root / testing / ptest / HPL_pdinfo.c
Historique | Voir | Annoter | Télécharger (43,51 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 | #ifdef STDC_HEADERS
|
| 53 | 1 | equemene | void HPL_pdinfo
|
| 54 | 1 | equemene | ( |
| 55 | 1 | equemene | HPL_T_test * TEST, |
| 56 | 1 | equemene | int * NS,
|
| 57 | 1 | equemene | int * N,
|
| 58 | 1 | equemene | int * NBS,
|
| 59 | 1 | equemene | int * NB,
|
| 60 | 1 | equemene | HPL_T_ORDER * PMAPPIN, |
| 61 | 1 | equemene | int * NPQS,
|
| 62 | 1 | equemene | int * P,
|
| 63 | 1 | equemene | int * Q,
|
| 64 | 1 | equemene | int * NPFS,
|
| 65 | 1 | equemene | HPL_T_FACT * PF, |
| 66 | 1 | equemene | int * NBMS,
|
| 67 | 1 | equemene | int * NBM,
|
| 68 | 1 | equemene | int * NDVS,
|
| 69 | 1 | equemene | int * NDV,
|
| 70 | 1 | equemene | int * NRFS,
|
| 71 | 1 | equemene | HPL_T_FACT * RF, |
| 72 | 1 | equemene | int * NTPS,
|
| 73 | 1 | equemene | HPL_T_TOP * TP, |
| 74 | 1 | equemene | int * NDHS,
|
| 75 | 1 | equemene | int * DH,
|
| 76 | 1 | equemene | HPL_T_SWAP * FSWAP, |
| 77 | 1 | equemene | int * TSWAP,
|
| 78 | 1 | equemene | int * L1NOTRAN,
|
| 79 | 1 | equemene | int * UNOTRAN,
|
| 80 | 1 | equemene | int * EQUIL,
|
| 81 | 1 | equemene | int * ALIGN
|
| 82 | 1 | equemene | ) |
| 83 | 1 | equemene | #else
|
| 84 | 1 | equemene | void HPL_pdinfo
|
| 85 | 1 | equemene | ( TEST, NS, N, NBS, NB, PMAPPIN, NPQS, P, Q, NPFS, PF, NBMS, NBM, NDVS, NDV, NRFS, RF, NTPS, TP, NDHS, DH, FSWAP, TSWAP, L1NOTRAN, UNOTRAN, EQUIL, ALIGN ) |
| 86 | 1 | equemene | HPL_T_test * TEST; |
| 87 | 1 | equemene | int * NS;
|
| 88 | 1 | equemene | int * N;
|
| 89 | 1 | equemene | int * NBS;
|
| 90 | 1 | equemene | int * NB;
|
| 91 | 1 | equemene | HPL_T_ORDER * PMAPPIN; |
| 92 | 1 | equemene | int * NPQS;
|
| 93 | 1 | equemene | int * P;
|
| 94 | 1 | equemene | int * Q;
|
| 95 | 1 | equemene | int * NPFS;
|
| 96 | 1 | equemene | HPL_T_FACT * PF; |
| 97 | 1 | equemene | int * NBMS;
|
| 98 | 1 | equemene | int * NBM;
|
| 99 | 1 | equemene | int * NDVS;
|
| 100 | 1 | equemene | int * NDV;
|
| 101 | 1 | equemene | int * NRFS;
|
| 102 | 1 | equemene | HPL_T_FACT * RF; |
| 103 | 1 | equemene | int * NTPS;
|
| 104 | 1 | equemene | HPL_T_TOP * TP; |
| 105 | 1 | equemene | int * NDHS;
|
| 106 | 1 | equemene | int * DH;
|
| 107 | 1 | equemene | HPL_T_SWAP * FSWAP; |
| 108 | 1 | equemene | int * TSWAP;
|
| 109 | 1 | equemene | int * L1NOTRAN;
|
| 110 | 1 | equemene | int * UNOTRAN;
|
| 111 | 1 | equemene | int * EQUIL;
|
| 112 | 1 | equemene | int * ALIGN;
|
| 113 | 1 | equemene | #endif
|
| 114 | 1 | equemene | {
|
| 115 | 1 | equemene | /*
|
| 116 | 1 | equemene | * Purpose
|
| 117 | 1 | equemene | * =======
|
| 118 | 1 | equemene | *
|
| 119 | 1 | equemene | * HPL_pdinfo reads the startup information for the various tests and
|
| 120 | 1 | equemene | * transmits it to all processes.
|
| 121 | 1 | equemene | *
|
| 122 | 1 | equemene | * Arguments
|
| 123 | 1 | equemene | * =========
|
| 124 | 1 | equemene | *
|
| 125 | 1 | equemene | * TEST (global output) HPL_T_test *
|
| 126 | 1 | equemene | * On entry, TEST points to a testing data structure. On exit,
|
| 127 | 1 | equemene | * the fields of this data structure are initialized as follows:
|
| 128 | 1 | equemene | * TEST->outfp specifies the output file where the results will
|
| 129 | 1 | equemene | * be printed. It is only defined and used by the process 0 of
|
| 130 | 1 | equemene | * the grid. TEST->thrsh specifies the threshhold value for the
|
| 131 | 1 | equemene | * test ratio. TEST->epsil is the relative machine precision of
|
| 132 | 1 | equemene | * the distributed computer. Finally the test counters, kfail,
|
| 133 | 1 | equemene | * kpass, kskip, ktest are initialized to zero.
|
| 134 | 1 | equemene | *
|
| 135 | 1 | equemene | * NS (global output) int *
|
| 136 | 1 | equemene | * On exit, NS specifies the number of different problem sizes
|
| 137 | 1 | equemene | * to be tested. NS is less than or equal to HPL_MAX_PARAM.
|
| 138 | 1 | equemene | *
|
| 139 | 1 | equemene | * N (global output) int *
|
| 140 | 1 | equemene | * On entry, N is an array of dimension HPL_MAX_PARAM. On exit,
|
| 141 | 1 | equemene | * the first NS entries of this array contain the problem sizes
|
| 142 | 1 | equemene | * to run the code with.
|
| 143 | 1 | equemene | *
|
| 144 | 1 | equemene | * NBS (global output) int *
|
| 145 | 1 | equemene | * On exit, NBS specifies the number of different distribution
|
| 146 | 1 | equemene | * blocking factors to be tested. NBS must be less than or equal
|
| 147 | 1 | equemene | * to HPL_MAX_PARAM.
|
| 148 | 1 | equemene | *
|
| 149 | 1 | equemene | * NB (global output) int *
|
| 150 | 1 | equemene | * On exit, PMAPPIN specifies the process mapping onto the no-
|
| 151 | 1 | equemene | * des of the MPI machine configuration. PMAPPIN defaults to
|
| 152 | 1 | equemene | * row-major ordering.
|
| 153 | 1 | equemene | *
|
| 154 | 1 | equemene | * PMAPPIN (global output) HPL_T_ORDER *
|
| 155 | 1 | equemene | * On entry, NB is an array of dimension HPL_MAX_PARAM. On exit,
|
| 156 | 1 | equemene | * the first NBS entries of this array contain the values of the
|
| 157 | 1 | equemene | * various distribution blocking factors, to run the code with.
|
| 158 | 1 | equemene | *
|
| 159 | 1 | equemene | * NPQS (global output) int *
|
| 160 | 1 | equemene | * On exit, NPQS specifies the number of different values that
|
| 161 | 1 | equemene | * can be used for P and Q, i.e., the number of process grids to
|
| 162 | 1 | equemene | * run the code with. NPQS must be less than or equal to
|
| 163 | 1 | equemene | * HPL_MAX_PARAM.
|
| 164 | 1 | equemene | *
|
| 165 | 1 | equemene | * P (global output) int *
|
| 166 | 1 | equemene | * On entry, P is an array of dimension HPL_MAX_PARAM. On exit,
|
| 167 | 1 | equemene | * the first NPQS entries of this array contain the values of P,
|
| 168 | 1 | equemene | * the number of process rows of the NPQS grids to run the code
|
| 169 | 1 | equemene | * with.
|
| 170 | 1 | equemene | *
|
| 171 | 1 | equemene | * Q (global output) int *
|
| 172 | 1 | equemene | * On entry, Q is an array of dimension HPL_MAX_PARAM. On exit,
|
| 173 | 1 | equemene | * the first NPQS entries of this array contain the values of Q,
|
| 174 | 1 | equemene | * the number of process columns of the NPQS grids to run the
|
| 175 | 1 | equemene | * code with.
|
| 176 | 1 | equemene | *
|
| 177 | 1 | equemene | * NPFS (global output) int *
|
| 178 | 1 | equemene | * On exit, NPFS specifies the number of different values that
|
| 179 | 1 | equemene | * can be used for PF : the panel factorization algorithm to run
|
| 180 | 1 | equemene | * the code with. NPFS is less than or equal to HPL_MAX_PARAM.
|
| 181 | 1 | equemene | *
|
| 182 | 1 | equemene | * PF (global output) HPL_T_FACT *
|
| 183 | 1 | equemene | * On entry, PF is an array of dimension HPL_MAX_PARAM. On exit,
|
| 184 | 1 | equemene | * the first NPFS entries of this array contain the various
|
| 185 | 1 | equemene | * panel factorization algorithms to run the code with.
