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
|
//------------------------------------------------------------------------------
// GB_subassign_08n_slice: slice the entries and vectors for GB_subassign_08n
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Constructs a set of tasks to compute C for GB_subassign_08n, based on
// slicing two input matrices (A and M). Fine tasks must also find their
// location in their vector C(:,jC).
// This method is used only by GB_subassign_08n. New zombies cannot be
// created, since no entries are deleted. Old zombies can be brought back to
// life, however.
// ===================== ==============
// M cmp rpl acc A S method: action
// ===================== ==============
// M - - + A - 08n: C(I,J)<M> += A, no S
// C, M, A: not bitmap. C can be full.
// If C is bitmap, then GB_bitmap_assign_M_accum is used instead.
// If M or A are bitmap, but C is sparse or hyper, then Method 08s is used
// instead (which handles both M and A as bitmap). As a result, this method
// does not need to consider the bitmap case for C, M, or A.
#include "GB_subassign_methods.h"
#include "GB_emult.h"
// Npending is set to NULL by the GB_EMPTY_TASKLIST macro, but unused here.
#include "GB_unused.h"
GrB_Info GB_subassign_08n_slice
(
// output:
GB_task_struct **p_TaskList, // array of structs, of size max_ntasks
size_t *p_TaskList_size, // size of TaskList
int *p_ntasks, // # of tasks constructed
int *p_nthreads, // # of threads to use
int64_t *p_Znvec, // # of vectors to compute in Z
const int64_t *restrict *Zh_handle, // Zh_shallow is A->h, M->h, or NULL
int64_t *restrict *Z_to_A_handle, // Z_to_A: size Znvec, or NULL
size_t *Z_to_A_size_handle,
int64_t *restrict *Z_to_M_handle, // Z_to_M: size Znvec, or NULL
size_t *Z_to_M_size_handle,
// input:
const GrB_Matrix C, // output matrix C
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 A, // matrix to slice
const GrB_Matrix M, // matrix to slice
GB_Context Context
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
ASSERT (!GB_IS_BITMAP (C)) ;
ASSERT (!GB_IS_BITMAP (M)) ; // Method 08n is not used for M bitmap
ASSERT (!GB_IS_BITMAP (A)) ; // Method 08n is not used for A bitmap
GB_EMPTY_TASKLIST
ASSERT (p_TaskList != NULL) ;
ASSERT (p_ntasks != NULL) ;
ASSERT (p_nthreads != NULL) ;
ASSERT_MATRIX_OK (C, "C for 08n_slice", GB0) ;
ASSERT_MATRIX_OK (M, "M for 08n_slice", GB0) ;
ASSERT_MATRIX_OK (A, "A for 08n_slice", GB0) ;
ASSERT (!GB_JUMBLED (C)) ;
ASSERT (!GB_JUMBLED (M)) ;
ASSERT (!GB_JUMBLED (A)) ;
ASSERT (p_Znvec != NULL) ;
ASSERT (Zh_handle != NULL) ;
ASSERT (Z_to_A_handle != NULL) ;
ASSERT (Z_to_M_handle != NULL) ;
(*p_TaskList ) = NULL ;
(*p_TaskList_size) = 0 ;
(*p_ntasks ) = 0 ;
(*p_nthreads ) = 1 ;
(*p_Znvec ) = 0 ;
(*Zh_handle ) = NULL ;
(*Z_to_A_handle) = NULL ;
(*Z_to_M_handle) = NULL ;
//--------------------------------------------------------------------------
// get inputs
//--------------------------------------------------------------------------
int64_t *restrict Ci = C->i ;
int64_t nzombies = C->nzombies ;
const int64_t Cnvec = C->nvec ;
const int64_t Cvlen = C->vlen ;
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 ;
const int64_t *restrict Mp = M->p ;
const int64_t *restrict Mh = M->h ;
const int64_t *restrict Mi = M->i ;
const int64_t Mvlen = M->vlen ;
const int64_t *restrict Ap = A->p ;
const int64_t *restrict Ah = A->h ;
const int64_t *restrict Ai = A->i ;
const int64_t Avlen = A->vlen ;
//--------------------------------------------------------------------------
// construct fine/coarse tasks for eWise multiply of A.*M
//--------------------------------------------------------------------------
// Compare with the first part of GB_emult for A.*B. Note that M in this
// function takes the place of B in GB_emult.
