1 /*
2 * Copyright (c) 2015 - 2017 Manojkumar Bhosale (Manojkumar.Bhosale@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 #include "libavutil/mips/generic_macros_msa.h"
22 #include "libavcodec/mips/hevcdsp_mips.h"
23 #include "libavcodec/mips/hevc_macros_msa.h"
24
25 static const uint8_t ff_hevc_mask_arr[16 * 2] __attribute__((aligned(0x40))) = {
26 /* 8 width cases */
27 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
28 /* 4 width cases */
29 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20
30 };
31
32 #define HEVC_UNIW_RND_CLIP2_MAX_SATU_H(in0_h, in1_h, wgt_w, offset_h, rnd_w, \
33 out0_h, out1_h) \
34 { \
35 v4i32 in0_r_m, in0_l_m, in1_r_m, in1_l_m; \
36 \
37 ILVRL_H2_SW(in0_h, in0_h, in0_r_m, in0_l_m); \
38 ILVRL_H2_SW(in1_h, in1_h, in1_r_m, in1_l_m); \
39 DOTP_SH4_SW(in0_r_m, in1_r_m, in0_l_m, in1_l_m, wgt_w, wgt_w, wgt_w, \
40 wgt_w, in0_r_m, in1_r_m, in0_l_m, in1_l_m); \
41 SRAR_W4_SW(in0_r_m, in1_r_m, in0_l_m, in1_l_m, rnd_w); \
42 PCKEV_H2_SH(in0_l_m, in0_r_m, in1_l_m, in1_r_m, out0_h, out1_h); \
43 ADDS_SH2_SH(out0_h, offset_h, out1_h, offset_h, out0_h, out1_h); \
44 CLIP_SH2_0_255(out0_h, out1_h); \
45 }
46
47 #define HEVC_UNIW_RND_CLIP4_MAX_SATU_H(in0_h, in1_h, in2_h, in3_h, wgt_w, \
48 offset_h, rnd_w, out0_h, out1_h, \
49 out2_h, out3_h) \
50 { \
51 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(in0_h, in1_h, wgt_w, offset_h, rnd_w, \
52 out0_h, out1_h); \
53 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(in2_h, in3_h, wgt_w, offset_h, rnd_w, \
54 out2_h, out3_h); \
55 }
56
hevc_uniwgt_copy_4w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)57 static void hevc_uniwgt_copy_4w_msa(uint8_t *src,
58 int32_t src_stride,
59 uint8_t *dst,
60 int32_t dst_stride,
61 int32_t height,
62 int32_t weight,
63 int32_t offset,
64 int32_t rnd_val)
65 {
66 uint32_t loop_cnt, tp0, tp1, tp2, tp3;
67 v16i8 zero = { 0 };
68 v16u8 out0, out1;
69 v16i8 src0 = { 0 }, src1 = { 0 };
70 v8i16 dst0, dst1, dst2, dst3, offset_vec;
71 v4i32 weight_vec, rnd_vec;
72
73 weight = weight & 0x0000FFFF;
74 weight_vec = __msa_fill_w(weight);
75 offset_vec = __msa_fill_h(offset);
76 rnd_vec = __msa_fill_w(rnd_val);
77
78 if (2 == height) {
79 v4i32 dst0_r, dst0_l;
80
81 LW2(src, src_stride, tp0, tp1);
82 INSERT_W2_SB(tp0, tp1, src0);
83 dst0 = (v8i16) __msa_ilvr_b(zero, src0);
84 dst0 <<= 6;
85
86 ILVRL_H2_SW(dst0, dst0, dst0_r, dst0_l);
87 DOTP_SH2_SW(dst0_r, dst0_l, weight_vec, weight_vec, dst0_r, dst0_l);
88 SRAR_W2_SW(dst0_r, dst0_l, rnd_vec);
89 dst0 = __msa_pckev_h((v8i16) dst0_l, (v8i16) dst0_r);
90 dst0 += offset_vec;
91 CLIP_SH_0_255(dst0);
92 out0 = (v16u8) __msa_pckev_b((v16i8) dst0, (v16i8) dst0);
93 ST_W2(out0, 0, 1, dst, dst_stride);
94 } else if (4 == height) {
95 LW4(src, src_stride, tp0, tp1, tp2, tp3);
96 INSERT_W4_SB(tp0, tp1, tp2, tp3, src0);
97 ILVRL_B2_SH(zero, src0, dst0, dst1);
98 SLLI_2V(dst0, dst1, 6);
99 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst0, dst1, weight_vec, offset_vec,
100 rnd_vec, dst0, dst1);
101 out0 = (v16u8) __msa_pckev_b((v16i8) dst1, (v16i8) dst0);
102 ST_W4(out0, 0, 1, 2, 3, dst, dst_stride);
103 } else if (0 == (height % 8)) {
104 for (loop_cnt = (height >> 3); loop_cnt--;) {
105 LW4(src, src_stride, tp0, tp1, tp2, tp3);
106 src += 4 * src_stride;
107 INSERT_W4_SB(tp0, tp1, tp2, tp3, src0);
108 LW4(src, src_stride, tp0, tp1, tp2, tp3);
109 src += 4 * src_stride;
110 INSERT_W4_SB(tp0, tp1, tp2, tp3, src1);
111 ILVRL_B2_SH(zero, src0, dst0, dst1);
112 ILVRL_B2_SH(zero, src1, dst2, dst3);
113 SLLI_4V(dst0, dst1, dst2, dst3, 6);
114 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
115 offset_vec, rnd_vec, dst0, dst1,
116 dst2, dst3);
117 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
118 ST_W8(out0, out1, 0, 1, 2, 3, 0, 1, 2, 3, dst, dst_stride);
119 dst += 8 * dst_stride;
120 }
121 }
122 }
123
hevc_uniwgt_copy_6w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)124 static void hevc_uniwgt_copy_6w_msa(uint8_t *src,
125 int32_t src_stride,
126 uint8_t *dst,
127 int32_t dst_stride,
128 int32_t height,
129 int32_t weight,
130 int32_t offset,
131 int32_t rnd_val)
132 {
133 uint32_t loop_cnt;
134 uint64_t tp0, tp1, tp2, tp3;
135 v16i8 zero = { 0 };
136 v16u8 out0, out1, out2, out3;
137 v16i8 src0, src1, src2, src3;
138 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, offset_vec;
139 v4i32 weight_vec, rnd_vec;
140
141 weight = weight & 0x0000FFFF;
142 weight_vec = __msa_fill_w(weight);
143 offset_vec = __msa_fill_h(offset);
144 rnd_vec = __msa_fill_w(rnd_val);
145
146 for (loop_cnt = (height >> 3); loop_cnt--;) {
147 LD4(src, src_stride, tp0, tp1, tp2, tp3);
148 src += (4 * src_stride);
149 INSERT_D2_SB(tp0, tp1, src0);
150 INSERT_D2_SB(tp2, tp3, src1);
151 LD4(src, src_stride, tp0, tp1, tp2, tp3);
152 src += (4 * src_stride);
153 INSERT_D2_SB(tp0, tp1, src2);
154 INSERT_D2_SB(tp2, tp3, src3);
155
156 ILVRL_B2_SH(zero, src0, dst0, dst1);
157 ILVRL_B2_SH(zero, src1, dst2, dst3);
158 ILVRL_B2_SH(zero, src2, dst4, dst5);
159 ILVRL_B2_SH(zero, src3, dst6, dst7);
160
161 SLLI_4V(dst0, dst1, dst2, dst3, 6);
162 SLLI_4V(dst4, dst5, dst6, dst7, 6);
163
164 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
165 offset_vec, rnd_vec, dst0, dst1, dst2,
166 dst3);
167 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
168 offset_vec, rnd_vec, dst4, dst5, dst6,
169 dst7);
170 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
171 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
172
173 ST_W2(out0, 0, 2, dst, dst_stride);
174 ST_H2(out0, 2, 6, dst + 4, dst_stride);
175 ST_W2(out1, 0, 2, dst + 2 * dst_stride, dst_stride);
176 ST_H2(out1, 2, 6, dst + 2 * dst_stride + 4, dst_stride);
177 dst += (4 * dst_stride);
178 ST_W2(out2, 0, 2, dst, dst_stride);
179 ST_H2(out2, 2, 6, dst + 4, dst_stride);
180 ST_W2(out3, 0, 2, dst + 2 * dst_stride, dst_stride);
181 ST_H2(out3, 2, 6, dst + 2 * dst_stride + 4, dst_stride);
182 dst += (4 * dst_stride);
183 }
184 }
185
hevc_uniwgt_copy_8w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)186 static void hevc_uniwgt_copy_8w_msa(uint8_t *src,
187 int32_t src_stride,
188 uint8_t *dst,
189 int32_t dst_stride,
190 int32_t height,
191 int32_t weight,
192 int32_t offset,
193 int32_t rnd_val)
194 {
195 uint32_t loop_cnt;
196 uint64_t tp0, tp1, tp2, tp3;
197 v16i8 src0 = { 0 }, src1 = { 0 }, src2 = { 0 }, src3 = { 0 };
198 v16i8 zero = { 0 };
199 v16u8 out0, out1, out2, out3;
200 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, offset_vec;
201 v4i32 weight_vec, rnd_vec;
202
203 weight = weight & 0x0000FFFF;
204 weight_vec = __msa_fill_w(weight);
205 offset_vec = __msa_fill_h(offset);
206 rnd_vec = __msa_fill_w(rnd_val);
207
208 if (2 == height) {
209 LD2(src, src_stride, tp0, tp1);
210 INSERT_D2_SB(tp0, tp1, src0);
211 ILVRL_B2_SH(zero, src0, dst0, dst1);
212 SLLI_2V(dst0, dst1, 6);
213 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst0, dst1, weight_vec, offset_vec,
214 rnd_vec, dst0, dst1);
215 out0 = (v16u8) __msa_pckev_b((v16i8) dst1, (v16i8) dst0);
216 ST_D2(out0, 0, 1, dst, dst_stride);
217 } else if (4 == height) {
218 LD4(src, src_stride, tp0, tp1, tp2, tp3);
219 INSERT_D2_SB(tp0, tp1, src0);
220 INSERT_D2_SB(tp2, tp3, src1);
221 ILVRL_B2_SH(zero, src0, dst0, dst1);
222 ILVRL_B2_SH(zero, src1, dst2, dst3);
223 SLLI_4V(dst0, dst1, dst2, dst3, 6);
224 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
225 offset_vec, rnd_vec, dst0, dst1, dst2,
226 dst3);
227 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
228 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
229 } else if (6 == height) {
230 LD4(src, src_stride, tp0, tp1, tp2, tp3);
231 src += 4 * src_stride;
232 INSERT_D2_SB(tp0, tp1, src0);
233 INSERT_D2_SB(tp2, tp3, src1);
234 LD2(src, src_stride, tp0, tp1);
235 INSERT_D2_SB(tp0, tp1, src2);
236 ILVRL_B2_SH(zero, src0, dst0, dst1);
237 ILVRL_B2_SH(zero, src1, dst2, dst3);
238 ILVRL_B2_SH(zero, src2, dst4, dst5);
239 SLLI_4V(dst0, dst1, dst2, dst3, 6);
240 SLLI_2V(dst4, dst5, 6);
241 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
242 offset_vec, rnd_vec, dst0, dst1, dst2,
243 dst3);
244 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec,
245 rnd_vec, dst4, dst5);
246 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
247 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
248 ST_D2(out2, 0, 1, dst + 4 * dst_stride, dst_stride);
249 } else if (0 == height % 8) {
250 for (loop_cnt = (height >> 3); loop_cnt--;) {
251 LD4(src, src_stride, tp0, tp1, tp2, tp3);
252 src += 4 * src_stride;
253 INSERT_D2_SB(tp0, tp1, src0);
254 INSERT_D2_SB(tp2, tp3, src1);
255 LD4(src, src_stride, tp0, tp1, tp2, tp3);
256 src += 4 * src_stride;
257 INSERT_D2_SB(tp0, tp1, src2);
258 INSERT_D2_SB(tp2, tp3, src3);
259
260 ILVRL_B2_SH(zero, src0, dst0, dst1);
261 ILVRL_B2_SH(zero, src1, dst2, dst3);
262 ILVRL_B2_SH(zero, src2, dst4, dst5);
263 ILVRL_B2_SH(zero, src3, dst6, dst7);
264 SLLI_4V(dst0, dst1, dst2, dst3, 6);
265 SLLI_4V(dst4, dst5, dst6, dst7, 6);
266 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
267 offset_vec, rnd_vec, dst0, dst1,
268 dst2, dst3);
269 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
270 offset_vec, rnd_vec, dst4, dst5,
271 dst6, dst7);
272 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
273 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
274 ST_D8(out0, out1, out2, out3, 0, 1, 0, 1, 0, 1, 0, 1,
275 dst, dst_stride);
276 dst += (8 * dst_stride);
277 }
278 }
279 }
280
hevc_uniwgt_copy_12w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)281 static void hevc_uniwgt_copy_12w_msa(uint8_t *src,
282 int32_t src_stride,
283 uint8_t *dst,
284 int32_t dst_stride,
285 int32_t height,
286 int32_t weight,
287 int32_t offset,
288 int32_t rnd_val)
289 {
290 uint32_t loop_cnt;
291 v16u8 out0, out1, out2;
292 v16i8 src0, src1, src2, src3;
293 v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
294 v8i16 offset_vec;
295 v16i8 zero = { 0 };
296 v4i32 weight_vec, rnd_vec;
297
298 weight = weight & 0x0000FFFF;
299 weight_vec = __msa_fill_w(weight);
300 offset_vec = __msa_fill_h(offset);
301 rnd_vec = __msa_fill_w(rnd_val);
302
303 for (loop_cnt = 4; loop_cnt--;) {
304 LD_SB4(src, src_stride, src0, src1, src2, src3);
305 src += (4 * src_stride);
306 ILVR_B4_SH(zero, src0, zero, src1, zero, src2, zero, src3,
307 dst0, dst1, dst2, dst3);
308
309 ILVL_W2_SB(src1, src0, src3, src2, src0, src1);
310 ILVR_B2_SH(zero, src0, zero, src1, dst4, dst5);
311 SLLI_4V(dst0, dst1, dst2, dst3, 6);
312 SLLI_2V(dst4, dst5, 6);
313 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
314 offset_vec, rnd_vec, dst0, dst1, dst2,
315 dst3);
316 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec,
317 rnd_vec, dst4, dst5);
318
319 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
320 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
321 ST_W4(out2, 0, 1, 2, 3, dst + 8, dst_stride);
322 dst += (4 * dst_stride);
323 }
324 }
325
hevc_uniwgt_copy_16w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)326 static void hevc_uniwgt_copy_16w_msa(uint8_t *src,
327 int32_t src_stride,
328 uint8_t *dst,
329 int32_t dst_stride,
330 int32_t height,
331 int32_t weight,
332 int32_t offset,
333 int32_t rnd_val)
334 {
335 uint32_t loop_cnt;
336 v16u8 out0, out1, out2, out3;
337 v16i8 src0, src1, src2, src3;
338 v16i8 zero = { 0 };
339 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, offset_vec;
340 v4i32 weight_vec, rnd_vec;
341
342 weight = weight & 0x0000FFFF;
343 weight_vec = __msa_fill_w(weight);
344 offset_vec = __msa_fill_h(offset);
345 rnd_vec = __msa_fill_w(rnd_val);
346
347 for (loop_cnt = height >> 2; loop_cnt--;) {
348 LD_SB4(src, src_stride, src0, src1, src2, src3);
349 src += (4 * src_stride);
350 ILVRL_B2_SH(zero, src0, dst0, dst1);
351 ILVRL_B2_SH(zero, src1, dst2, dst3);
352 ILVRL_B2_SH(zero, src2, dst4, dst5);
353 ILVRL_B2_SH(zero, src3, dst6, dst7);
354 SLLI_4V(dst0, dst1, dst2, dst3, 6);
355 SLLI_4V(dst4, dst5, dst6, dst7, 6);
356 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
357 offset_vec, rnd_vec, dst0, dst1, dst2,
358 dst3);
359 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
360 offset_vec, rnd_vec, dst4, dst5, dst6,
361 dst7);
362 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
363 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
364 ST_UB4(out0, out1, out2, out3, dst, dst_stride);
365 dst += (4 * dst_stride);
366 }
367 }
368
hevc_uniwgt_copy_24w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)369 static void hevc_uniwgt_copy_24w_msa(uint8_t *src,
370 int32_t src_stride,
371 uint8_t *dst,
372 int32_t dst_stride,
373 int32_t height,
374 int32_t weight,
375 int32_t offset,
376 int32_t rnd_val)
377 {
378 uint32_t loop_cnt;
379 v16u8 out0, out1, out2, out3, out4, out5;
380 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
381 v16i8 zero = { 0 };
382 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, offset_vec;
383 v8i16 dst8, dst9, dst10, dst11;
384 v4i32 weight_vec, rnd_vec;
385
386 weight = weight & 0x0000FFFF;
387 weight_vec = __msa_fill_w(weight);
388 offset_vec = __msa_fill_h(offset);
389 rnd_vec = __msa_fill_w(rnd_val);
390
391 for (loop_cnt = (height >> 2); loop_cnt--;) {
392 LD_SB4(src, src_stride, src0, src1, src4, src5);
393 LD_SB4(src + 16, src_stride, src2, src3, src6, src7);
394 src += (4 * src_stride);
395
396 ILVRL_B2_SH(zero, src0, dst0, dst1);
397 ILVRL_B2_SH(zero, src1, dst2, dst3);
398 ILVR_B2_SH(zero, src2, zero, src3, dst4, dst5);
399 ILVRL_B2_SH(zero, src4, dst6, dst7);
400 ILVRL_B2_SH(zero, src5, dst8, dst9);
401 ILVR_B2_SH(zero, src6, zero, src7, dst10, dst11);
402 SLLI_4V(dst0, dst1, dst2, dst3, 6);
403 SLLI_4V(dst4, dst5, dst6, dst7, 6);
404 SLLI_4V(dst8, dst9, dst10, dst11, 6);
405 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
406 offset_vec, rnd_vec, dst0, dst1, dst2,
407 dst3);
408 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
409 offset_vec, rnd_vec, dst4, dst5, dst6,
410 dst7);
411 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst8, dst9, dst10, dst11, weight_vec,
412 offset_vec, rnd_vec, dst8, dst9, dst10,
413 dst11);
414 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
415 PCKEV_B3_UB(dst7, dst6, dst9, dst8, dst11, dst10, out3, out4, out5);
416 ST_UB4(out0, out1, out3, out4, dst, dst_stride);
417 ST_D4(out2, out5, 0, 1, 0, 1, dst + 16, dst_stride);
418 dst += (4 * dst_stride);
419 }
420 }
421
hevc_uniwgt_copy_32w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)422 static void hevc_uniwgt_copy_32w_msa(uint8_t *src,
423 int32_t src_stride,
424 uint8_t *dst,
425 int32_t dst_stride,
426 int32_t height,
427 int32_t weight,
428 int32_t offset,
429 int32_t rnd_val)
430 {
431 uint32_t loop_cnt;
432 v16u8 out0, out1, out2, out3;
433 v16i8 src0, src1, src2, src3;
434 v16i8 zero = { 0 };
435 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, offset_vec;
436 v4i32 weight_vec, rnd_vec;
437
438 weight = weight & 0x0000FFFF;
439 weight_vec = __msa_fill_w(weight);
440 offset_vec = __msa_fill_h(offset);
441 rnd_vec = __msa_fill_w(rnd_val);
442
443 for (loop_cnt = (height >> 1); loop_cnt--;) {
444 LD_SB2(src, src_stride, src0, src1);
445 LD_SB2(src + 16, src_stride, src2, src3);
446 src += (2 * src_stride);
447
448 ILVRL_B2_SH(zero, src0, dst0, dst1);
449 ILVRL_B2_SH(zero, src1, dst2, dst3);
450 ILVRL_B2_SH(zero, src2, dst4, dst5);
451 ILVRL_B2_SH(zero, src3, dst6, dst7);
452 SLLI_4V(dst0, dst1, dst2, dst3, 6);
453 SLLI_4V(dst4, dst5, dst6, dst7, 6);
454 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
455 offset_vec, rnd_vec, dst0, dst1, dst2,
456 dst3);
457 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
458 offset_vec, rnd_vec, dst4, dst5, dst6,
459 dst7);
460 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
461 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
462 ST_UB2(out0, out1, dst, dst_stride);
463 ST_UB2(out2, out3, dst + 16, dst_stride);
464 dst += (2 * dst_stride);
465 }
466 }
467
hevc_uniwgt_copy_48w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)468 static void hevc_uniwgt_copy_48w_msa(uint8_t *src,
469 int32_t src_stride,
470 uint8_t *dst,
471 int32_t dst_stride,
472 int32_t height,
473 int32_t weight,
474 int32_t offset,
475 int32_t rnd_val)
476 {
477 uint32_t loop_cnt;
478 v16u8 out0, out1, out2, out3, out4, out5;
479 v16i8 src0, src1, src2, src3, src4, src5;
480 v16i8 zero = { 0 };
481 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, offset_vec;
482 v8i16 dst6, dst7, dst8, dst9, dst10, dst11;
483 v4i32 weight_vec, rnd_vec;
484
485 weight = weight & 0x0000FFFF;
486 weight_vec = __msa_fill_w(weight);
487 offset_vec = __msa_fill_h(offset);
488 rnd_vec = __msa_fill_w(rnd_val);
489
490 for (loop_cnt = (height >> 1); loop_cnt--;) {
491 LD_SB3(src, 16, src0, src1, src2);
492 src += src_stride;
493 LD_SB3(src, 16, src3, src4, src5);
494 src += src_stride;
495
496 ILVRL_B2_SH(zero, src0, dst0, dst1);
497 ILVRL_B2_SH(zero, src1, dst2, dst3);
498 ILVRL_B2_SH(zero, src2, dst4, dst5);
499 ILVRL_B2_SH(zero, src3, dst6, dst7);
500 ILVRL_B2_SH(zero, src4, dst8, dst9);
501 ILVRL_B2_SH(zero, src5, dst10, dst11);
502 SLLI_4V(dst0, dst1, dst2, dst3, 6);
503 SLLI_4V(dst4, dst5, dst6, dst7, 6);
504 SLLI_4V(dst8, dst9, dst10, dst11, 6);
505 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
506 offset_vec, rnd_vec, dst0, dst1, dst2,
507 dst3);
508 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
509 offset_vec, rnd_vec, dst4, dst5, dst6,
510 dst7);
511 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst8, dst9, dst10, dst11, weight_vec,
512 offset_vec, rnd_vec, dst8, dst9, dst10,
513 dst11);
514 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
515 PCKEV_B3_UB(dst7, dst6, dst9, dst8, dst11, dst10, out3, out4, out5);
516 ST_UB2(out0, out1, dst, 16);
517 ST_UB(out2, dst + 32);
518 dst += dst_stride;
519 ST_UB2(out3, out4, dst, 16);
520 ST_UB(out5, dst + 32);
521 dst += dst_stride;
522 }
523 }
524
hevc_uniwgt_copy_64w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)525 static void hevc_uniwgt_copy_64w_msa(uint8_t *src,
526 int32_t src_stride,
527 uint8_t *dst,
528 int32_t dst_stride,
529 int32_t height,
530 int32_t weight,
531 int32_t offset,
532 int32_t rnd_val)
533 {
534 uint32_t loop_cnt;
535 v16u8 out0, out1, out2, out3, out4, out5, out6, out7;
536 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
537 v16i8 zero = { 0 };
538 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, offset_vec;
539 v8i16 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
540 v4i32 weight_vec, rnd_vec;
541
542 weight = weight & 0x0000FFFF;
543 weight_vec = __msa_fill_w(weight);
544 offset_vec = __msa_fill_h(offset);
545 rnd_vec = __msa_fill_w(rnd_val);
546
547 for (loop_cnt = (height >> 1); loop_cnt--;) {
548 LD_SB4(src, 16, src0, src1, src2, src3);
549 src += src_stride;
550 LD_SB4(src, 16, src4, src5, src6, src7);
551 src += src_stride;
552
553 ILVRL_B2_SH(zero, src0, dst0, dst1);
554 ILVRL_B2_SH(zero, src1, dst2, dst3);
555 ILVRL_B2_SH(zero, src2, dst4, dst5);
556 ILVRL_B2_SH(zero, src3, dst6, dst7);
557 ILVRL_B2_SH(zero, src4, dst8, dst9);
558 ILVRL_B2_SH(zero, src5, dst10, dst11);
559 ILVRL_B2_SH(zero, src6, dst12, dst13);
560 ILVRL_B2_SH(zero, src7, dst14, dst15);
561 SLLI_4V(dst0, dst1, dst2, dst3, 6);
562 SLLI_4V(dst4, dst5, dst6, dst7, 6);
563 SLLI_4V(dst8, dst9, dst10, dst11, 6);
564 SLLI_4V(dst12, dst13, dst14, dst15, 6);
565 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
566 offset_vec, rnd_vec, dst0, dst1, dst2,
567 dst3);
568 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
569 offset_vec, rnd_vec, dst4, dst5, dst6,
570 dst7);
571 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst8, dst9, dst10, dst11, weight_vec,
572 offset_vec, rnd_vec, dst8, dst9, dst10,
573 dst11);
574 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst12, dst13, dst14, dst15, weight_vec,
575 offset_vec, rnd_vec, dst12, dst13, dst14,
576 dst15);
577 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
578 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
579 PCKEV_B2_UB(dst9, dst8, dst11, dst10, out4, out5);
580 PCKEV_B2_UB(dst13, dst12, dst15, dst14, out6, out7);
581 ST_UB4(out0, out1, out2, out3, dst, 16);
582 dst += dst_stride;
583 ST_UB4(out4, out5, out6, out7, dst, 16);
584 dst += dst_stride;
585 }
586 }
587
hevc_hz_uniwgt_8t_4w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)588 static void hevc_hz_uniwgt_8t_4w_msa(uint8_t *src,
589 int32_t src_stride,
590 uint8_t *dst,
591 int32_t dst_stride,
592 const int8_t *filter,
593 int32_t height,
594 int32_t weight,
595 int32_t offset,
596 int32_t rnd_val)
597 {
598 uint32_t loop_cnt;
599 v16u8 out0, out1;
600 v8i16 filt0, filt1, filt2, filt3;
601 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
602 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9, vec10;
603 v16i8 mask0, mask1, mask2, mask3, vec11, vec12, vec13, vec14, vec15;
604 v8i16 filter_vec, dst01, dst23, dst45, dst67;
605 v8i16 dst0, dst1, dst2, dst3, weight_vec_h, offset_vec, denom_vec;
606 v4i32 weight_vec, rnd_vec;
607
608 src -= 3;
609 weight = weight & 0x0000FFFF;
610
611 weight_vec = __msa_fill_w(weight);
612 rnd_vec = __msa_fill_w(rnd_val);
613
614 weight *= 128;
615 rnd_val -= 6;
616
617 weight_vec_h = __msa_fill_h(weight);
618 offset_vec = __msa_fill_h(offset);
619 denom_vec = __msa_fill_h(rnd_val);
620
621 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
622 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
623
624 filter_vec = LD_SH(filter);
625 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
626
627 mask0 = LD_SB(&ff_hevc_mask_arr[16]);
628 mask1 = mask0 + 2;
629 mask2 = mask0 + 4;
630 mask3 = mask0 + 6;
631
632 for (loop_cnt = (height >> 3); loop_cnt--;) {
633 LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
634 src += (8 * src_stride);
635 XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
636
637 VSHF_B4_SB(src0, src1, mask0, mask1, mask2, mask3,
638 vec0, vec1, vec2, vec3);
639 VSHF_B4_SB(src2, src3, mask0, mask1, mask2, mask3,
640 vec4, vec5, vec6, vec7);
641 VSHF_B4_SB(src4, src5, mask0, mask1, mask2, mask3,
642 vec8, vec9, vec10, vec11);
643 VSHF_B4_SB(src6, src7, mask0, mask1, mask2, mask3,
644 vec12, vec13, vec14, vec15);
645 dst01 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
646 filt3);
647 dst23 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
648 filt3);
649 dst45 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
650 filt3);
651 dst67 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
652 filt2, filt3);
653
654 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst01, dst23, dst45, dst67, weight_vec,
655 offset_vec, rnd_vec, dst0, dst1, dst2,
656 dst3);
657
658 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
659 ST_W8(out0, out1, 0, 1, 2, 3, 0, 1, 2, 3, dst, dst_stride);
660 dst += (8 * dst_stride);
661 }
662 }
663
hevc_hz_uniwgt_8t_8w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)664 static void hevc_hz_uniwgt_8t_8w_msa(uint8_t *src,
665 int32_t src_stride,
666 uint8_t *dst,
667 int32_t dst_stride,
668 const int8_t *filter,
669 int32_t height,
670 int32_t weight,
671 int32_t offset,
672 int32_t rnd_val)
673 {
674 uint32_t loop_cnt;
675 v16u8 out0, out1;
676 v16i8 src0, src1, src2, src3;
677 v8i16 filt0, filt1, filt2, filt3;
678 v16i8 mask0, mask1, mask2, mask3;
679 v8i16 filter_vec;
680 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
681 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
682 v8i16 dst0, dst1, dst2, dst3;
683 v8i16 weight_vec_h, offset_vec, denom_vec;
684 v4i32 weight_vec, rnd_vec;
685
686 src -= 3;
687 weight = weight & 0x0000FFFF;
688
689 weight_vec = __msa_fill_w(weight);
690 rnd_vec = __msa_fill_w(rnd_val);
691
692 weight *= 128;
693 rnd_val -= 6;
694
695 weight_vec_h = __msa_fill_h(weight);
696 offset_vec = __msa_fill_h(offset);
697 denom_vec = __msa_fill_h(rnd_val);
698
699 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
700 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
701
702 filter_vec = LD_SH(filter);
703 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
704
705 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
706 mask1 = mask0 + 2;
707 mask2 = mask0 + 4;
708 mask3 = mask0 + 6;
709
710 for (loop_cnt = (height >> 2); loop_cnt--;) {
711 LD_SB4(src, src_stride, src0, src1, src2, src3);
712 src += (4 * src_stride);
713 XORI_B4_128_SB(src0, src1, src2, src3);
714
715 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
716 vec0, vec1, vec2, vec3);
717 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
718 vec4, vec5, vec6, vec7);
719 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
720 vec8, vec9, vec10, vec11);
721 VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
722 vec12, vec13, vec14, vec15);
723 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
724 filt3);
725 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
