1 // Copyright 2016 Google Inc. All Rights Reserved.
2 //
3 // Use of this source code is governed by a BSD-style license
4 // that can be found in the COPYING file in the root of the source
5 // tree. An additional intellectual property rights grant can be found
6 // in the file PATENTS. All contributing project authors may
7 // be found in the AUTHORS file in the root of the source tree.
8 // -----------------------------------------------------------------------------
9 //
10 // SSE4 code common to several files.
11 //
12 // Author: Vincent Rabaud (vrabaud@google.com)
13
14 #ifndef WEBP_DSP_COMMON_SSE41_H_
15 #define WEBP_DSP_COMMON_SSE41_H_
16
17 #ifdef __cplusplus
18 extern "C" {
19 #endif
20
21 #if defined(WEBP_USE_SSE41)
22 #include <smmintrin.h>
23
24 //------------------------------------------------------------------------------
25 // Channel mixing.
26 // Shuffles the input buffer as A0 0 0 A1 0 0 A2 ...
27 #define WEBP_SSE41_SHUFF(OUT, IN0, IN1) \
28 OUT##0 = _mm_shuffle_epi8(*IN0, shuff0); \
29 OUT##1 = _mm_shuffle_epi8(*IN0, shuff1); \
30 OUT##2 = _mm_shuffle_epi8(*IN0, shuff2); \
31 OUT##3 = _mm_shuffle_epi8(*IN1, shuff0); \
32 OUT##4 = _mm_shuffle_epi8(*IN1, shuff1); \
33 OUT##5 = _mm_shuffle_epi8(*IN1, shuff2);
34
35 // Pack the planar buffers
36 // rrrr... rrrr... gggg... gggg... bbbb... bbbb....
37 // triplet by triplet in the output buffer rgb as rgbrgbrgbrgb ...
VP8PlanarTo24b_SSE41(__m128i * const in0,__m128i * const in1,__m128i * const in2,__m128i * const in3,__m128i * const in4,__m128i * const in5)38 static WEBP_INLINE void VP8PlanarTo24b_SSE41(
39 __m128i* const in0, __m128i* const in1, __m128i* const in2,
40 __m128i* const in3, __m128i* const in4, __m128i* const in5) {
41 __m128i R0, R1, R2, R3, R4, R5;
42 __m128i G0, G1, G2, G3, G4, G5;
43 __m128i B0, B1, B2, B3, B4, B5;
44
45 // Process R.
46 {
47 const __m128i shuff0 = _mm_set_epi8(
48 5, -1, -1, 4, -1, -1, 3, -1, -1, 2, -1, -1, 1, -1, -1, 0);
49 const __m128i shuff1 = _mm_set_epi8(
50 -1, 10, -1, -1, 9, -1, -1, 8, -1, -1, 7, -1, -1, 6, -1, -1);
51 const __m128i shuff2 = _mm_set_epi8(
52 -1, -1, 15, -1, -1, 14, -1, -1, 13, -1, -1, 12, -1, -1, 11, -1);
53 WEBP_SSE41_SHUFF(R, in0, in1)
54 }
55
56 // Process G.
57 {
58 // Same as before, just shifted to the left by one and including the right
59 // padding.
60 const __m128i shuff0 = _mm_set_epi8(
61 -1, -1, 4, -1, -1, 3, -1, -1, 2, -1, -1, 1, -1, -1, 0, -1);
62 const __m128i shuff1 = _mm_set_epi8(
63 10, -1, -1, 9, -1, -1, 8, -1, -1, 7, -1, -1, 6, -1, -1, 5);
64 const __m128i shuff2 = _mm_set_epi8(
65 -1, 15, -1, -1, 14, -1, -1, 13, -1, -1, 12, -1, -1, 11, -1, -1);
66 WEBP_SSE41_SHUFF(G, in2, in3)
67 }
68
69 // Process B.
70 {
71 const __m128i shuff0 = _mm_set_epi8(
72 -1, 4, -1, -1, 3, -1, -1, 2, -1, -1, 1, -1, -1, 0, -1, -1);
73 const __m128i shuff1 = _mm_set_epi8(
74 -1, -1, 9, -1, -1, 8, -1, -1, 7, -1, -1, 6, -1, -1, 5, -1);
75 const __m128i shuff2 = _mm_set_epi8(
76 15, -1, -1, 14, -1, -1, 13, -1, -1, 12, -1, -1, 11, -1, -1, 10);
77 WEBP_SSE41_SHUFF(B, in4, in5)
78 }
79
80 // OR the different channels.
81 {
82 const __m128i RG0 = _mm_or_si128(R0, G0);
83 const __m128i RG1 = _mm_or_si128(R1, G1);
84 const __m128i RG2 = _mm_or_si128(R2, G2);
85 const __m128i RG3 = _mm_or_si128(R3, G3);
86 const __m128i RG4 = _mm_or_si128(R4, G4);
87 const __m128i RG5 = _mm_or_si128(R5, G5);
88 *in0 = _mm_or_si128(RG0, B0);
89 *in1 = _mm_or_si128(RG1, B1);
90 *in2 = _mm_or_si128(RG2, B2);
91 *in3 = _mm_or_si128(RG3, B3);
92 *in4 = _mm_or_si128(RG4, B4);
93 *in5 = _mm_or_si128(RG5, B5);
94 }
95 }
96
97 #undef WEBP_SSE41_SHUFF
98
99 // Convert four packed four-channel buffers like argbargbargbargb... into the
100 // split channels aaaaa ... rrrr ... gggg .... bbbbb ......
VP8L32bToPlanar_SSE41(__m128i * const in0,__m128i * const in1,__m128i * const in2,__m128i * const in3)101 static WEBP_INLINE void VP8L32bToPlanar_SSE41(__m128i* const in0,
102 __m128i* const in1,
103 __m128i* const in2,
104 __m128i* const in3) {
105 // aaaarrrrggggbbbb
106 const __m128i shuff0 =
107 _mm_set_epi8(15, 11, 7, 3, 14, 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0);
108 const __m128i A0 = _mm_shuffle_epi8(*in0, shuff0);
109 const __m128i A1 = _mm_shuffle_epi8(*in1, shuff0);
110 const __m128i A2 = _mm_shuffle_epi8(*in2, shuff0);
111 const __m128i A3 = _mm_shuffle_epi8(*in3, shuff0);
112 // A0A1R0R1
113 // G0G1B0B1
114 // A2A3R2R3
115 // G0G1B0B1
116 const __m128i B0 = _mm_unpacklo_epi32(A0, A1);
117 const __m128i B1 = _mm_unpackhi_epi32(A0, A1);
118 const __m128i B2 = _mm_unpacklo_epi32(A2, A3);
119 const __m128i B3 = _mm_unpackhi_epi32(A2, A3);
120 *in3 = _mm_unpacklo_epi64(B0, B2);
121 *in2 = _mm_unpackhi_epi64(B0, B2);
122 *in1 = _mm_unpacklo_epi64(B1, B3);
123 *in0 = _mm_unpackhi_epi64(B1, B3);
124 }
125
126 #endif // WEBP_USE_SSE41
127
128 #ifdef __cplusplus
129 } // extern "C"
130 #endif
131
132 #endif // WEBP_DSP_COMMON_SSE41_H_
133