2 * Copyright © 2008 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 * Zou Nan hai <nanhai.zou@intel.com>
25 * Zhang Hua jun <huajun.zhang@intel.com>
26 * Xing Dong sheng <dongsheng.xing@intel.com>
29 mov (1) g115.8<1>UD 0x07000FUD {align1};
30 send (16) 0 g86.0<1>UW g115<16,8,1>UW read(surface_u, 2, 0, 2) mlen 1 rlen 4 {align1};//U
31 send (16) 0 g91.0<1>UW g115<16,8,1>UW read(surface_v, 2, 0, 2) mlen 1 rlen 4 {align1};//V
32 and (1) g2.24<1>UD g115.0<1,1,1>UD 3UD {align1};
33 mul (1) g2.24<1>UD g2.24<1,1,1>UD 25UD {align1};
34 mov (1) g115.8<1>UD 0x01000FUD {align1};
35 add (1) g115.4<1>UD g115.4<1,1,1>UD 8UD {align1};
36 send (16) 0 g90.0<1>UW g115<16,8,1>UW read(surface_u, 2, 0, 2) mlen 1 rlen 1 {align1};
37 send (16) 0 g95.0<1>UW g115<16,8,1>UW read(surface_v, 2, 0, 2) mlen 1 rlen 1 {align1};
40 add (16) g78.0<1>UW g86.0<16,8,1>UB g87.0<16,8,1>UB {align1};
41 add (16) g79.0<1>UW g87.0<16,8,1>UB g88.0<16,8,1>UB {align1};
42 add (16) g80.0<1>UW g88.0<16,8,1>UB g89.0<16,8,1>UB {align1};
43 add (16) g81.0<1>UW g89.0<16,8,1>UB g90.0<16,8,1>UB {align1};
45 add (16) g78.0<1>UW g78.0<16,8,1>UW g86.1<16,8,1>UB {align1};
46 add (16) g79.0<1>UW g79.0<16,8,1>UW g87.1<16,8,1>UB {align1};
47 add (16) g80.0<1>UW g80.0<16,8,1>UW g88.1<16,8,1>UB {align1};
48 add (16) g81.0<1>UW g81.0<16,8,1>UW g89.1<16,8,1>UB {align1};
50 add (16) g78.0<1>UW g78.0<16,8,1>UW g87.1<16,8,1>UB {align1};
51 add (16) g79.0<1>UW g79.0<16,8,1>UW g88.1<16,8,1>UB {align1};
52 add (16) g80.0<1>UW g80.0<16,8,1>UW g89.1<16,8,1>UB {align1};
53 add (16) g81.0<1>UW g81.0<16,8,1>UW g90.1<16,8,1>UB {align1};
55 add (16) g82.0<1>UW g91.0<16,8,1>UB g92.0<16,8,1>UB {align1};
56 add (16) g83.0<1>UW g92.0<16,8,1>UB g93.0<16,8,1>UB {align1};
57 add (16) g84.0<1>UW g93.0<16,8,1>UB g94.0<16,8,1>UB {align1};
58 add (16) g85.0<1>UW g94.0<16,8,1>UB g95.0<16,8,1>UB {align1};
60 add (16) g82.0<1>UW g82.0<16,8,1>UW g91.1<16,8,1>UB {align1};
61 add (16) g83.0<1>UW g83.0<16,8,1>UW g92.1<16,8,1>UB {align1};
62 add (16) g84.0<1>UW g84.0<16,8,1>UW g93.1<16,8,1>UB {align1};
63 add (16) g85.0<1>UW g85.0<16,8,1>UW g94.1<16,8,1>UB {align1};
65 add (16) g82.0<1>UW g82.0<16,8,1>UW g92.1<16,8,1>UB {align1};
66 add (16) g83.0<1>UW g83.0<16,8,1>UW g93.1<16,8,1>UB {align1};
67 add (16) g84.0<1>UW g84.0<16,8,1>UW g94.1<16,8,1>UB {align1};
68 add (16) g85.0<1>UW g85.0<16,8,1>UW g95.1<16,8,1>UB {align1};
71 add (16) g78.0<1>UW g86.1<16,8,1>UB g87.1<16,8,1>UB {align1};
72 add (16) g79.0<1>UW g87.1<16,8,1>UB g88.1<16,8,1>UB {align1};
