1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
|
/*
* Copyright (c) 2018, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <tmmintrin.h>
#include <assert.h>
#include "config/av1_rtcd.h"
#include "aom_dsp/x86/convolve_sse2.h"
#include "aom_dsp/x86/convolve_common_intrin.h"
void av1_highbd_convolve_y_sr_ssse3(const uint16_t *src, int src_stride,
uint16_t *dst, int dst_stride, int w, int h,
const InterpFilterParams *filter_params_y,
const int subpel_y_qn, int bd) {
int i, j;
const int fo_vert = filter_params_y->taps / 2 - 1;
const uint16_t *const src_ptr = src - fo_vert * src_stride;
const int bits = FILTER_BITS;
const __m128i round_shift_bits = _mm_cvtsi32_si128(bits);
const __m128i round_const_bits = _mm_set1_epi32((1 << bits) >> 1);
const __m128i clip_pixel =
_mm_set1_epi16(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
const __m128i zero = _mm_setzero_si128();
if (filter_params_y->taps == 12) {
__m128i s[24], coeffs_y[6];
prepare_coeffs_12tap(filter_params_y, subpel_y_qn, coeffs_y);
for (j = 0; j < w; j += 8) {
const uint16_t *data = &src_ptr[j];
/* Vertical filter */
__m128i s0 = _mm_loadu_si128((__m128i *)(data + 0 * src_stride));
__m128i s1 = _mm_loadu_si128((__m128i *)(data + 1 * src_stride));
__m128i s2 = _mm_loadu_si128((__m128i *)(data + 2 * src_stride));
__m128i s3 = _mm_loadu_si128((__m128i *)(data + 3 * src_stride));
__m128i s4 = _mm_loadu_si128((__m128i *)(data + 4 * src_stride));
__m128i s5 = _mm_loadu_si128((__m128i *)(data + 5 * src_stride));
__m128i s6 = _mm_loadu_si128((__m128i *)(data + 6 * src_stride));
__m128i s7 = _mm_loadu_si128((__m128i *)(data + 7 * src_stride));
__m128i s8 = _mm_loadu_si128((__m128i *)(data + 8 * src_stride));
__m128i s9 = _mm_loadu_si128((__m128i *)(data + 9 * src_stride));
__m128i s10 = _mm_loadu_si128((__m128i *)(data + 10 * src_stride));
s[0] = _mm_unpacklo_epi16(s0, s1);
s[1] = _mm_unpacklo_epi16(s2, s3);
s[2] = _mm_unpacklo_epi16(s4, s5);
s[3] = _mm_unpacklo_epi16(s6, s7);
s[4] = _mm_unpacklo_epi16(s8, s9);
s[6] = _mm_unpackhi_epi16(s0, s1);
s[7] = _mm_unpackhi_epi16(s2, s3);
s[8] = _mm_unpackhi_epi16(s4, s5);
s[9] = _mm_unpackhi_epi16(s6, s7);
s[10] = _mm_unpackhi_epi16(s8, s9);
s[12] = _mm_unpacklo_epi16(s1, s2);
s[13] = _mm_unpacklo_epi16(s3, s4);
s[14] = _mm_unpacklo_epi16(s5, s6);
s[15] = _mm_unpacklo_epi16(s7, s8);
s[16] = _mm_unpacklo_epi16(s9, s10);
s[18] = _mm_unpackhi_epi16(s1, s2);
s[19] = _mm_unpackhi_epi16(s3, s4);
s[20] = _mm_unpackhi_epi16(s5, s6);
s[21] = _mm_unpackhi_epi16(s7, s8);
s[22] = _mm_unpackhi_epi16(s9, s10);
for (i = 0; i < h; i += 2) {
data = &src_ptr[i * src_stride + j];
__m128i s11 = _mm_loadu_si128((__m128i *)(data + 11 * src_stride));
__m128i s12 = _mm_loadu_si128((__m128i *)(data + 12 * src_stride));
s[5] = _mm_unpacklo_epi16(s10, s11);
