root / src / pfact / HPL_dlocswpN.c @ 9
Historique | Voir | Annoter | Télécharger (17,35 ko)
| 1 | 1 | equemene | /*
|
|---|---|---|---|
| 2 | 1 | equemene | * -- High Performance Computing Linpack Benchmark (HPL)
|
| 3 | 1 | equemene | * HPL - 2.0 - September 10, 2008
|
| 4 | 1 | equemene | * Antoine P. Petitet
|
| 5 | 1 | equemene | * University of Tennessee, Knoxville
|
| 6 | 1 | equemene | * Innovative Computing Laboratory
|
| 7 | 1 | equemene | * (C) Copyright 2000-2008 All Rights Reserved
|
| 8 | 1 | equemene | *
|
| 9 | 1 | equemene | * -- Copyright notice and Licensing terms:
|
| 10 | 1 | equemene | *
|
| 11 | 1 | equemene | * Redistribution and use in source and binary forms, with or without
|
| 12 | 1 | equemene | * modification, are permitted provided that the following conditions
|
| 13 | 1 | equemene | * are met:
|
| 14 | 1 | equemene | *
|
| 15 | 1 | equemene | * 1. Redistributions of source code must retain the above copyright
|
| 16 | 1 | equemene | * notice, this list of conditions and the following disclaimer.
|
| 17 | 1 | equemene | *
|
| 18 | 1 | equemene | * 2. Redistributions in binary form must reproduce the above copyright
|
| 19 | 1 | equemene | * notice, this list of conditions, and the following disclaimer in the
|
| 20 | 1 | equemene | * documentation and/or other materials provided with the distribution.
|
| 21 | 1 | equemene | *
|
| 22 | 1 | equemene | * 3. All advertising materials mentioning features or use of this
|
| 23 | 1 | equemene | * software must display the following acknowledgement:
|
| 24 | 1 | equemene | * This product includes software developed at the University of
|
| 25 | 1 | equemene | * Tennessee, Knoxville, Innovative Computing Laboratory.
|
| 26 | 1 | equemene | *
|
| 27 | 1 | equemene | * 4. The name of the University, the name of the Laboratory, or the
|
| 28 | 1 | equemene | * names of its contributors may not be used to endorse or promote
|
| 29 | 1 | equemene | * products derived from this software without specific written
|
| 30 | 1 | equemene | * permission.
|
| 31 | 1 | equemene | *
|
| 32 | 1 | equemene | * -- Disclaimer:
|
| 33 | 1 | equemene | *
|
| 34 | 1 | equemene | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
| 35 | 1 | equemene | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
| 36 | 1 | equemene | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
| 37 | 1 | equemene | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY
|
| 38 | 1 | equemene | * OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
| 39 | 1 | equemene | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
| 40 | 1 | equemene | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
| 41 | 1 | equemene | * DATA OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
| 42 | 1 | equemene | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
| 43 | 1 | equemene | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
| 44 | 1 | equemene | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
| 45 | 1 | equemene | * ---------------------------------------------------------------------
|
| 46 | 1 | equemene | */
|
| 47 | 1 | equemene | /*
|
| 48 | 1 | equemene | * Include files
|
| 49 | 1 | equemene | */
|
| 50 | 1 | equemene | #include "hpl.h" |
| 51 | 1 | equemene | /*
|
| 52 | 1 | equemene | * Define default value for unrolling factor
|
| 53 | 1 | equemene | */
|
| 54 | 1 | equemene | #ifndef HPL_LOCSWP_DEPTH
|
| 55 | 1 | equemene | #define HPL_LOCSWP_DEPTH 32 |
| 56 | 1 | equemene | #define HPL_LOCSWP_LOG2_DEPTH 5 |
| 57 | 1 | equemene | #endif
|
| 58 | 1 | equemene | |
| 59 | 1 | equemene | #ifdef STDC_HEADERS
|
| 60 | 1 | equemene | void HPL_dlocswpN
|
| 61 | 1 | equemene | ( |
| 62 | 1 | equemene | HPL_T_panel * PANEL, |
| 63 | 1 | equemene | const int II, |
| 64 | 1 | equemene | const int JJ, |
| 65 | 1 | equemene | double * WORK
|
| 66 | 1 | equemene | ) |
| 67 | 1 | equemene | #else
|
| 68 | 1 | equemene | void HPL_dlocswpN
|
| 69 | 1 | equemene | ( PANEL, II, JJ, WORK ) |
| 70 | 1 | equemene | HPL_T_panel * PANEL; |
| 71 | 1 | equemene | const int II; |
| 72 | 1 | equemene | const int JJ; |
| 73 | 1 | equemene | double * WORK;
|
| 74 | 1 | equemene | #endif
|
| 75 | 1 | equemene | {
|
| 76 | 1 | equemene | /*
|
| 77 | 1 | equemene | * Purpose
|
| 78 | 1 | equemene | * =======
|
| 79 | 1 | equemene | *
|
| 80 | 1 | equemene | * HPL_dlocswpN performs the local swapping operations within a panel.
