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) g2.8<1>UD 0x007001FUD {align1};
30send (16) 0 g98.0<1>UW g2<8,8,1>UW read(input_surface, 2, 0, 2) mlen 1 rlen 8 {align1};
31
32and (1) g2.24<1>UD g2.0<1,1,1>UD 3UD {align1};
33mul(1) g2.24<1>UD g2.24<1,1,1>UD 17UD {align1};
34add (1) g2.4<1>UD g2.4<1,1,1>UD 8UD {align1};
35send (16) 0 g106.0<1>UW g2<8,8,1>UW read(input_surface, 2, 0, 2) mlen 1 rlen 8 {align1};
36add (1) g2.4<1>UD g2.4<1,1,1>UD 8UD {align1};
37mov (1) g2.8<1>UD 0x1FUD {align1};
38send (16) 0 g120.0<1>UW g2<8,8,1>UW read(input_surface, 2, 0, 2) mlen 1 rlen 1 {align1};
39
40jmpi g2.24<1,1,1>D;
41
42avg.sat (16) g28.0<1>UW g98.0<16,16,1>UB g99.0<16,16,1>UB {align1};
43avg.sat (16) g29.0<1>UW g99.0<16,16,1>UB g100.0<16,16,1>UB {align1};
44avg.sat (16) g30.0<1>UW g100.0<16,16,1>UB g101.0<16,16,1>UB {align1};
45avg.sat (16) g31.0<1>UW g101.0<16,16,1>UB g102.0<16,16,1>UB {align1};
46avg.sat (16) g32.0<1>UW g102.0<16,16,1>UB g103.0<16,16,1>UB {align1};
47avg.sat (16) g33.0<1>UW g103.0<16,16,1>UB g104.0<16,16,1>UB {align1};
48avg.sat (16) g34.0<1>UW g104.0<16,16,1>UB g105.0<16,16,1>UB {align1};
49avg.sat (16) g35.0<1>UW g105.0<16,16,1>UB g106.0<16,16,1>UB {align1};
50avg.sat (16) g36.0<1>UW g106.0<16,16,1>UB g107.0<16,16,1>UB {align1};
51avg.sat (16) g37.0<1>UW g107.0<16,16,1>UB g108.0<16,16,1>UB {align1};
52avg.sat (16) g38.0<1>UW g108.0<16,16,1>UB g109.0<16,16,1>UB {align1};
53avg.sat (16) g39.0<1>UW g109.0<16,16,1>UB g110.0<16,16,1>UB {align1};
54avg.sat (16) g40.0<1>UW g110.0<16,16,1>UB g111.0<16,16,1>UB {align1};
55avg.sat (16) g41.0<1>UW g111.0<16,16,1>UB g112.0<16,16,1>UB {align1};
56avg.sat (16) g42.0<1>UW g112.0<16,16,1>UB g113.0<16,16,1>UB {align1};
57avg.sat (16) g43.0<1>UW g113.0<16,16,1>UB g120.0<16,16,1>UB {align1};
58jmpi out;
59
60avg.sat (16) g28.0<1>UW g98.1<16,16,1>UB g99.1<16,16,1>UB {align1};
61avg.sat (16) g29.0<1>UW g99.1<16,16,1>UB g100.1<16,16,1>UB {align1};
62avg.sat (16) g30.0<1>UW g100.1<16,16,1>UB g101.1<16,16,1>UB {align1};
63avg.sat (16) g31.0<1>UW g101.1<16,16,1>UB g102.1<16,16,1>UB {align1};
64avg.sat (16) g32.0<1>UW g102.1<16,16,1>UB g103.1<16,16,1>UB {align1};
65avg.sat (16) g33.0<1>UW g103.1<16,16,1>UB g104.1<16,16,1>UB {align1};
66avg.sat (16) g34.0<1>UW g104.1<16,16,1>UB g105.1<16,16,1>UB {align1};
67avg.sat (16) g35.0<1>UW g105.1<16,16,1>UB g106.1<16,16,1>UB {align1};
68avg.sat (16) g36.0<1>UW g106.1<16,16,1>UB g107.1<16,16,1>UB {align1};
69avg.sat (16) g37.0<1>UW g107.1<16,16,1>UB g108.1<16,16,1>UB {align1};
70avg.sat (16) g38.0<1>UW g108.1<16,16,1>UB g109.1<16,16,1>UB {align1};
71avg.sat (16) g39.0<1>UW g109.1<16,16,1>UB g110.1<16,16,1>UB {align1};
