Line data Source code
1 : /*
2 : * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 : *
4 : * This source code is subject to the terms of the BSD 2 Clause License and
5 : * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 : * was not distributed with this source code in the LICENSE file, you can
7 : * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 : * Media Patent License 1.0 was not distributed with this source code in the
9 : * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 : */
11 :
12 : #include <smmintrin.h> // SSE4.1
13 :
14 : #include <assert.h>
15 :
16 : #include "aom/aom_integer.h"
17 : #include "aom_ports/mem.h"
18 : #include "aom_dsp/aom_dsp_common.h"
19 : #include "aom_dsp/blend.h"
20 :
21 : #include "aom_dsp/x86/synonyms.h"
22 : #include "aom_dsp/x86/blend_sse4.h"
23 :
24 : #include "./aom_dsp_rtcd.h"
25 :
26 : //////////////////////////////////////////////////////////////////////////////
27 : // No sub-sampling
28 : //////////////////////////////////////////////////////////////////////////////
29 :
30 0 : static void blend_a64_mask_w4_sse4_1(uint8_t *dst, uint32_t dst_stride,
31 : const uint8_t *src0, uint32_t src0_stride,
32 : const uint8_t *src1, uint32_t src1_stride,
33 : const uint8_t *mask, uint32_t mask_stride,
34 : int h, int w) {
35 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
36 :
37 : (void)w;
38 :
39 : do {
40 0 : const __m128i v_m0_b = xx_loadl_32(mask);
41 0 : const __m128i v_m0_w = _mm_cvtepu8_epi16(v_m0_b);
42 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
43 :
44 0 : const __m128i v_res_w = blend_4(src0, src1, v_m0_w, v_m1_w);
45 :
46 0 : const __m128i v_res_b = _mm_packus_epi16(v_res_w, v_res_w);
47 :
48 0 : xx_storel_32(dst, v_res_b);
49 :
50 0 : dst += dst_stride;
51 0 : src0 += src0_stride;
52 0 : src1 += src1_stride;
53 0 : mask += mask_stride;
54 0 : } while (--h);
55 0 : }
56 :
57 0 : static void blend_a64_mask_w8_sse4_1(uint8_t *dst, uint32_t dst_stride,
58 : const uint8_t *src0, uint32_t src0_stride,
59 : const uint8_t *src1, uint32_t src1_stride,
60 : const uint8_t *mask, uint32_t mask_stride,
61 : int h, int w) {
62 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
63 :
64 : (void)w;
65 :
66 : do {
67 0 : const __m128i v_m0_b = xx_loadl_64(mask);
68 0 : const __m128i v_m0_w = _mm_cvtepu8_epi16(v_m0_b);
69 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
70 :
71 0 : const __m128i v_res_w = blend_8(src0, src1, v_m0_w, v_m1_w);
72 :
73 0 : const __m128i v_res_b = _mm_packus_epi16(v_res_w, v_res_w);
74 :
75 0 : xx_storel_64(dst, v_res_b);
76 :
77 0 : dst += dst_stride;
78 0 : src0 += src0_stride;
79 0 : src1 += src1_stride;
80 0 : mask += mask_stride;
81 0 : } while (--h);
82 0 : }
83 :
84 0 : static void blend_a64_mask_w16n_sse4_1(
85 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
86 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
87 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
88 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
89 :
90 : do {
91 : int c;
92 0 : for (c = 0; c < w; c += 16) {
93 0 : const __m128i v_m0l_b = xx_loadl_64(mask + c);
94 0 : const __m128i v_m0h_b = xx_loadl_64(mask + c + 8);
95 0 : const __m128i v_m0l_w = _mm_cvtepu8_epi16(v_m0l_b);
96 0 : const __m128i v_m0h_w = _mm_cvtepu8_epi16(v_m0h_b);
97 0 : const __m128i v_m1l_w = _mm_sub_epi16(v_maxval_w, v_m0l_w);
98 0 : const __m128i v_m1h_w = _mm_sub_epi16(v_maxval_w, v_m0h_w);
99 :
100 0 : const __m128i v_resl_w = blend_8(src0 + c, src1 + c, v_m0l_w, v_m1l_w);
101 0 : const __m128i v_resh_w =
102 0 : blend_8(src0 + c + 8, src1 + c + 8, v_m0h_w, v_m1h_w);
103 :
104 0 : const __m128i v_res_b = _mm_packus_epi16(v_resl_w, v_resh_w);
105 :
106 0 : xx_storeu_128(dst + c, v_res_b);
107 : }
108 0 : dst += dst_stride;
109 0 : src0 += src0_stride;
110 0 : src1 += src1_stride;
111 0 : mask += mask_stride;
112 0 : } while (--h);
113 0 : }
114 :
115 : //////////////////////////////////////////////////////////////////////////////
116 : // Horizontal sub-sampling
117 : //////////////////////////////////////////////////////////////////////////////
118 :
119 0 : static void blend_a64_mask_sx_w4_sse4_1(
120 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
121 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
122 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
123 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
124 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
125 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
126 :
