|
1.2.3
by Ondrej Certik
Import upstream version 0.6.0 |
1 |
/*
|
2 |
* -- SuperLU routine (version 3.0) --
|
|
3 |
* Univ. of California Berkeley, Xerox Palo Alto Research Center,
|
|
4 |
* and Lawrence Berkeley National Lab.
|
|
5 |
* October 15, 2003
|
|
6 |
*
|
|
7 |
*/
|
|
8 |
/*
|
|
9 |
Copyright (c) 1994 by Xerox Corporation. All rights reserved.
|
|
10 |
|
|
11 |
THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY
|
|
12 |
EXPRESSED OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
|
|
13 |
|
|
14 |
Permission is hereby granted to use or copy this program for any
|
|
15 |
purpose, provided the above notices are retained on all copies.
|
|
16 |
Permission to modify the code and to distribute modified code is
|
|
17 |
granted, provided the above notices are retained, and a notice that
|
|
18 |
the code was modified is included with the above copyright notice.
|
|
19 |
*/
|
|
20 |
||
21 |
#include <stdio.h> |
|
22 |
#include <stdlib.h> |
|
23 |
#include "zsp_defs.h" |
|
24 |
||
25 |
/*
|
|
26 |
* Function prototypes
|
|
27 |
*/
|
|
28 |
void zlsolve(int, int, doublecomplex *, doublecomplex *); |
|
29 |
void zmatvec(int, int, int, doublecomplex *, doublecomplex *, doublecomplex *); |
|
30 |
extern void zcheck_tempv(); |
|
31 |
||
32 |
void
|
|
33 |
zpanel_bmod ( |
|
34 |
const int m, /* in - number of rows in the matrix */ |
|
35 |
const int w, /* in */ |
|
36 |
const int jcol, /* in */ |
|
37 |
const int nseg, /* in */ |
|
38 |
doublecomplex *dense, /* out, of size n by w */ |
|
39 |
doublecomplex *tempv, /* working array */ |
|
40 |
int *segrep, /* in */ |
|
41 |
int *repfnz, /* in, of size n by w */ |
|
42 |
GlobalLU_t *Glu, /* modified */ |
|
43 |
SuperLUStat_t *stat /* output */ |
|
44 |
)
|
|
45 |
{
|
|
46 |
/*
|
|
47 |
* Purpose
|
|
48 |
* =======
|
|
49 |
*
|
|
50 |
* Performs numeric block updates (sup-panel) in topological order.
|
|
51 |
* It features: col-col, 2cols-col, 3cols-col, and sup-col updates.
|
|
52 |
* Special processing on the supernodal portion of L\U[*,j]
|
|
53 |
*
|
|
54 |
* Before entering this routine, the original nonzeros in the panel
|
|
55 |
* were already copied into the spa[m,w].
|
|
56 |
*
|
|
57 |
* Updated/Output parameters-
|
|
58 |
* dense[0:m-1,w]: L[*,j:j+w-1] and U[*,j:j+w-1] are returned
|
|
59 |
* collectively in the m-by-w vector dense[*].
|
|
60 |
*
|
|
61 |
*/
|
|
62 |
||
63 |
#ifdef USE_VENDOR_BLAS
|
|
64 |
#ifdef _CRAY
|
|
65 |
_fcd ftcs1 = _cptofcd("L", strlen("L")), |
|
66 |
ftcs2 = _cptofcd("N", strlen("N")), |
|
67 |
ftcs3 = _cptofcd("U", strlen("U")); |
|
68 |
#endif
|
|
69 |
int incx = 1, incy = 1; |
|
70 |
doublecomplex alpha, beta; |
|
71 |
#endif
|
|
72 |
||
73 |
register int k, ksub; |
|
74 |
int fsupc, nsupc, nsupr, nrow; |
|
75 |
int krep, krep_ind; |
|
76 |
doublecomplex ukj, ukj1, ukj2; |
|
77 |
int luptr, luptr1, luptr2; |
|
78 |
int segsze; |
|
79 |
int block_nrow; /* no of rows in a block row */ |
