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
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
|
//------------------------------------------------------------------------------
// GB_AxB_saxpy3_template.h: C=A*B, C<M>=A*B, or C<!M>=A*B via saxpy3 method
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Definitions for GB_AxB_saxpy3_template.c. These do not depend on the
// sparsity of A and B.
#ifndef GB_AXB_SAXPY3_TEMPLATE_H
#define GB_AXB_SAXPY3_TEMPLATE_H
//------------------------------------------------------------------------------
// GB_GET_M_j: prepare to iterate over M(:,j)
//------------------------------------------------------------------------------
// prepare to iterate over the vector M(:,j), for the (kk)th vector of B
#define GB_GET_M_j \
int64_t pM_start, pM_end ; \
if (M_is_hyper) \
{ \
/* M is hypersparse: find M(:,j) in the M->Y hyper_hash */ \
GB_hyper_hash_lookup (Mp, M_Yp, M_Yi, M_Yx, M_hash_bits, \
GBH (Bh, kk), &pM_start, &pM_end) ; \
} \
else \
{ \
/* A is sparse, bitmap, or full */ \
int64_t j = GBH (Bh, kk) ; \
pM_start = GBP (Mp, j , mvlen) ; \
pM_end = GBP (Mp, j+1, mvlen) ; \
} \
const int64_t mjnz = pM_end - pM_start
//------------------------------------------------------------------------------
// GB_GET_M_j_RANGE
//------------------------------------------------------------------------------
#define GB_GET_M_j_RANGE(gamma) \
const int64_t mjnz_much = mjnz * gamma
//------------------------------------------------------------------------------
// GB_SCATTER_Mj_t: scatter M(:,j) of the given type into Gus. workspace
//------------------------------------------------------------------------------
#define GB_SCATTER_Mj_t(mask_t,pMstart,pMend,mark) \
{ \
const mask_t *restrict Mxx = (mask_t *) Mx ; \
if (M_is_bitmap) \
{ \
/* M is bitmap */ \
for (int64_t pM = pMstart ; pM < pMend ; pM++) \
{ \
/* if (M (i,j) == 1) mark Hf [i] */ \
if (Mb [pM] && Mxx [pM]) Hf [GBI (Mi, pM, mvlen)] = mark ; \
} \
} \
else \
{ \
/* M is hyper, sparse, or full */ \
for (int64_t pM = pMstart ; pM < pMend ; pM++) \
{ \
/* if (M (i,j) == 1) mark Hf [i] */ \
if (Mxx [pM]) Hf [GBI (Mi, pM, mvlen)] = mark ; \
} \
} \
} \
break ;
//------------------------------------------------------------------------------
// GB_SCATTER_M_j: scatter M(:,j) into the Gustavson workpace
//------------------------------------------------------------------------------
#define GB_SCATTER_M_j(pMstart,pMend,mark) \
if (Mx == NULL) \
{ \
/* M is structural, not valued */ \
if (M_is_bitmap) \
{ \
/* M is bitmap */ \
for (int64_t pM = pMstart ; pM < pMend ; pM++) \
{ \
/* if (M (i,j) is present) mark Hf [i] */ \
if (Mb [pM]) Hf [GBI (Mi, pM, mvlen)] = mark ; \
} \
} \
else \
{ \
/* M is hyper, sparse, or full */ \
for (int64_t pM = pMstart ; pM < pMend ; pM++) \
{ \
/* mark Hf [i] */ \
Hf [GBI (Mi, pM, mvlen)] = mark ; \
} \
} \
} \
else \
{ \
/* mask is valued, not structural */ \
switch (msize) \
{ \
default: \
case GB_1BYTE: GB_SCATTER_Mj_t (uint8_t , pMstart, pMend, mark) ; \
case GB_2BYTE: GB_SCATTER_Mj_t (uint16_t, pMstart, pMend, mark) ; \
case GB_4BYTE: GB_SCATTER_Mj_t (uint32_t, pMstart, pMend, mark) ; \
case GB_8BYTE: GB_SCATTER_Mj_t (uint64_t, pMstart, pMend, mark) ; \
case GB_16BYTE: \
{ \
const uint64_t *restrict Mxx = (uint64_t *) Mx ; \
for (int64_t pM = pMstart ; pM < pMend ; pM++) \
{ \
/* if (M (i,j) == 1) mark Hf [i] */ \
if (!GBB (Mb, pM)) continue ; \
if (Mxx [2*pM] || Mxx [2*pM+1]) \
{ \
/* Hf [i] = M(i,j) */ \
Hf [GBI (Mi, pM, mvlen)] = mark ; \
} \
} \
} \
} \
}
