root / testing / pmatgen / HPL_pdmatgen.c @ 9
Historique | Voir | Annoter | Télécharger (8,41 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_pdmatgen
|
| 54 | 1 | equemene | ( |
| 55 | 1 | equemene | const HPL_T_grid * GRID,
|
| 56 | 1 | equemene | const int M, |
| 57 | 1 | equemene | const int N, |
| 58 | 1 | equemene | const int NB, |
| 59 | 1 | equemene | double * A,
|
| 60 | 1 | equemene | const int LDA, |
| 61 | 1 | equemene | const int ISEED |
| 62 | 1 | equemene | ) |
| 63 | 1 | equemene | #else
|
| 64 | 1 | equemene | void HPL_pdmatgen
|
| 65 | 1 | equemene | ( GRID, M, N, NB, A, LDA, ISEED ) |
| 66 | 1 | equemene | const HPL_T_grid * GRID;
|
| 67 | 1 | equemene | const int M; |
| 68 | 1 | equemene | const int N; |
| 69 | 1 | equemene | const int NB; |
| 70 | 1 | equemene | double * A;
|
| 71 | 1 | equemene | const int LDA; |
| 72 | 1 | equemene | const int ISEED; |
| 73 | 1 | equemene | #endif
|
| 74 | 1 | equemene | {
|
| 75 | 1 | equemene | /*
|
| 76 | 1 | equemene | * Purpose
|
| 77 | 1 | equemene | * =======
|
| 78 | 1 | equemene | *
|
| 79 | 1 | equemene | * HPL_pdmatgen generates (or regenerates) a parallel random matrix A.
|
| 80 | 1 | equemene | *
|
| 81 | 1 | equemene | * The pseudo-random generator uses the linear congruential algorithm:
|
| 82 | 1 | equemene | * X(n+1) = (a * X(n) + c) mod m as described in the Art of Computer
|
| 83 | 1 | equemene | * Programming, Knuth 1973, Vol. 2.
|
| 84 | 1 | equemene | *
|
| 85 | 1 | equemene | * Arguments
|
| 86 | 1 | equemene | * =========
|
| 87 | 1 | equemene | *
|
| 88 | 1 | equemene | * GRID (local input) const HPL_T_grid *
|
| 89 | 1 | equemene | * On entry, GRID points to the data structure containing the
|
| 90 | 1 | equemene | * process grid information.
|
| 91 | 1 | equemene | *
|
| 92 | 1 | equemene | * M (global input) const int
|
| 93 | 1 | equemene | * On entry, M specifies the number of rows of the matrix A.
|
| 94 | 1 | equemene | * M must be at least zero.
|
| 95 | 1 | equemene | *
|
| 96 | 1 | equemene | * N (global input) const int
|
| 97 | 1 | equemene | * On entry, N specifies the number of columns of the matrix A.
|
| 98 | 1 | equemene | * N must be at least zero.
|
| 99 | 1 | equemene | *
|
| 100 | 1 | equemene | * NB (global input) const int
|
| 101 | 1 | equemene | * On entry, NB specifies the blocking factor used to partition
|
| 102 | 1 | equemene | * and distribute the matrix A. NB must be larger than one.
|
| 103 | 1 | equemene | *
|
| 104 | 1 | equemene | * A (local output) double *
|
| 105 | 1 | equemene | * On entry, A points to an array of dimension (LDA,LocQ(N)).
|
| 106 | 1 | equemene | * On exit, this array contains the coefficients of the randomly
|
| 107 | 1 | equemene | * generated matrix.
|
| 108 | 1 | equemene | *
|
| 109 | 1 | equemene | * LDA (local input) const int
|
| 110 | 1 | equemene | * On entry, LDA specifies the leading dimension of the array A.
|
| 111 | 1 | equemene | * LDA must be at least max(1,LocP(M)).
|
| 112 | 1 | equemene | *
|
| 113 | 1 | equemene | * ISEED (global input) const int
|
| 114 | 1 | equemene | * On entry, ISEED specifies the seed number to generate the
|
| 115 | 1 | equemene | * matrix A. ISEED must be at least zero.
|
| 116 | 1 | equemene | *
|
| 117 | 1 | equemene | * ---------------------------------------------------------------------
|
| 118 | 1 | equemene | */
|
| 119 | 1 | equemene | /*
|
| 120 | 1 | equemene | * .. Local Variables ..
|
| 121 | 1 | equemene | */
|
| 122 | 1 | equemene | int iadd [2], ia1 [2], ia2 [2], ia3 [2], |
| 123 | 1 | equemene | ia4 [2], ia5 [2], ib1 [2], ib2 [2], |
| 124 | 1 | equemene | ib3 [2], ic1 [2], ic2 [2], ic3 [2], |
| 125 | 1 | equemene | ic4 [2], ic5 [2], iran1[2], iran2[2], |
| 126 | 1 | equemene | iran3[2], iran4[2], itmp1[2], itmp2[2], |
| 127 | 1 | equemene | itmp3[2], jseed[2], mult [2]; |
| 128 | 1 | equemene | int ib, iblk, ik, jb, jblk, jk, jump1, jump2,
|
| 129 | 1 | equemene | jump3, jump4, jump5, jump6, jump7, lmb, |
| 130 | 1 | equemene | lnb, mblks, mp, mycol, myrow, nblks, |
| 131 | 1 | equemene | npcol, nprow, nq; |
| 132 | 1 | equemene | /* ..
|
| 133 | 1 | equemene | * .. Executable Statements ..
