summaryrefslogtreecommitdiffstats
path: root/media/libvpx/vp9/common/x86/vp9_idct_intrin_sse2.h
blob: 984363d40359d6a3c6e07b93c39c9b2f4a546821 (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
/*
 *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include <assert.h>
#include <emmintrin.h>  // SSE2
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_idct.h"

// perform 8x8 transpose
static INLINE void array_transpose_8x8(__m128i *in, __m128i *res) {
  const __m128i tr0_0 = _mm_unpacklo_epi16(in[0], in[1]);
  const __m128i tr0_1 = _mm_unpacklo_epi16(in[2], in[3]);
  const __m128i tr0_2 = _mm_unpackhi_epi16(in[0], in[1]);
  const __m128i tr0_3 = _mm_unpackhi_epi16(in[2], in[3]);
  const __m128i tr0_4 = _mm_unpacklo_epi16(in[4], in[5]);
  const __m128i tr0_5 = _mm_unpacklo_epi16(in[6], in[7]);
  const __m128i tr0_6 = _mm_unpackhi_epi16(in[4], in[5]);
  const __m128i tr0_7 = _mm_unpackhi_epi16(in[6], in[7]);

  const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
  const __m128i tr1_1 = _mm_unpacklo_epi32(tr0_4, tr0_5);
  const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
  const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_4, tr0_5);
  const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_2, tr0_3);
  const __m128i tr1_5 = _mm_unpacklo_epi32(tr0_6, tr0_7);
  const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_2, tr0_3);
  const __m128i tr1_7 = _mm_unpackhi_epi32(tr0_6, tr0_7);

  res[0] = _mm_unpacklo_epi64(tr1_0, tr1_1);
  res[1] = _mm_unpackhi_epi64(tr1_0, tr1_1);
  res[2] = _mm_unpacklo_epi64(tr1_2, tr1_3);
  res[3] = _mm_unpackhi_epi64(tr1_2, tr1_3);
  res[4] = _mm_unpacklo_epi64(tr1_4, tr1_5);
  res[5] = _mm_unpackhi_epi64(tr1_4, tr1_5);
  res[6] = _mm_unpacklo_epi64(tr1_6, tr1_7);
  res[7] = _mm_unpackhi_epi64(tr1_6, tr1_7);
}

#define TRANSPOSE_8X4(in0, in1, in2, in3, out0, out1) \
  {                                                     \
    const __m128i tr0_0 = _mm_unpacklo_epi16(in0, in1); \
    const __m128i tr0_1 = _mm_unpacklo_epi16(in2, in3); \
                                                        \
    in0 = _mm_unpacklo_epi32(tr0_0, tr0_1);  /* i1 i0 */  \
    in1 = _mm_unpackhi_epi32(tr0_0, tr0_1);  /* i3 i2 */  \
  }

static INLINE void array_transpose_4X8(__m128i *in, __m128i * out) {
  const __m128i tr0_0 = _mm_unpacklo_epi16(in[0], in[1]);
  const __m128i tr0_1 = _mm_unpacklo_epi16(in[2], in[3]);
  const __m128i tr0_4 = _mm_unpacklo_epi16(in[4], in[5]);
  const __m128i tr0_5 = _mm_unpacklo_epi16(in[6], in[7]);

  const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
  const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
  const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5);
  const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5);

  out[0] = _mm_unpacklo_epi64(tr1_0, tr1_4);
  out[1] = _mm_unpackhi_epi64(tr1_0, tr1_4);
  out[2] = _mm_unpacklo_epi64(tr1_2, tr1_6);
  out[3] = _mm_unpackhi_epi64(tr1_2, tr1_6);
}

static INLINE void array_transpose_16x16(__m128i *res0, __m128i *res1) {
  __m128i tbuf[8];
  array_transpose_8x8(res0, res0);
  array_transpose_8x8(res1, tbuf);
  array_transpose_8x8(res0 + 8, res1);
  array_transpose_8x8(res1 + 8, res1 + 8);

  res0[8] = tbuf[0];
  res0[9] = tbuf[1];
  res0[10] = tbuf[2];
  res0[11] = tbuf[3];
  res0[12] = tbuf[4];
  res0[13] = tbuf[5];
  res0[14] = tbuf[6];
  res0[15] = tbuf[7];
}

static INLINE void load_buffer_8x16(const int16_t *input, __m128i *in) {
  in[0]  = _mm_load_si128((const __m128i *)(input + 0 * 16));
  in[1]  = _mm_load_si128((const __m128i *)(input + 1 * 16));
  in[2]  = _mm_load_si128((const __m128i *)(input + 2 * 16));
  in[3]  = _mm_load_si128((const __m128i *)(input + 3 * 16));
  in[4]  = _mm_load_si128((const __m128i *)(input + 4 * 16));
  in[5]  = _mm_load_si128((const __m128i *)(input + 5 * 16));
  in[6]  = _mm_load_si128((const __m128i *)(input + 6 * 16));
  in[7]  = _mm_load_si128((const __m128i *)(input + 7 * 16));