|
80 |
register int lptr; /* Points to the row subscripts of a supernode */ |
|
81 |
int kfnz, irow, no_zeros; |
|
82 |
register int isub, isub1, i; |
|
83 |
register int jj; /* Index through each column in the panel */ |
|
84 |
int *xsup, *supno; |
|
85 |
int *lsub, *xlsub; |
|
86 |
doublecomplex *lusup; |
|
87 |
int *xlusup; |
|
88 |
int *repfnz_col; /* repfnz[] for a column in the panel */ |
|
89 |
doublecomplex *dense_col; /* dense[] for a column in the panel */ |
|
90 |
doublecomplex *tempv1; /* Used in 1-D update */ |
|
91 |
doublecomplex *TriTmp, *MatvecTmp; /* used in 2-D update */ |
|
92 |
doublecomplex zero = {0.0, 0.0}; |
|
93 |
doublecomplex one = {1.0, 0.0}; |
|
94 |
doublecomplex comp_temp, comp_temp1; |
|
95 |
register int ldaTmp; |
|
96 |
register int r_ind, r_hi; |
|
97 |
static int first = 1, maxsuper, rowblk, colblk; |
|
98 |
flops_t *ops = stat->ops; |
|
99 |
||
100 |
xsup = Glu->xsup; |
|
101 |
supno = Glu->supno; |
|
102 |
lsub = Glu->lsub; |
|
103 |
xlsub = Glu->xlsub; |
|
104 |
lusup = Glu->lusup; |
|
105 |
xlusup = Glu->xlusup; |
|
106 |
||
107 |
if ( first ) { |
|
108 |
maxsuper = sp_ienv(3); |
|
109 |
rowblk = sp_ienv(4); |
|
110 |
colblk = sp_ienv(5); |
|
111 |
first = 0; |
|
112 |
}
|
|
113 |
ldaTmp = maxsuper + rowblk; |
|
114 |
||
115 |
/*
|
|
116 |
* For each nonz supernode segment of U[*,j] in topological order
|
|
117 |
*/
|
|
118 |
k = nseg - 1; |
|
119 |
for (ksub = 0; ksub < nseg; ksub++) { /* for each updating supernode */ |
|
120 |
||
121 |
/* krep = representative of current k-th supernode
|
|
122 |
* fsupc = first supernodal column
|
|
123 |
* nsupc = no of columns in a supernode
|
|
124 |
* nsupr = no of rows in a supernode
|
|
125 |
*/
|
|
126 |
krep = segrep[k--]; |
|
127 |
fsupc = xsup[supno[krep]]; |
|
128 |
nsupc = krep - fsupc + 1; |
|
129 |
nsupr = xlsub[fsupc+1] - xlsub[fsupc]; |
|
130 |
nrow = nsupr - nsupc; |
|
131 |
lptr = xlsub[fsupc]; |
|
132 |
krep_ind = lptr + nsupc - 1; |
|
133 |
||
134 |
repfnz_col = repfnz; |
|
135 |
dense_col = dense; |
|
136 |
||
137 |
if ( nsupc >= colblk && nrow > rowblk ) { /* 2-D block update */ |
|
138 |
||
139 |
TriTmp = tempv; |
|
140 |
||
141 |
/* Sequence through each column in panel -- triangular solves */
|
|
142 |
for (jj = jcol; jj < jcol + w; jj++, |
|
143 |
repfnz_col += m, dense_col += m, TriTmp += ldaTmp ) { |
|
144 |
||
145 |
kfnz = repfnz_col[krep]; |
|
146 |
if ( kfnz == EMPTY ) continue; /* Skip any zero segment */ |
|
147 |
||
148 |
segsze = krep - kfnz + 1; |
|
149 |
luptr = xlusup[fsupc]; |
|
150 |
||
151 |
ops[TRSV] += 4 * segsze * (segsze - 1); |
|
152 |
ops[GEMV] += 8 * nrow * segsze; |
|
153 |
||
154 |
/* Case 1: Update U-segment of size 1 -- col-col update */
|
|
155 |
if ( segsze == 1 ) { |
|
156 |
ukj = dense_col[lsub[krep_ind]]; |
|
157 |
luptr += nsupr*(nsupc-1) + nsupc; |
|
158 |
||
159 |
for (i = lptr + nsupc; i < xlsub[fsupc+1]; i++) { |