127 : (void)w;
128 :
129 : do {
130 0 : const __m128i v_r_b = xx_loadl_64(mask);
131 0 : const __m128i v_a_b = _mm_avg_epu8(v_r_b, _mm_srli_si128(v_r_b, 1));
132 :
133 0 : const __m128i v_m0_w = _mm_and_si128(v_a_b, v_zmask_b);
134 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
135 :
136 0 : const __m128i v_res_w = blend_4(src0, src1, v_m0_w, v_m1_w);
137 :
138 0 : const __m128i v_res_b = _mm_packus_epi16(v_res_w, v_res_w);
139 :
140 0 : xx_storel_32(dst, v_res_b);
141 :
142 0 : dst += dst_stride;
143 0 : src0 += src0_stride;
144 0 : src1 += src1_stride;
145 0 : mask += mask_stride;
146 0 : } while (--h);
147 0 : }
148 :
149 0 : static void blend_a64_mask_sx_w8_sse4_1(
150 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
151 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
152 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
153 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
154 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
155 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
156 :
157 : (void)w;
158 :
159 : do {
160 0 : const __m128i v_r_b = xx_loadu_128(mask);
161 0 : const __m128i v_a_b = _mm_avg_epu8(v_r_b, _mm_srli_si128(v_r_b, 1));
162 :
163 0 : const __m128i v_m0_w = _mm_and_si128(v_a_b, v_zmask_b);
164 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
165 :
166 0 : const __m128i v_res_w = blend_8(src0, src1, v_m0_w, v_m1_w);
167 :
168 0 : const __m128i v_res_b = _mm_packus_epi16(v_res_w, v_res_w);
169 :
170 0 : xx_storel_64(dst, v_res_b);
171 :
172 0 : dst += dst_stride;
173 0 : src0 += src0_stride;
174 0 : src1 += src1_stride;
175 0 : mask += mask_stride;
176 0 : } while (--h);
177 0 : }
178 :
179 0 : static void blend_a64_mask_sx_w16n_sse4_1(
180 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
181 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
182 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
183 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
184 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
185 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
186 :
187 : do {
188 : int c;
189 0 : for (c = 0; c < w; c += 16) {
190 0 : const __m128i v_rl_b = xx_loadu_128(mask + 2 * c);
191 0 : const __m128i v_rh_b = xx_loadu_128(mask + 2 * c + 16);
192 0 : const __m128i v_al_b = _mm_avg_epu8(v_rl_b, _mm_srli_si128(v_rl_b, 1));
193 0 : const __m128i v_ah_b = _mm_avg_epu8(v_rh_b, _mm_srli_si128(v_rh_b, 1));
194 :
195 0 : const __m128i v_m0l_w = _mm_and_si128(v_al_b, v_zmask_b);
196 0 : const __m128i v_m0h_w = _mm_and_si128(v_ah_b, v_zmask_b);
197 0 : const __m128i v_m1l_w = _mm_sub_epi16(v_maxval_w, v_m0l_w);
198 0 : const __m128i v_m1h_w = _mm_sub_epi16(v_maxval_w, v_m0h_w);
199 :
200 0 : const __m128i v_resl_w = blend_8(src0 + c, src1 + c, v_m0l_w, v_m1l_w);
201 0 : const __m128i v_resh_w =
202 0 : blend_8(src0 + c + 8, src1 + c + 8, v_m0h_w, v_m1h_w);
203 :
204 0 : const __m128i v_res_b = _mm_packus_epi16(v_resl_w, v_resh_w);
205 :
206 0 : xx_storeu_128(dst + c, v_res_b);
207 : }
208 0 : dst += dst_stride;
209 0 : src0 += src0_stride;
210 0 : src1 += src1_stride;
211 0 : mask += mask_stride;
212 0 : } while (--h);
213 0 : }
214 :
215 : //////////////////////////////////////////////////////////////////////////////
216 : // Vertical sub-sampling
217 : //////////////////////////////////////////////////////////////////////////////
218 :
219 0 : static void blend_a64_mask_sy_w4_sse4_1(
220 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
221 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
222 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
223 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
224 :
225 : (void)w;
226 :
227 : do {
228 0 : const __m128i v_ra_b = xx_loadl_32(mask);
229 0 : const __m128i v_rb_b = xx_loadl_32(mask + mask_stride);
230 0 : const __m128i v_a_b = _mm_avg_epu8(v_ra_b, v_rb_b);
231 :
232 0 : const __m128i v_m0_w = _mm_cvtepu8_epi16(v_a_b);
233 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
234 :
235 0 : const __m128i v_res_w = blend_4(src0, src1, v_m0_w, v_m1_w);
236 :
237 0 : const __m128i v_res_b = _mm_packus_epi16(v_res_w, v_res_w);
238 :
239 0 : xx_storel_32(dst, v_res_b);
240 :
241 0 : dst += dst_stride;
242 0 : src0 += src0_stride;
243 0 : src1 += src1_stride;
244 0 : mask += 2 * mask_stride;
245 0 : } while (--h);
246 0 : }
247 :
248 0 : static void blend_a64_mask_sy_w8_sse4_1(
249 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
250 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
