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 0x01FUD {align1};
30send (16) 0 g78.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1};
31add (1) g115.4<1>UD g115.4<1,1,1>UD 2UD {align1};
32send (16) 0 g80.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1};
33add (1) g115.4<1>UD g115.4<1,1,1>UD 2UD {align1};
34send (16) 0 g82.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1};
35add (1) g115.4<1>UD g115.4<1,1,1>UD 2UD {align1};
36send (16) 0 g84.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1};
37add (1) g115.4<1>UD g115.4<1,1,1>UD 2UD {align1};
38mov (1) g115.8<1>UD 0x07001FUD {align1};
39send (16) 0 g86.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 8 {align1};
40add (1) g115.4<1>UD g115.4<1,1,1>UD 8UD {align1};
41mov (1) g115.8<1>UD 0x1FUD {align1};
42send (16) 0 g94.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1};
43and (1) g2.24<1>UD g115.0<1,1,1>UD 3UD {align1};
44mul (1) g2.24<1>UD g2.24<1,1,1>UD 25UD {align1};
45jmpi g2.24<1,1,1>D;
46
47add (16) g96.0<1>UW g78.0<16,16,1>UB g80.0<16,16,1>UB {align1};
48add (16) g97.0<1>UW g80.0<16,16,1>UB g82.0<16,16,1>UB {align1};
49add (16) g98.0<1>UW g82.0<16,16,1>UB g84.0<16,16,1>UB {align1};
50add (16) g99.0<1>UW g84.0<16,16,1>UB g86.0<16,16,1>UB {align1};
51add (16) g100.0<1>UW g86.0<16,16,1>UB g88.0<16,16,1>UB {align1};
52add (16) g101.0<1>UW g88.0<16,16,1>UB g90.0<16,16,1>UB {align1};
53add (16) g102.0<1>UW g90.0<16,16,1>UB g92.0<16,16,1>UB {align1};
54add (16) g103.0<1>UW g92.0<16,16,1>UB g94.0<16,16,1>UB {align1};
55
56add (16) g96.0<1>UW g96.0<16,16,1>UW g78.1<16,16,1>UB {align1};
57add (16) g97.0<1>UW g97.0<16,16,1>UW g80.1<16,16,1>UB {align1};
58add (16) g98.0<1>UW g98.0<16,16,1>UW g82.1<16,16,1>UB {align1};
59add (16) g99.0<1>UW g99.0<16,16,1>UW g84.1<16,16,1>UB {align1};
60add (16) g100.0<1>UW g100.0<16,16,1>UW g86.1<16,16,1>UB {align1};
61add (16) g101.0<1>UW g101.0<16,16,1>UW g88.1<16,16,1>UB {align1};
62add (16) g102.0<1>UW g102.0<16,16,1>UW g90.1<16,16,1>UB {align1};
63add (16) g103.0<1>UW g103.0<16,16,1>UW g92.1<16,16,1>UB {align1};
64
65add (16) g96.0<1>UW g96.0<16,16,1>UW g80.1<16,16,1>UB {align1};
66add (16) g97.0<1>UW g97.0<16,16,1>UW g82.1<16,16,1>UB {align1};
67add (16) g98.0<1>UW g98.0<16,16,1>UW g84.1<16,16,1>UB {align1};
68add (16) g99.0<1>UW g99.0<16,16,1>UW g86.1<16,16,1>UB {align1};
69add (16) g100.0<1>UW g100.0<16,16,1>UW g88.1<16,16,1>UB {align1};
70add (16) g101.0<1>UW g101.0<16,16,1>UW g90.1<16,16,1>UB {align1};
71add (16) g102.0<1>UW g102.0<16,16,1>UW g92.1<16,16,1>UB {align1};
72add (16) g103.0<1>UW g103.0<16,16,1>UW g94.1<16,16,1>UB {align1};
73jmpi out;
74
75add (16) g96.0<1>UW g78.1<16,16,1>UB g80.1<16,16,1>UB {align1};