|
| 186 | 1 | equemene | *
|
| 187 | 1 | equemene | * NBMS (global output) int *
|
| 188 | 1 | equemene | * On exit, NBMS specifies the number of various recursive
|
| 189 | 1 | equemene | * stopping criteria to be tested. NBMS must be less than or
|
| 190 | 1 | equemene | * equal to HPL_MAX_PARAM.
|
| 191 | 1 | equemene | *
|
| 192 | 1 | equemene | * NBM (global output) int *
|
| 193 | 1 | equemene | * On entry, NBM is an array of dimension HPL_MAX_PARAM. On
|
| 194 | 1 | equemene | * exit, the first NBMS entries of this array contain the values
|
| 195 | 1 | equemene | * of the various recursive stopping criteria to be tested.
|
| 196 | 1 | equemene | *
|
| 197 | 1 | equemene | * NDVS (global output) int *
|
| 198 | 1 | equemene | * On exit, NDVS specifies the number of various numbers of
|
| 199 | 1 | equemene | * panels in recursion to be tested. NDVS is less than or equal
|
| 200 | 1 | equemene | * to HPL_MAX_PARAM.
|
| 201 | 1 | equemene | *
|
| 202 | 1 | equemene | * NDV (global output) int *
|
| 203 | 1 | equemene | * On entry, NDV is an array of dimension HPL_MAX_PARAM. On
|
| 204 | 1 | equemene | * exit, the first NDVS entries of this array contain the values
|
| 205 | 1 | equemene | * of the various numbers of panels in recursion to be tested.
|
| 206 | 1 | equemene | *
|
| 207 | 1 | equemene | * NRFS (global output) int *
|
| 208 | 1 | equemene | * On exit, NRFS specifies the number of different values that
|
| 209 | 1 | equemene | * can be used for RF : the recursive factorization algorithm to
|
| 210 | 1 | equemene | * be tested. NRFS is less than or equal to HPL_MAX_PARAM.
|
| 211 | 1 | equemene | *
|
| 212 | 1 | equemene | * RF (global output) HPL_T_FACT *
|
| 213 | 1 | equemene | * On entry, RF is an array of dimension HPL_MAX_PARAM. On exit,
|
| 214 | 1 | equemene | * the first NRFS entries of this array contain the various
|
| 215 | 1 | equemene | * recursive factorization algorithms to run the code with.
|
| 216 | 1 | equemene | *
|
| 217 | 1 | equemene | * NTPS (global output) int *
|
| 218 | 1 | equemene | * On exit, NTPS specifies the number of different values that
|
| 219 | 1 | equemene | * can be used for the broadcast topologies to be tested. NTPS
|
| 220 | 1 | equemene | * is less than or equal to HPL_MAX_PARAM.
|
| 221 | 1 | equemene | *
|
| 222 | 1 | equemene | * TP (global output) HPL_T_TOP *
|
| 223 | 1 | equemene | * On entry, TP is an array of dimension HPL_MAX_PARAM. On exit,
|
| 224 | 1 | equemene | * the first NTPS entries of this array contain the various
|
| 225 | 1 | equemene | * broadcast (along rows) topologies to run the code with.
|
| 226 | 1 | equemene | *
|
| 227 | 1 | equemene | * NDHS (global output) int *
|
| 228 | 1 | equemene | * On exit, NDHS specifies the number of different values that
|
| 229 | 1 | equemene | * can be used for the lookahead depths to be tested. NDHS is
|
| 230 | 1 | equemene | * less than or equal to HPL_MAX_PARAM.
|
| 231 | 1 | equemene | *
|
| 232 | 1 | equemene | * DH (global output) int *
|
| 233 | 1 | equemene | * On entry, DH is an array of dimension HPL_MAX_PARAM. On
|
| 234 | 1 | equemene | * exit, the first NDHS entries of this array contain the values
|
| 235 | 1 | equemene | * of lookahead depths to run the code with. Such a value is at
|
| 236 | 1 | equemene | * least 0 (no-lookahead) or greater than zero.
|
| 237 | 1 | equemene | *
|
| 238 | 1 | equemene | * FSWAP (global output) HPL_T_SWAP *
|
| 239 | 1 | equemene | * On exit, FSWAP specifies the swapping algorithm to be used in
|
| 240 | 1 | equemene | * all tests.
|
| 241 | 1 | equemene | *
|
| 242 | 1 | equemene | * TSWAP (global output) int *
|
| 243 | 1 | equemene | * On exit, TSWAP specifies the swapping threshold as a number
|
| 244 | 1 | equemene | * of columns when the mixed swapping algorithm was chosen.
|
| 245 | 1 | equemene | *
|
| 246 | 1 | equemene | * L1NOTRA (global output) int *
|
| 247 | 1 | equemene | * On exit, L1NOTRAN specifies whether the upper triangle of the
|
| 248 | 1 | equemene | * panels of columns should be stored in no-transposed form
|
| 249 | 1 | equemene | * (L1NOTRAN=1) or in transposed form (L1NOTRAN=0).
|
| 250 | 1 | equemene | *
|
| 251 | 1 | equemene | * UNOTRAN (global output) int *
|
| 252 | 1 | equemene | * On exit, UNOTRAN specifies whether the panels of rows should
|
| 253 | 1 | equemene | * be stored in no-transposed form (UNOTRAN=1) or transposed
|
| 254 | 1 | equemene | * form (UNOTRAN=0) during their broadcast.
|
| 255 | 1 | equemene | *
|
| 256 | 1 | equemene | * EQUIL (global output) int *
|
| 257 | 1 | equemene | * On exit, EQUIL specifies whether equilibration during the
|
| 258 | 1 | equemene | * swap-broadcast of the panel of rows should be performed
|
| 259 | 1 | equemene | * (EQUIL=1) or not (EQUIL=0).
|
| 260 | 1 | equemene | *
|
| 261 | 1 | equemene | * ALIGN (global output) int *
|
| 262 | 1 | equemene | * On exit, ALIGN specifies the alignment of the dynamically
|
| 263 | 1 | equemene | * allocated buffers in double precision words. ALIGN is greater
|
| 264 | 1 | equemene | * than zero.
|
| 265 | 1 | equemene | *
|
| 266 | 1 | equemene | * ---------------------------------------------------------------------
|
| 267 | 1 | equemene | */
|
| 268 | 1 | equemene | /*
|
| 269 | 1 | equemene | * .. Local Variables ..
|
| 270 | 1 | equemene | */
|
| 271 | 1 | equemene | char file[HPL_LINE_MAX], line[HPL_LINE_MAX],
|
| 272 | 1 | equemene | auth[HPL_LINE_MAX], num [HPL_LINE_MAX]; |
| 273 | 1 | equemene | FILE * infp; |
| 274 | 1 | equemene | int * iwork = NULL; |
| 275 | 1 | equemene | char * lineptr;
|
| 276 | 1 | equemene | int error=0, fid, i, j, lwork, maxp, nprocs, |
| 277 | 1 | equemene | rank, size; |
| 278 | 1 | equemene | /* ..
|
| 279 | 1 | equemene | * .. Executable Statements ..
|
| 280 | 1 | equemene | */
|
| 281 | 1 | equemene | MPI_Comm_rank( MPI_COMM_WORLD, &rank ); |
| 282 | 1 | equemene | MPI_Comm_size( MPI_COMM_WORLD, &size ); |
| 283 | 1 | equemene | /*
|
| 284 | 1 | equemene | * Initialize the TEST data structure with default values
|
| 285 | 1 | equemene | */
|
| 286 | 1 | equemene | TEST->outfp = stderr; TEST->epsil = 2.0e-16; TEST->thrsh = 16.0; |
| 287 | 1 | equemene | TEST->kfail = TEST->kpass = TEST->kskip = TEST->ktest = 0;
|
| 288 | 1 | equemene | /*
|
| 289 | 1 | equemene | * Process 0 reads the input data, broadcasts to other processes and
|
| 290 | 1 | equemene | * writes needed information to TEST->outfp.
|
| 291 | 1 | equemene | */
|
| 292 | 1 | equemene | if( rank == 0 ) |
| 293 | 1 | equemene | {
|
| 294 | 1 | equemene | /*
|
| 295 | 1 | equemene | * Open file and skip data file header
|
| 296 | 1 | equemene | */
|
| 297 | 1 | equemene | if( ( infp = fopen( "HPL.dat", "r" ) ) == NULL ) |
| 298 | 1 | equemene | {
|
| 299 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 300 | 1 | equemene | "cannot open file HPL.dat" );
|
| 301 | 1 | equemene | error = 1; goto label_error; |
| 302 | 1 | equemene | } |
| 303 | 1 | equemene | |
| 304 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 305 | 1 | equemene | (void) fgets( auth, HPL_LINE_MAX - 2, infp ); |
| 306 | 1 | equemene | /*
|
| 307 | 1 | equemene | * Read name and unit number for summary output file
|
| 308 | 1 | equemene | */
|
| 309 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 310 | 1 | equemene | (void) sscanf( line, "%s", file ); |
| 311 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 312 | 1 | equemene | (void) sscanf( line, "%s", num ); |
| 313 | 1 | equemene | fid = atoi( num ); |
| 314 | 1 | equemene | if ( fid == 6 ) TEST->outfp = stdout; |
| 315 | 1 | equemene | else if( fid == 7 ) TEST->outfp = stderr; |
| 316 | 1 | equemene | else if( ( TEST->outfp = fopen( file, "w" ) ) == NULL ) |
| 317 | 1 | equemene | {
|
| 318 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "cannot open file %s.", |
| 319 | 1 | equemene | file ); |
| 320 | 1 | equemene | error = 1; goto label_error; |
| 321 | 1 | equemene | } |
| 322 | 1 | equemene | /*
|
| 323 | 1 | equemene | * Read and check the parameter values for the tests.