251 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
252 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
253 :
254 : (void)w;
255 :
256 : do {
257 0 : const __m128i v_ra_b = xx_loadl_64(mask);
258 0 : const __m128i v_rb_b = xx_loadl_64(mask + mask_stride);
259 0 : const __m128i v_a_b = _mm_avg_epu8(v_ra_b, v_rb_b);
260 :
261 0 : const __m128i v_m0_w = _mm_cvtepu8_epi16(v_a_b);
262 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
263 :
264 0 : const __m128i v_res_w = blend_8(src0, src1, v_m0_w, v_m1_w);
265 :
266 0 : const __m128i v_res_b = _mm_packus_epi16(v_res_w, v_res_w);
267 :
268 0 : xx_storel_64(dst, v_res_b);
269 :
270 0 : dst += dst_stride;
271 0 : src0 += src0_stride;
272 0 : src1 += src1_stride;
273 0 : mask += 2 * mask_stride;
274 0 : } while (--h);
275 0 : }
276 :
277 0 : static void blend_a64_mask_sy_w16n_sse4_1(
278 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
279 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
280 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
281 0 : const __m128i v_zero = _mm_setzero_si128();
282 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
283 :
284 : do {
285 : int c;
286 0 : for (c = 0; c < w; c += 16) {
287 0 : const __m128i v_ra_b = xx_loadu_128(mask + c);
288 0 : const __m128i v_rb_b = xx_loadu_128(mask + c + mask_stride);
289 0 : const __m128i v_a_b = _mm_avg_epu8(v_ra_b, v_rb_b);
290 :
291 0 : const __m128i v_m0l_w = _mm_cvtepu8_epi16(v_a_b);
292 0 : const __m128i v_m0h_w = _mm_unpackhi_epi8(v_a_b, v_zero);
293 0 : const __m128i v_m1l_w = _mm_sub_epi16(v_maxval_w, v_m0l_w);
294 0 : const __m128i v_m1h_w = _mm_sub_epi16(v_maxval_w, v_m0h_w);
295 :
296 0 : const __m128i v_resl_w = blend_8(src0 + c, src1 + c, v_m0l_w, v_m1l_w);
297 0 : const __m128i v_resh_w =
298 0 : blend_8(src0 + c + 8, src1 + c + 8, v_m0h_w, v_m1h_w);
299 :
300 0 : const __m128i v_res_b = _mm_packus_epi16(v_resl_w, v_resh_w);
301 :
302 0 : xx_storeu_128(dst + c, v_res_b);
303 : }
304 0 : dst += dst_stride;
305 0 : src0 += src0_stride;
306 0 : src1 += src1_stride;
307 0 : mask += 2 * mask_stride;
308 0 : } while (--h);
309 0 : }
310 :
311 : //////////////////////////////////////////////////////////////////////////////
312 : // Horizontal and Vertical sub-sampling
313 : //////////////////////////////////////////////////////////////////////////////
314 :
315 0 : static void blend_a64_mask_sx_sy_w4_sse4_1(
316 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
317 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
318 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
319 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
320 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
321 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
322 :
323 : (void)w;
324 :
325 : do {
326 0 : const __m128i v_ra_b = xx_loadl_64(mask);
327 0 : const __m128i v_rb_b = xx_loadl_64(mask + mask_stride);
328 0 : const __m128i v_rvs_b = _mm_add_epi8(v_ra_b, v_rb_b);
329 0 : const __m128i v_rvsa_w = _mm_and_si128(v_rvs_b, v_zmask_b);
330 0 : const __m128i v_rvsb_w =
331 0 : _mm_and_si128(_mm_srli_si128(v_rvs_b, 1), v_zmask_b);
332 0 : const __m128i v_rs_w = _mm_add_epi16(v_rvsa_w, v_rvsb_w);
333 :
334 0 : const __m128i v_m0_w = xx_roundn_epu16(v_rs_w, 2);
335 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
336 :
337 0 : const __m128i v_res_w = blend_4(src0, src1, v_m0_w, v_m1_w);
338 :
339 0 : const __m128i v_res_b = _mm_packus_epi16(v_res_w, v_res_w);
340 :
341 0 : xx_storel_32(dst, v_res_b);
342 :
343 0 : dst += dst_stride;
344 0 : src0 += src0_stride;
345 0 : src1 += src1_stride;
346 0 : mask += 2 * mask_stride;
347 0 : } while (--h);
348 0 : }
349 :
350 0 : static void blend_a64_mask_sx_sy_w8_sse4_1(
351 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
352 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
353 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
354 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
355 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
356 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
357 :
358 : (void)w;
359 :
360 : do {
361 0 : const __m128i v_ra_b = xx_loadu_128(mask);
362 0 : const __m128i v_rb_b = xx_loadu_128(mask + mask_stride);
363 0 : const __m128i v_rvs_b = _mm_add_epi8(v_ra_b, v_rb_b);
364 0 : const __m128i v_rvsa_w = _mm_and_si128(v_rvs_b, v_zmask_b);
365 0 : const __m128i v_rvsb_w =
366 0 : _mm_and_si128(_mm_srli_si128(v_rvs_b, 1), v_zmask_b);
