root / testing / ptest / HPL_pddriver.c
Historique | Voir | Annoter | Télécharger (11,84 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 | int main
|
| 54 | 1 | equemene | ( |
| 55 | 1 | equemene | int ARGC,
|
| 56 | 1 | equemene | char * * ARGV
|
| 57 | 1 | equemene | ) |
| 58 | 1 | equemene | #else
|
| 59 | 1 | equemene | int main( ARGC, ARGV )
|
| 60 | 1 | equemene | /*
|
| 61 | 1 | equemene | * .. Scalar Arguments ..
|
| 62 | 1 | equemene | */
|
| 63 | 1 | equemene | int ARGC;
|
| 64 | 1 | equemene | /*
|
| 65 | 1 | equemene | * .. Array Arguments ..
|
| 66 | 1 | equemene | */
|
| 67 | 1 | equemene | char * * ARGV;
|
| 68 | 1 | equemene | #endif
|
| 69 | 1 | equemene | {
|
| 70 | 1 | equemene | /*
|
| 71 | 1 | equemene | * Purpose
|
| 72 | 1 | equemene | * =======
|
| 73 | 1 | equemene | *
|
| 74 | 1 | equemene | * main is the main driver program for testing the HPL routines.
|
| 75 | 1 | equemene | * This program is driven by a short data file named "HPL.dat".
|
| 76 | 1 | equemene | *
|
| 77 | 1 | equemene | * ---------------------------------------------------------------------
|
| 78 | 1 | equemene | */
|
| 79 | 1 | equemene | /*
|
| 80 | 1 | equemene | * .. Local Variables ..
|
| 81 | 1 | equemene | */
|
| 82 | 1 | equemene | int nval [HPL_MAX_PARAM],
|
| 83 | 1 | equemene | nbval [HPL_MAX_PARAM], |
| 84 | 1 | equemene | pval [HPL_MAX_PARAM], |
| 85 | 1 | equemene | qval [HPL_MAX_PARAM], |
| 86 | 1 | equemene | nbmval[HPL_MAX_PARAM], |
| 87 | 1 | equemene | ndvval[HPL_MAX_PARAM], |
| 88 | 1 | equemene | ndhval[HPL_MAX_PARAM]; |
| 89 | 1 | equemene | |
| 90 | 1 | equemene | HPL_T_FACT pfaval[HPL_MAX_PARAM], |
| 91 | 1 | equemene | rfaval[HPL_MAX_PARAM]; |
| 92 | 1 | equemene | |
| 93 | 1 | equemene | HPL_T_TOP topval[HPL_MAX_PARAM]; |
| 94 | 1 | equemene | |
| 95 | 1 | equemene | HPL_T_grid grid; |
| 96 | 1 | equemene | HPL_T_palg algo; |
| 97 | 1 | equemene | HPL_T_test test; |
| 98 | 1 | equemene | int L1notran, Unotran, align, equil, in, inb,
|
| 99 | 1 | equemene | inbm, indh, indv, ipfa, ipq, irfa, itop, |
| 100 | 1 | equemene | mycol, myrow, ns, nbs, nbms, ndhs, ndvs, |
| 101 | 1 | equemene | npcol, npfs, npqs, nprow, nrfs, ntps, |
| 102 | 1 | equemene | rank, size, tswap; |
| 103 | 1 | equemene | HPL_T_ORDER pmapping; |
| 104 | 1 | equemene | HPL_T_FACT rpfa; |
| 105 | 1 | equemene | HPL_T_SWAP fswap; |
| 106 | 1 | equemene | /* ..
|
| 107 | 1 | equemene | * .. Executable Statements ..