|
| 324 | 1 | equemene | *
|
| 325 | 1 | equemene | * Problem size (>=0) (N)
|
| 326 | 1 | equemene | */
|
| 327 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 328 | 1 | equemene | (void) sscanf( line, "%s", num ); *NS = atoi( num ); |
| 329 | 1 | equemene | if( ( *NS < 1 ) || ( *NS > HPL_MAX_PARAM ) ) |
| 330 | 1 | equemene | {
|
| 331 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %d", |
| 332 | 1 | equemene | "Number of values of N is less than 1 or greater than",
|
| 333 | 1 | equemene | HPL_MAX_PARAM ); |
| 334 | 1 | equemene | error = 1; goto label_error; |
| 335 | 1 | equemene | } |
| 336 | 1 | equemene | |
| 337 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 338 | 1 | equemene | for( i = 0; i < *NS; i++ ) |
| 339 | 1 | equemene | {
|
| 340 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 341 | 1 | equemene | if( ( N[ i ] = atoi( num ) ) < 0 ) |
| 342 | 1 | equemene | {
|
| 343 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 344 | 1 | equemene | "Value of N less than 0" );
|
| 345 | 1 | equemene | error = 1; goto label_error; |
| 346 | 1 | equemene | } |
| 347 | 1 | equemene | } |
| 348 | 1 | equemene | /*
|
| 349 | 1 | equemene | * Block size (>=1) (NB)
|
| 350 | 1 | equemene | */
|
| 351 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 352 | 1 | equemene | (void) sscanf( line, "%s", num ); *NBS = atoi( num ); |
| 353 | 1 | equemene | if( ( *NBS < 1 ) || ( *NBS > HPL_MAX_PARAM ) ) |
| 354 | 1 | equemene | {
|
| 355 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %s %d", |
| 356 | 1 | equemene | "Number of values of NB is less than 1 or",
|
| 357 | 1 | equemene | "greater than", HPL_MAX_PARAM );
|
| 358 | 1 | equemene | error = 1; goto label_error; |
| 359 | 1 | equemene | } |
| 360 | 1 | equemene | |
| 361 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 362 | 1 | equemene | for( i = 0; i < *NBS; i++ ) |
| 363 | 1 | equemene | {
|
| 364 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 365 | 1 | equemene | if( ( NB[ i ] = atoi( num ) ) < 1 ) |
| 366 | 1 | equemene | {
|
| 367 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 368 | 1 | equemene | "Value of NB less than 1" );
|
| 369 | 1 | equemene | error = 1; goto label_error; |
| 370 | 1 | equemene | } |
| 371 | 1 | equemene | } |
| 372 | 1 | equemene | /*
|
| 373 | 1 | equemene | * Process grids, mapping, (>=1) (P, Q)
|
| 374 | 1 | equemene | */
|
| 375 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 376 | 1 | equemene | (void) sscanf( line, "%s", num ); |
| 377 | 1 | equemene | *PMAPPIN = ( atoi( num ) == 1 ? HPL_COLUMN_MAJOR : HPL_ROW_MAJOR );
|
| 378 | 1 | equemene | |
| 379 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 380 | 1 | equemene | (void) sscanf( line, "%s", num ); *NPQS = atoi( num ); |
| 381 | 1 | equemene | if( ( *NPQS < 1 ) || ( *NPQS > HPL_MAX_PARAM ) ) |
| 382 | 1 | equemene | {
|
| 383 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %s %d", |
| 384 | 1 | equemene | "Number of values of grids is less",
|
| 385 | 1 | equemene | "than 1 or greater than", HPL_MAX_PARAM );
|
| 386 | 1 | equemene | error = 1; goto label_error; |
| 387 | 1 | equemene | } |
| 388 | 1 | equemene | |
| 389 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 390 | 1 | equemene | for( i = 0; i < *NPQS; i++ ) |
| 391 | 1 | equemene | {
|
| 392 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 393 | 1 | equemene | if( ( P[ i ] = atoi( num ) ) < 1 ) |
| 394 | 1 | equemene | {
|
| 395 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 396 | 1 | equemene | "Value of P less than 1" );
|
| 397 | 1 | equemene | error = 1; goto label_error; |
| 398 | 1 | equemene | } |
| 399 | 1 | equemene | } |
| 400 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 401 | 1 | equemene | for( i = 0; i < *NPQS; i++ ) |
| 402 | 1 | equemene | {
|
| 403 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 404 | 1 | equemene | if( ( Q[ i ] = atoi( num ) ) < 1 ) |
| 405 | 1 | equemene | {
|
| 406 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 407 | 1 | equemene | "Value of Q less than 1" );
|
| 408 | 1 | equemene | error = 1; goto label_error; |
| 409 | 1 | equemene | } |
| 410 | 1 | equemene | } |
| 411 | 1 | equemene | /*
|
| 412 | 1 | equemene | * Check for enough processes in machine configuration
|
| 413 | 1 | equemene | */
|
| 414 | 1 | equemene | maxp = 0;
|
| 415 | 1 | equemene | for( i = 0; i < *NPQS; i++ ) |
| 416 | 1 | equemene | { nprocs = P[i] * Q[i]; maxp = Mmax( maxp, nprocs ); }
|
| 417 | 1 | equemene | if( maxp > size )
|
| 418 | 1 | equemene | {
|
| 419 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 420 | 1 | equemene | "Need at least %d processes for these tests", maxp );
|
| 421 | 1 | equemene | error = 1; goto label_error; |
| 422 | 1 | equemene | } |
| 423 | 1 | equemene | /*
|
| 424 | 1 | equemene | * Checking threshold value (TEST->thrsh)
|
| 425 | 1 | equemene | */
|
| 426 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 427 | 1 | equemene | (void) sscanf( line, "%s", num ); TEST->thrsh = atof( num ); |
| 428 | 1 | equemene | /*
|
| 429 | 1 | equemene | * Panel factorization algorithm (PF)
|
| 430 | 1 | equemene | */
|
| 431 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 432 | 1 | equemene | (void) sscanf( line, "%s", num ); *NPFS = atoi( num ); |
| 433 | 1 | equemene | if( ( *NPFS < 1 ) || ( *NPFS > HPL_MAX_PARAM ) ) |
| 434 | 1 | equemene | {
|
| 435 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %s %d", |
| 436 | 1 | equemene | "number of values of PFACT",
|
| 437 | 1 | equemene | "is less than 1 or greater than", HPL_MAX_PARAM );
|
| 438 | 1 | equemene | error = 1; goto label_error; |
| 439 | 1 | equemene | } |
| 440 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 441 | 1 | equemene | for( i = 0; i < *NPFS; i++ ) |
| 442 | 1 | equemene | {
|
| 443 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 444 | 1 | equemene | j = atoi( num ); |
| 445 | 1 | equemene | if( j == 0 ) PF[ i ] = HPL_LEFT_LOOKING; |
| 446 | 1 | equemene | else if( j == 1 ) PF[ i ] = HPL_CROUT; |
| 447 | 1 | equemene | else if( j == 2 ) PF[ i ] = HPL_RIGHT_LOOKING; |
| 448 | 1 | equemene | else PF[ i ] = HPL_RIGHT_LOOKING;
|
| 449 | 1 | equemene | } |
| 450 | 1 | equemene | /*
|
| 451 | 1 | equemene | * Recursive stopping criterium (>=1) (NBM)
|
| 452 | 1 | equemene | */
|
| 453 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 454 | 1 | equemene | (void) sscanf( line, "%s", num ); *NBMS = atoi( num ); |
| 455 | 1 | equemene | if( ( *NBMS < 1 ) || ( *NBMS > HPL_MAX_PARAM ) ) |
| 456 | 1 | equemene | {
|
| 457 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %s %d", |
| 458 | 1 | equemene | "Number of values of NBMIN",
|
| 459 | 1 | equemene | "is less than 1 or greater than", HPL_MAX_PARAM );
|
| 460 | 1 | equemene | error = 1; goto label_error; |
| 461 | 1 | equemene | } |
| 462 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 463 | 1 | equemene | for( i = 0; i < *NBMS; i++ ) |
| 464 | 1 | equemene | {
|