726 filt3);
727 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
728 filt3);
729 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
730 filt2, filt3);
731
732 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
733 offset_vec, rnd_vec, dst0, dst1, dst2,
734 dst3);
735
736 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
737 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
738 dst += (4 * dst_stride);
739 }
740 }
741
hevc_hz_uniwgt_8t_12w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)742 static void hevc_hz_uniwgt_8t_12w_msa(uint8_t *src,
743 int32_t src_stride,
744 uint8_t *dst,
745 int32_t dst_stride,
746 const int8_t *filter,
747 int32_t height,
748 int32_t weight,
749 int32_t offset,
750 int32_t rnd_val)
751 {
752 uint32_t loop_cnt;
753 v16u8 out0, out1, out2;
754 v8i16 filt0, filt1, filt2, filt3;
755 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
756 v16i8 mask0, mask1, mask2, mask3, mask4, mask5, mask6, mask7;
757 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
758 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
759 v8i16 filter_vec;
760 v8i16 dst01, dst23, dst0, dst1, dst2, dst3, dst4, dst5;
761 v8i16 weight_vec_h, offset_vec, denom_vec;
762 v4i32 weight_vec, rnd_vec;
763
764 src -= 3;
765 weight = weight & 0x0000FFFF;
766
767 weight_vec = __msa_fill_w(weight);
768 rnd_vec = __msa_fill_w(rnd_val);
769
770 weight *= 128;
771 rnd_val -= 6;
772
773 weight_vec_h = __msa_fill_h(weight);
774 offset_vec = __msa_fill_h(offset);
775 denom_vec = __msa_fill_h(rnd_val);
776
777 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
778 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
779
780 filter_vec = LD_SH(filter);
781 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
782
783 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
784 mask1 = mask0 + 2;
785 mask2 = mask0 + 4;
786 mask3 = mask0 + 6;
787 mask4 = LD_SB(&ff_hevc_mask_arr[16]);
788 mask5 = mask4 + 2;
789 mask6 = mask4 + 4;
790 mask7 = mask4 + 6;
791
792 for (loop_cnt = (height >> 2); loop_cnt--;) {
793 LD_SB4(src, src_stride, src0, src1, src2, src3);
794 LD_SB4(src + 8, src_stride, src4, src5, src6, src7);
795 src += (4 * src_stride);
796 XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
797
798 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
799 vec0, vec1, vec2, vec3);
800 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
801 vec4, vec5, vec6, vec7);
802 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
803 vec8, vec9, vec10, vec11);
804 VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
805 vec12, vec13, vec14, vec15);
806 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
807 filt3);
808 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
809 filt3);
810 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
811 filt3);
812 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
813 filt2, filt3);
814 VSHF_B4_SB(src4, src5, mask4, mask5, mask6, mask7,
815 vec0, vec1, vec2, vec3);
816 VSHF_B4_SB(src6, src7, mask4, mask5, mask6, mask7,
817 vec4, vec5, vec6, vec7);
818 dst01 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
819 filt3);
820 dst23 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
821 filt3);
822
823 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
824 offset_vec, rnd_vec, dst0, dst1, dst2,
825 dst3);
826 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst01, dst23, weight_vec, offset_vec,
827 rnd_vec, dst4, dst5);
828
829 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
830 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
831 ST_W4(out2, 0, 1, 2, 3, dst + 8, dst_stride);
832 dst += (4 * dst_stride);
833 }
834 }
835
hevc_hz_uniwgt_8t_16w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)836 static void hevc_hz_uniwgt_8t_16w_msa(uint8_t *src,
837 int32_t src_stride,
838 uint8_t *dst,
839 int32_t dst_stride,
840 const int8_t *filter,
841 int32_t height,
842 int32_t weight,
843 int32_t offset,
844 int32_t rnd_val)
845 {
846 uint32_t loop_cnt;
847 v16u8 out0, out1;
848 v16i8 src0, src1, src2, src3;
849 v8i16 filt0, filt1, filt2, filt3;
850 v16i8 mask0, mask1, mask2, mask3;
851 v8i16 filter_vec;
852 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
853 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
854 v8i16 dst0, dst1, dst2, dst3;
855 v8i16 weight_vec_h, offset_vec, denom_vec;
856 v4i32 weight_vec, rnd_vec;
857
858 src -= 3;
859
860 weight_vec = __msa_fill_w(weight);
861 rnd_vec = __msa_fill_w(rnd_val);
862
863 weight *= 128;
864 rnd_val -= 6;
865
866 weight_vec_h = __msa_fill_h(weight);
867 offset_vec = __msa_fill_h(offset);
868 denom_vec = __msa_fill_h(rnd_val);
869
870 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
871 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
872
873 filter_vec = LD_SH(filter);
874 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
875
876 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
877 mask1 = mask0 + 2;
878 mask2 = mask0 + 4;
879 mask3 = mask0 + 6;
880
881 for (loop_cnt = (height >> 1); loop_cnt--;) {
882 LD_SB2(src, src_stride, src0, src2);
883 LD_SB2(src + 8, src_stride, src1, src3);
884 src += (2 * src_stride);
885 XORI_B4_128_SB(src0, src1, src2, src3);
886
887 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
888 vec0, vec1, vec2, vec3);
889 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
890 vec4, vec5, vec6, vec7);
891 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
892 vec8, vec9, vec10, vec11);
893 VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
894 vec12, vec13, vec14, vec15);
895 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
896 filt3);
897 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
898 filt3);
899 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
900 filt3);
901 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
902 filt2, filt3);
903
904 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
905 offset_vec, rnd_vec, dst0, dst1, dst2,
906 dst3);
907
908 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
909 ST_UB2(out0, out1, dst, dst_stride);
910 dst += (2 * dst_stride);
911 }
912 }
913
hevc_hz_uniwgt_8t_24w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)914 static void hevc_hz_uniwgt_8t_24w_msa(uint8_t *src,
915 int32_t src_stride,
916 uint8_t *dst,
917 int32_t dst_stride,
918 const int8_t *filter,
919 int32_t height,
920 int32_t weight,
921 int32_t offset,
922 int32_t rnd_val)
923 {
924 uint32_t loop_cnt;
925 v16u8 out0, out1, out2;
926 v16i8 src0, src1, src2, src3;
927 v8i16 filt0, filt1, filt2, filt3;
928 v16i8 mask0, mask1, mask2, mask3, mask4, mask5, mask6, mask7;
929 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
930 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
931 v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
932 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
933 v4i32 weight_vec, rnd_vec;
934
935 src -= 3;
936
937 weight_vec = __msa_fill_w(weight);
938 rnd_vec = __msa_fill_w(rnd_val);
939
940 weight *= 128;
941 rnd_val -= 6;
942
943 weight_vec_h = __msa_fill_h(weight);
944 offset_vec = __msa_fill_h(offset);
945 denom_vec = __msa_fill_h(rnd_val);
946
947 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
948 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
949
950 filter_vec = LD_SH(filter);
951 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
952
953 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
954 mask1 = mask0 + 2;
955 mask2 = mask0 + 4;
956 mask3 = mask0 + 6;
957 mask4 = mask0 + 8;
958 mask5 = mask0 + 10;
959 mask6 = mask0 + 12;
960 mask7 = mask0 + 14;
961
962 for (loop_cnt = 16; loop_cnt--;) {
963 LD_SB2(src, 16, src0, src1);
964 src += src_stride;
965 LD_SB2(src, 16, src2, src3);
966 src += src_stride;
967 XORI_B4_128_SB(src0, src1, src2, src3);
968 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
969 vec0, vec1, vec2, vec3);
970 VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7,
971 vec4, vec5, vec6, vec7);
972 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
973 vec8, vec9, vec10, vec11);
974 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
975 vec12, vec13, vec14, vec15);
976 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
977 filt3);
978 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
979 filt3);
980 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
981 filt3);
982 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
983 filt2, filt3);
984
985 VSHF_B4_SB(src2, src3, mask4, mask5, mask6, mask7,
986 vec0, vec1, vec2, vec3);
987 VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
988 vec4, vec5, vec6, vec7);
989 dst4 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
990 filt3);
991 dst5 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
992 filt3);
993
994 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
995 offset_vec, rnd_vec, dst0, dst1, dst2,
996 dst3);
997 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec,
998 rnd_vec, dst4, dst5);
999
1000 PCKEV_B3_UB(dst1, dst0, dst4, dst3, dst5, dst2, out0, out1, out2);
1001 ST_UB2(out0, out1, dst, dst_stride);
1002 ST_D2(out2, 0, 1, dst + 16, dst_stride);
1003 dst += (2 * dst_stride);
1004 }
1005 }
1006
hevc_hz_uniwgt_8t_32w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1007 static void hevc_hz_uniwgt_8t_32w_msa(uint8_t *src,
1008 int32_t src_stride,
1009 uint8_t *dst,
1010 int32_t dst_stride,
1011 const int8_t *filter,
1012 int32_t height,
1013 int32_t weight,
1014 int32_t offset,
1015 int32_t rnd_val)
1016 {
1017 uint32_t loop_cnt;
1018 v16u8 out0, out1, out2, out3;
1019 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
1020 v8i16 filt0, filt1, filt2, filt3;
1021 v16i8 mask0, mask1, mask2, mask3;
1022 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
1023 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
1024 v8i16 filter_vec;
1025 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
1026 v8i16 weight_vec_h, offset_vec, denom_vec;
1027 v4i32 weight_vec, rnd_vec;
1028
1029 src -= 3;
1030
1031 weight_vec = __msa_fill_w(weight);
1032 rnd_vec = __msa_fill_w(rnd_val);
1033
1034 weight *= 128;
1035 rnd_val -= 6;
1036
1037 weight_vec_h = __msa_fill_h(weight);
1038 offset_vec = __msa_fill_h(offset);
1039 denom_vec = __msa_fill_h(rnd_val);
1040
1041 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
1042 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
1043
1044 filter_vec = LD_SH(filter);
1045 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1046
1047 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
1048 mask1 = mask0 + 2;
1049 mask2 = mask0 + 4;
1050 mask3 = mask0 + 6;
1051
1052 for (loop_cnt = height >> 1; loop_cnt--;) {
1053 LD_SB4(src, 8, src0, src1, src2, src3);
1054 src += src_stride;
1055 LD_SB4(src, 8, src4, src5, src6, src7);
1056 src += src_stride;
1057 XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
1058
1059 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
1060 vec0, vec1, vec2, vec3);
1061 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
1062 vec4, vec5, vec6, vec7);
1063 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
1064 vec8, vec9, vec10, vec11);
1065 VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
1066 vec12, vec13, vec14, vec15);
1067 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
1068 filt3);
1069 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
1070 filt3);
1071 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
1072 filt3);
1073 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
1074 filt2, filt3);
1075
1076 VSHF_B4_SB(src4, src4, mask0, mask1, mask2, mask3,
1077 vec0, vec1, vec2, vec3);
1078 VSHF_B4_SB(src5, src5, mask0, mask1, mask2, mask3,
1079 vec4, vec5, vec6, vec7);
1080 VSHF_B4_SB(src6, src6, mask0, mask1, mask2, mask3,
1081 vec8, vec9, vec10, vec11);
1082 VSHF_B4_SB(src7, src7, mask0, mask1, mask2, mask3,
1083 vec12, vec13, vec14, vec15);
1084 dst4 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
1085 filt3);
1086 dst5 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
1087 filt3);
1088 dst6 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
1089 filt3);
1090 dst7 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
1091 filt2, filt3);
1092
1093 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
1094 offset_vec, rnd_vec, dst0, dst1, dst2,
1095 dst3);
1096 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
1097 offset_vec, rnd_vec, dst4, dst5, dst6,
1098 dst7);
1099
1100 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
1101 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
1102 ST_UB2(out0, out1, dst, 16);
1103 dst += dst_stride;
1104 ST_UB2(out2, out3, dst, 16);
1105 dst += dst_stride;
1106 }
1107 }
1108
hevc_hz_uniwgt_8t_48w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1109 static void hevc_hz_uniwgt_8t_48w_msa(uint8_t *src,
1110 int32_t src_stride,
1111 uint8_t *dst,
1112 int32_t dst_stride,
1113 const int8_t *filter,
1114 int32_t height,
1115 int32_t weight,
1116 int32_t offset,
1117 int32_t rnd_val)
1118 {
1119 uint32_t loop_cnt;
1120 v16u8 out0, out1, out2;
1121 v16i8 src0, src1, src2, src3;
1122 v8i16 filt0, filt1, filt2, filt3;
1123 v16i8 mask0, mask1, mask2, mask3, mask4, mask5, mask6, mask7;
1124 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
1125 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
1126 v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
1127 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
1128 v4i32 weight_vec, rnd_vec;
1129
1130 src -= 3;
1131
1132 weight = weight & 0x0000FFFF;
1133 weight_vec = __msa_fill_w(weight);
1134 rnd_vec = __msa_fill_w(rnd_val);
1135
1136 weight *= 128;
1137 rnd_val -= 6;
1138
1139 weight_vec_h = __msa_fill_h(weight);
1140 offset_vec = __msa_fill_h(offset);
1141 denom_vec = __msa_fill_h(rnd_val);
1142
1143 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
1144 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
1145
1146 filter_vec = LD_SH(filter);
1147 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1148
1149 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
1150 mask1 = mask0 + 2;
1151 mask2 = mask0 + 4;
1152 mask3 = mask0 + 6;
1153 mask4 = mask0 + 8;
1154 mask5 = mask0 + 10;
1155 mask6 = mask0 + 12;
1156 mask7 = mask0 + 14;
1157
1158 for (loop_cnt = 64; loop_cnt--;) {
1159 LD_SB3(src, 16, src0, src1, src2);
1160 src3 = LD_SB(src + 40);
1161 src += src_stride;
1162 XORI_B4_128_SB(src0, src1, src2, src3);
1163
1164 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
1165 vec0, vec1, vec2, vec3);
1166 VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7,
1167 vec4, vec5, vec6, vec7);
1168 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
1169 vec8, vec9, vec10, vec11);
1170 VSHF_B4_SB(src1, src2, mask4, mask5, mask6, mask7,
1171 vec12, vec13, vec14, vec15);
1172 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
1173 filt3);
1174 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
1175 filt3);
1176 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
1177 filt3);
1178 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
1179 filt2, filt3);
1180
1181 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
1182 vec0, vec1, vec2, vec3);
1183 VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
1184 vec4, vec5, vec6, vec7);
1185 dst4 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
1186 filt3);
1187 dst5 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
1188 filt3);
1189
1190 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
1191 offset_vec, rnd_vec, dst0, dst1, dst2,
1192 dst3);
1193 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec,
1194 rnd_vec, dst4, dst5);
1195
1196 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
1197 ST_UB2(out0, out1, dst, 16);
1198 ST_UB(out2, dst + 32);
1199 dst += dst_stride;
1200 }
1201 }
1202
hevc_hz_uniwgt_8t_64w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1203 static void hevc_hz_uniwgt_8t_64w_msa(uint8_t *src,
1204 int32_t src_stride,
1205 uint8_t *dst,
1206 int32_t dst_stride,
1207 const int8_t *filter,
1208 int32_t height,
1209 int32_t weight,
1210 int32_t offset,
1211 int32_t rnd_val)
1212 {
1213 uint8_t *src_tmp;
1214 uint8_t *dst_tmp;
1215 uint32_t loop_cnt, cnt;
1216 v16u8 out0, out1;
1217 v16i8 src0, src1, src2;
1218 v8i16 filt0, filt1, filt2, filt3;
1219 v16i8 mask0, mask1, mask2, mask3, mask4, mask5, mask6, mask7;
1220 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
1221 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
1222 v8i16 dst0, dst1, dst2, dst3;
1223 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
1224 v4i32 weight_vec, rnd_vec;
1225
1226 src -= 3;
1227
1228 weight_vec = __msa_fill_w(weight);
1229 rnd_vec = __msa_fill_w(rnd_val);
1230
1231 weight *= 128;
1232 rnd_val -= 6;
1233
1234 weight_vec_h = __msa_fill_h(weight);
1235 offset_vec = __msa_fill_h(offset);
1236 denom_vec = __msa_fill_h(rnd_val);
1237
1238 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
1239 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
1240
1241 filter_vec = LD_SH(filter);
1242 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1243
1244 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
1245 mask1 = mask0 + 2;
1246 mask2 = mask0 + 4;
1247 mask3 = mask0 + 6;
1248 mask4 = mask0 + 8;
1249 mask5 = mask0 + 10;
1250 mask6 = mask0 + 12;
1251 mask7 = mask0 + 14;
1252
1253 for (loop_cnt = height; loop_cnt--;) {
1254 src_tmp = src;
1255 dst_tmp = dst;
1256
1257 for (cnt = 2; cnt--;) {
1258 LD_SB2(src_tmp, 16, src0, src1);
1259 src2 = LD_SB(src_tmp + 24);
1260 src_tmp += 32;
1261 XORI_B3_128_SB(src0, src1, src2);
1262
1263 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
1264 vec0, vec1, vec2, vec3);
1265 VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7,
1266 vec4, vec5, vec6, vec7);
1267 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
1268 vec8, vec9, vec10, vec11);
1269 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
1270 vec12, vec13, vec14, vec15);
1271 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1,
1272 filt2, filt3);
1273 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1,
1274 filt2, filt3);
1275 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1,
1276 filt2, filt3);
1277 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
1278 filt2, filt3);
1279
1280 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
1281 offset_vec, rnd_vec, dst0, dst1,
1282 dst2, dst3);
1283
1284 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
1285 ST_UB2(out0, out1, dst_tmp, 16);
1286 dst_tmp += 32;
1287 }
1288
1289 src += src_stride;
1290 dst += dst_stride;
1291 }
1292 }
1293
hevc_vt_uniwgt_8t_4w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1294 static void hevc_vt_uniwgt_8t_4w_msa(uint8_t *src,
1295 int32_t src_stride,
1296 uint8_t *dst,
1297 int32_t dst_stride,
1298 const int8_t *filter,
1299 int32_t height,
1300 int32_t weight,
1301 int32_t offset,
1302 int32_t rnd_val)
1303 {
1304 int32_t loop_cnt;
1305 v16u8 out0, out1;
1306 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
1307 v16i8 src9, src10, src11, src12, src13, src14;
1308 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r;
1309 v16i8 src21_r, src43_r, src65_r, src87_r, src109_r;
1310 v16i8 src1110_r, src1211_r, src1312_r, src1413_r;
1311 v16i8 src2110, src4332, src6554, src8776, src10998;
1312 v16i8 src12111110, src14131312;
1313 v8i16 filter_vec, dst01, dst23, dst45, dst67;
1314 v8i16 filt0, filt1, filt2, filt3;
1315 v8i16 dst0, dst1, dst2, dst3, weight_vec_h, offset_vec, denom_vec;
1316 v4i32 weight_vec, rnd_vec;
1317
1318 src -= (3 * src_stride);
1319
1320
1321 weight_vec = __msa_fill_w(weight);
1322 rnd_vec = __msa_fill_w(rnd_val);
1323
1324 weight *= 128;
1325 rnd_val -= 6;
1326
1327 weight_vec_h = __msa_fill_h(weight);
1328 offset_vec = __msa_fill_h(offset);
1329 denom_vec = __msa_fill_h(rnd_val);
1330
1331 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
1332 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
1333
1334 filter_vec = LD_SH(filter);
1335 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1336
1337 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
1338 src += (7 * src_stride);
1339
1340 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
1341 src10_r, src32_r, src54_r, src21_r);
1342
1343 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
1344
1345 ILVR_D3_SB(src21_r, src10_r, src43_r,
1346 src32_r, src65_r, src54_r, src2110, src4332, src6554);
1347
1348 XORI_B3_128_SB(src2110, src4332, src6554);
1349
1350 for (loop_cnt = (height >> 3); loop_cnt--;) {
1351 LD_SB8(src, src_stride,
1352 src7, src8, src9, src10, src11, src12, src13, src14);
1353 src += (8 * src_stride);
1354 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
1355 src76_r, src87_r, src98_r, src109_r);
1356 ILVR_B4_SB(src11, src10, src12, src11, src13, src12, src14, src13,
1357 src1110_r, src1211_r, src1312_r, src1413_r);
1358 ILVR_D4_SB(src87_r, src76_r, src109_r, src98_r, src1211_r, src1110_r,
1359 src1413_r, src1312_r,
1360 src8776, src10998, src12111110, src14131312);
1361 XORI_B4_128_SB(src8776, src10998, src12111110, src14131312);
1362 dst01 = HEVC_FILT_8TAP_SH(src2110, src4332, src6554, src8776, filt0,
1363 filt1, filt2, filt3);
1364 dst23 = HEVC_FILT_8TAP_SH(src4332, src6554, src8776, src10998, filt0,
1365 filt1, filt2, filt3);
1366 dst45 = HEVC_FILT_8TAP_SH(src6554, src8776, src10998, src12111110,
1367 filt0, filt1, filt2, filt3);
1368 dst67 = HEVC_FILT_8TAP_SH(src8776, src10998, src12111110, src14131312,
1369 filt0, filt1, filt2, filt3);
1370
1371 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst01, dst23, dst45, dst67, weight_vec,
1372 offset_vec, rnd_vec, dst0, dst1, dst2,
1373 dst3);
1374
1375 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
1376 ST_W8(out0, out1, 0, 1, 2, 3, 0, 1, 2, 3, dst, dst_stride);
1377 dst += (8 * dst_stride);
1378
1379 src2110 = src10998;
1380 src4332 = src12111110;
1381 src6554 = src14131312;
1382 src6 = src14;
1383 }
1384 }
1385
hevc_vt_uniwgt_8t_8w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1386 static void hevc_vt_uniwgt_8t_8w_msa(uint8_t *src,
1387 int32_t src_stride,
1388 uint8_t *dst,
1389 int32_t dst_stride,
1390 const int8_t *filter,
1391 int32_t height,
1392 int32_t weight,
1393 int32_t offset,
1394 int32_t rnd_val)
1395 {
1396 int32_t loop_cnt;
1397 v16u8 out0, out1;
1398 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
1399 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r;
1400 v16i8 src21_r, src43_r, src65_r, src87_r, src109_r;
1401 v8i16 filt0, filt1, filt2, filt3;
1402 v8i16 filter_vec;
1403 v8i16 dst0, dst1, dst2, dst3, weight_vec_h, offset_vec, denom_vec;
1404 v4i32 weight_vec, rnd_vec;
1405
1406 src -= (3 * src_stride);
1407
1408 weight_vec = __msa_fill_w(weight);
1409 rnd_vec = __msa_fill_w(rnd_val);
1410
1411 weight *= 128;
1412 rnd_val -= 6;
1413
1414 weight_vec_h = __msa_fill_h(weight);
1415 offset_vec = __msa_fill_h(offset);
1416 denom_vec = __msa_fill_h(rnd_val);
1417
1418 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
1419 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
1420
1421 filter_vec = LD_SH(filter);
1422 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1423
1424 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
1425 src += (7 * src_stride);
1426 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
1427
1428 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
1429 src10_r, src32_r, src54_r, src21_r);
1430 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
1431
1432 for (loop_cnt = (height >> 2); loop_cnt--;) {
1433 LD_SB4(src, src_stride, src7, src8, src9, src10);
1434 src += (4 * src_stride);
1435 XORI_B4_128_SB(src7, src8, src9, src10);
1436 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
1437 src76_r, src87_r, src98_r, src109_r);
1438 dst0 = HEVC_FILT_8TAP_SH(src10_r, src32_r, src54_r, src76_r, filt0,
1439 filt1, filt2, filt3);
1440 dst1 = HEVC_FILT_8TAP_SH(src21_r, src43_r, src65_r, src87_r, filt0,
1441 filt1, filt2, filt3);
1442 dst2 = HEVC_FILT_8TAP_SH(src32_r, src54_r, src76_r, src98_r, filt0,
1443 filt1, filt2, filt3);
1444 dst3 = HEVC_FILT_8TAP_SH(src43_r, src65_r, src87_r, src109_r, filt0,
1445 filt1, filt2, filt3);
1446
1447 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
1448 offset_vec, rnd_vec, dst0, dst1, dst2,
1449 dst3);
1450
1451 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
1452 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
1453 dst += (4 * dst_stride);
1454
1455 src10_r = src54_r;
1456 src32_r = src76_r;
1457 src54_r = src98_r;
1458 src21_r = src65_r;
1459 src43_r = src87_r;
1460 src65_r = src109_r;
1461 src6 = src10;
1462 }
1463 }
1464
hevc_vt_uniwgt_8t_12w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1465 static void hevc_vt_uniwgt_8t_12w_msa(uint8_t *src,
1466 int32_t src_stride,
