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
|
/*
-- MAGMA (version 2.9.0) --
Univ. of Tennessee, Knoxville
Univ. of California, Berkeley
Univ. of Colorado, Denver
@date January 2025
@generated from sparse/control/magma_zcsrsplit.cpp, normal z -> c, Wed Jan 22 14:42:26 2025
@author Hartwig Anzt
*/
#include "magmasparse_internal.h"
/**
Purpose
-------
Splits a CSR matrix into two matrices, one containing the diagonal blocks
with the diagonal element stored first, one containing the rest of the
original matrix.
Arguments
---------
@param[in]
offset magma_int_t
size of the first block
@param[in]
bsize magma_int_t
size of the diagonal blocks
@param[in]
A magma_c_matrix
CSR input matrix
@param[out]
D magma_c_matrix*
CSR matrix containing diagonal blocks
@param[out]
R magma_c_matrix*
CSR matrix containing rest
@param[in]
queue magma_queue_t
Queue to execute in.
@ingroup magmasparse_caux
********************************************************************/
extern "C" magma_int_t
magma_ccsrsplit(
magma_int_t offset,
magma_int_t bsize,
magma_c_matrix A,
magma_c_matrix *D,
magma_c_matrix *R,
magma_queue_t queue )
{
magma_int_t info = 0;
magma_int_t i, k, j, nnz_diag, nnz_offd;
// make sure the target structure is empty
magma_cmfree( D, queue );
magma_cmfree( R, queue );
D->ownership = MagmaTrue;
R->ownership = MagmaTrue;
D->val = NULL;
D->col = NULL;
D->row = NULL;
D->rowidx = NULL;
D->blockinfo = NULL;
D->diag = NULL;
D->dval = NULL;
D->dcol = NULL;
D->drow = NULL;
D->drowidx = NULL;
D->ddiag = NULL;
R->val = NULL;
R->col = NULL;
R->row = NULL;
R->rowidx = NULL;
R->blockinfo = NULL;
R->diag = NULL;
R->dval = NULL;
R->dcol = NULL;
R->drow = NULL;
R->drowidx = NULL;
R->ddiag = NULL;
if ( A.memory_location == Magma_CPU &&
( A.storage_type == Magma_CSR ||
A.storage_type == Magma_CSRCOO ) )
{
nnz_diag = nnz_offd = 0;
// Count the new number of nonzeroes in the two matrices
for( i=0; i<offset; i+=offset ){
for( k=i; k<min(A.num_rows,i+offset); k++ ){
int check = 0;
for( j=A.row[k]; j<A.row[k+1]; j++ ){
if ( A.col[j] < i )
nnz_offd++;
else if ( A.col[j] < i+offset ){
if( A.col[j] == k ){
check = 1;
}
nnz_diag++;
}
else
nnz_offd++;
}
if( check == 0 ){
printf("error: matrix contains zero on diagonal at (%d,%d).\n", int(i), int(i));
info = -1;
goto cleanup;
}
}
}
magma_int_t ii = i;
for( i=ii; i<A.num_rows; i+=bsize ){
for( k=i; k<min(A.num_rows,i+bsize); k++ ){
int check = 0;
for( j=A.row[k]; j<A.row[k+1]; j++ ){
if ( A.col[j] < i )
nnz_offd++;
else if ( A.col[j] < i+bsize ){
if( A.col[j] == k ){
check = 1;
}
nnz_diag++;
}
else
nnz_offd++;
}
if( check == 0 ){
printf("error: matrix contains zero on diagonal at (%d,%d).\n", int(i), int(i));
info = -1;
goto cleanup;
}
}
}
// Allocate memory for the new matrices
D->storage_type = Magma_CSRD;
D->memory_location = A.memory_location;
D->num_rows = A.num_rows;
D->num_cols = A.num_cols;
D->nnz = nnz_diag;
R->storage_type = Magma_CSR;
R->memory_location = A.memory_location;
R->num_rows = A.num_rows;
R->num_cols = A.num_cols;
R->nnz = nnz_offd;