| 465 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 466 | 1 | equemene | if( ( NBM[ i ] = atoi( num ) ) < 1 ) |
| 467 | 1 | equemene | {
|
| 468 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 469 | 1 | equemene | "Value of NBMIN less than 1" );
|
| 470 | 1 | equemene | error = 1; goto label_error; |
| 471 | 1 | equemene | } |
| 472 | 1 | equemene | } |
| 473 | 1 | equemene | /*
|
| 474 | 1 | equemene | * Number of panels in recursion (>=2) (NDV)
|
| 475 | 1 | equemene | */
|
| 476 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 477 | 1 | equemene | (void) sscanf( line, "%s", num ); *NDVS = atoi( num ); |
| 478 | 1 | equemene | if( ( *NDVS < 1 ) || ( *NDVS > HPL_MAX_PARAM ) ) |
| 479 | 1 | equemene | {
|
| 480 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %s %d", |
| 481 | 1 | equemene | "Number of values of NDIV",
|
| 482 | 1 | equemene | "is less than 1 or greater than", HPL_MAX_PARAM );
|
| 483 | 1 | equemene | error = 1; goto label_error; |
| 484 | 1 | equemene | } |
| 485 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 486 | 1 | equemene | for( i = 0; i < *NDVS; i++ ) |
| 487 | 1 | equemene | {
|
| 488 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 489 | 1 | equemene | if( ( NDV[ i ] = atoi( num ) ) < 2 ) |
| 490 | 1 | equemene | {
|
| 491 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 492 | 1 | equemene | "Value of NDIV less than 2" );
|
| 493 | 1 | equemene | error = 1; goto label_error; |
| 494 | 1 | equemene | } |
| 495 | 1 | equemene | } |
| 496 | 1 | equemene | /*
|
| 497 | 1 | equemene | * Recursive panel factorization (RF)
|
| 498 | 1 | equemene | */
|
| 499 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 500 | 1 | equemene | (void) sscanf( line, "%s", num ); *NRFS = atoi( num ); |
| 501 | 1 | equemene | if( ( *NRFS < 1 ) || ( *NRFS > HPL_MAX_PARAM ) ) |
| 502 | 1 | equemene | {
|
| 503 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %s %d", |
| 504 | 1 | equemene | "Number of values of RFACT",
|
| 505 | 1 | equemene | "is less than 1 or greater than", HPL_MAX_PARAM );
|
| 506 | 1 | equemene | error = 1; goto label_error; |
| 507 | 1 | equemene | } |
| 508 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 509 | 1 | equemene | for( i = 0; i < *NRFS; i++ ) |
| 510 | 1 | equemene | {
|
| 511 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 512 | 1 | equemene | j = atoi( num ); |
| 513 | 1 | equemene | if( j == 0 ) RF[ i ] = HPL_LEFT_LOOKING; |
| 514 | 1 | equemene | else if( j == 1 ) RF[ i ] = HPL_CROUT; |
| 515 | 1 | equemene | else if( j == 2 ) RF[ i ] = HPL_RIGHT_LOOKING; |
| 516 | 1 | equemene | else RF[ i ] = HPL_RIGHT_LOOKING;
|
| 517 | 1 | equemene | } |
| 518 | 1 | equemene | /*
|
| 519 | 1 | equemene | * Broadcast topology (TP) (0=rg, 1=2rg, 2=rgM, 3=2rgM, 4=L)
|
| 520 | 1 | equemene | */
|
| 521 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 522 | 1 | equemene | (void) sscanf( line, "%s", num ); *NTPS = atoi( num ); |
| 523 | 1 | equemene | if( ( *NTPS < 1 ) || ( *NTPS > HPL_MAX_PARAM ) ) |
| 524 | 1 | equemene | {
|
| 525 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %s %d", |
| 526 | 1 | equemene | "Number of values of BCAST",
|
| 527 | 1 | equemene | "is less than 1 or greater than", HPL_MAX_PARAM );
|
| 528 | 1 | equemene | error = 1; goto label_error; |
| 529 | 1 | equemene | } |
| 530 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 531 | 1 | equemene | for( i = 0; i < *NTPS; i++ ) |
| 532 | 1 | equemene | {
|
| 533 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); lineptr += strlen( num ) + 1; |
| 534 | 1 | equemene | j = atoi( num ); |
| 535 | 1 | equemene | if( j == 0 ) TP[ i ] = HPL_1RING; |
| 536 | 1 | equemene | else if( j == 1 ) TP[ i ] = HPL_1RING_M; |
| 537 | 1 | equemene | else if( j == 2 ) TP[ i ] = HPL_2RING; |
| 538 | 1 | equemene | else if( j == 3 ) TP[ i ] = HPL_2RING_M; |
| 539 | 1 | equemene | else if( j == 4 ) TP[ i ] = HPL_BLONG; |
| 540 | 1 | equemene | else if( j == 5 ) TP[ i ] = HPL_BLONG_M; |
| 541 | 1 | equemene | else TP[ i ] = HPL_1RING_M;
|
| 542 | 1 | equemene | } |
| 543 | 1 | equemene | /*
|
| 544 | 1 | equemene | * Lookahead depth (>=0) (NDH)
|
| 545 | 1 | equemene | */
|
| 546 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 547 | 1 | equemene | (void) sscanf( line, "%s", num ); *NDHS = atoi( num ); |
| 548 | 1 | equemene | if( ( *NDHS < 1 ) || ( *NDHS > HPL_MAX_PARAM ) ) |
| 549 | 1 | equemene | {
|
| 550 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo", "%s %s %d", |
| 551 | 1 | equemene | "Number of values of DEPTH",
|
| 552 | 1 | equemene | "is less than 1 or greater than", HPL_MAX_PARAM );
|
| 553 | 1 | equemene | error = 1; goto label_error; |
| 554 | 1 | equemene | } |
| 555 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); lineptr = line; |
| 556 | 1 | equemene | for( i = 0; i < *NDHS; i++ ) |
| 557 | 1 | equemene | {
|
| 558 | 1 | equemene | (void) sscanf( lineptr, "%s", num ); |
| 559 | 1 | equemene | lineptr += strlen( num ) + 1;
|
| 560 | 1 | equemene | if( ( DH[ i ] = atoi( num ) ) < 0 ) |
| 561 | 1 | equemene | {
|
| 562 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 563 | 1 | equemene | "Value of DEPTH less than 0" );
|
| 564 | 1 | equemene | error = 1; goto label_error; |
| 565 | 1 | equemene | } |
| 566 | 1 | equemene | } |
| 567 | 1 | equemene | /*
|
| 568 | 1 | equemene | * Swapping algorithm (0,1 or 2) (FSWAP)
|
| 569 | 1 | equemene | */
|
| 570 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 571 | 1 | equemene | (void) sscanf( line, "%s", num ); j = atoi( num ); |
| 572 | 1 | equemene | if( j == 0 ) *FSWAP = HPL_SWAP00; |
| 573 | 1 | equemene | else if( j == 1 ) *FSWAP = HPL_SWAP01; |
| 574 | 1 | equemene | else if( j == 2 ) *FSWAP = HPL_SW_MIX; |
| 575 | 1 | equemene | else *FSWAP = HPL_SWAP01;
|
| 576 | 1 | equemene | /*
|
| 577 | 1 | equemene | * Swapping threshold (>=0) (TSWAP)
|
| 578 | 1 | equemene | */
|
| 579 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 580 | 1 | equemene | (void) sscanf( line, "%s", num ); *TSWAP = atoi( num ); |
| 581 | 1 | equemene | if( *TSWAP <= 0 ) *TSWAP = 0; |
| 582 | 1 | equemene | /*
|
| 583 | 1 | equemene | * L1 in (no-)transposed form (0 or 1)
|
| 584 | 1 | equemene | */
|
| 585 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 586 | 1 | equemene | (void) sscanf( line, "%s", num ); *L1NOTRAN = atoi( num ); |
| 587 | 1 | equemene | if( ( *L1NOTRAN != 0 ) && ( *L1NOTRAN != 1 ) ) *L1NOTRAN = 0; |
| 588 | 1 | equemene | /*
|
| 589 | 1 | equemene | * U in (no-)transposed form (0 or 1)
|
| 590 | 1 | equemene | */
|
| 591 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 592 | 1 | equemene | (void) sscanf( line, "%s", num ); *UNOTRAN = atoi( num ); |
| 593 | 1 | equemene | if( ( *UNOTRAN != 0 ) && ( *UNOTRAN != 1 ) ) *UNOTRAN = 0; |
| 594 | 1 | equemene | /*
|
| 595 | 1 | equemene | * Equilibration (0=no, 1=yes)
|
| 596 | 1 | equemene | */
|
| 597 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 598 | 1 | equemene | (void) sscanf( line, "%s", num ); *EQUIL = atoi( num ); |