1467 uint8_t *dst,
1468 int32_t dst_stride,
1469 const int8_t *filter,
1470 int32_t height,
1471 int32_t weight,
1472 int32_t offset,
1473 int32_t rnd_val)
1474 {
1475 int32_t loop_cnt;
1476 v16u8 out0, out1, out2;
1477 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
1478 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r;
1479 v16i8 src21_r, src43_r, src65_r, src87_r, src109_r;
1480 v16i8 src10_l, src32_l, src54_l, src76_l, src98_l;
1481 v16i8 src21_l, src43_l, src65_l, src87_l, src109_l;
1482 v16i8 src2110, src4332, src6554, src8776, src10998;
1483 v8i16 filt0, filt1, filt2, filt3;
1484 v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
1485 v8i16 weight_vec_h, offset_vec, denom_vec, filter_vec;
1486 v4i32 weight_vec, rnd_vec;
1487
1488 src -= (3 * src_stride);
1489
1490 weight = weight & 0x0000FFFF;
1491 weight_vec = __msa_fill_w(weight);
1492 rnd_vec = __msa_fill_w(rnd_val);
1493
1494 weight *= 128;
1495 rnd_val -= 6;
1496
1497 weight_vec_h = __msa_fill_h(weight);
1498 offset_vec = __msa_fill_h(offset);
1499 denom_vec = __msa_fill_h(rnd_val);
1500
1501 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
1502 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
1503
1504 filter_vec = LD_SH(filter);
1505 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1506
1507 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
1508 src += (7 * src_stride);
1509 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
1510
1511 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
1512 src10_r, src32_r, src54_r, src21_r);
1513 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
1514 ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
1515 src10_l, src32_l, src54_l, src21_l);
1516 ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
1517 ILVR_D3_SB(src21_l, src10_l, src43_l, src32_l, src65_l, src54_l,
1518 src2110, src4332, src6554);
1519
1520 for (loop_cnt = 4; loop_cnt--;) {
1521 LD_SB4(src, src_stride, src7, src8, src9, src10);
1522 src += (4 * src_stride);
1523 XORI_B4_128_SB(src7, src8, src9, src10);
1524
1525 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
1526 src76_r, src87_r, src98_r, src109_r);
1527 ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
1528 src76_l, src87_l, src98_l, src109_l);
1529 ILVR_D2_SB(src87_l, src76_l, src109_l, src98_l, src8776, src10998);
1530
1531 dst0 = HEVC_FILT_8TAP_SH(src10_r, src32_r, src54_r, src76_r, filt0,
1532 filt1, filt2, filt3);
1533 dst1 = HEVC_FILT_8TAP_SH(src21_r, src43_r, src65_r, src87_r, filt0,
1534 filt1, filt2, filt3);
1535 dst2 = HEVC_FILT_8TAP_SH(src32_r, src54_r, src76_r, src98_r, filt0,
1536 filt1, filt2, filt3);
1537 dst3 = HEVC_FILT_8TAP_SH(src43_r, src65_r, src87_r, src109_r, filt0,
1538 filt1, filt2, filt3);
1539 dst4 = HEVC_FILT_8TAP_SH(src2110, src4332, src6554, src8776, filt0,
1540 filt1, filt2, filt3);
1541 dst5 = HEVC_FILT_8TAP_SH(src4332, src6554, src8776, src10998, filt0,
1542 filt1, filt2, filt3);
1543
1544 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
1545 offset_vec, rnd_vec, dst0, dst1, dst2,
1546 dst3);
1547 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec,
1548 rnd_vec, dst4, dst5);
1549
1550 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
1551 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
1552 ST_W4(out2, 0, 1, 2, 3, dst + 8, dst_stride);
1553 dst += (4 * dst_stride);
1554
1555 src10_r = src54_r;
1556 src32_r = src76_r;
1557 src54_r = src98_r;
1558 src21_r = src65_r;
1559 src43_r = src87_r;
1560 src65_r = src109_r;
1561 src2110 = src6554;
1562 src4332 = src8776;
1563 src6554 = src10998;
1564 src6 = src10;
1565 }
1566 }
1567
hevc_vt_uniwgt_8t_16multx4mult_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val,int32_t weightmul16)1568 static void hevc_vt_uniwgt_8t_16multx4mult_msa(uint8_t *src,
1569 int32_t src_stride,
1570 uint8_t *dst,
1571 int32_t dst_stride,
1572 const int8_t *filter,
1573 int32_t height,
1574 int32_t weight,
1575 int32_t offset,
1576 int32_t rnd_val,
1577 int32_t weightmul16)
1578 {
1579 uint8_t *src_tmp;
1580 uint8_t *dst_tmp;
1581 int32_t loop_cnt, cnt;
1582 v16u8 out0, out1, out2, out3;
1583 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
1584 v16i8 src10_r, src32_r, src54_r, src76_r;
1585 v16i8 src21_r, src43_r, src65_r, src87_r;
1586 v16i8 src10_l, src32_l, src54_l, src76_l;
1587 v16i8 src21_l, src43_l, src65_l, src87_l;
1588 v16i8 src98_r, src109_r, src98_l, src109_l;
1589 v8i16 filt0, filt1, filt2, filt3;
1590 v8i16 filter_vec;
1591 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
1592 v8i16 weight_vec_h, offset_vec, denom_vec;
1593 v4i32 weight_vec, rnd_vec;
1594
1595 src -= (3 * src_stride);
1596
1597 weight_vec = __msa_fill_w(weight);
1598 rnd_vec = __msa_fill_w(rnd_val);
1599
1600 weight *= 128;
1601 rnd_val -= 6;
1602
1603 weight_vec_h = __msa_fill_h(weight);
1604 offset_vec = __msa_fill_h(offset);
1605 denom_vec = __msa_fill_h(rnd_val);
1606
1607 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
1608 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
1609
1610 filter_vec = LD_SH(filter);
1611 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1612
1613 for (cnt = weightmul16; cnt--;) {
1614 src_tmp = src;
1615 dst_tmp = dst;
1616
1617 LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
1618 src_tmp += (7 * src_stride);
1619 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
1620
1621 for (loop_cnt = (height >> 2); loop_cnt--;) {
1622 LD_SB4(src_tmp, src_stride, src7, src8, src9, src10);
1623 src_tmp += (4 * src_stride);
1624 XORI_B4_128_SB(src7, src8, src9, src10);
1625
1626 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
1627 src10_r, src32_r, src54_r, src21_r);
1628 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
1629 ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1,
1630 src10_l, src32_l, src54_l, src21_l);
1631 ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
1632 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
1633 src76_r, src87_r, src98_r, src109_r);
1634 ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9,
1635 src76_l, src87_l, src98_l, src109_l);
1636
1637 dst0 = HEVC_FILT_8TAP_SH(src10_r, src32_r, src54_r, src76_r, filt0,
1638 filt1, filt2, filt3);
1639 dst1 = HEVC_FILT_8TAP_SH(src10_l, src32_l, src54_l, src76_l, filt0,
1640 filt1, filt2, filt3);
1641 dst2 = HEVC_FILT_8TAP_SH(src21_r, src43_r, src65_r, src87_r, filt0,
1642 filt1, filt2, filt3);
1643 dst3 = HEVC_FILT_8TAP_SH(src21_l, src43_l, src65_l, src87_l, filt0,
1644 filt1, filt2, filt3);
1645 dst4 = HEVC_FILT_8TAP_SH(src32_r, src54_r, src76_r, src98_r, filt0,
1646 filt1, filt2, filt3);
1647 dst5 = HEVC_FILT_8TAP_SH(src32_l, src54_l, src76_l, src98_l, filt0,
1648 filt1, filt2, filt3);
1649 dst6 = HEVC_FILT_8TAP_SH(src43_r, src65_r, src87_r, src109_r, filt0,
1650 filt1, filt2, filt3);
1651 dst7 = HEVC_FILT_8TAP_SH(src43_l, src65_l, src87_l, src109_l, filt0,
1652 filt1, filt2, filt3);
1653
1654 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
1655 offset_vec, rnd_vec, dst0, dst1,
1656 dst2, dst3);
1657 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
1658 offset_vec, rnd_vec, dst4, dst5,
1659 dst6, dst7);
1660 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
1661 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
1662 ST_UB4(out0, out1, out2, out3, dst_tmp, dst_stride);
1663 dst_tmp += (4 * dst_stride);
1664
1665 src0 = src4;
1666 src1 = src5;
1667 src2 = src6;
1668 src3 = src7;
1669 src4 = src8;
1670 src5 = src9;
1671 src6 = src10;
1672 }
1673
1674 src += 16;
1675 dst += 16;
1676 }
1677 }
1678
hevc_vt_uniwgt_8t_16w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1679 static void hevc_vt_uniwgt_8t_16w_msa(uint8_t *src,
1680 int32_t src_stride,
1681 uint8_t *dst,
1682 int32_t dst_stride,
1683 const int8_t *filter,
1684 int32_t height,
1685 int32_t weight,
1686 int32_t offset,
1687 int32_t rnd_val)
1688 {
1689 hevc_vt_uniwgt_8t_16multx4mult_msa(src, src_stride, dst, dst_stride,
1690 filter, height, weight,
1691 offset, rnd_val, 1);
1692 }
1693
hevc_vt_uniwgt_8t_24w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1694 static void hevc_vt_uniwgt_8t_24w_msa(uint8_t *src,
1695 int32_t src_stride,
1696 uint8_t *dst,
1697 int32_t dst_stride,
1698 const int8_t *filter,
1699 int32_t height,
1700 int32_t weight,
1701 int32_t offset,
1702 int32_t rnd_val)
1703 {
1704 hevc_vt_uniwgt_8t_16multx4mult_msa(src, src_stride, dst, dst_stride,
1705 filter, 32, weight,
1706 offset, rnd_val, 1);
1707
1708 hevc_vt_uniwgt_8t_8w_msa(src + 16, src_stride, dst + 16, dst_stride,
1709 filter, 32, weight, offset, rnd_val);
1710 }
1711
hevc_vt_uniwgt_8t_32w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1712 static void hevc_vt_uniwgt_8t_32w_msa(uint8_t *src,
1713 int32_t src_stride,
1714 uint8_t *dst,
1715 int32_t dst_stride,
1716 const int8_t *filter,
1717 int32_t height,
1718 int32_t weight,
1719 int32_t offset,
1720 int32_t rnd_val)
1721 {
1722 hevc_vt_uniwgt_8t_16multx4mult_msa(src, src_stride, dst, dst_stride,
1723 filter, height, weight,
1724 offset, rnd_val, 2);
1725 }
1726
hevc_vt_uniwgt_8t_48w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1727 static void hevc_vt_uniwgt_8t_48w_msa(uint8_t *src,
1728 int32_t src_stride,
1729 uint8_t *dst,
1730 int32_t dst_stride,
1731 const int8_t *filter,
1732 int32_t height,
1733 int32_t weight,
1734 int32_t offset,
1735 int32_t rnd_val)
1736 {
1737 hevc_vt_uniwgt_8t_16multx4mult_msa(src, src_stride, dst, dst_stride,
1738 filter, 64, weight,
1739 offset, rnd_val, 3);
1740 }
1741
hevc_vt_uniwgt_8t_64w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1742 static void hevc_vt_uniwgt_8t_64w_msa(uint8_t *src,
1743 int32_t src_stride,
1744 uint8_t *dst,
1745 int32_t dst_stride,
1746 const int8_t *filter,
1747 int32_t height,
1748 int32_t weight,
1749 int32_t offset,
1750 int32_t rnd_val)
1751 {
1752 hevc_vt_uniwgt_8t_16multx4mult_msa(src, src_stride, dst, dst_stride,
1753 filter, height, weight,
1754 offset, rnd_val, 4);
1755 }
1756
hevc_hv_uniwgt_8t_4w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)1757 static void hevc_hv_uniwgt_8t_4w_msa(uint8_t *src,
1758 int32_t src_stride,
1759 uint8_t *dst,
1760 int32_t dst_stride,
1761 const int8_t *filter_x,
1762 const int8_t *filter_y,
1763 int32_t height,
1764 int32_t weight,
1765 int32_t offset,
1766 int32_t rnd_val)
1767 {
1768 uint32_t loop_cnt;
1769 v16u8 out;
1770 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
1771 v8i16 filt0, filt1, filt2, filt3;
1772 v8i16 filt_h0, filt_h1, filt_h2, filt_h3;
1773 v16i8 mask1, mask2, mask3;
1774 v8i16 filter_vec;
1775 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
1776 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
1777 v8i16 dst30, dst41, dst52, dst63, dst66, dst97, dst108;
1778 v8i16 dst10_r, dst32_r, dst54_r, dst76_r, dst98_r;
1779 v8i16 dst21_r, dst43_r, dst65_r, dst87_r, dst109_r;
1780 v4i32 dst0_r, dst1_r, dst2_r, dst3_r;
1781 v4i32 weight_vec, offset_vec, rnd_vec, const_128, denom_vec;
1782 v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
1783
1784 src -= ((3 * src_stride) + 3);
1785 filter_vec = LD_SH(filter_x);
1786 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1787
1788 filter_vec = LD_SH(filter_y);
1789 UNPCK_R_SB_SH(filter_vec, filter_vec);
1790
1791 SPLATI_W4_SH(filter_vec, filt_h0, filt_h1, filt_h2, filt_h3);
1792
1793 mask1 = mask0 + 2;
1794 mask2 = mask0 + 4;
1795 mask3 = mask0 + 6;
1796
1797 weight_vec = __msa_fill_w(weight);
1798 offset_vec = __msa_fill_w(offset);
1799 rnd_vec = __msa_fill_w(rnd_val);
1800 denom_vec = rnd_vec - 6;
1801
1802 const_128 = __msa_ldi_w(128);
1803 const_128 *= weight_vec;
1804 offset_vec += __msa_srar_w(const_128, denom_vec);
1805
1806 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
1807 src += (7 * src_stride);
1808 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
1809
1810 /* row 0 row 1 row 2 row 3 */
1811 VSHF_B4_SB(src0, src3, mask0, mask1, mask2, mask3, vec0, vec1, vec2, vec3);
1812 VSHF_B4_SB(src1, src4, mask0, mask1, mask2, mask3, vec4, vec5, vec6, vec7);
1813 VSHF_B4_SB(src2, src5, mask0, mask1, mask2, mask3,
1814 vec8, vec9, vec10, vec11);
1815 VSHF_B4_SB(src3, src6, mask0, mask1, mask2, mask3,
1816 vec12, vec13, vec14, vec15);
1817 dst30 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
1818 filt3);
1819 dst41 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
1820 filt3);
1821 dst52 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
1822 filt3);
1823 dst63 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1, filt2,
1824 filt3);
1825
1826 ILVRL_H2_SH(dst41, dst30, dst10_r, dst43_r);
1827 ILVRL_H2_SH(dst52, dst41, dst21_r, dst54_r);
1828 ILVRL_H2_SH(dst63, dst52, dst32_r, dst65_r);
1829
1830 dst66 = (v8i16) __msa_splati_d((v2i64) dst63, 1);
1831
1832 for (loop_cnt = height >> 2; loop_cnt--;) {
1833 LD_SB4(src, src_stride, src7, src8, src9, src10);
1834 src += (4 * src_stride);
1835 XORI_B4_128_SB(src7, src8, src9, src10);
1836
1837 VSHF_B4_SB(src7, src9, mask0, mask1, mask2, mask3,
1838 vec0, vec1, vec2, vec3);
1839 VSHF_B4_SB(src8, src10, mask0, mask1, mask2, mask3,
1840 vec4, vec5, vec6, vec7);
1841 dst97 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
1842 filt3);
1843 dst108 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
1844 filt3);
1845
1846 dst76_r = __msa_ilvr_h(dst97, dst66);
1847 ILVRL_H2_SH(dst108, dst97, dst87_r, dst109_r);
1848 dst66 = (v8i16) __msa_splati_d((v2i64) dst97, 1);
1849 dst98_r = __msa_ilvr_h(dst66, dst108);
1850
1851 dst0_r = HEVC_FILT_8TAP(dst10_r, dst32_r, dst54_r, dst76_r, filt_h0,
1852 filt_h1, filt_h2, filt_h3);
1853 dst1_r = HEVC_FILT_8TAP(dst21_r, dst43_r, dst65_r, dst87_r, filt_h0,
1854 filt_h1, filt_h2, filt_h3);
1855 dst2_r = HEVC_FILT_8TAP(dst32_r, dst54_r, dst76_r, dst98_r, filt_h0,
1856 filt_h1, filt_h2, filt_h3);
1857 dst3_r = HEVC_FILT_8TAP(dst43_r, dst65_r, dst87_r, dst109_r, filt_h0,
1858 filt_h1, filt_h2, filt_h3);
1859
1860 SRA_4V(dst0_r, dst1_r, dst2_r, dst3_r, 6);
1861 MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
1862 MUL2(dst2_r, weight_vec, dst3_r, weight_vec, dst2_r, dst3_r);
1863 SRAR_W4_SW(dst0_r, dst1_r, dst2_r, dst3_r, rnd_vec);
1864 ADD2(dst0_r, offset_vec, dst1_r, offset_vec, dst0_r, dst1_r);
1865 ADD2(dst2_r, offset_vec, dst3_r, offset_vec, dst2_r, dst3_r);
1866 CLIP_SW4_0_255(dst0_r, dst1_r, dst2_r, dst3_r);
1867 PCKEV_H2_SW(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r);
1868 out = (v16u8) __msa_pckev_b((v16i8) dst1_r, (v16i8) dst0_r);
1869 ST_W4(out, 0, 1, 2, 3, dst, dst_stride);
1870 dst += (4 * dst_stride);
1871
1872 dst10_r = dst54_r;
1873 dst32_r = dst76_r;
1874 dst54_r = dst98_r;
1875 dst21_r = dst65_r;
1876 dst43_r = dst87_r;
1877 dst65_r = dst109_r;
1878 dst66 = (v8i16) __msa_splati_d((v2i64) dst108, 1);
1879 }
1880 }
1881
hevc_hv_uniwgt_8t_8multx2mult_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val,int32_t width)1882 static void hevc_hv_uniwgt_8t_8multx2mult_msa(uint8_t *src,
1883 int32_t src_stride,
1884 uint8_t *dst,
1885 int32_t dst_stride,
1886 const int8_t *filter_x,
1887 const int8_t *filter_y,
1888 int32_t height,
1889 int32_t weight,
1890 int32_t offset,
1891 int32_t rnd_val,
1892 int32_t width)
1893 {
1894 uint32_t loop_cnt, cnt;
1895 uint8_t *src_tmp;
1896 uint8_t *dst_tmp;
1897 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
1898 v8i16 filt0, filt1, filt2, filt3;
1899 v4i32 filt_h0, filt_h1, filt_h2, filt_h3;
1900 v16i8 mask1, mask2, mask3;
1901 v8i16 filter_vec;
1902 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
1903 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
1904 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
1905 v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
1906 v8i16 dst10_r, dst32_r, dst54_r, dst76_r;
1907 v8i16 dst10_l, dst32_l, dst54_l, dst76_l;
1908 v8i16 dst21_r, dst43_r, dst65_r, dst87_r;
1909 v8i16 dst21_l, dst43_l, dst65_l, dst87_l;
1910 v4i32 weight_vec, offset_vec, rnd_vec, const_128, denom_vec;
1911 v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
1912
1913 src -= ((3 * src_stride) + 3);
1914
1915 weight_vec = __msa_fill_w(weight);
1916 offset_vec = __msa_fill_w(offset);
1917 rnd_vec = __msa_fill_w(rnd_val);
1918 denom_vec = rnd_vec - 6;
1919
1920 const_128 = __msa_ldi_w(128);
1921 const_128 *= weight_vec;
1922 offset_vec += __msa_srar_w(const_128, denom_vec);
1923
1924 filter_vec = LD_SH(filter_x);
1925 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
1926
1927 filter_vec = LD_SH(filter_y);
1928 UNPCK_R_SB_SH(filter_vec, filter_vec);
1929 SPLATI_W4_SW(filter_vec, filt_h0, filt_h1, filt_h2, filt_h3);
1930
1931 mask1 = mask0 + 2;
1932 mask2 = mask0 + 4;
1933 mask3 = mask0 + 6;
1934
1935 for (cnt = width >> 3; cnt--;) {
1936 src_tmp = src;
1937 dst_tmp = dst;
1938
1939 LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
1940 src_tmp += (7 * src_stride);
1941 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
1942
1943 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3,
1944 vec0, vec1, vec2, vec3);
1945 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3,
1946 vec4, vec5, vec6, vec7);
1947 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3,
1948 vec8, vec9, vec10, vec11);
1949 VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3,
1950 vec12, vec13, vec14, vec15);
1951 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
1952 filt3);
1953 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
1954 filt3);
1955 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
1956 filt3);
1957 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
1958 filt2, filt3);
1959
1960 VSHF_B4_SB(src4, src4, mask0, mask1, mask2, mask3,
1961 vec0, vec1, vec2, vec3);
1962 VSHF_B4_SB(src5, src5, mask0, mask1, mask2, mask3,
1963 vec4, vec5, vec6, vec7);
1964 VSHF_B4_SB(src6, src6, mask0, mask1, mask2, mask3,
1965 vec8, vec9, vec10, vec11);
1966 dst4 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
1967 filt3);
1968 dst5 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
1969 filt3);
1970 dst6 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
1971 filt3);
1972
1973 ILVR_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst2, dst1,
1974 dst10_r, dst32_r, dst54_r, dst21_r);
1975 ILVR_H2_SH(dst4, dst3, dst6, dst5, dst43_r, dst65_r);
1976 ILVL_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst2, dst1,
1977 dst10_l, dst32_l, dst54_l, dst21_l);
1978 ILVL_H2_SH(dst4, dst3, dst6, dst5, dst43_l, dst65_l);
1979
1980 for (loop_cnt = height >> 1; loop_cnt--;) {
1981 LD_SB2(src_tmp, src_stride, src7, src8);
1982 src_tmp += 2 * src_stride;
1983 XORI_B2_128_SB(src7, src8);
1984
1985 VSHF_B4_SB(src7, src7, mask0, mask1, mask2, mask3,
1986 vec0, vec1, vec2, vec3);
1987 dst7 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1,
1988 filt2, filt3);
1989
1990 ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l);
1991 dst0_r = HEVC_FILT_8TAP(dst10_r, dst32_r, dst54_r, dst76_r,
1992 filt_h0, filt_h1, filt_h2, filt_h3);
1993 dst0_l = HEVC_FILT_8TAP(dst10_l, dst32_l, dst54_l, dst76_l,
1994 filt_h0, filt_h1, filt_h2, filt_h3);
1995 dst0_r >>= 6;
1996 dst0_l >>= 6;
1997
1998 /* row 8 */
1999 VSHF_B4_SB(src8, src8, mask0, mask1, mask2, mask3,
2000 vec0, vec1, vec2, vec3);
2001 dst8 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1,
2002 filt2, filt3);
2003
2004 ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_l);
2005 dst1_r = HEVC_FILT_8TAP(dst21_r, dst43_r, dst65_r, dst87_r,
2006 filt_h0, filt_h1, filt_h2, filt_h3);
2007 dst1_l = HEVC_FILT_8TAP(dst21_l, dst43_l, dst65_l, dst87_l,
2008 filt_h0, filt_h1, filt_h2, filt_h3);
2009 dst1_r >>= 6;
2010 dst1_l >>= 6;
2011
2012 MUL2(dst0_r, weight_vec, dst0_l, weight_vec, dst0_r, dst0_l);
2013 MUL2(dst1_r, weight_vec, dst1_l, weight_vec, dst1_r, dst1_l);
2014 SRAR_W4_SW(dst0_r, dst1_r, dst0_l, dst1_l, rnd_vec);
2015 ADD2(dst0_r, offset_vec, dst0_l, offset_vec, dst0_r, dst0_l);
2016 ADD2(dst1_r, offset_vec, dst1_l, offset_vec, dst1_r, dst1_l);
2017 CLIP_SW4_0_255(dst0_r, dst1_r, dst0_l, dst1_l);
2018
2019 PCKEV_H2_SW(dst0_l, dst0_r, dst1_l, dst1_r, dst0_r, dst1_r);
2020 dst0_r = (v4i32) __msa_pckev_b((v16i8) dst1_r, (v16i8) dst0_r);
2021 ST_D2(dst0_r, 0, 1, dst_tmp, dst_stride);
2022 dst_tmp += (2 * dst_stride);
2023
2024 dst10_r = dst32_r;
2025 dst32_r = dst54_r;
2026 dst54_r = dst76_r;
2027 dst10_l = dst32_l;
2028 dst32_l = dst54_l;
2029 dst54_l = dst76_l;
2030 dst21_r = dst43_r;
2031 dst43_r = dst65_r;
2032 dst65_r = dst87_r;
2033 dst21_l = dst43_l;
2034 dst43_l = dst65_l;
2035 dst65_l = dst87_l;
2036 dst6 = dst8;
2037 }
2038
2039 src += 8;
2040 dst += 8;
2041 }
2042 }
2043
hevc_hv_uniwgt_8t_8w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2044 static void hevc_hv_uniwgt_8t_8w_msa(uint8_t *src,
2045 int32_t src_stride,
2046 uint8_t *dst,
2047 int32_t dst_stride,
2048 const int8_t *filter_x,
2049 const int8_t *filter_y,
2050 int32_t height,
2051 int32_t weight,
2052 int32_t offset,
2053 int32_t rnd_val)
2054 {
2055 hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
2056 filter_x, filter_y, height, weight,
2057 offset, rnd_val, 8);
2058 }
2059
hevc_hv_uniwgt_8t_12w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2060 static void hevc_hv_uniwgt_8t_12w_msa(uint8_t *src,
2061 int32_t src_stride,
2062 uint8_t *dst,
2063 int32_t dst_stride,
2064 const int8_t *filter_x,
2065 const int8_t *filter_y,
2066 int32_t height,
2067 int32_t weight,
2068 int32_t offset,
2069 int32_t rnd_val)
2070 {
2071 uint32_t loop_cnt;
2072 uint8_t *src_tmp, *dst_tmp;
2073 v16u8 out;
2074 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
2075 v16i8 mask0, mask1, mask2, mask3, mask4, mask5, mask6, mask7;
2076 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
2077 v16i8 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
2078 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
2079 v8i16 dst30, dst41, dst52, dst63, dst66, dst97, dst108;
2080 v8i16 filt0, filt1, filt2, filt3, filt_h0, filt_h1, filt_h2, filt_h3;
2081 v8i16 dst10_r, dst32_r, dst54_r, dst76_r, dst10_l, dst32_l, dst54_l;
2082 v8i16 dst98_r, dst21_r, dst43_r, dst65_r, dst87_r, dst109_r;
2083 v8i16 dst76_l, filter_vec;
2084 v4i32 dst0_r, dst0_l, dst1_r, dst2_r, dst3_r;
2085 v4i32 weight_vec, offset_vec, rnd_vec, const_128, denom_vec;
2086
2087 src -= ((3 * src_stride) + 3);
2088
2089 filter_vec = LD_SH(filter_x);
2090 SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
2091
2092 filter_vec = LD_SH(filter_y);
2093 UNPCK_R_SB_SH(filter_vec, filter_vec);
2094
2095 SPLATI_W4_SH(filter_vec, filt_h0, filt_h1, filt_h2, filt_h3);
2096
2097 weight_vec = __msa_fill_w(weight);
2098 offset_vec = __msa_fill_w(offset);
2099 rnd_vec = __msa_fill_w(rnd_val);
2100 denom_vec = rnd_vec - 6;
2101
2102 const_128 = __msa_ldi_w(128);
2103 const_128 *= weight_vec;
2104 offset_vec += __msa_srar_w(const_128, denom_vec);
2105
2106 mask0 = LD_SB(ff_hevc_mask_arr);
2107 mask1 = mask0 + 2;
2108 mask2 = mask0 + 4;
2109 mask3 = mask0 + 6;
2110
2111 src_tmp = src;
2112 dst_tmp = dst;
2113
2114 LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
2115 src_tmp += (7 * src_stride);
2116 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
2117
2118 /* row 0 row 1 row 2 row 3 */
2119 VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, vec0, vec1, vec2, vec3);
2120 VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, vec4, vec5, vec6, vec7);
2121 VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, vec8, vec9, vec10,
2122 vec11);
2123 VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3, vec12, vec13, vec14,
2124 vec15);
2125 dst0 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
2126 filt3);
2127 dst1 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
2128 filt3);
2129 dst2 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
2130 filt3);
2131 dst3 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1,
2132 filt2, filt3);
2133 VSHF_B4_SB(src4, src4, mask0, mask1, mask2, mask3, vec0, vec1, vec2, vec3);
2134 VSHF_B4_SB(src5, src5, mask0, mask1, mask2, mask3, vec4, vec5, vec6, vec7);
2135 VSHF_B4_SB(src6, src6, mask0, mask1, mask2, mask3, vec8, vec9, vec10,
2136 vec11);
2137 dst4 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
2138 filt3);
2139 dst5 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
2140 filt3);
2141 dst6 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
2142 filt3);
2143
2144 for (loop_cnt = 16; loop_cnt--;) {
2145 src7 = LD_SB(src_tmp);
2146 src7 = (v16i8) __msa_xori_b((v16u8) src7, 128);
2147 src_tmp += src_stride;
2148
2149 VSHF_B4_SB(src7, src7, mask0, mask1, mask2, mask3, vec0, vec1, vec2,
2150 vec3);
2151 dst7 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
2152 filt3);
2153 ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
2154 ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
2155 ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
