2 * Copyright © <2010>, Intel Corporation.
\r
4 * This program is licensed under the terms and conditions of the
\r
5 * Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
\r
6 * http://www.opensource.org/licenses/eclipse-1.0.php.
\r
9 ////////// AVC ILDB filter horizontal UV ///////////////////////////////////////////////////////
\r
11 // This filter code prepares the src data and control data for ILDB filtering on all horizontal edges of UV.
\r
13 // It sssumes the data for horizontal de-blocking is already transposed.
\r
17 // +-------+-------+ H0 Edge
\r
21 // +-------+-------+ H1 Edge
\r
25 // +-------+-------+
\r
27 /////////////////////////////////////////////////////////////////////////////
\r
29 #if defined(_DEBUG)
\r
30 mov (1) EntrySignatureC:w 0xBBBC:w
\r
33 //=============== Chroma deblocking ================
\r
35 //---------- Deblock U external top edge ----------
\r
36 and.z.f0.0 (1) null:w r[ECM_AddrReg, BitFlags]:ub FilterTopMbEdgeFlag:w // Check for FilterTopMbEdgeFlag
\r
37 // (f0.0) jmpi BYPASS_EXT_TOP_EDGE_UV
\r
39 // Get horizontal border edge control data.
\r
41 //***** Need to take every other bit to form U maskA and mask B
\r
42 // Get Luma maskA and maskB
\r
43 shr (16) TempRow0(0)<1> r[ECM_AddrReg, wEdgeCntlMapA_ExtTopHorz0]<0;1,0>:uw RRampW(0)
\r
44 shr (16) TempRow1(0)<1> r[ECM_AddrReg, wEdgeCntlMapB_ExtTopHorz0]<0;1,0>:uw RRampW(0)
\r
46 (f0.0) jmpi ILDB_LABEL(BYPASS_EXT_TOP_EDGE_UV)
\r
48 // Extract UV MaskA and MaskB from every other bit of Y masks
\r
49 and.nz.f0.0 (8) null:w TempRow0(0)<16;8,2> 1:w
\r
50 and.nz.f0.1 (8) null:w TempRow1(0)<16;8,2> 1:w
\r
52 //---------- Deblock U external edge ----------
\r
53 // p1 = Prev MB U row 0
\r
54 // p0 = Prev MB U row 1
\r
55 // q0 = Cur MB U row 0
\r
56 // q1 = Cur MB U row 1
\r
57 // mov (1) P_AddrReg:w PREV_MB_U_BASE:w { NoDDClr }
\r
58 mov (1) P_AddrReg:w TOP_MB_U_BASE:w { NoDDClr }
\r
59 mov (1) Q_AddrReg:w SRC_MB_U_BASE:w { NoDDChk }
\r
61 // alpha = bAlphaTop0_Cb, beta = bBetaTop0_Cb
\r
62 mov (2) alpha<1>:w r[ECM_AddrReg, bAlphaTop0_Cb]<2;2,1>:ub { NoDDClr }
\r
63 // tc0 has bTc0_h03_0_Cb + bTc0_h02_0_Cb + bTc0_h01_0_Cb + bTc0_h00_0_Cb
\r
64 mov (4) tc0<1>:ub r[ECM_AddrReg, bTc0_h00_0_Cb]<4;4,1>:ub { NoDDChk }
\r
66 // UV MaskA and MaskB
\r
67 mov (2) MaskA<1>:uw f0.0<2;2,1>:uw
\r
71 //---------- Deblock V external top edge ----------
\r
72 // p1 = Prev MB V row 0
\r
73 // p0 = Prev MB V row 1
\r
74 // q0 = Cur MB V row 0
\r
75 // q1 = Cur MB V row 1
\r
76 // mov (1) P_AddrReg:w PREV_MB_V_BASE:w { NoDDClr }
\r
77 mov (1) P_AddrReg:w TOP_MB_V_BASE:w { NoDDClr }
\r
78 mov (1) Q_AddrReg:w SRC_MB_V_BASE:w { NoDDChk }
\r
80 // alpha = bAlphaTop0_Cr, beta = bBetaTop0_Cr
\r
81 mov (2) alpha<1>:w r[ECM_AddrReg, bAlphaTop0_Cr]<2;2,1>:ub { NoDDClr }
\r
83 // tc0 has bTc0_h03_0_Cr + bTc0_h02_0_Cr + bTc0_h01_0_Cr + bTc0_h00_0_Cr
\r
84 mov (4) tc0<1>:ub r[ECM_AddrReg, bTc0_h00_0_Cr]<4;4,1>:ub { NoDDChk }
\r
86 // UV MaskA and MaskB
\r
87 mov (2) f0.0<1>:uw MaskA<2;2,1>:uw
\r
91 ILDB_LABEL(BYPASS_EXT_TOP_EDGE_UV):
\r
93 // Set EdgeCntlMap2 = 0, so it always uses bS < 4 algorithm.
