root / src / pauxil / HPL_infog2l.c @ 1
Historique | Voir | Annoter | Télécharger (14,02 ko)
| 1 |
/*
|
|---|---|
| 2 |
* -- High Performance Computing Linpack Benchmark (HPL)
|
| 3 |
* HPL - 2.0 - September 10, 2008
|
| 4 |
* Antoine P. Petitet
|
| 5 |
* University of Tennessee, Knoxville
|
| 6 |
* Innovative Computing Laboratory
|
| 7 |
* (C) Copyright 2000-2008 All Rights Reserved
|
| 8 |
*
|
| 9 |
* -- Copyright notice and Licensing terms:
|
| 10 |
*
|
| 11 |
* Redistribution and use in source and binary forms, with or without
|
| 12 |
* modification, are permitted provided that the following conditions
|
| 13 |
* are met:
|
| 14 |
*
|
| 15 |
* 1. Redistributions of source code must retain the above copyright
|
| 16 |
* notice, this list of conditions and the following disclaimer.
|
| 17 |
*
|
| 18 |
* 2. Redistributions in binary form must reproduce the above copyright
|
| 19 |
* notice, this list of conditions, and the following disclaimer in the
|
| 20 |
* documentation and/or other materials provided with the distribution.
|
| 21 |
*
|
| 22 |
* 3. All advertising materials mentioning features or use of this
|
| 23 |
* software must display the following acknowledgement:
|
| 24 |
* This product includes software developed at the University of
|
| 25 |
* Tennessee, Knoxville, Innovative Computing Laboratory.
|
| 26 |
*
|
| 27 |
* 4. The name of the University, the name of the Laboratory, or the
|
| 28 |
* names of its contributors may not be used to endorse or promote
|
| 29 |
* products derived from this software without specific written
|
| 30 |
* permission.
|
| 31 |
*
|
| 32 |
* -- Disclaimer:
|
| 33 |
*
|
| 34 |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
| 35 |
* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
| 36 |
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
| 37 |
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY
|
| 38 |
* OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
| 39 |
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
| 40 |
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
| 41 |
* DATA OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
| 42 |
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
| 43 |
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
| 44 |
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
| 45 |
* ---------------------------------------------------------------------
|
| 46 |
*/
|
| 47 |
/*
|
| 48 |
* Include files
|
| 49 |
*/
|
| 50 |
#include "hpl.h" |
| 51 |
|
| 52 |
#ifdef STDC_HEADERS
|
| 53 |
void HPL_infog2l
|
| 54 |
( |
| 55 |
int I,
|
| 56 |
int J,
|
| 57 |
const int IMB, |
| 58 |
const int MB, |
| 59 |
const int INB, |
| 60 |
const int NB, |
| 61 |
const int RSRC, |
| 62 |
const int CSRC, |
| 63 |
const int MYROW, |
| 64 |
const int MYCOL, |
| 65 |
const int NPROW, |
| 66 |
const int NPCOL, |
| 67 |
int * II,
|
| 68 |
int * JJ,
|
| 69 |
int * PROW,
|
| 70 |
int * PCOL
|
| 71 |
) |
| 72 |
#else
|
| 73 |
void HPL_infog2l
|
| 74 |
( I, J, IMB, MB, INB, NB, RSRC, CSRC, MYROW, MYCOL, NPROW, NPCOL, II, JJ, PROW, PCOL ) |
| 75 |
int I;
|
| 76 |
int J;
|
| 77 |
const int IMB; |
| 78 |
const int MB; |
| 79 |
const int INB; |
| 80 |
const int NB; |
| 81 |
const int RSRC; |
| 82 |
const int CSRC; |
| 83 |
const int MYROW; |
| 84 |
const int MYCOL; |
| 85 |
const int NPROW; |
| 86 |
const int NPCOL; |
| 87 |
int * II;
|
| 88 |
int * JJ;
|
| 89 |
int * PROW;
|
| 90 |
int * PCOL;
|
| 91 |
#endif
|
| 92 |
{
|
| 93 |
/*
|
| 94 |
* Purpose
|
| 95 |
* =======
|
| 96 |
*
|
| 97 |
* HPL_infog2l computes the starting local index II, JJ corresponding to
|
| 98 |
* the submatrix starting globally at the entry pointed by I, J. This
|
| 99 |
* routine returns the coordinates in the grid of the process owning the
|
| 100 |
* matrix entry of global indexes I, J, namely PROW and PCOL.
