1/*
2 * Copyright © 2009 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 *    Yan Li <li.l.yan@intel.com>
26 *    Liu Xi bin<xibin.liu@intel.com>
27 */
28/* GRF allocation:
29   g1~g30: constant buffer
30           g1~g2:intra IQ matrix
31           g3~g4:non intra IQ matrix
32           g5~g20:IDCT table
33   g31:    thread payload 
34   g32:    message descriptor for reading reference data
35   g58~g81:reference data
36   g82:    thread payload backup
37   g83~g106:IDCT data                           */
38mov (1) g32.8<1>UD 0x007001FUD {align1};
39send (16) 0 g38.0<1>UW g32<8,8,1>UW read(input_surface, 2, 0, 2) mlen 1 rlen 8 {align1};
40add (1) g32.4<1>UD g32.4<1,1,1>UD 8UD {align1};
41send (16) 0 g46.0<1>UW g32<8,8,1>UW read(input_surface, 2, 0, 2) mlen 1 rlen 8 {align1};
42add (1) g32.4<1>UD g32.4<1,1,1>UD 8UD {align1};
43mov (1) g32.8<1>UD 0x1FUD {align1};
44send (16) 0 g54.0<1>UW g32<8,8,1>UW read(input_surface, 2, 0, 2) mlen 1 rlen 1 {align1};
45
46add (16) g58.0<1>UW g38.0<16,16,1>UB g38.1<16,16,1>UB {align1};
47add (16) g59.0<1>UW g39.0<16,16,1>UB g39.1<16,16,1>UB {align1};
48add (16) g60.0<1>UW g40.0<16,16,1>UB g40.1<16,16,1>UB {align1};
49add (16) g61.0<1>UW g41.0<16,16,1>UB g41.1<16,16,1>UB {align1};
50add (16) g62.0<1>UW g42.0<16,16,1>UB g42.1<16,16,1>UB {align1};
51add (16) g63.0<1>UW g43.0<16,16,1>UB g43.1<16,16,1>UB {align1};
52add (16) g64.0<1>UW g44.0<16,16,1>UB g44.1<16,16,1>UB {align1};
53add (16) g65.0<1>UW g45.0<16,16,1>UB g45.1<16,16,1>UB {align1};
54add (16) g66.0<1>UW g46.0<16,16,1>UB g46.1<16,16,1>UB {align1};
55add (16) g67.0<1>UW g47.0<16,16,1>UB g47.1<16,16,1>UB {align1};
56add (16) g68.0<1>UW g48.0<16,16,1>UB g48.1<16,16,1>UB {align1};
57add (16) g69.0<1>UW g49.0<16,16,1>UB g49.1<16,16,1>UB {align1};
58add (16) g70.0<1>UW g50.0<16,16,1>UB g50.1<16,16,1>UB {align1};
59add (16) g71.0<1>UW g51.0<16,16,1>UB g51.1<16,16,1>UB {align1};
60add (16) g72.0<1>UW g52.0<16,16,1>UB g52.1<16,16,1>UB {align1};
61add (16) g73.0<1>UW g53.0<16,16,1>UB g53.1<16,16,1>UB {align1};
62
63add (16) g58.0<1>UW g58.0<16,16,1>UW g39.0<16,16,1>UB {align1};
64add (16) g59.0<1>UW g59.0<16,16,1>UW g40.0<16,16,1>UB {align1};
65add (16) g60.0<1>UW g60.0<16,16,1>UW g41.0<16,16,1>UB {align1};
66add (16) g61.0<1>UW g61.0<16,16,1>UW g42.0<16,16,1>UB {align1};
67add (16) g62.0<1>UW g62.0<16,16,1>UW g43.0<16,16,1>UB {align1};
68add (16) g63.0<1>UW g63.0<16,16,1>UW g44.0<16,16,1>UB {align1};
69add (16) g64.0<1>UW g64.0<16,16,1>UW g45.0<16,16,1>UB {align1};
70add (16) g65.0<1>UW g65.0<16,16,1>UW g46.0<16,16,1>UB {align1};