s[11] = _mm_unpackhi_epi16(s10, s11);
s[17] = _mm_unpacklo_epi16(s11, s12);
s[23] = _mm_unpackhi_epi16(s11, s12);
const __m128i res_a0 = convolve_12tap(s, coeffs_y);
__m128i res_a_round0 = _mm_sra_epi32(
_mm_add_epi32(res_a0, round_const_bits), round_shift_bits);
const __m128i res_a1 = convolve_12tap(s + 12, coeffs_y);
__m128i res_a_round1 = _mm_sra_epi32(
_mm_add_epi32(res_a1, round_const_bits), round_shift_bits);
if (w - j > 4) {
const __m128i res_b0 = convolve_12tap(s + 6, coeffs_y);
__m128i res_b_round0 = _mm_sra_epi32(
_mm_add_epi32(res_b0, round_const_bits), round_shift_bits);
const __m128i res_b1 = convolve_12tap(s + 18, coeffs_y);
__m128i res_b_round1 = _mm_sra_epi32(
_mm_add_epi32(res_b1, round_const_bits), round_shift_bits);
__m128i res_16bit0 = _mm_packs_epi32(res_a_round0, res_b_round0);
res_16bit0 = _mm_min_epi16(res_16bit0, clip_pixel);
res_16bit0 = _mm_max_epi16(res_16bit0, zero);
__m128i res_16bit1 = _mm_packs_epi32(res_a_round1, res_b_round1);
res_16bit1 = _mm_min_epi16(res_16bit1, clip_pixel);
res_16bit1 = _mm_max_epi16(res_16bit1, zero);
_mm_storeu_si128((__m128i *)&dst[i * dst_stride + j], res_16bit0);
_mm_storeu_si128((__m128i *)&dst[i * dst_stride + j + dst_stride],
res_16bit1);
} else if (w == 4) {
res_a_round0 = _mm_packs_epi32(res_a_round0, res_a_round0);
res_a_round0 = _mm_min_epi16(res_a_round0, clip_pixel);
res_a_round0 = _mm_max_epi16(res_a_round0, zero);
res_a_round1 = _mm_packs_epi32(res_a_round1, res_a_round1);
res_a_round1 = _mm_min_epi16(res_a_round1, clip_pixel);
res_a_round1 = _mm_max_epi16(res_a_round1, zero);
_mm_storel_epi64((__m128i *)&dst[i * dst_stride + j], res_a_round0);
_mm_storel_epi64((__m128i *)&dst[i * dst_stride + j + dst_stride],
res_a_round1);
} else {
res_a_round0 = _mm_packs_epi32(res_a_round0, res_a_round0);
res_a_round0 = _mm_min_epi16(res_a_round0, clip_pixel);
res_a_round0 = _mm_max_epi16(res_a_round0, zero);
res_a_round1 = _mm_packs_epi32(res_a_round1, res_a_round1);
res_a_round1 = _mm_min_epi16(res_a_round1, clip_pixel);
res_a_round1 = _mm_max_epi16(res_a_round1, zero);
*((int *)(&dst[i * dst_stride + j])) =
_mm_cvtsi128_si32(res_a_round0);
*((int *)(&dst[i * dst_stride + j + dst_stride])) =
_mm_cvtsi128_si32(res_a_round1);
}
s[0] = s[1];
s[1] = s[2];
s[2] = s[3];
s[3] = s[4];
s[4] = s[5];
s[6] = s[7];
s[7] = s[8];
s[8] = s[9];
s[9] = s[10];
s[10] = s[11];
s[12] = s[13];
s[13] = s[14];
s[14] = s[15];
s[15] = s[16];
s[16] = s[17];
s[18] = s[19];
s[19] = s[20];
s[20] = s[21];
s[21] = s[22];
s[22] = s[23];
s10 = s12;
}
}
} else {
__m128i s[16], coeffs_y[4];
prepare_coeffs(filter_params_y, subpel_y_qn, coeffs_y);
