highbd_convolve_2d_ssse3.c 16 KB
Newer Older
David Barker's avatar
David Barker committed
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
/*
 * Copyright (c) 2016, 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 "./aom_dsp_rtcd.h"
#include "aom_dsp/aom_convolve.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/aom_filter.h"
#include "av1/common/convolve.h"

#if CONFIG_COMPOUND_ROUND
void av1_highbd_convolve_2d_ssse3(const uint16_t *src, int src_stride,
                                  CONV_BUF_TYPE *dst, int dst_stride, int w,
                                  int h, InterpFilterParams *filter_params_x,
                                  InterpFilterParams *filter_params_y,
                                  const int subpel_x_q4, const int subpel_y_q4,
                                  ConvolveParams *conv_params, int bd) {
  DECLARE_ALIGNED(16, int16_t,
                  im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE]);
  int im_h = h + filter_params_y->taps - 1;
  int im_stride = MAX_SB_SIZE;
  int i, j;
  const int fo_vert = filter_params_y->taps / 2 - 1;
  const int fo_horiz = filter_params_x->taps / 2 - 1;
  const uint16_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;

  /* Horizontal filter */
  {
    const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
        *filter_params_x, subpel_x_q4 & SUBPEL_MASK);
    const __m128i coeffs_x = _mm_loadu_si128((__m128i *)x_filter);

    // coeffs 0 1 0 1 2 3 2 3
    const __m128i tmp_0 = _mm_unpacklo_epi32(coeffs_x, coeffs_x);
    // coeffs 4 5 4 5 6 7 6 7
    const __m128i tmp_1 = _mm_unpackhi_epi32(coeffs_x, coeffs_x);

    // coeffs 0 1 0 1 0 1 0 1
    const __m128i coeff_01 = _mm_unpacklo_epi64(tmp_0, tmp_0);
    // coeffs 2 3 2 3 2 3 2 3
    const __m128i coeff_23 = _mm_unpackhi_epi64(tmp_0, tmp_0);
    // coeffs 4 5 4 5 4 5 4 5
    const __m128i coeff_45 = _mm_unpacklo_epi64(tmp_1, tmp_1);
    // coeffs 6 7 6 7 6 7 6 7
    const __m128i coeff_67 = _mm_unpackhi_epi64(tmp_1, tmp_1);

    const __m128i round_const =
        _mm_set1_epi32((1 << conv_params->round_0) >> 1);
    const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_0);

    for (i = 0; i < im_h; ++i) {
      for (j = 0; j < w; j += 8) {
        const __m128i data =
            _mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j]);
        const __m128i data2 =
            _mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j + 8]);

        // Filter even-index pixels
        const __m128i res_0 = _mm_madd_epi16(data, coeff_01);
        const __m128i res_2 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 4), coeff_23);
        const __m128i res_4 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 8), coeff_45);
        const __m128i res_6 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 12), coeff_67);

        __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_4),
                                         _mm_add_epi32(res_2, res_6));
        res_even =
            _mm_sra_epi32(_mm_add_epi32(res_even, round_const), round_shift);

        // Filter odd-index pixels
        const __m128i res_1 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 2), coeff_01);
        const __m128i res_3 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 6), coeff_23);
        const __m128i res_5 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 10), coeff_45);
        const __m128i res_7 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 14), coeff_67);

        __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_5),
                                        _mm_add_epi32(res_3, res_7));
        res_odd =
            _mm_sra_epi32(_mm_add_epi32(res_odd, round_const), round_shift);

        // Pack in the column order 0, 2, 4, 6, 1, 3, 5, 7
        const __m128i maxval = _mm_set1_epi16((1 << bd) - 1);
        __m128i res = _mm_packs_epi32(res_even, res_odd);
        res = _mm_max_epi16(_mm_min_epi16(res, maxval), _mm_setzero_si128());
        _mm_storeu_si128((__m128i *)&im_block[i * im_stride + j], res);
      }
    }
  }