CHECK( magma_cmalloc_cpu( &D->val, nnz_diag ));
CHECK( magma_index_malloc_cpu( &D->row, A.num_rows+1 ));
CHECK( magma_index_malloc_cpu( &D->col, nnz_diag ));
CHECK( magma_cmalloc_cpu( &R->val, nnz_offd ));
CHECK( magma_index_malloc_cpu( &R->row, A.num_rows+1 ));
CHECK( magma_index_malloc_cpu( &R->col, nnz_offd ));
// Fill up the new sparse matrices
D->row[0] = 0;
R->row[0] = 0;
nnz_offd = nnz_diag = 0;
for( i=0; i<offset; i+=offset) {
for( k=i; k<min(A.num_rows,i+offset); k++ ) {
D->row[k+1] = D->row[k];
R->row[k+1] = R->row[k];
for( j=A.row[k]; j<A.row[k+1]; j++ ) {
if ( A.col[j] < i ) {
R->val[nnz_offd] = A.val[j];
R->col[nnz_offd] = A.col[j];
R->row[k+1]++;
nnz_offd++;
}
else if ( A.col[j] < i+offset ) {
// larger than diagonal remain as before
if ( A.col[j]>k ) {
D->val[nnz_diag] = A.val[ j ];
D->col[nnz_diag] = A.col[ j ];
D->row[k+1]++;
}
// diagonal is written first
else if ( A.col[j]==k ) {
D->val[D->row[k]] = A.val[ j ];
D->col[D->row[k]] = A.col[ j ];
D->row[k+1]++;
}
// smaller than diagonal are shifted one to the right
// to have room for the diagonal
else {
D->val[nnz_diag+1] = A.val[ j ];
D->col[nnz_diag+1] = A.col[ j ];
D->row[k+1]++;
}
nnz_diag++;
}
else {
R->val[nnz_offd] = A.val[j];
R->col[nnz_offd] = A.col[j];
R->row[k+1]++;
nnz_offd++;
}
}
}
}
ii = i;
for( i=ii; i<A.num_rows; i+=bsize) {
for( k=i; k<min(A.num_rows,i+bsize); k++ ) {
D->row[k+1] = D->row[k];
R->row[k+1] = R->row[k];
for( j=A.row[k]; j<A.row[k+1]; j++ ) {
if ( A.col[j] < i ) {
R->val[nnz_offd] = A.val[j];
R->col[nnz_offd] = A.col[j];
R->row[k+1]++;
nnz_offd++;
}
else if ( A.col[j] < i+bsize ) {
// larger than diagonal remain as before
if ( A.col[j]>k ) {
D->val[nnz_diag] = A.val[ j ];
D->col[nnz_diag] = A.col[ j ];
D->row[k+1]++;
}
// diagonal is written first
else if ( A.col[j]==k ) {
D->val[D->row[k]] = A.val[ j ];
D->col[D->row[k]] = A.col[ j ];
D->row[k+1]++;
}
// smaller than diagonal are shifted one to the right
// to have room for the diagonal
else {
D->val[nnz_diag+1] = A.val[ j ];
D->col[nnz_diag+1] = A.col[ j ];
D->row[k+1]++;
}
nnz_diag++;
}
else {
R->val[nnz_offd] = A.val[j];
R->col[nnz_offd] = A.col[j];
R->row[k+1]++;
nnz_offd++;
}
}
}
}
}
else {
magma_c_matrix Ah={Magma_CSR}, ACSR={Magma_CSR}, DCSR={Magma_CSR}, RCSR={Magma_CSR}, Dh={Magma_CSR}, Rh={Magma_CSR};
CHECK( magma_cmtransfer( A, &Ah, A.memory_location, Magma_CPU, queue ));
CHECK( magma_cmconvert( Ah, &ACSR, A.storage_type, Magma_CSR, queue ));
CHECK( magma_ccsrsplit( offset, bsize, ACSR, &DCSR, &RCSR, queue ));
CHECK( magma_cmconvert( DCSR, &Dh, Magma_CSR, A.storage_type, queue ));
CHECK( magma_cmconvert( RCSR, &Rh, Magma_CSR, A.storage_type, queue ));
CHECK( magma_cmtransfer( Dh, D, Magma_CPU, A.memory_location, queue ));
CHECK( magma_cmtransfer( Rh, R, Magma_CPU, A.memory_location, queue ));
magma_cmfree( &Ah, queue );
magma_cmfree( &ACSR, queue );
magma_cmfree( &Dh, queue );
magma_cmfree( &DCSR, queue );
magma_cmfree( &Rh, queue );
magma_cmfree( &RCSR, queue );
}
cleanup:
if( info != 0 ){
magma_cmfree( D, queue );
magma_cmfree( R, queue );
}
return info;
}
|
