1 /*
2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <assert.h>
12 #include "./vpx_dsp_rtcd.h"
13 #include "vpx_dsp/mips/vpx_convolve_msa.h"
14
common_vt_8t_and_aver_dst_4w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)15 static void common_vt_8t_and_aver_dst_4w_msa(const uint8_t *src,
16 int32_t src_stride,
17 uint8_t *dst,
18 int32_t dst_stride,
19 int8_t *filter,
20 int32_t height) {
21 uint32_t loop_cnt;
22 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
23 v16u8 dst0, dst1, dst2, dst3, out;
24 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
25 v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776;
26 v16i8 src10998, filt0, filt1, filt2, filt3;
27 v8i16 filt, out10, out32;
28
29 src -= (3 * src_stride);
30
31 filt = LD_SH(filter);
32 SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
33
34 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
35 src += (7 * src_stride);
36
37 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
38 src54_r, src21_r);
39 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
40 ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src2110,
41 src4332, src6554);
42 XORI_B3_128_SB(src2110, src4332, src6554);
43
44 for (loop_cnt = (height >> 2); loop_cnt--;) {
45 LD_SB4(src, src_stride, src7, src8, src9, src10);
46 src += (4 * src_stride);
47
48 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
49 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
50 src87_r, src98_r, src109_r);
51 ILVR_D2_SB(src87_r, src76_r, src109_r, src98_r, src8776, src10998);
52 XORI_B2_128_SB(src8776, src10998);
53 out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, filt0,
54 filt1, filt2, filt3);
55 out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, filt0,
56 filt1, filt2, filt3);
57 SRARI_H2_SH(out10, out32, FILTER_BITS);
58 SAT_SH2_SH(out10, out32, 7);
59 out = PCKEV_XORI128_UB(out10, out32);
60 ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
61
62 dst0 = (v16u8)__msa_ilvr_d((v2i64)dst2, (v2i64)dst0);
63 out = __msa_aver_u_b(out, dst0);
64
65 ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
66 dst += (4 * dst_stride);
67
68 src2110 = src6554;
69 src4332 = src8776;
70 src6554 = src10998;
71 src6 = src10;
72 }
73 }
74
common_vt_8t_and_aver_dst_8w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)75 static void common_vt_8t_and_aver_dst_8w_msa(const uint8_t *src,
76 int32_t src_stride,
77 uint8_t *dst,
78 int32_t dst_stride,
79 int8_t *filter,
80 int32_t height) {
81 uint32_t loop_cnt;
82 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
83 v16u8 dst0, dst1, dst2, dst3;
84 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
85 v16i8 src65_r, src87_r, src109_r, filt0, filt1, filt2, filt3;
86 v8i16 filt, out0, out1, out2, out3;
87
88 src -= (3 * src_stride);
89
90 filt = LD_SH(filter);
91 SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
92
93 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
94 src += (7 * src_stride);
95
96 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
97 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
98 src54_r, src21_r);
99 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
100
101 for (loop_cnt = (height >> 2); loop_cnt--;) {
102 LD_SB4(src, src_stride, src7, src8, src9, src10);
103 src += (4 * src_stride);
104
105 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
106 XORI_B4_128_SB(src7, src8, src9, src10);
107 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
108 src87_r, src98_r, src109_r);
109 out0 = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
110 filt1, filt2, filt3);
111 out1 = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
112 filt1, filt2, filt3);
113 out2 = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
114 filt1, filt2, filt3);
115 out3 = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
116 filt1, filt2, filt3);
117 SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
118 SAT_SH4_SH(out0, out1, out2, out3, 7);
119 CONVERT_UB_AVG_ST8x4_UB(out0, out1, out2, out3, dst0, dst1, dst2, dst3,
120 dst, dst_stride);
121 dst += (4 * dst_stride);
122
123 src10_r = src54_r;
124 src32_r = src76_r;
125 src54_r = src98_r;
126 src21_r = src65_r;
127 src43_r = src87_r;
128 src65_r = src109_r;
129 src6 = src10;
130 }
131 }
132
common_vt_8t_and_aver_dst_16w_mult_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height,int32_t width)133 static void common_vt_8t_and_aver_dst_16w_mult_msa(const uint8_t *src,
134 int32_t src_stride,
135 uint8_t *dst,
136 int32_t dst_stride,
137 int8_t *filter,
138 int32_t height,
139 int32_t width) {
140 const uint8_t *src_tmp;
141 uint8_t *dst_tmp;
142 uint32_t loop_cnt, cnt;
143 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
144 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
145 v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l;
146 v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l;
147 v16i8 filt0, filt1, filt2, filt3;
148 v16u8 dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3;
149 v8i16 out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l, filt;
150
151 src -= (3 * src_stride);
152
153 filt = LD_SH(filter);
154 SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
155
156 for (cnt = (width >> 4); cnt--;) {
