1/*
2 * Copyright © 2008 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * SOFTWARE.
22 *
23 * Author:
24 *    Zou Nan hai <nanhai.zou@intel.com>
25 *    Zhang Hua jun <huajun.zhang@intel.com>
26 *    Xing Dong sheng <dongsheng.xing@intel.com>
27 *
28 */
29mov (1) g115.8<1>UD 0x07000FUD {align1};
30send (16) 0 g86.0<1>UW g115<16,8,1>UW read(surface_u, 2, 0, 2) mlen 1 rlen 4 {align1};//U
31send (16) 0 g91.0<1>UW g115<16,8,1>UW read(surface_v, 2, 0, 2) mlen 1 rlen 4 {align1};//V
32and (1) g2.24<1>UD g115.0<1,1,1>UD 3UD {align1};
33mul (1) g2.24<1>UD g2.24<1,1,1>UD 25UD {align1};
34mov (1) g115.8<1>UD 0x01000FUD {align1};
35add (1) g115.4<1>UD g115.4<1,1,1>UD 8UD {align1};
36send (16) 0 g90.0<1>UW g115<16,8,1>UW read(surface_u, 2, 0, 2) mlen 1 rlen 1 {align1};
37send (16) 0 g95.0<1>UW g115<16,8,1>UW read(surface_v, 2, 0, 2) mlen 1 rlen 1 {align1};
38jmpi g2.24<1,1,1>D;
39//U
40add (16) g78.0<1>UW g86.0<16,8,1>UB g87.0<16,8,1>UB {align1};
41add (16) g79.0<1>UW g87.0<16,8,1>UB g88.0<16,8,1>UB {align1};
42add (16) g80.0<1>UW g88.0<16,8,1>UB g89.0<16,8,1>UB {align1};
43add (16) g81.0<1>UW g89.0<16,8,1>UB g90.0<16,8,1>UB {align1};
44
45add (16) g78.0<1>UW g78.0<16,8,1>UW g86.1<16,8,1>UB {align1};
46add (16) g79.0<1>UW g79.0<16,8,1>UW g87.1<16,8,1>UB {align1};
47add (16) g80.0<1>UW g80.0<16,8,1>UW g88.1<16,8,1>UB {align1};
48add (16) g81.0<1>UW g81.0<16,8,1>UW g89.1<16,8,1>UB {align1};
49
50add (16) g78.0<1>UW g78.0<16,8,1>UW g87.1<16,8,1>UB {align1};
51add (16) g79.0<1>UW g79.0<16,8,1>UW g88.1<16,8,1>UB {align1};
52add (16) g80.0<1>UW g80.0<16,8,1>UW g89.1<16,8,1>UB {align1};
53add (16) g81.0<1>UW g81.0<16,8,1>UW g90.1<16,8,1>UB {align1};
54//V
55add (16) g82.0<1>UW g91.0<16,8,1>UB g92.0<16,8,1>UB {align1};
56add (16) g83.0<1>UW g92.0<16,8,1>UB g93.0<16,8,1>UB {align1};
57add (16) g84.0<1>UW g93.0<16,8,1>UB g94.0<16,8,1>UB {align1};
58add (16) g85.0<1>UW g94.0<16,8,1>UB g95.0<16,8,1>UB {align1};
59
60add (16) g82.0<1>UW g82.0<16,8,1>UW g91.1<16,8,1>UB {align1};
61add (16) g83.0<1>UW g83.0<16,8,1>UW g92.1<16,8,1>UB {align1};
62add (16) g84.0<1>UW g84.0<16,8,1>UW g93.1<16,8,1>UB {align1};
63add (16) g85.0<1>UW g85.0<16,8,1>UW g94.1<16,8,1>UB {align1};
64
65add (16) g82.0<1>UW g82.0<16,8,1>UW g92.1<16,8,1>UB {align1};
66add (16) g83.0<1>UW g83.0<16,8,1>UW g93.1<16,8,1>UB {align1};
67add (16) g84.0<1>UW g84.0<16,8,1>UW g94.1<16,8,1>UB {align1};
68add (16) g85.0<1>UW g85.0<16,8,1>UW g95.1<16,8,1>UB {align1};
69jmpi out;
70//U
71add (16) g78.0<1>UW g86.1<16,8,1>UB g87.1<16,8,1>UB {align1};
72add (16) g79.0<1>UW g87.1<16,8,1>UB g88.1<16,8,1>UB {align1};