  in[8]  = _mm_load_si128((const __m128i *)(input + 8 * 16));
  in[9]  = _mm_load_si128((const __m128i *)(input + 9 * 16));
  in[10]  = _mm_load_si128((const __m128i *)(input + 10 * 16));
  in[11]  = _mm_load_si128((const __m128i *)(input + 11 * 16));
  in[12]  = _mm_load_si128((const __m128i *)(input + 12 * 16));
  in[13]  = _mm_load_si128((const __m128i *)(input + 13 * 16));
  in[14]  = _mm_load_si128((const __m128i *)(input + 14 * 16));
  in[15]  = _mm_load_si128((const __m128i *)(input + 15 * 16));
}

#define RECON_AND_STORE(dest, in_x) \
  {                                                     \
     __m128i d0 = _mm_loadl_epi64((__m128i *)(dest)); \
      d0 = _mm_unpacklo_epi8(d0, zero); \
      d0 = _mm_add_epi16(in_x, d0); \
      d0 = _mm_packus_epi16(d0, d0); \
      _mm_storel_epi64((__m128i *)(dest), d0); \
  }

static INLINE void write_buffer_8x16(uint8_t *dest, __m128i *in, int stride) {
  const __m128i final_rounding = _mm_set1_epi16(1<<5);
  const __m128i zero = _mm_setzero_si128();
  // Final rounding and shift
  in[0] = _mm_adds_epi16(in[0], final_rounding);
  in[1] = _mm_adds_epi16(in[1], final_rounding);
  in[2] = _mm_adds_epi16(in[2], final_rounding);
  in[3] = _mm_adds_epi16(in[3], final_rounding);
  in[4] = _mm_adds_epi16(in[4], final_rounding);
  in[5] = _mm_adds_epi16(in[5], final_rounding);
  in[6] = _mm_adds_epi16(in[6], final_rounding);
  in[7] = _mm_adds_epi16(in[7], final_rounding);
  in[8] = _mm_adds_epi16(in[8], final_rounding);
  in[9] = _mm_adds_epi16(in[9], final_rounding);
  in[10] = _mm_adds_epi16(in[10], final_rounding);
  in[11] = _mm_adds_epi16(in[11], final_rounding);
  in[12] = _mm_adds_epi16(in[12], final_rounding);
  in[13] = _mm_adds_epi16(in[13], final_rounding);
  in[14] = _mm_adds_epi16(in[14], final_rounding);
  in[15] = _mm_adds_epi16(in[15], final_rounding);

  in[0] = _mm_srai_epi16(in[0], 6);
  in[1] = _mm_srai_epi16(in[1], 6);
  in[2] = _mm_srai_epi16(in[2], 6);
  in[3] = _mm_srai_epi16(in[3], 6);
  in[4] = _mm_srai_epi16(in[4], 6);
  in[5] = _mm_srai_epi16(in[5], 6);
  in[6] = _mm_srai_epi16(in[6], 6);
  in[7] = _mm_srai_epi16(in[7], 6);
  in[8] = _mm_srai_epi16(in[8], 6);
  in[9] = _mm_srai_epi16(in[9], 6);
  in[10] = _mm_srai_epi16(in[10], 6);
  in[11] = _mm_srai_epi16(in[11], 6);
  in[12] = _mm_srai_epi16(in[12], 6);
  in[13] = _mm_srai_epi16(in[13], 6);
  in[14] = _mm_srai_epi16(in[14], 6);
  in[15] = _mm_srai_epi16(in[15], 6);

  RECON_AND_STORE(dest +  0 * stride, in[0]);
  RECON_AND_STORE(dest +  1 * stride, in[1]);
  RECON_AND_STORE(dest +  2 * stride, in[2]);
  RECON_AND_STORE(dest +  3 * stride, in[3]);
  RECON_AND_STORE(dest +  4 * stride, in[4]);
  RECON_AND_STORE(dest +  5 * stride, in[5]);
  RECON_AND_STORE(dest +  6 * stride, in[6]);
  RECON_AND_STORE(dest +  7 * stride, in[7]);
  RECON_AND_STORE(dest +  8 * stride, in[8]);
  RECON_AND_STORE(dest +  9 * stride, in[9]);
  RECON_AND_STORE(dest + 10 * stride, in[10]);
  RECON_AND_STORE(dest + 11 * stride, in[11]);
  RECON_AND_STORE(dest + 12 * stride, in[12]);
  RECON_AND_STORE(dest + 13 * stride, in[13]);
  RECON_AND_STORE(dest + 14 * stride, in[14]);
  RECON_AND_STORE(dest + 15 * stride, in[15]);
}