157 src_tmp = src;
158 dst_tmp = dst;
159
160 LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
161 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
162 src_tmp += (7 * src_stride);
163
164 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
165 src54_r, src21_r);
166 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
167 ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l, src32_l,
168 src54_l, src21_l);
169 ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
170
171 for (loop_cnt = (height >> 2); loop_cnt--;) {
172 LD_SB4(src_tmp, src_stride, src7, src8, src9, src10);
173 src_tmp += (4 * src_stride);
174
175 LD_UB4(dst_tmp, dst_stride, dst0, dst1, dst2, dst3);
176 XORI_B4_128_SB(src7, src8, src9, src10);
177 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
178 src87_r, src98_r, src109_r);
179 ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l,
180 src87_l, src98_l, src109_l);
181 out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
182 filt1, filt2, filt3);
183 out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
184 filt1, filt2, filt3);
185 out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
186 filt1, filt2, filt3);
187 out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
188 filt1, filt2, filt3);
189 out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0,
190 filt1, filt2, filt3);
191 out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0,
192 filt1, filt2, filt3);
193 out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0,
194 filt1, filt2, filt3);
195 out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0,
196 filt1, filt2, filt3);
197 SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
198 SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS);
199 SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
200 SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7);
201 PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l,
202 out3_r, tmp0, tmp1, tmp2, tmp3);
203 XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3);
204 AVER_UB4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, dst0, dst1,
205 dst2, dst3);
206 ST_UB4(dst0, dst1, dst2, dst3, dst_tmp, dst_stride);
207 dst_tmp += (4 * dst_stride);
208
209 src10_r = src54_r;
210 src32_r = src76_r;
211 src54_r = src98_r;
212 src21_r = src65_r;
213 src43_r = src87_r;
214 src65_r = src109_r;
215 src10_l = src54_l;
216 src32_l = src76_l;
217 src54_l = src98_l;
218 src21_l = src65_l;
219 src43_l = src87_l;
220 src65_l = src109_l;
221 src6 = src10;
222 }
223
224 src += 16;
225 dst += 16;
226 }
227 }
228
common_vt_8t_and_aver_dst_16w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)229 static void common_vt_8t_and_aver_dst_16w_msa(const uint8_t *src,
230 int32_t src_stride,
231 uint8_t *dst,
232 int32_t dst_stride,
233 int8_t *filter,
234 int32_t height) {
235 common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
236 filter, height, 16);
237 }
238
common_vt_8t_and_aver_dst_32w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)239 static void common_vt_8t_and_aver_dst_32w_msa(const uint8_t *src,
240 int32_t src_stride,
241 uint8_t *dst,
242 int32_t dst_stride,
243 int8_t *filter,
244 int32_t height) {
245 common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
246 filter, height, 32);
247 }
248
common_vt_8t_and_aver_dst_64w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)249 static void common_vt_8t_and_aver_dst_64w_msa(const uint8_t *src,
250 int32_t src_stride,
251 uint8_t *dst,
252 int32_t dst_stride,
253 int8_t *filter,
254 int32_t height) {
255 common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
256 filter, height, 64);
257 }
258
common_vt_2t_and_aver_dst_4x4_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter)259 static void common_vt_2t_and_aver_dst_4x4_msa(const uint8_t *src,
260 int32_t src_stride,
261 uint8_t *dst,
262 int32_t dst_stride,
263 int8_t *filter) {
264 v16i8 src0, src1, src2, src3, src4;
265 v16u8 dst0, dst1, dst2, dst3, out, filt0, src2110, src4332;
266 v16i8 src10_r, src32_r, src21_r, src43_r;
267 v8i16 filt;
268 v8u16 tmp0, tmp1;
269
270 filt = LD_SH(filter);
271 filt0 = (v16u8)__msa_splati_h(filt, 0);
272
273 LD_SB4(src, src_stride, src0, src1, src2, src3);
274 src += (4 * src_stride);
275
276 src4 = LD_SB(src);
277 src += src_stride;
278
279 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
280 ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
281 dst0 = (v16u8)__msa_ilvr_d((v2i64)dst1, (v2i64)dst0);
282 ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
283 src32_r, src43_r);
284 ILVR_D2_UB(src21_r, src10_r, src43_r, src32_r, src2110, src4332);
285 DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1);
286 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
287
288 out = (v16u8)__msa_pckev_b((v16i8)tmp1, (v16i8)tmp0);
289 out = __msa_aver_u_b(out, dst0);
290
291 ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
292 }
293
common_vt_2t_and_aver_dst_4x8_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter)294 static void common_vt_2t_and_aver_dst_4x8_msa(const uint8_t *src,
295 int32_t src_stride,
296 uint8_t *dst,
297 int32_t dst_stride,
298 int8_t *filter) {
299 v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