  /* Vertical filter */
  {
    const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
        *filter_params_y, subpel_y_q4 & SUBPEL_MASK);
    const __m128i coeffs_y = _mm_loadu_si128((__m128i *)y_filter);

    // coeffs 0 1 0 1 2 3 2 3
    const __m128i tmp_0 = _mm_unpacklo_epi32(coeffs_y, coeffs_y);
    // coeffs 4 5 4 5 6 7 6 7
    const __m128i tmp_1 = _mm_unpackhi_epi32(coeffs_y, coeffs_y);

    // coeffs 0 1 0 1 0 1 0 1
    const __m128i coeff_01 = _mm_unpacklo_epi64(tmp_0, tmp_0);
    // coeffs 2 3 2 3 2 3 2 3
    const __m128i coeff_23 = _mm_unpackhi_epi64(tmp_0, tmp_0);
    // coeffs 4 5 4 5 4 5 4 5
    const __m128i coeff_45 = _mm_unpacklo_epi64(tmp_1, tmp_1);
    // coeffs 6 7 6 7 6 7 6 7
    const __m128i coeff_67 = _mm_unpackhi_epi64(tmp_1, tmp_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);

    for (i = 0; i < h; ++i) {
      for (j = 0; j < w; j += 8) {
        // Filter even-index pixels
        const int16_t *data = &im_block[i * im_stride + j];
        const __m128i src_0 =
            _mm_unpacklo_epi16(*(__m128i *)(data + 0 * im_stride),
                               *(__m128i *)(data + 1 * im_stride));
        const __m128i src_2 =
            _mm_unpacklo_epi16(*(__m128i *)(data + 2 * im_stride),
                               *(__m128i *)(data + 3 * im_stride));
        const __m128i src_4 =
            _mm_unpacklo_epi16(*(__m128i *)(data + 4 * im_stride),
                               *(__m128i *)(data + 5 * im_stride));
        const __m128i src_6 =
            _mm_unpacklo_epi16(*(__m128i *)(data + 6 * im_stride),
                               *(__m128i *)(data + 7 * im_stride));

        const __m128i res_0 = _mm_madd_epi16(src_0, coeff_01);
        const __m128i res_2 = _mm_madd_epi16(src_2, coeff_23);
        const __m128i res_4 = _mm_madd_epi16(src_4, coeff_45);
        const __m128i res_6 = _mm_madd_epi16(src_6, coeff_67);

        const __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
                                               _mm_add_epi32(res_4, res_6));

        // Filter odd-index pixels
        const __m128i src_1 =
            _mm_unpackhi_epi16(*(__m128i *)(data + 0 * im_stride),
                               *(__m128i *)(data + 1 * im_stride));
        const __m128i src_3 =
            _mm_unpackhi_epi16(*(__m128i *)(data + 2 * im_stride),
                               *(__m128i *)(data + 3 * im_stride));
        const __m128i src_5 =
            _mm_unpackhi_epi16(*(__m128i *)(data + 4 * im_stride),
                               *(__m128i *)(data + 5 * im_stride));
        const __m128i src_7 =
            _mm_unpackhi_epi16(*(__m128i *)(data + 6 * im_stride),
                               *(__m128i *)(data + 7 * im_stride));

        const __m128i res_1 = _mm_madd_epi16(src_1, coeff_01);
        const __m128i res_3 = _mm_madd_epi16(src_3, coeff_23);
        const __m128i res_5 = _mm_madd_epi16(src_5, coeff_45);
        const __m128i res_7 = _mm_madd_epi16(src_7, coeff_67);

        const __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
                                              _mm_add_epi32(res_5, res_7));

        // 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);