2156 ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l);
2157
2158 dst0_r = HEVC_FILT_8TAP(dst10_r, dst32_r, dst54_r, dst76_r,
2159 filt_h0, filt_h1, filt_h2, filt_h3);
2160 dst0_l = HEVC_FILT_8TAP(dst10_l, dst32_l, dst54_l, dst76_l,
2161 filt_h0, filt_h1, filt_h2, filt_h3);
2162 dst0_r >>= 6;
2163 dst0_l >>= 6;
2164
2165 MUL2(dst0_r, weight_vec, dst0_l, weight_vec, dst0_r, dst0_l);
2166 SRAR_W2_SW(dst0_r, dst0_l, rnd_vec);
2167 ADD2(dst0_r, offset_vec, dst0_l, offset_vec, dst0_r, dst0_l);
2168 CLIP_SW2_0_255(dst0_r, dst0_l);
2169 dst0_r = (v4i32) __msa_pckev_h((v8i16) dst0_l, (v8i16) dst0_r);
2170 out = (v16u8) __msa_pckev_b((v16i8) dst0_r, (v16i8) dst0_r);
2171 ST_D1(out, 0, dst_tmp);
2172 dst_tmp += dst_stride;
2173
2174 dst0 = dst1;
2175 dst1 = dst2;
2176 dst2 = dst3;
2177 dst3 = dst4;
2178 dst4 = dst5;
2179 dst5 = dst6;
2180 dst6 = dst7;
2181 }
2182
2183 src += 8;
2184 dst += 8;
2185
2186 mask4 = LD_SB(ff_hevc_mask_arr + 16);
2187 mask5 = mask4 + 2;
2188 mask6 = mask4 + 4;
2189 mask7 = mask4 + 6;
2190
2191 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
2192 src += (7 * src_stride);
2193 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
2194
2195 VSHF_B4_SB(src0, src3, mask4, mask5, mask6, mask7, vec0, vec1, vec2, vec3);
2196 VSHF_B4_SB(src1, src4, mask4, mask5, mask6, mask7, vec4, vec5, vec6, vec7);
2197 VSHF_B4_SB(src2, src5, mask4, mask5, mask6, mask7, vec8, vec9, vec10,
2198 vec11);
2199 VSHF_B4_SB(src3, src6, mask4, mask5, mask6, mask7, vec12, vec13, vec14,
2200 vec15);
2201 dst30 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
2202 filt3);
2203 dst41 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
2204 filt3);
2205 dst52 = HEVC_FILT_8TAP_SH(vec8, vec9, vec10, vec11, filt0, filt1, filt2,
2206 filt3);
2207 dst63 = HEVC_FILT_8TAP_SH(vec12, vec13, vec14, vec15, filt0, filt1, filt2,
2208 filt3);
2209 ILVRL_H2_SH(dst41, dst30, dst10_r, dst43_r);
2210 ILVRL_H2_SH(dst52, dst41, dst21_r, dst54_r);
2211 ILVRL_H2_SH(dst63, dst52, dst32_r, dst65_r);
2212
2213 dst66 = (v8i16) __msa_splati_d((v2i64) dst63, 1);
2214
2215 for (loop_cnt = 4; loop_cnt--;) {
2216 LD_SB4(src, src_stride, src7, src8, src9, src10);
2217 src += (4 * src_stride);
2218 XORI_B4_128_SB(src7, src8, src9, src10);
2219
2220 VSHF_B4_SB(src7, src9, mask4, mask5, mask6, mask7, vec0, vec1, vec2,
2221 vec3);
2222 VSHF_B4_SB(src8, src10, mask4, mask5, mask6, mask7, vec4, vec5, vec6,
2223 vec7);
2224 dst97 = HEVC_FILT_8TAP_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2,
2225 filt3);
2226 dst108 = HEVC_FILT_8TAP_SH(vec4, vec5, vec6, vec7, filt0, filt1, filt2,
2227 filt3);
2228
2229 dst76_r = __msa_ilvr_h(dst97, dst66);
2230 ILVRL_H2_SH(dst108, dst97, dst87_r, dst109_r);
2231 dst66 = (v8i16) __msa_splati_d((v2i64) dst97, 1);
2232 dst98_r = __msa_ilvr_h(dst66, dst108);
2233
2234 dst0_r = HEVC_FILT_8TAP(dst10_r, dst32_r, dst54_r, dst76_r, filt_h0,
2235 filt_h1, filt_h2, filt_h3);
2236 dst1_r = HEVC_FILT_8TAP(dst21_r, dst43_r, dst65_r, dst87_r, filt_h0,
2237 filt_h1, filt_h2, filt_h3);
2238 dst2_r = HEVC_FILT_8TAP(dst32_r, dst54_r, dst76_r, dst98_r, filt_h0,
2239 filt_h1, filt_h2, filt_h3);
2240 dst3_r = HEVC_FILT_8TAP(dst43_r, dst65_r, dst87_r, dst109_r, filt_h0,
2241 filt_h1, filt_h2, filt_h3);
2242
2243 SRA_4V(dst0_r, dst1_r, dst2_r, dst3_r, 6);
2244 MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
2245 MUL2(dst2_r, weight_vec, dst3_r, weight_vec, dst2_r, dst3_r);
2246 SRAR_W4_SW(dst0_r, dst1_r, dst2_r, dst3_r, rnd_vec);
2247 ADD2(dst0_r, offset_vec, dst1_r, offset_vec, dst0_r, dst1_r);
2248 ADD2(dst2_r, offset_vec, dst3_r, offset_vec, dst2_r, dst3_r);
2249 CLIP_SW4_0_255(dst0_r, dst1_r, dst2_r, dst3_r);
2250 PCKEV_H2_SW(dst1_r, dst0_r, dst3_r, dst2_r, dst0_r, dst1_r);
2251 out = (v16u8) __msa_pckev_b((v16i8) dst1_r, (v16i8) dst0_r);
2252 ST_W4(out, 0, 1, 2, 3, dst, dst_stride);
2253 dst += (4 * dst_stride);
2254
2255 dst10_r = dst54_r;
2256 dst32_r = dst76_r;
2257 dst54_r = dst98_r;
2258 dst21_r = dst65_r;
2259 dst43_r = dst87_r;
2260 dst65_r = dst109_r;
2261 dst66 = (v8i16) __msa_splati_d((v2i64) dst108, 1);
2262 }
2263 }
2264
hevc_hv_uniwgt_8t_16w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2265 static void hevc_hv_uniwgt_8t_16w_msa(uint8_t *src,
2266 int32_t src_stride,
2267 uint8_t *dst,
2268 int32_t dst_stride,
2269 const int8_t *filter_x,
2270 const int8_t *filter_y,
2271 int32_t height,
2272 int32_t weight,
2273 int32_t offset,
2274 int32_t rnd_val)
2275 {
2276 hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
2277 filter_x, filter_y, height, weight,
2278 offset, rnd_val, 16);
2279 }
2280
hevc_hv_uniwgt_8t_24w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2281 static void hevc_hv_uniwgt_8t_24w_msa(uint8_t *src,
2282 int32_t src_stride,
2283 uint8_t *dst,
2284 int32_t dst_stride,
2285 const int8_t *filter_x,
2286 const int8_t *filter_y,
2287 int32_t height,
2288 int32_t weight,
2289 int32_t offset,
2290 int32_t rnd_val)
2291 {
2292 hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
2293 filter_x, filter_y, height, weight,
2294 offset, rnd_val, 24);
2295 }
2296
hevc_hv_uniwgt_8t_32w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2297 static void hevc_hv_uniwgt_8t_32w_msa(uint8_t *src,
2298 int32_t src_stride,
2299 uint8_t *dst,
2300 int32_t dst_stride,
2301 const int8_t *filter_x,
2302 const int8_t *filter_y,
2303 int32_t height,
2304 int32_t weight,
2305 int32_t offset,
2306 int32_t rnd_val)
2307 {
2308 hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
2309 filter_x, filter_y, height, weight,
2310 offset, rnd_val, 32);
2311 }
2312
hevc_hv_uniwgt_8t_48w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2313 static void hevc_hv_uniwgt_8t_48w_msa(uint8_t *src,
2314 int32_t src_stride,
2315 uint8_t *dst,
2316 int32_t dst_stride,
2317 const int8_t *filter_x,
2318 const int8_t *filter_y,
2319 int32_t height,
2320 int32_t weight,
2321 int32_t offset,
2322 int32_t rnd_val)
2323 {
2324 hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
2325 filter_x, filter_y, height, weight,
2326 offset, rnd_val, 48);
2327 }
2328
hevc_hv_uniwgt_8t_64w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2329 static void hevc_hv_uniwgt_8t_64w_msa(uint8_t *src,
2330 int32_t src_stride,
2331 uint8_t *dst,
2332 int32_t dst_stride,
2333 const int8_t *filter_x,
2334 const int8_t *filter_y,
2335 int32_t height,
2336 int32_t weight,
2337 int32_t offset,
2338 int32_t rnd_val)
2339 {
2340 hevc_hv_uniwgt_8t_8multx2mult_msa(src, src_stride, dst, dst_stride,
2341 filter_x, filter_y, height, weight,
2342 offset, rnd_val, 64);
2343 }
2344
hevc_hz_uniwgt_4t_4x2_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)2345 static void hevc_hz_uniwgt_4t_4x2_msa(uint8_t *src,
2346 int32_t src_stride,
2347 uint8_t *dst,
2348 int32_t dst_stride,
2349 const int8_t *filter,
2350 int32_t weight,
2351 int32_t offset,
2352 int32_t rnd_val)
2353 {
2354 v16u8 out;
2355 v8i16 filt0, filt1;
2356 v16i8 src0, src1, vec0, vec1;
2357 v16i8 mask1;
2358 v8i16 dst0;
2359 v4i32 dst0_r, dst0_l;
2360 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
2361 v4i32 weight_vec, rnd_vec;
2362 v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[16]);
2363
2364 src -= 1;
2365
2366 filter_vec = LD_SH(filter);
2367 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2368
2369 mask1 = mask0 + 2;
2370
2371 weight = weight & 0x0000FFFF;
2372
2373 weight_vec = __msa_fill_w(weight);
2374 rnd_vec = __msa_fill_w(rnd_val);
2375
2376 weight *= 128;
2377 rnd_val -= 6;
2378
2379 weight_vec_h = __msa_fill_h(weight);
2380 offset_vec = __msa_fill_h(offset);
2381 denom_vec = __msa_fill_h(rnd_val);
2382
2383 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2384 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2385
2386 LD_SB2(src, src_stride, src0, src1);
2387 XORI_B2_128_SB(src0, src1);
2388
2389 VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
2390 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2391
2392 ILVRL_H2_SW(dst0, dst0, dst0_r, dst0_l);
2393 DOTP_SH2_SW(dst0_r, dst0_l, weight_vec, weight_vec, dst0_r, dst0_l);
2394 SRAR_W2_SW(dst0_r, dst0_l, rnd_vec);
2395 dst0 = __msa_pckev_h((v8i16) dst0_l, (v8i16) dst0_r);
2396 dst0 = __msa_adds_s_h(dst0, offset_vec);
2397 CLIP_SH_0_255(dst0);
2398 out = (v16u8) __msa_pckev_b((v16i8) dst0, (v16i8) dst0);
2399 ST_W2(out, 0, 1, dst, dst_stride);
2400 dst += (4 * dst_stride);
2401 }
2402
hevc_hz_uniwgt_4t_4x4_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)2403 static void hevc_hz_uniwgt_4t_4x4_msa(uint8_t *src,
2404 int32_t src_stride,
2405 uint8_t *dst,
2406 int32_t dst_stride,
2407 const int8_t *filter,
2408 int32_t weight,
2409 int32_t offset,
2410 int32_t rnd_val)
2411 {
2412 v16u8 out;
2413 v8i16 filt0, filt1;
2414 v16i8 src0, src1, src2, src3;
2415 v16i8 mask1, vec0, vec1, vec2, vec3;
2416 v8i16 dst0, dst1;
2417 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
2418 v4i32 weight_vec, rnd_vec;
2419 v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[16]);
2420
2421 src -= 1;
2422
2423 /* rearranging filter */
2424 filter_vec = LD_SH(filter);
2425 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2426
2427 mask1 = mask0 + 2;
2428
2429 weight = weight & 0x0000FFFF;
2430
2431 weight_vec = __msa_fill_w(weight);
2432 rnd_vec = __msa_fill_w(rnd_val);
2433
2434 weight *= 128;
2435 rnd_val -= 6;
2436
2437 weight_vec_h = __msa_fill_h(weight);
2438 offset_vec = __msa_fill_h(offset);
2439 denom_vec = __msa_fill_h(rnd_val);
2440
2441 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2442 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2443
2444 LD_SB4(src, src_stride, src0, src1, src2, src3);
2445 XORI_B4_128_SB(src0, src1, src2, src3);
2446
2447 VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
2448 VSHF_B2_SB(src2, src3, src2, src3, mask0, mask1, vec2, vec3);
2449 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2450 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2451
2452 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst0, dst1, weight_vec, offset_vec, rnd_vec,
2453 dst0, dst1);
2454
2455 out = (v16u8) __msa_pckev_b((v16i8) dst1, (v16i8) dst0);
2456 ST_W4(out, 0, 1, 2, 3, dst, dst_stride);
2457 dst += (4 * dst_stride);
2458 }
2459
hevc_hz_uniwgt_4t_4x8multiple_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2460 static void hevc_hz_uniwgt_4t_4x8multiple_msa(uint8_t *src,
2461 int32_t src_stride,
2462 uint8_t *dst,
2463 int32_t dst_stride,
2464 const int8_t *filter,
2465 int32_t height,
2466 int32_t weight,
2467 int32_t offset,
2468 int32_t rnd_val)
2469 {
2470 uint32_t loop_cnt;
2471 v16u8 out0, out1;
2472 v8i16 filt0, filt1;
2473 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
2474 v16i8 mask1, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
2475 v8i16 dst0, dst1, dst2, dst3;
2476 v8i16 filter_vec;
2477 v8i16 weight_vec_h, offset_vec, denom_vec;
2478 v4i32 weight_vec, rnd_vec;
2479 v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[16]);
2480
2481 src -= 1;
2482
2483 filter_vec = LD_SH(filter);
2484 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2485
2486 weight = weight & 0x0000FFFF;
2487
2488 weight_vec = __msa_fill_w(weight);
2489 rnd_vec = __msa_fill_w(rnd_val);
2490
2491 weight *= 128;
2492 rnd_val -= 6;
2493
2494 weight_vec_h = __msa_fill_h(weight);
2495 offset_vec = __msa_fill_h(offset);
2496 denom_vec = __msa_fill_h(rnd_val);
2497
2498 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2499 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2500
2501 mask1 = mask0 + 2;
2502
2503 for (loop_cnt = (height >> 3); loop_cnt--;) {
2504 LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
2505 src += (8 * src_stride);
2506
2507 XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
2508
2509 VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
2510 VSHF_B2_SB(src2, src3, src2, src3, mask0, mask1, vec2, vec3);
2511 VSHF_B2_SB(src4, src5, src4, src5, mask0, mask1, vec4, vec5);
2512 VSHF_B2_SB(src6, src7, src6, src7, mask0, mask1, vec6, vec7);
2513 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2514 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2515 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
2516 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
2517
2518 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
2519 weight_vec, offset_vec, rnd_vec,
2520 dst0, dst1, dst2, dst3);
2521
2522 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
2523 ST_W8(out0, out1, 0, 1, 2, 3, 0, 1, 2, 3, dst, dst_stride);
2524 dst += (8 * dst_stride);
2525 }
2526 }
2527
hevc_hz_uniwgt_4t_4w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2528 static void hevc_hz_uniwgt_4t_4w_msa(uint8_t *src,
2529 int32_t src_stride,
2530 uint8_t *dst,
2531 int32_t dst_stride,
2532 const int8_t *filter,
2533 int32_t height,
2534 int32_t weight,
2535 int32_t offset,
2536 int32_t rnd_val)
2537 {
2538 if (2 == height) {
2539 hevc_hz_uniwgt_4t_4x2_msa(src, src_stride, dst, dst_stride,
2540 filter, weight, offset, rnd_val);
2541 } else if (4 == height) {
2542 hevc_hz_uniwgt_4t_4x4_msa(src, src_stride, dst, dst_stride,
2543 filter, weight, offset, rnd_val);
2544 } else if (8 == height || 16 == height) {
2545 hevc_hz_uniwgt_4t_4x8multiple_msa(src, src_stride, dst, dst_stride,
2546 filter, height, weight,
2547 offset, rnd_val);
2548 }
2549 }
2550
hevc_hz_uniwgt_4t_6w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2551 static void hevc_hz_uniwgt_4t_6w_msa(uint8_t *src,
2552 int32_t src_stride,
2553 uint8_t *dst,
2554 int32_t dst_stride,
2555 const int8_t *filter,
2556 int32_t height,
2557 int32_t weight,
2558 int32_t offset,
2559 int32_t rnd_val)
2560 {
2561 v16u8 out0, out1, out2, out3;
2562 v8i16 filt0, filt1;
2563 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
2564 v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
2565 v16i8 mask1;
2566 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
2567 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
2568 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
2569 v4i32 weight_vec, rnd_vec;
2570
2571 src -= 1;
2572
2573 filter_vec = LD_SH(filter);
2574 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2575
2576 weight = weight & 0x0000FFFF;
2577
2578 weight_vec = __msa_fill_w(weight);
2579 rnd_vec = __msa_fill_w(rnd_val);
2580
2581 weight *= 128;
2582 rnd_val -= 6;
2583
2584 weight_vec_h = __msa_fill_h(weight);
2585 offset_vec = __msa_fill_h(offset);
2586 denom_vec = __msa_fill_h(rnd_val);
2587
2588 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2589 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2590
2591 mask1 = mask0 + 2;
2592
2593 LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
2594 XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
2595 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
2596 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
2597 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
2598 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
2599 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2600 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2601 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
2602 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
2603 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
2604 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec2, vec3);
2605 VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec4, vec5);
2606 VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec6, vec7);
2607 dst4 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2608 dst5 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2609 dst6 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
2610 dst7 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
2611
2612 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
2613 weight_vec, offset_vec, rnd_vec,
2614 dst0, dst1, dst2, dst3);
2615 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7,
2616 weight_vec, offset_vec, rnd_vec,
2617 dst4, dst5, dst6, dst7);
2618
2619 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
2620 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
2621 ST_W2(out0, 0, 2, dst, dst_stride);
2622 ST_H2(out0, 2, 6, dst + 4, dst_stride);
2623 ST_W2(out1, 0, 2, dst + 2 * dst_stride, dst_stride);
2624 ST_H2(out1, 2, 6, dst + 2 * dst_stride + 4, dst_stride);
2625 dst += (4 * dst_stride);
2626 ST_W2(out2, 0, 2, dst, dst_stride);
2627 ST_H2(out2, 2, 6, dst + 4, dst_stride);
2628 ST_W2(out3, 0, 2, dst + 2 * dst_stride, dst_stride);
2629 ST_H2(out3, 2, 6, dst + 2 * dst_stride + 4, dst_stride);
2630 }
2631
hevc_hz_uniwgt_4t_8x2_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)2632 static void hevc_hz_uniwgt_4t_8x2_msa(uint8_t *src,
2633 int32_t src_stride,
2634 uint8_t *dst,
2635 int32_t dst_stride,
2636 const int8_t *filter,
2637 int32_t weight,
2638 int32_t offset,
2639 int32_t rnd_val)
2640 {
2641 v16u8 out;
2642 v8i16 filt0, filt1, dst0, dst1;
2643 v16i8 src0, src1;
2644 v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
2645 v16i8 mask1;
2646 v16i8 vec0, vec1, vec2, vec3;
2647 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
2648 v4i32 weight_vec, rnd_vec;
2649
2650 src -= 1;
2651
2652 filter_vec = LD_SH(filter);
2653 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2654
2655 weight = weight & 0x0000FFFF;
2656
2657 weight_vec = __msa_fill_w(weight);
2658 rnd_vec = __msa_fill_w(rnd_val);
2659
2660 weight *= 128;
2661 rnd_val -= 6;
2662
2663 weight_vec_h = __msa_fill_h(weight);
2664 offset_vec = __msa_fill_h(offset);
2665 denom_vec = __msa_fill_h(rnd_val);
2666
2667 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2668 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2669
2670 mask1 = mask0 + 2;
2671
2672 LD_SB2(src, src_stride, src0, src1);
2673 XORI_B2_128_SB(src0, src1);
2674
2675 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
2676 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
2677 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2678 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2679
2680 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst0, dst1, weight_vec, offset_vec, rnd_vec,
2681 dst0, dst1);
2682
2683 out = (v16u8) __msa_pckev_b((v16i8) dst1, (v16i8) dst0);
2684 ST_D2(out, 0, 1, dst, dst_stride);
2685 }
2686
hevc_hz_uniwgt_4t_8x4_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)2687 static void hevc_hz_uniwgt_4t_8x4_msa(uint8_t *src,
2688 int32_t src_stride,
2689 uint8_t *dst,
2690 int32_t dst_stride,
2691 const int8_t *filter,
2692 int32_t weight,
2693 int32_t offset,
2694 int32_t rnd_val)
2695 {
2696 v16u8 out0, out1;
2697 v16i8 src0, src1, src2, src3;
2698 v16i8 mask0, mask1, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
2699 v8i16 filt0, filt1, dst0, dst1, dst2, dst3;
2700 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
2701 v4i32 weight_vec, rnd_vec;
2702
2703 src -= 1;
2704
2705 filter_vec = LD_SH(filter);
2706 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2707
2708 weight = weight & 0x0000FFFF;
2709 weight_vec = __msa_fill_w(weight);
2710 rnd_vec = __msa_fill_w(rnd_val);
2711
2712 weight *= 128;
2713 rnd_val -= 6;
2714
2715 weight_vec_h = __msa_fill_h(weight);
2716 offset_vec = __msa_fill_h(offset);
2717 denom_vec = __msa_fill_h(rnd_val);
2718
2719 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2720 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2721
2722 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
2723 mask1 = mask0 + 2;
2724
2725 LD_SB4(src, src_stride, src0, src1, src2, src3);
2726 XORI_B4_128_SB(src0, src1, src2, src3);
2727 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
2728 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
2729 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
2730 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
2731 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2732 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2733 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
2734 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
2735
2736 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
2737 weight_vec, offset_vec, rnd_vec,
2738 dst0, dst1, dst2, dst3);
2739
2740 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
2741 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
2742 }
2743
hevc_hz_uniwgt_4t_8x6_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)2744 static void hevc_hz_uniwgt_4t_8x6_msa(uint8_t *src,
2745 int32_t src_stride,
2746 uint8_t *dst,
2747 int32_t dst_stride,
2748 const int8_t *filter,
2749 int32_t weight,
2750 int32_t offset,
2751 int32_t rnd_val)
2752 {
2753 v16u8 out0, out1, out2;
2754 v8i16 filt0, filt1;
2755 v16i8 src0, src1, src2, src3, src4, src5;
2756 v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
2757 v16i8 mask1;
2758 v16i8 vec11;
2759 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9, vec10;
2760 v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
2761 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
2762 v4i32 weight_vec, rnd_vec;
2763
2764 src -= 1;
2765
2766 filter_vec = LD_SH(filter);
2767 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2768
2769 weight = weight & 0x0000FFFF;
2770
2771 weight_vec = __msa_fill_w(weight);
2772 rnd_vec = __msa_fill_w(rnd_val);
2773
2774 weight *= 128;
2775 rnd_val -= 6;
2776
2777 weight_vec_h = __msa_fill_h(weight);
2778 offset_vec = __msa_fill_h(offset);
2779 denom_vec = __msa_fill_h(rnd_val);
2780
2781 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2782 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2783
2784 mask1 = mask0 + 2;
2785
2786 LD_SB6(src, src_stride, src0, src1, src2, src3, src4, src5);
2787 XORI_B6_128_SB(src0, src1, src2, src3, src4, src5);
2788
2789 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
2790 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
2791 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
2792 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
2793 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec8, vec9);
2794 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec10, vec11);
2795 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2796 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2797 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
2798 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
2799 dst4 = HEVC_FILT_4TAP_SH(vec8, vec9, filt0, filt1);
2800 dst5 = HEVC_FILT_4TAP_SH(vec10, vec11, filt0, filt1);
2801
2802 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
2803 weight_vec, offset_vec, rnd_vec,
2804 dst0, dst1, dst2, dst3);
2805
2806 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec, rnd_vec,
2807 dst4, dst5);
2808
2809 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
2810 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
2811 ST_D2(out2, 0, 1, dst + 4 * dst_stride, dst_stride);
2812 }
2813
hevc_hz_uniwgt_4t_8x8multiple_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2814 static void hevc_hz_uniwgt_4t_8x8multiple_msa(uint8_t *src,
2815 int32_t src_stride,
2816 uint8_t *dst,
2817 int32_t dst_stride,
2818 const int8_t *filter,
2819 int32_t height,
2820 int32_t weight,
2821 int32_t offset,
2822 int32_t rnd_val)
2823 {
2824 uint32_t loop_cnt;
2825 v8i16 filt0, filt1;
2826 v16u8 out0, out1, out2, out3;
2827 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
2828 v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
2829 v16i8 mask1;
2830 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
2831 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
2832 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
2833 v4i32 weight_vec, rnd_vec;
2834
2835 src -= 1;
2836
2837 filter_vec = LD_SH(filter);
2838 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2839
2840 weight = weight & 0x0000FFFF;
2841
2842 weight_vec = __msa_fill_w(weight);
2843 rnd_vec = __msa_fill_w(rnd_val);
2844
2845 weight *= 128;
2846 rnd_val -= 6;
2847
2848 weight_vec_h = __msa_fill_h(weight);
2849 offset_vec = __msa_fill_h(offset);
2850 denom_vec = __msa_fill_h(rnd_val);
2851
2852 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2853 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2854
2855 mask1 = mask0 + 2;
2856
2857 for (loop_cnt = (height >> 3); loop_cnt--;) {
2858 LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
2859 src += (8 * src_stride);
2860 XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
2861
2862 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
2863 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
2864 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
2865 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
2866 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2867 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2868 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