|
| 101 |
*
|
| 102 |
* Arguments
|
| 103 |
* =========
|
| 104 |
*
|
| 105 |
* I (global input) int
|
| 106 |
* On entry, I specifies the global row index of the matrix
|
| 107 |
* entry. I must be at least zero.
|
| 108 |
*
|
| 109 |
* J (global input) int
|
| 110 |
* On entry, J specifies the global column index of the matrix
|
| 111 |
* entry. J must be at least zero.
|
| 112 |
*
|
| 113 |
* IMB (global input) const int
|
| 114 |
* On entry, IMB specifies the size of the first row block of
|
| 115 |
* the global matrix. IMB must be at least one.
|
| 116 |
*
|
| 117 |
* MB (global input) const int
|
| 118 |
* On entry, MB specifies the blocking factor used to partition
|
| 119 |
* and distribute the rows of the matrix A. MB must be larger
|
| 120 |
* than one.
|
| 121 |
*
|
| 122 |
* INB (global input) const int
|
| 123 |
* On entry, INB specifies the size of the first column block of
|
| 124 |
* the global matrix. INB must be at least one.
|
| 125 |
*
|
| 126 |
* NB (global input) const int
|
| 127 |
* On entry, NB specifies the blocking factor used to partition
|
| 128 |
* and distribute the columns of the matrix A. NB must be larger
|
| 129 |
* than one.
|
| 130 |
*
|
| 131 |
* RSRC (global input) const int
|
| 132 |
* On entry, RSRC specifies the row coordinate of the process
|
| 133 |
* that possesses the row I. RSRC must be at least zero and
|
| 134 |
* strictly less than NPROW.
|
| 135 |
*
|
| 136 |
* CSRC (global input) const int
|
| 137 |
* On entry, CSRC specifies the column coordinate of the process
|
| 138 |
* that possesses the column J. CSRC must be at least zero and
|
| 139 |
* strictly less than NPCOL.
|
| 140 |
*
|
| 141 |
* MYROW (local input) const int
|
| 142 |
* On entry, MYROW specifies my row process coordinate in the
|
| 143 |
* grid. MYROW is greater than or equal to zero and less than
|
| 144 |
* NPROW.
|
| 145 |
*
|
| 146 |
* MYCOL (local input) const int
|
| 147 |
* On entry, MYCOL specifies my column process coordinate in the
|
| 148 |
* grid. MYCOL is greater than or equal to zero and less than
|
| 149 |
* NPCOL.
|
| 150 |
*
|
| 151 |
* NPROW (global input) const int
|
| 152 |
* On entry, NPROW specifies the number of process rows in the
|
| 153 |
* grid. NPROW is at least one.
|
| 154 |
*
|
| 155 |
* NPCOL (global input) const int
|
| 156 |
* On entry, NPCOL specifies the number of process columns in
|
| 157 |
* the grid. NPCOL is at least one.
|
| 158 |
*
|
| 159 |
* II (local output) int *
|
| 160 |
* On exit, II specifies the local starting row index of the
|
| 161 |
* submatrix. On exit, II is at least 0.
|
| 162 |
*
|
| 163 |
* JJ (local output) int *
|
| 164 |
* On exit, JJ specifies the local starting column index of the
|
| 165 |
* submatrix. On exit, JJ is at least 0.
|
| 166 |
*
|
| 167 |
* PROW (global output) int *
|
| 168 |
* On exit, PROW is the row coordinate of the process owning the
|
| 169 |
* entry specified by the global index I. PROW is at least zero
|
| 170 |
* and less than NPROW.