        // Accumulate values into the destination buffer
        __m128i *const p = (__m128i *)&dst[i * dst_stride + j];
        _mm_storeu_si128(p, _mm_add_epi32(_mm_loadu_si128(p), res_lo_round));
        _mm_storeu_si128(p + 1,
                         _mm_add_epi32(_mm_loadu_si128(p + 1), res_hi_round));
      }
    }
  }
}
#else
void av1_highbd_convolve_2d_ssse3(const uint16_t *src, int src_stride,
                                  CONV_BUF_TYPE *dst, int dst_stride, int w,
                                  int h, InterpFilterParams *filter_params_x,
                                  InterpFilterParams *filter_params_y,
                                  const int subpel_x_q4, const int subpel_y_q4,
                                  ConvolveParams *conv_params, int bd) {
  DECLARE_ALIGNED(16, int16_t,
                  im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE]);
  int im_h = h + filter_params_y->taps - 1;
  int im_stride = MAX_SB_SIZE;
  int i, j;
  const int fo_vert = filter_params_y->taps / 2 - 1;
  const int fo_horiz = filter_params_x->taps / 2 - 1;
  const uint16_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;

  // Check that, even with 12-bit input, the intermediate values will fit
  // into an unsigned 15-bit intermediate array.
  assert(conv_params->round_0 >= 5);

  /* Horizontal filter */
  {
    const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
        *filter_params_x, subpel_x_q4 & SUBPEL_MASK);
    const __m128i coeffs_x = _mm_loadu_si128((__m128i *)x_filter);

    // coeffs 0 1 0 1 2 3 2 3
    const __m128i tmp_0 = _mm_unpacklo_epi32(coeffs_x, coeffs_x);
    // coeffs 4 5 4 5 6 7 6 7
    const __m128i tmp_1 = _mm_unpackhi_epi32(coeffs_x, coeffs_x);

    // coeffs 0 1 0 1 0 1 0 1
    const __m128i coeff_01 = _mm_unpacklo_epi64(tmp_0, tmp_0);
    // coeffs 2 3 2 3 2 3 2 3
    const __m128i coeff_23 = _mm_unpackhi_epi64(tmp_0, tmp_0);
    // coeffs 4 5 4 5 4 5 4 5
    const __m128i coeff_45 = _mm_unpacklo_epi64(tmp_1, tmp_1);
    // coeffs 6 7 6 7 6 7 6 7
    const __m128i coeff_67 = _mm_unpackhi_epi64(tmp_1, tmp_1);

    const __m128i round_const = _mm_set1_epi32(
        ((1 << conv_params->round_0) >> 1) + (1 << (bd + FILTER_BITS - 1)));
    const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_0);

    for (i = 0; i < im_h; ++i) {
      for (j = 0; j < w; j += 8) {
        const __m128i data =
            _mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j]);
        const __m128i data2 =
            _mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j + 8]);

        // Filter even-index pixels
        const __m128i res_0 = _mm_madd_epi16(data, coeff_01);
        const __m128i res_2 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 4), coeff_23);
        const __m128i res_4 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 8), coeff_45);
        const __m128i res_6 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 12), coeff_67);

        __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_4),
                                         _mm_add_epi32(res_2, res_6));
        res_even =
            _mm_sra_epi32(_mm_add_epi32(res_even, round_const), round_shift);

        // Filter odd-index pixels
        const __m128i res_1 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 2), coeff_01);
        const __m128i res_3 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 6), coeff_23);
        const __m128i res_5 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 10), coeff_45);
        const __m128i res_7 =
            _mm_madd_epi16(_mm_alignr_epi8(data2, data, 14), coeff_67);

        __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_5),
                                        _mm_add_epi32(res_3, res_7));
        res_odd =
            _mm_sra_epi32(_mm_add_epi32(res_odd, round_const), round_shift);

        // Pack in the column order 0, 2, 4, 6, 1, 3, 5, 7
        __m128i res = _mm_packs_epi32(res_even, res_odd);
        _mm_storeu_si128((__m128i *)&im_block[i * im_stride + j], res);
      }
    }
  }

  /* Vertical filter */
  {
    const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
        *filter_params_y, subpel_y_q4 & SUBPEL_MASK);
    const __m128i coeffs_y = _mm_loadu_si128((__m128i *)y_filter);