367 0 : const __m128i v_rs_w = _mm_add_epi16(v_rvsa_w, v_rvsb_w);
368 :
369 0 : const __m128i v_m0_w = xx_roundn_epu16(v_rs_w, 2);
370 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
371 :
372 0 : const __m128i v_res_w = blend_8(src0, src1, v_m0_w, v_m1_w);
373 :
374 0 : const __m128i v_res_b = _mm_packus_epi16(v_res_w, v_res_w);
375 :
376 0 : xx_storel_64(dst, v_res_b);
377 :
378 0 : dst += dst_stride;
379 0 : src0 += src0_stride;
380 0 : src1 += src1_stride;
381 0 : mask += 2 * mask_stride;
382 0 : } while (--h);
383 0 : }
384 :
385 0 : static void blend_a64_mask_sx_sy_w16n_sse4_1(
386 : uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
387 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
388 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
389 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
390 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
391 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
392 :
393 : do {
394 : int c;
395 0 : for (c = 0; c < w; c += 16) {
396 0 : const __m128i v_ral_b = xx_loadu_128(mask + 2 * c);
397 0 : const __m128i v_rah_b = xx_loadu_128(mask + 2 * c + 16);
398 0 : const __m128i v_rbl_b = xx_loadu_128(mask + mask_stride + 2 * c);
399 0 : const __m128i v_rbh_b = xx_loadu_128(mask + mask_stride + 2 * c + 16);
400 0 : const __m128i v_rvsl_b = _mm_add_epi8(v_ral_b, v_rbl_b);
401 0 : const __m128i v_rvsh_b = _mm_add_epi8(v_rah_b, v_rbh_b);
402 0 : const __m128i v_rvsal_w = _mm_and_si128(v_rvsl_b, v_zmask_b);
403 0 : const __m128i v_rvsah_w = _mm_and_si128(v_rvsh_b, v_zmask_b);
404 0 : const __m128i v_rvsbl_w =
405 0 : _mm_and_si128(_mm_srli_si128(v_rvsl_b, 1), v_zmask_b);
406 0 : const __m128i v_rvsbh_w =
407 0 : _mm_and_si128(_mm_srli_si128(v_rvsh_b, 1), v_zmask_b);
408 0 : const __m128i v_rsl_w = _mm_add_epi16(v_rvsal_w, v_rvsbl_w);
409 0 : const __m128i v_rsh_w = _mm_add_epi16(v_rvsah_w, v_rvsbh_w);
410 :
411 0 : const __m128i v_m0l_w = xx_roundn_epu16(v_rsl_w, 2);
412 0 : const __m128i v_m0h_w = xx_roundn_epu16(v_rsh_w, 2);
413 0 : const __m128i v_m1l_w = _mm_sub_epi16(v_maxval_w, v_m0l_w);
414 0 : const __m128i v_m1h_w = _mm_sub_epi16(v_maxval_w, v_m0h_w);
415 :
416 0 : const __m128i v_resl_w = blend_8(src0 + c, src1 + c, v_m0l_w, v_m1l_w);
417 0 : const __m128i v_resh_w =
418 0 : blend_8(src0 + c + 8, src1 + c + 8, v_m0h_w, v_m1h_w);
419 :
420 0 : const __m128i v_res_b = _mm_packus_epi16(v_resl_w, v_resh_w);
421 :
422 0 : xx_storeu_128(dst + c, v_res_b);
423 : }
424 0 : dst += dst_stride;
425 0 : src0 += src0_stride;
426 0 : src1 += src1_stride;
427 0 : mask += 2 * mask_stride;
428 0 : } while (--h);
429 0 : }
430 :
431 : //////////////////////////////////////////////////////////////////////////////
432 : // Dispatch
433 : //////////////////////////////////////////////////////////////////////////////
434 :
435 0 : void aom_blend_a64_mask_sse4_1(uint8_t *dst, uint32_t dst_stride,
436 : const uint8_t *src0, uint32_t src0_stride,
437 : const uint8_t *src1, uint32_t src1_stride,
438 : const uint8_t *mask, uint32_t mask_stride, int h,
439 : int w, int suby, int subx) {
440 : typedef void (*blend_fn)(
441 : uint8_t * dst, uint32_t dst_stride, const uint8_t *src0,
442 : uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
443 : const uint8_t *mask, uint32_t mask_stride, int h, int w);
444 :
445 : // Dimensions are: width_index X subx X suby
446 : static const blend_fn blend[3][2][2] = {
447 : { // w % 16 == 0
448 : { blend_a64_mask_w16n_sse4_1, blend_a64_mask_sy_w16n_sse4_1 },
449 : { blend_a64_mask_sx_w16n_sse4_1, blend_a64_mask_sx_sy_w16n_sse4_1 } },
450 : { // w == 4
451 : { blend_a64_mask_w4_sse4_1, blend_a64_mask_sy_w4_sse4_1 },
452 : { blend_a64_mask_sx_w4_sse4_1, blend_a64_mask_sx_sy_w4_sse4_1 } },
453 : { // w == 8
454 : { blend_a64_mask_w8_sse4_1, blend_a64_mask_sy_w8_sse4_1 },
455 : { blend_a64_mask_sx_w8_sse4_1, blend_a64_mask_sx_sy_w8_sse4_1 } }
456 : };
457 :
458 0 : assert(IMPLIES(src0 == dst, src0_stride == dst_stride));
459 0 : assert(IMPLIES(src1 == dst, src1_stride == dst_stride));
460 :
461 0 : assert(h >= 1);
462 0 : assert(w >= 1);
463 0 : assert(IS_POWER_OF_TWO(h));
464 0 : assert(IS_POWER_OF_TWO(w));
465 :
466 0 : if (UNLIKELY((h | w) & 3)) { // if (w <= 2 || h <= 2)
467 0 : aom_blend_a64_mask_c(dst, dst_stride, src0, src0_stride, src1, src1_stride,
468 : mask, mask_stride, h, w, suby, subx);
469 : } else {
470 0 : blend[(w >> 2) & 3][subx != 0][suby != 0](dst, dst_stride, src0,
471 : src0_stride, src1, src1_stride,
472 : mask, mask_stride, h, w);
473 : }