71add (16) g66.0<1>UW g66.0<16,16,1>UW g47.0<16,16,1>UB {align1};
72add (16) g67.0<1>UW g67.0<16,16,1>UW g48.0<16,16,1>UB {align1};
73add (16) g68.0<1>UW g68.0<16,16,1>UW g49.0<16,16,1>UB {align1};
74add (16) g69.0<1>UW g69.0<16,16,1>UW g50.0<16,16,1>UB {align1};
75add (16) g70.0<1>UW g70.0<16,16,1>UW g51.0<16,16,1>UB {align1};
76add (16) g71.0<1>UW g71.0<16,16,1>UW g52.0<16,16,1>UB {align1};
77add (16) g72.0<1>UW g72.0<16,16,1>UW g53.0<16,16,1>UB {align1};
78add (16) g73.0<1>UW g73.0<16,16,1>UW g54.0<16,16,1>UB {align1};
79
80add (16) g58.0<1>UW g58.0<16,16,1>UW g39.1<16,16,1>UB {align1};
81add (16) g59.0<1>UW g59.0<16,16,1>UW g40.1<16,16,1>UB {align1};
82add (16) g60.0<1>UW g60.0<16,16,1>UW g41.1<16,16,1>UB {align1};
83add (16) g61.0<1>UW g61.0<16,16,1>UW g42.1<16,16,1>UB {align1};
84add (16) g62.0<1>UW g62.0<16,16,1>UW g43.1<16,16,1>UB {align1};
85add (16) g63.0<1>UW g63.0<16,16,1>UW g44.1<16,16,1>UB {align1};
86add (16) g64.0<1>UW g64.0<16,16,1>UW g45.1<16,16,1>UB {align1};
87add (16) g65.0<1>UW g65.0<16,16,1>UW g46.1<16,16,1>UB {align1};
88add (16) g66.0<1>UW g66.0<16,16,1>UW g47.1<16,16,1>UB {align1};
89add (16) g67.0<1>UW g67.0<16,16,1>UW g48.1<16,16,1>UB {align1};
90add (16) g68.0<1>UW g68.0<16,16,1>UW g49.1<16,16,1>UB {align1};
91add (16) g69.0<1>UW g69.0<16,16,1>UW g50.1<16,16,1>UB {align1};
92add (16) g70.0<1>UW g70.0<16,16,1>UW g51.1<16,16,1>UB {align1};
93add (16) g71.0<1>UW g71.0<16,16,1>UW g52.1<16,16,1>UB {align1};
94add (16) g72.0<1>UW g72.0<16,16,1>UW g53.1<16,16,1>UB {align1};
95add (16) g73.0<1>UW g73.0<16,16,1>UW g54.1<16,16,1>UB {align1};
96
97shr.sat (16) g58.0<1>UW g58.0<16,16,1>UW 2UW {align1};
98shr.sat (16) g59.0<1>UW g59.0<16,16,1>UW 2UW {align1};
99shr.sat (16) g60.0<1>UW g60.0<16,16,1>UW 2UW {align1};
100shr.sat (16) g61.0<1>UW g61.0<16,16,1>UW 2UW {align1};
101shr.sat (16) g62.0<1>UW g62.0<16,16,1>UW 2UW {align1};
102shr.sat (16) g63.0<1>UW g63.0<16,16,1>UW 2UW {align1};
103shr.sat (16) g64.0<1>UW g64.0<16,16,1>UW 2UW {align1};
104shr.sat (16) g65.0<1>UW g65.0<16,16,1>UW 2UW {align1};
105shr.sat (16) g66.0<1>UW g66.0<16,16,1>UW 2UW {align1};
106shr.sat (16) g67.0<1>UW g67.0<16,16,1>UW 2UW {align1};
107shr.sat (16) g68.0<1>UW g68.0<16,16,1>UW 2UW {align1};
108shr.sat (16) g69.0<1>UW g69.0<16,16,1>UW 2UW {align1};
109shr.sat (16) g70.0<1>UW g70.0<16,16,1>UW 2UW {align1};
110shr.sat (16) g71.0<1>UW g71.0<16,16,1>UW 2UW {align1};
111shr.sat (16) g72.0<1>UW g72.0<16,16,1>UW 2UW {align1};
112shr.sat (16) g73.0<1>UW g73.0<16,16,1>UW 2UW {align1};
113