300 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src87_r;
301 v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r, src65_r;
302 v16u8 src2110, src4332, src6554, src8776, filt0;
303 v8u16 tmp0, tmp1, tmp2, tmp3;
304 v8i16 filt;
305
306 filt = LD_SH(filter);
307 filt0 = (v16u8)__msa_splati_h(filt, 0);
308
309 LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
310 src += (8 * src_stride);
311 src8 = LD_SB(src);
312
313 LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
314 ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst1,
315 dst2, dst3);
316 ILVR_D2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
317 ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
318 src32_r, src43_r);
319 ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r,
320 src76_r, src87_r);
321 ILVR_D4_UB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r,
322 src87_r, src76_r, src2110, src4332, src6554, src8776);
323 DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0,
324 tmp0, tmp1, tmp2, tmp3);
325 SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
326 PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, src2110, src4332);
327 AVER_UB2_UB(src2110, dst0, src4332, dst1, src2110, src4332);
328 ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
329 dst += (4 * dst_stride);
330 ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst, dst_stride);
331 }
332
common_vt_2t_and_aver_dst_4w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)333 static void common_vt_2t_and_aver_dst_4w_msa(const uint8_t *src,
334 int32_t src_stride,
335 uint8_t *dst,
336 int32_t dst_stride,
337 int8_t *filter,
338 int32_t height) {
339 if (4 == height) {
340 common_vt_2t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
341 } else if (8 == height) {
342 common_vt_2t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
343 }
344 }
345
common_vt_2t_and_aver_dst_8x4_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter)346 static void common_vt_2t_and_aver_dst_8x4_msa(const uint8_t *src,
347 int32_t src_stride,
348 uint8_t *dst,
349 int32_t dst_stride,
350 int8_t *filter) {
351 v16u8 src0, src1, src2, src3, src4;
352 v16u8 dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3, filt0;
353 v8u16 tmp0, tmp1, tmp2, tmp3;
354 v8i16 filt;
355
356 /* rearranging filter_y */
357 filt = LD_SH(filter);
358 filt0 = (v16u8)__msa_splati_h(filt, 0);
359
360 LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
361 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
362 ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1);
363 ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3);
364 DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
365 tmp2, tmp3);
366 SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
367 PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3,
368 dst, dst_stride);
369 }
370
common_vt_2t_and_aver_dst_8x8mult_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)371 static void common_vt_2t_and_aver_dst_8x8mult_msa(const uint8_t *src,
372 int32_t src_stride,
373 uint8_t *dst,
374 int32_t dst_stride,
375 int8_t *filter,
376 int32_t height) {
377 uint32_t loop_cnt;
378 v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
379 v16u8 dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
380 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
381 v8u16 tmp0, tmp1, tmp2, tmp3;
382 v8i16 filt;
383
384 /* rearranging filter_y */
385 filt = LD_SH(filter);
386 filt0 = (v16u8)__msa_splati_h(filt, 0);
387
388 src0 = LD_UB(src);
389 src += src_stride;
390
391 for (loop_cnt = (height >> 3); loop_cnt--;) {
392 LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8);
393 src += (8 * src_stride);
394 LD_UB8(dst, dst_stride, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8);
395
396 ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3, vec0, vec1,
397 vec2, vec3);
398 ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7, vec4, vec5,
399 vec6, vec7);
400 DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
401 tmp2, tmp3);
402 SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
403 PCKEV_AVG_ST8x4_UB(tmp0, dst1, tmp1, dst2, tmp2, dst3, tmp3, dst4,
404 dst, dst_stride);
405 dst += (4 * dst_stride);
406
407 DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, tmp0, tmp1,
408 tmp2, tmp3);
409 SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
410 PCKEV_AVG_ST8x4_UB(tmp0, dst5, tmp1, dst6, tmp2, dst7, tmp3, dst8,
411 dst, dst_stride);
412 dst += (4 * dst_stride);
413
414 src0 = src8;
415 }
416 }
417
common_vt_2t_and_aver_dst_8w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)418 static void common_vt_2t_and_aver_dst_8w_msa(const uint8_t *src,
419 int32_t src_stride,
420 uint8_t *dst,
421 int32_t dst_stride,
422 int8_t *filter,
423 int32_t height) {
424 if (4 == height) {
425 common_vt_2t_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride, filter);
426 } else {
427 common_vt_2t_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
428 filter, height);
429 }
430 }
431
common_vt_2t_and_aver_dst_16w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)432 static void common_vt_2t_and_aver_dst_16w_msa(const uint8_t *src,
433 int32_t src_stride,
434 uint8_t *dst,