73 add (16) g80.0<1>UW g88.1<16,8,1>UB g89.1<16,8,1>UB {align1};
74 add (16) g81.0<1>UW g89.1<16,8,1>UB g90.1<16,8,1>UB {align1};
76 add (16) g78.0<1>UW g78.0<16,8,1>UW g86.2<16,8,1>UB {align1};
77 add (16) g79.0<1>UW g79.0<16,8,1>UW g87.2<16,8,1>UB {align1};
78 add (16) g80.0<1>UW g80.0<16,8,1>UW g88.2<16,8,1>UB {align1};
79 add (16) g81.0<1>UW g81.0<16,8,1>UW g89.2<16,8,1>UB {align1};
81 add (16) g78.0<1>UW g78.0<16,8,1>UW g87.2<16,8,1>UB {align1};
82 add (16) g79.0<1>UW g79.0<16,8,1>UW g88.2<16,8,1>UB {align1};
83 add (16) g80.0<1>UW g80.0<16,8,1>UW g89.2<16,8,1>UB {align1};
84 add (16) g81.0<1>UW g81.0<16,8,1>UW g90.2<16,8,1>UB {align1};
86 add (16) g82.0<1>UW g91.1<16,8,1>UB g92.1<16,8,1>UB {align1};
87 add (16) g83.0<1>UW g92.1<16,8,1>UB g93.1<16,8,1>UB {align1};
88 add (16) g84.0<1>UW g93.1<16,8,1>UB g94.1<16,8,1>UB {align1};
89 add (16) g85.0<1>UW g94.1<16,8,1>UB g95.1<16,8,1>UB {align1};
91 add (16) g82.0<1>UW g82.0<16,8,1>UW g91.2<16,8,1>UB {align1};
92 add (16) g83.0<1>UW g83.0<16,8,1>UW g92.2<16,8,1>UB {align1};
93 add (16) g84.0<1>UW g84.0<16,8,1>UW g93.2<16,8,1>UB {align1};
94 add (16) g85.0<1>UW g85.0<16,8,1>UW g94.2<16,8,1>UB {align1};
96 add (16) g82.0<1>UW g82.0<16,8,1>UW g92.2<16,8,1>UB {align1};
97 add (16) g83.0<1>UW g83.0<16,8,1>UW g93.2<16,8,1>UB {align1};
98 add (16) g84.0<1>UW g84.0<16,8,1>UW g94.2<16,8,1>UB {align1};
99 add (16) g85.0<1>UW g85.0<16,8,1>UW g95.2<16,8,1>UB {align1};
102 add (16) g78.0<1>UW g86.2<16,8,1>UB g87.2<16,8,1>UB {align1};
103 add (16) g79.0<1>UW g87.2<16,8,1>UB g88.2<16,8,1>UB {align1};
104 add (16) g80.0<1>UW g88.2<16,8,1>UB g89.2<16,8,1>UB {align1};
105 add (16) g81.0<1>UW g89.2<16,8,1>UB g90.2<16,8,1>UB {align1};
107 add (16) g78.0<1>UW g78.0<16,8,1>UW g86.3<16,8,1>UB {align1};
108 add (16) g79.0<1>UW g79.0<16,8,1>UW g87.3<16,8,1>UB {align1};
109 add (16) g80.0<1>UW g80.0<16,8,1>UW g88.3<16,8,1>UB {align1};
110 add (16) g81.0<1>UW g81.0<16,8,1>UW g89.3<16,8,1>UB {align1};
112 add (16) g78.0<1>UW g78.0<16,8,1>UW g87.3<16,8,1>UB {align1};
113 add (16) g79.0<1>UW g79.0<16,8,1>UW g88.3<16,8,1>UB {align1};
114 add (16) g80.0<1>UW g80.0<16,8,1>UW g89.3<16,8,1>UB {align1};
115 add (16) g81.0<1>UW g81.0<16,8,1>UW g90.3<16,8,1>UB {align1};
117 add (16) g82.0<1>UW g91.2<16,8,1>UB g92.2<16,8,1>UB {align1};
118 add (16) g83.0<1>UW g92.2<16,8,1>UB g93.2<16,8,1>UB {align1};
119 add (16) g84.0<1>UW g93.2<16,8,1>UB g94.2<16,8,1>UB {align1};
120 add (16) g85.0<1>UW g94.2<16,8,1>UB g95.2<16,8,1>UB {align1};
122 add (16) g82.0<1>UW g82.0<16,8,1>UW g91.3<16,8,1>UB {align1};
