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
|
/* ************************************************************************
* Copyright 2013 Advanced Micro Devices, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
* ************************************************************************/
static const char *GEMM_HELPER = "
void getBlockNumber(uint nBlocks, uint blockID, uint *bidY, uint *bidX, uint flag)
{
#ifndef HERK
{
if(flag) //Column Major ordering for NT kernels
{
*bidY = ( blockID % ( nBlocks));
*bidX = ( blockID / ( nBlocks));
}
else //Row Major ordering for TN kernels
{
*bidX = ( blockID % ( nBlocks));
*bidY = ( blockID / ( nBlocks));
}
}
#else
{
volatile uint _i = 0, _j = 0;
for ( _j = (blockID / nBlocks); _j < nBlocks; _j++)
{
_i = blockID - ((_j*((2* nBlocks) + 1 - _j))/2) + _j;
if ( _i < nBlocks && ( _i >= 0) )
{
break;
}
}
#ifdef HERK_LOWER_TRIANGLE
*bidY = _i;
*bidX = _j;
#else
*bidY = _j;
*bidX = _i;
#endif
}
#endif
}
//
// mapWorkGroupToTileNumber() - Maps a workgroup number to a Tile position in output matrix
// Groups the full tiles together and half-tiles together and maps the workgroup number
// such that full tiles are processed wholly by consecutive workgroups and half-tiles are
// processed by consecutive workgroups
//
// ASSUMPTION:
// Assumes column major numbering of workgroup
//
// Observation:
// This new grouping yielded worse results than normal column-major order.
// Tested with GEMM NN kernel. So, we will not be using this function.
// This is here just for completeness sake
//
void mapWorkGroupToTileNumber(uint M, uint N, uint *bidY, uint *bidX)
{
uint fullTilesOnY, numTilesOnX;
uint relativeGroupId;
numTilesOnX = ((N-1) / ((get_local_size(0) / %WIDTH) * %ITEMX)) + 1;
fullTilesOnY = (M / (%WIDTH * %ITEMY));
if (get_group_id(0) < (numTilesOnX * fullTilesOnY) )
{
*bidY = ( get_group_id(0) % ( fullTilesOnY));
*bidX = ( get_group_id(0) / ( fullTilesOnY));
} else {
relativeGroupId = get_group_id(0) - (numTilesOnX * fullTilesOnY);
*bidY = fullTilesOnY;
*bidX = relativeGroupId;
}
}
";
|