|
| 81 | 1 | equemene | * The lower triangular N0-by-N0 upper block of the panel is stored in
|
| 82 | 1 | equemene | * no-transpose form (i.e. just like the input matrix itself).
|
| 83 | 1 | equemene | *
|
| 84 | 1 | equemene | * Arguments
|
| 85 | 1 | equemene | * =========
|
| 86 | 1 | equemene | *
|
| 87 | 1 | equemene | * PANEL (local input/output) HPL_T_panel *
|
| 88 | 1 | equemene | * On entry, PANEL points to the data structure containing the
|
| 89 | 1 | equemene | * panel information.
|
| 90 | 1 | equemene | *
|
| 91 | 1 | equemene | * II (local input) const int
|
| 92 | 1 | equemene | * On entry, II specifies the row offset where the column to be
|
| 93 | 1 | equemene | * operated on starts with respect to the panel.
|
| 94 | 1 | equemene | *
|
| 95 | 1 | equemene | * JJ (local input) const int
|
| 96 | 1 | equemene | * On entry, JJ specifies the column offset where the column to
|
| 97 | 1 | equemene | * be operated on starts with respect to the panel.
|
| 98 | 1 | equemene | *
|
| 99 | 1 | equemene | * WORK (local workspace) double *
|
| 100 | 1 | equemene | * On entry, WORK is a workarray of size at least 2 * (4+2*N0).
|
| 101 | 1 | equemene | * WORK[0] contains the local maximum absolute value scalar,
|
| 102 | 1 | equemene | * WORK[1] contains the corresponding local row index, WORK[2]
|
| 103 | 1 | equemene | * contains the corresponding global row index, and WORK[3] is
|
| 104 | 1 | equemene | * the coordinate of process owning this max. The N0 length max
|
| 105 | 1 | equemene | * row is stored in WORK[4:4+N0-1]; Note that this is also the
|
| 106 | 1 | equemene | * JJth row (or column) of L1. The remaining part of this array
|
| 107 | 1 | equemene | * is used as workspace.
|
| 108 | 1 | equemene | *
|
| 109 | 1 | equemene | * ---------------------------------------------------------------------
|
| 110 | 1 | equemene | */
|
| 111 | 1 | equemene | /*
|
| 112 | 1 | equemene | * .. Local Variables ..
|
| 113 | 1 | equemene | */
|
| 114 | 1 | equemene | double gmax;
|
| 115 | 1 | equemene | double * A1, * A2, * L, * Wr0, * Wmx;
|
| 116 | 1 | equemene | int ilindx, lda, myrow, n0, nr, nu;
|
| 117 | 1 | equemene | register int i; |
| 118 | 1 | equemene | /* ..
|
| 119 | 1 | equemene | * .. Executable Statements ..
|
| 120 | 1 | equemene | */
|
| 121 | 1 | equemene | myrow = PANEL->grid->myrow; n0 = PANEL->jb; lda = PANEL->lda; |
| 122 | 1 | equemene | |
| 123 | 1 | equemene | Wr0 = ( Wmx = WORK + 4 ) + n0; Wmx[JJ] = gmax = WORK[0]; |
| 124 | 1 | equemene | nu = (int)( ( (unsigned int)(n0) >> HPL_LOCSWP_LOG2_DEPTH ) |
| 125 | 1 | equemene | << HPL_LOCSWP_LOG2_DEPTH ); |
| 126 | 1 | equemene | nr = n0 - nu; |
| 127 | 1 | equemene | /*
|
| 128 | 1 | equemene | * Replicated swap and copy of the current (new) row of A into L1
|
| 129 | 1 | equemene | */
|
| 130 | 1 | equemene | L = Mptr( PANEL->L1, JJ, 0, n0 );
|
| 131 | 1 | equemene | /*
|
| 132 | 1 | equemene | * If the pivot is non-zero ...