//------------------------------------------------------------------------------
// GB_HASH_M_j: scatter M(:,j) for a coarse hash task
//------------------------------------------------------------------------------
// hash M(:,j) into Hf and Hi for coarse hash task, C<M>=A*B or C<!M>=A*B
#define GB_HASH_M_j \
for (int64_t pM = pM_start ; pM < pM_end ; pM++) \
{ \
GB_GET_M_ij (pM) ; /* get M(i,j) */ \
if (!mij) continue ; /* skip if M(i,j)=0 */ \
const int64_t i = GBI (Mi, pM, mvlen) ; \
for (GB_HASH (i)) /* find i in hash */ \
{ \
if (Hf [hash] < mark) \
{ \
Hf [hash] = mark ; /* insert M(i,j)=1 */ \
Hi [hash] = i ; \
break ; \
} \
} \
}
//------------------------------------------------------------------------------
// GB_GET_T_FOR_SECONDJ: define t for SECONDJ and SECONDJ1 semirings
//------------------------------------------------------------------------------
#if GB_IS_SECONDJ_MULTIPLIER
#define GB_GET_T_FOR_SECONDJ \
GB_CIJ_DECLARE (t) ; \
GB_MULT (t, ignore, ignore, ignore, ignore, j) ;
#else
#define GB_GET_T_FOR_SECONDJ
#endif
//------------------------------------------------------------------------------
// GB_GET_B_j_FOR_ALL_FORMATS: prepare to iterate over B(:,j)
//------------------------------------------------------------------------------
// prepare to iterate over the vector B(:,j), the (kk)th vector in B, where
// j == GBH (Bh, kk). This macro works regardless of the sparsity of A and B.
#define GB_GET_B_j_FOR_ALL_FORMATS(A_is_hyper,B_is_sparse,B_is_hyper) \
const int64_t j = (B_is_hyper) ? Bh [kk] : kk ; \
GB_GET_T_FOR_SECONDJ ; /* t = j for SECONDJ, or j+1 for SECONDJ1 */ \
int64_t pB = (B_is_sparse || B_is_hyper) ? Bp [kk] : (kk * bvlen) ; \
const int64_t pB_end = \
(B_is_sparse || B_is_hyper) ? Bp [kk+1] : (pB+bvlen) ; \
const int64_t bjnz = pB_end - pB ; /* nnz (B (:,j) */
//------------------------------------------------------------------------------
// GB_GET_B_kj: get the numeric value of B(k,j)
//------------------------------------------------------------------------------
#if GB_IS_FIRSTJ_MULTIPLIER
// FIRSTJ or FIRSTJ1 multiplier
// t = aik * bkj = k or k+1
#define GB_GET_B_kj \
GB_CIJ_DECLARE (t) ; \
GB_MULT (t, ignore, ignore, ignore, k, ignore)
#else
#define GB_GET_B_kj \
GB_GETB (bkj, Bx, pB, B_iso) /* bkj = Bx [pB] */
#endif
//------------------------------------------------------------------------------
// GB_GET_A_k_FOR_ALL_FORMATS: prepare to iterate over the vector A(:,k)
//------------------------------------------------------------------------------
#define GB_GET_A_k_FOR_ALL_FORMATS(A_is_hyper) \
int64_t pA_start, pA_end ; \
if (A_is_hyper) \
{ \
/* A is hypersparse: find A(:,k) in the A->Y hyper_hash */ \
GB_hyper_hash_lookup (Ap, A_Yp, A_Yi, A_Yx, A_hash_bits, \
k, &pA_start, &pA_end) ; \
} \
else \
{ \
/* A is sparse, bitmap, or full */ \
pA_start = GBP (Ap, k , avlen) ; \
pA_end = GBP (Ap, k+1, avlen) ; \
} \
const int64_t aknz = pA_end - pA_start
//------------------------------------------------------------------------------
// GB_GET_M_ij: get the numeric value of M(i,j)
//------------------------------------------------------------------------------
#define GB_GET_M_ij(pM) \
/* get M(i,j), at Mi [pM] and Mx [pM] */ \
bool mij = GBB (Mb, pM) && GB_mcast (Mx, pM, msize)
//------------------------------------------------------------------------------
// GB_MULT_A_ik_B_kj: declare t and compute t = A(i,k) * B(k,j)
//------------------------------------------------------------------------------
#if GB_IS_PAIR_MULTIPLIER
// PAIR multiplier: t is always 1; no numeric work to do to compute t.