|
| 108 | 1 | equemene | */
|
| 109 | 1 | equemene | MPI_Init( &ARGC, &ARGV ); |
| 110 | 1 | equemene | #ifdef HPL_CALL_VSIPL
|
| 111 | 1 | equemene | vsip_init((void*)0); |
| 112 | 1 | equemene | #endif
|
| 113 | 1 | equemene | MPI_Comm_rank( MPI_COMM_WORLD, &rank ); |
| 114 | 1 | equemene | MPI_Comm_size( MPI_COMM_WORLD, &size ); |
| 115 | 1 | equemene | /*
|
| 116 | 1 | equemene | * Read and check validity of test parameters from input file
|
| 117 | 1 | equemene | *
|
| 118 | 1 | equemene | * HPL Version 1.0, Linpack benchmark input file
|
| 119 | 1 | equemene | * Your message here
|
| 120 | 1 | equemene | * HPL.out output file name (if any)
|
| 121 | 1 | equemene | * 6 device out (6=stdout,7=stderr,file)
|
| 122 | 1 | equemene | * 4 # of problems sizes (N)
|
| 123 | 1 | equemene | * 29 30 34 35 Ns
|
| 124 | 1 | equemene | * 4 # of NBs
|
| 125 | 1 | equemene | * 1 2 3 4 NBs
|
| 126 | 1 | equemene | * 0 PMAP process mapping (0=Row-,1=Column-major)
|
| 127 | 1 | equemene | * 3 # of process grids (P x Q)
|
| 128 | 1 | equemene | * 2 1 4 Ps
|
| 129 | 1 | equemene | * 2 4 1 Qs
|
| 130 | 1 | equemene | * 16.0 threshold
|
| 131 | 1 | equemene | * 3 # of panel fact
|
| 132 | 1 | equemene | * 0 1 2 PFACTs (0=left, 1=Crout, 2=Right)
|
| 133 | 1 | equemene | * 2 # of recursive stopping criterium
|
| 134 | 1 | equemene | * 2 4 NBMINs (>= 1)
|
| 135 | 1 | equemene | * 1 # of panels in recursion
|
| 136 | 1 | equemene | * 2 NDIVs
|
| 137 | 1 | equemene | * 3 # of recursive panel fact.
|
| 138 | 1 | equemene | * 0 1 2 RFACTs (0=left, 1=Crout, 2=Right)
|
| 139 | 1 | equemene | * 1 # of broadcast
|
| 140 | 1 | equemene | * 0 BCASTs (0=1rg,1=1rM,2=2rg,3=2rM,4=Lng,5=LnM)
|
| 141 | 1 | equemene | * 1 # of lookahead depth
|
| 142 | 1 | equemene | * 0 DEPTHs (>=0)
|
| 143 | 1 | equemene | * 2 SWAP (0=bin-exch,1=long,2=mix)
|
| 144 | 1 | equemene | * 4 swapping threshold
|
| 145 | 1 | equemene | * 0 L1 in (0=transposed,1=no-transposed) form
|
| 146 | 1 | equemene | * 0 U in (0=transposed,1=no-transposed) form
|
| 147 | 1 | equemene | * 1 Equilibration (0=no,1=yes)
|
| 148 | 1 | equemene | * 8 memory alignment in double (> 0)
|
| 149 | 1 | equemene | */
|
| 150 | 1 | equemene | HPL_pdinfo( &test, &ns, nval, &nbs, nbval, &pmapping, &npqs, pval, qval, |
| 151 | 1 | equemene | &npfs, pfaval, &nbms, nbmval, &ndvs, ndvval, &nrfs, rfaval, |
| 152 | 1 | equemene | &ntps, topval, &ndhs, ndhval, &fswap, &tswap, &L1notran, |
| 153 | 1 | equemene | &Unotran, &equil, &align ); |
| 154 | 1 | equemene | /*
|
| 155 | 1 | equemene | * Loop over different process grids - Define process grid. Go to bottom
|
| 156 | 1 | equemene | * of process grid loop if this case does not use my process.
