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
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
|
//------------------------------------------------------------------------------
// GB_subassign_08n: C(I,J)<M> += A ; no S
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Method 08n: C(I,J)<M> += A ; no S
// M: present
// Mask_comp: false
// C_replace: false
// accum: present
// A: matrix
// S: none
// C not bitmap; C can be full since no zombies are inserted in that case.
// If C is bitmap, then GB_bitmap_assign_M_accum is used instead.
// M, A: not bitmap; Method 08s is used instead if M or A are bitmap.
#include "GB_subassign_methods.h"
//------------------------------------------------------------------------------
// GB_PHASE1_ACTION
//------------------------------------------------------------------------------
// action to take for phase 1 when A(i,j) exists and M(i,j)=1
#define GB_PHASE1_ACTION \
{ \
if (cjdense) \
{ \
/* direct lookup of C(iC,jC) */ \
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_matrix ; \
} \
else \
{ \
/* binary search for 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_matrix ; \
} \
else \
{ \
/* ----[. A 1]-------------------------------------- */ \
/* [. A 1]: action: ( insert ) */ \
task_pending++ ; \
} \
} \
}
//------------------------------------------------------------------------------
// GB_PHASE2_ACTION
//------------------------------------------------------------------------------
// action to take for phase 2 when A(i,j) exists and M(i,j)=1
#define GB_PHASE2_ACTION \
{ \
ASSERT (!cjdense) ; \
{ \
/* binary search for C(iC,jC) in C(:,jC) */ \
GB_iC_BINARY_SEARCH ; \
if (!cij_found) \
{ \
/* ----[. A 1]-------------------------------------- */ \
/* [. A 1]: action: ( insert ) */ \
GB_PENDING_INSERT_aij ; \
} \
} \
}
//------------------------------------------------------------------------------
// GB_subassign_08n: C(I,J)<M> += A ; no S
//------------------------------------------------------------------------------
GrB_Info GB_subassign_08n
(
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 GrB_Matrix A,
GB_Context Context
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
ASSERT (!GB_IS_BITMAP (C)) ;
ASSERT (!GB_IS_BITMAP (M)) ; // Method 08s is used if M is bitmap
ASSERT (!GB_IS_BITMAP (A)) ; // Method 08s is used if A is bitmap
ASSERT (!GB_aliased (C, M)) ; // NO ALIAS of C==M
ASSERT (!GB_aliased (C, A)) ; // NO ALIAS of C==A
//--------------------------------------------------------------------------
// get inputs
//--------------------------------------------------------------------------
GB_EMPTY_TASKLIST ;
GB_MATRIX_WAIT_IF_JUMBLED (C) ;
GB_MATRIX_WAIT_IF_JUMBLED (M) ;
GB_MATRIX_WAIT_IF_JUMBLED (A) ;
GB_GET_C ; // C must not be bitmap
int64_t zorig = C->nzombies ;
const int64_t Cnvec = C->nvec ;
const int64_t *restrict Ch = C->h ;
const int64_t *restrict Cp = C->p ;
const bool C_is_hyper = (Ch != NULL) ;
GB_GET_C_HYPER_HASH ;
GB_GET_MASK ;
GB_GET_A ;
const int64_t *restrict Ah = A->h ;
GB_GET_ACCUM ;
//--------------------------------------------------------------------------
// Method 08n: C(I,J)<M> += A ; no S
//--------------------------------------------------------------------------
// Time: Close to optimal. Omega (sum_j (min (nnz (A(:,j)), nnz (M(:,j)))),
// since only the intersection of A.*M needs to be considered. If either
// M(:,j) or A(:,j) are very sparse compared to the other, then the shorter
// is traversed with a linear-time scan and a binary search is used for the
// other. If the number of nonzeros is comparable, a linear-time scan is
// used for both. Once two entries M(i,j)=1 and A(i,j) are found with the
// same index i, the entry A(i,j) is accumulated or inserted into C.
// The algorithm is very much like the eWise multiplication of A.*M, so the
// parallel scheduling relies on GB_emult_phase0 and GB_ewise_slice.