// The LXOR_PAIR and PLUS_PAIR semirings need the value t = 1 to use in
// their monoid operator, however.
#define t (GB_CTYPE_CAST (1, 0))
#define GB_MULT_A_ik_B_kj
#elif ( GB_IS_FIRSTJ_MULTIPLIER || GB_IS_SECONDJ_MULTIPLIER )
// nothing to do; t = aik*bkj already defined in an outer loop
#define GB_MULT_A_ik_B_kj
#else
// typical semiring
#define GB_MULT_A_ik_B_kj \
GB_GETA (aik, Ax, pA, A_iso) ; /* aik = Ax [pA] ; */ \
GB_CIJ_DECLARE (t) ; /* ctype t ; */ \
GB_MULT (t, aik, bkj, i, k, j) /* t = aik * bkj ; */
#endif
//------------------------------------------------------------------------------
// GB_GATHER_ALL_C_j: gather the values and pattern of C(:,j)
//------------------------------------------------------------------------------
// gather the pattern and values of C(:,j) for a coarse Gustavson task;
// the pattern is not flagged as jumbled.
#if GB_IS_ANY_PAIR_SEMIRING
// ANY_PAIR: result is purely symbolic; no numeric work to do
#define GB_GATHER_ALL_C_j(mark) \
for (int64_t i = 0 ; i < cvlen ; i++) \
{ \
if (Hf [i] == mark) \
{ \
Ci [pC++] = i ; \
} \
}
#else
// typical semiring
#define GB_GATHER_ALL_C_j(mark) \
for (int64_t i = 0 ; i < cvlen ; i++) \
{ \
if (Hf [i] == mark) \
{ \
GB_CIJ_GATHER (pC, i) ; /* Cx [pC] = Hx [i] */ \
Ci [pC++] = i ; \
} \
}
#endif
//------------------------------------------------------------------------------
// GB_SORT_C_j_PATTERN: sort C(:,j) for a coarse task, or flag as jumbled
//------------------------------------------------------------------------------
// Only coarse tasks do the optional sort. Fine hash tasks always leave C
// jumbled.
#define GB_SORT_C_j_PATTERN \
if (do_sort) \
{ \
/* sort the pattern of C(:,j) (non-default) */ \
GB_qsort_1 (Ci + Cp [kk], cjnz) ; \
} \
else \
{ \
/* lazy sort: C(:,j) is now jumbled (default) */ \
task_C_jumbled = true ; \
}
//------------------------------------------------------------------------------
// GB_SORT_AND_GATHER_C_j: sort and gather C(:,j) for a coarse Gustavson task
//------------------------------------------------------------------------------
// gather the values of C(:,j) for a coarse Gustavson task
#if GB_IS_ANY_PAIR_SEMIRING
// ANY_PAIR: result is purely symbolic
#define GB_SORT_AND_GATHER_C_j \
GB_SORT_C_j_PATTERN ;
#else
// typical semiring
#define GB_SORT_AND_GATHER_C_j \
GB_SORT_C_j_PATTERN ; \
/* gather the values into C(:,j) */ \
for (int64_t pC = Cp [kk] ; pC < Cp [kk+1] ; pC++) \
{ \
const int64_t i = Ci [pC] ; \
GB_CIJ_GATHER (pC, i) ; /* Cx [pC] = Hx [i] */ \
}
#endif
//------------------------------------------------------------------------------
// GB_SORT_AND_GATHER_HASHED_C_j: sort and gather C(:,j) for a coarse hash task
//------------------------------------------------------------------------------
#if GB_IS_ANY_PAIR_SEMIRING
// ANY_PAIR: result is purely symbolic
#define GB_SORT_AND_GATHER_HASHED_C_j(hash_mark) \
GB_SORT_C_j_PATTERN ;
#else
// gather the values of C(:,j) for a coarse hash task
#define GB_SORT_AND_GATHER_HASHED_C_j(hash_mark) \
GB_SORT_C_j_PATTERN ; \
for (int64_t pC = Cp [kk] ; pC < Cp [kk+1] ; pC++) \
{ \
const int64_t i = Ci [pC] ; \
for (GB_HASH (i)) /* find i in hash table */ \
{ \
if (Hf [hash] == (hash_mark) && (Hi [hash] == i)) \
{ \
/* i found in the hash table */ \
/* Cx [pC] = Hx [hash] ; */ \
GB_CIJ_GATHER (pC, hash) ; \
break ; \
} \
} \
}
#endif
//------------------------------------------------------------------------------
// GB_ATOMIC_UPDATE_HX: Hx [i] += t
//------------------------------------------------------------------------------
#if GB_IS_ANY_MONOID
//--------------------------------------------------------------------------
// The update Hx [i] += t can be skipped entirely, for the ANY monoid.