for (j = 0; j < w; j += 8) {
const uint16_t *data = &src_ptr[j];
/* Vertical filter */
{
__m128i s0 = _mm_loadu_si128((__m128i *)(data + 0 * src_stride));
__m128i s1 = _mm_loadu_si128((__m128i *)(data + 1 * src_stride));
__m128i s2 = _mm_loadu_si128((__m128i *)(data + 2 * src_stride));
__m128i s3 = _mm_loadu_si128((__m128i *)(data + 3 * src_stride));
__m128i s4 = _mm_loadu_si128((__m128i *)(data + 4 * src_stride));
__m128i s5 = _mm_loadu_si128((__m128i *)(data + 5 * src_stride));
__m128i s6 = _mm_loadu_si128((__m128i *)(data + 6 * src_stride));
s[0] = _mm_unpacklo_epi16(s0, s1);
s[1] = _mm_unpacklo_epi16(s2, s3);
s[2] = _mm_unpacklo_epi16(s4, s5);
s[4] = _mm_unpackhi_epi16(s0, s1);
s[5] = _mm_unpackhi_epi16(s2, s3);
s[6] = _mm_unpackhi_epi16(s4, s5);
s[0 + 8] = _mm_unpacklo_epi16(s1, s2);
s[1 + 8] = _mm_unpacklo_epi16(s3, s4);
s[2 + 8] = _mm_unpacklo_epi16(s5, s6);
s[4 + 8] = _mm_unpackhi_epi16(s1, s2);
s[5 + 8] = _mm_unpackhi_epi16(s3, s4);
s[6 + 8] = _mm_unpackhi_epi16(s5, s6);
for (i = 0; i < h; i += 2) {
data = &src_ptr[i * src_stride + j];
__m128i s7 = _mm_loadu_si128((__m128i *)(data + 7 * src_stride));
__m128i s8 = _mm_loadu_si128((__m128i *)(data + 8 * src_stride));
s[3] = _mm_unpacklo_epi16(s6, s7);
s[7] = _mm_unpackhi_epi16(s6, s7);
s[3 + 8] = _mm_unpacklo_epi16(s7, s8);
s[7 + 8] = _mm_unpackhi_epi16(s7, s8);
const __m128i res_a0 = convolve(s, coeffs_y);
__m128i res_a_round0 = _mm_sra_epi32(
_mm_add_epi32(res_a0, round_const_bits), round_shift_bits);
const __m128i res_a1 = convolve(s + 8, coeffs_y);
__m128i res_a_round1 = _mm_sra_epi32(
_mm_add_epi32(res_a1, round_const_bits), round_shift_bits);
if (w - j > 4) {
const __m128i res_b0 = convolve(s + 4, coeffs_y);
__m128i res_b_round0 = _mm_sra_epi32(
_mm_add_epi32(res_b0, round_const_bits), round_shift_bits);
const __m128i res_b1 = convolve(s + 4 + 8, coeffs_y);
__m128i res_b_round1 = _mm_sra_epi32(
_mm_add_epi32(res_b1, round_const_bits), round_shift_bits);
__m128i res_16bit0 = _mm_packs_epi32(res_a_round0, res_b_round0);
res_16bit0 = _mm_min_epi16(res_16bit0, clip_pixel);
res_16bit0 = _mm_max_epi16(res_16bit0, zero);
__m128i res_16bit1 = _mm_packs_epi32(res_a_round1, res_b_round1);
res_16bit1 = _mm_min_epi16(res_16bit1, clip_pixel);
res_16bit1 = _mm_max_epi16(res_16bit1, zero);
_mm_storeu_si128((__m128i *)&dst[i * dst_stride + j], res_16bit0);
_mm_storeu_si128((__m128i *)&dst[i * dst_stride + j + dst_stride],
res_16bit1);
} else if (w == 4) {
res_a_round0 = _mm_packs_epi32(res_a_round0, res_a_round0);
res_a_round0 = _mm_min_epi16(res_a_round0, clip_pixel);
res_a_round0 = _mm_max_epi16(res_a_round0, zero);
res_a_round1 = _mm_packs_epi32(res_a_round1, res_a_round1);
res_a_round1 = _mm_min_epi16(res_a_round1, clip_pixel);
