2
* Copyright © 2009 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
* Yan Li <li.l.yan@intel.com>
26
* Liu Xi bin<xibin.liu@intel.com>
29
g1~g30: constant buffer
31
g3~g4:non intra IQ matrix
34
g58~g81:reference data
35
g82: thread payload backup
37
g32: message descriptor for reading reference data */
39
mov (8) g82.0<1>UD g31.0<8,8,1>UD {align1};
40
mov (2) g31.0<1>UD g82.12<2,2,1>UW {align1};
41
mov (1) g126.8<1>UD ip {align1};
42
mov (1) ip g21.0<1,1,1>UD {align1};
45
asr (2) g31.14<1>W g82.16<2,2,1>W 1W {align1};
46
add (2) g32.0<1>UD g31.0<2,2,1>UD g31.14<2,2,1>W {align1};
47
mov (1) g32.28<1>UD 0UD {align1};
48
and.nz (1) null g82.2<1,1,1>UW 0x1000UW {align1}; //motion vertical field select
49
(f0) mov (1) g32.28<1>UD 1UD {align1};
50
shl (1) g32.4<1>UD g32.4<1,1,1>UD 1UD {align1};
51
add (1) g32.4<1>UD g32.4<1,1,1>UD g32.28<1,1,1>UD {align1};
53
define(`mv1',`g82.16')
54
define(`mv2',`g82.18')
55
include(`field_motion_y.g4i')
56
mov (16) g108.0<1>UD g58.0<16,16,1>UD {align1 compr};
57
mov (16) g110.0<1>UD g60.0<16,16,1>UD {align1 compr};
58
mov (16) g112.0<1>UD g62.0<16,16,1>UD {align1 compr};
59
mov (16) g114.0<1>UD g64.0<16,16,1>UD {align1 compr};
60
mov (16) g116.0<1>UD g66.0<16,16,1>UD {align1 compr};
61
mov (16) g118.0<1>UD g68.0<16,16,1>UD {align1 compr};
62
mov (16) g120.0<1>UD g70.0<16,16,1>UD {align1 compr};
63
mov (16) g122.0<1>UD g72.0<16,16,1>UD {align1 compr};
66
asr (2) g31.14<1>W g82.20<2,2,1>W 1W {align1};
67
add (2) g32.0<1>UD g31.0<2,2,1>UD g31.14<2,2,1>W {align1};
68
mov (1) g32.28<1>UD 0UD {align1};
69
and.nz (1) null g82.2<1,1,1>UW 0x2000UW {align1}; //motion vertical field select
70
(f0) mov (1) g32.28<1>UD 1UD {align1};
71
shl (1) g32.4<1>UD g32.4<1,1,1>UD 1UD {align1};
72
add (1) g32.4<1>UD g32.4<1,1,1>UD g32.28<1,1,1>UD {align1};
74
define(`mv1',`g82.20')
75
define(`mv2',`g82.22')
76
include(`field_motion_y.g4i')
77
avg (16) g58.0<1>UW g58.0<16,16,1>UW g108.0<16,16,1>UW {align1};
78
avg (16) g59.0<1>UW g59.0<16,16,1>UW g109.0<16,16,1>UW {align1};
79
avg (16) g60.0<1>UW g60.0<16,16,1>UW g110.0<16,16,1>UW {align1};
80
avg (16) g61.0<1>UW g61.0<16,16,1>UW g111.0<16,16,1>UW {align1};
81
avg (16) g62.0<1>UW g62.0<16,16,1>UW g112.0<16,16,1>UW {align1};
82
avg (16) g63.0<1>UW g63.0<16,16,1>UW g113.0<16,16,1>UW {align1};
83
avg (16) g64.0<1>UW g64.0<16,16,1>UW g114.0<16,16,1>UW {align1};
84
avg (16) g65.0<1>UW g65.0<16,16,1>UW g115.0<16,16,1>UW {align1};
85
avg (16) g66.0<1>UW g66.0<16,16,1>UW g116.0<16,16,1>UW {align1};
86