//--------------------------------------------------------------------------
// Parallel: slice the eWiseMult of Z=A.*M (Method 08n only)
//--------------------------------------------------------------------------
// Method 08n only. If C is sparse, it is sliced for a fine task, so that
// it can do a binary search via GB_iC_BINARY_SEARCH. But if C(:,jC) is
// dense, C(:,jC) is not sliced, so the fine task must do a direct lookup
// via GB_iC_DENSE_LOOKUP. Otherwise a race condition will occur.
// The Z matrix is not constructed, except for its hyperlist (Zh_shallow)
// and mapping to A and M.
// No matrix (C, M, or A) can be bitmap. C, M, A can be sparse/hyper/full,
// in any combination.
int64_t Znvec ;
const int64_t *restrict Zh_shallow = NULL ;
GB_OK (GB_subassign_08n_slice (
&TaskList, &TaskList_size, &ntasks, &nthreads,
&Znvec, &Zh_shallow, &Z_to_A, &Z_to_A_size, &Z_to_M, &Z_to_M_size,
C, I, nI, Ikind, Icolon, J, nJ, Jkind, Jcolon,
A, M, Context)) ;
GB_ALLOCATE_NPENDING_WERK ;
//--------------------------------------------------------------------------
// 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 A(:,j) and M(:,j)
//------------------------------------------------------------------
int64_t j = GBH (Zh_shallow, k) ;
GB_GET_EVEC (pA, pA_end, pA, pA_end, Ap, Ah, j, k, Z_to_A, Avlen) ;
GB_GET_EVEC (pM, pM_end, pB, pB_end, Mp, Mh, j, k, Z_to_M, Mvlen) ;
//------------------------------------------------------------------
// quick checks for empty intersection of A(:,j) and M(:,j)
//------------------------------------------------------------------
int64_t ajnz = pA_end - pA ;
int64_t mjnz = pM_end - pM ;
if (ajnz == 0 || mjnz == 0) continue ;
int64_t iA_first = GBI (Ai, pA, Avlen) ;
int64_t iA_last = GBI (Ai, pA_end-1, Avlen) ;
int64_t iM_first = GBI (Mi, pM, Mvlen) ;
int64_t iM_last = GBI (Mi, pM_end-1, Mvlen) ;
if (iA_last < iM_first || iM_last < iA_first) continue ;
int64_t pM_start = pM ;
//------------------------------------------------------------------
// 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)> += A(:,j) ; no S
//------------------------------------------------------------------
if (ajnz > 32 * mjnz)
{
//--------------------------------------------------------------
// A(:,j) is much denser than M(:,j)
//--------------------------------------------------------------
for ( ; pM < pM_end ; pM++)
{
if (GB_mcast (Mx, pM, msize))
{
int64_t iA = GBI (Mi, pM, Mvlen) ;
// find iA in A(:,j)
int64_t pright = pA_end - 1 ;
bool found ;
// FUTURE::: exploit dense A(:,j)
GB_BINARY_SEARCH (iA, Ai, pA, pright, found) ;
if (found) GB_PHASE1_ACTION ;
}
}
}
else if (mjnz > 32 * ajnz)
{
//--------------------------------------------------------------
// M(:,j) is much denser than A(:,j)
//--------------------------------------------------------------
// FUTURE::: exploit dense mask
bool mjdense = false ;
for ( ; pA < pA_end ; pA++)
{
int64_t iA = GBI (Ai, pA, Avlen) ;
GB_MIJ_BINARY_SEARCH_OR_DENSE_LOOKUP (iA) ;
if (mij) GB_PHASE1_ACTION ;
}
}
else
{
//----------------------------------------------------------
// A(:,j) and M(:,j) have about the same # of entries
//----------------------------------------------------------
// linear-time scan of A(:,j) and M(:,j)
while (pA < pA_end && pM < pM_end)
{
int64_t iA = GBI (Ai, pA, Avlen) ;