2869 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
2870 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
2871 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec2, vec3);
2872 VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec4, vec5);
2873 VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec6, vec7);
2874 dst4 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2875 dst5 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2876 dst6 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
2877 dst7 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
2878
2879 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
2880 weight_vec, offset_vec, rnd_vec,
2881 dst0, dst1, dst2, dst3);
2882
2883 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7,
2884 weight_vec, offset_vec, rnd_vec,
2885 dst4, dst5, dst6, dst7);
2886
2887 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
2888 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
2889 ST_D8(out0, out1, out2, out3, 0, 1, 0, 1, 0, 1, 0, 1, dst, dst_stride);
2890 dst += (8 * dst_stride);
2891 }
2892 }
2893
hevc_hz_uniwgt_4t_8w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2894 static void hevc_hz_uniwgt_4t_8w_msa(uint8_t *src,
2895 int32_t src_stride,
2896 uint8_t *dst,
2897 int32_t dst_stride,
2898 const int8_t *filter,
2899 int32_t height,
2900 int32_t weight,
2901 int32_t offset,
2902 int32_t rnd_val)
2903 {
2904 if (2 == height) {
2905 hevc_hz_uniwgt_4t_8x2_msa(src, src_stride, dst, dst_stride,
2906 filter, weight, offset, rnd_val);
2907 } else if (4 == height) {
2908 hevc_hz_uniwgt_4t_8x4_msa(src, src_stride, dst, dst_stride,
2909 filter, weight, offset, rnd_val);
2910 } else if (6 == height) {
2911 hevc_hz_uniwgt_4t_8x6_msa(src, src_stride, dst, dst_stride,
2912 filter, weight, offset, rnd_val);
2913 } else {
2914 hevc_hz_uniwgt_4t_8x8multiple_msa(src, src_stride, dst, dst_stride,
2915 filter, height, weight, offset,
2916 rnd_val);
2917 }
2918 }
2919
hevc_hz_uniwgt_4t_12w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)2920 static void hevc_hz_uniwgt_4t_12w_msa(uint8_t *src,
2921 int32_t src_stride,
2922 uint8_t *dst,
2923 int32_t dst_stride,
2924 const int8_t *filter,
2925 int32_t height,
2926 int32_t weight,
2927 int32_t offset,
2928 int32_t rnd_val)
2929 {
2930 uint32_t loop_cnt;
2931 v16u8 out0, out1, out2;
2932 v8i16 filt0, filt1;
2933 v16i8 src0, src1, src2, src3;
2934 v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
2935 v16i8 mask2 = { 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28
2936 };
2937 v16i8 mask1;
2938 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9, vec10;
2939 v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
2940 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
2941 v16i8 mask3, vec11;
2942 v4i32 weight_vec, rnd_vec;
2943
2944 src -= 1;
2945
2946 filter_vec = LD_SH(filter);
2947 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
2948
2949 weight = weight & 0x0000FFFF;
2950
2951 weight_vec = __msa_fill_w(weight);
2952 rnd_vec = __msa_fill_w(rnd_val);
2953
2954 weight *= 128;
2955 rnd_val -= 6;
2956
2957 weight_vec_h = __msa_fill_h(weight);
2958 offset_vec = __msa_fill_h(offset);
2959 denom_vec = __msa_fill_h(rnd_val);
2960
2961 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
2962 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
2963
2964 mask1 = mask0 + 2;
2965 mask3 = mask2 + 2;
2966
2967 for (loop_cnt = 4; loop_cnt--;) {
2968 LD_SB4(src, src_stride, src0, src1, src2, src3);
2969 src += (4 * src_stride);
2970
2971 XORI_B4_128_SB(src0, src1, src2, src3);
2972
2973 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
2974 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
2975 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
2976 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
2977 VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec8, vec9);
2978 VSHF_B2_SB(src2, src3, src2, src3, mask2, mask3, vec10, vec11);
2979 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
2980 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
2981 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
2982 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
2983 dst4 = HEVC_FILT_4TAP_SH(vec8, vec9, filt0, filt1);
2984 dst5 = HEVC_FILT_4TAP_SH(vec10, vec11, filt0, filt1);
2985
2986 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
2987 weight_vec, offset_vec, rnd_vec,
2988 dst0, dst1, dst2, dst3);
2989
2990 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec,
2991 rnd_vec, dst4, dst5);
2992
2993 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
2994 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
2995 ST_W4(out2, 0, 1, 2, 3, dst + 8, dst_stride);
2996 dst += (4 * dst_stride);
2997 }
2998 }
2999
hevc_hz_uniwgt_4t_16w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3000 static void hevc_hz_uniwgt_4t_16w_msa(uint8_t *src,
3001 int32_t src_stride,
3002 uint8_t *dst,
3003 int32_t dst_stride,
3004 const int8_t *filter,
3005 int32_t height,
3006 int32_t weight,
3007 int32_t offset,
3008 int32_t rnd_val)
3009 {
3010 uint32_t loop_cnt;
3011 v16u8 out0, out1, out2, out3;
3012 v16i8 src0, src1, src2, src3, src4, src5, src6, src7;
3013 v8i16 filt0, filt1;
3014 v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
3015 v16i8 mask1;
3016 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
3017 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
3018 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3019 v4i32 weight_vec, rnd_vec;
3020
3021 src -= 1;
3022
3023 filter_vec = LD_SH(filter);
3024 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3025
3026 weight = weight & 0x0000FFFF;
3027
3028 weight_vec = __msa_fill_w(weight);
3029 rnd_vec = __msa_fill_w(rnd_val);
3030
3031 weight *= 128;
3032 rnd_val -= 6;
3033
3034 weight_vec_h = __msa_fill_h(weight);
3035 offset_vec = __msa_fill_h(offset);
3036 denom_vec = __msa_fill_h(rnd_val);
3037
3038 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3039 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3040
3041 mask1 = mask0 + 2;
3042
3043 for (loop_cnt = (height >> 2); loop_cnt--;) {
3044 LD_SB4(src, src_stride, src0, src2, src4, src6);
3045 LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
3046 src += (4 * src_stride);
3047
3048 XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7);
3049
3050 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
3051 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
3052 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
3053 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
3054 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
3055 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
3056 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
3057 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
3058 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec0, vec1);
3059 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec2, vec3);
3060 VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec4, vec5);
3061 VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec6, vec7);
3062 dst4 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
3063 dst5 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
3064 dst6 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
3065 dst7 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
3066
3067 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
3068 weight_vec, offset_vec, rnd_vec,
3069 dst0, dst1, dst2, dst3);
3070
3071 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7,
3072 weight_vec, offset_vec, rnd_vec,
3073 dst4, dst5, dst6, dst7);
3074
3075 PCKEV_B4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6,
3076 out0, out1, out2, out3);
3077
3078 ST_UB4(out0, out1, out2, out3, dst, dst_stride);
3079 dst += (4 * dst_stride);
3080 }
3081 }
3082
hevc_hz_uniwgt_4t_24w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3083 static void hevc_hz_uniwgt_4t_24w_msa(uint8_t *src,
3084 int32_t src_stride,
3085 uint8_t *dst,
3086 int32_t dst_stride,
3087 const int8_t *filter,
3088 int32_t height,
3089 int32_t weight,
3090 int32_t offset,
3091 int32_t rnd_val)
3092 {
3093 uint32_t loop_cnt;
3094 v16u8 out0, out1, out2;
3095 v16i8 src0, src1, src2, src3;
3096 v8i16 filt0, filt1;
3097 v16i8 mask0, mask1, mask2, mask3;
3098 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
3099 v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
3100 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3101 v4i32 weight_vec, rnd_vec;
3102
3103 src -= 1;
3104
3105 filter_vec = LD_SH(filter);
3106 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3107
3108 weight = weight & 0x0000FFFF;
3109 weight_vec = __msa_fill_w(weight);
3110 rnd_vec = __msa_fill_w(rnd_val);
3111
3112 weight *= 128;
3113 rnd_val -= 6;
3114
3115 weight_vec_h = __msa_fill_h(weight);
3116 offset_vec = __msa_fill_h(offset);
3117 denom_vec = __msa_fill_h(rnd_val);
3118
3119 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3120 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3121
3122 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
3123 mask1 = mask0 + 2;
3124 mask2 = mask0 + 8;
3125 mask3 = mask0 + 10;
3126
3127 for (loop_cnt = 16; loop_cnt--;) {
3128 LD_SB2(src, src_stride, src0, src2);
3129 LD_SB2(src + 16, src_stride, src1, src3);
3130 src += (2 * src_stride);
3131
3132 XORI_B4_128_SB(src0, src1, src2, src3);
3133
3134 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
3135 VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec2, vec3);
3136 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
3137 VSHF_B2_SB(src2, src3, src2, src3, mask2, mask3, vec6, vec7);
3138 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
3139 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
3140 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
3141 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
3142 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1);
3143 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec2, vec3);
3144 dst4 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
3145 dst5 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
3146
3147 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
3148 weight_vec, offset_vec, rnd_vec,
3149 dst0, dst1, dst2, dst3);
3150
3151 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec,
3152 rnd_vec, dst4, dst5);
3153
3154 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
3155 ST_UB2(out0, out1, dst, dst_stride);
3156 ST_D2(out2, 0, 1, dst + 16, dst_stride);
3157 dst += (2 * dst_stride);
3158 }
3159 }
3160
hevc_hz_uniwgt_4t_32w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3161 static void hevc_hz_uniwgt_4t_32w_msa(uint8_t *src,
3162 int32_t src_stride,
3163 uint8_t *dst,
3164 int32_t dst_stride,
3165 const int8_t *filter,
3166 int32_t height,
3167 int32_t weight,
3168 int32_t offset,
3169 int32_t rnd_val)
3170 {
3171 uint32_t loop_cnt;
3172 v16u8 out0, out1, out2, out3;
3173 v16i8 src0, src1, src2, src3, src4, src5;
3174 v8i16 filt0, filt1;
3175 v16i8 mask0 = LD_SB(&ff_hevc_mask_arr[0]);
3176 v16i8 mask1, mask2, mask3;
3177 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
3178 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
3179 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3180 v4i32 weight_vec, rnd_vec;
3181
3182 src -= 1;
3183
3184 filter_vec = LD_SH(filter);
3185 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3186
3187 weight = weight & 0x0000FFFF;
3188
3189 weight_vec = __msa_fill_w(weight);
3190 rnd_vec = __msa_fill_w(rnd_val);
3191
3192 weight *= 128;
3193 rnd_val -= 6;
3194
3195 weight_vec_h = __msa_fill_h(weight);
3196 offset_vec = __msa_fill_h(offset);
3197 denom_vec = __msa_fill_h(rnd_val);
3198
3199 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3200 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3201
3202 mask1 = mask0 + 2;
3203 mask2 = mask0 + 8;
3204 mask3 = mask0 + 10;
3205
3206 for (loop_cnt = (height >> 1); loop_cnt--;) {
3207 LD_SB2(src, 16, src0, src1);
3208 src2 = LD_SB(src + 24);
3209 src += src_stride;
3210 LD_SB2(src, 16, src3, src4);
3211 src5 = LD_SB(src + 24);
3212 src += src_stride;
3213 XORI_B6_128_SB(src0, src1, src2, src3, src4, src5);
3214 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
3215 VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec2, vec3);
3216 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec4, vec5);
3217 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec6, vec7);
3218 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
3219 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
3220 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
3221 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
3222 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
3223 VSHF_B2_SB(src3, src4, src3, src4, mask2, mask3, vec2, vec3);
3224 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec4, vec5);
3225 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec6, vec7);
3226 dst4 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
3227 dst5 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
3228 dst6 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
3229 dst7 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
3230
3231 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
3232 weight_vec, offset_vec, rnd_vec,
3233 dst0, dst1, dst2, dst3);
3234
3235 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7,
3236 weight_vec, offset_vec, rnd_vec,
3237 dst4, dst5, dst6, dst7);
3238
3239 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
3240 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
3241 ST_UB2(out0, out1, dst, 16);
3242 dst += dst_stride;
3243 ST_UB2(out2, out3, dst, 16);
3244 dst += dst_stride;
3245 }
3246 }
3247
hevc_vt_uniwgt_4t_4x2_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)3248 static void hevc_vt_uniwgt_4t_4x2_msa(uint8_t *src,
3249 int32_t src_stride,
3250 uint8_t *dst,
3251 int32_t dst_stride,
3252 const int8_t *filter,
3253 int32_t weight,
3254 int32_t offset,
3255 int32_t rnd_val)
3256 {
3257 v16u8 out;
3258 v16i8 src0, src1, src2, src3, src4;
3259 v16i8 src10_r, src32_r, src21_r, src43_r;
3260 v16i8 src2110, src4332;
3261 v8i16 dst0;
3262 v4i32 dst0_r, dst0_l;
3263 v8i16 filt0, filt1;
3264 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3265 v4i32 weight_vec, rnd_vec;
3266
3267 src -= src_stride;
3268
3269 weight = weight & 0x0000FFFF;
3270
3271 weight_vec = __msa_fill_w(weight);
3272 rnd_vec = __msa_fill_w(rnd_val);
3273
3274 weight *= 128;
3275 rnd_val -= 6;
3276
3277 weight_vec_h = __msa_fill_h(weight);
3278 offset_vec = __msa_fill_h(offset);
3279 denom_vec = __msa_fill_h(rnd_val);
3280
3281 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3282 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3283
3284 filter_vec = LD_SH(filter);
3285 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3286
3287 LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
3288 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3289 ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
3290 ILVR_D2_SB(src21_r, src10_r, src43_r, src32_r, src2110, src4332);
3291 XORI_B2_128_SB(src2110, src4332);
3292 dst0 = HEVC_FILT_4TAP_SH(src2110, src4332, filt0, filt1);
3293 ILVRL_H2_SW(dst0, dst0, dst0_r, dst0_l);
3294 DOTP_SH2_SW(dst0_r, dst0_l, weight_vec, weight_vec, dst0_r, dst0_l);
3295 SRAR_W2_SW(dst0_r, dst0_l, rnd_vec);
3296 dst0 = __msa_pckev_h((v8i16) dst0_l, (v8i16) dst0_r);
3297 dst0 = __msa_adds_s_h(dst0, offset_vec);
3298 CLIP_SH_0_255(dst0);
3299 out = (v16u8) __msa_pckev_b((v16i8) dst0, (v16i8) dst0);
3300 ST_W2(out, 0, 1, dst, dst_stride);
3301 }
3302
hevc_vt_uniwgt_4t_4x4_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)3303 static void hevc_vt_uniwgt_4t_4x4_msa(uint8_t *src,
3304 int32_t src_stride,
3305 uint8_t *dst,
3306 int32_t dst_stride,
3307 const int8_t *filter,
3308 int32_t weight,
3309 int32_t offset,
3310 int32_t rnd_val)
3311 {
3312 v16u8 out;
3313 v16i8 src0, src1, src2, src3, src4, src5, src6;
3314 v16i8 src10_r, src32_r, src54_r, src21_r, src43_r, src65_r;
3315 v16i8 src2110, src4332, src6554;
3316 v8i16 dst0, dst1;
3317 v8i16 filt0, filt1;
3318 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3319 v4i32 weight_vec, rnd_vec;
3320
3321 src -= src_stride;
3322
3323 weight = weight & 0x0000FFFF;
3324
3325 weight_vec = __msa_fill_w(weight);
3326 rnd_vec = __msa_fill_w(rnd_val);
3327
3328 weight *= 128;
3329 rnd_val -= 6;
3330
3331 weight_vec_h = __msa_fill_h(weight);
3332 offset_vec = __msa_fill_h(offset);
3333 denom_vec = __msa_fill_h(rnd_val);
3334
3335 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3336 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3337
3338 filter_vec = LD_SH(filter);
3339 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3340
3341 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
3342 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3343 ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5,
3344 src32_r, src43_r, src54_r, src65_r);
3345 ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r,
3346 src2110, src4332, src6554);
3347 XORI_B3_128_SB(src2110, src4332, src6554);
3348 dst0 = HEVC_FILT_4TAP_SH(src2110, src4332, filt0, filt1);
3349 dst1 = HEVC_FILT_4TAP_SH(src4332, src6554, filt0, filt1);
3350 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst0, dst1, weight_vec, offset_vec, rnd_vec,
3351 dst0, dst1);
3352
3353 out = (v16u8) __msa_pckev_b((v16i8) dst1, (v16i8) dst0);
3354 ST_W4(out, 0, 1, 2, 3, dst, dst_stride);
3355 }
3356
hevc_vt_uniwgt_4t_4x8multiple_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3357 static void hevc_vt_uniwgt_4t_4x8multiple_msa(uint8_t *src,
3358 int32_t src_stride,
3359 uint8_t *dst,
3360 int32_t dst_stride,
3361 const int8_t *filter,
3362 int32_t height,
3363 int32_t weight,
3364 int32_t offset,
3365 int32_t rnd_val)
3366 {
3367 int32_t loop_cnt;
3368 v16u8 out0, out1;
3369 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
3370 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r;
3371 v16i8 src21_r, src43_r, src65_r, src87_r, src109_r;
3372 v16i8 src2110, src4332, src6554, src8776;
3373 v16i8 src10998;
3374 v8i16 dst0, dst1, dst2, dst3, filt0, filt1;
3375 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3376 v4i32 weight_vec, rnd_vec;
3377
3378 src -= src_stride;
3379
3380 weight = weight & 0x0000FFFF;
3381
3382 weight_vec = __msa_fill_w(weight);
3383 rnd_vec = __msa_fill_w(rnd_val);
3384
3385 weight *= 128;
3386 rnd_val -= 6;
3387
3388 weight_vec_h = __msa_fill_h(weight);
3389 offset_vec = __msa_fill_h(offset);
3390 denom_vec = __msa_fill_h(rnd_val);
3391
3392 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3393 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3394
3395 filter_vec = LD_SH(filter);
3396 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3397
3398 LD_SB3(src, src_stride, src0, src1, src2);
3399 src += (3 * src_stride);
3400 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3401 src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r);
3402 src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128);
3403
3404 for (loop_cnt = (height >> 3); loop_cnt--;) {
3405 LD_SB8(src, src_stride,
3406 src3, src4, src5, src6, src7, src8, src9, src10);
3407 src += (8 * src_stride);
3408 ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5,
3409 src32_r, src43_r, src54_r, src65_r);
3410 ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r);
3411 ILVR_B2_SB(src9, src8, src10, src9, src98_r, src109_r);
3412 ILVR_D4_SB(src43_r, src32_r, src65_r, src54_r, src87_r, src76_r,
3413 src109_r, src98_r, src4332, src6554, src8776, src10998);
3414 XORI_B4_128_SB(src4332, src6554, src8776, src10998);
3415 dst0 = HEVC_FILT_4TAP_SH(src2110, src4332, filt0, filt1);
3416 dst1 = HEVC_FILT_4TAP_SH(src4332, src6554, filt0, filt1);
3417 dst2 = HEVC_FILT_4TAP_SH(src6554, src8776, filt0, filt1);
3418 dst3 = HEVC_FILT_4TAP_SH(src8776, src10998, filt0, filt1);
3419
3420 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
3421 weight_vec, offset_vec, rnd_vec,
3422 dst0, dst1, dst2, dst3);
3423
3424 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
3425 ST_W8(out0, out1, 0, 1, 2, 3, 0, 1, 2, 3, dst, dst_stride);
3426 dst += (8 * dst_stride);
3427
3428 src2 = src10;
3429 src2110 = src10998;
3430 }
3431 }
3432
hevc_vt_uniwgt_4t_4w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3433 static void hevc_vt_uniwgt_4t_4w_msa(uint8_t *src,
3434 int32_t src_stride,
3435 uint8_t *dst,
3436 int32_t dst_stride,
3437 const int8_t *filter,
3438 int32_t height,
3439 int32_t weight,
3440 int32_t offset,
3441 int32_t rnd_val)
3442 {
3443 if (2 == height) {
3444 hevc_vt_uniwgt_4t_4x2_msa(src, src_stride, dst, dst_stride,
3445 filter, weight, offset, rnd_val);
3446 } else if (4 == height) {
3447 hevc_vt_uniwgt_4t_4x4_msa(src, src_stride, dst, dst_stride,
3448 filter, weight, offset, rnd_val);
3449 } else if (0 == (height % 8)) {
3450 hevc_vt_uniwgt_4t_4x8multiple_msa(src, src_stride, dst, dst_stride,
3451 filter, height, weight, offset,
3452 rnd_val);
3453 }
3454 }
3455
hevc_vt_uniwgt_4t_6w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3456 static void hevc_vt_uniwgt_4t_6w_msa(uint8_t *src,
3457 int32_t src_stride,
3458 uint8_t *dst,
3459 int32_t dst_stride,
3460 const int8_t *filter,
3461 int32_t height,
3462 int32_t weight,
3463 int32_t offset,
3464 int32_t rnd_val)
3465 {
3466 v16u8 out0, out1, out2, out3;
3467 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
3468 v16i8 src10_r, src32_r, src21_r, src43_r;
3469 v16i8 src54_r, src65_r, src76_r, src87_r, src98_r, src109_r;
3470 v8i16 filt0, filt1;
3471 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
3472 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3473 v4i32 weight_vec, rnd_vec;
3474
3475 src -= src_stride;
3476
3477 weight = weight & 0x0000FFFF;
3478
3479 weight_vec = __msa_fill_w(weight);
3480 rnd_vec = __msa_fill_w(rnd_val);
3481
3482 weight *= 128;
3483 rnd_val -= 6;
3484
3485 weight_vec_h = __msa_fill_h(weight);
3486 offset_vec = __msa_fill_h(offset);
3487 denom_vec = __msa_fill_h(rnd_val);
3488
3489 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3490 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3491
3492 filter_vec = LD_SH(filter);
3493 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3494
3495 LD_SB3(src, src_stride, src0, src1, src2);
3496 src += (3 * src_stride);
3497 LD_SB8(src, src_stride, src3, src4, src5, src6, src7, src8, src9, src10);
3498 XORI_B3_128_SB(src0, src1, src2);
3499 XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
3500 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3501 ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
3502 ILVR_B2_SB(src5, src4, src6, src5, src54_r, src65_r);
3503 ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r);
3504 ILVR_B2_SB(src9, src8, src10, src9, src98_r, src109_r);
3505 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
3506 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
3507 dst2 = HEVC_FILT_4TAP_SH(src32_r, src54_r, filt0, filt1);
3508 dst3 = HEVC_FILT_4TAP_SH(src43_r, src65_r, filt0, filt1);
3509 dst4 = HEVC_FILT_4TAP_SH(src54_r, src76_r, filt0, filt1);
3510 dst5 = HEVC_FILT_4TAP_SH(src65_r, src87_r, filt0, filt1);
3511 dst6 = HEVC_FILT_4TAP_SH(src76_r, src98_r, filt0, filt1);
3512 dst7 = HEVC_FILT_4TAP_SH(src87_r, src109_r, filt0, filt1);
3513
3514 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3,
3515 weight_vec, offset_vec, rnd_vec,
3516 dst0, dst1, dst2, dst3);
3517 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7,
3518 weight_vec, offset_vec, rnd_vec,
3519 dst4, dst5, dst6, dst7);
3520
3521 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
3522 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
3523 ST_W2(out0, 0, 2, dst, dst_stride);
3524 ST_H2(out0, 2, 6, dst + 4, dst_stride);
3525 ST_W2(out1, 0, 2, dst + 2 * dst_stride, dst_stride);
3526 ST_H2(out1, 2, 6, dst + 2 * dst_stride + 4, dst_stride);
3527 dst += (4 * dst_stride);
3528 ST_W2(out2, 0, 2, dst, dst_stride);
3529 ST_H2(out2, 2, 6, dst + 4, dst_stride);
3530 ST_W2(out3, 0, 2, dst + 2 * dst_stride, dst_stride);
3531 ST_H2(out3, 2, 6, dst + 2 * dst_stride + 4, dst_stride);
3532 }
3533
hevc_vt_uniwgt_4t_8x2_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)3534 static void hevc_vt_uniwgt_4t_8x2_msa(uint8_t *src,
3535 int32_t src_stride,
3536 uint8_t *dst,
3537 int32_t dst_stride,
3538 const int8_t *filter,
3539 int32_t weight,
3540 int32_t offset,
3541 int32_t rnd_val)
3542 {
3543 v16u8 out;
3544 v16i8 src0, src1, src2, src3, src4;
3545 v16i8 src10_r, src32_r, src21_r, src43_r;
3546 v8i16 dst0, dst1;
3547 v8i16 filt0, filt1;
3548 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3549 v4i32 weight_vec, rnd_vec;