|
160 |
irow = lsub[i]; |
|
161 |
zz_mult(&comp_temp, &ukj, &lusup[luptr]); |
|
162 |
z_sub(&dense_col[irow], &dense_col[irow], &comp_temp); |
|
163 |
++luptr; |
|
164 |
}
|
|
165 |
||
166 |
} else if ( segsze <= 3 ) { |
|
167 |
ukj = dense_col[lsub[krep_ind]]; |
|
168 |
ukj1 = dense_col[lsub[krep_ind - 1]]; |
|
169 |
luptr += nsupr*(nsupc-1) + nsupc-1; |
|
170 |
luptr1 = luptr - nsupr; |
|
171 |
||
172 |
if ( segsze == 2 ) { |
|
173 |
zz_mult(&comp_temp, &ukj1, &lusup[luptr1]); |
|
174 |
z_sub(&ukj, &ukj, &comp_temp); |
|
175 |
dense_col[lsub[krep_ind]] = ukj; |
|
176 |
for (i = lptr + nsupc; i < xlsub[fsupc+1]; ++i) { |
|
177 |
irow = lsub[i]; |
|
178 |
luptr++; luptr1++; |
|
179 |
zz_mult(&comp_temp, &ukj, &lusup[luptr]); |
|
180 |
zz_mult(&comp_temp1, &ukj1, &lusup[luptr1]); |
|
181 |
z_add(&comp_temp, &comp_temp, &comp_temp1); |
|
182 |
z_sub(&dense_col[irow], &dense_col[irow], &comp_temp); |
|
183 |
}
|
|
184 |
} else { |
|
185 |
ukj2 = dense_col[lsub[krep_ind - 2]]; |
|
186 |
luptr2 = luptr1 - nsupr; |
|
187 |
zz_mult(&comp_temp, &ukj2, &lusup[luptr2-1]); |
|
188 |
z_sub(&ukj1, &ukj1, &comp_temp); |
|
189 |
||
190 |
zz_mult(&comp_temp, &ukj1, &lusup[luptr1]); |
|
191 |
zz_mult(&comp_temp1, &ukj2, &lusup[luptr2]); |
|
192 |
z_add(&comp_temp, &comp_temp, &comp_temp1); |
|
193 |
z_sub(&ukj, &ukj, &comp_temp); |
|
194 |
dense_col[lsub[krep_ind]] = ukj; |
|
195 |
dense_col[lsub[krep_ind-1]] = ukj1; |
|
196 |
for (i = lptr + nsupc; i < xlsub[fsupc+1]; ++i) { |
|
197 |
irow = lsub[i]; |
|
198 |
luptr++; luptr1++; luptr2++; |
|
199 |
zz_mult(&comp_temp, &ukj, &lusup[luptr]); |
|
200 |
zz_mult(&comp_temp1, &ukj1, &lusup[luptr1]); |
|
201 |
z_add(&comp_temp, &comp_temp, &comp_temp1); |
|
202 |
zz_mult(&comp_temp1, &ukj2, &lusup[luptr2]); |
|
203 |
z_add(&comp_temp, &comp_temp, &comp_temp1); |
|
204 |
z_sub(&dense_col[irow], &dense_col[irow], &comp_temp); |
|
205 |
}
|
|
206 |
}
|
|
207 |
||
208 |
} else { /* segsze >= 4 */ |
|
209 |
||
210 |
/* Copy U[*,j] segment from dense[*] to TriTmp[*], which
|
|
211 |
holds the result of triangular solves. */
|
|
212 |
no_zeros = kfnz - fsupc; |
|
213 |
isub = lptr + no_zeros; |
|
214 |
for (i = 0; i < segsze; ++i) { |
|
215 |
irow = lsub[isub]; |
|
216 |
TriTmp[i] = dense_col[irow]; /* Gather */ |
|
217 |
++isub; |
|
218 |
}
|
|
219 |
||
220 |
/* start effective triangle */
|
|
221 |
luptr += nsupr * no_zeros + no_zeros; |
|
222 |
||
223 |
#ifdef USE_VENDOR_BLAS
|
|
224 |
#ifdef _CRAY
|
|
225 |
CTRSV( ftcs1, ftcs2, ftcs3, &segsze, &lusup[luptr], |
|
226 |
&nsupr, TriTmp, &incx ); |
|
227 |
#else
|
|
228 |
ztrsv_( "L", "N", "U", &segsze, &lusup[luptr], |
|
229 |
&nsupr, TriTmp, &incx ); |
|
230 |
#endif
|
|
231 |
#else
|
|
232 |
zlsolve ( nsupr, segsze, &lusup[luptr], TriTmp ); |
|
233 |
#endif
|
|
234 |
||
235 |
||
236 |
} /* else ... */ |
|
237 |
||
238 |
} /* for jj ... end tri-solves */ |
|
239 |
||
240 |
