UB(vec4, vec4, 0, 1, 2, 3, src, pitch);
324 ST2x4_UB(vec2, 4, src + 4, pitch);
325 src += (4 * pitch);
326 ST4x4_UB(vec6, vec6, 0, 1, 2, 3, src, pitch);
327 ST2x4_UB(vec5, 0, src + 4, pitch);
328 src += (4 * pitch);
329 ST4x4_UB(vec7, vec7, 0, 1, 2, 3, src, pitch);
330 ST2x4_UB(vec5, 4, src + 4, pitch);
331 }
332 }
333