|
| 133 | 1 | equemene | */
|
| 134 | 1 | equemene | if( gmax != HPL_rzero )
|
| 135 | 1 | equemene | {
|
| 136 | 1 | equemene | /*
|
| 137 | 1 | equemene | * and if I own the current row of A ...
|
| 138 | 1 | equemene | */
|
| 139 | 1 | equemene | if( myrow == PANEL->prow )
|
| 140 | 1 | equemene | {
|
| 141 | 1 | equemene | /*
|
| 142 | 1 | equemene | * and if I also own the row to be swapped with the current row of A ...
|
| 143 | 1 | equemene | */
|
| 144 | 1 | equemene | if( myrow == (int)(WORK[3]) ) |
| 145 | 1 | equemene | {
|
| 146 | 1 | equemene | /*
|
| 147 | 1 | equemene | * and if the current row of A is not to swapped with itself ...
|
| 148 | 1 | equemene | */
|
| 149 | 1 | equemene | if( ( ilindx = (int)(WORK[1]) ) != 0 ) |
| 150 | 1 | equemene | {
|
| 151 | 1 | equemene | /*
|
| 152 | 1 | equemene | * then copy the max row into L1 and locally swap the 2 rows of A.
|
| 153 | 1 | equemene | */
|
| 154 | 1 | equemene | A1 = Mptr( PANEL->A, II, 0, lda );
|
| 155 | 1 | equemene | A2 = Mptr( A1, ilindx, 0, lda );
|
| 156 | 1 | equemene | |
| 157 | 1 | equemene | for( i = 0; i < nu; i += HPL_LOCSWP_DEPTH, |
| 158 | 1 | equemene | Wmx += HPL_LOCSWP_DEPTH, Wr0 += HPL_LOCSWP_DEPTH ) |
| 159 | 1 | equemene | {
|
| 160 | 1 | equemene | *L=*A1=Wmx[ 0]; *A2=Wr0[ 0]; L+=n0; A1+=lda; A2+=lda; |
| 161 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 1 ) |
| 162 | 1 | equemene | *L=*A1=Wmx[ 1]; *A2=Wr0[ 1]; L+=n0; A1+=lda; A2+=lda; |
| 163 | 1 | equemene | #endif
|
| 164 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 2 ) |
| 165 | 1 | equemene | *L=*A1=Wmx[ 2]; *A2=Wr0[ 2]; L+=n0; A1+=lda; A2+=lda; |
| 166 | 1 | equemene | *L=*A1=Wmx[ 3]; *A2=Wr0[ 3]; L+=n0; A1+=lda; A2+=lda; |
| 167 | 1 | equemene | #endif
|
| 168 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 4 ) |
| 169 | 1 | equemene | *L=*A1=Wmx[ 4]; *A2=Wr0[ 4]; L+=n0; A1+=lda; A2+=lda; |
| 170 | 1 | equemene | *L=*A1=Wmx[ 5]; *A2=Wr0[ 5]; L+=n0; A1+=lda; A2+=lda; |
| 171 | 1 | equemene | *L=*A1=Wmx[ 6]; *A2=Wr0[ 6]; L+=n0; A1+=lda; A2+=lda; |
| 172 | 1 | equemene | *L=*A1=Wmx[ 7]; *A2=Wr0[ 7]; L+=n0; A1+=lda; A2+=lda; |
| 173 | 1 | equemene | #endif
|
| 174 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 8 ) |
| 175 | 1 | equemene | *L=*A1=Wmx[ 8]; *A2=Wr0[ 8]; L+=n0; A1+=lda; A2+=lda; |
