root / src / pauxil / HPL_pdlaprnt.c @ 1
Historique | Voir | Annoter | Télécharger (7,99 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_pdlaprnt
|
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 IAROW, |
62 | 1 | equemene | const int IACOL, |
63 | 1 | equemene | const char * CMATNM |
64 | 1 | equemene | ) |
65 | 1 | equemene | #else
|
66 | 1 | equemene | void HPL_pdlaprnt
|
67 | 1 | equemene | ( GRID, M, N, NB, A, LDA, IAROW, IACOL, CMATNM ) |
68 | 1 | equemene | const HPL_T_grid * GRID;
|
69 | 1 | equemene | const int M; |
70 | 1 | equemene | const int N; |
71 | 1 | equemene | const int NB; |
72 | 1 | equemene | double * A;
|
73 | 1 | equemene | const int LDA; |
74 | 1 | equemene | const int IAROW; |
75 | 1 | equemene | const int IACOL; |
76 | 1 | equemene | const char * CMATNM; |
77 | 1 | equemene | #endif
|
78 | 1 | equemene | { |
79 | 1 | equemene | /*
|
80 | 1 | equemene | * Purpose
|
81 | 1 | equemene | * =======
|
82 | 1 | equemene | *
|
83 | 1 | equemene | * HPL_pdlaprnt prints to standard error a distributed matrix A. The
|
84 | 1 | equemene | * local pieces of A are sent to the process of coordinates (0,0) in
|
85 | 1 | equemene | * the grid and then printed.
|
86 | 1 | equemene | *
|
87 | 1 | equemene | * Arguments
|
88 | 1 | equemene | * =========
|
89 | 1 | equemene | *
|
90 | 1 | equemene | * GRID (local input) const HPL_T_grid *
|
91 | 1 | equemene | * On entry, GRID points to the data structure containing the
|
92 | 1 | equemene | * process grid information.
|
93 | 1 | equemene | *
|
94 | 1 | equemene | * M (global input) const int
|
95 | 1 | equemene | * On entry, M specifies the number of rows of the coefficient
|
96 | 1 | equemene | * matrix A. M must be at least zero.
|
97 | 1 | equemene | *
|
98 | 1 | equemene | * N (global input) const int
|
99 | 1 | equemene | * On entry, N specifies the number of columns of the
|
100 | 1 | equemene | * coefficient matrix A. N must be at least zero.
|
101 | 1 | equemene | *
|
102 | 1 | equemene | * NB (global input) const int
|
103 | 1 | equemene | * On entry, NB specifies the blocking factor used to partition
|
104 | 1 | equemene | * and distribute the matrix. NB must be larger than one.
|
105 | 1 | equemene | *
|
106 | 1 | equemene | * A (local input) double *
|
107 | 1 | equemene | * On entry, A points to an array of dimension (LDA,LocQ(N)).
|
108 | 1 | equemene | * This array contains the coefficient matrix to be printed.
|
109 | 1 | equemene | *
|
110 | 1 | equemene | * LDA (local input) const int
|
111 | 1 | equemene | * On entry, LDA specifies the leading dimension of the array A.
|
112 | 1 | equemene | * LDA must be at least max(1,LocP(M)).
|
113 | 1 | equemene | *
|
114 | 1 | equemene | * IAROW (global input) const int
|
115 | 1 | equemene | * On entry, IAROW specifies the row process coordinate owning
|
116 | 1 | equemene | * the first row of A. IAROW must be larger than or equal to
|
117 | 1 | equemene | * zero and less than NPROW.
|
118 | 1 | equemene | *
|
119 | 1 | equemene | * IACOL (global input) const int
|
120 | 1 | equemene | * On entry, IACOL specifies the column process coordinate
|
121 | 1 | equemene | * owning the first column of A. IACOL must be larger than or
|
122 | 1 | equemene | * equal to zero and less than NPCOL.
|
123 | 1 | equemene | *
|
124 | 1 | equemene | * CMATNM (global input) const char *
|
125 | 1 | equemene | * On entry, CMATNM is the name of the matrix to be printed.
|
126 | 1 | equemene | *
|
127 | 1 | equemene | * ---------------------------------------------------------------------
|
128 | 1 | equemene | */
|
129 | 1 | equemene | /*
|
130 | 1 | equemene | * .. Local Variables ..