76add (16) g97.0<1>UW g80.1<16,16,1>UB g82.1<16,16,1>UB {align1};
77add (16) g98.0<1>UW g82.1<16,16,1>UB g84.1<16,16,1>UB {align1};
78add (16) g99.0<1>UW g84.1<16,16,1>UB g86.1<16,16,1>UB {align1};
79add (16) g100.0<1>UW g86.1<16,16,1>UB g88.1<16,16,1>UB {align1};
80add (16) g101.0<1>UW g88.1<16,16,1>UB g90.1<16,16,1>UB {align1};
81add (16) g102.0<1>UW g90.1<16,16,1>UB g92.1<16,16,1>UB {align1};
82add (16) g103.0<1>UW g92.1<16,16,1>UB g94.1<16,16,1>UB {align1};
83
84add (16) g96.0<1>UW g96.0<16,16,1>UW g78.2<16,16,1>UB {align1};
85add (16) g97.0<1>UW g97.0<16,16,1>UW g80.2<16,16,1>UB {align1};
86add (16) g98.0<1>UW g98.0<16,16,1>UW g82.2<16,16,1>UB {align1};
87add (16) g99.0<1>UW g99.0<16,16,1>UW g84.2<16,16,1>UB {align1};
88add (16) g100.0<1>UW g100.0<16,16,1>UW g86.2<16,16,1>UB {align1};
89add (16) g101.0<1>UW g101.0<16,16,1>UW g88.2<16,16,1>UB {align1};
90add (16) g102.0<1>UW g102.0<16,16,1>UW g90.2<16,16,1>UB {align1};
91add (16) g103.0<1>UW g103.0<16,16,1>UW g92.2<16,16,1>UB {align1};
92
93add (16) g96.0<1>UW g96.0<16,16,1>UW g80.2<16,16,1>UB {align1};
94add (16) g97.0<1>UW g97.0<16,16,1>UW g82.2<16,16,1>UB {align1};
95add (16) g98.0<1>UW g98.0<16,16,1>UW g84.2<16,16,1>UB {align1};
96add (16) g99.0<1>UW g99.0<16,16,1>UW g86.2<16,16,1>UB {align1};
97add (16) g100.0<1>UW g100.0<16,16,1>UW g88.2<16,16,1>UB {align1};
98add (16) g101.0<1>UW g101.0<16,16,1>UW g90.2<16,16,1>UB {align1};
99add (16) g102.0<1>UW g102.0<16,16,1>UW g92.2<16,16,1>UB {align1};
100add (16) g103.0<1>UW g103.0<16,16,1>UW g94.2<16,16,1>UB {align1};
101jmpi out;
102
103add (16) g96.0<1>UW g78.2<16,16,1>UB g80.2<16,16,1>UB {align1};
104add (16) g97.0<1>UW g80.2<16,16,1>UB g82.2<16,16,1>UB {align1};
105add (16) g98.0<1>UW g82.2<16,16,1>UB g84.2<16,16,1>UB {align1};
106add (16) g99.0<1>UW g84.2<16,16,1>UB g86.2<16,16,1>UB {align1};
107add (16) g100.0<1>UW g86.2<16,16,1>UB g88.2<16,16,1>UB {align1};
108add (16) g101.0<1>UW g88.2<16,16,1>UB g90.2<16,16,1>UB {align1};
109add (16) g102.0<1>UW g90.2<16,16,1>UB g92.2<16,16,1>UB {align1};
110add (16) g103.0<1>UW g92.2<16,16,1>UB g94.2<16,16,1>UB {align1};
111
112add (16) g96.0<1>UW g96.0<16,16,1>UW g78.3<16,16,1>UB {align1};
113add (16) g97.0<1>UW g97.0<16,16,1>UW g80.3<16,16,1>UB {align1};
114add (16) g98.0<1>UW g98.0<16,16,1>UW g82.3<16,16,1>UB {align1};
115add (16) g99.0<1>UW g99.0<16,16,1>UW g84.3<16,16,1>UB {align1};
116add (16) g100.0<1>UW g100.0<16,16,1>UW g86.3<16,16,1>UB {align1};
117add (16) g101.0<1>UW g101.0<16,16,1>UW g88.3<16,16,1>UB {align1};
118add (16) g102.0<1>UW g102.0<16,16,1>UW g90.3<16,16,1>UB {align1};
119add (16) g103.0<1>UW g103.0<16,16,1>UW g92.3<16,16,1>UB {align1};