| 599 | 1 | equemene | if( ( *EQUIL != 0 ) && ( *EQUIL != 1 ) ) *EQUIL = 1; |
| 600 | 1 | equemene | /*
|
| 601 | 1 | equemene | * Memory alignment in bytes (> 0) (ALIGN)
|
| 602 | 1 | equemene | */
|
| 603 | 1 | equemene | (void) fgets( line, HPL_LINE_MAX - 2, infp ); |
| 604 | 1 | equemene | (void) sscanf( line, "%s", num ); *ALIGN = atoi( num ); |
| 605 | 1 | equemene | if( *ALIGN <= 0 ) *ALIGN = 4; |
| 606 | 1 | equemene | /*
|
| 607 | 1 | equemene | * Close input file
|
| 608 | 1 | equemene | */
|
| 609 | 1 | equemene | label_error:
|
| 610 | 1 | equemene | (void) fclose( infp );
|
| 611 | 1 | equemene | } |
| 612 | 1 | equemene | else { TEST->outfp = NULL; } |
| 613 | 1 | equemene | /*
|
| 614 | 1 | equemene | * Check for error on reading input file
|
| 615 | 1 | equemene | */
|
| 616 | 1 | equemene | (void) HPL_all_reduce( (void *)(&error), 1, HPL_INT, HPL_max, |
| 617 | 1 | equemene | MPI_COMM_WORLD ); |
| 618 | 1 | equemene | if( error )
|
| 619 | 1 | equemene | {
|
| 620 | 1 | equemene | if( rank == 0 ) |
| 621 | 1 | equemene | HPL_pwarn( stderr, __LINE__, "HPL_pdinfo",
|
| 622 | 1 | equemene | "Illegal input in file HPL.dat. Exiting ..." );
|
| 623 | 1 | equemene | MPI_Finalize(); |
| 624 | 1 | equemene | #ifdef HPL_CALL_VSIPL
|
| 625 | 1 | equemene | (void) vsip_finalize( NULL ); |
| 626 | 1 | equemene | #endif
|
| 627 | 1 | equemene | exit( 1 );
|
| 628 | 1 | equemene | } |
| 629 | 1 | equemene | /*
|
| 630 | 1 | equemene | * Compute and broadcast machine epsilon
|
| 631 | 1 | equemene | */
|
| 632 | 1 | equemene | TEST->epsil = HPL_pdlamch( MPI_COMM_WORLD, HPL_MACH_EPS ); |
| 633 | 1 | equemene | /*
|
| 634 | 1 | equemene | * Pack information arrays and broadcast
|
| 635 | 1 | equemene | */
|
| 636 | 1 | equemene | (void) HPL_broadcast( (void *)(&(TEST->thrsh)), 1, HPL_DOUBLE, 0, |
| 637 | 1 | equemene | MPI_COMM_WORLD ); |
| 638 | 1 | equemene | /*
|
| 639 | 1 | equemene | * Broadcast array sizes
|
| 640 | 1 | equemene | */
|
| 641 | 1 | equemene | iwork = (int *)malloc( (size_t)(15) * sizeof( int ) ); |
| 642 | 1 | equemene | if( rank == 0 ) |
| 643 | 1 | equemene | {
|
| 644 | 1 | equemene | iwork[ 0] = *NS; iwork[ 1] = *NBS; |
| 645 | 1 | equemene | iwork[ 2] = ( *PMAPPIN == HPL_ROW_MAJOR ? 0 : 1 ); |
| 646 | 1 | equemene | iwork[ 3] = *NPQS; iwork[ 4] = *NPFS; iwork[ 5] = *NBMS; |
| 647 | 1 | equemene | iwork[ 6] = *NDVS; iwork[ 7] = *NRFS; iwork[ 8] = *NTPS; |
| 648 | 1 | equemene | iwork[ 9] = *NDHS; iwork[10] = *TSWAP; iwork[11] = *L1NOTRAN; |
| 649 | 1 | equemene | iwork[12] = *UNOTRAN; iwork[13] = *EQUIL; iwork[14] = *ALIGN; |
| 650 | 1 | equemene | } |
| 651 | 1 | equemene | (void) HPL_broadcast( (void *)iwork, 15, HPL_INT, 0, MPI_COMM_WORLD ); |
| 652 | 1 | equemene | if( rank != 0 ) |
| 653 | 1 | equemene | {
|
| 654 | 1 | equemene | *NS = iwork[ 0]; *NBS = iwork[ 1]; |
| 655 | 1 | equemene | *PMAPPIN = ( iwork[ 2] == 0 ? HPL_ROW_MAJOR : HPL_COLUMN_MAJOR ); |
| 656 | 1 | equemene | *NPQS = iwork[ 3]; *NPFS = iwork[ 4]; *NBMS = iwork[ 5]; |
| 657 | 1 | equemene | *NDVS = iwork[ 6]; *NRFS = iwork[ 7]; *NTPS = iwork[ 8]; |
| 658 | 1 | equemene | *NDHS = iwork[ 9]; *TSWAP = iwork[10]; *L1NOTRAN = iwork[11]; |
| 659 | 1 | equemene | *UNOTRAN = iwork[12]; *EQUIL = iwork[13]; *ALIGN = iwork[14]; |
| 660 | 1 | equemene | } |
| 661 | 1 | equemene | if( iwork ) free( iwork );
|
| 662 | 1 | equemene | /*
|
| 663 | 1 | equemene | * Pack information arrays and broadcast
|
| 664 | 1 | equemene | */
|
| 665 | 1 | equemene | lwork = (*NS) + (*NBS) + 2 * (*NPQS) + (*NPFS) + (*NBMS) +
|
| 666 | 1 | equemene | (*NDVS) + (*NRFS) + (*NTPS) + (*NDHS) + 1;
|
| 667 | 1 | equemene | iwork = (int *)malloc( (size_t)(lwork) * sizeof( int ) ); |
| 668 | 1 | equemene | if( rank == 0 ) |
| 669 | 1 | equemene | {
|
| 670 | 1 | equemene | j = 0;
|
| 671 | 1 | equemene | for( i = 0; i < *NS; i++ ) { iwork[j] = N [i]; j++; } |
| 672 | 1 | equemene | for( i = 0; i < *NBS; i++ ) { iwork[j] = NB[i]; j++; } |
| 673 | 1 | equemene | for( i = 0; i < *NPQS; i++ ) { iwork[j] = P [i]; j++; } |
| 674 | 1 | equemene | for( i = 0; i < *NPQS; i++ ) { iwork[j] = Q [i]; j++; } |
| 675 | 1 | equemene | for( i = 0; i < *NPFS; i++ ) |
| 676 | 1 | equemene | {
|
| 677 | 1 | equemene | if( PF[i] == HPL_LEFT_LOOKING ) iwork[j] = 0; |
| 678 | 1 | equemene | else if( PF[i] == HPL_CROUT ) iwork[j] = 1; |
| 679 | 1 | equemene | else if( PF[i] == HPL_RIGHT_LOOKING ) iwork[j] = 2; |
| 680 | 1 | equemene | j++; |
| 681 | 1 | equemene | } |
| 682 | 1 | equemene | for( i = 0; i < *NBMS; i++ ) { iwork[j] = NBM[i]; j++; } |
| 683 | 1 | equemene | for( i = 0; i < *NDVS; i++ ) { iwork[j] = NDV[i]; j++; } |
| 684 | 1 | equemene | for( i = 0; i < *NRFS; i++ ) |
| 685 | 1 | equemene | {
|
| 686 | 1 | equemene | if( RF[i] == HPL_LEFT_LOOKING ) iwork[j] = 0; |
| 687 | 1 | equemene | else if( RF[i] == HPL_CROUT ) iwork[j] = 1; |
| 688 | 1 | equemene | else if( RF[i] == HPL_RIGHT_LOOKING ) iwork[j] = 2; |
| 689 | 1 | equemene | j++; |
| 690 | 1 | equemene | } |
| 691 | 1 | equemene | for( i = 0; i < *NTPS; i++ ) |
| 692 | 1 | equemene | {
|
| 693 | 1 | equemene | if( TP[i] == HPL_1RING ) iwork[j] = 0; |
| 694 | 1 | equemene | else if( TP[i] == HPL_1RING_M ) iwork[j] = 1; |
| 695 | 1 | equemene | else if( TP[i] == HPL_2RING ) iwork[j] = 2; |
| 696 | 1 | equemene | else if( TP[i] == HPL_2RING_M ) iwork[j] = 3; |
| 697 | 1 | equemene | else if( TP[i] == HPL_BLONG ) iwork[j] = 4; |
| 698 | 1 | equemene | else if( TP[i] == HPL_BLONG_M ) iwork[j] = 5; |
| 699 | 1 | equemene | j++; |
| 700 | 1 | equemene | } |
| 701 | 1 | equemene | for( i = 0; i < *NDHS; i++ ) { iwork[j] = DH[i]; j++; } |
| 702 | 1 | equemene | |
| 703 | 1 | equemene | if( *FSWAP == HPL_SWAP00 ) iwork[j] = 0; |
| 704 | 1 | equemene | else if( *FSWAP == HPL_SWAP01 ) iwork[j] = 1; |
| 705 | 1 | equemene | else if( *FSWAP == HPL_SW_MIX ) iwork[j] = 2; |
| 706 | 1 | equemene | j++; |
| 707 | 1 | equemene | } |
| 708 | 1 | equemene | (void) HPL_broadcast( (void*)iwork, lwork, HPL_INT, 0, |
| 709 | 1 | equemene | MPI_COMM_WORLD ); |
| 710 | 1 | equemene | if( rank != 0 ) |
| 711 | 1 | equemene | {
|
| 712 | 1 | equemene | j = 0;
|
| 713 | 1 | equemene | for( i = 0; i < *NS; i++ ) { N [i] = iwork[j]; j++; } |
| 714 | 1 | equemene | for( i = 0; i < *NBS; i++ ) { NB[i] = iwork[j]; j++; } |
| 715 | 1 | equemene | for( i = 0; i < *NPQS; i++ ) { P [i] = iwork[j]; j++; } |
| 716 | 1 | equemene | for( i = 0; i < *NPQS; i++ ) { Q [i] = iwork[j]; j++; } |
| 717 | 1 | equemene | |
| 718 | 1 | equemene | for( i = 0; i < *NPFS; i++ ) |
| 719 | 1 | equemene | {
|
| 720 | 1 | equemene | if( iwork[j] == 0 ) PF[i] = HPL_LEFT_LOOKING; |
| 721 | 1 | equemene | else if( iwork[j] == 1 ) PF[i] = HPL_CROUT; |
| 722 | 1 | equemene | else if( iwork[j] == 2 ) PF[i] = HPL_RIGHT_LOOKING; |
| 723 | 1 | equemene | j++; |
| 724 | 1 | equemene | } |
| 725 | 1 | equemene | for( i = 0; i < *NBMS; i++ ) { NBM[i] = iwork[j]; j++; } |