3550
3551 src -= src_stride;
3552
3553 weight = weight & 0x0000FFFF;
3554
3555 weight_vec = __msa_fill_w(weight);
3556 rnd_vec = __msa_fill_w(rnd_val);
3557
3558 weight *= 128;
3559 rnd_val -= 6;
3560
3561 weight_vec_h = __msa_fill_h(weight);
3562 offset_vec = __msa_fill_h(offset);
3563 denom_vec = __msa_fill_h(rnd_val);
3564
3565 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3566 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3567
3568 filter_vec = LD_SH(filter);
3569 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3570
3571 LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
3572 XORI_B5_128_SB(src0, src1, src2, src3, src4);
3573 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3574 ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
3575 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
3576 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
3577
3578 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst0, dst1, weight_vec, offset_vec, rnd_vec,
3579 dst0, dst1);
3580
3581 out = (v16u8) __msa_pckev_b((v16i8) dst1, (v16i8) dst0);
3582 ST_D2(out, 0, 1, dst, dst_stride);
3583 }
3584
hevc_vt_uniwgt_4t_8x4_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)3585 static void hevc_vt_uniwgt_4t_8x4_msa(uint8_t *src,
3586 int32_t src_stride,
3587 uint8_t *dst,
3588 int32_t dst_stride,
3589 const int8_t *filter,
3590 int32_t weight,
3591 int32_t offset,
3592 int32_t rnd_val)
3593 {
3594 v16u8 out0, out1;
3595 v16i8 src0, src1, src2, src3, src4;
3596 v16i8 src10_r, src32_r, src21_r, src43_r;
3597 v16i8 src5, src6, src54_r, src65_r;
3598 v8i16 filt0, filt1;
3599 v8i16 dst0, dst1, dst2, dst3;
3600 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3601 v4i32 weight_vec, rnd_vec;
3602
3603 src -= src_stride;
3604
3605 weight = weight & 0x0000FFFF;
3606
3607 weight_vec = __msa_fill_w(weight);
3608 rnd_vec = __msa_fill_w(rnd_val);
3609
3610 weight *= 128;
3611 rnd_val -= 6;
3612
3613 weight_vec_h = __msa_fill_h(weight);
3614 offset_vec = __msa_fill_h(offset);
3615 denom_vec = __msa_fill_h(rnd_val);
3616
3617 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3618 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3619
3620 filter_vec = LD_SH(filter);
3621 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3622
3623 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
3624 src += (3 * src_stride);
3625 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
3626 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3627 ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
3628 ILVR_B2_SB(src5, src4, src6, src5, src54_r, src65_r);
3629 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
3630 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
3631 dst2 = HEVC_FILT_4TAP_SH(src32_r, src54_r, filt0, filt1);
3632 dst3 = HEVC_FILT_4TAP_SH(src43_r, src65_r, filt0, filt1);
3633 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
3634 offset_vec, rnd_vec, dst0, dst1, dst2,
3635 dst3);
3636 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
3637 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
3638 }
3639
hevc_vt_uniwgt_4t_8x6_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t weight,int32_t offset,int32_t rnd_val)3640 static void hevc_vt_uniwgt_4t_8x6_msa(uint8_t *src,
3641 int32_t src_stride,
3642 uint8_t *dst,
3643 int32_t dst_stride,
3644 const int8_t *filter,
3645 int32_t weight,
3646 int32_t offset,
3647 int32_t rnd_val)
3648 {
3649 v16u8 out0, out1, out2;
3650 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
3651 v16i8 src10_r, src32_r, src54_r, src76_r;
3652 v16i8 src21_r, src43_r, src65_r, src87_r;
3653 v8i16 dst0, dst1, dst2, dst3, dst4, dst5;
3654 v8i16 filt0, filt1;
3655 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3656 v4i32 weight_vec, rnd_vec;
3657
3658 src -= src_stride;
3659
3660 weight = weight & 0x0000FFFF;
3661
3662 weight_vec = __msa_fill_w(weight);
3663 rnd_vec = __msa_fill_w(rnd_val);
3664
3665 weight *= 128;
3666 rnd_val -= 6;
3667
3668 weight_vec_h = __msa_fill_h(weight);
3669 offset_vec = __msa_fill_h(offset);
3670 denom_vec = __msa_fill_h(rnd_val);
3671
3672 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3673 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3674
3675 filter_vec = LD_SH(filter);
3676 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3677
3678 LD_SB3(src, src_stride, src0, src1, src2);
3679 src += (3 * src_stride);
3680 LD_SB6(src, src_stride, src3, src4, src5, src6, src7, src8);
3681
3682 XORI_B3_128_SB(src0, src1, src2);
3683 XORI_B6_128_SB(src3, src4, src5, src6, src7, src8);
3684 ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
3685 src32_r, src43_r);
3686 ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r,
3687 src76_r, src87_r);
3688 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
3689 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
3690 dst2 = HEVC_FILT_4TAP_SH(src32_r, src54_r, filt0, filt1);
3691 dst3 = HEVC_FILT_4TAP_SH(src43_r, src65_r, filt0, filt1);
3692 dst4 = HEVC_FILT_4TAP_SH(src54_r, src76_r, filt0, filt1);
3693 dst5 = HEVC_FILT_4TAP_SH(src65_r, src87_r, filt0, filt1);
3694 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
3695 offset_vec, rnd_vec, dst0, dst1, dst2, dst3);
3696 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec, rnd_vec,
3697 dst4, dst5);
3698 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
3699 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
3700 ST_D2(out2, 0, 1, dst + 4 * dst_stride, dst_stride);
3701 }
3702
hevc_vt_uniwgt_4t_8x8mult_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3703 static void hevc_vt_uniwgt_4t_8x8mult_msa(uint8_t *src,
3704 int32_t src_stride,
3705 uint8_t *dst,
3706 int32_t dst_stride,
3707 const int8_t *filter,
3708 int32_t height,
3709 int32_t weight,
3710 int32_t offset,
3711 int32_t rnd_val)
3712 {
3713 int32_t loop_cnt;
3714 v16u8 out0, out1, out2, out3;
3715 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
3716 v16i8 src10_r, src32_r, src21_r, src43_r;
3717 v16i8 src54_r, src65_r, src76_r, src87_r, src98_r, src109_r;
3718 v8i16 filt0, filt1;
3719 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
3720 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3721 v4i32 weight_vec, rnd_vec;
3722
3723 src -= src_stride;
3724
3725 weight = weight & 0x0000FFFF;
3726
3727 weight_vec = __msa_fill_w(weight);
3728 rnd_vec = __msa_fill_w(rnd_val);
3729
3730 weight *= 128;
3731 rnd_val -= 6;
3732
3733 weight_vec_h = __msa_fill_h(weight);
3734 offset_vec = __msa_fill_h(offset);
3735 denom_vec = __msa_fill_h(rnd_val);
3736
3737 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3738 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3739
3740 filter_vec = LD_SH(filter);
3741 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3742
3743 LD_SB3(src, src_stride, src0, src1, src2);
3744 src += (3 * src_stride);
3745 XORI_B3_128_SB(src0, src1, src2);
3746 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3747
3748 for (loop_cnt = (height >> 3); loop_cnt--;) {
3749 LD_SB8(src, src_stride,
3750 src3, src4, src5, src6, src7, src8, src9, src10);
3751 src += (8 * src_stride);
3752 XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
3753 ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
3754 ILVR_B2_SB(src5, src4, src6, src5, src54_r, src65_r);
3755 ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r);
3756 ILVR_B2_SB(src9, src8, src10, src9, src98_r, src109_r);
3757 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
3758 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
3759 dst2 = HEVC_FILT_4TAP_SH(src32_r, src54_r, filt0, filt1);
3760 dst3 = HEVC_FILT_4TAP_SH(src43_r, src65_r, filt0, filt1);
3761 dst4 = HEVC_FILT_4TAP_SH(src54_r, src76_r, filt0, filt1);
3762 dst5 = HEVC_FILT_4TAP_SH(src65_r, src87_r, filt0, filt1);
3763 dst6 = HEVC_FILT_4TAP_SH(src76_r, src98_r, filt0, filt1);
3764 dst7 = HEVC_FILT_4TAP_SH(src87_r, src109_r, filt0, filt1);
3765 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
3766 offset_vec, rnd_vec, dst0, dst1, dst2,
3767 dst3);
3768 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
3769 offset_vec, rnd_vec, dst4, dst5, dst6,
3770 dst7);
3771 PCKEV_B2_UB(dst1, dst0, dst3, dst2, out0, out1);
3772 PCKEV_B2_UB(dst5, dst4, dst7, dst6, out2, out3);
3773 ST_D8(out0, out1, out2, out3, 0, 1, 0, 1, 0, 1, 0, 1, dst, dst_stride);
3774 dst += (8 * dst_stride);
3775
3776 src2 = src10;
3777 src10_r = src98_r;
3778 src21_r = src109_r;
3779 }
3780 }
3781
hevc_vt_uniwgt_4t_8w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3782 static void hevc_vt_uniwgt_4t_8w_msa(uint8_t *src,
3783 int32_t src_stride,
3784 uint8_t *dst,
3785 int32_t dst_stride,
3786 const int8_t *filter,
3787 int32_t height,
3788 int32_t weight,
3789 int32_t offset,
3790 int32_t rnd_val)
3791 {
3792 if (2 == height) {
3793 hevc_vt_uniwgt_4t_8x2_msa(src, src_stride, dst, dst_stride,
3794 filter, weight, offset, rnd_val);
3795 } else if (4 == height) {
3796 hevc_vt_uniwgt_4t_8x4_msa(src, src_stride, dst, dst_stride,
3797 filter, weight, offset, rnd_val);
3798 } else if (6 == height) {
3799 hevc_vt_uniwgt_4t_8x6_msa(src, src_stride, dst, dst_stride,
3800 filter, weight, offset, rnd_val);
3801 } else {
3802 hevc_vt_uniwgt_4t_8x8mult_msa(src, src_stride, dst, dst_stride,
3803 filter, height, weight, offset,
3804 rnd_val);
3805 }
3806 }
3807
hevc_vt_uniwgt_4t_12w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3808 static void hevc_vt_uniwgt_4t_12w_msa(uint8_t *src,
3809 int32_t src_stride,
3810 uint8_t *dst,
3811 int32_t dst_stride,
3812 const int8_t *filter,
3813 int32_t height,
3814 int32_t weight,
3815 int32_t offset,
3816 int32_t rnd_val)
3817 {
3818 int32_t loop_cnt;
3819 v16u8 out0, out1, out2, out3, out4, out5;
3820 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
3821 v16i8 src10_r, src32_r, src21_r, src43_r;
3822 v16i8 src10_l, src32_l, src54_l, src21_l, src43_l, src65_l;
3823 v16i8 src2110, src4332;
3824 v16i8 src54_r, src76_r, src98_r, src65_r, src87_r, src109_r;
3825 v16i8 src76_l, src98_l, src87_l, src109_l, src6554, src8776, src10998;
3826 v8i16 filt0, filt1;
3827 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
3828 v8i16 dst9, dst10, dst11, filter_vec, weight_vec_h, offset_vec, denom_vec;
3829 v4i32 weight_vec, rnd_vec;
3830
3831 src -= (1 * src_stride);
3832
3833 weight = weight & 0x0000FFFF;
3834
3835 weight_vec = __msa_fill_w(weight);
3836 rnd_vec = __msa_fill_w(rnd_val);
3837
3838 weight *= 128;
3839 rnd_val -= 6;
3840
3841 weight_vec_h = __msa_fill_h(weight);
3842 offset_vec = __msa_fill_h(offset);
3843 denom_vec = __msa_fill_h(rnd_val);
3844
3845 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3846 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3847
3848 filter_vec = LD_SH(filter);
3849 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3850
3851 LD_SB3(src, src_stride, src0, src1, src2);
3852 src += (3 * src_stride);
3853 XORI_B3_128_SB(src0, src1, src2);
3854 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3855 ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l);
3856 src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_l, (v2i64) src10_l);
3857
3858 for (loop_cnt = 2; loop_cnt--;) {
3859 LD_SB8(src, src_stride, src3, src4, src5, src6, src7, src8, src9, src10);
3860 src += (8 * src_stride);
3861 XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
3862 ILVRL_B2_SB(src3, src2, src32_r, src32_l);
3863 ILVRL_B2_SB(src4, src3, src43_r, src43_l);
3864 ILVRL_B2_SB(src5, src4, src54_r, src54_l);
3865 ILVRL_B2_SB(src6, src5, src65_r, src65_l);
3866 src4332 = (v16i8) __msa_ilvr_d((v2i64) src43_l, (v2i64) src32_l);
3867 src6554 = (v16i8) __msa_ilvr_d((v2i64) src65_l, (v2i64) src54_l);
3868 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
3869 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
3870 dst2 = HEVC_FILT_4TAP_SH(src32_r, src54_r, filt0, filt1);
3871 dst3 = HEVC_FILT_4TAP_SH(src43_r, src65_r, filt0, filt1);
3872 dst4 = HEVC_FILT_4TAP_SH(src2110, src4332, filt0, filt1);
3873 dst5 = HEVC_FILT_4TAP_SH(src4332, src6554, filt0, filt1);
3874 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
3875 offset_vec, rnd_vec, dst0, dst1, dst2,
3876 dst3);
3877 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst4, dst5, weight_vec, offset_vec,
3878 rnd_vec, dst4, dst5);
3879 PCKEV_B3_UB(dst1, dst0, dst3, dst2, dst5, dst4, out0, out1, out2);
3880 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
3881 ST_W4(out2, 0, 1, 2, 3, dst + 8, dst_stride);
3882 dst += (4 * dst_stride);
3883
3884 ILVRL_B2_SB(src7, src6, src76_r, src76_l);
3885 ILVRL_B2_SB(src8, src7, src87_r, src87_l);
3886 ILVRL_B2_SB(src9, src8, src98_r, src98_l);
3887 ILVRL_B2_SB(src10, src9, src109_r, src109_l);
3888 src8776 = (v16i8) __msa_ilvr_d((v2i64) src87_l, (v2i64) src76_l);
3889 src10998 = (v16i8) __msa_ilvr_d((v2i64) src109_l, (v2i64) src98_l);
3890 dst6 = HEVC_FILT_4TAP_SH(src54_r, src76_r, filt0, filt1);
3891 dst7 = HEVC_FILT_4TAP_SH(src65_r, src87_r, filt0, filt1);
3892 dst8 = HEVC_FILT_4TAP_SH(src76_r, src98_r, filt0, filt1);
3893 dst9 = HEVC_FILT_4TAP_SH(src87_r, src109_r, filt0, filt1);
3894 dst10 = HEVC_FILT_4TAP_SH(src6554, src8776, filt0, filt1);
3895 dst11 = HEVC_FILT_4TAP_SH(src8776, src10998, filt0, filt1);
3896 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst6, dst7, dst8, dst9, weight_vec,
3897 offset_vec, rnd_vec, dst6, dst7, dst8,
3898 dst9);
3899 HEVC_UNIW_RND_CLIP2_MAX_SATU_H(dst10, dst11, weight_vec, offset_vec,
3900 rnd_vec, dst10, dst11);
3901 PCKEV_B3_UB(dst7, dst6, dst9, dst8, dst11, dst10, out3, out4, out5);
3902 ST_D4(out3, out4, 0, 1, 0, 1, dst, dst_stride);
3903 ST_W4(out5, 0, 1, 2, 3, dst + 8, dst_stride);
3904 dst += (4 * dst_stride);
3905
3906 src2 = src10;
3907 src10_r = src98_r;
3908 src21_r = src109_r;
3909 src2110 = src10998;
3910 }
3911 }
3912
hevc_vt_uniwgt_4t_16w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3913 static void hevc_vt_uniwgt_4t_16w_msa(uint8_t *src,
3914 int32_t src_stride,
3915 uint8_t *dst,
3916 int32_t dst_stride,
3917 const int8_t *filter,
3918 int32_t height,
3919 int32_t weight,
3920 int32_t offset,
3921 int32_t rnd_val)
3922 {
3923 int32_t loop_cnt;
3924 v16u8 out0, out1, out2, out3;
3925 v16i8 src0, src1, src2, src3, src4, src5;
3926 v16i8 src10_r, src32_r, src21_r, src43_r;
3927 v16i8 src10_l, src32_l, src21_l, src43_l;
3928 v16i8 src54_r, src54_l, src65_r, src65_l, src6;
3929 v8i16 filt0, filt1;
3930 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
3931 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
3932 v4i32 weight_vec, rnd_vec;
3933
3934 src -= src_stride;
3935
3936 weight = weight & 0x0000FFFF;
3937
3938 weight_vec = __msa_fill_w(weight);
3939 rnd_vec = __msa_fill_w(rnd_val);
3940
3941 weight *= 128;
3942 rnd_val -= 6;
3943
3944 weight_vec_h = __msa_fill_h(weight);
3945 offset_vec = __msa_fill_h(offset);
3946 denom_vec = __msa_fill_h(rnd_val);
3947
3948 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
3949 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
3950
3951 filter_vec = LD_SH(filter);
3952 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
3953
3954 LD_SB3(src, src_stride, src0, src1, src2);
3955 src += (3 * src_stride);
3956 XORI_B3_128_SB(src0, src1, src2);
3957 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
3958 ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l);
3959
3960 for (loop_cnt = (height >> 2); loop_cnt--;) {
3961 LD_SB4(src, src_stride, src3, src4, src5, src6);
3962 src += (4 * src_stride);
3963 XORI_B4_128_SB(src3, src4, src5, src6);
3964 ILVRL_B2_SB(src3, src2, src32_r, src32_l);
3965 ILVRL_B2_SB(src4, src3, src43_r, src43_l);
3966 ILVRL_B2_SB(src5, src4, src54_r, src54_l);
3967 ILVRL_B2_SB(src6, src5, src65_r, src65_l);
3968 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
3969 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
3970 dst2 = HEVC_FILT_4TAP_SH(src32_r, src54_r, filt0, filt1);
3971 dst3 = HEVC_FILT_4TAP_SH(src43_r, src65_r, filt0, filt1);
3972 dst4 = HEVC_FILT_4TAP_SH(src10_l, src32_l, filt0, filt1);
3973 dst5 = HEVC_FILT_4TAP_SH(src21_l, src43_l, filt0, filt1);
3974 dst6 = HEVC_FILT_4TAP_SH(src32_l, src54_l, filt0, filt1);
3975 dst7 = HEVC_FILT_4TAP_SH(src43_l, src65_l, filt0, filt1);
3976 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
3977 offset_vec, rnd_vec, dst0, dst1, dst2,
3978 dst3);
3979 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
3980 offset_vec, rnd_vec, dst4, dst5, dst6,
3981 dst7);
3982 PCKEV_B4_UB(dst4, dst0, dst5, dst1, dst6, dst2, dst7, dst3, out0, out1,
3983 out2, out3);
3984 ST_UB4(out0, out1, out2, out3, dst, dst_stride);
3985 dst += (4 * dst_stride);
3986
3987 src2 = src6;
3988 src10_r = src54_r;
3989 src21_r = src65_r;
3990 src10_l = src54_l;
3991 src21_l = src65_l;
3992 }
3993 }
3994
hevc_vt_uniwgt_4t_24w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)3995 static void hevc_vt_uniwgt_4t_24w_msa(uint8_t *src,
3996 int32_t src_stride,
3997 uint8_t *dst,
3998 int32_t dst_stride,
3999 const int8_t *filter,
4000 int32_t height,
4001 int32_t weight,
4002 int32_t offset,
4003 int32_t rnd_val)
4004 {
4005 uint32_t loop_cnt;
4006 v16u8 out0, out1, out2, out3, out4, out5;
4007 v16i8 src0, src1, src2, src3, src4, src5;
4008 v16i8 src6, src7, src8, src9, src10, src11, src12, src13;
4009 v16i8 src10_r, src32_r, src54_r, src21_r, src43_r, src65_r;
4010 v16i8 src10_l, src32_l, src54_l, src21_l, src43_l, src65_l;
4011 v16i8 src87_r, src98_r, src109_r, src1110_r, src1211_r, src1312_r;
4012 v8i16 filt0, filt1;
4013 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8, dst9, dst10;
4014 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec, dst11;
4015 v4i32 weight_vec, rnd_vec;
4016
4017 src -= src_stride;
4018
4019 weight = weight & 0x0000FFFF;
4020
4021 weight_vec = __msa_fill_w(weight);
4022 rnd_vec = __msa_fill_w(rnd_val);
4023
4024 weight *= 128;
4025 rnd_val -= 6;
4026
4027 weight_vec_h = __msa_fill_h(weight);
4028 offset_vec = __msa_fill_h(offset);
4029 denom_vec = __msa_fill_h(rnd_val);
4030
4031 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
4032 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
4033
4034 filter_vec = LD_SH(filter);
4035 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4036
4037 LD_SB3(src, src_stride, src0, src1, src2);
4038 LD_SB3(src + 16, src_stride, src7, src8, src9);
4039 src += (3 * src_stride);
4040 XORI_B3_128_SB(src0, src1, src2);
4041 XORI_B3_128_SB(src7, src8, src9);
4042 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
4043 ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l);
4044 ILVR_B2_SB(src8, src7, src9, src8, src87_r, src98_r);
4045
4046 for (loop_cnt = 8; loop_cnt--;) {
4047 LD_SB4(src, src_stride, src3, src4, src5, src6);
4048 LD_SB4(src + 16, src_stride, src10, src11, src12, src13);
4049 src += (4 * src_stride);
4050 XORI_B4_128_SB(src3, src4, src5, src6);
4051 XORI_B4_128_SB(src10, src11, src12, src13);
4052 ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
4053 ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l);
4054 ILVRL_B2_SB(src5, src4, src54_r, src54_l);
4055 ILVRL_B2_SB(src6, src5, src65_r, src65_l);
4056 ILVR_B2_SB(src10, src9, src11, src10, src109_r, src1110_r);
4057 ILVR_B2_SB(src12, src11, src13, src12, src1211_r, src1312_r);
4058 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
4059 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
4060 dst2 = HEVC_FILT_4TAP_SH(src32_r, src54_r, filt0, filt1);
4061 dst3 = HEVC_FILT_4TAP_SH(src43_r, src65_r, filt0, filt1);
4062 dst4 = HEVC_FILT_4TAP_SH(src10_l, src32_l, filt0, filt1);
4063 dst5 = HEVC_FILT_4TAP_SH(src21_l, src43_l, filt0, filt1);
4064 dst6 = HEVC_FILT_4TAP_SH(src32_l, src54_l, filt0, filt1);
4065 dst7 = HEVC_FILT_4TAP_SH(src43_l, src65_l, filt0, filt1);
4066 dst8 = HEVC_FILT_4TAP_SH(src87_r, src109_r, filt0, filt1);
4067 dst9 = HEVC_FILT_4TAP_SH(src98_r, src1110_r, filt0, filt1);
4068 dst10 = HEVC_FILT_4TAP_SH(src109_r, src1211_r, filt0, filt1);
4069 dst11 = HEVC_FILT_4TAP_SH(src1110_r, src1312_r, filt0, filt1);
4070 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
4071 offset_vec, rnd_vec, dst0, dst1, dst2,
4072 dst3);
4073 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
4074 offset_vec, rnd_vec, dst4, dst5, dst6,
4075 dst7);
4076 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst8, dst9, dst10, dst11, weight_vec,
4077 offset_vec, rnd_vec, dst8, dst9, dst10,
4078 dst11);
4079 PCKEV_B4_UB(dst4, dst0, dst5, dst1, dst6, dst2, dst7, dst3, out0, out1,
4080 out2, out3);
4081 PCKEV_B2_UB(dst9, dst8, dst11, dst10, out4, out5);
4082 ST_UB4(out0, out1, out2, out3, dst, dst_stride);
4083 ST_D4(out4, out5, 0, 1, 0, 1, dst + 16, dst_stride);
4084 dst += (4 * dst_stride);
4085
4086 src2 = src6;
4087 src9 = src13;
4088 src10_r = src54_r;
4089 src21_r = src65_r;
4090 src10_l = src54_l;
4091 src21_l = src65_l;
4092 src87_r = src1211_r;
4093 src98_r = src1312_r;
4094 }
4095 }
4096
hevc_vt_uniwgt_4t_32w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)4097 static void hevc_vt_uniwgt_4t_32w_msa(uint8_t *src,
4098 int32_t src_stride,
4099 uint8_t *dst,
4100 int32_t dst_stride,
4101 const int8_t *filter,
4102 int32_t height,
4103 int32_t weight,
4104 int32_t offset,
4105 int32_t rnd_val)
4106 {
4107 uint32_t loop_cnt;
4108 v16u8 out0, out1, out2, out3;
4109 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
4110 v16i8 src10_r, src32_r, src76_r, src98_r;
4111 v16i8 src21_r, src43_r, src65_r, src87_r;
4112 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
4113 v16i8 src10_l, src32_l, src76_l, src98_l;
4114 v16i8 src21_l, src43_l, src65_l, src87_l;
4115 v8i16 filt0, filt1;
4116 v8i16 filter_vec, weight_vec_h, offset_vec, denom_vec;
4117 v4i32 weight_vec, rnd_vec;
4118
4119 src -= src_stride;
4120
4121 weight = weight & 0x0000FFFF;
4122
4123 weight_vec = __msa_fill_w(weight);
4124 rnd_vec = __msa_fill_w(rnd_val);
4125
4126 weight *= 128;
4127 rnd_val -= 6;
4128
4129 weight_vec_h = __msa_fill_h(weight);
4130 offset_vec = __msa_fill_h(offset);
4131 denom_vec = __msa_fill_h(rnd_val);
4132
4133 weight_vec_h = __msa_srar_h(weight_vec_h, denom_vec);
4134 offset_vec = __msa_adds_s_h(offset_vec, weight_vec_h);
4135
4136 filter_vec = LD_SH(filter);
4137 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4138
4139 LD_SB3(src, src_stride, src0, src1, src2);
4140 LD_SB3(src + 16, src_stride, src5, src6, src7);
4141 src += (3 * src_stride);
4142 XORI_B6_128_SB(src0, src1, src2, src5, src6, src7);
4143 ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r);
4144 ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l);
4145 ILVR_B2_SB(src6, src5, src7, src6, src65_r, src76_r);
4146 ILVL_B2_SB(src6, src5, src7, src6, src65_l, src76_l);
4147
4148 for (loop_cnt = (height >> 1); loop_cnt--;) {
4149 LD_SB2(src, src_stride, src3, src4);
4150 LD_SB2(src + 16, src_stride, src8, src9);
4151 src += (2 * src_stride);
4152 XORI_B4_128_SB(src3, src4, src8, src9);
4153 ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r);
4154 ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l);
4155 ILVRL_B2_SB(src8, src7, src87_r, src87_l);
4156 ILVRL_B2_SB(src9, src8, src98_r, src98_l);
4157 dst0 = HEVC_FILT_4TAP_SH(src10_r, src32_r, filt0, filt1);
4158 dst1 = HEVC_FILT_4TAP_SH(src21_r, src43_r, filt0, filt1);
4159 dst2 = HEVC_FILT_4TAP_SH(src10_l, src32_l, filt0, filt1);
4160 dst3 = HEVC_FILT_4TAP_SH(src21_l, src43_l, filt0, filt1);
4161 dst4 = HEVC_FILT_4TAP_SH(src65_r, src87_r, filt0, filt1);
4162 dst5 = HEVC_FILT_4TAP_SH(src76_r, src98_r, filt0, filt1);
4163 dst6 = HEVC_FILT_4TAP_SH(src65_l, src87_l, filt0, filt1);
4164 dst7 = HEVC_FILT_4TAP_SH(src76_l, src98_l, filt0, filt1);
4165 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst0, dst1, dst2, dst3, weight_vec,
4166 offset_vec, rnd_vec, dst0, dst1, dst2,
4167 dst3);
4168 HEVC_UNIW_RND_CLIP4_MAX_SATU_H(dst4, dst5, dst6, dst7, weight_vec,
4169 offset_vec, rnd_vec, dst4, dst5, dst6,
4170 dst7);
4171 PCKEV_B4_UB(dst2, dst0, dst3, dst1, dst6, dst4, dst7, dst5, out0, out1,
4172 out2, out3);
4173 ST_UB2(out0, out2, dst, 16);
4174 dst += dst_stride;
4175 ST_UB2(out1, out3, dst, 16);
4176 dst += dst_stride;
4177
4178 src2 = src4;
4179 src7 = src9;
4180 src10_r = src32_r;
4181 src21_r = src43_r;
4182 src10_l = src32_l;
4183 src21_l = src43_l;
4184 src65_r = src87_r;
4185 src76_r = src98_r;
4186 src65_l = src87_l;
4187 src76_l = src98_l;
4188 }
4189 }
4190
hevc_hv_uniwgt_4t_4x2_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t weight,int32_t offset,int32_t rnd_val)4191 static void hevc_hv_uniwgt_4t_4x2_msa(uint8_t *src,
4192 int32_t src_stride,
4193 uint8_t *dst,
4194 int32_t dst_stride,
4195 const int8_t *filter_x,
4196 const int8_t *filter_y,
4197 int32_t weight,
4198 int32_t offset,
4199 int32_t rnd_val)
4200 {
4201 v16u8 out;
4202 v16i8 src0, src1, src2, src3, src4;