|
| 157 | 1 | equemene | */
|
| 158 | 1 | equemene | for( ipq = 0; ipq < npqs; ipq++ ) |
| 159 | 1 | equemene | {
|
| 160 | 1 | equemene | (void) HPL_grid_init( MPI_COMM_WORLD, pmapping, pval[ipq], qval[ipq],
|
| 161 | 1 | equemene | &grid ); |
| 162 | 1 | equemene | (void) HPL_grid_info( &grid, &nprow, &npcol, &myrow, &mycol );
|
| 163 | 1 | equemene | |
| 164 | 1 | equemene | if( ( myrow < 0 ) || ( myrow >= nprow ) || |
| 165 | 1 | equemene | ( mycol < 0 ) || ( mycol >= npcol ) ) goto label_end_of_npqs; |
| 166 | 1 | equemene | |
| 167 | 1 | equemene | for( in = 0; in < ns; in++ ) |
| 168 | 1 | equemene | { /* Loop over various problem sizes */
|
| 169 | 1 | equemene | for( inb = 0; inb < nbs; inb++ ) |
| 170 | 1 | equemene | { /* Loop over various blocking factors */
|
| 171 | 1 | equemene | for( indh = 0; indh < ndhs; indh++ ) |
| 172 | 1 | equemene | { /* Loop over various lookahead depths */
|
| 173 | 1 | equemene | for( itop = 0; itop < ntps; itop++ ) |
| 174 | 1 | equemene | { /* Loop over various broadcast topologies */
|
| 175 | 1 | equemene | for( irfa = 0; irfa < nrfs; irfa++ ) |
| 176 | 1 | equemene | { /* Loop over various recursive factorizations */
|
| 177 | 1 | equemene | for( ipfa = 0; ipfa < npfs; ipfa++ ) |
| 178 | 1 | equemene | { /* Loop over various panel factorizations */
|
| 179 | 1 | equemene | for( inbm = 0; inbm < nbms; inbm++ ) |
| 180 | 1 | equemene | { /* Loop over various recursive stopping criteria */
|
| 181 | 1 | equemene | for( indv = 0; indv < ndvs; indv++ ) |
| 182 | 1 | equemene | { /* Loop over various # of panels in recursion */
|
| 183 | 1 | equemene | /*
|
| 184 | 1 | equemene | * Set up the algorithm parameters
|
| 185 | 1 | equemene | */
|
| 186 | 1 | equemene | algo.btopo = topval[itop]; algo.depth = ndhval[indh]; |
| 187 | 1 | equemene | algo.nbmin = nbmval[inbm]; algo.nbdiv = ndvval[indv]; |
| 188 | 1 | equemene | |
| 189 | 1 | equemene | algo.pfact = rpfa = pfaval[ipfa]; |
| 190 | 1 | equemene | |
| 191 | 1 | equemene | if( L1notran != 0 ) |
| 192 | 1 | equemene | {
|
| 193 | 1 | equemene | if( rpfa == HPL_LEFT_LOOKING ) algo.pffun = HPL_pdpanllN;
|
| 194 | 1 | equemene | else if( rpfa == HPL_CROUT ) algo.pffun = HPL_pdpancrN; |
| 195 | 1 | equemene | else algo.pffun = HPL_pdpanrlN;
|
| 196 | 1 | equemene | |
| 197 | 1 | equemene | algo.rfact = rpfa = rfaval[irfa]; |
| 198 | 1 | equemene | if( rpfa == HPL_LEFT_LOOKING ) algo.rffun = HPL_pdrpanllN;
|
| 199 | 1 | equemene | else if( rpfa == HPL_CROUT ) algo.rffun = HPL_pdrpancrN; |
| 200 | 1 | equemene | else algo.rffun = HPL_pdrpanrlN;
|
| 201 | 1 | equemene | |
| 202 | 1 | equemene | if( Unotran != 0 ) algo.upfun = HPL_pdupdateNN; |
| 203 | 1 | equemene | else algo.upfun = HPL_pdupdateNT;
|
| 204 | 1 | equemene | } |
| 205 | 1 | equemene | else
|
| 206 | 1 | equemene | {
|
| 207 | 1 | equemene | if( rpfa == HPL_LEFT_LOOKING ) algo.pffun = HPL_pdpanllT;
|
| 208 | 1 | equemene | else if( rpfa == HPL_CROUT ) algo.pffun = HPL_pdpancrT; |
| 209 | 1 | equemene | else algo.pffun = HPL_pdpanrlT;
|
| 210 | 1 | equemene | |
| 211 | 1 | equemene | algo.rfact = rpfa = rfaval[irfa]; |
| 212 | 1 | equemene | if( rpfa == HPL_LEFT_LOOKING ) algo.rffun = HPL_pdrpanllT;
|
| 213 | 1 | equemene | else if( rpfa == HPL_CROUT ) algo.rffun = HPL_pdrpancrT; |
| 214 | 1 | equemene | else algo.rffun = HPL_pdrpanrlT;
|
| 215 | 1 | equemene | |
| 216 | 1 | equemene | if( Unotran != 0 ) algo.upfun = HPL_pdupdateTN; |
| 217 | 1 | equemene | else algo.upfun = HPL_pdupdateTT;
|
| 218 | 1 | equemene | } |
| 219 | 1 | equemene | |
| 220 | 1 | equemene | algo.fswap = fswap; algo.fsthr = tswap; |
| 221 | 1 | equemene | algo.equil = equil; algo.align = align; |
| 222 | 1 | equemene | |
| 223 | 1 | equemene | HPL_pdtest( &test, &grid, &algo, nval[in], nbval[inb] ); |
| 224 | 1 | equemene | |
| 225 | 1 | equemene | } |
| 226 | 1 | equemene | } |
| 227 | 1 | equemene | } |
| 228 | 1 | equemene | } |
| 229 | 1 | equemene | } |
| 230 | 1 | equemene | } |
| 231 | 1 | equemene | } |
| 232 | 1 | equemene | } |
| 233 | 1 | equemene | (void) HPL_grid_exit( &grid );
|
| 234 | 1 | equemene | label_end_of_npqs: ;
|
| 235 | 1 | equemene | } |
| 236 | 1 | equemene | /*
|
| 237 | 1 | equemene | * Print ending messages, close output file, exit.