120
121add (16) g96.0<1>UW g96.0<16,16,1>UW g80.3<16,16,1>UB {align1};
122add (16) g97.0<1>UW g97.0<16,16,1>UW g82.3<16,16,1>UB {align1};
123add (16) g98.0<1>UW g98.0<16,16,1>UW g84.3<16,16,1>UB {align1};
124add (16) g99.0<1>UW g99.0<16,16,1>UW g86.3<16,16,1>UB {align1};
125add (16) g100.0<1>UW g100.0<16,16,1>UW g88.3<16,16,1>UB {align1};
126add (16) g101.0<1>UW g101.0<16,16,1>UW g90.3<16,16,1>UB {align1};
127add (16) g102.0<1>UW g102.0<16,16,1>UW g92.3<16,16,1>UB {align1};
128add (16) g103.0<1>UW g103.0<16,16,1>UW g94.3<16,16,1>UB {align1};
129jmpi out;
130
131add (16) g96.0<1>UW g78.3<16,16,1>UB g80.3<16,16,1>UB {align1};
132add (16) g97.0<1>UW g80.3<16,16,1>UB g82.3<16,16,1>UB {align1};
133add (16) g98.0<1>UW g82.3<16,16,1>UB g84.3<16,16,1>UB {align1};
134add (16) g99.0<1>UW g84.3<16,16,1>UB g86.3<16,16,1>UB {align1};
135add (16) g100.0<1>UW g86.3<16,16,1>UB g88.3<16,16,1>UB {align1};
136add (16) g101.0<1>UW g88.3<16,16,1>UB g90.3<16,16,1>UB {align1};
137add (16) g102.0<1>UW g90.3<16,16,1>UB g92.3<16,16,1>UB {align1};
138add (16) g103.0<1>UW g92.3<16,16,1>UB g94.3<16,16,1>UB {align1};
139
140add (16) g96.0<1>UW g96.0<16,16,1>UW g78.4<16,16,1>UB {align1};
141add (16) g97.0<1>UW g97.0<16,16,1>UW g80.4<16,16,1>UB {align1};
142add (16) g98.0<1>UW g98.0<16,16,1>UW g82.4<16,16,1>UB {align1};
143add (16) g99.0<1>UW g99.0<16,16,1>UW g84.4<16,16,1>UB {align1};
144add (16) g100.0<1>UW g100.0<16,16,1>UW g86.4<16,16,1>UB {align1};
145add (16) g101.0<1>UW g101.0<16,16,1>UW g88.4<16,16,1>UB {align1};
146add (16) g102.0<1>UW g102.0<16,16,1>UW g90.4<16,16,1>UB {align1};
147add (16) g103.0<1>UW g103.0<16,16,1>UW g92.4<16,16,1>UB {align1};
148
149add (16) g96.0<1>UW g96.0<16,16,1>UW g80.4<16,16,1>UB {align1};
150add (16) g97.0<1>UW g97.0<16,16,1>UW g82.4<16,16,1>UB {align1};
151add (16) g98.0<1>UW g98.0<16,16,1>UW g84.4<16,16,1>UB {align1};
152add (16) g99.0<1>UW g99.0<16,16,1>UW g86.4<16,16,1>UB {align1};
153add (16) g100.0<1>UW g100.0<16,16,1>UW g88.4<16,16,1>UB {align1};
154add (16) g101.0<1>UW g101.0<16,16,1>UW g90.4<16,16,1>UB {align1};
155add (16) g102.0<1>UW g102.0<16,16,1>UW g92.4<16,16,1>UB {align1};
156add (16) g103.0<1>UW g103.0<16,16,1>UW g94.4<16,16,1>UB {align1};
157
158out:
159shr (16) g96.0<1>UW g96.0<16,16,1>UW 2UW {align1};
160shr (16) g97.0<1>UW g97.0<16,16,1>UW 2UW {align1};
161shr (16) g98.0<1>UW g98.0<16,16,1>UW 2UW {align1};
162shr (16) g99.0<1>UW g99.0<16,16,1>UW 2UW {align1};
163shr (16) g100.0<1>UW g100.0<16,16,1>UW 2UW {align1};
164shr (16) g101.0<1>UW g101.0<16,16,1>UW 2UW {align1};
165shr (16) g102.0<1>UW g102.0<16,16,1>UW 2UW {align1};
166shr (16) g103.0<1>UW g103.0<16,16,1>UW 2UW {align1};
167