|
131 | 1 | equemene | */
|
132 | 1 | equemene | MPI_Comm Acomm; |
133 | 1 | equemene | double * buf = NULL; |
134 | 1 | equemene | int h, i, ib, icurcol=IACOL, icurrow=IAROW,
|
135 | 1 | equemene | ii=0, j, jb, jj=0, mycol, myrow, npcol, |
136 | 1 | equemene | nprow, src; |
137 | 1 | equemene | /* ..
|
138 | 1 | equemene | * .. Executable Statements ..
|
139 | 1 | equemene | */
|
140 | 1 | equemene | (void) HPL_grid_info( GRID, &nprow, &npcol, &myrow, &mycol );
|
141 | 1 | equemene | Acomm = GRID->all_comm; |
142 | 1 | equemene | if( ( myrow == 0 ) && ( mycol == 0 ) ) |
143 | 1 | equemene | buf = (double*)malloc( (size_t)(NB) * sizeof( double ) ); |
144 | 1 | equemene | |
145 | 1 | equemene | for( j = 0; j < N; j += NB ) |
146 | 1 | equemene | { |
147 | 1 | equemene | jb = N-j; jb = Mmin( jb, NB ); |
148 | 1 | equemene | for( h = 0; h < jb; h++ ) |
149 | 1 | equemene | { |
150 | 1 | equemene | (void) HPL_barrier( Acomm );
|
151 | 1 | equemene | |
152 | 1 | equemene | for( i = 0; i < M; i += NB ) |
153 | 1 | equemene | { |
154 | 1 | equemene | ib = M-i; ib = Mmin( ib, NB ); |
155 | 1 | equemene | if( ( icurrow == 0 ) && ( icurcol == 0 ) ) |
156 | 1 | equemene | { |
157 | 1 | equemene | if( ( myrow == 0 ) && ( mycol == 0 ) ) |
158 | 1 | equemene | HPL_dlaprnt( ib, 1, Mptr( A, ii, jj+h, LDA ), i+1, |
159 | 1 | equemene | j+h+1, LDA, CMATNM );
|
160 | 1 | equemene | } |
161 | 1 | equemene | else
|
162 | 1 | equemene | { |
163 | 1 | equemene | if( ( myrow == icurrow ) && ( mycol == icurcol ) )
|
164 | 1 | equemene | { |
165 | 1 | equemene | (void) HPL_send( Mptr( A, ii, jj+h, LDA ), ib, 0, |
166 | 1 | equemene | 9000+(j+h)*M+i, Acomm );
|
167 | 1 | equemene | } |
168 | 1 | equemene | else if( ( myrow == 0 ) && ( mycol == 0 ) ) |
169 | 1 | equemene | { |
170 | 1 | equemene | src = HPL_pnum( GRID, icurrow, icurcol ); |
171 | 1 | equemene | (void) HPL_recv( buf, ib, src, 9000+(j+h)*M+i, |
172 | 1 | equemene | Acomm ); |
173 | 1 | equemene | HPL_dlaprnt( ib, 1, buf, i+1, j+h+1, NB, CMATNM ); |
174 | 1 | equemene | } |
175 | 1 | equemene | } |
176 | 1 | equemene | if( myrow == icurrow ) ii += ib;
|
177 | 1 | equemene | icurrow = MModAdd1( icurrow, nprow ); |
178 | 1 | equemene | (void) HPL_barrier( Acomm );
|
179 | 1 | equemene | } |
180 | 1 | equemene | ii = 0; icurrow = IAROW;
|
181 | 1 | equemene | } |
182 | 1 | equemene | if( mycol == icurcol ) jj += jb;
|
183 | 1 | equemene | icurcol = MModAdd1( icurcol, npcol ); |
184 | 1 | equemene | (void) HPL_barrier( Acomm );
|
185 | 1 | equemene | } |
186 | 1 | equemene | if( ( myrow == 0 ) && ( mycol == 0 ) && ( buf ) ) free( buf ); |
187 | 1 | equemene | /*
|
188 | 1 | equemene | * End of HPL_pdlaprnt
|
189 | 1 | equemene | */
|
190 | 1 | equemene | } |