| 726 | 1 | equemene | for( i = 0; i < *NDVS; i++ ) { NDV[i] = iwork[j]; j++; } |
| 727 | 1 | equemene | for( i = 0; i < *NRFS; i++ ) |
| 728 | 1 | equemene | {
|
| 729 | 1 | equemene | if( iwork[j] == 0 ) RF[i] = HPL_LEFT_LOOKING; |
| 730 | 1 | equemene | else if( iwork[j] == 1 ) RF[i] = HPL_CROUT; |
| 731 | 1 | equemene | else if( iwork[j] == 2 ) RF[i] = HPL_RIGHT_LOOKING; |
| 732 | 1 | equemene | j++; |
| 733 | 1 | equemene | } |
| 734 | 1 | equemene | for( i = 0; i < *NTPS; i++ ) |
| 735 | 1 | equemene | {
|
| 736 | 1 | equemene | if( iwork[j] == 0 ) TP[i] = HPL_1RING; |
| 737 | 1 | equemene | else if( iwork[j] == 1 ) TP[i] = HPL_1RING_M; |
| 738 | 1 | equemene | else if( iwork[j] == 2 ) TP[i] = HPL_2RING; |
| 739 | 1 | equemene | else if( iwork[j] == 3 ) TP[i] = HPL_2RING_M; |
| 740 | 1 | equemene | else if( iwork[j] == 4 ) TP[i] = HPL_BLONG; |
| 741 | 1 | equemene | else if( iwork[j] == 5 ) TP[i] = HPL_BLONG_M; |
| 742 | 1 | equemene | j++; |
| 743 | 1 | equemene | } |
| 744 | 1 | equemene | for( i = 0; i < *NDHS; i++ ) { DH[i] = iwork[j]; j++; } |
| 745 | 1 | equemene | |
| 746 | 1 | equemene | if( iwork[j] == 0 ) *FSWAP = HPL_SWAP00; |
| 747 | 1 | equemene | else if( iwork[j] == 1 ) *FSWAP = HPL_SWAP01; |
| 748 | 1 | equemene | else if( iwork[j] == 2 ) *FSWAP = HPL_SW_MIX; |
| 749 | 1 | equemene | j++; |
| 750 | 1 | equemene | } |
| 751 | 1 | equemene | if( iwork ) free( iwork );
|
| 752 | 1 | equemene | /*
|
| 753 | 1 | equemene | * regurgitate input
|
| 754 | 1 | equemene | */
|
| 755 | 1 | equemene | if( rank == 0 ) |
| 756 | 1 | equemene | {
|
| 757 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s%s\n",
|
| 758 | 1 | equemene | "========================================",
|
| 759 | 1 | equemene | "========================================" );
|
| 760 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s%s\n",
|
| 761 | 1 | equemene | "HPLinpack 2.0 -- High-Performance Linpack benchmark -- ",
|
| 762 | 1 | equemene | " September 10, 2008" );
|
| 763 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s%s\n",
|
| 764 | 1 | equemene | "Written by A. Petitet and R. Clint Whaley, ",
|
| 765 | 1 | equemene | "Innovative Computing Laboratory, UTK" );
|
| 766 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s%s\n",
|
| 767 | 1 | equemene | "Modified by Piotr Luszczek, ",
|
| 768 | 1 | equemene | "Innovative Computing Laboratory, UTK" );
|
| 769 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s%s\n",
|
| 770 | 1 | equemene | "Modified by Julien Langou, ",
|
| 771 | 1 | equemene | "University of Colorado Denver");
|
| 772 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s%s\n",
|
| 773 | 1 | equemene | "========================================",
|
| 774 | 1 | equemene | "========================================" );
|
| 775 | 1 | equemene | |
| 776 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n%s\n",
|
| 777 | 1 | equemene | "An explanation of the input/output parameters follows:" );
|
| 778 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 779 | 1 | equemene | "T/V : Wall time / encoded variant." );
|
| 780 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 781 | 1 | equemene | "N : The order of the coefficient matrix A." );
|
| 782 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 783 | 1 | equemene | "NB : The partitioning blocking factor." );
|
| 784 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 785 | 1 | equemene | "P : The number of process rows." );
|
| 786 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 787 | 1 | equemene | "Q : The number of process columns." );
|
| 788 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 789 | 1 | equemene | "Time : Time in seconds to solve the linear system." );
|
| 790 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n\n",
|
| 791 | 1 | equemene | "Gflops : Rate of execution for solving the linear system." );
|
| 792 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 793 | 1 | equemene | "The following parameter values will be used:" );
|
| 794 | 1 | equemene | /*
|
| 795 | 1 | equemene | * Problem size
|
| 796 | 1 | equemene | */
|
| 797 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nN :" );
|
| 798 | 1 | equemene | for( i = 0; i < Mmin( 8, *NS ); i++ ) |
| 799 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", N[i] );
|
| 800 | 1 | equemene | if( *NS > 8 ) |
| 801 | 1 | equemene | {
|
| 802 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 803 | 1 | equemene | for( i = 8; i < Mmin( 16, *NS ); i++ ) |
| 804 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", N[i] );
|
| 805 | 1 | equemene | if( *NS > 16 ) |
| 806 | 1 | equemene | {
|
| 807 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 808 | 1 | equemene | for( i = 16; i < *NS; i++ ) |
| 809 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", N[i] );
|
| 810 | 1 | equemene | } |
| 811 | 1 | equemene | } |
| 812 | 1 | equemene | /*
|
| 813 | 1 | equemene | * Distribution blocking factor
|
| 814 | 1 | equemene | */
|
| 815 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nNB :" );
|
| 816 | 1 | equemene | for( i = 0; i < Mmin( 8, *NBS ); i++ ) |
| 817 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NB[i] );
|
| 818 | 1 | equemene | if( *NBS > 8 ) |
| 819 | 1 | equemene | {
|
| 820 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 821 | 1 | equemene | for( i = 8; i < Mmin( 16, *NBS ); i++ ) |
| 822 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NB[i] );
|
| 823 | 1 | equemene | if( *NBS > 16 ) |
| 824 | 1 | equemene | {
|
| 825 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 826 | 1 | equemene | for( i = 16; i < *NBS; i++ ) |
| 827 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NB[i] );
|
| 828 | 1 | equemene | } |
| 829 | 1 | equemene | } |
| 830 | 1 | equemene | /*
|
| 831 | 1 | equemene | * Process mapping
|
| 832 | 1 | equemene | */
|
| 833 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nPMAP :" );
|
| 834 | 1 | equemene | if( *PMAPPIN == HPL_ROW_MAJOR )
|
| 835 | 1 | equemene | HPL_fprintf( TEST->outfp, " Row-major process mapping" );
|
| 836 | 1 | equemene | else if( *PMAPPIN == HPL_COLUMN_MAJOR ) |
| 837 | 1 | equemene | HPL_fprintf( TEST->outfp, " Column-major process mapping" );
|
| 838 | 1 | equemene | /*
|
| 839 | 1 | equemene | * Process grid
|
| 840 | 1 | equemene | */
|
| 841 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nP :" );
|
| 842 | 1 | equemene | for( i = 0; i < Mmin( 8, *NPQS ); i++ ) |
| 843 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", P[i] );
|
| 844 | 1 | equemene | if( *NPQS > 8 ) |
| 845 | 1 | equemene | {
|
| 846 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 847 | 1 | equemene | for( i = 8; i < Mmin( 16, *NPQS ); i++ ) |
| 848 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", P[i] );
|
| 849 | 1 | equemene | if( *NPQS > 16 ) |
| 850 | 1 | equemene | {
|
| 851 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 852 | 1 | equemene | for( i = 16; i < *NPQS; i++ ) |