72avg.sat (16) g40.0<1>UW g110.1<16,16,1>UB g111.1<16,16,1>UB {align1};
73avg.sat (16) g41.0<1>UW g111.1<16,16,1>UB g112.1<16,16,1>UB {align1};
74avg.sat (16) g42.0<1>UW g112.1<16,16,1>UB g113.1<16,16,1>UB {align1};
75avg.sat (16) g43.0<1>UW g113.1<16,16,1>UB g120.1<16,16,1>UB {align1};
76jmpi out;
77
78avg.sat (16) g28.0<1>UW g98.2<16,16,1>UB g99.2<16,16,1>UB {align1};
79avg.sat (16) g29.0<1>UW g99.2<16,16,1>UB g100.2<16,16,1>UB {align1};
80avg.sat (16) g30.0<1>UW g100.2<16,16,1>UB g101.2<16,16,1>UB {align1};
81avg.sat (16) g31.0<1>UW g101.2<16,16,1>UB g102.2<16,16,1>UB {align1};
82avg.sat (16) g32.0<1>UW g102.2<16,16,1>UB g103.2<16,16,1>UB {align1};
83avg.sat (16) g33.0<1>UW g103.2<16,16,1>UB g104.2<16,16,1>UB {align1};
84avg.sat (16) g34.0<1>UW g104.2<16,16,1>UB g105.2<16,16,1>UB {align1};
85avg.sat (16) g35.0<1>UW g105.2<16,16,1>UB g106.2<16,16,1>UB {align1};
86avg.sat (16) g36.0<1>UW g106.2<16,16,1>UB g107.2<16,16,1>UB {align1};
87avg.sat (16) g37.0<1>UW g107.2<16,16,1>UB g108.2<16,16,1>UB {align1};
88avg.sat (16) g38.0<1>UW g108.2<16,16,1>UB g109.2<16,16,1>UB {align1};
89avg.sat (16) g39.0<1>UW g109.2<16,16,1>UB g110.2<16,16,1>UB {align1};
90avg.sat (16) g40.0<1>UW g110.2<16,16,1>UB g111.2<16,16,1>UB {align1};
91avg.sat (16) g41.0<1>UW g111.2<16,16,1>UB g112.2<16,16,1>UB {align1};
92avg.sat (16) g42.0<1>UW g112.2<16,16,1>UB g113.2<16,16,1>UB {align1};
93avg.sat (16) g43.0<1>UW g113.2<16,16,1>UB g120.2<16,16,1>UB {align1};
94jmpi out;
95
96avg.sat (16) g28.0<1>UW g98.3<16,16,1>UB g99.3<16,16,1>UB {align1};
97avg.sat (16) g29.0<1>UW g99.3<16,16,1>UB g100.3<16,16,1>UB {align1};
98avg.sat (16) g30.0<1>UW g100.3<16,16,1>UB g101.3<16,16,1>UB {align1};
99avg.sat (16) g31.0<1>UW g101.3<16,16,1>UB g102.3<16,16,1>UB {align1};
100avg.sat (16) g32.0<1>UW g102.3<16,16,1>UB g103.3<16,16,1>UB {align1};
101avg.sat (16) g33.0<1>UW g103.3<16,16,1>UB g104.3<16,16,1>UB {align1};
102avg.sat (16) g34.0<1>UW g104.3<16,16,1>UB g105.3<16,16,1>UB {align1};
103avg.sat (16) g35.0<1>UW g105.3<16,16,1>UB g106.3<16,16,1>UB {align1};
104avg.sat (16) g36.0<1>UW g106.3<16,16,1>UB g107.3<16,16,1>UB {align1};
105avg.sat (16) g37.0<1>UW g107.3<16,16,1>UB g108.3<16,16,1>UB {align1};
106avg.sat (16) g38.0<1>UW g108.3<16,16,1>UB g109.3<16,16,1>UB {align1};
107avg.sat (16) g39.0<1>UW g109.3<16,16,1>UB g110.3<16,16,1>UB {align1};
108avg.sat (16) g40.0<1>UW g110.3<16,16,1>UB g111.3<16,16,1>UB {align1};
109avg.sat (16) g41.0<1>UW g111.3<16,16,1>UB g112.3<16,16,1>UB {align1};
110avg.sat (16) g42.0<1>UW g112.3<16,16,1>UB g113.3<16,16,1>UB {align1};
111avg.sat (16) g43.0<1>UW g113.3<16,16,1>UB g120.3<16,16,1>UB {align1};
112
113out:
114