474 0 : }
475 :
476 : #if CONFIG_HIGHBITDEPTH
477 : //////////////////////////////////////////////////////////////////////////////
478 : // No sub-sampling
479 : //////////////////////////////////////////////////////////////////////////////
480 :
481 0 : static INLINE void blend_a64_mask_bn_w4_sse4_1(
482 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
483 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
484 : const uint8_t *mask, uint32_t mask_stride, int h, blend_unit_fn blend) {
485 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
486 :
487 : do {
488 0 : const __m128i v_m0_b = xx_loadl_32(mask);
489 0 : const __m128i v_m0_w = _mm_cvtepu8_epi16(v_m0_b);
490 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
491 :
492 0 : const __m128i v_res_w = blend(src0, src1, v_m0_w, v_m1_w);
493 :
494 0 : xx_storel_64(dst, v_res_w);
495 :
496 0 : dst += dst_stride;
497 0 : src0 += src0_stride;
498 0 : src1 += src1_stride;
499 0 : mask += mask_stride;
500 0 : } while (--h);
501 0 : }
502 :
503 0 : static void blend_a64_mask_b10_w4_sse4_1(
504 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
505 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
506 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
507 : (void)w;
508 0 : blend_a64_mask_bn_w4_sse4_1(dst, dst_stride, src0, src0_stride, src1,
509 : src1_stride, mask, mask_stride, h, blend_4_b10);
510 0 : }
511 :
512 0 : static void blend_a64_mask_b12_w4_sse4_1(
513 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
514 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
515 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
516 : (void)w;
517 0 : blend_a64_mask_bn_w4_sse4_1(dst, dst_stride, src0, src0_stride, src1,
518 : src1_stride, mask, mask_stride, h, blend_4_b12);
519 0 : }
520 :
521 0 : static INLINE void blend_a64_mask_bn_w8n_sse4_1(
522 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
523 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
524 : const uint8_t *mask, uint32_t mask_stride, int h, int w,
525 : blend_unit_fn blend) {
526 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
527 :
528 : do {
529 : int c;
530 0 : for (c = 0; c < w; c += 8) {
531 0 : const __m128i v_m0_b = xx_loadl_64(mask + c);
532 0 : const __m128i v_m0_w = _mm_cvtepu8_epi16(v_m0_b);
533 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
534 :
535 0 : const __m128i v_res_w = blend(src0 + c, src1 + c, v_m0_w, v_m1_w);
536 :
537 0 : xx_storeu_128(dst + c, v_res_w);
538 : }
539 0 : dst += dst_stride;
540 0 : src0 += src0_stride;
541 0 : src1 += src1_stride;
542 0 : mask += mask_stride;
543 0 : } while (--h);
544 0 : }
545 :
546 0 : static void blend_a64_mask_b10_w8n_sse4_1(
547 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
548 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
549 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
550 0 : blend_a64_mask_bn_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
551 : src1_stride, mask, mask_stride, h, w,
552 : blend_8_b10);
553 0 : }
554 :
555 0 : static void blend_a64_mask_b12_w8n_sse4_1(
556 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
557 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
558 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
559 0 : blend_a64_mask_bn_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
560 : src1_stride, mask, mask_stride, h, w,
561 : blend_8_b12);
562 0 : }
563 :
564 : //////////////////////////////////////////////////////////////////////////////
565 : // Horizontal sub-sampling
566 : //////////////////////////////////////////////////////////////////////////////
567 :
568 0 : static INLINE void blend_a64_mask_bn_sx_w4_sse4_1(
569 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
570 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
571 : const uint8_t *mask, uint32_t mask_stride, int h, blend_unit_fn blend) {
572 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
573 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
574 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
575 :
576 : do {
577 0 : const __m128i v_r_b = xx_loadl_64(mask);
578 0 : const __m128i v_a_b = _mm_avg_epu8(v_r_b, _mm_srli_si128(v_r_b, 1));
579 :
580 0 : const __m128i v_m0_w = _mm_and_si128(v_a_b, v_zmask_b);
581 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
582 :
583 0 : const __m128i v_res_w = blend(src0, src1, v_m0_w, v_m1_w);
584 :
585 0 : xx_storel_64(dst, v_res_w);
586 :
587 0 : dst += dst_stride;
588 0 : src0 += src0_stride;