| 853 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", P[i] );
|
| 854 | 1 | equemene | } |
| 855 | 1 | equemene | } |
| 856 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nQ :" );
|
| 857 | 1 | equemene | for( i = 0; i < Mmin( 8, *NPQS ); i++ ) |
| 858 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", Q[i] );
|
| 859 | 1 | equemene | if( *NPQS > 8 ) |
| 860 | 1 | equemene | {
|
| 861 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 862 | 1 | equemene | for( i = 8; i < Mmin( 16, *NPQS ); i++ ) |
| 863 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", Q[i] );
|
| 864 | 1 | equemene | if( *NPQS > 16 ) |
| 865 | 1 | equemene | {
|
| 866 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 867 | 1 | equemene | for( i = 16; i < *NPQS; i++ ) |
| 868 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", Q[i] );
|
| 869 | 1 | equemene | } |
| 870 | 1 | equemene | } |
| 871 | 1 | equemene | /*
|
| 872 | 1 | equemene | * Panel Factorization
|
| 873 | 1 | equemene | */
|
| 874 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nPFACT :" );
|
| 875 | 1 | equemene | for( i = 0; i < Mmin( 8, *NPFS ); i++ ) |
| 876 | 1 | equemene | {
|
| 877 | 1 | equemene | if( PF[i] == HPL_LEFT_LOOKING )
|
| 878 | 1 | equemene | HPL_fprintf( TEST->outfp, " Left " );
|
| 879 | 1 | equemene | else if( PF[i] == HPL_CROUT ) |
| 880 | 1 | equemene | HPL_fprintf( TEST->outfp, " Crout " );
|
| 881 | 1 | equemene | else if( PF[i] == HPL_RIGHT_LOOKING ) |
| 882 | 1 | equemene | HPL_fprintf( TEST->outfp, " Right " );
|
| 883 | 1 | equemene | } |
| 884 | 1 | equemene | if( *NPFS > 8 ) |
| 885 | 1 | equemene | {
|
| 886 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 887 | 1 | equemene | for( i = 8; i < Mmin( 16, *NPFS ); i++ ) |
| 888 | 1 | equemene | {
|
| 889 | 1 | equemene | if( PF[i] == HPL_LEFT_LOOKING )
|
| 890 | 1 | equemene | HPL_fprintf( TEST->outfp, " Left " );
|
| 891 | 1 | equemene | else if( PF[i] == HPL_CROUT ) |
| 892 | 1 | equemene | HPL_fprintf( TEST->outfp, " Crout " );
|
| 893 | 1 | equemene | else if( PF[i] == HPL_RIGHT_LOOKING ) |
| 894 | 1 | equemene | HPL_fprintf( TEST->outfp, " Right " );
|
| 895 | 1 | equemene | } |
| 896 | 1 | equemene | if( *NPFS > 16 ) |
| 897 | 1 | equemene | {
|
| 898 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 899 | 1 | equemene | for( i = 16; i < *NPFS; i++ ) |
| 900 | 1 | equemene | {
|
| 901 | 1 | equemene | if( PF[i] == HPL_LEFT_LOOKING )
|
| 902 | 1 | equemene | HPL_fprintf( TEST->outfp, " Left " );
|
| 903 | 1 | equemene | else if( PF[i] == HPL_CROUT ) |
| 904 | 1 | equemene | HPL_fprintf( TEST->outfp, " Crout " );
|
| 905 | 1 | equemene | else if( PF[i] == HPL_RIGHT_LOOKING ) |
| 906 | 1 | equemene | HPL_fprintf( TEST->outfp, " Right " );
|
| 907 | 1 | equemene | } |
| 908 | 1 | equemene | } |
| 909 | 1 | equemene | } |
| 910 | 1 | equemene | /*
|
| 911 | 1 | equemene | * Recursive stopping criterium
|
| 912 | 1 | equemene | */
|
| 913 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nNBMIN :" );
|
| 914 | 1 | equemene | for( i = 0; i < Mmin( 8, *NBMS ); i++ ) |
| 915 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NBM[i] );
|
| 916 | 1 | equemene | if( *NBMS > 8 ) |
| 917 | 1 | equemene | {
|
| 918 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 919 | 1 | equemene | for( i = 8; i < Mmin( 16, *NBMS ); i++ ) |
| 920 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NBM[i] );
|
| 921 | 1 | equemene | if( *NBMS > 16 ) |
| 922 | 1 | equemene | {
|
| 923 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 924 | 1 | equemene | for( i = 16; i < *NBMS; i++ ) |
| 925 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NBM[i] );
|
| 926 | 1 | equemene | } |
| 927 | 1 | equemene | } |
| 928 | 1 | equemene | /*
|
| 929 | 1 | equemene | * Number of panels in recursion
|
| 930 | 1 | equemene | */
|
| 931 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nNDIV :" );
|
| 932 | 1 | equemene | for( i = 0; i < Mmin( 8, *NDVS ); i++ ) |
| 933 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NDV[i] );
|
| 934 | 1 | equemene | if( *NDVS > 8 ) |
| 935 | 1 | equemene | {
|
| 936 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 937 | 1 | equemene | for( i = 8; i < Mmin( 16, *NDVS ); i++ ) |
| 938 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NDV[i] );
|
| 939 | 1 | equemene | if( *NDVS > 16 ) |
| 940 | 1 | equemene | {
|
| 941 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 942 | 1 | equemene | for( i = 16; i < *NDVS; i++ ) |
| 943 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", NDV[i] );
|
| 944 | 1 | equemene | } |
| 945 | 1 | equemene | } |
| 946 | 1 | equemene | /*
|
| 947 | 1 | equemene | * Recursive Factorization
|
| 948 | 1 | equemene | */
|
| 949 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nRFACT :" );
|
| 950 | 1 | equemene | for( i = 0; i < Mmin( 8, *NRFS ); i++ ) |
| 951 | 1 | equemene | {
|
| 952 | 1 | equemene | if( RF[i] == HPL_LEFT_LOOKING )
|
| 953 | 1 | equemene | HPL_fprintf( TEST->outfp, " Left " );
|
| 954 | 1 | equemene | else if( RF[i] == HPL_CROUT ) |
| 955 | 1 | equemene | HPL_fprintf( TEST->outfp, " Crout " );
|
| 956 | 1 | equemene | else if( RF[i] == HPL_RIGHT_LOOKING ) |
| 957 | 1 | equemene | HPL_fprintf( TEST->outfp, " Right " );
|
| 958 | 1 | equemene | } |
| 959 | 1 | equemene | if( *NRFS > 8 ) |
| 960 | 1 | equemene | {
|
| 961 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 962 | 1 | equemene | for( i = 8; i < Mmin( 16, *NRFS ); i++ ) |
| 963 | 1 | equemene | {
|
| 964 | 1 | equemene | if( RF[i] == HPL_LEFT_LOOKING )
|
| 965 | 1 | equemene | HPL_fprintf( TEST->outfp, " Left " );
|
| 966 | 1 | equemene | else if( RF[i] == HPL_CROUT ) |
| 967 | 1 | equemene | HPL_fprintf( TEST->outfp, " Crout " );
|
| 968 | 1 | equemene | else if( RF[i] == HPL_RIGHT_LOOKING ) |
| 969 | 1 | equemene | HPL_fprintf( TEST->outfp, " Right " );
|
| 970 | 1 | equemene | } |
| 971 | 1 | equemene | if( *NRFS > 16 ) |
| 972 | 1 | equemene | {
|
| 973 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 974 | 1 | equemene | for( i = 16; i < *NRFS; i++ ) |
| 975 | 1 | equemene | {
|
| 976 | 1 | equemene | if( RF[i] == HPL_LEFT_LOOKING )
|
| 977 | 1 | equemene | HPL_fprintf( TEST->outfp, " Left " );
|
| 978 | 1 | equemene | else if( RF[i] == HPL_CROUT ) |
| 979 | 1 | equemene | HPL_fprintf( TEST->outfp, " Crout " );
|
| 980 | 1 | equemene | else if( RF[i] == HPL_RIGHT_LOOKING ) |
| 981 | 1 | equemene | HPL_fprintf( TEST->outfp, " Right " );
|
| 982 | 1 | equemene | } |
| 983 | 1 | equemene | } |
| 984 | 1 | equemene | } |
| 985 | 1 | equemene | /*
|
| 986 | 1 | equemene | * Broadcast topology
|
| 987 | 1 | equemene | */
|
| 988 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nBCAST :" );
|
| 989 | 1 | equemene | for( i = 0; i < Mmin( 8, *NTPS ); i++ ) |
| 990 | 1 | equemene | {
|
| 991 | 1 | equemene | if( TP[i] == HPL_1RING )
|
| 992 | 1 | equemene | HPL_fprintf( TEST->outfp, " 1ring " );
|
| 993 | 1 | equemene | else if( TP[i] == HPL_1RING_M ) |