|
| 238 | 1 | equemene | */
|
| 239 | 1 | equemene | if( rank == 0 ) |
| 240 | 1 | equemene | {
|
| 241 | 1 | equemene | test.ktest = test.kpass + test.kfail + test.kskip; |
| 242 | 1 | equemene | #ifndef HPL_DETAILED_TIMING
|
| 243 | 1 | equemene | HPL_fprintf( test.outfp, "%s%s\n",
|
| 244 | 1 | equemene | "========================================",
|
| 245 | 1 | equemene | "========================================" );
|
| 246 | 1 | equemene | #else
|
| 247 | 1 | equemene | if( test.thrsh > HPL_rzero )
|
| 248 | 1 | equemene | HPL_fprintf( test.outfp, "%s%s\n",
|
| 249 | 1 | equemene | "========================================",
|
| 250 | 1 | equemene | "========================================" );
|
| 251 | 1 | equemene | #endif
|
| 252 | 1 | equemene | |
| 253 | 1 | equemene | HPL_fprintf( test.outfp, "\n%s %6d %s\n", "Finished", test.ktest, |
| 254 | 1 | equemene | "tests with the following results:" );
|
| 255 | 1 | equemene | if( test.thrsh > HPL_rzero )
|
| 256 | 1 | equemene | {
|
| 257 | 1 | equemene | HPL_fprintf( test.outfp, " %6d %s\n", test.kpass,
|
| 258 | 1 | equemene | "tests completed and passed residual checks," );
|
| 259 | 1 | equemene | HPL_fprintf( test.outfp, " %6d %s\n", test.kfail,
|
| 260 | 1 | equemene | "tests completed and failed residual checks," );
|
| 261 | 1 | equemene | HPL_fprintf( test.outfp, " %6d %s\n", test.kskip,
|
| 262 | 1 | equemene | "tests skipped because of illegal input values." );
|
| 263 | 1 | equemene | } |
| 264 | 1 | equemene | else
|
| 265 | 1 | equemene | {
|
| 266 | 1 | equemene | HPL_fprintf( test.outfp, " %6d %s\n", test.kpass,
|
| 267 | 1 | equemene | "tests completed without checking," );
|
| 268 | 1 | equemene | HPL_fprintf( test.outfp, " %6d %s\n", test.kskip,
|
| 269 | 1 | equemene | "tests skipped because of illegal input values." );
|
| 270 | 1 | equemene | } |
| 271 | 1 | equemene | |
| 272 | 1 | equemene | HPL_fprintf( test.outfp, "%s%s\n",
|
| 273 | 1 | equemene | "----------------------------------------",
|
| 274 | 1 | equemene | "----------------------------------------" );
|
| 275 | 1 | equemene | HPL_fprintf( test.outfp, "\nEnd of Tests.\n" );
|
| 276 | 1 | equemene | HPL_fprintf( test.outfp, "%s%s\n",
|
| 277 | 1 | equemene | "========================================",
|
| 278 | 1 | equemene | "========================================" );
|
| 279 | 1 | equemene | |
| 280 | 1 | equemene | if( ( test.outfp != stdout ) && ( test.outfp != stderr ) )
|
| 281 | 1 | equemene | (void) fclose( test.outfp );
|
| 282 | 1 | equemene | } |
| 283 | 1 | equemene | #ifdef HPL_CALL_VSIPL
|
| 284 | 1 | equemene | vsip_finalize((void*)0); |
| 285 | 1 | equemene | #endif
|
| 286 | 1 | equemene | MPI_Finalize(); |
| 287 | 1 | equemene | exit( 0 );
|
| 288 | 1 | equemene | |
| 289 | 1 | equemene | return( 0 ); |
| 290 | 1 | equemene | /*
|
| 291 | 1 | equemene | * End of main
|
| 292 | 1 | equemene | */
|
| 293 | 1 | equemene | } |