\r
95 // Bypass deblocking if FilterInternal4x4EdgesFlag = 0
\r
96 // and.z.f0.0 (1) null:w r[ECM_AddrReg, BitFlags]:ub FilterInternal4x4EdgesFlag:w // Check for FilterInternal4x4EdgesFlag
\r
97 // (f0.0) jmpi BYPASS_4x4_DEBLOCK_H
\r
99 //---------- Deblock U internal horz middle edge ----------
\r
101 //***** Need to take every other bit to form U maskA
\r
102 // Get Luma maskA and maskB
\r
103 shr (16) TempRow0(0)<1> r[ECM_AddrReg, wEdgeCntlMap_IntMidHorz]<0;1,0>:uw RRampW(0)
\r
105 // p1 = Cur MB U row 2
\r
106 // p0 = Cur MB U row 3
\r
107 // q0 = Cur MB U row 4
\r
108 // q1 = Cur MB U row 5
\r
109 mov (1) P_AddrReg:w 4*UV_ROW_WIDTH+SRC_MB_U_BASE:w { NoDDClr } // Skip 2 U rows and 2 V rows
\r
110 mov (1) Q_AddrReg:w 8*UV_ROW_WIDTH+SRC_MB_U_BASE:w { NoDDChk }
\r
112 // alpha = bAlphaInternal_Cb, beta = bBetaInternal_Cb
\r
113 mov (2) alpha<1>:w r[ECM_AddrReg, bAlphaInternal_Cb]<2;2,1>:ub { NoDDClr }
\r
114 // tc0 has bTc0_h23_Cb + bTc0_h22_Cb + bTc0_h21_Cb + bTc0_h20_Cb
\r
115 mov (4) tc0<1>:ub r[ECM_AddrReg, bTc0_h20_Cb]<4;4,1>:ub { NoDDChk }
\r
117 // Extract UV MaskA and MaskB from every other bit of Y masks
\r
118 and.nz.f0.0 (8) null:w TempRow0(0)<16;8,2> 1:w
\r
120 // UV MaskA and MaskB
\r
121 mov (1) f0.1:uw 0:w
\r
122 mov (1) MaskB:uw 0:w { NoDDClr }
\r
123 mov (1) MaskA:uw f0.0:uw { NoDDChk }
\r
125 CALL(FILTER_UV, 1)
\r
127 //---------- Deblock V internal horz middle edge ----------
\r
128 // p1 = Cur MB V row 2
\r
129 // p0 = Cur MB V row 3
\r
130 // q0 = Cur MB V row 4
\r
131 // q1 = Cur MB V row 5
\r
132 mov (1) P_AddrReg:w 4*UV_ROW_WIDTH+SRC_MB_V_BASE:w { NoDDClr } // Skip 2 U rows and 2 V rows
\r
133 mov (1) Q_AddrReg:w 8*UV_ROW_WIDTH+SRC_MB_V_BASE:w { NoDDChk }
\r
135 // alpha = bAlphaInternal_Cr, beta = bBetaInternal_Cr
\r
136 mov (2) alpha<1>:w r[ECM_AddrReg, bAlphaInternal_Cr]<2;2,1>:ub { NoDDClr }
\r
137 // tc0 has bTc0_h23_Cr + bTc0_h22_Cr + bTc0_h21_Cr + bTc0_h20_Cr
\r
138 mov (4) tc0<1>:ub r[ECM_AddrReg, bTc0_h20_Cr]<4;4,1>:ub { NoDDChk }
\r
140 // UV MaskA and MaskB
\r
141 mov (2) f0.0<1>:uw MaskA<2;2,1>:uw
\r
143 CALL(FILTER_UV, 1)
\r
145 //BYPASS_4x4_DEBLOCK_H:
\r