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
|
#include "GB_cuda_select.hpp"
#undef GB_FREE_WORKSPACE
#define GB_FREE_WORKSPACE \
{ \
if (stream != nullptr) \
{ \
cudaStreamSynchronize (stream) ; \
cudaStreamDestroy (stream) ; \
} \
stream = nullptr ; \
}
#undef GB_FREE_ALL
#define GB_FREE_ALL \
{ \
GB_phybix_free (C) ; \
GB_FREE_WORKSPACE ; \
}
#define BLOCK_SIZE 512
#define LOG2_BLOCK_SIZE 9
GrB_Info GB_cuda_select_sparse
(
GrB_Matrix C,
const bool C_iso,
const GrB_IndexUnaryOp op,
const bool flipij,
const GrB_Matrix A,
const GB_void *athunk,
const GB_void *ythunk,
GB_Werk Werk
)
{
// check inputs
GrB_Info info = GrB_NO_VALUE ;
ASSERT (C != NULL && !(C->header_size == 0)) ;
ASSERT (A != NULL && !(A->header_size == 0)) ;
// FIXME: use the stream pool
cudaStream_t stream = nullptr ;
CUDA_OK (cudaStreamCreate (&stream)) ;
GrB_Index anz = GB_nnz_held (A) ;
int32_t number_of_sms = GB_Global_gpu_sm_get (0) ;
int64_t raw_gridsz = GB_ICEIL (anz, BLOCK_SIZE) ;
int32_t gridsz = std::min (raw_gridsz, (int64_t) (number_of_sms * 256)) ;
gridsz = std::max (gridsz, 1) ;
// determine the p_is_32, j_is_32, and i_is_32 settings for the new matrix
int csparsity = GxB_HYPERSPARSE ;
bool Cp_is_32, Cj_is_32, Ci_is_32 ;
GB_determine_pji_is_32 (&Cp_is_32, &Cj_is_32, &Ci_is_32,
csparsity, anz, A->vlen, A->vdim, Werk) ;
// Initialize C to be a user-returnable hypersparse empty matrix.
// If needed, we handle the hyper->sparse conversion below.
GB_OK (GB_new (&C, // sparse or hyper (from A), existing header
A->type, A->vlen, A->vdim, GB_ph_calloc, A->is_csc,
csparsity, A->hyper_switch, /* C->plen: revised later: */ 1,
Cp_is_32, Cj_is_32, Ci_is_32)) ;
C->jumbled = A->jumbled ;
C->iso = C_iso ;
GB_OK (GB_cuda_select_sparse_jit (C, A,
flipij, ythunk, op, stream, gridsz, BLOCK_SIZE)) ;
GB_FREE_WORKSPACE ;
ASSERT (C->x != NULL) ;
if (C_iso)
{
// If C is iso, initialize the iso entry
GB_select_iso ((GB_void *) C->x, op->opcode, athunk,
(GB_void *) A->x, A->type->size) ;
}
if (C->nvec == C->vdim)
{
// C hypersparse with all vectors present; quick convert to sparse
GB_FREE_MEMORY (&(C->h), C->h_size) ;
}
ASSERT_MATRIX_OK (C, "C output of cuda_select_sparse", GB0) ;
return GrB_SUCCESS ;
}
|