//--------------------------------------------------------------------------
#define GB_ATOMIC_UPDATE_HX(i,t)
#elif GB_HAS_ATOMIC
//--------------------------------------------------------------------------
// Hx [i] += t via atomic update
//--------------------------------------------------------------------------
// for built-in MIN/MAX monoids only, on built-in types
#define GB_MINMAX(i,t,done) \
{ \
GB_CTYPE xold, xnew, *px = Hx + (i) ; \
do \
{ \
/* xold = Hx [i] via atomic read */ \
GB_ATOMIC_READ \
xold = (*px) ; \
/* done if xold <= t for MIN, or xold >= t for MAX, */ \
/* but not done if xold is NaN */ \
if (done) break ; \
xnew = t ; /* t should be assigned; it is not NaN */ \
} \
while (!GB_ATOMIC_COMPARE_EXCHANGE (px, xold, xnew)) ; \
}
#if GB_IS_IMIN_MONOID
// built-in MIN monoids for signed and unsigned integers
#define GB_ATOMIC_UPDATE_HX(i,t) \
GB_MINMAX (i, t, xold <= t)
#elif GB_IS_IMAX_MONOID
// built-in MAX monoids for signed and unsigned integers
#define GB_ATOMIC_UPDATE_HX(i,t) \
GB_MINMAX (i, t, xold >= t)
#elif GB_IS_FMIN_MONOID
// built-in MIN monoids for float and double, with omitnan behavior.
// The update is skipped entirely if t is NaN. Otherwise, if t is not
// NaN, xold is checked. If xold is NaN, islessequal (xold, t) is
// always false, so the non-NaN t must be always be assigned to Hx [i].
// If both terms are not NaN, then islessequal (xold,t) is just
// xold <= t. If that is true, there is no work to do and
// the loop breaks. Otherwise, t is smaller than xold and so it must
// be assigned to Hx [i].
#define GB_ATOMIC_UPDATE_HX(i,t) \
{ \
if (!isnan (t)) \
{ \
GB_MINMAX (i, t, islessequal (xold, t)) ; \
} \
}
#elif GB_IS_FMAX_MONOID
// built-in MAX monoids for float and double, with omitnan behavior.
#define GB_ATOMIC_UPDATE_HX(i,t) \
{ \
if (!isnan (t)) \
{ \
GB_MINMAX (i, t, isgreaterequal (xold, t)) ; \
} \
}
#elif GB_IS_PLUS_FC32_MONOID
// built-in PLUS_FC32 monoid can be done as two independent atomics
#define GB_ATOMIC_UPDATE_HX(i,t) \
GB_ATOMIC_UPDATE \
Hx_real [2*(i)] += crealf (t) ; \
GB_ATOMIC_UPDATE \
Hx_imag [2*(i)] += cimagf (t) ;
#elif GB_IS_PLUS_FC64_MONOID
// built-in PLUS_FC64 monoid can be done as two independent atomics
#define GB_ATOMIC_UPDATE_HX(i,t) \
GB_ATOMIC_UPDATE \
Hx_real [2*(i)] += creal (t) ; \
GB_ATOMIC_UPDATE \
Hx_imag [2*(i)] += cimag (t) ;
#elif GB_HAS_OMP_ATOMIC
// built-in PLUS and TIMES for integers and real, and boolean LOR,
// LAND, LXOR monoids can be implemented with an OpenMP pragma.