| 176 | 1 | equemene | *L=*A1=Wmx[ 9]; *A2=Wr0[ 9]; L+=n0; A1+=lda; A2+=lda; |
| 177 | 1 | equemene | *L=*A1=Wmx[10]; *A2=Wr0[10]; L+=n0; A1+=lda; A2+=lda; |
| 178 | 1 | equemene | *L=*A1=Wmx[11]; *A2=Wr0[11]; L+=n0; A1+=lda; A2+=lda; |
| 179 | 1 | equemene | *L=*A1=Wmx[12]; *A2=Wr0[12]; L+=n0; A1+=lda; A2+=lda; |
| 180 | 1 | equemene | *L=*A1=Wmx[13]; *A2=Wr0[13]; L+=n0; A1+=lda; A2+=lda; |
| 181 | 1 | equemene | *L=*A1=Wmx[14]; *A2=Wr0[14]; L+=n0; A1+=lda; A2+=lda; |
| 182 | 1 | equemene | *L=*A1=Wmx[15]; *A2=Wr0[15]; L+=n0; A1+=lda; A2+=lda; |
| 183 | 1 | equemene | #endif
|
| 184 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 16 ) |
| 185 | 1 | equemene | *L=*A1=Wmx[16]; *A2=Wr0[16]; L+=n0; A1+=lda; A2+=lda; |
| 186 | 1 | equemene | *L=*A1=Wmx[17]; *A2=Wr0[17]; L+=n0; A1+=lda; A2+=lda; |
| 187 | 1 | equemene | *L=*A1=Wmx[18]; *A2=Wr0[18]; L+=n0; A1+=lda; A2+=lda; |
| 188 | 1 | equemene | *L=*A1=Wmx[19]; *A2=Wr0[19]; L+=n0; A1+=lda; A2+=lda; |
| 189 | 1 | equemene | *L=*A1=Wmx[20]; *A2=Wr0[20]; L+=n0; A1+=lda; A2+=lda; |
| 190 | 1 | equemene | *L=*A1=Wmx[21]; *A2=Wr0[21]; L+=n0; A1+=lda; A2+=lda; |
| 191 | 1 | equemene | *L=*A1=Wmx[22]; *A2=Wr0[22]; L+=n0; A1+=lda; A2+=lda; |
| 192 | 1 | equemene | *L=*A1=Wmx[23]; *A2=Wr0[23]; L+=n0; A1+=lda; A2+=lda; |
| 193 | 1 | equemene | *L=*A1=Wmx[24]; *A2=Wr0[24]; L+=n0; A1+=lda; A2+=lda; |
| 194 | 1 | equemene | *L=*A1=Wmx[25]; *A2=Wr0[25]; L+=n0; A1+=lda; A2+=lda; |
| 195 | 1 | equemene | *L=*A1=Wmx[26]; *A2=Wr0[26]; L+=n0; A1+=lda; A2+=lda; |
| 196 | 1 | equemene | *L=*A1=Wmx[27]; *A2=Wr0[27]; L+=n0; A1+=lda; A2+=lda; |
| 197 | 1 | equemene | *L=*A1=Wmx[28]; *A2=Wr0[28]; L+=n0; A1+=lda; A2+=lda; |
| 198 | 1 | equemene | *L=*A1=Wmx[29]; *A2=Wr0[29]; L+=n0; A1+=lda; A2+=lda; |
| 199 | 1 | equemene | *L=*A1=Wmx[30]; *A2=Wr0[30]; L+=n0; A1+=lda; A2+=lda; |
| 200 | 1 | equemene | *L=*A1=Wmx[31]; *A2=Wr0[31]; L+=n0; A1+=lda; A2+=lda; |
| 201 | 1 | equemene | #endif
|
| 202 | 1 | equemene | } |
| 203 | 1 | equemene | for( i = 0; i < nr; i++, L += n0, A1 += lda, A2 += lda ) |
| 204 | 1 | equemene | { *L = *A1 = Wmx[i]; *A2 = Wr0[i]; }
|
| 205 | 1 | equemene | } |
| 206 | 1 | equemene | else
|
| 207 | 1 | equemene | {
|
| 208 | 1 | equemene | /*
|
| 209 | 1 | equemene | * otherwise the current row of A is swapped with itself, so just copy
|
| 210 | 1 | equemene | * the current of A into L1.