435 int32_t dst_stride,
436 int8_t *filter,
437 int32_t height) {
438 uint32_t loop_cnt;
439 v16u8 src0, src1, src2, src3, src4, dst0, dst1, dst2, dst3, filt0;
440 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
441 v8u16 tmp0, tmp1, tmp2, tmp3, filt;
442
443 /* rearranging filter_y */
444 filt = LD_UH(filter);
445 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
446
447 src0 = LD_UB(src);
448 src += src_stride;
449
450 for (loop_cnt = (height >> 2); loop_cnt--;) {
451 LD_UB4(src, src_stride, src1, src2, src3, src4);
452 src += (4 * src_stride);
453
454 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
455 ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
456 ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
457 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
458 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
459 PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
460 dst += dst_stride;
461
462 ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
463 ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
464 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
465 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
466 PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst);
467 dst += dst_stride;
468
469 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
470 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
471 PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst);
472 dst += dst_stride;
473
474 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
475 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
476 PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst);
477 dst += dst_stride;
478
479 src0 = src4;
480 }
481 }
482
common_vt_2t_and_aver_dst_32w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)483 static void common_vt_2t_and_aver_dst_32w_msa(const uint8_t *src,
484 int32_t src_stride,
485 uint8_t *dst,
486 int32_t dst_stride,
487 int8_t *filter,
488 int32_t height) {
489 uint32_t loop_cnt;
490 v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
491 v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
492 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
493 v8u16 tmp0, tmp1, tmp2, tmp3, filt;
494
495 /* rearranging filter_y */
496 filt = LD_UH(filter);
497 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
498
499 LD_UB2(src, 16, src0, src5);
500 src += src_stride;
501
502 for (loop_cnt = (height >> 2); loop_cnt--;) {
503 LD_UB4(src, src_stride, src1, src2, src3, src4);
504 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
505 ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
506 ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
507
508 LD_UB4(src + 16, src_stride, src6, src7, src8, src9);
509 LD_UB4(dst + 16, dst_stride, dst4, dst5, dst6, dst7);
510 src += (4 * src_stride);
511
512 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
513 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
514 PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
515
516 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
517 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
518 PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride);
519
520 ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
521 ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
522 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
523 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
524 PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst + 2 * dst_stride);
525
526 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
527 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
528 PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst + 3 * dst_stride);
529
530 ILVR_B2_UB(src6, src5, src7, src6, vec0, vec2);
531 ILVL_B2_UB(src6, src5, src7, src6, vec1, vec3);
532 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
533 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
534 PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 16);
535
536 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
537 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
538 PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 16 + dst_stride);
539
540 ILVR_B2_UB(src8, src7, src9, src8, vec4, vec6);
541 ILVL_B2_UB(src8, src7, src9, src8, vec5, vec7);
542 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
543 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
544 PCKEV_AVG_ST_UB(tmp1, tmp0, dst6, dst + 16 + 2 * dst_stride);
545
546 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
547 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
548 PCKEV_AVG_ST_UB(tmp3, tmp2, dst7, dst + 16 + 3 * dst_stride);
549 dst += (4 * dst_stride);
550
551 src0 = src4;
552 src5 = src9;
553 }
554 }
555
common_vt_2t_and_aver_dst_64w_msa(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,int8_t * filter,int32_t height)556 static void common_vt_2t_and_aver_dst_64w_msa(const uint8_t *src,
557 int32_t src_stride,
558 uint8_t *dst,
559 int32_t dst_stride,
560 int8_t *filter,
561 int32_t height) {
562 uint32_t loop_cnt;
563 v16u8 src0, src1, src2, src3, src4, src5;
564 v16u8 src6, src7, src8, src9, src10, src11, filt0;
565 v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