|
| 171 |
*
|
| 172 |
* PCOL (global output) int *
|
| 173 |
* On exit, PCOL is the column coordinate of the process owning
|
| 174 |
* the entry specified by the global index J. PCOL is at least
|
| 175 |
* zero and less than NPCOL.
|
| 176 |
*
|
| 177 |
* ---------------------------------------------------------------------
|
| 178 |
*/
|
| 179 |
/*
|
| 180 |
* .. Local Variables ..
|
| 181 |
*/
|
| 182 |
int ilocblk, imb, inb, mb, mydist, nb, nblocks, csrc, rsrc;
|
| 183 |
/* ..
|
| 184 |
* .. Executable Statements ..
|
| 185 |
*/
|
| 186 |
imb = IMB; |
| 187 |
*PROW = RSRC; |
| 188 |
|
| 189 |
if( ( *PROW == -1 ) || ( NPROW == 1 ) ) |
| 190 |
{
|
| 191 |
/*
|
| 192 |
* The data is not distributed, or there is just one process row in the
|
| 193 |
* grid.
|
| 194 |
*/
|
| 195 |
*II = I; |
| 196 |
} |
| 197 |
else if( I < imb ) |
| 198 |
{
|
| 199 |
/*
|
| 200 |
* I refers to an entry in the first block of rows
|
| 201 |
*/
|
| 202 |
*II = ( MYROW == *PROW ? I : 0 );
|
| 203 |
} |
| 204 |
else
|
| 205 |
{
|
| 206 |
mb = MB; |
| 207 |
rsrc = *PROW; |
| 208 |
/*
|
| 209 |
* The discussion goes as follows: compute my distance from the source
|
| 210 |
* process so that within this process coordinate system, the source
|
| 211 |
* process is the process such that mydist = 0, or equivalently
|
| 212 |
* MYROW == rsrc.
|
| 213 |
*
|
| 214 |
* Find out the global coordinate of the block I belongs to (nblocks),
|
| 215 |
* as well as the minimum local number of blocks that every process has.
|
| 216 |
*
|
| 217 |
* when mydist < nblocks-ilocblk*NPROCS, I own ilocblk + 1 full blocks,
|
| 218 |
* when mydist > nblocks-ilocblk*NPROCS, I own ilocblk full blocks,
|
| 219 |
* when mydist = nblocks-ilocblk*NPROCS, I own ilocblk full blocks
|
| 220 |
* but not I, or I own ilocblk + 1 blocks and the entry I refers to.
|
| 221 |
*/
|
| 222 |
if( MYROW == rsrc )
|
| 223 |
{
|
| 224 |
/*
|
| 225 |
* I refers to an entry that is not in the first block, find out which
|
| 226 |
* process has it.
|
| 227 |
*/
|
| 228 |
nblocks = ( I - imb ) / mb + 1;
|
| 229 |
*PROW += nblocks; |
| 230 |
*PROW -= ( *PROW / NPROW ) * NPROW; |
| 231 |
/*
|
| 232 |
* Since mydist = 0 and nblocks - ilocblk * NPROW >= 0, there are only
|
| 233 |
* three possible cases:
|
| 234 |
*
|
| 235 |
* 1) When 0 = mydist = nblocks - ilocblk * NPROW = 0 and I do not own
|
| 236 |
* I, in which case II = IMB + ( ilocblk - 1 ) * MB. Note that this
|
| 237 |
* case cannot happen when ilocblk is zero, since nblocks is at
|
| 238 |
* least one.
|
| 239 |
*
|
| 240 |
* 2) When 0 = mydist = nblocks - ilocblk * NPROW = 0 and I own I, in
|
| 241 |
* which case I and II can respectively be written as IMB +
|
| 242 |
* (nblocks-1)*NB + IL and IMB + (ilocblk-1) * MB + IL. That is
|
| 243 |
* II = I + (ilocblk-nblocks)*MB. Note that this case cannot happen
|
| 244 |
* when ilocblk is zero, since nblocks is at least one.