|
| 211 | 1 | equemene | */
|
| 212 | 1 | equemene | *Mptr( PANEL->A, II, JJ, lda ) = gmax; |
| 213 | 1 | equemene | |
| 214 | 1 | equemene | for( i = 0; i < nu; i += HPL_LOCSWP_DEPTH, |
| 215 | 1 | equemene | Wmx += HPL_LOCSWP_DEPTH, Wr0 += HPL_LOCSWP_DEPTH ) |
| 216 | 1 | equemene | {
|
| 217 | 1 | equemene | *L = Wmx[ 0]; L+=n0;
|
| 218 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 1 ) |
| 219 | 1 | equemene | *L = Wmx[ 1]; L+=n0;
|
| 220 | 1 | equemene | #endif
|
| 221 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 2 ) |
| 222 | 1 | equemene | *L = Wmx[ 2]; L+=n0; *L = Wmx[ 3]; L+=n0; |
| 223 | 1 | equemene | #endif
|
| 224 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 4 ) |
| 225 | 1 | equemene | *L = Wmx[ 4]; L+=n0; *L = Wmx[ 5]; L+=n0; |
| 226 | 1 | equemene | *L = Wmx[ 6]; L+=n0; *L = Wmx[ 7]; L+=n0; |
| 227 | 1 | equemene | #endif
|
| 228 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 8 ) |
| 229 | 1 | equemene | *L = Wmx[ 8]; L+=n0; *L = Wmx[ 9]; L+=n0; |
| 230 | 1 | equemene | *L = Wmx[10]; L+=n0; *L = Wmx[11]; L+=n0; |
| 231 | 1 | equemene | *L = Wmx[12]; L+=n0; *L = Wmx[13]; L+=n0; |
| 232 | 1 | equemene | *L = Wmx[14]; L+=n0; *L = Wmx[15]; L+=n0; |
| 233 | 1 | equemene | #endif
|
| 234 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 16 ) |
| 235 | 1 | equemene | *L = Wmx[16]; L+=n0; *L = Wmx[17]; L+=n0; |
| 236 | 1 | equemene | *L = Wmx[18]; L+=n0; *L = Wmx[19]; L+=n0; |
| 237 | 1 | equemene | *L = Wmx[20]; L+=n0; *L = Wmx[21]; L+=n0; |
| 238 | 1 | equemene | *L = Wmx[22]; L+=n0; *L = Wmx[23]; L+=n0; |
| 239 | 1 | equemene | *L = Wmx[24]; L+=n0; *L = Wmx[25]; L+=n0; |
| 240 | 1 | equemene | *L = Wmx[26]; L+=n0; *L = Wmx[27]; L+=n0; |
| 241 | 1 | equemene | *L = Wmx[28]; L+=n0; *L = Wmx[29]; L+=n0; |
| 242 | 1 | equemene | *L = Wmx[30]; L+=n0; *L = Wmx[31]; L+=n0; |
| 243 | 1 | equemene | #endif
|
| 244 | 1 | equemene | } |
| 245 | 1 | equemene | for( i = 0; i < nr; i++, L += n0 ) { *L = Wmx[i]; } |
| 246 | 1 | equemene | } |
| 247 | 1 | equemene | } |
| 248 | 1 | equemene | else
|
| 249 | 1 | equemene | {
|
| 250 | 1 | equemene | /*
|
| 251 | 1 | equemene | * otherwise, the row to be swapped with the current row of A is in Wmx,
|
| 252 | 1 | equemene | * so copy Wmx into L1 and A.
|
| 253 | 1 | equemene | */
|
| 254 | 1 | equemene | A1 = Mptr( PANEL->A, II, 0, lda );
|
| 255 | 1 | equemene | |
| 256 | 1 | equemene | for( i = 0; i < nu; i += HPL_LOCSWP_DEPTH, |
| 257 | 1 | equemene | Wmx += HPL_LOCSWP_DEPTH ) |
| 258 | 1 | equemene | {
|
| 259 | 1 | equemene | *L = *A1 = Wmx[ 0]; L += n0; A1 += lda;
|
| 260 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 1 ) |
| 261 | 1 | equemene | *L = *A1 = Wmx[ 1]; L += n0; A1 += lda;
|
| 262 | 1 | equemene | #endif
|
| 263 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 2 ) |
| 264 | 1 | equemene | *L = *A1 = Wmx[ 2]; L += n0; A1 += lda;
|
| 265 | 1 | equemene | *L = *A1 = Wmx[ 3]; L += n0; A1 += lda;
|
| 266 | 1 | equemene | #endif
|
| 267 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 4 ) |
| 268 | 1 | equemene | *L = *A1 = Wmx[ 4]; L += n0; A1 += lda;
|
| 269 | 1 | equemene | *L = *A1 = Wmx[ 5]; L += n0; A1 += lda;
|
| 270 | 1 | equemene | *L = *A1 = Wmx[ 6]; L += n0; A1 += lda;
|
| 271 | 1 | equemene | *L = *A1 = Wmx[ 7]; L += n0; A1 += lda;
|
| 272 | 1 | equemene | #endif
|