/* Block row updates; push all the way into dense[*] block */
|
|
241 |
for ( r_ind = 0; r_ind < nrow; r_ind += rowblk ) { |
|
242 |
||
243 |
r_hi = SUPERLU_MIN(nrow, r_ind + rowblk); |
|
244 |
block_nrow = SUPERLU_MIN(rowblk, r_hi - r_ind); |
|
245 |
luptr = xlusup[fsupc] + nsupc + r_ind; |
|
246 |
isub1 = lptr + nsupc + r_ind; |
|
247 |
||
248 |
repfnz_col = repfnz; |
|
249 |
TriTmp = tempv; |
|
250 |
dense_col = dense; |
|
251 |
||
252 |
/* Sequence through each column in panel -- matrix-vector */
|
|
253 |
for (jj = jcol; jj < jcol + w; jj++, |
|
254 |
repfnz_col += m, dense_col += m, TriTmp += ldaTmp) { |
|
255 |
||
256 |
kfnz = repfnz_col[krep]; |
|
257 |
if ( kfnz == EMPTY ) continue; /* Skip any zero segment */ |
|
258 |
||
259 |
segsze = krep - kfnz + 1; |
|
260 |
if ( segsze <= 3 ) continue; /* skip unrolled cases */ |
|
261 |
||
262 |
/* Perform a block update, and scatter the result of
|
|
263 |
matrix-vector to dense[]. */
|
|
264 |
no_zeros = kfnz - fsupc; |
|
265 |
luptr1 = luptr + nsupr * no_zeros; |
|
266 |
MatvecTmp = &TriTmp[maxsuper]; |
|
267 |
||
268 |
#ifdef USE_VENDOR_BLAS
|
|
269 |
alpha = one; |
|
270 |
beta = zero; |
|
271 |
#ifdef _CRAY
|
|
272 |
CGEMV(ftcs2, &block_nrow, &segsze, &alpha, &lusup[luptr1], |
|
273 |
&nsupr, TriTmp, &incx, &beta, MatvecTmp, &incy); |
|
274 |
#else
|
|
275 |
zgemv_("N", &block_nrow, &segsze, &alpha, &lusup[luptr1], |
|
276 |
&nsupr, TriTmp, &incx, &beta, MatvecTmp, &incy); |
|
277 |
#endif
|
|
278 |
#else
|
|
279 |
zmatvec(nsupr, block_nrow, segsze, &lusup[luptr1], |
|
280 |
TriTmp, MatvecTmp); |
|
281 |
#endif
|
|
282 |
||
283 |
/* Scatter MatvecTmp[*] into SPA dense[*] temporarily
|
|
284 |
* such that MatvecTmp[*] can be re-used for the
|
|
285 |
* the next blok row update. dense[] will be copied into
|
|
286 |
* global store after the whole panel has been finished.
|
|
287 |
*/
|
|
288 |
isub = isub1; |
|
289 |
for (i = 0; i < block_nrow; i++) { |
|
290 |
irow = lsub[isub]; |
|
291 |
z_sub(&dense_col[irow], &dense_col[irow], |
|
292 |
&MatvecTmp[i]); |
|
293 |
MatvecTmp[i] = zero; |
|
294 |
++isub; |
|
295 |
}
|
|
296 |
||
297 |
} /* for jj ... */ |
|
298 |
||
299 |
} /* for each block row ... */ |
|
300 |
||
301 |
/* Scatter the triangular solves into SPA dense[*] */
|
|
302 |
repfnz_col = repfnz; |
|
303 |
TriTmp = tempv; |
|
304 |
dense_col = dense; |
|
305 |
||
306 |
for (jj = jcol; jj < jcol + w; jj++, |
|
307 |
repfnz_col += m, dense_col += m, TriTmp += ldaTmp) { |
|
308 |
kfnz = repfnz_col[krep]; |
|
309 |
if ( kfnz == EMPTY ) continue; /* Skip any zero segment */ |
|
310 |
||
311 |
segsze = krep - kfnz + 1; |
|
312 |
if ( segsze <= 3 ) continue; /* skip unrolled cases */ |
|
313 |
||
314 |
no_zeros = kfnz - fsupc; |
|
315 |
isub = lptr + no_zeros; |
|
316 |
for (i = 0; i < segsze; i++) { |
|
317 |
irow = lsub[isub]; |
|
318 |
dense_col[irow] = TriTmp[i]; |
|
319 |
TriTmp[i] = zero; |
|
320 |
++isub; |
|
321 |
