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
|
// =================================================================================================
// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
// width of 100 characters per line.
//
// Author(s):
// Cedric Nugteren <www.cedricnugteren.nl>
//
// This file contains the Xcopy kernel. It contains one fast vectorized version in case of unit
// strides (incx=incy=1) and no offsets (offx=offy=0). Another version is more general, but doesn't
// support vector data-types.
//
// This kernel uses the level-1 BLAS common tuning parameters.
//
// =================================================================================================
// Enables loading of this file using the C++ pre-processor's #include (C++11 standard raw string
// literal). Comment-out this line for syntax-highlighting when developing.
R"(
// =================================================================================================
// Full version of the kernel with offsets and strided accesses
__kernel __attribute__((reqd_work_group_size(WGS, 1, 1)))
void Xcopy(const int n,
const __global real* restrict xgm, const int x_offset, const int x_inc,
__global real* ygm, const int y_offset, const int y_inc) {
// Loops over the work that needs to be done (allows for an arbitrary number of threads)
for (int id = get_global_id(0); id<n; id += get_global_size(0)) {
ygm[id*y_inc + y_offset] = xgm[id*x_inc + x_offset];
}
}
// =================================================================================================
// Faster version of the kernel without offsets and strided accesses. Also assumes that 'n' is
// dividable by 'VW', 'WGS' and 'WPT'.
__kernel __attribute__((reqd_work_group_size(WGS, 1, 1)))
void XcopyFast(const int n,
const __global realV* restrict xgm,
__global realV* ygm) {
#pragma unroll
for (int _w = 0; _w < WPT; _w += 1) {
const int id = _w*get_global_size(0) + get_global_id(0);
ygm[id] = xgm[id];
}
}
// =================================================================================================
// End of the C++11 raw string literal
)"
// =================================================================================================
|