|
| 134 | 1 | equemene | */
|
| 135 | 1 | equemene | (void) HPL_grid_info( GRID, &nprow, &npcol, &myrow, &mycol );
|
| 136 | 1 | equemene | |
| 137 | 1 | equemene | mult [0] = HPL_MULT0; mult [1] = HPL_MULT1; |
| 138 | 1 | equemene | iadd [0] = HPL_IADD0; iadd [1] = HPL_IADD1; |
| 139 | 1 | equemene | jseed[0] = ISEED; jseed[1] = 0; |
| 140 | 1 | equemene | /*
|
| 141 | 1 | equemene | * Generate an M by N matrix starting in process (0,0)
|
| 142 | 1 | equemene | */
|
| 143 | 1 | equemene | Mnumroc( mp, M, NB, NB, myrow, 0, nprow );
|
| 144 | 1 | equemene | Mnumroc( nq, N, NB, NB, mycol, 0, npcol );
|
| 145 | 1 | equemene | |
| 146 | 1 | equemene | if( ( mp <= 0 ) || ( nq <= 0 ) ) return; |
| 147 | 1 | equemene | /*
|
| 148 | 1 | equemene | * Local number of blocks and size of the last one
|
| 149 | 1 | equemene | */
|
| 150 | 1 | equemene | mblks = ( mp + NB - 1 ) / NB; lmb = mp - ( ( mp - 1 ) / NB ) * NB; |
| 151 | 1 | equemene | nblks = ( nq + NB - 1 ) / NB; lnb = nq - ( ( nq - 1 ) / NB ) * NB; |
| 152 | 1 | equemene | /*
|
| 153 | 1 | equemene | * Compute multiplier/adder for various jumps in random sequence
|
| 154 | 1 | equemene | */
|
| 155 | 1 | equemene | jump1 = 1; jump2 = nprow * NB; jump3 = M; jump4 = npcol * NB;
|
| 156 | 1 | equemene | jump5 = NB; jump6 = mycol; jump7 = myrow * NB; |
| 157 | 1 | equemene | |
| 158 | 1 | equemene | HPL_xjumpm( jump1, mult, iadd, jseed, iran1, ia1, ic1 ); |
| 159 | 1 | equemene | HPL_xjumpm( jump2, mult, iadd, iran1, itmp1, ia2, ic2 ); |
| 160 | 1 | equemene | HPL_xjumpm( jump3, mult, iadd, iran1, itmp1, ia3, ic3 ); |
| 161 | 1 | equemene | HPL_xjumpm( jump4, ia3, ic3, iran1, itmp1, ia4, ic4 ); |
| 162 | 1 | equemene | HPL_xjumpm( jump5, ia3, ic3, iran1, itmp1, ia5, ic5 ); |
| 163 | 1 | equemene | HPL_xjumpm( jump6, ia5, ic5, iran1, itmp3, itmp1, itmp2 ); |
| 164 | 1 | equemene | HPL_xjumpm( jump7, mult, iadd, itmp3, iran1, itmp1, itmp2 ); |
| 165 | 1 | equemene | HPL_setran( 0, iran1 ); HPL_setran( 1, ia1 ); HPL_setran( 2, ic1 ); |
| 166 | 1 | equemene | /*
|
| 167 | 1 | equemene | * Save value of first number in sequence
|
| 168 | 1 | equemene | */
|
| 169 | 1 | equemene | ib1[0] = iran1[0]; ib1[1] = iran1[1]; |
| 170 | 1 | equemene | ib2[0] = iran1[0]; ib2[1] = iran1[1]; |
| 171 | 1 | equemene | ib3[0] = iran1[0]; ib3[1] = iran1[1]; |
| 172 | 1 | equemene | |
| 173 | 1 | equemene | for( jblk = 0; jblk < nblks; jblk++ ) |
| 174 | 1 | equemene | {
|
| 175 | 1 | equemene | jb = ( jblk == nblks - 1 ? lnb : NB );
|
| 176 | 1 | equemene | for( jk = 0; jk < jb; jk++ ) |
| 177 | 1 | equemene | {
|
| 178 | 1 | equemene | for( iblk = 0; iblk < mblks; iblk++ ) |
| 179 | 1 | equemene | {
|
| 180 | 1 | equemene | ib = ( iblk == mblks - 1 ? lmb : NB );
|
| 181 | 1 | equemene | for( ik = 0; ik < ib; A++, ik++ ) *A = HPL_rand(); |
| 182 | 1 | equemene | HPL_jumpit( ia2, ic2, ib1, iran2 ); |
| 183 | 1 | equemene | ib1[0] = iran2[0]; ib1[1] = iran2[1]; |
| 184 | 1 | equemene | } |
| 185 | 1 | equemene | A += LDA - mp; |
| 186 | 1 | equemene | HPL_jumpit( ia3, ic3, ib2, iran3 ); |
| 187 | 1 | equemene | ib1[0] = iran3[0]; ib1[1] = iran3[1]; |
| 188 | 1 | equemene | ib2[0] = iran3[0]; ib2[1] = iran3[1]; |
| 189 | 1 | equemene | } |
| 190 | 1 | equemene | HPL_jumpit( ia4, ic4, ib3, iran4 ); |
| 191 | 1 | equemene | ib1[0] = iran4[0]; ib1[1] = iran4[1]; |
| 192 | 1 | equemene | ib2[0] = iran4[0]; ib2[1] = iran4[1]; |
| 193 | 1 | equemene | ib3[0] = iran4[0]; ib3[1] = iran4[1]; |
| 194 | 1 | equemene | } |
| 195 | 1 | equemene | /*
|
| 196 | 1 | equemene | * End of HPL_pdmatgen
|
| 197 | 1 | equemene | */
|
| 198 | 1 | equemene | } |