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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
|
/*
* 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.
*
*/
// Modual name: IntraFrame_gen8.asm
//
// Make intra predition estimation for Intra frame on Gen8
//
//
// Now, begin source code....
//
/*
* __START
*/
__INTRA_START:
mov (16) tmp_reg0.0<1>:UD 0x0:UD {align1};
mov (16) tmp_reg2.0<1>:UD 0x0:UD {align1};
mov (16) tmp_reg4.0<1>:UD 0x0:UD {align1} ;
mov (16) tmp_reg6.0<1>:UD 0x0:UD {align1} ;
shl (2) read0_header.0<1>:D orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* (x, y) * 16 */
add (1) read0_header.0<1>:D read0_header.0<0,1,0>:D -8:W {align1}; /* X offset */
add (1) read0_header.4<1>:D read0_header.4<0,1,0>:D -1:W {align1}; /* Y offset */
mov (1) read0_header.8<1>:UD BLOCK_32X1 {align1};
mov (1) read0_header.20<1>:UB thread_id_ub {align1}; /* dispatch id */
shl (2) read1_header.0<1>:D orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* (x, y) * 16 */
add (1) read1_header.0<1>:D read1_header.0<0,1,0>:D -4:W {align1}; /* X offset */
mov (1) read1_header.8<1>:UD BLOCK_4X16 {align1};
mov (1) read1_header.20<1>:UB thread_id_ub {align1}; /* dispatch id */
shl (2) vme_m0.8<1>:UW orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* (x, y) * 16 */
mov (1) vme_m0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
mul (1) obw_m0.8<1>:UD w_in_mb_uw<0,1,0>:UW orig_y_ub<0,1,0>:UB {align1};
add (1) obw_m0.8<1>:UD obw_m0.8<0,1,0>:UD orig_x_ub<0,1,0>:UB {align1};
mul (1) obw_m0.8<1>:UD obw_m0.8<0,1,0>:UD 0x02:UD {align1};
mov (1) obw_m0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
/*
* Media Read Message -- fetch Luma neighbor edge pixels
*/
/* ROW */
mov (8) msg_reg0.0<1>:UD read0_header.0<8,8,1>:UD {align1};
send (8) msg_ind INEP_ROW<1>:UB null read(BIND_IDX_INEP, 0, 0, 4) mlen 1 rlen 1 {align1};
/* COL */
mov (8) msg_reg0.0<1>:UD read1_header.0<8,8,1>:UD {align1};
send (8) msg_ind INEP_COL0<1>:UB null read(BIND_IDX_INEP, 0, 0, 4) mlen 1 rlen 2 {align1};
/*
* Media Read Message -- fetch Chroma neighbor edge pixels
*/
/* ROW */
shl (2) read0_header.0<1>:D orig_xy_ub<2,2,1>:UB 3:UW {align1}; /* x * 16 , y * 8 */
mul (1) read0_header.0<1>:D read0_header.0<0,1,0>:D 2:W {align1};
add (1) read0_header.0<1>:D read0_header.0<0,1,0>:D -8:W {align1}; /* X offset */
add (1) read0_header.4<1>:D read0_header.4<0,1,0>:D -1:W {align1}; /* Y offset */
mov (8) msg_reg0.0<1>:UD read0_header.0<8,8,1>:UD {align1};
send (8) msg_ind CHROMA_ROW<1>:UB null read(BIND_IDX_CBCR, 0, 0, 4) mlen 1 rlen 1 {align1};
/* COL */
shl (2) read1_header.0<1>:D orig_xy_ub<2,2,1>:UB 3:UW {align1}; /* x * 16, y * 8 */
mul (1) read1_header.0<1>:D read1_header.0<0,1,0>:D 2:W {align1};
add (1) read1_header.0<1>:D read1_header.0<0,1,0>:D -4:W {align1}; /* X offset */
mov (1) read1_header.8<1>:UD BLOCK_8X4 {align1};
mov (8) msg_reg0.0<1>:UD read1_header.0<8,8,1>:UD {align1};
send (8) msg_ind CHROMA_COL<1>:UB null read(BIND_IDX_CBCR, 0, 0, 4) mlen 1 rlen 1 {align1};
