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
|
/*
* 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 <emmintrin.h>
#include "config/aom_dsp_rtcd.h"
#include "aom_dsp/aom_filter.h"
#include "aom_dsp/x86/convolve_sse2.h"
void av1_jnt_convolve_x_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
int dst_stride0, int w, int h,
const InterpFilterParams *filter_params_x,
const InterpFilterParams *filter_params_y,
const int subpel_x_q4, const int subpel_y_q4,
ConvolveParams *conv_params) {
const int bd = 8;
CONV_BUF_TYPE *dst = conv_params->dst;
const int dst_stride = conv_params->dst_stride;
const int fo_horiz = filter_params_x->taps / 2 - 1;
const uint8_t *src_ptr = src - fo_horiz;
const int bits = FILTER_BITS - conv_params->round_1;
const __m128i left_shift = _mm_cvtsi32_si128(bits);
const __m128i round_const = _mm_set1_epi32((1 << conv_params->round_0) >> 1);
const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_0);
const int w0 = conv_params->fwd_offset;
const int w1 = conv_params->bck_offset;
const __m128i wt0 = _mm_set1_epi16(w0);
const __m128i wt1 = _mm_set1_epi16(w1);
const __m128i wt = _mm_unpacklo_epi16(wt0, wt1);
const int do_average = conv_params->do_average;
const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
const int offset_0 =
bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
const __m128i offset_const = _mm_set1_epi16(offset);
const int rounding_shift =
2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
const __m128i rounding_const = _mm_set1_epi16((1 << rounding_shift) >> 1);
__m128i coeffs[4];
(void)filter_params_y;
(void)subpel_y_q4;
prepare_coeffs(filter_params_x, subpel_x_q4, coeffs);
if (w == 4) {
do {
const __m128i data = _mm_loadu_si128((__m128i *)src_ptr);
__m128i s[4];
s[0] = _mm_unpacklo_epi8(data, _mm_srli_si128(data, 1));
s[1] =
_mm_unpacklo_epi8(_mm_srli_si128(data, 2), _mm_srli_si128(data, 3));
s[2] =
_mm_unpacklo_epi8(_mm_srli_si128(data, 4), _mm_srli_si128(data, 5));
s[3] =
_mm_unpacklo_epi8(_mm_srli_si128(data, 6), _mm_srli_si128(data, 7));
const __m128i res_lo = convolve_lo_x(s, coeffs);
const __m128i res_lo_round =
_mm_sra_epi32(_mm_add_epi32(res_lo, round_const), round_shift);
const __m128i res_lo_shift = _mm_sll_epi32(res_lo_round, left_shift);
const __m128i res_16b = _mm_packs_epi32(res_lo_shift, res_lo_shift);
const __m128i res_unsigned = _mm_add_epi16(res_16b, offset_const);
// Accumulate values into the destination buffer
if (do_average) {
const __m128i data_ref_0 = _mm_loadu_si128((__m128i *)dst);
const __m128i comp_avg_res =
comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
const __m128i round_result = convolve_rounding(
&comp_avg_res, &offset_const, &rounding_const, rounding_shift);
const __m128i res_8 = _mm_packus_epi16(round_result, round_result);
*(uint32_t *)(&dst0[0]) = _mm_cvtsi128_si32(res_8);
} else {
_mm_store_si128((__m128i *)(&dst[0]), res_unsigned);
}
src_ptr += src_stride;
dst += dst_stride;
dst0 += dst_stride0;
} while (--h);
} else {
assert(!(w % 8));
int i = 0;
do {
int j = 0;
do {
const __m128i data =
_mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j]);
__m128i s[4];
// Filter even-index pixels
s[0] = data;
s[1] = _mm_srli_si128(data, 2);
s[2] = _mm_srli_si128(data, 4);
s[3] = _mm_srli_si128(data, 6);
const __m128i res_even = convolve_lo_x(s, coeffs);
// Filter odd-index pixels
s[0] = _mm_srli_si128(data, 1);
s[1] = _mm_srli_si128(data, 3);
s[2] = _mm_srli_si128(data, 5);
s[3] = _mm_srli_si128(data, 7);
const __m128i res_odd = convolve_lo_x(s, coeffs);
// Rearrange pixels back into the order 0 ... 7
const __m128i res_lo = _mm_unpacklo_epi32(res_even, res_odd);
const __m128i res_hi = _mm_unpackhi_epi32(res_even, res_odd);
const __m128i res_lo_round =
_mm_sra_epi32(_mm_add_epi32(res_lo, round_const), round_shift);
const __m128i res_hi_round =
_mm_sra_epi32(_mm_add_epi32(res_hi, round_const), round_shift);
const __m128i res_lo_shift = _mm_sll_epi32(res_lo_round, left_shift);
const __m128i res_hi_shift = _mm_sll_epi32(res_hi_round, left_shift);
const __m128i res_16b = _mm_packs_epi32(res_lo_shift, res_hi_shift);
const __m128i res_unsigned = _mm_add_epi16(res_16b, offset_const);
// Accumulate values into the destination buffer
if (do_average) {
const __m128i data_ref_0 =
_mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