4203 v8i16 filt0, filt1;
4204 v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
4205 v16i8 mask1;
4206 v8i16 filt_h0, filt_h1, filter_vec, tmp;
4207 v16i8 vec0, vec1, vec2, vec3, vec4, vec5;
4208 v8i16 dst20, dst31, dst42, dst10, dst32, dst21, dst43;
4209 v8i16 offset_vec, const_128, denom_vec;
4210 v4i32 dst0, dst1, weight_vec, rnd_vec;
4211
4212 src -= (src_stride + 1);
4213
4214 filter_vec = LD_SH(filter_x);
4215 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4216
4217 filter_vec = LD_SH(filter_y);
4218 UNPCK_R_SB_SH(filter_vec, filter_vec);
4219
4220 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
4221
4222 mask1 = mask0 + 2;
4223
4224 weight_vec = __msa_fill_w(weight);
4225 rnd_vec = __msa_fill_w(rnd_val);
4226
4227 offset_vec = __msa_fill_h(offset);
4228 denom_vec = __msa_fill_h(rnd_val - 6);
4229 const_128 = __msa_fill_h((128 * weight));
4230 offset_vec += __msa_srar_h(const_128, denom_vec);
4231
4232 LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
4233 XORI_B5_128_SB(src0, src1, src2, src3, src4);
4234 VSHF_B2_SB(src0, src2, src0, src2, mask0, mask1, vec0, vec1);
4235 VSHF_B2_SB(src1, src3, src1, src3, mask0, mask1, vec2, vec3);
4236 VSHF_B2_SB(src2, src4, src2, src4, mask0, mask1, vec4, vec5);
4237 dst20 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4238 dst31 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4239 dst42 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4240 ILVRL_H2_SH(dst31, dst20, dst10, dst32);
4241 ILVRL_H2_SH(dst42, dst31, dst21, dst43);
4242 dst0 = HEVC_FILT_4TAP(dst10, dst32, filt_h0, filt_h1);
4243 dst1 = HEVC_FILT_4TAP(dst21, dst43, filt_h0, filt_h1);
4244 dst0 >>= 6;
4245 dst1 >>= 6;
4246 MUL2(dst0, weight_vec, dst1, weight_vec, dst0, dst1);
4247 SRAR_W2_SW(dst0, dst1, rnd_vec);
4248 tmp = __msa_pckev_h((v8i16) dst1, (v8i16) dst0);
4249 tmp += offset_vec;
4250 CLIP_SH_0_255(tmp);
4251 out = (v16u8) __msa_pckev_b((v16i8) tmp, (v16i8) tmp);
4252 ST_W2(out, 0, 1, dst, dst_stride);
4253 }
4254
hevc_hv_uniwgt_4t_4x4_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t weight,int32_t offset,int32_t rnd_val)4255 static void hevc_hv_uniwgt_4t_4x4_msa(uint8_t *src,
4256 int32_t src_stride,
4257 uint8_t *dst,
4258 int32_t dst_stride,
4259 const int8_t *filter_x,
4260 const int8_t *filter_y,
4261 int32_t weight,
4262 int32_t offset,
4263 int32_t rnd_val)
4264 {
4265 v16u8 out;
4266 v16i8 src0, src1, src2, src3, src4, src5, src6;
4267 v8i16 filt0, filt1;
4268 v8i16 filt_h0, filt_h1, filter_vec, tmp0, tmp1;
4269 v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
4270 v16i8 mask1;
4271 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
4272 v8i16 dst30, dst41, dst52, dst63, dst10, dst32, dst54, dst21, dst43, dst65;
4273 v8i16 offset_vec, const_128, denom_vec;
4274 v4i32 dst0, dst1, dst2, dst3, weight_vec, rnd_vec;
4275
4276 src -= (src_stride + 1);
4277
4278 filter_vec = LD_SH(filter_x);
4279 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4280
4281 filter_vec = LD_SH(filter_y);
4282 UNPCK_R_SB_SH(filter_vec, filter_vec);
4283
4284 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
4285
4286 mask1 = mask0 + 2;
4287
4288 weight_vec = __msa_fill_w(weight);
4289 rnd_vec = __msa_fill_w(rnd_val);
4290
4291 offset_vec = __msa_fill_h(offset);
4292 denom_vec = __msa_fill_h(rnd_val - 6);
4293 const_128 = __msa_fill_h((128 * weight));
4294 offset_vec += __msa_srar_h(const_128, denom_vec);
4295
4296 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
4297 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
4298 VSHF_B2_SB(src0, src3, src0, src3, mask0, mask1, vec0, vec1);
4299 VSHF_B2_SB(src1, src4, src1, src4, mask0, mask1, vec2, vec3);
4300 VSHF_B2_SB(src2, src5, src2, src5, mask0, mask1, vec4, vec5);
4301 VSHF_B2_SB(src3, src6, src3, src6, mask0, mask1, vec6, vec7);
4302 dst30 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4303 dst41 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4304 dst52 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4305 dst63 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
4306 ILVRL_H2_SH(dst41, dst30, dst10, dst43);
4307 ILVRL_H2_SH(dst52, dst41, dst21, dst54);
4308 ILVRL_H2_SH(dst63, dst52, dst32, dst65);
4309 dst0 = HEVC_FILT_4TAP(dst10, dst32, filt_h0, filt_h1);
4310 dst1 = HEVC_FILT_4TAP(dst21, dst43, filt_h0, filt_h1);
4311 dst2 = HEVC_FILT_4TAP(dst32, dst54, filt_h0, filt_h1);
4312 dst3 = HEVC_FILT_4TAP(dst43, dst65, filt_h0, filt_h1);
4313 SRA_4V(dst0, dst1, dst2, dst3, 6);
4314 MUL2(dst0, weight_vec, dst1, weight_vec, dst0, dst1);
4315 MUL2(dst2, weight_vec, dst3, weight_vec, dst2, dst3);
4316 SRAR_W4_SW(dst0, dst1, dst2, dst3, rnd_vec);
4317 PCKEV_H2_SH(dst1, dst0, dst3, dst2, tmp0, tmp1);
4318 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
4319 CLIP_SH2_0_255(tmp0, tmp1);
4320 out = (v16u8) __msa_pckev_b((v16i8) tmp1, (v16i8) tmp0);
4321 ST_W4(out, 0, 1, 2, 3, dst, dst_stride);
4322 }
4323
hevc_hv_uniwgt_4t_4multx8mult_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)4324 static void hevc_hv_uniwgt_4t_4multx8mult_msa(uint8_t *src,
4325 int32_t src_stride,
4326 uint8_t *dst,
4327 int32_t dst_stride,
4328 const int8_t *filter_x,
4329 const int8_t *filter_y,
4330 int32_t height,
4331 int32_t weight,
4332 int32_t offset,
4333 int32_t rnd_val)
4334 {
4335 uint32_t loop_cnt;
4336 v16u8 out0, out1;
4337 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
4338 v8i16 filt0, filt1;
4339 v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16);
4340 v16i8 mask1;
4341 v8i16 filt_h0, filt_h1, filter_vec, tmp0, tmp1, tmp2, tmp3;
4342 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
4343 v8i16 dst10, dst21, dst22, dst73, dst84, dst95, dst106;
4344 v8i16 dst10_r, dst32_r, dst54_r, dst76_r;
4345 v8i16 dst21_r, dst43_r, dst65_r, dst87_r;
4346 v8i16 dst98_r, dst109_r, offset_vec, const_128, denom_vec;
4347 v4i32 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, weight_vec, rnd_vec;
4348
4349 src -= (src_stride + 1);
4350
4351 filter_vec = LD_SH(filter_x);
4352 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4353
4354 filter_vec = LD_SH(filter_y);
4355 UNPCK_R_SB_SH(filter_vec, filter_vec);
4356
4357 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
4358
4359 mask1 = mask0 + 2;
4360
4361 weight_vec = __msa_fill_w(weight);
4362 rnd_vec = __msa_fill_w(rnd_val);
4363
4364 offset_vec = __msa_fill_h(offset);
4365 denom_vec = __msa_fill_h(rnd_val - 6);
4366 const_128 = __msa_fill_h((128 * weight));
4367 offset_vec += __msa_srar_h(const_128, denom_vec);
4368
4369 LD_SB3(src, src_stride, src0, src1, src2);
4370 src += (3 * src_stride);
4371 XORI_B3_128_SB(src0, src1, src2);
4372
4373 VSHF_B2_SB(src0, src1, src0, src1, mask0, mask1, vec0, vec1);
4374 VSHF_B2_SB(src1, src2, src1, src2, mask0, mask1, vec2, vec3);
4375 dst10 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4376 dst21 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4377 ILVRL_H2_SH(dst21, dst10, dst10_r, dst21_r);
4378 dst22 = (v8i16) __msa_splati_d((v2i64) dst21, 1);
4379
4380 for (loop_cnt = height >> 3; loop_cnt--;) {
4381 LD_SB8(src, src_stride,
4382 src3, src4, src5, src6, src7, src8, src9, src10);
4383 src += (8 * src_stride);
4384 XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
4385
4386 VSHF_B2_SB(src3, src7, src3, src7, mask0, mask1, vec0, vec1);
4387 VSHF_B2_SB(src4, src8, src4, src8, mask0, mask1, vec2, vec3);
4388 VSHF_B2_SB(src5, src9, src5, src9, mask0, mask1, vec4, vec5);
4389 VSHF_B2_SB(src6, src10, src6, src10, mask0, mask1, vec6, vec7);
4390 dst73 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4391 dst84 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4392 dst95 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4393 dst106 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
4394 dst32_r = __msa_ilvr_h(dst73, dst22);
4395 ILVRL_H2_SH(dst84, dst73, dst43_r, dst87_r);
4396 ILVRL_H2_SH(dst95, dst84, dst54_r, dst98_r);
4397 ILVRL_H2_SH(dst106, dst95, dst65_r, dst109_r);
4398 dst22 = (v8i16) __msa_splati_d((v2i64) dst73, 1);
4399 dst76_r = __msa_ilvr_h(dst22, dst106);
4400 dst0 = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
4401 dst1 = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
4402 dst2 = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
4403 dst3 = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
4404 dst4 = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
4405 dst5 = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
4406 dst6 = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
4407 dst7 = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
4408 SRA_4V(dst0, dst1, dst2, dst3, 6);
4409 SRA_4V(dst4, dst5, dst6, dst7, 6);
4410 MUL2(dst0, weight_vec, dst1, weight_vec, dst0, dst1);
4411 MUL2(dst2, weight_vec, dst3, weight_vec, dst2, dst3);
4412 MUL2(dst4, weight_vec, dst5, weight_vec, dst4, dst5);
4413 MUL2(dst6, weight_vec, dst7, weight_vec, dst6, dst7);
4414 SRAR_W4_SW(dst0, dst1, dst2, dst3, rnd_vec);
4415 SRAR_W4_SW(dst4, dst5, dst6, dst7, rnd_vec);
4416 PCKEV_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, tmp0, tmp1,
4417 tmp2, tmp3);
4418 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
4419 ADD2(tmp2, offset_vec, tmp3, offset_vec, tmp2, tmp3);
4420 CLIP_SH4_0_255(tmp0, tmp1, tmp2, tmp3);
4421 PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
4422 ST_W8(out0, out1, 0, 1, 2, 3, 0, 1, 2, 3, dst, dst_stride);
4423 dst += (8 * dst_stride);
4424
4425 dst10_r = dst98_r;
4426 dst21_r = dst109_r;
4427 dst22 = (v8i16) __msa_splati_d((v2i64) dst106, 1);
4428 }
4429 }
4430
hevc_hv_uniwgt_4t_4w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)4431 static void hevc_hv_uniwgt_4t_4w_msa(uint8_t *src,
4432 int32_t src_stride,
4433 uint8_t *dst,
4434 int32_t dst_stride,
4435 const int8_t *filter_x,
4436 const int8_t *filter_y,
4437 int32_t height,
4438 int32_t weight,
4439 int32_t offset,
4440 int32_t rnd_val)
4441 {
4442 if (2 == height) {
4443 hevc_hv_uniwgt_4t_4x2_msa(src, src_stride, dst, dst_stride,
4444 filter_x, filter_y, weight,
4445 offset, rnd_val);
4446 } else if (4 == height) {
4447 hevc_hv_uniwgt_4t_4x4_msa(src, src_stride, dst, dst_stride,
4448 filter_x,filter_y, weight,
4449 offset, rnd_val);
4450 } else if (0 == (height % 8)) {
4451 hevc_hv_uniwgt_4t_4multx8mult_msa(src, src_stride, dst, dst_stride,
4452 filter_x, filter_y, height, weight,
4453 offset, rnd_val);
4454 }
4455 }
4456
hevc_hv_uniwgt_4t_6w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)4457 static void hevc_hv_uniwgt_4t_6w_msa(uint8_t *src,
4458 int32_t src_stride,
4459 uint8_t *dst,
4460 int32_t dst_stride,
4461 const int8_t *filter_x,
4462 const int8_t *filter_y,
4463 int32_t height,
4464 int32_t weight,
4465 int32_t offset,
4466 int32_t rnd_val)
4467 {
4468 v16u8 out0, out1, out2;
4469 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
4470 v8i16 filt0, filt1;
4471 v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
4472 v16i8 mask1;
4473 v8i16 filt_h0, filt_h1, filter_vec;
4474 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
4475 v8i16 dsth0, dsth1, dsth2, dsth3, dsth4, dsth5, dsth6, dsth7, dsth8, dsth9;
4476 v8i16 dsth10, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
4477 v8i16 dst10_r, dst32_r, dst54_r, dst76_r, dst98_r, dst21_r, dst43_r;
4478 v8i16 dst65_r, dst87_r, dst109_r, dst10_l, dst32_l, dst54_l, dst76_l;
4479 v8i16 dst98_l, dst21_l, dst43_l, dst65_l, dst87_l, dst109_l;
4480 v8i16 dst1021_l, dst3243_l, dst5465_l, dst7687_l, dst98109_l;
4481 v8i16 offset_vec, const_128, denom_vec;
4482 v4i32 dst0_r, dst1_r, dst2_r, dst3_r, dst4_r, dst5_r, dst6_r, dst7_r;
4483 v4i32 dst0_l, dst1_l, dst2_l, dst3_l, weight_vec, rnd_vec;
4484
4485 src -= (src_stride + 1);
4486
4487 filter_vec = LD_SH(filter_x);
4488 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4489
4490 filter_vec = LD_SH(filter_y);
4491 UNPCK_R_SB_SH(filter_vec, filter_vec);
4492
4493 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
4494
4495 mask1 = mask0 + 2;
4496
4497 weight_vec = __msa_fill_w(weight);
4498 rnd_vec = __msa_fill_w(rnd_val);
4499
4500 offset_vec = __msa_fill_h(offset);
4501 denom_vec = __msa_fill_h(rnd_val - 6);
4502 const_128 = __msa_fill_h((128 * weight));
4503 offset_vec += __msa_srar_h(const_128, denom_vec);
4504
4505 LD_SB3(src, src_stride, src0, src1, src2);
4506 src += (3 * src_stride);
4507 XORI_B3_128_SB(src0, src1, src2);
4508
4509 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
4510 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
4511 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
4512 dsth0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4513 dsth1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4514 dsth2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4515 ILVRL_H2_SH(dsth1, dsth0, dst10_r, dst10_l);
4516 ILVRL_H2_SH(dsth2, dsth1, dst21_r, dst21_l);
4517
4518 LD_SB8(src, src_stride, src3, src4, src5, src6, src7, src8, src9, src10);
4519 XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
4520 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
4521 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
4522 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
4523 VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
4524 dsth3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4525 dsth4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4526 dsth5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4527 dsth6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
4528 VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec0, vec1);
4529 VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec2, vec3);
4530 VSHF_B2_SB(src9, src9, src9, src9, mask0, mask1, vec4, vec5);
4531 VSHF_B2_SB(src10, src10, src10, src10, mask0, mask1, vec6, vec7);
4532 dsth7 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4533 dsth8 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4534 dsth9 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4535 dsth10 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
4536 ILVRL_H2_SH(dsth3, dsth2, dst32_r, dst32_l);
4537 ILVRL_H2_SH(dsth4, dsth3, dst43_r, dst43_l);
4538 ILVRL_H2_SH(dsth5, dsth4, dst54_r, dst54_l);
4539 ILVRL_H2_SH(dsth6, dsth5, dst65_r, dst65_l);
4540 ILVRL_H2_SH(dsth7, dsth6, dst76_r, dst76_l);
4541 ILVRL_H2_SH(dsth8, dsth7, dst87_r, dst87_l);
4542 ILVRL_H2_SH(dsth9, dsth8, dst98_r, dst98_l);
4543 ILVRL_H2_SH(dsth10, dsth9, dst109_r, dst109_l);
4544 PCKEV_D2_SH(dst21_l, dst10_l, dst43_l, dst32_l, dst1021_l, dst3243_l);
4545 PCKEV_D2_SH(dst65_l, dst54_l, dst87_l, dst76_l, dst5465_l, dst7687_l);
4546 dst98109_l = (v8i16) __msa_pckev_d((v2i64) dst109_l, (v2i64) dst98_l);
4547 dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
4548 dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
4549 dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
4550 dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
4551 dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
4552 dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
4553 dst6_r = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
4554 dst7_r = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
4555 dst0_l = HEVC_FILT_4TAP(dst1021_l, dst3243_l, filt_h0, filt_h1);
4556 dst1_l = HEVC_FILT_4TAP(dst3243_l, dst5465_l, filt_h0, filt_h1);
4557 dst2_l = HEVC_FILT_4TAP(dst5465_l, dst7687_l, filt_h0, filt_h1);
4558 dst3_l = HEVC_FILT_4TAP(dst7687_l, dst98109_l, filt_h0, filt_h1);
4559 SRA_4V(dst0_r, dst1_r, dst2_r, dst3_r, 6);
4560 SRA_4V(dst4_r, dst5_r, dst6_r, dst7_r, 6);
4561 SRA_4V(dst0_l, dst1_l, dst2_l, dst3_l, 6);
4562 MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
4563 MUL2(dst2_r, weight_vec, dst3_r, weight_vec, dst2_r, dst3_r);
4564 MUL2(dst4_r, weight_vec, dst5_r, weight_vec, dst4_r, dst5_r);
4565 MUL2(dst6_r, weight_vec, dst7_r, weight_vec, dst6_r, dst7_r);
4566 MUL2(dst0_l, weight_vec, dst1_l, weight_vec, dst0_l, dst1_l);
4567 MUL2(dst2_l, weight_vec, dst3_l, weight_vec, dst2_l, dst3_l);
4568 SRAR_W4_SW(dst0_r, dst1_r, dst2_r, dst3_r, rnd_vec);
4569 SRAR_W4_SW(dst4_r, dst5_r, dst6_r, dst7_r, rnd_vec);
4570 SRAR_W4_SW(dst0_l, dst1_l, dst2_l, dst3_l, rnd_vec);
4571 PCKEV_H2_SH(dst1_r, dst0_r, dst3_r, dst2_r, tmp0, tmp1);
4572 PCKEV_H2_SH(dst5_r, dst4_r, dst7_r, dst6_r, tmp2, tmp3);
4573 PCKEV_H2_SH(dst1_l, dst0_l, dst3_l, dst2_l, tmp4, tmp5);
4574 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
4575 ADD2(tmp2, offset_vec, tmp3, offset_vec, tmp2, tmp3);
4576 ADD2(tmp4, offset_vec, tmp5, offset_vec, tmp4, tmp5);
4577 CLIP_SH4_0_255(tmp0, tmp1, tmp2, tmp3);
4578 CLIP_SH2_0_255(tmp4, tmp5);
4579 PCKEV_B3_UB(tmp1, tmp0, tmp3, tmp2, tmp5, tmp4, out0, out1, out2);
4580 ST_W8(out0, out1, 0, 1, 2, 3, 0, 1, 2, 3, dst, dst_stride);
4581 ST_H8(out2, 0, 1, 2, 3, 4, 5, 6, 7, dst + 4, dst_stride);
4582 }
4583
hevc_hv_uniwgt_4t_8x2_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t weight,int32_t offset,int32_t rnd_val)4584 static void hevc_hv_uniwgt_4t_8x2_msa(uint8_t *src,
4585 int32_t src_stride,
4586 uint8_t *dst,
4587 int32_t dst_stride,
4588 const int8_t *filter_x,
4589 const int8_t *filter_y,
4590 int32_t weight,
4591 int32_t offset,
4592 int32_t rnd_val)
4593 {
4594 v16u8 out;
4595 v16i8 src0, src1, src2, src3, src4;
4596 v8i16 filt0, filt1;
4597 v8i16 filt_h0, filt_h1, filter_vec;
4598 v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
4599 v16i8 mask1;
4600 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9;
4601 v8i16 dst0, dst1, dst2, dst3, dst4;
4602 v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
4603 v8i16 dst10_r, dst32_r, dst21_r, dst43_r;
4604 v8i16 dst10_l, dst32_l, dst21_l, dst43_l;
4605 v8i16 tmp0, tmp1;
4606 v8i16 offset_vec, const_128, denom_vec;
4607 v4i32 weight_vec, rnd_vec;
4608
4609 src -= (src_stride + 1);
4610
4611 filter_vec = LD_SH(filter_x);
4612 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4613
4614 filter_vec = LD_SH(filter_y);
4615 UNPCK_R_SB_SH(filter_vec, filter_vec);
4616
4617 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
4618
4619 mask1 = mask0 + 2;
4620
4621 weight_vec = __msa_fill_w(weight);
4622 rnd_vec = __msa_fill_w(rnd_val);
4623
4624 offset_vec = __msa_fill_h(offset);
4625 denom_vec = __msa_fill_h(rnd_val - 6);
4626 const_128 = __msa_fill_h((128 * weight));
4627 offset_vec += __msa_srar_h(const_128, denom_vec);
4628
4629 LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
4630 XORI_B5_128_SB(src0, src1, src2, src3, src4);
4631 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
4632 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
4633 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
4634 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
4635 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec8, vec9);
4636 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4637 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4638 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4639 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
4640 dst4 = HEVC_FILT_4TAP_SH(vec8, vec9, filt0, filt1);
4641 ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
4642 ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
4643 ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
4644 ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
4645 dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
4646 dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
4647 dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
4648 dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
4649 SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
4650 MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
4651 MUL2(dst0_l, weight_vec, dst1_l, weight_vec, dst0_l, dst1_l);
4652 SRAR_W4_SW(dst0_r, dst0_l, dst1_r, dst1_l, rnd_vec);
4653 PCKEV_H2_SH(dst0_l, dst0_r, dst1_l, dst1_r, tmp0, tmp1);
4654 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
4655 CLIP_SH2_0_255(tmp0, tmp1);
4656 out = (v16u8) __msa_pckev_b((v16i8) tmp1, (v16i8) tmp0);
4657 ST_D2(out, 0, 1, dst, dst_stride);
4658 }
4659
hevc_hv_uniwgt_4t_8multx4_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t width8mult,int32_t weight,int32_t offset,int32_t rnd_val)4660 static void hevc_hv_uniwgt_4t_8multx4_msa(uint8_t *src,
4661 int32_t src_stride,
4662 uint8_t *dst,
4663 int32_t dst_stride,
4664 const int8_t *filter_x,
4665 const int8_t *filter_y,
4666 int32_t width8mult,
4667 int32_t weight,
4668 int32_t offset,
4669 int32_t rnd_val)
4670 {
4671 uint32_t cnt;
4672 v16u8 out0, out1;
4673 v16i8 src0, src1, src2, src3, src4, src5, src6, mask0, mask1;
4674 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
4675 v8i16 filt0, filt1, filt_h0, filt_h1, filter_vec;
4676 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, tmp0, tmp1, tmp2, tmp3;
4677 v8i16 dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
4678 v8i16 dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
4679 v8i16 offset_vec, const_128, denom_vec;
4680 v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
4681 v4i32 weight_vec, rnd_vec;
4682
4683 src -= (src_stride + 1);
4684
4685 filter_vec = LD_SH(filter_x);
4686 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4687
4688 filter_vec = LD_SH(filter_y);
4689 UNPCK_R_SB_SH(filter_vec, filter_vec);
4690
4691 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
4692
4693 mask0 = LD_SB(ff_hevc_mask_arr);
4694 mask1 = mask0 + 2;
4695
4696 weight_vec = __msa_fill_w(weight);
4697 rnd_vec = __msa_fill_w(rnd_val);
4698
4699 offset_vec = __msa_fill_h(offset);
4700 denom_vec = __msa_fill_h(rnd_val - 6);
4701 const_128 = __msa_fill_h((128 * weight));
4702 offset_vec += __msa_srar_h(const_128, denom_vec);
4703
4704 for (cnt = width8mult; cnt--;) {
4705 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
4706 src += 8;
4707 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
4708 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
4709 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
4710 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
4711 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4712 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4713 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4714 ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
4715 ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
4716 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
4717 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
4718 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
4719 VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
4720 dst3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4721 dst4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4722 dst5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4723 dst6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
4724 ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
4725 ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
4726 ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
4727 ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
4728 dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
4729 dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
4730 dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
4731 dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
4732 dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
4733 dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
4734 dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
4735 dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
4736 SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
4737 SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
4738 MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
4739 MUL2(dst2_r, weight_vec, dst3_r, weight_vec, dst2_r, dst3_r);
4740 MUL2(dst0_l, weight_vec, dst1_l, weight_vec, dst0_l, dst1_l);
4741 MUL2(dst2_l, weight_vec, dst3_l, weight_vec, dst2_l, dst3_l);
4742 SRAR_W4_SW(dst0_r, dst0_l, dst1_r, dst1_l, rnd_vec);
4743 SRAR_W4_SW(dst2_r, dst2_l, dst3_r, dst3_l, rnd_vec);
4744 PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l,
4745 dst3_r, tmp0, tmp1, tmp2, tmp3);
4746 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
4747 ADD2(tmp2, offset_vec, tmp3, offset_vec, tmp2, tmp3);
4748 CLIP_SH4_0_255(tmp0, tmp1, tmp2, tmp3);
4749 PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
4750 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
4751 dst += 8;
4752 }
4753 }
4754
hevc_hv_uniwgt_4t_8x6_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t weight,int32_t offset,int32_t rnd_val)4755 static void hevc_hv_uniwgt_4t_8x6_msa(uint8_t *src,
4756 int32_t src_stride,
4757 uint8_t *dst,
4758 int32_t dst_stride,
4759 const int8_t *filter_x,
4760 const int8_t *filter_y,
4761 int32_t weight,
4762 int32_t offset,
4763 int32_t rnd_val)
4764 {
4765 v16u8 out0, out1, out2;