    // coeffs 0 1 0 1 2 3 2 3
    const __m128i tmp_0 = _mm_unpacklo_epi32(coeffs_y, coeffs_y);
    // coeffs 4 5 4 5 6 7 6 7
    const __m128i tmp_1 = _mm_unpackhi_epi32(coeffs_y, coeffs_y);

    // coeffs 0 1 0 1 0 1 0 1
    const __m128i coeff_01 = _mm_unpacklo_epi64(tmp_0, tmp_0);
    // coeffs 2 3 2 3 2 3 2 3
    const __m128i coeff_23 = _mm_unpackhi_epi64(tmp_0, tmp_0);
    // coeffs 4 5 4 5 4 5 4 5
    const __m128i coeff_45 = _mm_unpacklo_epi64(tmp_1, tmp_1);
    // coeffs 6 7 6 7 6 7 6 7
    const __m128i coeff_67 = _mm_unpackhi_epi64(tmp_1, tmp_1);

    const __m128i round_const = _mm_set1_epi32(
        ((1 << conv_params->round_1) >> 1) -
        (1 << (bd + 2 * FILTER_BITS - conv_params->round_0 - 1)));
    const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_1);

    for (i = 0; i < h; ++i) {
      for (j = 0; j < w; j += 8) {
        // Filter even-index pixels
        const int16_t *data = &im_block[i * im_stride + j];
        const __m128i src_0 =
            _mm_unpacklo_epi16(*(__m128i *)(data + 0 * im_stride),
                               *(__m128i *)(data + 1 * im_stride));
        const __m128i src_2 =
            _mm_unpacklo_epi16(*(__m128i *)(data + 2 * im_stride),
                               *(__m128i *)(data + 3 * im_stride));
        const __m128i src_4 =
            _mm_unpacklo_epi16(*(__m128i *)(data + 4 * im_stride),
                               *(__m128i *)(data + 5 * im_stride));
        const __m128i src_6 =
            _mm_unpacklo_epi16(*(__m128i *)(data + 6 * im_stride),
                               *(__m128i *)(data + 7 * im_stride));

        const __m128i res_0 = _mm_madd_epi16(src_0, coeff_01);
        const __m128i res_2 = _mm_madd_epi16(src_2, coeff_23);
        const __m128i res_4 = _mm_madd_epi16(src_4, coeff_45);
        const __m128i res_6 = _mm_madd_epi16(src_6, coeff_67);

        const __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
                                               _mm_add_epi32(res_4, res_6));

        // Filter odd-index pixels
        const __m128i src_1 =
            _mm_unpackhi_epi16(*(__m128i *)(data + 0 * im_stride),
                               *(__m128i *)(data + 1 * im_stride));
        const __m128i src_3 =
            _mm_unpackhi_epi16(*(__m128i *)(data + 2 * im_stride),
                               *(__m128i *)(data + 3 * im_stride));
        const __m128i src_5 =
            _mm_unpackhi_epi16(*(__m128i *)(data + 4 * im_stride),
                               *(__m128i *)(data + 5 * im_stride));
        const __m128i src_7 =
            _mm_unpackhi_epi16(*(__m128i *)(data + 6 * im_stride),
                               *(__m128i *)(data + 7 * im_stride));

        const __m128i res_1 = _mm_madd_epi16(src_1, coeff_01);
        const __m128i res_3 = _mm_madd_epi16(src_3, coeff_23);
        const __m128i res_5 = _mm_madd_epi16(src_5, coeff_45);
        const __m128i res_7 = _mm_madd_epi16(src_7, coeff_67);

        const __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
                                              _mm_add_epi32(res_5, res_7));

        // 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);

        // Accumulate values into the destination buffer
        __m128i *const p = (__m128i *)&dst[i * dst_stride + j];
        _mm_storeu_si128(p, _mm_add_epi32(_mm_loadu_si128(p), res_lo_round));
        _mm_storeu_si128(p + 1,
                         _mm_add_epi32(_mm_loadu_si128(p + 1), res_hi_round));
      }
    }
  }
}
#endif