#define GB_ATOMIC_UPDATE_HX(i,t) \
GB_ATOMIC_UPDATE \
GB_HX_UPDATE (i, t)
#else
// all other atomic monoids (EQ, XNOR) on boolean, signed and unsigned
// integers, float, and double (not used for single and double
// complex).
#define GB_ATOMIC_UPDATE_HX(i,t) \
{ \
GB_CTYPE xold, xnew, *px = Hx + (i) ; \
do \
{ \
/* xold = Hx [i] via atomic read */ \
GB_ATOMIC_READ \
xold = (*px) ; \
/* xnew = xold + t */ \
xnew = GB_ADD_FUNCTION (xold, t) ; \
} \
while (!GB_ATOMIC_COMPARE_EXCHANGE (px, xold, xnew)) ; \
}
#endif
#else
//--------------------------------------------------------------------------
// Hx [i] += t can only be done inside the critical section
//--------------------------------------------------------------------------
// all user-defined monoids go here, and all complex monoids (except PLUS)
#define GB_ATOMIC_UPDATE_HX(i,t) \
GB_OMP_FLUSH \
GB_HX_UPDATE (i, t) ; \
GB_OMP_FLUSH
#endif
#define GB_IS_MINMAX_MONOID \
(GB_IS_IMIN_MONOID || GB_IS_IMAX_MONOID || \
GB_IS_FMIN_MONOID || GB_IS_FMAX_MONOID)
//------------------------------------------------------------------------------
// GB_ATOMIC_WRITE_HX: Hx [i] = t
//------------------------------------------------------------------------------
#if GB_IS_ANY_PAIR_SEMIRING
//--------------------------------------------------------------------------
// ANY_PAIR: result is purely symbolic; no numeric work to do
//--------------------------------------------------------------------------
#define GB_ATOMIC_WRITE_HX(i,t)
#elif GB_HAS_ATOMIC
//--------------------------------------------------------------------------
// Hx [i] = t via atomic write
//--------------------------------------------------------------------------
#if GB_IS_PLUS_FC32_MONOID
// built-in PLUS_FC32 monoid
#define GB_ATOMIC_WRITE_HX(i,t) \
GB_ATOMIC_WRITE \
Hx_real [2*(i)] = crealf (t) ; \
GB_ATOMIC_WRITE \
Hx_imag [2*(i)] = cimagf (t) ;
#elif GB_IS_PLUS_FC64_MONOID
// built-in PLUS_FC64 monoid
#define GB_ATOMIC_WRITE_HX(i,t) \
GB_ATOMIC_WRITE \
Hx_real [2*(i)] = creal (t) ; \
GB_ATOMIC_WRITE \
Hx_imag [2*(i)] = cimag (t) ;
#else
// all other atomic monoids
#define GB_ATOMIC_WRITE_HX(i,t) \
GB_ATOMIC_WRITE \
GB_HX_WRITE (i, t)
#endif
#else
//--------------------------------------------------------------------------
// Hx [i] = t via critical section
//--------------------------------------------------------------------------
#define GB_ATOMIC_WRITE_HX(i,t) \
GB_OMP_FLUSH \
GB_HX_WRITE (i, t) ; \
GB_OMP_FLUSH
#endif
//------------------------------------------------------------------------------
// hash iteration
//------------------------------------------------------------------------------
// to iterate over the hash table, looking for index i:
//
// for (GB_HASH (i))
// {
// ...
// }
//
// which expands into the following, where f(i) is the GB_HASHF(i) hash
// function:
//
// for (int64_t hash = f(i) ; ; hash = (hash+1)&(hash_size-1))
// {
// ...
// }
#define GB_HASH(i) \
int64_t hash = GB_HASHF (i,hash_bits) ; ; GB_REHASH (hash,i,hash_bits)
//------------------------------------------------------------------------------
// define macros for any sparsity of A and B
//------------------------------------------------------------------------------
#undef GB_META16
#include "GB_meta16_definitions.h"
#endif
|
