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
|
//------------------------------------------------------------------------------
// GB_full_add_template: phase2 for C=A+B, C<M>=A+B, C<!M>=A+B, C is full
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// C is full. The mask M is not present (otherwise, C would be sparse,
// hypersparse, or bitmap). All of these methods are asymptotically optimal.
// ------------------------------------------
// C = A + B
// ------------------------------------------
// full . sparse full
// full . bitmap full
// full . full sparse
// full . full bitmap
// full . full full
// If C is iso and full, this phase has nothing to do.
#ifndef GB_ISO_ADD
{
int64_t p ;
ASSERT (M == NULL) ;
ASSERT (A_is_full || B_is_full) ;
ASSERT (C_sparsity == GxB_FULL) ;
if (A_is_full && B_is_full)
{
//----------------------------------------------------------------------
// Method30: C, A, B are all full
//----------------------------------------------------------------------
#pragma omp parallel for num_threads(C_nthreads) schedule(static)
for (p = 0 ; p < cnz ; p++)
{
// C (i,j) = A (i,j) + B (i,j)
GB_LOAD_A (aij, Ax, p, A_iso) ;
GB_LOAD_B (bij, Bx, p, B_iso) ;
GB_BINOP (GB_CX (p), aij, bij, p % vlen, p / vlen) ;
}
}
else if (A_is_full)
{
//----------------------------------------------------------------------
// C and A are full; B is hypersparse, sparse, or bitmap
//----------------------------------------------------------------------
if (B_is_bitmap)
{
//------------------------------------------------------------------
// Method31: C and A are full; B is bitmap
//------------------------------------------------------------------
#pragma omp parallel for num_threads(C_nthreads) schedule(static)
for (p = 0 ; p < cnz ; p++)
{
if (Bb [p])
{
// C (i,j) = A (i,j) + B (i,j)
GB_LOAD_A (aij, Ax, p, A_iso) ;
GB_LOAD_B (bij, Bx, p, B_iso) ;
GB_BINOP (GB_CX (p), aij, bij, p % vlen, p / vlen) ;
}
else
{
#ifdef GB_EWISEUNION
{
// C (i,j) = A(i,j) + beta
GB_LOAD_A (aij, Ax, p, A_iso) ;
GB_BINOP (GB_CX (p), aij, beta_scalar,
p % vlen, p / vlen);
}
#else
{
// C (i,j) = A (i,j)
GB_COPY_A_TO_C (GB_CX (p), Ax, p, A_iso) ;
}
#endif
}
}
}
else
{
//------------------------------------------------------------------
// Method32: C and A are full; B is sparse or hypersparse
//------------------------------------------------------------------
#pragma omp parallel for num_threads(C_nthreads) schedule(static)
for (p = 0 ; p < cnz ; p++)
{
#ifdef GB_EWISEUNION
{
// C (i,j) = A(i,j) + beta
GB_LOAD_A (aij, Ax, p, A_iso) ;
GB_BINOP (GB_CX (p), aij, beta_scalar,
p % vlen, p / vlen) ;
}
#else
{
// C (i,j) = A (i,j)
GB_COPY_A_TO_C (GB_CX (p), Ax, p, A_iso) ;
}
#endif
}
GB_SLICE_MATRIX (B, 8, chunk) ;
#pragma omp parallel for num_threads(B_nthreads) schedule(dynamic,1)
for (taskid = 0 ; taskid < B_ntasks ; taskid++)
{
int64_t kfirst = kfirst_Bslice [taskid] ;
int64_t klast = klast_Bslice [taskid] ;
for (int64_t k = kfirst ; k <= klast ; k++)
{
// find the part of B(:,k) for this task
int64_t j = GBH (Bh, k) ;
int64_t pB_start, pB_end ;
GB_get_pA (&pB_start, &pB_end, taskid, k, kfirst,
klast, pstart_Bslice, Bp, vlen) ;
int64_t pC_start = j * vlen ;
// traverse over B(:,j), the kth vector of B
for (int64_t pB = pB_start ; pB < pB_end ; pB++)
{
// C (i,j) = A (i,j) + B (i,j)
int64_t i = Bi [pB] ;
int64_t p = pC_start + i ;
GB_LOAD_A (aij, Ax, p , A_iso) ;
GB_LOAD_B (bij, Bx, pB, B_iso) ;
GB_BINOP (GB_CX (p), aij, bij, i, j) ;
}
}
}
}
}
else
{
//----------------------------------------------------------------------
// C and B are full; A is hypersparse, sparse, or bitmap
//----------------------------------------------------------------------
if (A_is_bitmap)
{
//------------------------------------------------------------------
// Method33: C and B are full; A is bitmap
//------------------------------------------------------------------
#pragma omp parallel for num_threads(C_nthreads) schedule(static)
for (p = 0 ; p < cnz ; p++)
{
if (Ab [p])
{
// C (i,j) = A (i,j) + B (i,j)
GB_LOAD_A (aij, Ax, p, A_iso) ;
GB_LOAD_B (bij, Bx, p, B_iso) ;
GB_BINOP (GB_CX (p), aij, bij, p % vlen, p / vlen) ;
}
else
{
#ifdef GB_EWISEUNION
{
// C (i,j) = alpha + B(i,j)
GB_LOAD_B (bij, Bx, p, B_iso) ;
GB_BINOP (GB_CX (p), alpha_scalar,
bij, p % vlen, p / vlen);
}
#else
{
// C (i,j) = B (i,j)
GB_COPY_B_TO_C (GB_CX (p), Bx, p, B_iso) ;
}
#endif
}
}
}
else
{
//------------------------------------------------------------------
// Method34: C and B are full; A is hypersparse or sparse
//------------------------------------------------------------------
#pragma omp parallel for num_threads(C_nthreads) schedule(static)
for (p = 0 ; p < cnz ; p++)
{
#ifdef GB_EWISEUNION
{
// C (i,j) = alpha + B(i,j)
GB_LOAD_B (bij, Bx, p, B_iso) ;
GB_BINOP (GB_CX (p), alpha_scalar, bij,
p % vlen, p / vlen) ;
}
#else
{
// C (i,j) = B (i,j)
GB_COPY_B_TO_C (GB_CX (p), Bx, p, B_iso) ;
}
#endif
}
GB_SLICE_MATRIX (A, 8, chunk) ;
#pragma omp parallel for num_threads(A_nthreads) schedule(dynamic,1)
for (taskid = 0 ; taskid < A_ntasks ; taskid++)
{
int64_t kfirst = kfirst_Aslice [taskid] ;
int64_t klast = klast_Aslice [taskid] ;
for (int64_t k = kfirst ; k <= klast ; k++)
{
// find the part of A(:,k) for this task
int64_t j = GBH (Ah, k) ;
int64_t pA_start, pA_end ;
GB_get_pA (&pA_start, &pA_end, taskid, k, kfirst,
klast, pstart_Aslice, Ap, vlen) ;
int64_t pC_start = j * vlen ;
// traverse over A(:,j), the kth vector of A
for (int64_t pA = pA_start ; pA < pA_end ; pA++)
{
// C (i,j) = A (i,j) + B (i,j)
int64_t i = Ai [pA] ;
int64_t p = pC_start + i ;
GB_LOAD_A (aij, Ax, pA, A_iso) ;
GB_LOAD_B (bij, Bx, p , B_iso) ;
GB_BINOP (GB_CX (p), aij, bij, i, j) ;
}
}
}
}
}
}
#endif
|