589 0 : src1 += src1_stride;
590 0 : mask += mask_stride;
591 0 : } while (--h);
592 0 : }
593 :
594 0 : static void blend_a64_mask_b10_sx_w4_sse4_1(
595 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
596 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
597 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
598 : (void)w;
599 0 : blend_a64_mask_bn_sx_w4_sse4_1(dst, dst_stride, src0, src0_stride, src1,
600 : src1_stride, mask, mask_stride, h,
601 : blend_4_b10);
602 0 : }
603 :
604 0 : static void blend_a64_mask_b12_sx_w4_sse4_1(
605 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
606 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
607 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
608 : (void)w;
609 0 : blend_a64_mask_bn_sx_w4_sse4_1(dst, dst_stride, src0, src0_stride, src1,
610 : src1_stride, mask, mask_stride, h,
611 : blend_4_b12);
612 0 : }
613 :
614 0 : static INLINE void blend_a64_mask_bn_sx_w8n_sse4_1(
615 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
616 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
617 : const uint8_t *mask, uint32_t mask_stride, int h, int w,
618 : blend_unit_fn blend) {
619 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
620 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
621 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
622 :
623 : do {
624 : int c;
625 0 : for (c = 0; c < w; c += 8) {
626 0 : const __m128i v_r_b = xx_loadu_128(mask + 2 * c);
627 0 : const __m128i v_a_b = _mm_avg_epu8(v_r_b, _mm_srli_si128(v_r_b, 1));
628 :
629 0 : const __m128i v_m0_w = _mm_and_si128(v_a_b, v_zmask_b);
630 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
631 :
632 0 : const __m128i v_res_w = blend(src0 + c, src1 + c, v_m0_w, v_m1_w);
633 :
634 0 : xx_storeu_128(dst + c, v_res_w);
635 : }
636 0 : dst += dst_stride;
637 0 : src0 += src0_stride;
638 0 : src1 += src1_stride;
639 0 : mask += mask_stride;
640 0 : } while (--h);
641 0 : }
642 :
643 0 : static void blend_a64_mask_b10_sx_w8n_sse4_1(
644 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
645 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
646 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
647 0 : blend_a64_mask_bn_sx_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
648 : src1_stride, mask, mask_stride, h, w,
649 : blend_8_b10);
650 0 : }
651 :
652 0 : static void blend_a64_mask_b12_sx_w8n_sse4_1(
653 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
654 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
655 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
656 0 : blend_a64_mask_bn_sx_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
657 : src1_stride, mask, mask_stride, h, w,
658 : blend_8_b12);
659 0 : }
660 :
661 : //////////////////////////////////////////////////////////////////////////////
662 : // Vertical sub-sampling
663 : //////////////////////////////////////////////////////////////////////////////
664 :
665 0 : static INLINE void blend_a64_mask_bn_sy_w4_sse4_1(
666 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
667 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
668 : const uint8_t *mask, uint32_t mask_stride, int h, blend_unit_fn blend) {
669 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
670 :
671 : do {
672 0 : const __m128i v_ra_b = xx_loadl_32(mask);
673 0 : const __m128i v_rb_b = xx_loadl_32(mask + mask_stride);
674 0 : const __m128i v_a_b = _mm_avg_epu8(v_ra_b, v_rb_b);
675 :
676 0 : const __m128i v_m0_w = _mm_cvtepu8_epi16(v_a_b);
677 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
678 :
679 0 : const __m128i v_res_w = blend(src0, src1, v_m0_w, v_m1_w);
680 :
681 0 : xx_storel_64(dst, v_res_w);
682 :
683 0 : dst += dst_stride;
684 0 : src0 += src0_stride;
685 0 : src1 += src1_stride;
686 0 : mask += 2 * mask_stride;
687 0 : } while (--h);
688 0 : }
689 :
690 0 : static void blend_a64_mask_b10_sy_w4_sse4_1(
691 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
692 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
693 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
694 : (void)w;
695 0 : blend_a64_mask_bn_sy_w4_sse4_1(dst, dst_stride, src0, src0_stride, src1,
696 : src1_stride, mask, mask_stride, h,
697 : blend_4_b10);
698 0 : }
699 :
700 0 : static void blend_a64_mask_b12_sy_w4_sse4_1(
701 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
702 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
703 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
704 : (void)w;