int64_t Znvec ;
const int64_t *restrict Zh_shallow = NULL ;
int Z_sparsity = GxB_SPARSE ;
GB_OK (GB_emult_phase0 (&Znvec, &Zh_shallow, &Zh_size, NULL, NULL,
&Z_to_A, &Z_to_A_size, &Z_to_M, &Z_to_M_size, &Z_sparsity, NULL, A, M,
Context)) ;
// Z is still sparse or hypersparse, not bitmap or full
ASSERT (Z_sparsity == GxB_SPARSE || Z_sparsity == GxB_HYPERSPARSE) ;
GB_OK (GB_ewise_slice (
&TaskList, &TaskList_size, &ntasks, &nthreads,
Znvec, Zh_shallow, NULL, Z_to_A, Z_to_M, false,
NULL, A, M, Context)) ;
//--------------------------------------------------------------------------
// slice C(:,jC) for each fine task
//--------------------------------------------------------------------------
// Each fine task that operates on C(:,jC) must be limited to just its
// portion of C(:,jC). Otherwise, one task could bring a zombie to life,
// at the same time another is attempting to do a binary search on that
// entry. This is safe as long as a 64-bit integer read/write is always
// atomic, but there is no gaurantee that this is true for all
// architectures. Note that GB_subassign_08n cannot create new zombies.
// This work could be done in parallel, but each task does at most 2 binary
// searches. The total work for all the binary searches will likely be
// small. So do the work with a single thread.
for (taskid = 0 ; taskid < ntasks ; taskid++)
{
//----------------------------------------------------------------------
// get the task descriptor
//----------------------------------------------------------------------
GB_GET_TASK_DESCRIPTOR ;
//----------------------------------------------------------------------
// do the binary search for this fine task
//----------------------------------------------------------------------
if (fine_task)
{
//------------------------------------------------------------------
// get A(:,j) and M(:,j)
//------------------------------------------------------------------
int64_t k = kfirst ;
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 ;
//------------------------------------------------------------------
// get jC, the corresponding vector of C
//------------------------------------------------------------------
GB_LOOKUP_VECTOR_jC (false, 0) ;
bool cjdense = (pC_end - pC_start == Cvlen) ;
//------------------------------------------------------------------
// slice C(:,jC) for this fine task
//------------------------------------------------------------------
if (cjdense)
{
// do not slice C(:,jC) if it is dense
TaskList [taskid].pC = pC_start ;
TaskList [taskid].pC_end = pC_end ;
}
else
{
// find where this task starts and ends in C(:,jC)
int64_t iA_start = GB_IMIN (iA_first, iM_first) ;
int64_t iC1 = GB_ijlist (I, iA_start, Ikind, Icolon) ;
int64_t iA_end = GB_IMAX (iA_last, iM_last) ;
int64_t iC2 = GB_ijlist (I, iA_end, Ikind, Icolon) ;
// If I is an explicit list, it must be already sorted
// in ascending order, and thus iC1 <= iC2. If I is
// GB_ALL or GB_STRIDE with inc >= 0, then iC1 < iC2.
// But if inc < 0, then iC1 > iC2. iC_start and iC_end
// are used for a binary search bracket, so iC_start <=
// iC_end must hold.
int64_t iC_start = GB_IMIN (iC1, iC2) ;
int64_t iC_end = GB_IMAX (iC1, iC2) ;
// this task works on Ci,Cx [pC:pC_end-1]
int64_t pleft = pC_start ;
int64_t pright = pC_end - 1 ;
bool found, is_zombie ;
GB_SPLIT_BINARY_SEARCH_ZOMBIE (iC_start, Ci, pleft, pright,
found, nzombies, is_zombie) ;
TaskList [taskid].pC = pleft ;
pleft = pC_start ;
pright = pC_end - 1 ;
GB_SPLIT_BINARY_SEARCH_ZOMBIE (iC_end, Ci, pleft, pright,
found, nzombies, is_zombie) ;
TaskList [taskid].pC_end = (found) ? (pleft+1) : pleft ;
}
ASSERT (TaskList [taskid].pC <= TaskList [taskid].pC_end) ;
}
}
//--------------------------------------------------------------------------
// return result
//--------------------------------------------------------------------------
(*p_TaskList ) = TaskList ;
(*p_TaskList_size) = TaskList_size ;
(*p_ntasks ) = ntasks ;
(*p_nthreads ) = nthreads ;
(*p_Znvec ) = Znvec ;
(*Zh_handle ) = Zh_shallow ;
(*Z_to_A_handle) = Z_to_A ; (*Z_to_A_size_handle) = Z_to_A_size ;
(*Z_to_M_handle) = Z_to_M ; (*Z_to_M_size_handle) = Z_to_M_size ;
return (GrB_SUCCESS) ;
}
|