123 add (16) g83.0<1>UW g83.0<16,8,1>UW g92.3<16,8,1>UB {align1};
124 add (16) g84.0<1>UW g84.0<16,8,1>UW g93.3<16,8,1>UB {align1};
125 add (16) g85.0<1>UW g85.0<16,8,1>UW g94.3<16,8,1>UB {align1};
127 add (16) g82.0<1>UW g82.0<16,8,1>UW g92.3<16,8,1>UB {align1};
128 add (16) g83.0<1>UW g83.0<16,8,1>UW g93.3<16,8,1>UB {align1};
129 add (16) g84.0<1>UW g84.0<16,8,1>UW g94.3<16,8,1>UB {align1};
130 add (16) g85.0<1>UW g85.0<16,8,1>UW g95.3<16,8,1>UB {align1};
133 add (16) g78.0<1>UW g86.3<16,8,1>UB g87.3<16,8,1>UB {align1};
134 add (16) g79.0<1>UW g87.3<16,8,1>UB g88.3<16,8,1>UB {align1};
135 add (16) g80.0<1>UW g88.3<16,8,1>UB g89.3<16,8,1>UB {align1};
136 add (16) g81.0<1>UW g89.3<16,8,1>UB g90.3<16,8,1>UB {align1};
138 add (16) g78.0<1>UW g78.0<16,8,1>UW g86.4<16,8,1>UB {align1};
139 add (16) g79.0<1>UW g79.0<16,8,1>UW g87.4<16,8,1>UB {align1};
140 add (16) g80.0<1>UW g80.0<16,8,1>UW g88.4<16,8,1>UB {align1};
141 add (16) g81.0<1>UW g81.0<16,8,1>UW g89.4<16,8,1>UB {align1};
143 add (16) g78.0<1>UW g78.0<16,8,1>UW g87.4<16,8,1>UB {align1};
144 add (16) g79.0<1>UW g79.0<16,8,1>UW g88.4<16,8,1>UB {align1};
145 add (16) g80.0<1>UW g80.0<16,8,1>UW g89.4<16,8,1>UB {align1};
146 add (16) g81.0<1>UW g81.0<16,8,1>UW g90.4<16,8,1>UB {align1};
148 add (16) g82.0<1>UW g91.3<16,8,1>UB g92.3<16,8,1>UB {align1};
149 add (16) g83.0<1>UW g92.3<16,8,1>UB g93.3<16,8,1>UB {align1};
150 add (16) g84.0<1>UW g93.3<16,8,1>UB g94.3<16,8,1>UB {align1};
151 add (16) g85.0<1>UW g94.3<16,8,1>UB g95.3<16,8,1>UB {align1};
153 add (16) g82.0<1>UW g82.0<16,8,1>UW g91.4<16,8,1>UB {align1};
154 add (16) g83.0<1>UW g83.0<16,8,1>UW g92.4<16,8,1>UB {align1};
155 add (16) g84.0<1>UW g84.0<16,8,1>UW g93.4<16,8,1>UB {align1};
156 add (16) g85.0<1>UW g85.0<16,8,1>UW g94.4<16,8,1>UB {align1};
158 add (16) g82.0<1>UW g82.0<16,8,1>UW g92.4<16,8,1>UB {align1};
159 add (16) g83.0<1>UW g83.0<16,8,1>UW g93.4<16,8,1>UB {align1};
160 add (16) g84.0<1>UW g84.0<16,8,1>UW g94.4<16,8,1>UB {align1};
161 add (16) g85.0<1>UW g85.0<16,8,1>UW g95.4<16,8,1>UB {align1};
164 shr.sat (16) g78.0<1>UW g78.0<16,16,1>UW 2UW {align1};
165 shr.sat (16) g79.0<1>UW g79.0<16,16,1>UW 2UW {align1};
166 shr.sat (16) g80.0<1>UW g80.0<16,16,1>UW 2UW {align1};
167 shr.sat (16) g81.0<1>UW g81.0<16,16,1>UW 2UW {align1};
168 shr.sat (16) g82.0<1>UW g82.0<16,16,1>UW 2UW {align1};
169 shr.sat (16) g83.0<1>UW g83.0<16,16,1>UW 2UW {align1};
170 shr.sat (16) g84.0<1>UW g84.0<16,16,1>UW 2UW {align1};
171 shr.sat (16) g85.0<1>UW g85.0<16,16,1>UW 2UW {align1};