4766 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
4767 v8i16 filt0, filt1;
4768 v8i16 filt_h0, filt_h1, filter_vec;
4769 v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
4770 v16i8 mask1;
4771 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, vec9;
4772 v16i8 vec10, vec11, vec12, vec13, vec14, vec15, vec16, vec17;
4773 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
4774 v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
4775 v4i32 dst4_r, dst4_l, dst5_r, dst5_l, weight_vec, rnd_vec;
4776 v8i16 dst10_r, dst32_r, dst10_l, dst32_l;
4777 v8i16 dst21_r, dst43_r, dst21_l, dst43_l;
4778 v8i16 dst54_r, dst54_l, dst65_r, dst65_l;
4779 v8i16 dst76_r, dst76_l, dst87_r, dst87_l;
4780 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
4781 v8i16 offset_vec, const_128, denom_vec;
4782
4783 src -= (src_stride + 1);
4784
4785 filter_vec = LD_SH(filter_x);
4786 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4787
4788 filter_vec = LD_SH(filter_y);
4789 UNPCK_R_SB_SH(filter_vec, filter_vec);
4790
4791 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
4792
4793 mask1 = mask0 + 2;
4794
4795 weight_vec = __msa_fill_w(weight);
4796 rnd_vec = __msa_fill_w(rnd_val);
4797
4798 offset_vec = __msa_fill_h(offset);
4799 denom_vec = __msa_fill_h(rnd_val - 6);
4800 const_128 = __msa_fill_h((128 * weight));
4801 offset_vec += __msa_srar_h(const_128, denom_vec);
4802
4803 LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
4804 src += (5 * src_stride);
4805 LD_SB4(src, src_stride, src5, src6, src7, src8);
4806 XORI_B5_128_SB(src0, src1, src2, src3, src4);
4807 XORI_B4_128_SB(src5, src6, src7, src8);
4808 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
4809 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
4810 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
4811 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec6, vec7);
4812 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec8, vec9);
4813 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec10, vec11);
4814 VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec12, vec13);
4815 VSHF_B2_SB(src7, src7, src7, src7, mask0, mask1, vec14, vec15);
4816 VSHF_B2_SB(src8, src8, src8, src8, mask0, mask1, vec16, vec17);
4817 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4818 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4819 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4820 dst3 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
4821 dst4 = HEVC_FILT_4TAP_SH(vec8, vec9, filt0, filt1);
4822 dst5 = HEVC_FILT_4TAP_SH(vec10, vec11, filt0, filt1);
4823 dst6 = HEVC_FILT_4TAP_SH(vec12, vec13, filt0, filt1);
4824 dst7 = HEVC_FILT_4TAP_SH(vec14, vec15, filt0, filt1);
4825 dst8 = HEVC_FILT_4TAP_SH(vec16, vec17, filt0, filt1);
4826 ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
4827 ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
4828 ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
4829 ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
4830 ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
4831 ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
4832 ILVRL_H2_SH(dst7, dst6, dst76_r, dst76_l);
4833 ILVRL_H2_SH(dst8, dst7, dst87_r, dst87_l);
4834 dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
4835 dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
4836 dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
4837 dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
4838 dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
4839 dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
4840 dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
4841 dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
4842 dst4_r = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
4843 dst4_l = HEVC_FILT_4TAP(dst54_l, dst76_l, filt_h0, filt_h1);
4844 dst5_r = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
4845 dst5_l = HEVC_FILT_4TAP(dst65_l, dst87_l, filt_h0, filt_h1);
4846 SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
4847 SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
4848 SRA_4V(dst4_r, dst4_l, dst5_r, dst5_l, 6);
4849 MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
4850 MUL2(dst2_r, weight_vec, dst3_r, weight_vec, dst2_r, dst3_r);
4851 MUL2(dst4_r, weight_vec, dst5_r, weight_vec, dst4_r, dst5_r);
4852 MUL2(dst0_l, weight_vec, dst1_l, weight_vec, dst0_l, dst1_l);
4853 MUL2(dst2_l, weight_vec, dst3_l, weight_vec, dst2_l, dst3_l);
4854 MUL2(dst4_l, weight_vec, dst5_l, weight_vec, dst4_l, dst5_l);
4855 SRAR_W4_SW(dst0_r, dst0_l, dst1_r, dst1_l, rnd_vec);
4856 SRAR_W4_SW(dst2_r, dst2_l, dst3_r, dst3_l, rnd_vec);
4857 SRAR_W4_SW(dst4_r, dst4_l, dst5_r, dst5_l, rnd_vec);
4858 PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l, dst3_r,
4859 tmp0, tmp1, tmp2, tmp3);
4860 PCKEV_H2_SH(dst4_l, dst4_r, dst5_l, dst5_r, tmp4, tmp5);
4861 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
4862 ADD2(tmp2, offset_vec, tmp3, offset_vec, tmp2, tmp3);
4863 ADD2(tmp4, offset_vec, tmp5, offset_vec, tmp4, tmp5);
4864 CLIP_SH4_0_255(tmp0, tmp1, tmp2, tmp3);
4865 CLIP_SH2_0_255(tmp4, tmp5);
4866 PCKEV_B3_UB(tmp1, tmp0, tmp3, tmp2, tmp5, tmp4, out0, out1, out2);
4867 ST_D4(out0, out1, 0, 1, 0, 1, dst, dst_stride);
4868 ST_D2(out2, 0, 1, dst + 4 * dst_stride, dst_stride);
4869 }
4870
hevc_hv_uniwgt_4t_8multx4mult_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val,int32_t width8mult)4871 static void hevc_hv_uniwgt_4t_8multx4mult_msa(uint8_t *src,
4872 int32_t src_stride,
4873 uint8_t *dst,
4874 int32_t dst_stride,
4875 const int8_t *filter_x,
4876 const int8_t *filter_y,
4877 int32_t height,
4878 int32_t weight,
4879 int32_t offset,
4880 int32_t rnd_val,
4881 int32_t width8mult)
4882 {
4883 uint32_t loop_cnt, cnt;
4884 uint8_t *src_tmp;
4885 uint8_t *dst_tmp;
4886 v16u8 out0, out1;
4887 v16i8 src0, src1, src2, src3, src4, src5, src6;
4888 v8i16 filt0, filt1;
4889 v8i16 filt_h0, filt_h1, filter_vec;
4890 v16i8 mask0 = LD_SB(ff_hevc_mask_arr);
4891 v16i8 mask1;
4892 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
4893 v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, tmp0, tmp1, tmp2, tmp3;
4894 v8i16 dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
4895 v8i16 dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
4896 v4i32 dst0_r, dst0_l, dst1_r, dst1_l;
4897 v8i16 offset_vec, const_128, denom_vec;
4898 v4i32 dst2_r, dst2_l, dst3_r, dst3_l;
4899 v4i32 weight_vec, rnd_vec;
4900
4901 src -= (src_stride + 1);
4902
4903 filter_vec = LD_SH(filter_x);
4904 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
4905
4906 filter_vec = LD_SH(filter_y);
4907 UNPCK_R_SB_SH(filter_vec, filter_vec);
4908
4909 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
4910
4911 mask1 = mask0 + 2;
4912
4913 weight_vec = __msa_fill_w(weight);
4914 rnd_vec = __msa_fill_w(rnd_val);
4915
4916 offset_vec = __msa_fill_h(offset);
4917 denom_vec = __msa_fill_h(rnd_val - 6);
4918 const_128 = __msa_fill_h((128 * weight));
4919 offset_vec += __msa_srar_h(const_128, denom_vec);
4920
4921 for (cnt = width8mult; cnt--;) {
4922 src_tmp = src;
4923 dst_tmp = dst;
4924
4925 LD_SB3(src_tmp, src_stride, src0, src1, src2);
4926 src_tmp += (3 * src_stride);
4927 XORI_B3_128_SB(src0, src1, src2);
4928
4929 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
4930 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
4931 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
4932 dst0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4933 dst1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4934 dst2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4935
4936 ILVRL_H2_SH(dst1, dst0, dst10_r, dst10_l);
4937 ILVRL_H2_SH(dst2, dst1, dst21_r, dst21_l);
4938
4939 for (loop_cnt = height >> 2; loop_cnt--;) {
4940 LD_SB4(src_tmp, src_stride, src3, src4, src5, src6);
4941 src_tmp += (4 * src_stride);
4942 XORI_B4_128_SB(src3, src4, src5, src6);
4943
4944 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
4945 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
4946 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
4947 VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
4948 dst3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
4949 dst4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
4950 dst5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
4951 dst6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
4952 ILVRL_H2_SH(dst3, dst2, dst32_r, dst32_l);
4953 ILVRL_H2_SH(dst4, dst3, dst43_r, dst43_l);
4954 ILVRL_H2_SH(dst5, dst4, dst54_r, dst54_l);
4955 ILVRL_H2_SH(dst6, dst5, dst65_r, dst65_l);
4956 dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
4957 dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
4958 dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
4959 dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
4960 dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
4961 dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
4962 dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
4963 dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
4964 SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
4965 SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
4966 MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
4967 MUL2(dst2_r, weight_vec, dst3_r, weight_vec, dst2_r, dst3_r);
4968 MUL2(dst0_l, weight_vec, dst1_l, weight_vec, dst0_l, dst1_l);
4969 MUL2(dst2_l, weight_vec, dst3_l, weight_vec, dst2_l, dst3_l);
4970 SRAR_W4_SW(dst0_r, dst0_l, dst1_r, dst1_l, rnd_vec);
4971 SRAR_W4_SW(dst2_r, dst2_l, dst3_r, dst3_l, rnd_vec);
4972 PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l,
4973 dst3_r, tmp0, tmp1, tmp2, tmp3);
4974 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
4975 ADD2(tmp2, offset_vec, tmp3, offset_vec, tmp2, tmp3);
4976 CLIP_SH4_0_255(tmp0, tmp1, tmp2, tmp3);
4977 PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
4978 ST_D4(out0, out1, 0, 1, 0, 1, dst_tmp, dst_stride);
4979 dst_tmp += (4 * dst_stride);
4980
4981 dst10_r = dst54_r;
4982 dst10_l = dst54_l;
4983 dst21_r = dst65_r;
4984 dst21_l = dst65_l;
4985 dst2 = dst6;
4986 }
4987
4988 src += 8;
4989 dst += 8;
4990 }
4991 }
4992
hevc_hv_uniwgt_4t_8w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)4993 static void hevc_hv_uniwgt_4t_8w_msa(uint8_t *src,
4994 int32_t src_stride,
4995 uint8_t *dst,
4996 int32_t dst_stride,
4997 const int8_t *filter_x,
4998 const int8_t *filter_y,
4999 int32_t height,
5000 int32_t weight,
5001 int32_t offset,
5002 int32_t rnd_val)
5003 {
5004
5005 if (2 == height) {
5006 hevc_hv_uniwgt_4t_8x2_msa(src, src_stride, dst, dst_stride,
5007 filter_x, filter_y, weight,
5008 offset, rnd_val);
5009 } else if (4 == height) {
5010 hevc_hv_uniwgt_4t_8multx4_msa(src, src_stride, dst, dst_stride,
5011 filter_x, filter_y, 1, weight,
5012 offset, rnd_val);
5013 } else if (6 == height) {
5014 hevc_hv_uniwgt_4t_8x6_msa(src, src_stride, dst, dst_stride,
5015 filter_x, filter_y, weight,
5016 offset, rnd_val);
5017 } else if (0 == (height % 4)) {
5018 hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
5019 filter_x, filter_y, height, weight,
5020 offset, rnd_val, 1);
5021 }
5022 }
5023
hevc_hv_uniwgt_4t_12w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)5024 static void hevc_hv_uniwgt_4t_12w_msa(uint8_t *src,
5025 int32_t src_stride,
5026 uint8_t *dst,
5027 int32_t dst_stride,
5028 const int8_t *filter_x,
5029 const int8_t *filter_y,
5030 int32_t height,
5031 int32_t weight,
5032 int32_t offset,
5033 int32_t rnd_val)
5034 {
5035 uint32_t loop_cnt;
5036 uint8_t *src_tmp, *dst_tmp;
5037 v16u8 out0, out1;
5038 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
5039 v16i8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
5040 v16i8 mask0, mask1, mask2, mask3;
5041 v8i16 filt0, filt1, filt_h0, filt_h1, filter_vec, tmp0, tmp1, tmp2, tmp3;
5042 v8i16 dsth0, dsth1, dsth2, dsth3, dsth4, dsth5, dsth6;
5043 v8i16 dst10, dst21, dst22, dst73, dst84, dst95, dst106;
5044 v8i16 dst76_r, dst98_r, dst87_r, dst109_r;
5045 v8i16 dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
5046 v8i16 dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
5047 v8i16 offset_vec, const_128, denom_vec;
5048 v4i32 dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
5049 v4i32 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, weight_vec, rnd_vec;
5050
5051 src -= (src_stride + 1);
5052
5053 filter_vec = LD_SH(filter_x);
5054 SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1);
5055
5056 filter_vec = LD_SH(filter_y);
5057 UNPCK_R_SB_SH(filter_vec, filter_vec);
5058
5059 SPLATI_W2_SH(filter_vec, 0, filt_h0, filt_h1);
5060
5061 mask0 = LD_SB(ff_hevc_mask_arr);
5062 mask1 = mask0 + 2;
5063
5064 weight_vec = __msa_fill_w(weight);
5065 rnd_vec = __msa_fill_w(rnd_val);
5066
5067 offset_vec = __msa_fill_h(offset);
5068 denom_vec = __msa_fill_h(rnd_val - 6);
5069 const_128 = __msa_fill_h((128 * weight));
5070 offset_vec += __msa_srar_h(const_128, denom_vec);
5071
5072 src_tmp = src;
5073 dst_tmp = dst;
5074
5075 LD_SB3(src_tmp, src_stride, src0, src1, src2);
5076 src_tmp += (3 * src_stride);
5077 XORI_B3_128_SB(src0, src1, src2);
5078 VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1);
5079 VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec2, vec3);
5080 VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec4, vec5);
5081 dsth0 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
5082 dsth1 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
5083 dsth2 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
5084 ILVRL_H2_SH(dsth1, dsth0, dst10_r, dst10_l);
5085 ILVRL_H2_SH(dsth2, dsth1, dst21_r, dst21_l);
5086
5087 for (loop_cnt = 4; loop_cnt--;) {
5088 LD_SB4(src_tmp, src_stride, src3, src4, src5, src6);
5089 src_tmp += (4 * src_stride);
5090 XORI_B4_128_SB(src3, src4, src5, src6);
5091 VSHF_B2_SB(src3, src3, src3, src3, mask0, mask1, vec0, vec1);
5092 VSHF_B2_SB(src4, src4, src4, src4, mask0, mask1, vec2, vec3);
5093 VSHF_B2_SB(src5, src5, src5, src5, mask0, mask1, vec4, vec5);
5094 VSHF_B2_SB(src6, src6, src6, src6, mask0, mask1, vec6, vec7);
5095 dsth3 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
5096 dsth4 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
5097 dsth5 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
5098 dsth6 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
5099 ILVRL_H2_SH(dsth3, dsth2, dst32_r, dst32_l);
5100 ILVRL_H2_SH(dsth4, dsth3, dst43_r, dst43_l);
5101 ILVRL_H2_SH(dsth5, dsth4, dst54_r, dst54_l);
5102 ILVRL_H2_SH(dsth6, dsth5, dst65_r, dst65_l);
5103 dst0_r = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
5104 dst0_l = HEVC_FILT_4TAP(dst10_l, dst32_l, filt_h0, filt_h1);
5105 dst1_r = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
5106 dst1_l = HEVC_FILT_4TAP(dst21_l, dst43_l, filt_h0, filt_h1);
5107 dst2_r = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
5108 dst2_l = HEVC_FILT_4TAP(dst32_l, dst54_l, filt_h0, filt_h1);
5109 dst3_r = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
5110 dst3_l = HEVC_FILT_4TAP(dst43_l, dst65_l, filt_h0, filt_h1);
5111 SRA_4V(dst0_r, dst0_l, dst1_r, dst1_l, 6);
5112 SRA_4V(dst2_r, dst2_l, dst3_r, dst3_l, 6);
5113 MUL2(dst0_r, weight_vec, dst1_r, weight_vec, dst0_r, dst1_r);
5114 MUL2(dst2_r, weight_vec, dst3_r, weight_vec, dst2_r, dst3_r);
5115 MUL2(dst0_l, weight_vec, dst1_l, weight_vec, dst0_l, dst1_l);
5116 MUL2(dst2_l, weight_vec, dst3_l, weight_vec, dst2_l, dst3_l);
5117 SRAR_W4_SW(dst0_r, dst0_l, dst1_r, dst1_l, rnd_vec);
5118 SRAR_W4_SW(dst2_r, dst2_l, dst3_r, dst3_l, rnd_vec);
5119 PCKEV_H4_SH(dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r, dst3_l,
5120 dst3_r, tmp0, tmp1, tmp2, tmp3);
5121 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
5122 ADD2(tmp2, offset_vec, tmp3, offset_vec, tmp2, tmp3);
5123 CLIP_SH4_0_255(tmp0, tmp1, tmp2, tmp3);
5124 PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
5125 ST_D4(out0, out1, 0, 1, 0, 1, dst_tmp, dst_stride);
5126 dst_tmp += (4 * dst_stride);
5127
5128 dst10_r = dst54_r;
5129 dst10_l = dst54_l;
5130 dst21_r = dst65_r;
5131 dst21_l = dst65_l;
5132 dsth2 = dsth6;
5133 }
5134
5135 src += 8;
5136 dst += 8;
5137
5138 mask2 = LD_SB(ff_hevc_mask_arr + 16);
5139 mask3 = mask2 + 2;
5140
5141 LD_SB3(src, src_stride, src0, src1, src2);
5142 src += (3 * src_stride);
5143 XORI_B3_128_SB(src0, src1, src2);
5144 VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1);
5145 VSHF_B2_SB(src1, src2, src1, src2, mask2, mask3, vec2, vec3);
5146 dst10 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
5147 dst21 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
5148 ILVRL_H2_SH(dst21, dst10, dst10_r, dst21_r);
5149 dst22 = (v8i16) __msa_splati_d((v2i64) dst21, 1);
5150
5151 for (loop_cnt = 2; loop_cnt--;) {
5152 LD_SB8(src, src_stride, src3, src4, src5, src6, src7, src8, src9,
5153 src10);
5154 src += (8 * src_stride);
5155 XORI_B8_128_SB(src3, src4, src5, src6, src7, src8, src9, src10);
5156 VSHF_B2_SB(src3, src7, src3, src7, mask2, mask3, vec0, vec1);
5157 VSHF_B2_SB(src4, src8, src4, src8, mask2, mask3, vec2, vec3);
5158 VSHF_B2_SB(src5, src9, src5, src9, mask2, mask3, vec4, vec5);
5159 VSHF_B2_SB(src6, src10, src6, src10, mask2, mask3, vec6, vec7);
5160 dst73 = HEVC_FILT_4TAP_SH(vec0, vec1, filt0, filt1);
5161 dst84 = HEVC_FILT_4TAP_SH(vec2, vec3, filt0, filt1);
5162 dst95 = HEVC_FILT_4TAP_SH(vec4, vec5, filt0, filt1);
5163 dst106 = HEVC_FILT_4TAP_SH(vec6, vec7, filt0, filt1);
5164 dst32_r = __msa_ilvr_h(dst73, dst22);
5165 ILVRL_H2_SH(dst84, dst73, dst43_r, dst87_r);
5166 ILVRL_H2_SH(dst95, dst84, dst54_r, dst98_r);
5167 ILVRL_H2_SH(dst106, dst95, dst65_r, dst109_r);
5168 dst22 = (v8i16) __msa_splati_d((v2i64) dst73, 1);
5169 dst76_r = __msa_ilvr_h(dst22, dst106);
5170 dst0 = HEVC_FILT_4TAP(dst10_r, dst32_r, filt_h0, filt_h1);
5171 dst1 = HEVC_FILT_4TAP(dst21_r, dst43_r, filt_h0, filt_h1);
5172 dst2 = HEVC_FILT_4TAP(dst32_r, dst54_r, filt_h0, filt_h1);
5173 dst3 = HEVC_FILT_4TAP(dst43_r, dst65_r, filt_h0, filt_h1);
5174 dst4 = HEVC_FILT_4TAP(dst54_r, dst76_r, filt_h0, filt_h1);
5175 dst5 = HEVC_FILT_4TAP(dst65_r, dst87_r, filt_h0, filt_h1);
5176 dst6 = HEVC_FILT_4TAP(dst76_r, dst98_r, filt_h0, filt_h1);
5177 dst7 = HEVC_FILT_4TAP(dst87_r, dst109_r, filt_h0, filt_h1);
5178 SRA_4V(dst0, dst1, dst2, dst3, 6);
5179 SRA_4V(dst4, dst5, dst6, dst7, 6);
5180 MUL2(dst0, weight_vec, dst1, weight_vec, dst0, dst1);
5181 MUL2(dst2, weight_vec, dst3, weight_vec, dst2, dst3);
5182 MUL2(dst4, weight_vec, dst5, weight_vec, dst4, dst5);
5183 MUL2(dst6, weight_vec, dst7, weight_vec, dst6, dst7);
5184 SRAR_W4_SW(dst0, dst1, dst2, dst3, rnd_vec);
5185 SRAR_W4_SW(dst4, dst5, dst6, dst7, rnd_vec);
5186 PCKEV_H4_SH(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, tmp0, tmp1,
5187 tmp2, tmp3);
5188 ADD2(tmp0, offset_vec, tmp1, offset_vec, tmp0, tmp1);
5189 ADD2(tmp2, offset_vec, tmp3, offset_vec, tmp2, tmp3);
5190 CLIP_SH4_0_255(tmp0, tmp1, tmp2, tmp3);
5191 PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, out0, out1);
5192 ST_W8(out0, out1, 0, 1, 2, 3, 0, 1, 2, 3, dst, dst_stride);
5193 dst += (8 * dst_stride);
5194
5195 dst10_r = dst98_r;
5196 dst21_r = dst109_r;
5197 dst22 = (v8i16) __msa_splati_d((v2i64) dst106, 1);
5198 }
5199 }
5200
hevc_hv_uniwgt_4t_16w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)5201 static void hevc_hv_uniwgt_4t_16w_msa(uint8_t *src,
5202 int32_t src_stride,
5203 uint8_t *dst,
5204 int32_t dst_stride,
5205 const int8_t *filter_x,
5206 const int8_t *filter_y,
5207 int32_t height,
5208 int32_t weight,
5209 int32_t offset,
5210 int32_t rnd_val)
5211 {
5212 if (4 == height) {
5213 hevc_hv_uniwgt_4t_8multx4_msa(src, src_stride, dst, dst_stride,
5214 filter_x, filter_y, 2, weight, offset,
5215 rnd_val);
5216 } else {
5217 hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
5218 filter_x, filter_y, height, weight,
5219 offset, rnd_val, 2);
5220 }
5221 }
5222
hevc_hv_uniwgt_4t_24w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)5223 static void hevc_hv_uniwgt_4t_24w_msa(uint8_t *src,
5224 int32_t src_stride,
5225 uint8_t *dst,
5226 int32_t dst_stride,
5227 const int8_t *filter_x,
5228 const int8_t *filter_y,
5229 int32_t height,
5230 int32_t weight,
5231 int32_t offset,
5232 int32_t rnd_val)
5233 {
5234 hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
5235 filter_x, filter_y, height, weight,
5236 offset, rnd_val, 3);
5237 }
5238
hevc_hv_uniwgt_4t_32w_msa(uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int8_t * filter_x,const int8_t * filter_y,int32_t height,int32_t weight,int32_t offset,int32_t rnd_val)5239 static void hevc_hv_uniwgt_4t_32w_msa(uint8_t *src,
5240 int32_t src_stride,
5241 uint8_t *dst,
5242 int32_t dst_stride,
5243 const int8_t *filter_x,
5244 const int8_t *filter_y,
5245 int32_t height,
5246 int32_t weight,
5247 int32_t offset,
5248 int32_t rnd_val)
5249 {
5250 hevc_hv_uniwgt_4t_8multx4mult_msa(src, src_stride, dst, dst_stride,
5251 filter_x, filter_y, height, weight,
5252 offset, rnd_val, 4);
5253 }
5254
5255 #define UNIWGT_MC_COPY(WIDTH) \
5256 void ff_hevc_put_hevc_uni_w_pel_pixels##WIDTH##_8_msa(uint8_t *dst, \
5257 ptrdiff_t dst_stride, \
5258 uint8_t *src, \
5259 ptrdiff_t src_stride, \
5260 int height, \
5261 int denom, \
5262 int weight, \
5263 int offset, \
5264 intptr_t mx, \
5265 intptr_t my, \
5266 int width) \
5267 { \
5268 int shift = denom + 14 - 8; \
5269 hevc_uniwgt_copy_##WIDTH##w_msa(src, src_stride, dst, dst_stride, \
5270 height, weight, offset, shift); \
5271 }
5272
5273 UNIWGT_MC_COPY(4);
5274 UNIWGT_MC_COPY(6);
5275 UNIWGT_MC_COPY(8);
5276 UNIWGT_MC_COPY(12);
5277 UNIWGT_MC_COPY(16);
5278 UNIWGT_MC_COPY(24);
5279 UNIWGT_MC_COPY(32);
5280 UNIWGT_MC_COPY(48);
5281 UNIWGT_MC_COPY(64);
5282
5283 #undef UNIWGT_MC_COPY
5284
5285 #define UNI_W_MC(PEL, DIR, WIDTH, TAP, DIR1, FILT_DIR) \
5286 void ff_hevc_put_hevc_uni_w_##PEL##_##DIR##WIDTH##_8_msa(uint8_t *dst, \
5287 ptrdiff_t \
5288 dst_stride, \
5289 uint8_t *src, \
5290 ptrdiff_t \
5291 src_stride, \
5292 int height, \
5293 int denom, \
5294 int weight, \
5295 int offset, \
5296 intptr_t mx, \
5297 intptr_t my, \
5298 int width) \
5299 { \
5300 const int8_t *filter = ff_hevc_##PEL##_filters[FILT_DIR - 1]; \
5301 int shift = denom + 14 - 8; \
5302 \
5303 hevc_##DIR1##_uniwgt_##TAP##t_##WIDTH##w_msa(src, src_stride, dst, \
5304 dst_stride, filter, height, \
5305 weight, offset, shift); \
5306 }
5307
5308 UNI_W_MC(qpel, h, 4, 8, hz, mx);
5309 UNI_W_MC(qpel, h, 8, 8, hz, mx);
5310 UNI_W_MC(qpel, h, 12, 8, hz, mx);
5311 UNI_W_MC(qpel, h, 16, 8, hz, mx);
5312 UNI_W_MC(qpel, h, 24, 8, hz, mx);
5313 UNI_W_MC(qpel, h, 32, 8, hz, mx);
5314 UNI_W_MC(qpel, h, 48, 8, hz, mx);
5315 UNI_W_MC(qpel, h, 64, 8, hz, mx);
5316
5317 UNI_W_MC(qpel, v, 4, 8, vt, my);
5318 UNI_W_MC(qpel, v, 8, 8, vt, my);
5319 UNI_W_MC(qpel, v, 12, 8, vt, my);
5320 UNI_W_MC(qpel, v, 16, 8, vt, my);
5321 UNI_W_MC(qpel, v, 24, 8, vt, my);
5322 UNI_W_MC(qpel, v, 32, 8, vt, my);
5323 UNI_W_MC(qpel, v, 48, 8, vt, my);
5324 UNI_W_MC(qpel, v, 64, 8, vt, my);
5325
5326 UNI_W_MC(epel, h, 4, 4, hz, mx);
5327 UNI_W_MC(epel, h, 6, 4, hz, mx);
5328 UNI_W_MC(epel, h, 8, 4, hz, mx);
5329 UNI_W_MC(epel, h, 12, 4, hz, mx);
5330 UNI_W_MC(epel, h, 16, 4, hz, mx);
5331 UNI_W_MC(epel, h, 24, 4, hz, mx);
5332 UNI_W_MC(epel, h, 32, 4, hz, mx);
5333
5334 UNI_W_MC(epel, v, 4, 4, vt, my);
5335 UNI_W_MC(epel, v, 6, 4, vt, my);
5336 UNI_W_MC(epel, v, 8, 4, vt, my);
5337 UNI_W_MC(epel, v, 12, 4, vt, my);
5338 UNI_W_MC(epel, v, 16, 4, vt, my);
5339 UNI_W_MC(epel, v, 24, 4, vt, my);
5340 UNI_W_MC(epel, v, 32, 4, vt, my);
5341
5342 #undef UNI_W_MC
5343
5344 #define UNI_W_MC_HV(PEL, WIDTH, TAP) \
5345 void ff_hevc_put_hevc_uni_w_##PEL##_hv##WIDTH##_8_msa(uint8_t *dst, \
5346 ptrdiff_t dst_stride, \
5347 uint8_t *src, \
5348 ptrdiff_t src_stride, \
5349 int height, \
5350 int denom, \
5351 int weight, \
5352 int offset, \
5353 intptr_t mx, \
5354 intptr_t my, \
5355 int width) \
5356 { \
5357 const int8_t *filter_x = ff_hevc_##PEL##_filters[mx - 1]; \
5358 const int8_t *filter_y = ff_hevc_##PEL##_filters[my - 1]; \
5359 int shift = denom + 14 - 8; \
5360 \
5361 hevc_hv_uniwgt_##TAP##t_##WIDTH##w_msa(src, src_stride, dst, dst_stride, \
5362 filter_x, filter_y, height, \
5363 weight, offset, shift); \
5364 }
5365
5366 UNI_W_MC_HV(qpel, 4, 8);
5367 UNI_W_MC_HV(qpel, 8, 8);
5368 UNI_W_MC_HV(qpel, 12, 8);
5369 UNI_W_MC_HV(qpel, 16, 8);
5370 UNI_W_MC_HV(qpel, 24, 8);
5371 UNI_W_MC_HV(qpel, 32, 8);
5372 UNI_W_MC_HV(qpel, 48, 8);
5373 UNI_W_MC_HV(qpel, 64, 8);
5374
5375 UNI_W_MC_HV(epel, 4, 4);
5376 UNI_W_MC_HV(epel, 6, 4);
5377 UNI_W_MC_HV(epel, 8, 4);
5378 UNI_W_MC_HV(epel, 12, 4);
5379 UNI_W_MC_HV(epel, 16, 4);
5380 UNI_W_MC_HV(epel, 24, 4);
5381 UNI_W_MC_HV(epel, 32, 4);
5382
5383 #undef UNI_W_MC_HV
5384