| 273 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 8 ) |
| 274 | 1 | equemene | *L = *A1 = Wmx[ 8]; L += n0; A1 += lda;
|
| 275 | 1 | equemene | *L = *A1 = Wmx[ 9]; L += n0; A1 += lda;
|
| 276 | 1 | equemene | *L = *A1 = Wmx[10]; L += n0; A1 += lda;
|
| 277 | 1 | equemene | *L = *A1 = Wmx[11]; L += n0; A1 += lda;
|
| 278 | 1 | equemene | *L = *A1 = Wmx[12]; L += n0; A1 += lda;
|
| 279 | 1 | equemene | *L = *A1 = Wmx[13]; L += n0; A1 += lda;
|
| 280 | 1 | equemene | *L = *A1 = Wmx[14]; L += n0; A1 += lda;
|
| 281 | 1 | equemene | *L = *A1 = Wmx[15]; L += n0; A1 += lda;
|
| 282 | 1 | equemene | #endif
|
| 283 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 16 ) |
| 284 | 1 | equemene | *L = *A1 = Wmx[16]; L += n0; A1 += lda;
|
| 285 | 1 | equemene | *L = *A1 = Wmx[17]; L += n0; A1 += lda;
|
| 286 | 1 | equemene | *L = *A1 = Wmx[18]; L += n0; A1 += lda;
|
| 287 | 1 | equemene | *L = *A1 = Wmx[19]; L += n0; A1 += lda;
|
| 288 | 1 | equemene | *L = *A1 = Wmx[20]; L += n0; A1 += lda;
|
| 289 | 1 | equemene | *L = *A1 = Wmx[21]; L += n0; A1 += lda;
|
| 290 | 1 | equemene | *L = *A1 = Wmx[22]; L += n0; A1 += lda;
|
| 291 | 1 | equemene | *L = *A1 = Wmx[23]; L += n0; A1 += lda;
|
| 292 | 1 | equemene | *L = *A1 = Wmx[24]; L += n0; A1 += lda;
|
| 293 | 1 | equemene | *L = *A1 = Wmx[25]; L += n0; A1 += lda;
|
| 294 | 1 | equemene | *L = *A1 = Wmx[26]; L += n0; A1 += lda;
|
| 295 | 1 | equemene | *L = *A1 = Wmx[27]; L += n0; A1 += lda;
|
| 296 | 1 | equemene | *L = *A1 = Wmx[28]; L += n0; A1 += lda;
|
| 297 | 1 | equemene | *L = *A1 = Wmx[29]; L += n0; A1 += lda;
|
| 298 | 1 | equemene | *L = *A1 = Wmx[30]; L += n0; A1 += lda;
|
| 299 | 1 | equemene | *L = *A1 = Wmx[31]; L += n0; A1 += lda;
|
| 300 | 1 | equemene | #endif
|
| 301 | 1 | equemene | } |
| 302 | 1 | equemene | |
| 303 | 1 | equemene | for( i = 0; i < nr; i++, L += n0, A1 += lda ) |
| 304 | 1 | equemene | { *L = *A1 = Wmx[i]; }
|
| 305 | 1 | equemene | } |
| 306 | 1 | equemene | } |
| 307 | 1 | equemene | else
|
| 308 | 1 | equemene | {
|
| 309 | 1 | equemene | /*
|
| 310 | 1 | equemene | * otherwise I do not own the current row of A, so copy the max row Wmx
|
| 311 | 1 | equemene | * into L1.
|
| 312 | 1 | equemene | */
|
| 313 | 1 | equemene | for( i = 0; i < nu; i += HPL_LOCSWP_DEPTH, |
| 314 | 1 | equemene | Wmx += HPL_LOCSWP_DEPTH ) |
| 315 | 1 | equemene | {
|
| 316 | 1 | equemene | *L = Wmx[ 0]; L+=n0;
|
| 317 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 1 ) |
| 318 | 1 | equemene | *L = Wmx[ 1]; L+=n0;
|
| 319 | 1 | equemene | #endif
|
| 320 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 2 ) |
| 321 | 1 | equemene | *L = Wmx[ 2]; L+=n0; *L = Wmx[ 3]; L+=n0; |
| 322 | 1 | equemene | #endif
|
| 323 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 4 ) |
| 324 | 1 | equemene | *L = Wmx[ 4]; L+=n0; *L = Wmx[ 5]; L+=n0; |
| 325 | 1 | equemene | *L = Wmx[ 6]; L+=n0; *L = Wmx[ 7]; L+=n0; |
| 326 | 1 | equemene | #endif
|
| 327 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 8 ) |
| 328 | 1 | equemene | *L = Wmx[ 8]; L+=n0; *L = Wmx[ 9]; L+=n0; |
| 329 | 1 | equemene | *L = Wmx[10]; L+=n0; *L = Wmx[11]; L+=n0; |
| 330 | 1 | equemene | *L = Wmx[12]; L+=n0; *L = Wmx[13]; L+=n0; |
| 331 | 1 | equemene | *L = Wmx[14]; L+=n0; *L = Wmx[15]; L+=n0; |
| 332 | 1 | equemene | #endif
|