705 0 : blend_a64_mask_bn_sy_w4_sse4_1(dst, dst_stride, src0, src0_stride, src1,
706 : src1_stride, mask, mask_stride, h,
707 : blend_4_b12);
708 0 : }
709 :
710 0 : static INLINE void blend_a64_mask_bn_sy_w8n_sse4_1(
711 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
712 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
713 : const uint8_t *mask, uint32_t mask_stride, int h, int w,
714 : blend_unit_fn blend) {
715 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
716 :
717 : do {
718 : int c;
719 0 : for (c = 0; c < w; c += 8) {
720 0 : const __m128i v_ra_b = xx_loadl_64(mask + c);
721 0 : const __m128i v_rb_b = xx_loadl_64(mask + c + mask_stride);
722 0 : const __m128i v_a_b = _mm_avg_epu8(v_ra_b, v_rb_b);
723 :
724 0 : const __m128i v_m0_w = _mm_cvtepu8_epi16(v_a_b);
725 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
726 :
727 0 : const __m128i v_res_w = blend(src0 + c, src1 + c, v_m0_w, v_m1_w);
728 :
729 0 : xx_storeu_128(dst + c, v_res_w);
730 : }
731 0 : dst += dst_stride;
732 0 : src0 += src0_stride;
733 0 : src1 += src1_stride;
734 0 : mask += 2 * mask_stride;
735 0 : } while (--h);
736 0 : }
737 :
738 0 : static void blend_a64_mask_b10_sy_w8n_sse4_1(
739 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
740 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
741 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
742 0 : blend_a64_mask_bn_sy_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
743 : src1_stride, mask, mask_stride, h, w,
744 : blend_8_b10);
745 0 : }
746 :
747 0 : static void blend_a64_mask_b12_sy_w8n_sse4_1(
748 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
749 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
750 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
751 0 : blend_a64_mask_bn_sy_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
752 : src1_stride, mask, mask_stride, h, w,
753 : blend_8_b12);
754 0 : }
755 :
756 : //////////////////////////////////////////////////////////////////////////////
757 : // Horizontal and Vertical sub-sampling
758 : //////////////////////////////////////////////////////////////////////////////
759 :
760 0 : static INLINE void blend_a64_mask_bn_sx_sy_w4_sse4_1(
761 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
762 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
763 : const uint8_t *mask, uint32_t mask_stride, int h, blend_unit_fn blend) {
764 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
765 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
766 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
767 :
768 : do {
769 0 : const __m128i v_ra_b = xx_loadl_64(mask);
770 0 : const __m128i v_rb_b = xx_loadl_64(mask + mask_stride);
771 0 : const __m128i v_rvs_b = _mm_add_epi8(v_ra_b, v_rb_b);
772 0 : const __m128i v_rvsa_w = _mm_and_si128(v_rvs_b, v_zmask_b);
773 0 : const __m128i v_rvsb_w =
774 0 : _mm_and_si128(_mm_srli_si128(v_rvs_b, 1), v_zmask_b);
775 0 : const __m128i v_rs_w = _mm_add_epi16(v_rvsa_w, v_rvsb_w);
776 :
777 0 : const __m128i v_m0_w = xx_roundn_epu16(v_rs_w, 2);
778 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
779 :
780 0 : const __m128i v_res_w = blend(src0, src1, v_m0_w, v_m1_w);
781 :
782 0 : xx_storel_64(dst, v_res_w);
783 :
784 0 : dst += dst_stride;
785 0 : src0 += src0_stride;
786 0 : src1 += src1_stride;
787 0 : mask += 2 * mask_stride;
788 0 : } while (--h);
789 0 : }
790 :
791 0 : static void blend_a64_mask_b10_sx_sy_w4_sse4_1(
792 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
793 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
794 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
795 : (void)w;
796 0 : blend_a64_mask_bn_sx_sy_w4_sse4_1(dst, dst_stride, src0, src0_stride, src1,
797 : src1_stride, mask, mask_stride, h,
798 : blend_4_b10);
799 0 : }
800 :
801 0 : static void blend_a64_mask_b12_sx_sy_w4_sse4_1(
802 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
803 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
804 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
805 : (void)w;
806 0 : blend_a64_mask_bn_sx_sy_w4_sse4_1(dst, dst_stride, src0, src0_stride, src1,
807 : src1_stride, mask, mask_stride, h,
808 : blend_4_b12);
809 0 : }
810 :
811 0 : static INLINE void blend_a64_mask_bn_sx_sy_w8n_sse4_1(
812 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
813 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
814 : const uint8_t *mask, uint32_t mask_stride, int h, int w,
815 : blend_unit_fn blend) {
