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
|
/*
* Intra predict 16x16 luma block
* 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.
*
*/
// Module name: intra_Pred_16x16_Y.asm
//
// Intra predict 16x16 luma block
//
and (1) PINTRAPRED_Y<1>:w INTRA_PRED_MODE(0)REGION(1,0) 0x0F:w
// WA for "jmpi" restriction
mov (1) REG_INTRA_TEMP_1<1>:ud r[PINTRAPRED_Y, INTRA_16X16_OFFSET]:ub
jmpi (1) REG_INTRA_TEMP_1<0;1,0>:d
// Mode 0
INTRA_16x16_VERTICAL:
$for(0; <16; 2) {
mov (32) PRED_YW(%1)<1> INTRA_REF_TOP(0) {Compr}
}
jmpi (1) End_intra_Pred_16x16_Y
// Mode 1
INTRA_16x16_HORIZONTAL:
mov (1) PREF_LEFT_UD<1>:ud INTRA_REF_LEFT_ID*GRFWIB*0x00010001+0x00040000:ud // Set address registers for instruction compression
$for(0,0; <16; 2,8) {
mov (32) PRED_YW(%1)<1> r[PREF_LEFT,%2+3]<0;1,0>:ub {Compr} // Actual left column reference data start at offset 3
}
jmpi (1) End_intra_Pred_16x16_Y
// Mode 2
INTRA_16x16_DC:
and.nz.f0.0 (8) NULLREG REG_INTRA_PRED_AVAIL_FLAG INTRA_PRED_UP_AVAIL_FLAG:ud // Top macroblock available for intra prediction?
and (8) acc0<1>:ud REG_INTRA_PRED_AVAIL_FLAG INTRA_PRED_LEFT_TH_AVAIL_FLAG+INTRA_PRED_LEFT_BH_AVAIL_FLAG:ud // Left macroblock available for intra prediction?
xor.z.f0.1 (8) NULLREG acc0:ud INTRA_PRED_LEFT_TH_AVAIL_FLAG+INTRA_PRED_LEFT_BH_AVAIL_FLAG:ud // Left macroblock available for intra prediction?
// Rearrange reference samples for unified DC prediction code
//
(-f0.0.any8h) mov (8) INTRA_REF_TOP_W(0)<1> 0x8080:uw
(-f0.1.any8h) mov (8) INTRA_REF_LEFT(0)<4> INTRA_REF_TOP(0)REGION(8,1)
(-f0.1.any8h) mov (8) INTRA_REF_LEFT(1)<4> INTRA_REF_TOP(0,8)REGION(8,1)
(-f0.0.any8h) mov (8) INTRA_REF_TOP(0)<1> INTRA_REF_LEFT(0)REGION(8,4)
(-f0.0.any8h) mov (8) INTRA_REF_TOP(0,8)<1> INTRA_REF_LEFT(1)REGION(8,4) // Split due to HW limitation
// Perform DC prediction
//
add (16) PRED_YW(15)<1> INTRA_REF_LEFT(0)REGION(8,4) INTRA_REF_TOP(0)REGION(16,1)
add (8) PRED_YW(15)<1> PRED_YW(15)REGION(8,1) PRED_YW(15,8)REGION(8,1)
add (4) PRED_YW(15)<1> PRED_YW(15)REGION(4,1) PRED_YW(15,4)REGION(4,1)
add (2) PRED_YW(15)<1> PRED_YW(15)REGION(2,1) PRED_YW(15,2)REGION(2,1)
add (32) acc0<1>:w PRED_YW(15)REGION(1,0) PRED_YW(15,1)REGION(1,0) {Compr} // Set up both acc0 and acc1
add (32) acc0<1>:w acc0:w 16:w {Compr}
$for(0; <16; 2) {
shr (32) PRED_YW(%1)<1> acc0:w 5:w {Compr}
}
jmpi (1) End_intra_Pred_16x16_Y
// Mode 3
INTRA_16x16_PLANE:
// Refer to H.264/AVC spec Section 8.3.3.4
#define A REG_INTRA_TEMP_2.0 // All are WORD type
#define B REG_INTRA_TEMP_3.0
#define C REG_INTRA_TEMP_3.1
#define YP REG_INTRA_TEMP_0 // Store intermediate results of c*(y-7). Make sure it's an even GRF
#define YP1 REG_INTRA_TEMP_1 // Store intermediate results of c*(y-7). Make sure it's an odd GRF, used in {Comp}
#define XP REG_INTRA_TEMP_5 // Store intermediate results of a+b*(x-7)+16. Make sure it's an odd GRF
// First Calculate constants H and V
// H1 = sum((-x'-1)*p[8+x',-1]), x'=0,1,...7
// H2 = sum((-x'-1)*p[6-x',-1]), x'=7,6,...0
// H = -H1 + H2
// The same calculation holds for V
//
mul (8) H1(0)<1> INTRA_REF_TOP(0,8)REGION(8,1) 0x89ABCDEF:v
mul (8) H2(0)<1> INTRA_REF_TOP(0,-1)REGION(8,1) 0xFEDCBA98:v
mul (8) V1(0)<1> INTRA_REF_LEFT(0,8*4)REGION(8,4) 0x89ABCDEF:v
mul (8) V2(0)<1> INTRA_REF_LEFT(0)REGION(8,4) 0x0FEDCBA9:v
mul (1) V2(0,7)<1> INTRA_REF_TOP(0,-1)<0;1,0> -8:w // Replace 0*p[-1,7] with -8*p[-1,-1]
// Now, REG_INTRA_TEMP_0 holds [H2, -H1] and REG_INTRA_TEMP_1 holds [V2, -V1]
// Sum up [H2, -H1] and [V2, -V1] using instruction compression
// ExecSize = 16 is restricted by B-spec for instruction compression
// Actual intermediate results are in lower sub-registers after each summing step
add (16) H1(0)<1> -H1(0) H2(0) {Compr} // Results in lower 8 WORDs
add (16) H1(0)<1> H1(0) H1(0,4) {Compr} // Results in lower 4 WORDs
add (16) H1(0)<1> H1(0) H1(0,2) {Compr} // Results in lower 2 WORDs
add (16) H1(0)<1> H1(0) H1(0,1) {Compr} // Results in lower 1 WORD
// Calculate a, b, c and further derivations
mov (16) acc0<1>:w 32:w
mac (2) acc0<1>:w H1(0)<16;1,0> 5:w
shr (2) B<1>:w acc0:w 6:w // Done b,c
mov (16) acc0<1>:w 16:w
mac (16) acc0<1>:w INTRA_REF_TOP(0,15)<0;1,0> 16:w
mac (16) A<1>:w INTRA_REF_LEFT(0,15*4)<0;1,0> 16:w // A = a+16
mac (16) XP<1>:w B<0;1,0>:w XY_7<16;16,1>:b // XP = A+b*(x-7)
mul (8) YP<1>:w C<0;1,0>:w XY_7<16;8,2>:b // YP = c*(y-7), even portion
mul (8) YP1<1>:w C<0;1,0>:w XY_7_1<16;8,2>:b // YP = c*(y-7), odd portion
// Finally the intra_16x16 plane prediction
$for(0,0; <16; 2,1) {
add (32) acc0<1>:w XP<16;16,1>:w YP.%2<16;16,0>:w {Compr} // Set Width!= 1 to trick EU to use YP_1.%2 for 2nd instruction
shr.sat (32) PRED_Y(%1)<2> acc0<16;16,1>:w 5:w {Compr}
}
End_intra_Pred_16x16_Y:
// End of intra_Pred_16x16_Y
|