| 994 | 1 | equemene | HPL_fprintf( TEST->outfp, " 1ringM " );
|
| 995 | 1 | equemene | else if( TP[i] == HPL_2RING ) |
| 996 | 1 | equemene | HPL_fprintf( TEST->outfp, " 2ring " );
|
| 997 | 1 | equemene | else if( TP[i] == HPL_2RING_M ) |
| 998 | 1 | equemene | HPL_fprintf( TEST->outfp, " 2ringM " );
|
| 999 | 1 | equemene | else if( TP[i] == HPL_BLONG ) |
| 1000 | 1 | equemene | HPL_fprintf( TEST->outfp, " Blong " );
|
| 1001 | 1 | equemene | else if( TP[i] == HPL_BLONG_M ) |
| 1002 | 1 | equemene | HPL_fprintf( TEST->outfp, " BlongM " );
|
| 1003 | 1 | equemene | } |
| 1004 | 1 | equemene | if( *NTPS > 8 ) |
| 1005 | 1 | equemene | {
|
| 1006 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 1007 | 1 | equemene | for( i = 8; i < Mmin( 16, *NTPS ); i++ ) |
| 1008 | 1 | equemene | {
|
| 1009 | 1 | equemene | if( TP[i] == HPL_1RING )
|
| 1010 | 1 | equemene | HPL_fprintf( TEST->outfp, " 1ring " );
|
| 1011 | 1 | equemene | else if( TP[i] == HPL_1RING_M ) |
| 1012 | 1 | equemene | HPL_fprintf( TEST->outfp, " 1ringM " );
|
| 1013 | 1 | equemene | else if( TP[i] == HPL_2RING ) |
| 1014 | 1 | equemene | HPL_fprintf( TEST->outfp, " 2ring " );
|
| 1015 | 1 | equemene | else if( TP[i] == HPL_2RING_M ) |
| 1016 | 1 | equemene | HPL_fprintf( TEST->outfp, " 2ringM " );
|
| 1017 | 1 | equemene | else if( TP[i] == HPL_BLONG ) |
| 1018 | 1 | equemene | HPL_fprintf( TEST->outfp, " Blong " );
|
| 1019 | 1 | equemene | else if( TP[i] == HPL_BLONG_M ) |
| 1020 | 1 | equemene | HPL_fprintf( TEST->outfp, " BlongM " );
|
| 1021 | 1 | equemene | } |
| 1022 | 1 | equemene | if( *NTPS > 16 ) |
| 1023 | 1 | equemene | {
|
| 1024 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 1025 | 1 | equemene | for( i = 16; i < *NTPS; i++ ) |
| 1026 | 1 | equemene | {
|
| 1027 | 1 | equemene | if( TP[i] == HPL_1RING )
|
| 1028 | 1 | equemene | HPL_fprintf( TEST->outfp, " 1ring " );
|
| 1029 | 1 | equemene | else if( TP[i] == HPL_1RING_M ) |
| 1030 | 1 | equemene | HPL_fprintf( TEST->outfp, " 1ringM " );
|
| 1031 | 1 | equemene | else if( TP[i] == HPL_2RING ) |
| 1032 | 1 | equemene | HPL_fprintf( TEST->outfp, " 2ring " );
|
| 1033 | 1 | equemene | else if( TP[i] == HPL_2RING_M ) |
| 1034 | 1 | equemene | HPL_fprintf( TEST->outfp, " 2ringM " );
|
| 1035 | 1 | equemene | else if( TP[i] == HPL_BLONG ) |
| 1036 | 1 | equemene | HPL_fprintf( TEST->outfp, " Blong " );
|
| 1037 | 1 | equemene | else if( TP[i] == HPL_BLONG_M ) |
| 1038 | 1 | equemene | HPL_fprintf( TEST->outfp, " BlongM " );
|
| 1039 | 1 | equemene | } |
| 1040 | 1 | equemene | } |
| 1041 | 1 | equemene | } |
| 1042 | 1 | equemene | /*
|
| 1043 | 1 | equemene | * Lookahead depths
|
| 1044 | 1 | equemene | */
|
| 1045 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nDEPTH :" );
|
| 1046 | 1 | equemene | for( i = 0; i < Mmin( 8, *NDHS ); i++ ) |
| 1047 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", DH[i] );
|
| 1048 | 1 | equemene | if( *NDHS > 8 ) |
| 1049 | 1 | equemene | {
|
| 1050 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 1051 | 1 | equemene | for( i = 8; i < Mmin( 16, *NDHS ); i++ ) |
| 1052 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", DH[i] );
|
| 1053 | 1 | equemene | if( *NDHS > 16 ) |
| 1054 | 1 | equemene | {
|
| 1055 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n " );
|
| 1056 | 1 | equemene | for( i = 16; i < *NDHS; i++ ) |
| 1057 | 1 | equemene | HPL_fprintf( TEST->outfp, "%8d ", DH[i] );
|
| 1058 | 1 | equemene | } |
| 1059 | 1 | equemene | } |
| 1060 | 1 | equemene | /*
|
| 1061 | 1 | equemene | * Swapping algorithm
|
| 1062 | 1 | equemene | */
|
| 1063 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nSWAP :" );
|
| 1064 | 1 | equemene | if( *FSWAP == HPL_SWAP00 )
|
| 1065 | 1 | equemene | HPL_fprintf( TEST->outfp, " Binary-exchange" );
|
| 1066 | 1 | equemene | else if( *FSWAP == HPL_SWAP01 ) |
| 1067 | 1 | equemene | HPL_fprintf( TEST->outfp, " Spread-roll (long)" );
|
| 1068 | 1 | equemene | else if( *FSWAP == HPL_SW_MIX ) |
| 1069 | 1 | equemene | HPL_fprintf( TEST->outfp, " Mix (threshold = %d)", *TSWAP );
|
| 1070 | 1 | equemene | /*
|
| 1071 | 1 | equemene | * L1 storage form
|
| 1072 | 1 | equemene | */
|
| 1073 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nL1 :" );
|
| 1074 | 1 | equemene | if( *L1NOTRAN != 0 ) |
| 1075 | 1 | equemene | HPL_fprintf( TEST->outfp, " no-transposed form" );
|
| 1076 | 1 | equemene | else
|
| 1077 | 1 | equemene | HPL_fprintf( TEST->outfp, " transposed form" );
|
| 1078 | 1 | equemene | /*
|
| 1079 | 1 | equemene | * U storage form
|
| 1080 | 1 | equemene | */
|
| 1081 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nU :" );
|
| 1082 | 1 | equemene | if( *UNOTRAN != 0 ) |
| 1083 | 1 | equemene | HPL_fprintf( TEST->outfp, " no-transposed form" );
|
| 1084 | 1 | equemene | else
|
| 1085 | 1 | equemene | HPL_fprintf( TEST->outfp, " transposed form" );
|
| 1086 | 1 | equemene | /*
|
| 1087 | 1 | equemene | * Equilibration
|
| 1088 | 1 | equemene | */
|
| 1089 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nEQUIL :" );
|
| 1090 | 1 | equemene | if( *EQUIL != 0 ) |
| 1091 | 1 | equemene | HPL_fprintf( TEST->outfp, " yes" );
|
| 1092 | 1 | equemene | else
|
| 1093 | 1 | equemene | HPL_fprintf( TEST->outfp, " no" );
|
| 1094 | 1 | equemene | /*
|
| 1095 | 1 | equemene | * Alignment
|
| 1096 | 1 | equemene | */
|
| 1097 | 1 | equemene | HPL_fprintf( TEST->outfp, "\nALIGN : %d double precision words",
|
| 1098 | 1 | equemene | *ALIGN ); |
| 1099 | 1 | equemene | |
| 1100 | 1 | equemene | HPL_fprintf( TEST->outfp, "\n\n" );
|
| 1101 | 1 | equemene | /*
|
| 1102 | 1 | equemene | * For testing only
|
| 1103 | 1 | equemene | */
|
| 1104 | 1 | equemene | if( TEST->thrsh > HPL_rzero )
|
| 1105 | 1 | equemene | {
|
| 1106 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s%s\n\n",
|
| 1107 | 1 | equemene | "----------------------------------------",
|
| 1108 | 1 | equemene | "----------------------------------------" );
|
| 1109 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 1110 | 1 | equemene | "- The matrix A is randomly generated for each test." );
|
| 1111 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 1112 | 1 | equemene | "- The following scaled residual check will be computed:" );
|
| 1113 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s\n",
|
| 1114 | 1 | equemene | " ||Ax-b||_oo / ( eps * ( || x ||_oo * || A ||_oo + || b ||_oo ) * N )" );
|
| 1115 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s %21.6e\n",
|
| 1116 | 1 | equemene | "- The relative machine precision (eps) is taken to be ",
|
| 1117 | 1 | equemene | TEST->epsil ); |
| 1118 | 1 | equemene | HPL_fprintf( TEST->outfp, "%s %11.1f\n\n",
|
| 1119 | 1 | equemene | "- Computational tests pass if scaled residuals are less than ",
|
| 1120 | 1 | equemene | TEST->thrsh ); |
| 1121 | 1 | equemene | } |
| 1122 | 1 | equemene | } |
| 1123 | 1 | equemene | /*
|
| 1124 | 1 | equemene | * End of HPL_pdinfo
|
| 1125 | 1 | equemene | */
|
| 1126 | 1 | equemene | } |