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
|
//------------------------------------------------------------------------------
// GB_reduce_panel: s=reduce(A), reduce a matrix to a scalar
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Reduce a matrix to a scalar using a panel-based method for built-in
// operators. No typecasting is performed. A must be sparse, hypersparse,
// or full (it cannot be bitmap). A cannot have any zombies. If A has zombies
// or is bitmap, GB_reduce_to_scalar_template is used instead.
{
//--------------------------------------------------------------------------
// get A
//--------------------------------------------------------------------------
const GB_ATYPE *restrict Ax = (GB_ATYPE *) A->x ;
ASSERT (!A->iso) ;
int64_t anz = GB_nnz (A) ;
ASSERT (anz > 0) ;
ASSERT (!GB_IS_BITMAP (A)) ;
ASSERT (A->nzombies == 0) ;
#if GB_IS_ANY_MONOID
// the ANY monoid can take any entry, and terminate immediately
s = Ax [anz-1] ;
#else
//--------------------------------------------------------------------------
// reduce A to a scalar
//--------------------------------------------------------------------------
if (nthreads == 1)
{
//----------------------------------------------------------------------
// load the Panel with the first entries
//----------------------------------------------------------------------
GB_ATYPE Panel [GB_PANEL] ;
int64_t first_panel_size = GB_IMIN (GB_PANEL, anz) ;
for (int64_t k = 0 ; k < first_panel_size ; k++)
{
Panel [k] = Ax [k] ;
}
#if GB_HAS_TERMINAL
int panel_count = 0 ;
#endif
//----------------------------------------------------------------------
// reduce all entries to the Panel
//----------------------------------------------------------------------
for (int64_t p = GB_PANEL ; p < anz ; p += GB_PANEL)
{
if (p + GB_PANEL > anz)
{
// last partial panel
for (int64_t k = 0 ; k < anz-p ; k++)
{
// Panel [k] = op (Panel [k], Ax [p+k]) ;
GB_ADD_ARRAY_TO_ARRAY (Panel, k, Ax, p+k) ;
}
}
else
{
// whole panel
for (int64_t k = 0 ; k < GB_PANEL ; k++)
{
// Panel [k] = op (Panel [k], Ax [p+k]) ;
GB_ADD_ARRAY_TO_ARRAY (Panel, k, Ax, p+k) ;
}
#if GB_HAS_TERMINAL
panel_count-- ;
if (panel_count <= 0)
{
// check for early exit only every 256 panels
panel_count = 256 ;
int count = 0 ;
for (int64_t k = 0 ; k < GB_PANEL ; k++)
{
count += (Panel [k] == GB_TERMINAL_VALUE) ;
}
if (count > 0)
{
break ;
}
}
#endif
}
}
//----------------------------------------------------------------------
// s = reduce (Panel)
//----------------------------------------------------------------------
s = Panel [0] ;
for (int64_t k = 1 ; k < first_panel_size ; k++)
{
// s = op (s, Panel [k]) ;
GB_ADD_ARRAY_TO_SCALAR (s, Panel, k) ;
}
}
else
{
//----------------------------------------------------------------------
// all tasks share a single early_exit flag
//----------------------------------------------------------------------
// If this flag gets set, all tasks can terminate early
#if GB_HAS_TERMINAL
bool early_exit = false ;
#endif
//----------------------------------------------------------------------
// each thread reduces its own slice in parallel
//----------------------------------------------------------------------
int tid ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (tid = 0 ; tid < ntasks ; tid++)
{
//------------------------------------------------------------------
// determine the work for this task
//------------------------------------------------------------------
// Task tid reduces Ax [pstart:pend-1] to the scalar W [tid]
int64_t pstart, pend ;
GB_PARTITION (pstart, pend, anz, tid, ntasks) ;
GB_ATYPE t = Ax [pstart] ;
//------------------------------------------------------------------
// skip this task if the terminal value has already been reached
//------------------------------------------------------------------
#if GB_HAS_TERMINAL
// check if another task has called for an early exit
bool my_exit ;
GB_ATOMIC_READ
my_exit = early_exit ;
if (!my_exit)
#endif
//------------------------------------------------------------------
// do the reductions for this task
//------------------------------------------------------------------
{
//--------------------------------------------------------------
// load the Panel with the first entries
//--------------------------------------------------------------
GB_ATYPE Panel [GB_PANEL] ;
int64_t my_anz = pend - pstart ;
int64_t first_panel_size = GB_IMIN (GB_PANEL, my_anz) ;
for (int64_t k = 0 ; k < first_panel_size ; k++)
{
Panel [k] = Ax [pstart + k] ;
}
#if GB_HAS_TERMINAL
int panel_count = 0 ;
#endif
//--------------------------------------------------------------
// reduce all entries to the Panel
//--------------------------------------------------------------
for (int64_t p = pstart + GB_PANEL ; p < pend ; p += GB_PANEL)
{
if (p + GB_PANEL > pend)
{
// last partial panel
for (int64_t k = 0 ; k < pend-p ; k++)
{
// Panel [k] = op (Panel [k], Ax [p+k]) ;
GB_ADD_ARRAY_TO_ARRAY (Panel, k, Ax, p+k) ;
}
}
else
{
// whole panel
for (int64_t k = 0 ; k < GB_PANEL ; k++)
{
// Panel [k] = op (Panel [k], Ax [p+k]) ;
GB_ADD_ARRAY_TO_ARRAY (Panel, k, Ax, p+k) ;
}
#if GB_HAS_TERMINAL
panel_count-- ;
if (panel_count <= 0)
{
// check for early exit only every 256 panels
panel_count = 256 ;
int count = 0 ;
for (int64_t k = 0 ; k < GB_PANEL ; k++)
{
count += (Panel [k] == GB_TERMINAL_VALUE) ;
}
if (count > 0)
{
break ;
}
}
#endif
}
}
//--------------------------------------------------------------
// t = reduce (Panel)
//--------------------------------------------------------------
t = Panel [0] ;
for (int64_t k = 1 ; k < first_panel_size ; k++)
{
// t = op (t, Panel [k]) ;
GB_ADD_ARRAY_TO_SCALAR (t, Panel, k) ;
}
#if GB_HAS_TERMINAL
if (t == GB_TERMINAL_VALUE)
{
// tell all other tasks to exit early
GB_ATOMIC_WRITE
early_exit = true ;
}
#endif
}
//------------------------------------------------------------------
// save the results of this task
//------------------------------------------------------------------
W [tid] = t ;
}
//----------------------------------------------------------------------
// sum up the results of each slice using a single thread
//----------------------------------------------------------------------
s = W [0] ;
for (int tid = 1 ; tid < ntasks ; tid++)
{
// s = op (s, W [tid]), no typecast
GB_ADD_ARRAY_TO_SCALAR (s, W, tid) ;
}
}
#endif
}
|
