1 /*
2 * Copyright (c) 2017 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #ifndef VPX_VPX_DSP_PPC_TRANSPOSE_VSX_H_
12 #define VPX_VPX_DSP_PPC_TRANSPOSE_VSX_H_
13
14 #include "./vpx_config.h"
15 #include "vpx_dsp/ppc/types_vsx.h"
16
vpx_transpose_s16_8x8(int16x8_t v[8])17 static INLINE void vpx_transpose_s16_8x8(int16x8_t v[8]) {
18 // d = vec_mergeh(a,b):
19 // The even elements of the result are obtained left-to-right,
20 // from the high elements of a.
21 // The odd elements of the result are obtained left-to-right,
22 // from the high elements of b.
23 //
24 // d = vec_mergel(a,b):
25 // The even elements of the result are obtained left-to-right,
26 // from the low elements of a.
27 // The odd elements of the result are obtained left-to-right,
28 // from the low elements of b.
29
30 // Example, starting with:
31 // v[0]: 00 01 02 03 04 05 06 07
32 // v[1]: 10 11 12 13 14 15 16 17
33 // v[2]: 20 21 22 23 24 25 26 27
34 // v[3]: 30 31 32 33 34 35 36 37
35 // v[4]: 40 41 42 43 44 45 46 47
36 // v[5]: 50 51 52 53 54 55 56 57
37 // v[6]: 60 61 62 63 64 65 66 67
38 // v[7]: 70 71 72 73 74 75 76 77
39
40 int16x8_t b0, b1, b2, b3, b4, b5, b6, b7;
41 int16x8_t c0, c1, c2, c3, c4, c5, c6, c7;
42
43 b0 = vec_mergeh(v[0], v[4]);
44 b1 = vec_mergel(v[0], v[4]);
45 b2 = vec_mergeh(v[1], v[5]);
46 b3 = vec_mergel(v[1], v[5]);
47 b4 = vec_mergeh(v[2], v[6]);
48 b5 = vec_mergel(v[2], v[6]);
49 b6 = vec_mergeh(v[3], v[7]);
50 b7 = vec_mergel(v[3], v[7]);
51
52 // After first merge operation
53 // b0: 00 40 01 41 02 42 03 43
54 // b1: 04 44 05 45 06 46 07 47
55 // b2: 10 50 11 51 12 52 13 53
56 // b3: 14 54 15 55 16 56 17 57
57 // b4: 20 60 21 61 22 62 23 63
58 // b5: 24 64 25 65 26 66 27 67
59 // b6: 30 70 31 71 32 62 33 73
60 // b7: 34 74 35 75 36 76 37 77
61
62 c0 = vec_mergeh(b0, b4);
63 c1 = vec_mergel(b0, b4);
64 c2 = vec_mergeh(b1, b5);
65 c3 = vec_mergel(b1, b5);
66 c4 = vec_mergeh(b2, b6);
67 c5 = vec_mergel(b2, b6);
68 c6 = vec_mergeh(b3, b7);
69 c7 = vec_mergel(b3, b7);
70
71 // After second merge operation
72 // c0: 00 20 40 60 01 21 41 61
73 // c1: 02 22 42 62 03 23 43 63
74 // c2: 04 24 44 64 05 25 45 65
75 // c3: 06 26 46 66 07 27 47 67
76 // c4: 10 30 50 70 11 31 51 71
77 // c5: 12 32 52 72 13 33 53 73
78 // c6: 14 34 54 74 15 35 55 75
79 // c7: 16 36 56 76 17 37 57 77
80
81 v[0] = vec_mergeh(c0, c4);
82 v[1] = vec_mergel(c0, c4);
83 v[2] = vec_mergeh(c1, c5);
84 v[3] = vec_mergel(c1, c5);
85 v[4] = vec_mergeh(c2, c6);
86 v[5] = vec_mergel(c2, c6);
87 v[6] = vec_mergeh(c3, c7);
88 v[7] = vec_mergel(c3, c7);
89
90 // After last merge operation
91 // v[0]: 00 10 20 30 40 50 60 70
92 // v[1]: 01 11 21 31 41 51 61 71
93 // v[2]: 02 12 22 32 42 52 62 72
94 // v[3]: 03 13 23 33 43 53 63 73
95 // v[4]: 04 14 24 34 44 54 64 74
96 // v[5]: 05 15 25 35 45 55 65 75
97 // v[6]: 06 16 26 36 46 56 66 76
98 // v[7]: 07 17 27 37 47 57 67 77
99 }
100
transpose_8x8(const int16x8_t * a,int16x8_t * b)101 static INLINE void transpose_8x8(const int16x8_t *a, int16x8_t *b) {
102 // Stage 1
103 const int16x8_t s1_0 = vec_mergeh(a[0], a[4]);
104 const int16x8_t s1_1 = vec_mergel(a[0], a[4]);
105 const int16x8_t s1_2 = vec_mergeh(a[1], a[5]);
106 const int16x8_t s1_3 = vec_mergel(a[1], a[5]);
107 const int16x8_t s1_4 = vec_mergeh(a[2], a[6]);
108 const int16x8_t s1_5 = vec_mergel(a[2], a[6]);
109 const int16x8_t s1_6 = vec_mergeh(a[3], a[7]);
110 const int16x8_t s1_7 = vec_mergel(a[3], a[7]);
111
112 // Stage 2
113 const int16x8_t s2_0 = vec_mergeh(s1_0, s1_4);
114 const int16x8_t s2_1 = vec_mergel(s1_0, s1_4);
115 const int16x8_t s2_2 = vec_mergeh(s1_1, s1_5);
116 const int16x8_t s2_3 = vec_mergel(s1_1, s1_5);
117 const int16x8_t s2_4 = vec_mergeh(s1_2, s1_6);
118 const int16x8_t s2_5 = vec_mergel(s1_2, s1_6);
119 const int16x8_t s2_6 = vec_mergeh(s1_3, s1_7);
120 const int16x8_t s2_7 = vec_mergel(s1_3, s1_7);
121
122 // Stage 2
123 b[0] = vec_mergeh(s2_0, s2_4);
124 b[1] = vec_mergel(s2_0, s2_4);
125 b[2] = vec_mergeh(s2_1, s2_5);
126 b[3] = vec_mergel(s2_1, s2_5);
127 b[4] = vec_mergeh(s2_2, s2_6);
128 b[5] = vec_mergel(s2_2, s2_6);
129 b[6] = vec_mergeh(s2_3, s2_7);
130 b[7] = vec_mergel(s2_3, s2_7);
131 }
132
133 #endif // VPX_VPX_DSP_PPC_TRANSPOSE_VSX_H_
134