1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #ifndef AOM_AV1_COMMON_X86_HIGHBD_TXFM_UTILITY_SSE4_H_
13 #define AOM_AV1_COMMON_X86_HIGHBD_TXFM_UTILITY_SSE4_H_
14
15 #include <smmintrin.h> /* SSE4.1 */
16
17 #define TRANSPOSE_4X4(x0, x1, x2, x3, y0, y1, y2, y3) \
18 do { \
19 __m128i u0, u1, u2, u3; \
20 u0 = _mm_unpacklo_epi32(x0, x1); \
21 u1 = _mm_unpackhi_epi32(x0, x1); \
22 u2 = _mm_unpacklo_epi32(x2, x3); \
23 u3 = _mm_unpackhi_epi32(x2, x3); \
24 y0 = _mm_unpacklo_epi64(u0, u2); \
25 y1 = _mm_unpackhi_epi64(u0, u2); \
26 y2 = _mm_unpacklo_epi64(u1, u3); \
27 y3 = _mm_unpackhi_epi64(u1, u3); \
28 } while (0)
29
transpose_8x8(const __m128i * in,__m128i * out)30 static INLINE void transpose_8x8(const __m128i *in, __m128i *out) {
31 TRANSPOSE_4X4(in[0], in[2], in[4], in[6], out[0], out[2], out[4], out[6]);
32 TRANSPOSE_4X4(in[1], in[3], in[5], in[7], out[8], out[10], out[12], out[14]);
33 TRANSPOSE_4X4(in[8], in[10], in[12], in[14], out[1], out[3], out[5], out[7]);
34 TRANSPOSE_4X4(in[9], in[11], in[13], in[15], out[9], out[11], out[13],
35 out[15]);
36 }
37
transpose_16x16(const __m128i * in,__m128i * out)38 static INLINE void transpose_16x16(const __m128i *in, __m128i *out) {
39 // Upper left 8x8
40 TRANSPOSE_4X4(in[0], in[4], in[8], in[12], out[0], out[4], out[8], out[12]);
41 TRANSPOSE_4X4(in[1], in[5], in[9], in[13], out[16], out[20], out[24],
42 out[28]);
43 TRANSPOSE_4X4(in[16], in[20], in[24], in[28], out[1], out[5], out[9],
44 out[13]);
45 TRANSPOSE_4X4(in[17], in[21], in[25], in[29], out[17], out[21], out[25],
46 out[29]);
47
48 // Upper right 8x8
49 TRANSPOSE_4X4(in[2], in[6], in[10], in[14], out[32], out[36], out[40],
50 out[44]);
51 TRANSPOSE_4X4(in[3], in[7], in[11], in[15], out[48], out[52], out[56],
52 out[60]);
53 TRANSPOSE_4X4(in[18], in[22], in[26], in[30], out[33], out[37], out[41],
54 out[45]);
55 TRANSPOSE_4X4(in[19], in[23], in[27], in[31], out[49], out[53], out[57],
56 out[61]);
57
58 // Lower left 8x8
59 TRANSPOSE_4X4(in[32], in[36], in[40], in[44], out[2], out[6], out[10],
60 out[14]);
61 TRANSPOSE_4X4(in[33], in[37], in[41], in[45], out[18], out[22], out[26],
62 out[30]);
63 TRANSPOSE_4X4(in[48], in[52], in[56], in[60], out[3], out[7], out[11],
64 out[15]);
65 TRANSPOSE_4X4(in[49], in[53], in[57], in[61], out[19], out[23], out[27],
66 out[31]);
67 // Lower right 8x8
68 TRANSPOSE_4X4(in[34], in[38], in[42], in[46], out[34], out[38], out[42],
69 out[46]);
70 TRANSPOSE_4X4(in[35], in[39], in[43], in[47], out[50], out[54], out[58],
71 out[62]);
72 TRANSPOSE_4X4(in[50], in[54], in[58], in[62], out[35], out[39], out[43],
73 out[47]);
74 TRANSPOSE_4X4(in[51], in[55], in[59], in[63], out[51], out[55], out[59],
75 out[63]);
76 }
77
transpose_8nx8n(const __m128i * input,__m128i * output,const int width,const int height)78 static INLINE void transpose_8nx8n(const __m128i *input, __m128i *output,
79 const int width, const int height) {
80 const int numcol = height >> 2;
81 const int numrow = width >> 2;
82 for (int j = 0; j < numrow; j++) {
83 for (int i = 0; i < numcol; i++) {
84 TRANSPOSE_4X4(input[i * width + j + (numrow * 0)],
85 input[i * width + j + (numrow * 1)],
86 input[i * width + j + (numrow * 2)],
87 input[i * width + j + (numrow * 3)],
88 output[j * height + i + (numcol * 0)],
89 output[j * height + i + (numcol * 1)],
90 output[j * height + i + (numcol * 2)],
91 output[j * height + i + (numcol * 3)]);
92 }
93 }
94 }
95
96 // Note:
97 // rounding = 1 << (bit - 1)
half_btf_sse4_1(const __m128i * w0,const __m128i * n0,const __m128i * w1,const __m128i * n1,const __m128i * rounding,int bit)98 static INLINE __m128i half_btf_sse4_1(const __m128i *w0, const __m128i *n0,
99 const __m128i *w1, const __m128i *n1,
100 const __m128i *rounding, int bit) {
101 __m128i x, y;
102
103 x = _mm_mullo_epi32(*w0, *n0);
104 y = _mm_mullo_epi32(*w1, *n1);
105 x = _mm_add_epi32(x, y);
106 x = _mm_add_epi32(x, *rounding);
107 x = _mm_srai_epi32(x, bit);
108 return x;
109 }
110
half_btf_0_sse4_1(const __m128i * w0,const __m128i * n0,const __m128i * rounding,int bit)111 static INLINE __m128i half_btf_0_sse4_1(const __m128i *w0, const __m128i *n0,
112 const __m128i *rounding, int bit) {
113 __m128i x;
114
115 x = _mm_mullo_epi32(*w0, *n0);
116 x = _mm_add_epi32(x, *rounding);
117 x = _mm_srai_epi32(x, bit);
118 return x;
119 }
120
121 typedef void (*transform_1d_sse4_1)(__m128i *in, __m128i *out, int bit,
122 int do_cols, int bd, int out_shift);
123
124 typedef void (*fwd_transform_1d_sse4_1)(__m128i *in, __m128i *out, int bit,
125 const int num_cols);
126
127 void av1_highbd_inv_txfm2d_add_universe_sse4_1(const int32_t *input,
128 uint8_t *output, int stride,
129 TX_TYPE tx_type, TX_SIZE tx_size,
130 int eob, const int bd);
131
132 #endif // AOM_AV1_COMMON_X86_HIGHBD_TXFM_UTILITY_SSE4_H_
133