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
|
#include "GB_cuda_apply.hpp"
#undef GB_FREE_WORKSPACE
#define GB_FREE_WORKSPACE \
{ \
GB_FREE_MEMORY (&ythunk_cuda, ythunk_cuda_size) ; \
}
#undef GB_FREE_ALL
#define GB_FREE_ALL \
{ \
GB_FREE_WORKSPACE \
GB_cuda_release_stream (&stream) ; \
}
#define BLOCK_SIZE 512
#define LOG2_BLOCK_SIZE 9
GrB_Info GB_cuda_apply_unop
(
GB_void *Cx,
const GrB_Type ctype,
const GB_Operator op,
const bool flipij,
const GrB_Matrix A,
const GB_void *ythunk
)
{
GrB_Info info ;
GB_void *ythunk_cuda = NULL ;
size_t ythunk_cuda_size = 0 ;
cudaStream_t stream = nullptr ;
GrB_Index anz = GB_nnz_held (A) ;
if (anz == 0) return (GrB_SUCCESS) ;
// get a stream on the current device
GB_OK (GB_cuda_acquire_stream (&stream)) ;
// FIXME: make this a CUDA helper function
if (ythunk != NULL && op != NULL && op->ytype != NULL)
{
// make a copy of ythunk, since ythunk might be allocated on
// the CPU stack and thus not accessible to the CUDA kernel.
ythunk_cuda = (GB_void *) GB_MALLOC_MEMORY (1, op->ytype->size,
&ythunk_cuda_size) ;
if (ythunk_cuda == NULL)
{
GB_FREE_ALL ;
return (GrB_OUT_OF_MEMORY) ;
}
memcpy (ythunk_cuda, ythunk, op->ytype->size) ;
}
int32_t number_of_sms = GB_Global_gpu_sm_get (0) ;
int64_t raw_gridsz = GB_ICEIL (anz, BLOCK_SIZE) ;
// cap #of blocks to 256 * #of sms
int32_t gridsz = std::min (raw_gridsz, (int64_t) (number_of_sms * 256)) ;
GB_OK (GB_cuda_apply_unop_jit (Cx, ctype, op, flipij, A,
ythunk_cuda, stream, gridsz, BLOCK_SIZE)) ;
GB_FREE_WORKSPACE ;
GB_OK (GB_cuda_release_stream (&stream)) ;
return GrB_SUCCESS ;
}
|
