1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
|
/*
* Copyright © <2010>, Intel Corporation.
*
* This program is licensed under the terms and conditions of the
* Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
* http://www.opensource.org/licenses/eclipse-1.0.php.
*
*/
#if !defined(__AVC_ILDB_CHROMA_CORE__) // Make sure this file is only included once
#define __AVC_ILDB_CHROMA_CORE__
////////// AVC ILDB Chroma Core /////////////////////////////////////////////////////////////////////////////////
//
// This core performs AVC U or V ILDB filtering on one horizontal edge (8 pixels) of a MB.
// If data is transposed, it can also de-block a vertical edge.
//
// Bafore calling this subroutine, caller needs to set the following parameters.
//
// - EdgeCntlMap1 // Edge control map A
// - EdgeCntlMap2 // Edge control map B
// - P_AddrReg // Src and dest address register for P pixels
// - Q_AddrReg // Src and dest address register for Q pixels
// - alpha // alpha corresponding to the edge to be filtered
// - beta // beta corresponding to the edge to be filtered
// - tc0 // tc0 corresponding to the edge to be filtered
//
// U or V:
// +----+----+----+----+
// | P1 | p0 | q0 | q1 |
// +----+----+----+----+
//
// p1 = r[P_AddrReg, 0]<16;8,2>
// p0 = r[P_AddrReg, 16]<16;8,2>
// q0 = r[Q_AddrReg, 0]<16;8,2>
// q1 = r[Q_AddrReg, 16]<16;8,2>
//
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// The region is both src and dest
// P0-P3 and Q0-Q3 should be only used if they have not been modified to new values
#undef P1
#undef P0
#undef Q0
#undef Q1
#define P1 r[P_AddrReg, 0]<16;8,2>:ub
#define P0 r[P_AddrReg, 16]<16;8,2>:ub
#define Q0 r[Q_AddrReg, 0]<16;8,2>:ub
#define Q1 r[Q_AddrReg, 16]<16;8,2>:ub
// New region as dest
#undef NewP0
#undef NewQ0
#define NewP0 r[P_AddrReg, 16]<2>:ub
#define NewQ0 r[Q_AddrReg, 0]<2>:ub
// Filter one chroma edge
FILTER_UV:
#if defined(_DEBUG)
mov (1) EntrySignatureC:w 0x1112:w
#endif
//---------- Derive filterSampleflag in AVC spec, equition (8-469) ----------
// bS is in MaskA
// Src copy of the p1, p0, q0, q1
// mov (8) p1(0)<1> r[P_AddrReg, 0]<16;8,2>:ub
// mov (8) p0(0)<1> r[P_AddrReg, 16]<16;8,2>:ub
// mov (8) q0(0)<1> r[Q_AddrReg, 0]<16;8,2>:ub
// mov (8) q1(0)<1> r[Q_AddrReg, 16]<16;8,2>:ub
// mov (1) f0.0:uw MaskA:uw
add (8) q0_p0(0)<1> Q0 -P0 // q0-p0
add (8) TempRow0(0)<1> P1 -P0 // p1-p0
add (8) TempRow1(0)<1> Q1 -Q0 // q1-q0
// Build FilterSampleFlag
// abs(q0-p0) < alpha
(f0.0) cmp.l.f0.0 (16) null:w (abs)q0_p0(0) alpha:w
// abs(p1-p0) < Beta
(f0.0) cmp.l.f0.0 (16) null:w (abs)TempRow0(0) beta:w
// abs(q1-q0) < Beta
(f0.0) cmp.l.f0.0 (16) null:w (abs)TempRow1(0) beta:w
//-----------------------------------------------------------------------------------------
// if
(f0.0) if (8) UV_ENDIF1
// For channels whose edge control map1 = 1 ---> perform de-blocking
// mov (1) f0.1:w MaskB:w {NoMask} // Now check for which algorithm to apply
(f0.1) if (8) UV_ELSE2
// For channels whose edge control map2 = 1 ---> bS = 4 algorithm
// p0' = (2*p1 + p0 + q1 + 2) >> 2
// q0' = (2*q1 + q0 + p1 + 2) >> 2
// Optimized version:
// A = (p1 + q1 + 2)
// p0' = (p0 + p1 + A) >> 2
// q0' = (q0 + q1 + A) >> 2
//------------------------------------------------------------------------------------
// p0' = (2*p1 + p0 + q1 + 2) >> 2
add (8) acc0<1>:w Q1 2:w
mac (8) acc0<1>:w P1 2:w
add (8) acc0<1>:w acc0<8;8,1>:w P0
shr.sat (8) TempRow0B(0)<2> acc0<8;8,1>:w 2:w
// q0' = (2*q1 + q0 + p1 + 2) >> 2
add (8) acc0<1>:w P1 2:w
mac (8) acc0<1>:w Q1 2:w
add (8) acc0<1>:w acc0<8;8,1>:w Q0
shr.sat (8) TempRow1B(0)<2> acc0<8;8,1>:w 2:w
mov (8) NewP0 TempRow0B(0) // p0'
mov (8) NewQ0 TempRow1B(0) // q0'
UV_ELSE2:
else (8) UV_ENDIF2
// For channels whose edge control map2 = 0 ---> bS < 4 algorithm
// Expand tc0 (tc0 has 4 bytes)
// mov (8) tc0_exp(0)<1> tc0<1;2,0>:ub {NoMask} // tc0_exp = tc0, each tc0 is duplicated 2 times for 2 adjcent pixels
mov (8) acc0<1>:w tc0<1;2,0>:ub {NoMask} // tc0_exp = tc0, each tc0 is duplicated 2 times for 2 adjcent pixels
// tc_exp = tc0_exp + 1
// add (8) tc_exp(0)<1> tc0_exp(0) 1:w
add (8) tc_exp(0)<1> acc0<8;8,1>:w 1:w
// delta = Clip3(-tc, tc, ((((q0 - p0)<<2) + (p1-q1) + 4) >> 3))
// 4 * (q0-p0) + p1 - q1 + 4
add (8) acc0<1>:w P1 4:w
mac (8) acc0<1>:w q0_p0(0) 4:w
add (8) acc0<1>:w acc0<8;8,1>:w -Q1
shr (8) TempRow0(0)<1> acc0<8;8,1>:w 3:w
// tc clip
cmp.g.f0.0 (8) null:w TempRow0(0) tc_exp(0) // Clip if > tc0
cmp.l.f0.1 (8) null:w TempRow0(0) -tc_exp(0) // Clip if < -tc0
(f0.0) mov (8) TempRow0(0)<1> tc_exp(0)
(f0.1) mov (8) TempRow0(0)<1> -tc_exp(0)
// p0' = Clip1(p0 + delta) = Clip3(0, 0xFF, p0 + delta)
add.sat (8) TempRow1B(0)<2> P0 TempRow0(0) // p0+delta
// q0' = Clip1(q0 - delta) = Clip3(0, 0xFF, q0 - delta)
add.sat (8) TempRow0B(0)<2> Q0 -TempRow0(0) // q0-delta
mov (8) NewP0 TempRow1B(0) // p0'
mov (8) NewQ0 TempRow0B(0) // q0'
endif
UV_ENDIF2:
UV_ENDIF1:
endif
RETURN
#endif // !defined(__AVC_ILDB_CHROMA_CORE__)
|
