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
|
// =============================================================================
// === spqrgpu_buildAssemblyMaps ===============================================
// =============================================================================
// SPQRGPU, Copyright (c) 2008-2022, Timothy A Davis, Sanjay Ranka,
// Sencer Nuri Yeralan, and Wissam Sid-Lakhdar, All Rights Reserved.
// SPDX-License-Identifier: GPL-2.0+
//------------------------------------------------------------------------------
#include "spqr.hpp"
#ifdef SPQR_HAS_CUDA
template <typename Int>
void spqrgpu_buildAssemblyMaps
(
Int numFronts,
Int n,
Int *Fmap,
Int *Post,
Int *Super,
Int *Rp,
Int *Rj,
Int *Sleft,
Int *Sp,
Int *Sj,
double *Sx,
Int *Fm,
Int *Cm,
Int *Childp,
Int *Child,
Int *CompleteStair,
int *CompleteRjmap,
Int *RjmapOffsets,
int *CompleteRimap,
Int *RimapOffsets,
SEntry *cpuS
)
{
PR (("GPU: building assembly maps:\n")) ;
/* Use Fmap and Stair to map a front's local rows to global rows. */
Int sindex = 0;
for(Int pf=0; pf<numFronts; pf++) // iterate in post-order
{
Int f = Post[pf];
/* Build Fmap for front f. */
Int pstart = Rp[f], pend = Rp[f+1];
for (Int p=pstart; p<pend ; p++)
{
Fmap[Rj[p]] = p - pstart;
}
/* Get workspaces for offset front members */
Int *Stair = CompleteStair + Rp[f];
// ---------------------------------------------------------------------
// initialize the staircase for front F
// ---------------------------------------------------------------------
// initialize the staircase with original rows of S
Int col1 = Super[f], col2 = Super[f+1];
Int fp = col2 - col1;
Int fn = Rp[f+1] - Rp[f];
for (Int j = 0 ; j < fp ; j++)
{
// global column j+col1 is the jth pivot column of front F
Int col = j + col1 ;
Stair[j] = Sleft [col+1] - Sleft [col] ;
PR (("GPU init rows, j: %ld count %ld\n", j, Stair[j])) ;
}
// contribution blocks from children will be added here
for (Int j = fp ; j < fn ; j++){ Stair[j] = 0; }
/* Build Rjmap and Stair for each child of the current front. */
for(Int p=Childp[f]; p<Childp[f+1]; p++)
{
Int c = Child[p];
PR (("child %ld\n", c)) ;
ASSERT (c >= 0 && c < f) ;
int *Rjmap = CompleteRjmap + RjmapOffsets[c];
Int pc = Rp [c] ;
Int pcend = Rp [c+1] ;
Int fnc = pcend - pc ; // total # cols in child F
Int fpc = Super [c+1] - Super [c] ; // # of pivot cols in child
Int cm = Cm [c] ; // # of rows in child C
Int cn = fnc - fpc ; // # of cols in child C
// Create the relative column indices of the child's contributions
// to the parent. The global column 'col' in the child is a
// contribution to the jth column of its parent front. j is
// a relative column index.