const __m128i comp_avg_res =
comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
const __m128i round_result = convolve_rounding(
&comp_avg_res, &offset_const, &rounding_const, rounding_shift);
const __m128i res_8 = _mm_packus_epi16(round_result, round_result);
_mm_storel_epi64((__m128i *)(&dst0[i * dst_stride0 + j]), res_8);
} else {
_mm_store_si128((__m128i *)(&dst[i * dst_stride + j]), res_unsigned);
}
j += 8;
} while (j < w);
} while (++i < h);
}
}
void av1_jnt_convolve_y_sse2(const uint8_t *src, int src_stride, uint8_t *dst0,
int dst_stride0, int w, int h,
const InterpFilterParams *filter_params_x,
const InterpFilterParams *filter_params_y,
const int subpel_x_q4, const int subpel_y_q4,
ConvolveParams *conv_params) {
const int bd = 8;
CONV_BUF_TYPE *dst = conv_params->dst;
const int dst_stride = conv_params->dst_stride;
const int fo_vert = filter_params_y->taps / 2 - 1;
const uint8_t *src_ptr = src - fo_vert * src_stride;
const int bits = FILTER_BITS - conv_params->round_0;
const __m128i left_shift = _mm_cvtsi32_si128(bits);
const __m128i wt0 = _mm_set1_epi16(conv_params->fwd_offset);
const __m128i wt1 = _mm_set1_epi16(conv_params->bck_offset);
const __m128i wt = _mm_unpacklo_epi16(wt0, wt1);
const int do_average = conv_params->do_average;
const int use_jnt_comp_avg = conv_params->use_jnt_comp_avg;
const int offset_0 =
bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
const __m128i offset_const = _mm_set1_epi16(offset);
const int rounding_shift =
2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
const __m128i rounding_const = _mm_set1_epi16((1 << rounding_shift) >> 1);
const __m128i round_const = _mm_set1_epi32((1 << conv_params->round_1) >> 1);
const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_1);
__m128i coeffs[4];
(void)filter_params_x;
(void)subpel_x_q4;
prepare_coeffs(filter_params_y, subpel_y_q4, coeffs);
if (w == 4) {
__m128i s[8], src6, res, res_shift;
src6 = _mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 6 * src_stride));
s[0] = _mm_unpacklo_epi8(
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 0 * src_stride)),
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 1 * src_stride)));
s[1] = _mm_unpacklo_epi8(
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 1 * src_stride)),
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 2 * src_stride)));
s[2] = _mm_unpacklo_epi8(
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 2 * src_stride)),
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 3 * src_stride)));
s[3] = _mm_unpacklo_epi8(
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 3 * src_stride)),
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 4 * src_stride)));
s[4] = _mm_unpacklo_epi8(
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 4 * src_stride)),
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 5 * src_stride)));
s[5] = _mm_unpacklo_epi8(
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 5 * src_stride)), src6);
do {
s[6] = _mm_unpacklo_epi8(
src6, _mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 7 * src_stride)));
src6 = _mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 8 * src_stride));
s[7] = _mm_unpacklo_epi8(
_mm_cvtsi32_si128(*(uint32_t *)(src_ptr + 7 * src_stride)), src6);
res = convolve_lo_y(s + 0, coeffs);
res_shift = _mm_sll_epi32(res, left_shift);
res_shift =
_mm_sra_epi32(_mm_add_epi32(res_shift, round_const), round_shift);
__m128i res_16b = _mm_packs_epi32(res_shift, res_shift);
__m128i res_unsigned = _mm_add_epi16(res_16b, offset_const);
// Accumulate values into the destination buffer
if (do_average) {
const __m128i data_ref_0 = _mm_loadu_si128((__m128i *)dst);
const __m128i comp_avg_res =
comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
const __m128i round_result = convolve_rounding(
&comp_avg_res, &offset_const, &rounding_const, rounding_shift);
const __m128i res_8 = _mm_packus_epi16(round_result, round_result);
*(uint32_t *)(&dst0[0]) = _mm_cvtsi128_si32(res_8);
} else {
_mm_store_si128((__m128i *)dst, res_unsigned);
}
src_ptr += src_stride;
dst += dst_stride;
dst0 += dst_stride0;
res = convolve_lo_y(s + 1, coeffs);
res_shift = _mm_sll_epi32(res, left_shift);
res_shift =
_mm_sra_epi32(_mm_add_epi32(res_shift, round_const), round_shift);