int64_t iM = GBI (Mi, pM, Mvlen) ;
if (iA < iM)
{
// A(i,j) exists but not M(i,j)
GB_NEXT (A) ;
}
else if (iM < iA)
{
// M(i,j) exists but not A(i,j)
GB_NEXT (M) ;
}
else
{
// both A(i,j) and M(i,j) exist
if (GB_mcast (Mx, pM, msize)) GB_PHASE1_ACTION ;
GB_NEXT (A) ;
GB_NEXT (M) ;
}
}
}
}
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 A(:,j) and M(:,j)
//------------------------------------------------------------------
int64_t j = GBH (Zh_shallow, k) ;
GB_GET_EVEC (pA, pA_end, pA, pA_end, Ap, Ah, j, k, Z_to_A, Avlen) ;
GB_GET_EVEC (pM, pM_end, pB, pB_end, Mp, Mh, j, k, Z_to_M, Mvlen) ;
//------------------------------------------------------------------
// quick checks for empty intersection of A(:,j) and M(:,j)
//------------------------------------------------------------------
int64_t ajnz = pA_end - pA ;
int64_t mjnz = pM_end - pM ;
if (ajnz == 0 || mjnz == 0) continue ;
int64_t iA_first = GBI (Ai, pA, Avlen) ;
int64_t iA_last = GBI (Ai, pA_end-1, Avlen) ;
int64_t iM_first = GBI (Mi, pM, Mvlen) ;
int64_t iM_last = GBI (Mi, pM_end-1, Mvlen) ;
if (iA_last < iM_first || iM_last < iA_first) continue ;
int64_t pM_start = pM ;
//------------------------------------------------------------------
// get jC, the corresponding vector of C
//------------------------------------------------------------------
GB_LOOKUP_VECTOR_jC (fine_task, taskid) ;
bool cjdense = (pC_end - pC_start == Cvlen) ;
if (cjdense) continue ;
//------------------------------------------------------------------
// C(I,jC)<M(:,j)> += A(:,j) ; no S
//------------------------------------------------------------------
if (ajnz > 32 * mjnz)
{
//--------------------------------------------------------------
// A(:,j) is much denser than M(:,j)
//--------------------------------------------------------------
for ( ; pM < pM_end ; pM++)
{
if (GB_mcast (Mx, pM, msize))
{
int64_t iA = GBI (Mi, pM, Mvlen) ;
// find iA in A(:,j)
int64_t pright = pA_end - 1 ;
bool found ;
// FUTURE::: exploit dense A(:,j)
GB_BINARY_SEARCH (iA, Ai, pA, pright, found) ;
if (found) GB_PHASE2_ACTION ;
}
}
}
else if (mjnz > 32 * ajnz)
{
//--------------------------------------------------------------
// M(:,j) is much denser than A(:,j)
//--------------------------------------------------------------
// FUTURE::: exploit dense mask
bool mjdense = false ;
for ( ; pA < pA_end ; pA++)
{
int64_t iA = GBI (Ai, pA, Avlen) ;
GB_MIJ_BINARY_SEARCH_OR_DENSE_LOOKUP (iA) ;
if (mij) GB_PHASE2_ACTION ;
}
}
else
{
//----------------------------------------------------------
// A(:,j) and M(:,j) have about the same # of entries
//----------------------------------------------------------
// linear-time scan of A(:,j) and M(:,j)
while (pA < pA_end && pM < pM_end)
{
int64_t iA = GBI (Ai, pA, Avlen) ;
int64_t iM = GBI (Mi, pM, Mvlen) ;
if (iA < iM)
{
// A(i,j) exists but not M(i,j)
GB_NEXT (A) ;
}
else if (iM < iA)
{
// M(i,j) exists but not A(i,j)
GB_NEXT (M) ;
}
else
{
// both A(i,j) and M(i,j) exist
if (GB_mcast (Mx, pM, msize)) GB_PHASE2_ACTION ;
GB_NEXT (A) ;
GB_NEXT (M) ;
}
}
}
}
GB_PHASE2_TASK_WRAPUP ;
}
//--------------------------------------------------------------------------
// finalize the matrix and return result
//--------------------------------------------------------------------------
GB_SUBASSIGN_WRAPUP ;
}
|