for (Int pp=0 ; pp<cn ; pp++)
{
Int col = Rj [pc+fpc+pp] ; // global column index
Int j = Fmap [col] ; // relative column index
Rjmap [pp] = j ; // Save this column
}
ASSERT (cm >= 0 && cm <= cn) ;
pc += fpc ; // pointer to column indices in C
ASSERT (pc + cn == Rp [c+1]) ;
PR ((" cm %ld cn %ld\n", cm, cn)) ;
// add the child rows to the staircase
for (Int ci = 0 ; ci < cm ; ci++)
{
Int col = Rj [pc + ci] ; // leftmost col of this row of C
Int j = Fmap [col] ; // global col is jth col of F
PR ((" child col %ld j %ld\n", col, j)) ;
ASSERT (j >= 0 && j < fn) ;
Stair[j]++ ; // add this row to jth staircase
}
}
// ---------------------------------------------------------------------
// replace Stair with cumsum ([0 Stair]), and find # rows of F
// ---------------------------------------------------------------------
Int fm = 0 ;
for (Int j = 0 ; j < fn ; j++)
{
Int t = fm ;
fm += Stair[j] ;
Stair[j] = t ;
}
PR (("fm %ld %ld\n", fm, Stair[fn-1])) ;
// ---------------------------------------------------------------------
// pack all the S values into the cpuS workspace & advance scalar stair
// ---------------------------------------------------------------------
Int Scount = MAX(0, Sp[Sleft[Super[f+1]]] - Sp[Sleft[Super[f]]]);
if(Scount > 0)
{
for(Int k=0 ; k<fp ; k++)
{
/* pack all rows whose leftmost global column index is k+col1 */
Int leftcol = k + col1 ;
/* for each row of S whose leftmost column is leftcol */
for (Int row = Sleft [leftcol]; row < Sleft [leftcol+1]; row++)
{
// get the location of this row in F & advance the staircase
Int i = Stair[k]++;
/* Pack into S */
for (Int p=Sp[row] ; p<Sp[row+1] ; p++)
{
Int j = Sj[p];
cpuS[sindex].findex = i*fn + j;
cpuS[sindex].value = Sx[p];
sindex++;
}
}
}
}
// ---------------------------------------------------------------------
// build Rimap
// ---------------------------------------------------------------------
for (Int p = Childp [f] ; p < Childp [f+1] ; p++)
{
/* Get child details */
Int c = Child [p] ; // get the child c of front F
int *Rimap = CompleteRimap + RimapOffsets[c];
int *Rjmap = CompleteRjmap + RjmapOffsets[c];
Int pc = Rp [c] ;
// Int pcend = Rp [c+1] ;
// Int fnc = pcend - pc ; // total # cols in child F
Int fpc = Super [c+1] - Super [c] ; // # of pivot cols in child
Int cm = Cm [c] ; // # of rows in child C
// Int cn = (fnc - fpc) ; // cn =# of cols in child C
// ASSERT (cm >= 0 && cm <= cn) ;
pc += fpc ; // pointer to column indices in C
// ASSERT (pc + cn == Rp [c+1]) ;
/* -------------------------------------------------------------- */
/* construct the Rimap */
/* -------------------------------------------------------------- */
for (Int ci = 0 ; ci < cm ; ci++)
{
Int j = Rjmap[ci] ; // global col is jth col of F
ASSERT (j >= 0 && j < fn) ;
Int i = Stair[j]++ ; // add row F(i,:) to jth staircase
ASSERT (i >= 0 && i < fm) ;
Rimap[ci] = i ; // keep track of the mapping
}
}
}
}
template void spqrgpu_buildAssemblyMaps
(
int32_t numFronts,
int32_t n,
int32_t *Fmap,
int32_t *Post,
int32_t *Super,
int32_t *Rp,
int32_t *Rj,
int32_t *Sleft,
int32_t *Sp,
int32_t *Sj,
double *Sx,
int32_t *Fm,
int32_t *Cm,
int32_t *Childp,
int32_t *Child,
int32_t *CompleteStair,
int *CompleteRjmap,
int32_t *RjmapOffsets,
int *CompleteRimap,
int32_t *RimapOffsets,
SEntry *cpuS
) ;
template void spqrgpu_buildAssemblyMaps
(
int64_t numFronts,
int64_t n,
int64_t *Fmap,
int64_t *Post,
int64_t *Super,
int64_t *Rp,
int64_t *Rj,
int64_t *Sleft,
int64_t *Sp,
int64_t *Sj,
double *Sx,
int64_t *Fm,
int64_t *Cm,
int64_t *Childp,
int64_t *Child,
int64_t *CompleteStair,
int *CompleteRjmap,
int64_t *RjmapOffsets,
int *CompleteRimap,
int64_t *RimapOffsets,
SEntry *cpuS
) ;
#endif
|