| 333 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 16 ) |
| 334 | 1 | equemene | *L = Wmx[16]; L+=n0; *L = Wmx[17]; L+=n0; |
| 335 | 1 | equemene | *L = Wmx[18]; L+=n0; *L = Wmx[19]; L+=n0; |
| 336 | 1 | equemene | *L = Wmx[20]; L+=n0; *L = Wmx[21]; L+=n0; |
| 337 | 1 | equemene | *L = Wmx[22]; L+=n0; *L = Wmx[23]; L+=n0; |
| 338 | 1 | equemene | *L = Wmx[24]; L+=n0; *L = Wmx[25]; L+=n0; |
| 339 | 1 | equemene | *L = Wmx[26]; L+=n0; *L = Wmx[27]; L+=n0; |
| 340 | 1 | equemene | *L = Wmx[28]; L+=n0; *L = Wmx[29]; L+=n0; |
| 341 | 1 | equemene | *L = Wmx[30]; L+=n0; *L = Wmx[31]; L+=n0; |
| 342 | 1 | equemene | #endif
|
| 343 | 1 | equemene | } |
| 344 | 1 | equemene | for( i = 0; i < nr; i++, L += n0 ) { *L = Wmx[i]; } |
| 345 | 1 | equemene | /*
|
| 346 | 1 | equemene | * and if I own the max row, overwrite it with the current row Wr0.
|
| 347 | 1 | equemene | */
|
| 348 | 1 | equemene | if( myrow == (int)(WORK[3]) ) |
| 349 | 1 | equemene | {
|
| 350 | 1 | equemene | A2 = Mptr( PANEL->A, II + (size_t)(WORK[1]), 0, lda ); |
| 351 | 1 | equemene | |
| 352 | 1 | equemene | for( i = 0; i < nu; i += HPL_LOCSWP_DEPTH, |
| 353 | 1 | equemene | Wr0 += HPL_LOCSWP_DEPTH ) |
| 354 | 1 | equemene | {
|
| 355 | 1 | equemene | *A2 = Wr0[ 0]; A2+=lda;
|
| 356 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 1 ) |
| 357 | 1 | equemene | *A2 = Wr0[ 1]; A2+=lda;
|
| 358 | 1 | equemene | #endif
|
| 359 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 2 ) |
| 360 | 1 | equemene | *A2 = Wr0[ 2]; A2+=lda; *A2 = Wr0[ 3]; A2+=lda; |
| 361 | 1 | equemene | #endif
|
| 362 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 4 ) |
| 363 | 1 | equemene | *A2 = Wr0[ 4]; A2+=lda; *A2 = Wr0[ 5]; A2+=lda; |
| 364 | 1 | equemene | *A2 = Wr0[ 6]; A2+=lda; *A2 = Wr0[ 7]; A2+=lda; |
| 365 | 1 | equemene | #endif
|
| 366 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 8 ) |
| 367 | 1 | equemene | *A2 = Wr0[ 8]; A2+=lda; *A2 = Wr0[ 9]; A2+=lda; |
| 368 | 1 | equemene | *A2 = Wr0[10]; A2+=lda; *A2 = Wr0[11]; A2+=lda; |
| 369 | 1 | equemene | *A2 = Wr0[12]; A2+=lda; *A2 = Wr0[13]; A2+=lda; |
| 370 | 1 | equemene | *A2 = Wr0[14]; A2+=lda; *A2 = Wr0[15]; A2+=lda; |
| 371 | 1 | equemene | #endif
|
| 372 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 16 ) |
| 373 | 1 | equemene | *A2 = Wr0[16]; A2+=lda; *A2 = Wr0[17]; A2+=lda; |
| 374 | 1 | equemene | *A2 = Wr0[18]; A2+=lda; *A2 = Wr0[19]; A2+=lda; |
| 375 | 1 | equemene | *A2 = Wr0[20]; A2+=lda; *A2 = Wr0[21]; A2+=lda; |
| 376 | 1 | equemene | *A2 = Wr0[22]; A2+=lda; *A2 = Wr0[23]; A2+=lda; |
| 377 | 1 | equemene | *A2 = Wr0[24]; A2+=lda; *A2 = Wr0[25]; A2+=lda; |
| 378 | 1 | equemene | *A2 = Wr0[26]; A2+=lda; *A2 = Wr0[27]; A2+=lda; |
| 379 | 1 | equemene | *A2 = Wr0[28]; A2+=lda; *A2 = Wr0[29]; A2+=lda; |
| 380 | 1 | equemene | *A2 = Wr0[30]; A2+=lda; *A2 = Wr0[31]; A2+=lda; |
| 381 | 1 | equemene | #endif
|
| 382 | 1 | equemene | } |
| 383 | 1 | equemene | |
| 384 | 1 | equemene | for( i = 0; i < nr; i++, A2 += lda ) { *A2 = Wr0[i]; } |
| 385 | 1 | equemene | } |
| 386 | 1 | equemene | } |
| 387 | 1 | equemene | } |
| 388 | 1 | equemene | else
|
| 389 | 1 | equemene | {
|
| 390 | 1 | equemene | /*
|
| 391 | 1 | equemene | * Otherwise the max element in the current column is zero, simply copy
|
| 392 | 1 | equemene | * the current row Wr0 into L1. The matrix is singular.
