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
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
|
//------------------------------------------------------------------------------
// GB_hyper_hash_build: construct A->Y for a hypersparse matrix A
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
#define GB_FREE_WORKSPACE \
{ \
GB_FREE (&I_work, I_work_size) ; \
GB_FREE (&J_work, J_work_size) ; \
GB_FREE (&X_work, X_work_size) ; \
}
#define GB_FREE_ALL \
{ \
GB_FREE_WORKSPACE ; \
GB_phybix_free (A) ; \
}
#include "GB_build.h"
#include "GB_hash.h"
GrB_Info GB_hyper_hash_build // construct A->Y if not already constructed
(
GrB_Matrix A,
GB_Context Context
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
if (A == NULL || !GB_NEED_HYPER_HASH (A))
{
// quick return: A is NULL, not hypersparse, or A->Y already computed
return (GrB_SUCCESS) ;
}
GrB_Info info ;
int64_t *restrict I_work = NULL ; size_t I_work_size = 0 ;
int64_t *restrict J_work = NULL ; size_t J_work_size = 0 ;
uint64_t *restrict X_work = NULL ; size_t X_work_size = 0 ;
ASSERT_MATRIX_OK (A, "A for hyper_hash", GB0) ;
GB_BURBLE_MATRIX (A, "(build hyper hash) ") ;
//--------------------------------------------------------------------------
// allocate A->Y
//--------------------------------------------------------------------------
// A->Y is (A->vdim)-by-(hash table size for A->h), with one vector per
// hash bucket.
const int64_t *restrict Ah = A->h ;
int64_t anvec = A->nvec ;
// this ensures a load factor of 0.5 to 1:
int64_t yvdim = ((uint64_t) 1) << (GB_FLOOR_LOG2 (anvec) + 1) ;
// divide by 4 to get a load factor of 2 to 4:
yvdim = yvdim / 4 ;
yvdim = GB_IMAX (yvdim, 4) ;
int64_t yvlen = A->vdim ;
int64_t hash_bits = (yvdim - 1) ; // yvdim is always a power of 2
GB_OK (GB_new (&(A->Y), // new dynamic header, do not allocate any content
GrB_UINT64, yvlen, yvdim, GB_Ap_null, true, GxB_SPARSE,
-1, 0, Context)) ;
GrB_Matrix Y = A->Y ;
//--------------------------------------------------------------------------
// create the tuples for A->Y
//--------------------------------------------------------------------------
I_work = GB_MALLOC (anvec, int64_t, &I_work_size) ;
J_work = GB_MALLOC (anvec, int64_t, &J_work_size) ;
X_work = GB_MALLOC (anvec, uint64_t, &X_work_size) ;
if (I_work == NULL || J_work == NULL || X_work == NULL)
{
// out of memory
GB_FREE_ALL ;
return (GrB_OUT_OF_MEMORY) ;
}
GB_GET_NTHREADS_MAX (nthreads_max, chunk, Context) ;
int nthreads = GB_nthreads (anvec, chunk, nthreads_max) ;
int64_t k ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (k = 0 ; k < anvec ; k++)
{
int64_t j = Ah [k] ;
I_work [k] = j ;
J_work [k] = GB_HASHF2 (j, hash_bits) ; // in range 0 to yvdim-1
X_work [k] = (uint64_t) k ;
}
//--------------------------------------------------------------------------
// build A->Y, initially hypersparse
//--------------------------------------------------------------------------
GB_OK (GB_builder (
Y, // create Y using a dynamic header
GrB_UINT64, // Y->type
yvlen, // Y->vlen
yvdim, // Y->vdim
true, // Y->is_csc
(int64_t **) &I_work, // row indices
&I_work_size,
(int64_t **) &J_work, // column indices
&J_work_size,
(GB_void **) &X_work, // values
&X_work_size,
false, // tuples need to be sorted
true, // no duplicates
anvec, // size of I_work and J_work in # of tuples
true, // is_matrix: unused
NULL, NULL, // original I,J tuples
NULL, // no scalar iso value
false, // Y is never iso
anvec, // # of tuples
NULL, // no duplicates, so dup is NUL
GrB_UINT64, // the type of X_work
false, // no burble (already burbled above)
Context
)) ;
Y->hyper_switch = -1 ; // never make Y hypersparse
Y->sparsity_control = GxB_SPARSE ; // Y is always sparse CSC
ASSERT (GB_IS_HYPERSPARSE (Y)) ; // Y is currently hypersparse
// workspace has been freed by GB_builder, or transplanted in to Y
ASSERT (I_work == NULL) ;
ASSERT (J_work == NULL) ;
ASSERT (X_work == NULL) ;
//--------------------------------------------------------------------------
// convert A->Y to sparse
//--------------------------------------------------------------------------
// GB_builder always constructs its matrix as hypersparse. Y is now
// conformed to its required sparsity format: always sparse. No burble;
// (already burbled above).
GB_OK (GB_convert_hyper_to_sparse (Y, false, Context)) ;
ASSERT (anvec == GB_nnz (Y)) ;
ASSERT (GB_IS_SPARSE (Y)) ; // Y is now sparse and will remain so
//--------------------------------------------------------------------------
// return result
//--------------------------------------------------------------------------
ASSERT_MATRIX_OK (A, "A from hyper_hash", GB0) ;
ASSERT (!GB_ZOMBIES (Y)) ;
ASSERT (!GB_JUMBLED (Y)) ;
ASSERT (!GB_PENDING (Y)) ;
return (GrB_SUCCESS) ;
}
|
