summaryrefslogtreecommitdiffstats
path: root/third_party/aom/av1/common/x86/highbd_convolve_2d_ssse3.c
blob: 1d029db39a52d169bec754e0b2ef71ff425fae1a (plain)
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
/*
 * 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 "config/aom_dsp_rtcd.h"

#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/aom_filter.h"
#include "aom_dsp/x86/convolve_sse2.h"
#include "av1/common/convolve.h"

void av1_highbd_convolve_2d_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 InterpFilterParams *filter_params_y, const int subpel_x_q4,
    const int subpel_y_q4, ConvolveParams *conv_params, int bd) {
  DECLARE_ALIGNED(32, int16_t, im_block[(MAX_SB_SIZE + MAX_FILTER_TAP) * 8]);
  int im_h = h + filter_params_y->taps - 1;
  int im_stride = 8;
  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 16-bit intermediate array.
  assert(bd + FILTER_BITS + 2 - conv_params->round_0 <= 16);
  __m128i coeffs_x[4], coeffs_y[4], s[16];

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

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

  const int bits =
      FILTER_BITS * 2 - conv_params->round_0 - conv_params->round_1;
  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();

  prepare_coeffs(filter_params_x, subpel_x_q4, coeffs_x);
  prepare_coeffs(filter_params_y, subpel_y_q4, coeffs_y);

  for (j = 0; j < w; j += 8) {
    /* Horizontal filter */
    {
      for (i = 0; i < im_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);

        __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);

        _mm_store_si128((__m128i *)&im_block[i * im_stride], res);
      }
    }
    /* Vertical filter */
    {
      __m128i s0 = _mm_loadu_si128((__m128i *)(im_block + 0 * im_stride));
      __m128i s1 = _mm_loadu_si128((__m128i *)(im_block + 1 * im_stride));
      __m128i s2 = _mm_loadu_si128((__m128i *)(im_block + 2 * im_stride));
      __m128i s3 = _mm_loadu_si128((__m128i *)(im_block + 3 * im_stride));
      __m128i s4 = _mm_loadu_si128((__m128i *)(im_block + 4 * im_stride));
      __m128i s5 = _mm_loadu_si128((__m128i *)(im_block + 5 * im_stride));
      __m128i s6 = _mm_loadu_si128((__m128i *)(im_block + 6 * im_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) {
        const int16_t *data = &im_block[i * im_stride];

        __m128i s7 = _mm_loadu_si128((__m128i *)(data + 7 * im_stride));
        __m128i s8 = _mm_loadu_si128((__m128i *)(data + 8 * im_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_y), round_shift_y);
        res_a_round0 = _mm_sra_epi32(
            _mm_add_epi32(res_a_round0, 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_y), round_shift_y);
        res_a_round1 = _mm_sra_epi32(
            _mm_add_epi32(res_a_round1, 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_y), round_shift_y);
          res_b_round0 = _mm_sra_epi32(
              _mm_add_epi32(res_b_round0, 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_y), round_shift_y);
          res_b_round1 = _mm_sra_epi32(
              _mm_add_epi32(res_b_round1, 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);

          *((uint32_t *)(&dst[i * dst_stride + j])) =
              _mm_cvtsi128_si32(res_a_round0);

          *((uint32_t *)(&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;
      }
    }
  }
}