res_16b = _mm_packs_epi32(res_shift, res_shift);
res_unsigned = _mm_add_epi16(res_16b, offset_const);
// Accumulate values into the destination buffer
if (do_average) {
const __m128i data_ref_0 = _mm_loadu_si128((__m128i *)dst);
const __m128i comp_avg_res =
comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
const __m128i round_result = convolve_rounding(
&comp_avg_res, &offset_const, &rounding_const, rounding_shift);
const __m128i res_8 = _mm_packus_epi16(round_result, round_result);
*(uint32_t *)(&dst0[0]) = _mm_cvtsi128_si32(res_8);
} else {
_mm_store_si128((__m128i *)dst, res_unsigned);
}
src_ptr += src_stride;
dst += dst_stride;
dst0 += dst_stride0;
s[0] = s[2];
s[1] = s[3];
s[2] = s[4];
s[3] = s[5];
s[4] = s[6];
s[5] = s[7];
h -= 2;
} while (h);
} else {
assert(!(w % 8));
int j = 0;
do {
__m128i s[8], src6, res_lo, res_hi, res_lo_shift, res_hi_shift;
const uint8_t *data = &src_ptr[j];
src6 = _mm_loadl_epi64((__m128i *)(data + 6 * src_stride));
s[0] = _mm_unpacklo_epi8(
_mm_loadl_epi64((__m128i *)(data + 0 * src_stride)),
_mm_loadl_epi64((__m128i *)(data + 1 * src_stride)));
s[1] = _mm_unpacklo_epi8(
_mm_loadl_epi64((__m128i *)(data + 1 * src_stride)),
_mm_loadl_epi64((__m128i *)(data + 2 * src_stride)));
s[2] = _mm_unpacklo_epi8(
_mm_loadl_epi64((__m128i *)(data + 2 * src_stride)),
_mm_loadl_epi64((__m128i *)(data + 3 * src_stride)));
s[3] = _mm_unpacklo_epi8(
_mm_loadl_epi64((__m128i *)(data + 3 * src_stride)),
_mm_loadl_epi64((__m128i *)(data + 4 * src_stride)));
s[4] = _mm_unpacklo_epi8(
_mm_loadl_epi64((__m128i *)(data + 4 * src_stride)),
_mm_loadl_epi64((__m128i *)(data + 5 * src_stride)));
s[5] = _mm_unpacklo_epi8(
_mm_loadl_epi64((__m128i *)(data + 5 * src_stride)), src6);
int i = 0;
do {
data = &src_ptr[i * src_stride + j];
s[6] = _mm_unpacklo_epi8(
src6, _mm_loadl_epi64((__m128i *)(data + 7 * src_stride)));
src6 = _mm_loadl_epi64((__m128i *)(data + 8 * src_stride));
s[7] = _mm_unpacklo_epi8(
_mm_loadl_epi64((__m128i *)(data + 7 * src_stride)), src6);
res_lo = convolve_lo_y(s, coeffs); // Filter low index pixels
res_hi = convolve_hi_y(s, coeffs); // Filter high index pixels
res_lo_shift = _mm_sll_epi32(res_lo, left_shift);
res_hi_shift = _mm_sll_epi32(res_hi, left_shift);
res_lo_shift = _mm_sra_epi32(_mm_add_epi32(res_lo_shift, round_const),
round_shift);
res_hi_shift = _mm_sra_epi32(_mm_add_epi32(res_hi_shift, round_const),
round_shift);
__m128i res_16b = _mm_packs_epi32(res_lo_shift, res_hi_shift);
__m128i res_unsigned = _mm_add_epi16(res_16b, offset_const);
// Accumulate values into the destination buffer
if (do_average) {
const __m128i data_ref_0 =
_mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
const __m128i comp_avg_res =
comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
const __m128i round_result = convolve_rounding(
&comp_avg_res, &offset_const, &rounding_const, rounding_shift);
const __m128i res_8 = _mm_packus_epi16(round_result, round_result);
_mm_storel_epi64((__m128i *)(&dst0[i * dst_stride0 + j]), res_8);
} else {
_mm_store_si128((__m128i *)(&dst[i * dst_stride + j]), res_unsigned);
}
i++;
res_lo = convolve_lo_y(s + 1, coeffs); // Filter low index pixels
res_hi = convolve_hi_y(s + 1, coeffs); // Filter high index pixels
res_lo_shift = _mm_sll_epi32(res_lo, left_shift);
res_hi_shift = _mm_sll_epi32(res_hi, left_shift);
res_lo_shift = _mm_sra_epi32(_mm_add_epi32(res_lo_shift, round_const),
round_shift);
res_hi_shift = _mm_sra_epi32(_mm_add_epi32(res_hi_shift, round_const),
round_shift);
res_16b = _mm_packs_epi32(res_lo_shift, res_hi_shift);
res_unsigned = _mm_add_epi16(res_16b, offset_const);
// Accumulate values into the destination buffer
if (do_average) {
__m128i data_ref_0 =
_mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
const __m128i comp_avg_res =
comp_avg(&data_ref_0, &res_unsigned, &wt, use_jnt_comp_avg);
const __m128i round_result = convolve_rounding(
&comp_avg_res, &offset_const, &rounding_const, rounding_shift);
const __m128i res_8 = _mm_packus_epi16(round_result, round_result);
_mm_storel_epi64((__m128i *)(&dst0[i * dst_stride0 + j]), res_8);
} else {
_mm_store_si128((__m128i *)(&dst[i * dst_stride + j]), res_unsigned);
}
i++;
s[0] = s[2];
s[1] = s[3];
s[2] = s[4];
s[3] = s[5];
s[4] = s[6];
s[5] = s[7];
} while (i < h);
j += 8;
} while (j < w);
}
}
|