}
|
|
322 |
||
323 |
} /* for jj ... */ |
|
324 |
||
325 |
} else { /* 1-D block modification */ |
|
326 |
||
327 |
||
328 |
/* Sequence through each column in the panel */
|
|
329 |
for (jj = jcol; jj < jcol + w; jj++, |
|
330 |
repfnz_col += m, dense_col += m) { |
|
331 |
||
332 |
kfnz = repfnz_col[krep]; |
|
333 |
if ( kfnz == EMPTY ) continue; /* Skip any zero segment */ |
|
334 |
||
335 |
segsze = krep - kfnz + 1; |
|
336 |
luptr = xlusup[fsupc]; |
|
337 |
||
338 |
ops[TRSV] += 4 * segsze * (segsze - 1); |
|
339 |
ops[GEMV] += 8 * nrow * segsze; |
|
340 |
||
341 |
/* Case 1: Update U-segment of size 1 -- col-col update */
|
|
342 |
if ( segsze == 1 ) { |
|
343 |
ukj = dense_col[lsub[krep_ind]]; |
|
344 |
luptr += nsupr*(nsupc-1) + nsupc; |
|
345 |
||
346 |
for (i = lptr + nsupc; i < xlsub[fsupc+1]; i++) { |
|
347 |
irow = lsub[i]; |
|
348 |
zz_mult(&comp_temp, &ukj, &lusup[luptr]); |
|
349 |
z_sub(&dense_col[irow], &dense_col[irow], &comp_temp); |
|
350 |
++luptr; |
|
351 |
}
|
|
352 |
||
353 |
} else if ( segsze <= 3 ) { |
|
354 |
ukj = dense_col[lsub[krep_ind]]; |
|
355 |
luptr += nsupr*(nsupc-1) + nsupc-1; |
|
356 |
ukj1 = dense_col[lsub[krep_ind - 1]]; |
|
357 |
luptr1 = luptr - nsupr; |
|
358 |
||
359 |
if ( segsze == 2 ) { |
|
360 |
zz_mult(&comp_temp, &ukj1, &lusup[luptr1]); |
|
361 |
z_sub(&ukj, &ukj, &comp_temp); |
|
362 |
dense_col[lsub[krep_ind]] = ukj; |
|
363 |
for (i = lptr + nsupc; i < xlsub[fsupc+1]; ++i) { |
|
364 |
irow = lsub[i]; |
|
365 |
++luptr; ++luptr1; |
|
366 |
zz_mult(&comp_temp, &ukj, &lusup[luptr]); |
|
367 |
zz_mult(&comp_temp1, &ukj1, &lusup[luptr1]); |
|
368 |
z_add(&comp_temp, &comp_temp, &comp_temp1); |
|
369 |
z_sub(&dense_col[irow], &dense_col[irow], &comp_temp); |
|
370 |
}
|
|
371 |
} else { |
|
372 |
ukj2 = dense_col[lsub[krep_ind - 2]]; |
|
373 |
luptr2 = luptr1 - nsupr; |
|
374 |
zz_mult(&comp_temp, &ukj2, &lusup[luptr2-1]); |
|
375 |
z_sub(&ukj1, &ukj1, &comp_temp); |
|
376 |
||
377 |
zz_mult(&comp_temp, &ukj1, &lusup[luptr1]); |
|
378 |
zz_mult(&comp_temp1, &ukj2, &lusup[luptr2]); |
|
379 |
z_add(&comp_temp, &comp_temp, &comp_temp1); |
|
380 |
z_sub(&ukj, &ukj, &comp_temp); |
|
381 |
dense_col[lsub[krep_ind]] = ukj; |
|
382 |
dense_col[lsub[krep_ind-1]] = ukj1; |
|
383 |
for (i = lptr + nsupc; i < xlsub[fsupc+1]; ++i) { |
|
384 |
irow = lsub[i]; |
|
385 |
++luptr; ++luptr1; ++luptr2; |
|
386 |
zz_mult(&comp_temp, &ukj, &lusup[luptr]); |
|
387 |
zz_mult(&comp_temp1, &ukj1, &lusup[luptr1]); |
|
388 |
z_add(&comp_temp, &comp_temp, &comp_temp1); |
|
389 |
zz_mult(&comp_temp1, &ukj2, &lusup[luptr2]); |
|
390 |
z_add(&comp_temp, &comp_temp, &comp_temp1); |
|
391 |
z_sub(&dense_col[irow], &dense_col[irow], &comp_temp); |
|
392 |
}
|
|
393 |
}
|
|
394 |
||
395 |
} else { /* segsze >= 4 */ |
|
396 |
/*
|
|
397 |
* Perform a triangular solve and block update,
|
|
398 |
* then scatter the result of sup-col update to dense[].