816 0 : const __m128i v_zmask_b = _mm_set_epi8(0, 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0,
817 : 0xff, 0, 0xff, 0, 0xff, 0, 0xff);
818 0 : const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
819 :
820 : do {
821 : int c;
822 0 : for (c = 0; c < w; c += 8) {
823 0 : const __m128i v_ra_b = xx_loadu_128(mask + 2 * c);
824 0 : const __m128i v_rb_b = xx_loadu_128(mask + 2 * c + mask_stride);
825 0 : const __m128i v_rvs_b = _mm_add_epi8(v_ra_b, v_rb_b);
826 0 : const __m128i v_rvsa_w = _mm_and_si128(v_rvs_b, v_zmask_b);
827 0 : const __m128i v_rvsb_w =
828 0 : _mm_and_si128(_mm_srli_si128(v_rvs_b, 1), v_zmask_b);
829 0 : const __m128i v_rs_w = _mm_add_epi16(v_rvsa_w, v_rvsb_w);
830 :
831 0 : const __m128i v_m0_w = xx_roundn_epu16(v_rs_w, 2);
832 0 : const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);
833 :
834 0 : const __m128i v_res_w = blend(src0 + c, src1 + c, v_m0_w, v_m1_w);
835 :
836 0 : xx_storeu_128(dst + c, v_res_w);
837 : }
838 0 : dst += dst_stride;
839 0 : src0 += src0_stride;
840 0 : src1 += src1_stride;
841 0 : mask += 2 * mask_stride;
842 0 : } while (--h);
843 0 : }
844 :
845 0 : static void blend_a64_mask_b10_sx_sy_w8n_sse4_1(
846 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
847 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
848 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
849 0 : blend_a64_mask_bn_sx_sy_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
850 : src1_stride, mask, mask_stride, h, w,
851 : blend_8_b10);
852 0 : }
853 :
854 0 : static void blend_a64_mask_b12_sx_sy_w8n_sse4_1(
855 : uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
856 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
857 : const uint8_t *mask, uint32_t mask_stride, int h, int w) {
858 0 : blend_a64_mask_bn_sx_sy_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
859 : src1_stride, mask, mask_stride, h, w,
860 : blend_8_b12);
861 0 : }
862 :
863 : //////////////////////////////////////////////////////////////////////////////
864 : // Dispatch
865 : //////////////////////////////////////////////////////////////////////////////
866 :
867 0 : void aom_highbd_blend_a64_mask_sse4_1(uint8_t *dst_8, uint32_t dst_stride,
868 : const uint8_t *src0_8,
869 : uint32_t src0_stride,
870 : const uint8_t *src1_8,
871 : uint32_t src1_stride, const uint8_t *mask,
872 : uint32_t mask_stride, int h, int w,
873 : int suby, int subx, int bd) {
874 : typedef void (*blend_fn)(
875 : uint16_t * dst, uint32_t dst_stride, const uint16_t *src0,
876 : uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
877 : const uint8_t *mask, uint32_t mask_stride, int h, int w);
878 :
879 : // Dimensions are: bd_index X width_index X subx X suby
880 : static const blend_fn blend[2][2][2][2] = {
881 : { // bd == 8 or 10
882 : { // w % 8 == 0
883 : { blend_a64_mask_b10_w8n_sse4_1, blend_a64_mask_b10_sy_w8n_sse4_1 },
884 : { blend_a64_mask_b10_sx_w8n_sse4_1,
885 : blend_a64_mask_b10_sx_sy_w8n_sse4_1 } },
886 : { // w == 4
887 : { blend_a64_mask_b10_w4_sse4_1, blend_a64_mask_b10_sy_w4_sse4_1 },
888 : { blend_a64_mask_b10_sx_w4_sse4_1,
889 : blend_a64_mask_b10_sx_sy_w4_sse4_1 } } },
890 : { // bd == 12
891 : { // w % 8 == 0
892 : { blend_a64_mask_b12_w8n_sse4_1, blend_a64_mask_b12_sy_w8n_sse4_1 },
893 : { blend_a64_mask_b12_sx_w8n_sse4_1,
894 : blend_a64_mask_b12_sx_sy_w8n_sse4_1 } },
895 : { // w == 4
896 : { blend_a64_mask_b12_w4_sse4_1, blend_a64_mask_b12_sy_w4_sse4_1 },
897 : { blend_a64_mask_b12_sx_w4_sse4_1,
898 : blend_a64_mask_b12_sx_sy_w4_sse4_1 } } }
899 : };
900 :
901 0 : assert(IMPLIES(src0_8 == dst_8, src0_stride == dst_stride));
902 0 : assert(IMPLIES(src1_8 == dst_8, src1_stride == dst_stride));
903 :
904 0 : assert(h >= 1);
905 0 : assert(w >= 1);
906 0 : assert(IS_POWER_OF_TWO(h));
907 0 : assert(IS_POWER_OF_TWO(w));
908 :
909 0 : assert(bd == 8 || bd == 10 || bd == 12);
910 0 : if (UNLIKELY((h | w) & 3)) { // if (w <= 2 || h <= 2)
911 0 : aom_highbd_blend_a64_mask_c(dst_8, dst_stride, src0_8, src0_stride, src1_8,
912 : src1_stride, mask, mask_stride, h, w, suby,
913 : subx, bd);
914 : } else {
915 0 : uint16_t *const dst = CONVERT_TO_SHORTPTR(dst_8);
916 0 : const uint16_t *const src0 = CONVERT_TO_SHORTPTR(src0_8);
917 0 : const uint16_t *const src1 = CONVERT_TO_SHORTPTR(src1_8);
918 :
919 0 : blend[bd == 12][(w >> 2) & 1][subx != 0][suby != 0](
920 : dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
921 : mask_stride, h, w);
922 : }
923 0 : }
924 : #endif // CONFIG_HIGHBITDEPTH
|