res_a_round1 = _mm_max_epi16(res_a_round1, zero);
_mm_storel_epi64((__m128i *)&dst[i * dst_stride + j], res_a_round0);
_mm_storel_epi64((__m128i *)&dst[i * dst_stride + j + dst_stride],
res_a_round1);
} else {
res_a_round0 = _mm_packs_epi32(res_a_round0, res_a_round0);
res_a_round0 = _mm_min_epi16(res_a_round0, clip_pixel);
res_a_round0 = _mm_max_epi16(res_a_round0, zero);
res_a_round1 = _mm_packs_epi32(res_a_round1, res_a_round1);
res_a_round1 = _mm_min_epi16(res_a_round1, clip_pixel);
res_a_round1 = _mm_max_epi16(res_a_round1, zero);
*((int *)(&dst[i * dst_stride + j])) =
_mm_cvtsi128_si32(res_a_round0);
*((int *)(&dst[i * dst_stride + j + dst_stride])) =
_mm_cvtsi128_si32(res_a_round1);
}
s[0] = s[1];
s[1] = s[2];
s[2] = s[3];
s[4] = s[5];
s[5] = s[6];
s[6] = s[7];
s[0 + 8] = s[1 + 8];
s[1 + 8] = s[2 + 8];
s[2 + 8] = s[3 + 8];
s[4 + 8] = s[5 + 8];
s[5 + 8] = s[6 + 8];
s[6 + 8] = s[7 + 8];
s6 = s8;
}
}
}
}
}
void av1_highbd_convolve_x_sr_ssse3(const uint16_t *src, int src_stride,
uint16_t *dst, int dst_stride, int w, int h,
const InterpFilterParams *filter_params_x,
const int subpel_x_qn,
ConvolveParams *conv_params, int bd) {
int i, j;
const int fo_horiz = filter_params_x->taps / 2 - 1;
const uint16_t *const src_ptr = src - fo_horiz;
// Check that, even with 12-bit input, the intermediate values will fit
// into an unsigned 16-bit intermediate array.
assert(bd + FILTER_BITS + 2 - conv_params->round_0 <= 16);
const __m128i round_const_x =
_mm_set1_epi32(((1 << conv_params->round_0) >> 1));
const __m128i round_shift_x = _mm_cvtsi32_si128(conv_params->round_0);
const int bits = FILTER_BITS - conv_params->round_0;
const __m128i round_shift_bits = _mm_cvtsi32_si128(bits);
const __m128i round_const_bits = _mm_set1_epi32((1 << bits) >> 1);
const __m128i clip_pixel =
_mm_set1_epi16(bd == 10 ? 1023 : (bd == 12 ? 4095 : 255));
const __m128i zero = _mm_setzero_si128();
if (filter_params_x->taps == 12) {
__m128i s[6], coeffs_x[6];
prepare_coeffs_12tap(filter_params_x, subpel_x_qn, coeffs_x);
for (j = 0; j < w; j += 8) {
/* Horizontal filter */
{
for (i = 0; i < h; i += 1) {
const __m128i row00 =
_mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j]);
const __m128i row01 =
_mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + (j + 8)]);
const __m128i row02 =
_mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + (j + 16)]);
// even pixels
s[0] = _mm_alignr_epi8(row01, row00, 0);
s[1] = _mm_alignr_epi8(row01, row00, 4);
s[2] = _mm_alignr_epi8(row01, row00, 8);
s[3] = _mm_alignr_epi8(row01, row00, 12);
s[4] = _mm_alignr_epi8(row02, row01, 0);
s[5] = _mm_alignr_epi8(row02, row01, 4);
__m128i res_even = convolve_12tap(s, coeffs_x);
res_even = _mm_sra_epi32(_mm_add_epi32(res_even, round_const_x),
