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
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
|
//------------------------------------------------------------------------------
// GB_subassign_07: C(I,J)<M> += scalar ; no S
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Method 07: C(I,J)<M> += scalar ; no S
// M: present
// Mask_comp: false
// C_replace: false
// accum: present
// A: scalar
// S: none
// C: not bitmap
// M: any sparsity
#include "GB_subassign_methods.h"
GrB_Info GB_subassign_07
(
GrB_Matrix C,
// input:
const GrB_Index *I,
const int64_t nI,
const int Ikind,
const int64_t Icolon [3],
const GrB_Index *J,
const int64_t nJ,
const int Jkind,
const int64_t Jcolon [3],
const GrB_Matrix M,
const bool Mask_struct,
const GrB_BinaryOp accum,
const void *scalar,
const GrB_Type atype,
GB_Context Context
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
ASSERT (!GB_IS_BITMAP (C)) ;
ASSERT (!GB_aliased (C, M)) ; // NO ALIAS of C==M
//--------------------------------------------------------------------------
// get inputs
//--------------------------------------------------------------------------
GB_EMPTY_TASKLIST ;
GB_MATRIX_WAIT_IF_JUMBLED (C) ;
GB_MATRIX_WAIT_IF_JUMBLED (M) ;
GB_GET_C ; // C must not be bitmap
int64_t zorig = C->nzombies ;
const int64_t *restrict Ch = C->h ;
const int64_t *restrict Cp = C->p ;
const bool C_is_hyper = (Ch != NULL) ;
const int64_t Cnvec = C->nvec ;
GB_GET_C_HYPER_HASH ;
GB_GET_MASK ;
GB_GET_ACCUM_SCALAR ;
//--------------------------------------------------------------------------
// Method 07: C(I,J)<M> += scalar ; no S
//--------------------------------------------------------------------------
// Time: Close to Optimal: same as Method 05.
// Method 05 and Method 07 are very similar. Also compare with Method 06n.
//--------------------------------------------------------------------------
// Parallel: slice M into coarse/fine tasks (Method 05, 06n, 07)
//--------------------------------------------------------------------------
GB_SUBASSIGN_ONE_SLICE (M) ; // M cannot be jumbled
//--------------------------------------------------------------------------
// phase 1: undelete zombies, update entries, and count pending tuples
//--------------------------------------------------------------------------
#pragma omp parallel for num_threads(nthreads) schedule(dynamic,1) \
reduction(+:nzombies)
for (taskid = 0 ; taskid < ntasks ; taskid++)
{
//----------------------------------------------------------------------
// get the task descriptor
//----------------------------------------------------------------------
GB_GET_TASK_DESCRIPTOR_PHASE1 ;
//----------------------------------------------------------------------
// compute all vectors in this task
//----------------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//------------------------------------------------------------------
// get j, the kth vector of M
//------------------------------------------------------------------
int64_t j = GBH (Mh, k) ;
GB_GET_VECTOR (pM, pM_end, pA, pA_end, Mp, k, Mvlen) ;
int64_t mjnz = pM_end - pM ;
if (mjnz == 0) continue ;
//------------------------------------------------------------------
// get jC, the corresponding vector of C
//------------------------------------------------------------------
GB_LOOKUP_VECTOR_jC (fine_task, taskid) ;
int64_t cjnz = pC_end - pC_start ;
bool cjdense = (cjnz == Cvlen) ;
//------------------------------------------------------------------
// C(I,jC)<M(:,j)> += scalar ; no S
//------------------------------------------------------------------
if (cjdense)
{
//--------------------------------------------------------------
// C(:,jC) is dense so the binary search of C is not needed
//--------------------------------------------------------------
for ( ; pM < pM_end ; pM++)
{
//----------------------------------------------------------
// update C(iC,jC), but only if M(iA,j) allows it
//----------------------------------------------------------
bool mij = GBB (Mb, pM) && GB_mcast (Mx, pM, msize) ;
if (mij)
{
int64_t iA = GBI (Mi, pM, Mvlen) ;
GB_iC_DENSE_LOOKUP ;
// ----[C A 1] or [X A 1]-------------------------------
// [C A 1]: action: ( =C+A ): apply accum
// [X A 1]: action: ( undelete ): zombie lives
GB_withaccum_C_A_1_scalar ;
}
}
}
else
{
//--------------------------------------------------------------
// C(:,jC) is sparse; use binary search for C
//--------------------------------------------------------------
for ( ; pM < pM_end ; pM++)
{
//----------------------------------------------------------
// update C(iC,jC), but only if M(iA,j) allows it
//----------------------------------------------------------
bool mij = GBB (Mb, pM) && GB_mcast (Mx, pM, msize) ;
if (mij)
{
int64_t iA = GBI (Mi, pM, Mvlen) ;
// find C(iC,jC) in C(:,jC)
GB_iC_BINARY_SEARCH ;
if (cij_found)
{
// ----[C A 1] or [X A 1]---------------------------
// [C A 1]: action: ( =C+A ): apply accum
// [X A 1]: action: ( undelete ): zombie lives
GB_withaccum_C_A_1_scalar ;
}
else
{
// ----[. A 1]--------------------------------------
// [. A 1]: action: ( insert )
task_pending++ ;
}
}
}
}
}
GB_PHASE1_TASK_WRAPUP ;
}
//--------------------------------------------------------------------------
// phase 2: insert pending tuples
//--------------------------------------------------------------------------
GB_PENDING_CUMSUM ;
zorig = C->nzombies ;
#pragma omp parallel for num_threads(nthreads) schedule(dynamic,1) \
reduction(&&:pending_sorted)
for (taskid = 0 ; taskid < ntasks ; taskid++)
{
//----------------------------------------------------------------------
// get the task descriptor
//----------------------------------------------------------------------
GB_GET_TASK_DESCRIPTOR_PHASE2 ;
//----------------------------------------------------------------------
// compute all vectors in this task
//----------------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//------------------------------------------------------------------
// get j, the kth vector of M
//------------------------------------------------------------------
int64_t j = GBH (Mh, k) ;
GB_GET_VECTOR (pM, pM_end, pA, pA_end, Mp, k, Mvlen) ;
int64_t mjnz = pM_end - pM ;
if (mjnz == 0) continue ;
//------------------------------------------------------------------
// get jC, the corresponding vector of C
//------------------------------------------------------------------
GB_LOOKUP_VECTOR_jC (fine_task, taskid) ;
bool cjdense = ((pC_end - pC_start) == Cvlen) ;
//------------------------------------------------------------------
// C(I,jC)<M(:,j)> += scalar ; no S
//------------------------------------------------------------------
if (!cjdense)
{
//--------------------------------------------------------------
// C(:,jC) is sparse; use binary search for C
//--------------------------------------------------------------
for ( ; pM < pM_end ; pM++)
{
//----------------------------------------------------------
// update C(iC,jC), but only if M(iA,j) allows it
//----------------------------------------------------------
bool mij = GBB (Mb, pM) && GB_mcast (Mx, pM, msize) ;
if (mij)
{
int64_t iA = GBI (Mi, pM, Mvlen) ;
// find C(iC,jC) in C(:,jC)
GB_iC_BINARY_SEARCH ;
if (!cij_found)
{
// ----[. A 1]--------------------------------------
// [. A 1]: action: ( insert )
GB_PENDING_INSERT (scalar) ;
}
}
}
}
}
GB_PHASE2_TASK_WRAPUP ;
}
//--------------------------------------------------------------------------
// finalize the matrix and return result
//--------------------------------------------------------------------------
GB_SUBASSIGN_WRAPUP ;
}
|