73add (16) g80.0<1>UW g88.1<16,8,1>UB g89.1<16,8,1>UB {align1};
74add (16) g81.0<1>UW g89.1<16,8,1>UB g90.1<16,8,1>UB {align1};
75
76add (16) g78.0<1>UW g78.0<16,8,1>UW g86.2<16,8,1>UB {align1};
77add (16) g79.0<1>UW g79.0<16,8,1>UW g87.2<16,8,1>UB {align1};
78add (16) g80.0<1>UW g80.0<16,8,1>UW g88.2<16,8,1>UB {align1};
79add (16) g81.0<1>UW g81.0<16,8,1>UW g89.2<16,8,1>UB {align1};
80
81add (16) g78.0<1>UW g78.0<16,8,1>UW g87.2<16,8,1>UB {align1};
82add (16) g79.0<1>UW g79.0<16,8,1>UW g88.2<16,8,1>UB {align1};
83add (16) g80.0<1>UW g80.0<16,8,1>UW g89.2<16,8,1>UB {align1};
84add (16) g81.0<1>UW g81.0<16,8,1>UW g90.2<16,8,1>UB {align1};
85//V
86add (16) g82.0<1>UW g91.1<16,8,1>UB g92.1<16,8,1>UB {align1};
87add (16) g83.0<1>UW g92.1<16,8,1>UB g93.1<16,8,1>UB {align1};
88add (16) g84.0<1>UW g93.1<16,8,1>UB g94.1<16,8,1>UB {align1};
89add (16) g85.0<1>UW g94.1<16,8,1>UB g95.1<16,8,1>UB {align1};
90
91add (16) g82.0<1>UW g82.0<16,8,1>UW g91.2<16,8,1>UB {align1};
92add (16) g83.0<1>UW g83.0<16,8,1>UW g92.2<16,8,1>UB {align1};
93add (16) g84.0<1>UW g84.0<16,8,1>UW g93.2<16,8,1>UB {align1};
94add (16) g85.0<1>UW g85.0<16,8,1>UW g94.2<16,8,1>UB {align1};
95
96add (16) g82.0<1>UW g82.0<16,8,1>UW g92.2<16,8,1>UB {align1};
97add (16) g83.0<1>UW g83.0<16,8,1>UW g93.2<16,8,1>UB {align1};
98add (16) g84.0<1>UW g84.0<16,8,1>UW g94.2<16,8,1>UB {align1};
99add (16) g85.0<1>UW g85.0<16,8,1>UW g95.2<16,8,1>UB {align1};
100jmpi out;
101//U
102add (16) g78.0<1>UW g86.2<16,8,1>UB g87.2<16,8,1>UB {align1};
103add (16) g79.0<1>UW g87.2<16,8,1>UB g88.2<16,8,1>UB {align1};
104add (16) g80.0<1>UW g88.2<16,8,1>UB g89.2<16,8,1>UB {align1};
105add (16) g81.0<1>UW g89.2<16,8,1>UB g90.2<16,8,1>UB {align1};
106
107add (16) g78.0<1>UW g78.0<16,8,1>UW g86.3<16,8,1>UB {align1};
108add (16) g79.0<1>UW g79.0<16,8,1>UW g87.3<16,8,1>UB {align1};
109add (16) g80.0<1>UW g80.0<16,8,1>UW g88.3<16,8,1>UB {align1};
110add (16) g81.0<1>UW g81.0<16,8,1>UW g89.3<16,8,1>UB {align1};
111
112add (16) g78.0<1>UW g78.0<16,8,1>UW g87.3<16,8,1>UB {align1};
113add (16) g79.0<1>UW g79.0<16,8,1>UW g88.3<16,8,1>UB {align1};
114add (16) g80.0<1>UW g80.0<16,8,1>UW g89.3<16,8,1>UB {align1};
115add (16) g81.0<1>UW g81.0<16,8,1>UW g90.3<16,8,1>UB {align1};
116//V
117add (16) g82.0<1>UW g91.2<16,8,1>UB g92.2<16,8,1>UB {align1};
118add (16) g83.0<1>UW g92.2<16,8,1>UB g93.2<16,8,1>UB {align1};
119add (16) g84.0<1>UW g93.2<16,8,1>UB g94.2<16,8,1>UB {align1};
120add (16) g85.0<1>UW g94.2<16,8,1>UB g95.2<16,8,1>UB {align1};
121
122add (16) g82.0<1>UW g82.0<16,8,1>UW g91.3<16,8,1>UB {align1};