|
| 393 | 1 | equemene | */
|
| 394 | 1 | equemene | for( i = 0; i < nu; i += HPL_LOCSWP_DEPTH, |
| 395 | 1 | equemene | Wr0 += HPL_LOCSWP_DEPTH ) |
| 396 | 1 | equemene | {
|
| 397 | 1 | equemene | *L = Wr0[ 0]; L+=n0;
|
| 398 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 1 ) |
| 399 | 1 | equemene | *L = Wr0[ 1]; L+=n0;
|
| 400 | 1 | equemene | #endif
|
| 401 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 2 ) |
| 402 | 1 | equemene | *L = Wr0[ 2]; L+=n0; *L = Wr0[ 3]; L+=n0; |
| 403 | 1 | equemene | #endif
|
| 404 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 4 ) |
| 405 | 1 | equemene | *L = Wr0[ 4]; L+=n0; *L = Wr0[ 5]; L+=n0; |
| 406 | 1 | equemene | *L = Wr0[ 6]; L+=n0; *L = Wr0[ 7]; L+=n0; |
| 407 | 1 | equemene | #endif
|
| 408 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 8 ) |
| 409 | 1 | equemene | *L = Wr0[ 8]; L+=n0; *L = Wr0[ 9]; L+=n0; |
| 410 | 1 | equemene | *L = Wr0[10]; L+=n0; *L = Wr0[11]; L+=n0; |
| 411 | 1 | equemene | *L = Wr0[12]; L+=n0; *L = Wr0[13]; L+=n0; |
| 412 | 1 | equemene | *L = Wr0[14]; L+=n0; *L = Wr0[15]; L+=n0; |
| 413 | 1 | equemene | #endif
|
| 414 | 1 | equemene | #if ( HPL_LOCSWP_DEPTH > 16 ) |
| 415 | 1 | equemene | *L = Wr0[16]; L+=n0; *L = Wr0[17]; L+=n0; |
| 416 | 1 | equemene | *L = Wr0[18]; L+=n0; *L = Wr0[19]; L+=n0; |
| 417 | 1 | equemene | *L = Wr0[20]; L+=n0; *L = Wr0[21]; L+=n0; |
| 418 | 1 | equemene | *L = Wr0[22]; L+=n0; *L = Wr0[23]; L+=n0; |
| 419 | 1 | equemene | *L = Wr0[24]; L+=n0; *L = Wr0[25]; L+=n0; |
| 420 | 1 | equemene | *L = Wr0[26]; L+=n0; *L = Wr0[27]; L+=n0; |
| 421 | 1 | equemene | *L = Wr0[28]; L+=n0; *L = Wr0[29]; L+=n0; |
| 422 | 1 | equemene | *L = Wr0[30]; L+=n0; *L = Wr0[31]; L+=n0; |
| 423 | 1 | equemene | #endif
|
| 424 | 1 | equemene | } |
| 425 | 1 | equemene | |
| 426 | 1 | equemene | for( i = 0; i < nr; i++, L += n0 ) { *L = Wr0[i]; } |
| 427 | 1 | equemene | /*
|
| 428 | 1 | equemene | * set INFO.
|
| 429 | 1 | equemene | */
|
| 430 | 1 | equemene | if( *(PANEL->DINFO) == 0.0 ) |
| 431 | 1 | equemene | *(PANEL->DINFO) = (double)(PANEL->ia + JJ + 1); |
| 432 | 1 | equemene | } |
| 433 | 1 | equemene | /*
|
| 434 | 1 | equemene | * End of HPL_dlocswpN
|
| 435 | 1 | equemene | */
|
| 436 | 1 | equemene | } |