round_shift_x);
res_even = _mm_sra_epi32(_mm_add_epi32(res_even, round_const_bits),
round_shift_bits);
// odd pixels
s[0] = _mm_alignr_epi8(row01, row00, 2);
s[1] = _mm_alignr_epi8(row01, row00, 6);
s[2] = _mm_alignr_epi8(row01, row00, 10);
s[3] = _mm_alignr_epi8(row01, row00, 14);
s[4] = _mm_alignr_epi8(row02, row01, 2);
s[5] = _mm_alignr_epi8(row02, row01, 6);
__m128i res_odd = convolve_12tap(s, coeffs_x);
res_odd = _mm_sra_epi32(_mm_add_epi32(res_odd, round_const_x),
round_shift_x);
res_odd = _mm_sra_epi32(_mm_add_epi32(res_odd, round_const_bits),
round_shift_bits);
__m128i res_even1 = _mm_packs_epi32(res_even, res_even);
__m128i res_odd1 = _mm_packs_epi32(res_odd, res_odd);
__m128i res = _mm_unpacklo_epi16(res_even1, res_odd1);
res = _mm_min_epi16(res, clip_pixel);
res = _mm_max_epi16(res, zero);
if (w - j > 4) {
_mm_storeu_si128((__m128i *)&dst[i * dst_stride + j], res);
} else if (w == 4) {
_mm_storel_epi64((__m128i *)&dst[i * dst_stride + j], res);
} else {
*((int *)(&dst[i * dst_stride + j])) = _mm_cvtsi128_si32(res);
}
}
}
}
} else {
__m128i s[4], coeffs_x[4];
prepare_coeffs(filter_params_x, subpel_x_qn, coeffs_x);
for (j = 0; j < w; j += 8) {
/* Horizontal filter */
{
for (i = 0; i < h; i += 1) {
const __m128i row00 =
_mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j]);
const __m128i row01 =
_mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + (j + 8)]);
// even pixels
s[0] = _mm_alignr_epi8(row01, row00, 0);
s[1] = _mm_alignr_epi8(row01, row00, 4);
s[2] = _mm_alignr_epi8(row01, row00, 8);
s[3] = _mm_alignr_epi8(row01, row00, 12);
__m128i res_even = convolve(s, coeffs_x);
res_even = _mm_sra_epi32(_mm_add_epi32(res_even, round_const_x),
round_shift_x);
// odd pixels
s[0] = _mm_alignr_epi8(row01, row00, 2);
s[1] = _mm_alignr_epi8(row01, row00, 6);
s[2] = _mm_alignr_epi8(row01, row00, 10);
s[3] = _mm_alignr_epi8(row01, row00, 14);
__m128i res_odd = convolve(s, coeffs_x);
res_odd = _mm_sra_epi32(_mm_add_epi32(res_odd, round_const_x),
round_shift_x);
res_even = _mm_sra_epi32(_mm_add_epi32(res_even, round_const_bits),
round_shift_bits);
res_odd = _mm_sra_epi32(_mm_add_epi32(res_odd, round_const_bits),
round_shift_bits);
__m128i res_even1 = _mm_packs_epi32(res_even, res_even);
__m128i res_odd1 = _mm_packs_epi32(res_odd, res_odd);
__m128i res = _mm_unpacklo_epi16(res_even1, res_odd1);
res = _mm_min_epi16(res, clip_pixel);
res = _mm_max_epi16(res, zero);
if (w - j > 4) {
_mm_storeu_si128((__m128i *)&dst[i * dst_stride + j], res);
} else if (w == 4) {
_mm_storel_epi64((__m128i *)&dst[i * dst_stride + j], res);
} else {
*((int *)(&dst[i * dst_stride + j])) = _mm_cvtsi128_si32(res);
}
}
}
}
}
}
|