123add (16) g83.0<1>UW g83.0<16,8,1>UW g92.3<16,8,1>UB {align1};
124add (16) g84.0<1>UW g84.0<16,8,1>UW g93.3<16,8,1>UB {align1};
125add (16) g85.0<1>UW g85.0<16,8,1>UW g94.3<16,8,1>UB {align1};
126
127add (16) g82.0<1>UW g82.0<16,8,1>UW g92.3<16,8,1>UB {align1};
128add (16) g83.0<1>UW g83.0<16,8,1>UW g93.3<16,8,1>UB {align1};
129add (16) g84.0<1>UW g84.0<16,8,1>UW g94.3<16,8,1>UB {align1};
130add (16) g85.0<1>UW g85.0<16,8,1>UW g95.3<16,8,1>UB {align1};
131jmpi out;
132//U
133add (16) g78.0<1>UW g86.3<16,8,1>UB g87.3<16,8,1>UB {align1};
134add (16) g79.0<1>UW g87.3<16,8,1>UB g88.3<16,8,1>UB {align1};
135add (16) g80.0<1>UW g88.3<16,8,1>UB g89.3<16,8,1>UB {align1};
136add (16) g81.0<1>UW g89.3<16,8,1>UB g90.3<16,8,1>UB {align1};
137
138add (16) g78.0<1>UW g78.0<16,8,1>UW g86.4<16,8,1>UB {align1};
139add (16) g79.0<1>UW g79.0<16,8,1>UW g87.4<16,8,1>UB {align1};
140add (16) g80.0<1>UW g80.0<16,8,1>UW g88.4<16,8,1>UB {align1};
141add (16) g81.0<1>UW g81.0<16,8,1>UW g89.4<16,8,1>UB {align1};
142
143add (16) g78.0<1>UW g78.0<16,8,1>UW g87.4<16,8,1>UB {align1};
144add (16) g79.0<1>UW g79.0<16,8,1>UW g88.4<16,8,1>UB {align1};
145add (16) g80.0<1>UW g80.0<16,8,1>UW g89.4<16,8,1>UB {align1};
146add (16) g81.0<1>UW g81.0<16,8,1>UW g90.4<16,8,1>UB {align1};
147//V
148add (16) g82.0<1>UW g91.3<16,8,1>UB g92.3<16,8,1>UB {align1};
149add (16) g83.0<1>UW g92.3<16,8,1>UB g93.3<16,8,1>UB {align1};
150add (16) g84.0<1>UW g93.3<16,8,1>UB g94.3<16,8,1>UB {align1};
151add (16) g85.0<1>UW g94.3<16,8,1>UB g95.3<16,8,1>UB {align1};
152
153add (16) g82.0<1>UW g82.0<16,8,1>UW g91.4<16,8,1>UB {align1};
154add (16) g83.0<1>UW g83.0<16,8,1>UW g92.4<16,8,1>UB {align1};
155add (16) g84.0<1>UW g84.0<16,8,1>UW g93.4<16,8,1>UB {align1};
156add (16) g85.0<1>UW g85.0<16,8,1>UW g94.4<16,8,1>UB {align1};
157
158add (16) g82.0<1>UW g82.0<16,8,1>UW g92.4<16,8,1>UB {align1};
159add (16) g83.0<1>UW g83.0<16,8,1>UW g93.4<16,8,1>UB {align1};
160add (16) g84.0<1>UW g84.0<16,8,1>UW g94.4<16,8,1>UB {align1};
161add (16) g85.0<1>UW g85.0<16,8,1>UW g95.4<16,8,1>UB {align1};
162
163out:
164shr.sat (16) g78.0<1>UW g78.0<16,16,1>UW 2UW {align1};
165shr.sat (16) g79.0<1>UW g79.0<16,16,1>UW 2UW {align1};
166shr.sat (16) g80.0<1>UW g80.0<16,16,1>UW 2UW {align1};
167shr.sat (16) g81.0<1>UW g81.0<16,16,1>UW 2UW {align1};
168shr.sat (16) g82.0<1>UW g82.0<16,16,1>UW 2UW {align1};
169shr.sat (16) g83.0<1>UW g83.0<16,16,1>UW 2UW {align1};
170shr.sat (16) g84.0<1>UW g84.0<16,16,1>UW 2UW {align1};
171shr.sat (16) g85.0<1>UW g85.0<16,16,1>UW 2UW {align1};
172
173