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
|
/*
* All Video Processing kernels
* 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: PL8x8_Save_PA.asm
//
// Save planar YUV422 to packed YUV422 format data
//
// Note: SRC_* must reference to regions with data type "BYTE"
// in order to save to byte-aligned byte location
#include "PL8x8_Save_PA.inc"
add (4) pCF_Y_OFFSET<1>:uw ubDEST_CF_OFFSET<4;4,1>:ub nDEST_YUV_REG*nGRFWIB:w // Initial Y,U,V offset in YUV422 block
// Pack Y
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (16) r[pCF_Y_OFFSET, %1*nGRFWIB]<2> ubSRC_Y(0,%1*32)
}
// Pack U/V
$for(0; <nUV_NUM_OF_ROWS; 1) {
mov (8) r[pCF_U_OFFSET, %1*nGRFWIB]<4> ubSRC_U(0, %1*16)
mov (8) r[pCF_V_OFFSET, %1*nGRFWIB]<4> ubSRC_V(0, %1*16)
}
shl (1) rMSGSRC.0<1>:d wORIX<0;1,0>:w 1:w { NoDDClr } // H. block origin need to be doubled
mov (1) rMSGSRC.1<1>:d wORIY<0;1,0>:w { NoDDClr, NoDDChk } // Block origin
mov (1) rMSGSRC.2<1>:ud nDPW_BLOCK_SIZE_YUV:ud { NoDDChk } // Block width and height (32x8)
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Use the mask to determine which pixels shouldn't be over-written
and (1) acc0.0<1>:ud udBLOCK_MASK<0;1,0>:ud 0x00FFFFFF:ud
cmp.ge.f0.0(1) dNULLREG acc0.0<0;1,0>:ud 0x00FFFFFF:ud //Check if all pixels in the block need to be modified
(f0.0) jmpi WritePackedToDataPort
//If mask is not all 1's, then load the entire 32x8 block
//so that only those bytes may be modified that need to be (using the mask)
// Load 32x8 packed YUV 422 ----------------------------------------------------
send (8) udSRC_YUV(0)<1> mMSGHDR udDUMMY_NULL nDATAPORT_READ nDPMR_MSGDSC+nDPR_MSG_SIZE_YUV+nBI_DESTINATION_YUV:ud
mov (8) mMSGHDR<1>:ud rMSGSRC<8;8,1>:ud
//Merge the data
mov (1) f0.0:uw ubBLOCK_MASK_V:ub //Load the mask on flag reg
(f0.0) mov (8) rMASK_TEMP<1>:uw uwBLOCK_MASK_H:uw
(-f0.0) mov (8) rMASK_TEMP<1>:uw 0:uw
// Destination is Byte aligned
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (1) f0.1:uw uwMASK_TEMP(0,%1)<0;1,0>
(-f0.1) mov (16) uwDEST_YUV(%1)<1> uwSRC_YUV(%1) //check the UV merge - vK
}
WritePackedToDataPort:
// Packed YUV data are stored in one of the I/O regions before moving to MRF
// Note: This is necessary since indirect addressing is not supported for MRF.
// Packed data block should be saved as 32x8 pixel block
$for(0; <nY_NUM_OF_ROWS; 1) {
mov (8) mudMSGPAYLOAD(%1)<1> udDEST_YUV(%1)REGION(8,1)
}
send (8) dNULLREG mMSGHDR udDUMMY_NULL nDATAPORT_WRITE nDPMW_MSGDSC+nDPW_MSG_SIZE_YUV+nBI_DESTINATION_YUV:ud
// End of PL8x8_Save_PA
|