|
|
399 |
*/
|
|
400 |
no_zeros = kfnz - fsupc; |
|
401 |
||
402 |
/* Copy U[*,j] segment from dense[*] to tempv[*]:
|
|
403 |
* The result of triangular solve is in tempv[*];
|
|
404 |
* The result of matrix vector update is in dense_col[*]
|
|
405 |
*/
|
|
406 |
isub = lptr + no_zeros; |
|
407 |
for (i = 0; i < segsze; ++i) { |
|
408 |
irow = lsub[isub]; |
|
409 |
tempv[i] = dense_col[irow]; /* Gather */ |
|
410 |
++isub; |
|
411 |
}
|
|
412 |
||
413 |
/* start effective triangle */
|
|
414 |
luptr += nsupr * no_zeros + no_zeros; |
|
415 |
||
416 |
#ifdef USE_VENDOR_BLAS
|
|
417 |
#ifdef _CRAY
|
|
418 |
CTRSV( ftcs1, ftcs2, ftcs3, &segsze, &lusup[luptr], |
|
419 |
&nsupr, tempv, &incx ); |
|
420 |
#else
|
|
421 |
ztrsv_( "L", "N", "U", &segsze, &lusup[luptr], |
|
422 |
&nsupr, tempv, &incx ); |
|
423 |
#endif
|
|
424 |
||
425 |
luptr += segsze; /* Dense matrix-vector */ |
|
426 |
tempv1 = &tempv[segsze]; |
|
427 |
alpha = one; |
|
428 |
beta = zero; |
|
429 |
#ifdef _CRAY
|
|
430 |
CGEMV( ftcs2, &nrow, &segsze, &alpha, &lusup[luptr], |
|
431 |
&nsupr, tempv, &incx, &beta, tempv1, &incy ); |
|
432 |
#else
|
|
433 |
zgemv_( "N", &nrow, &segsze, &alpha, &lusup[luptr], |
|
434 |
&nsupr, tempv, &incx, &beta, tempv1, &incy ); |
|
435 |
#endif
|
|
436 |
#else
|
|
437 |
zlsolve ( nsupr, segsze, &lusup[luptr], tempv ); |
|
438 |
||
439 |
luptr += segsze; /* Dense matrix-vector */ |
|
440 |
tempv1 = &tempv[segsze]; |
|
441 |
zmatvec (nsupr, nrow, segsze, &lusup[luptr], tempv, tempv1); |
|
442 |
#endif
|
|
443 |
||
444 |
/* Scatter tempv[*] into SPA dense[*] temporarily, such
|
|
445 |
* that tempv[*] can be used for the triangular solve of
|
|
446 |
* the next column of the panel. They will be copied into
|
|
447 |
* ucol[*] after the whole panel has been finished.
|
|
448 |
*/
|
|
449 |
isub = lptr + no_zeros; |
|
450 |
for (i = 0; i < segsze; i++) { |
|
451 |
irow = lsub[isub]; |
|
452 |
dense_col[irow] = tempv[i]; |
|
453 |
tempv[i] = zero; |
|
454 |
isub++; |
|
455 |
}
|
|
456 |
||
457 |
/* Scatter the update from tempv1[*] into SPA dense[*] */
|
|
458 |
/* Start dense rectangular L */
|
|
459 |
for (i = 0; i < nrow; i++) { |
|
460 |
irow = lsub[isub]; |
|
461 |
z_sub(&dense_col[irow], &dense_col[irow], &tempv1[i]); |
|
462 |
tempv1[i] = zero; |
|
463 |
++isub; |
|
464 |
}
|
|
465 |
||
466 |
} /* else segsze>=4 ... */ |
|
467 |
||
468 |
} /* for each column in the panel... */ |
|
469 |
||
470 |
} /* else 1-D update ... */ |
|
471 |
||
472 |
} /* for each updating supernode ... */ |
|
473 |
||
474 |
}
|
|
475 |
||
476 |
||
477 |
||
478 |