avg (16) g67.0<1>UW g67.0<16,16,1>UW g117.0<16,16,1>UW {align1};
87
avg (16) g68.0<1>UW g68.0<16,16,1>UW g118.0<16,16,1>UW {align1};
88
avg (16) g69.0<1>UW g69.0<16,16,1>UW g119.0<16,16,1>UW {align1};
89
avg (16) g70.0<1>UW g70.0<16,16,1>UW g120.0<16,16,1>UW {align1};
90
avg (16) g71.0<1>UW g71.0<16,16,1>UW g121.0<16,16,1>UW {align1};
91
avg (16) g72.0<1>UW g72.0<16,16,1>UW g122.0<16,16,1>UW {align1};
92
avg (16) g73.0<1>UW g73.0<16,16,1>UW g123.0<16,16,1>UW {align1};
95
shr (2) g31.0<1>UD g31.0<2,2,1>UD 1UD {align1};
96
asr (2) g31.14<1>W g82.16<2,2,1>W 2W {align1};
97
add (2) g32.0<1>UD g31.0<2,2,1>UD g31.14<2,2,1>W {align1};
98
mov (1) g32.28<1>UD 0UD {align1};
99
and.nz (1) null g82.2<1,1,1>UW 0x1000UW {align1}; //motion vertical field select
100
(f0) mov (1) g32.28<1>UD 1UD {align1};
101
shl (1) g32.4<1>UD g32.4<1,1,1>UD 1UD {align1};
102
add (1) g32.4<1>UD g32.4<1,1,1>UD g32.28<1,1,1>UD {align1};
103
define(`surface_u', `5')
104
define(`surface_v', `6')
105
define(`mv1',`g82.16')
106
define(`mv2',`g82.18')
107
include(`field_motion_uv.g4i')
109
mov (16) g108.0<1>UD g74.0<16,16,1>UD {align1 compr};
110
mov (16) g110.0<1>UD g76.0<16,16,1>UD {align1 compr};
111
mov (16) g112.0<1>UD g78.0<16,16,1>UD {align1 compr};
112
mov (16) g114.0<1>UD g80.0<16,16,1>UD {align1 compr};
115
asr (2) g31.14<1>W g82.20<2,2,1>W 2W {align1};
116
add (2) g32.0<1>UD g31.0<2,2,1>UD g31.14<2,2,1>W {align1};
117
mov (1) g32.28<1>UD 0UD {align1};
118
and.nz (1) null g82.2<1,1,1>UW 0x2000UW {align1}; //motion vertical field select
119
(f0) mov (1) g32.28<1>UD 1UD {align1};
120
shl (1) g32.4<1>UD g32.4<1,1,1>UD 1UD {align1};
121
add (1) g32.4<1>UD g32.4<1,1,1>UD g32.28<1,1,1>UD {align1};
122
define(`surface_u', `8')
123
define(`surface_v', `9')
124
define(`mv1',`g82.20')
125
define(`mv2',`g82.22')
126
include(`field_motion_uv.g4i')
127
//Average Forward and Backward
128
avg (16) g74.0<1>UW g74.0<16,16,1>UW g108.0<16,16,1>UW {align1};
129
avg (16) g75.0<1>UW g75.0<16,16,1>UW g109.0<16,16,1>UW {align1};
130
avg (16) g76.0<1>UW g76.0<16,16,1>UW g110.0<16,16,1>UW {align1};
131
avg (16) g77.0<1>UW g77.0<16,16,1>UW g111.0<16,16,1>UW {align1};
132
avg (16) g78.0<1>UW g78.0<16,16,1>UW g112.0<16,16,1>UW {align1};
133
avg (16) g79.0<1>UW g79.0<16,16,1>UW g113.0<16,16,1>UW {align1};
134
avg (16) g80.0<1>UW g80.0<16,16,1>UW g114.0<16,16,1>UW {align1};
135
avg (16) g81.0<1>UW g81.0<16,16,1>UW g115.0<16,16,1>UW {align1};
138
include(`field_addidct.g4i')
140
send (16) 0 acc0<1>UW g0<8,8,1>UW
141
thread_spawner(0, 0, 0) mlen 1 rlen 0 { align1 EOT};