/* m2, get the MV/Mb cost passed by constant buffer
when creating EU thread by MEDIA_OBJECT */
mov (8) vme_msg_2<1>:UD r1.0<8,8,1>:UD {align1};
/* m3. This is changed for FWD/BWD cost center */
mov (8) vme_msg_3<1>:UD 0x0:UD {align1};
/* m4.*/
mov (8) vme_msg_4<1>:ud 0x0:ud {align1};
/* m5 */
mov (1) INEP_ROW.0<1>:UD 0x0:UD {align1};
and (1) INEP_ROW.4<1>:UD INEP_ROW.4<0,1,0>:UD 0xFF000000:UD {align1};
mov (8) vme_msg_5<1>:UD INEP_ROW.0<8,8,1>:UD {align1};
mov (1) tmp_reg0.0<1>:UW LUMA_CHROMA_MODE:UW {align1};
/* Use the Luma mode */
mov (1) vme_msg_5.5<1>:UB tmp_reg0.0<0,1,0>:UB {align1};
/* m6 */
mov (8) vme_msg_6<1>:UD 0x0:UD {align1};
mov (16) vme_msg_6.0<1>:UB INEP_COL0.3<32,8,4>:UB {align1};
mov (1) vme_msg_6.16<1>:UD INTRA_PREDICTORE_MODE {align1};
/* the penalty for Intra mode */
mov (1) vme_msg_6.28<1>:UD 0x010101:UD {align1};
mov (1) vme_msg_6.20<1>:UW CHROMA_ROW.6<0,1,0>:UW {align1};
/* m7 */
mov (4) vme_msg_7.16<1>:UD CHROMA_ROW.8<4,4,1>:UD {align1};
mov (8) vme_msg_7.0<1>:UW CHROMA_COL.2<16,8,2>:UW {align1};
/*
* VME message
*/
/* m1 */
mov (1) intra_flag<1>:UW 0x0:UW {align1} ;
and.z.f0.0 (1) null<1>:UW transform_8x8_ub<0,1,0>:UB 1:UW {align1};
(f0.0) mov (1) intra_part_mask_ub<1>:UB LUMA_INTRA_8x8_DISABLE {align1};
/* assign MB intra struct from the thread payload*/
mov (1) mb_intra_struct_ub<1>:UB input_mb_intra_ub<0,1,0>:UB {align1};
/* Disable DC HAAR component when calculating HARR SATD block */
mov (1) tmp_reg0.0<1>:UW DC_HARR_DISABLE:UW {align1};
mov (1) vme_m1.30<1>:UB tmp_reg0.0<0,1,0>:UB {align1};
mov (8) vme_msg_1<1>:UD vme_m1.0<8,8,1>:UD {align1};
/* m0 */
/* 16x16 Source, Intra_harr */
add (1) vme_m0.12<1>:UD vme_m0.12<0,1,0>:ud INTRA_SAD_HAAR:UD {align1};
mov (8) vme_msg_0<1>:UD vme_m0.0<8,8,1>:UD {align1};
/* after verification it will be passed by using payload */
send (8)
vme_msg_ind
vme_wb<1>:UD
null
cre(
BIND_IDX_VME,
VME_SIC_MESSAGE_TYPE
)
mlen sic_vme_msg_length
rlen vme_wb_length
{align1};
/*
* Oword Block Write message
*/
mov (8) msg_reg0.0<1>:UD obw_m0<8,8,1>:UD {align1};
mov (1) msg_reg1.0<1>:UD vme_wb.0<0,1,0>:UD {align1};
mov (1) msg_reg1.4<1>:UD vme_wb.16<0,1,0>:UD {align1};
mov (1) msg_reg1.8<1>:UD vme_wb.20<0,1,0>:UD {align1};
mov (1) msg_reg1.12<1>:UD vme_wb.24<0,1,0>:UD {align1};
/* Distortion, Intra (17-16), */
mov (1) msg_reg1.16<1>:UW vme_wb.12<0,1,0>:UW {align1};
mov (1) msg_reg1.20<1>:UD vme_wb.8<0,1,0>:UD {align1};
/* VME clock counts */
mov (1) msg_reg1.24<1>:UD vme_wb.28<0,1,0>:UD {align1};
mov (1) msg_reg1.28<1>:UD obw_m0.8<0,1,0>:UD {align1};
/* bind index 3, write 2 oword (32bytes), msg type: 8(OWord Block Write) */
send (16)
msg_ind
obw_wb
null
data_port(
OBW_CACHE_TYPE,
OBW_MESSAGE_TYPE,
OBW_CONTROL_2,
OBW_BIND_IDX,
OBW_WRITE_COMMIT_CATEGORY,
OBW_HEADER_PRESENT
)
mlen 2
rlen obw_wb_length
{align1};
__EXIT:
/*
* kill thread
*/
mov (8) ts_msg_reg0<1>:UD r0<8,8,1>:UD {align1};
send (16) ts_msg_ind acc0<1>UW null thread_spawner(0, 0, 1) mlen 1 rlen 0 {align1 EOT};
|