566 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
567 v8u16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
568 v8u16 filt;
569
570 /* rearranging filter_y */
571 filt = LD_UH(filter);
572 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
573
574 LD_UB4(src, 16, src0, src3, src6, src9);
575 src += src_stride;
576
577 for (loop_cnt = (height >> 1); loop_cnt--;) {
578 LD_UB2(src, src_stride, src1, src2);
579 LD_UB2(dst, dst_stride, dst0, dst1);
580 LD_UB2(src + 16, src_stride, src4, src5);
581 LD_UB2(dst + 16, dst_stride, dst2, dst3);
582 LD_UB2(src + 32, src_stride, src7, src8);
583 LD_UB2(dst + 32, dst_stride, dst4, dst5);
584 LD_UB2(src + 48, src_stride, src10, src11);
585 LD_UB2(dst + 48, dst_stride, dst6, dst7);
586 src += (2 * src_stride);
587
588 ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
589 ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
590 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
591 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
592 PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
593
594 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
595 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
596 PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride);
597
598 ILVR_B2_UB(src4, src3, src5, src4, vec4, vec6);
599 ILVL_B2_UB(src4, src3, src5, src4, vec5, vec7);
600 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
601 SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
602 PCKEV_AVG_ST_UB(tmp5, tmp4, dst2, dst + 16);
603
604 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
605 SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
606 PCKEV_AVG_ST_UB(tmp7, tmp6, dst3, dst + 16 + dst_stride);
607
608 ILVR_B2_UB(src7, src6, src8, src7, vec0, vec2);
609 ILVL_B2_UB(src7, src6, src8, src7, vec1, vec3);
610 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
611 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
612 PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 32);
613
614 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
615 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
616 PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 32 + dst_stride);
617
618 ILVR_B2_UB(src10, src9, src11, src10, vec4, vec6);
619 ILVL_B2_UB(src10, src9, src11, src10, vec5, vec7);
620 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
621 SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
622 PCKEV_AVG_ST_UB(tmp5, tmp4, dst6, (dst + 48));
623
624 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
625 SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
626 PCKEV_AVG_ST_UB(tmp7, tmp6, dst7, dst + 48 + dst_stride);
627 dst += (2 * dst_stride);
628
629 src0 = src2;
630 src3 = src5;
631 src6 = src8;
632 src9 = src11;
633 }
634 }
635
vpx_convolve8_avg_vert_msa(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const int16_t * filter_x,int x_step_q4,const int16_t * filter_y,int y_step_q4,int w,int h)636 void vpx_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
637 uint8_t *dst, ptrdiff_t dst_stride,
638 const int16_t *filter_x, int x_step_q4,
639 const int16_t *filter_y, int y_step_q4,
640 int w, int h) {
641 int8_t cnt, filt_ver[8];
642
643 assert(y_step_q4 == 16);
644 assert(((const int32_t *)filter_y)[1] != 0x800000);
645
646 for (cnt = 0; cnt < 8; ++cnt) {
647 filt_ver[cnt] = filter_y[cnt];
648 }
649
650 if (((const int32_t *)filter_y)[0] == 0) {
651 switch (w) {
652 case 4:
653 common_vt_2t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
654 dst, (int32_t)dst_stride,
655 &filt_ver[3], h);
656 break;
657 case 8:
658 common_vt_2t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
659 dst, (int32_t)dst_stride,
660 &filt_ver[3], h);
661 break;
662 case 16:
663 common_vt_2t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
664 dst, (int32_t)dst_stride,
665 &filt_ver[3], h);
666 break;
667 case 32:
668 common_vt_2t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
669 dst, (int32_t)dst_stride,
670 &filt_ver[3], h);
671 break;
672 case 64:
673 common_vt_2t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
674 dst, (int32_t)dst_stride,
675 &filt_ver[3], h);
676 break;
677 default:
678 vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
679 filter_x, x_step_q4, filter_y, y_step_q4,
680 w, h);
681 break;
682 }
683 } else {
684 switch (w) {
685 case 4:
686 common_vt_8t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
687 dst, (int32_t)dst_stride,
688 filt_ver, h);
689 break;
690 case 8:
691 common_vt_8t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
692 dst, (int32_t)dst_stride,
693 filt_ver, h);
694 break;
695 case 16:
696 common_vt_8t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
697 dst, (int32_t)dst_stride,
698 filt_ver, h);
699
700 break;
701 case 32:
702 common_vt_8t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
703 dst, (int32_t)dst_stride,
704 filt_ver, h);
705 break;
706 case 64:
707 common_vt_8t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
708 dst, (int32_t)dst_stride,
709 filt_ver, h);
710 break;
711 default:
712 vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
713 filter_x, x_step_q4, filter_y, y_step_q4,
714 w, h);
715 break;
716 }
717 }
718 }
719