|
| 245 |
*
|
| 246 |
* 3) mydist = 0 < nblocks - ilocblk * NPROW, the source process owns
|
| 247 |
* ilocblk+1 full blocks, and therefore II = IMB + ilocblk * MB.
|
| 248 |
* Note that when ilocblk is zero, II is just IMB.
|
| 249 |
*/
|
| 250 |
if( nblocks < NPROW )
|
| 251 |
{
|
| 252 |
*II = imb; |
| 253 |
} |
| 254 |
else
|
| 255 |
{
|
| 256 |
ilocblk = nblocks / NPROW; |
| 257 |
if( ilocblk * NPROW >= nblocks )
|
| 258 |
{
|
| 259 |
*II = ( ( MYROW == *PROW ) ? |
| 260 |
I + ( ilocblk - nblocks ) * mb : |
| 261 |
imb + ( ilocblk - 1 ) * mb );
|
| 262 |
} |
| 263 |
else
|
| 264 |
{
|
| 265 |
*II = imb + ilocblk * mb; |
| 266 |
} |
| 267 |
} |
| 268 |
} |
| 269 |
else
|
| 270 |
{
|
| 271 |
/*
|
| 272 |
* I refers to an entry that is not in the first block, find out which
|
| 273 |
* process has it.
|
| 274 |
*/
|
| 275 |
nblocks = ( I -= imb ) / mb + 1;
|
| 276 |
*PROW += nblocks; |
| 277 |
*PROW -= ( *PROW / NPROW ) * NPROW; |
| 278 |
/*
|
| 279 |
* Compute my distance from the source process so that within this pro-
|
| 280 |
* cess coordinate system, the source process is the process such that
|
| 281 |
* mydist=0.
|
| 282 |
*/
|
| 283 |
if( ( mydist = MYROW - rsrc ) < 0 ) mydist += NPROW; |
| 284 |
/*
|
| 285 |
* When mydist < nblocks - ilocblk * NPROW, I own ilocblk+1 full blocks
|
| 286 |
* of size MB since I am not the source process, i.e. II=(ilocblk+1)*MB.
|
| 287 |
* When mydist>=nblocks-ilocblk*NPROW and I do not own I, I own ilocblk
|
| 288 |
* full blocks of size MB, i.e. II = ilocblk*MB, otherwise I own ilocblk
|
| 289 |
* blocks and I, in which case I can be written as IMB + (nblocks-1)*MB
|
| 290 |
* + IL and II = ilocblk*MB + IL = I - IMB + (ilocblk - nblocks + 1)*MB.
|
| 291 |
*/
|
| 292 |
if( nblocks < NPROW )
|
| 293 |
{
|
| 294 |
mydist -= nblocks; |
| 295 |
*II = ( ( mydist < 0 ) ? mb :
|
| 296 |
( ( MYROW == *PROW ) ? |
| 297 |
I + ( 1 - nblocks ) * mb : 0 ) ); |
| 298 |
} |
| 299 |
else
|
| 300 |
{
|
| 301 |
ilocblk = nblocks / NPROW; |
| 302 |
mydist -= nblocks - ilocblk * NPROW; |
| 303 |
*II = ( ( mydist < 0 ) ? ( ilocblk + 1 ) * mb : |
| 304 |
( ( MYROW == *PROW ) ? |
| 305 |
( ilocblk - nblocks + 1 ) * mb + I :
|
| 306 |
ilocblk * mb ) ); |
| 307 |
} |
| 308 |
} |
| 309 |
} |
| 310 |
/*
|
| 311 |
* Idem for the columns
|
| 312 |
*/
|
| 313 |
inb = INB; |
| 314 |
*PCOL = CSRC; |
| 315 |
|
| 316 |
if( ( *PCOL == -1 ) || ( NPCOL == 1 ) ) |
| 317 |
{
|
| 318 |
*JJ = J; |
| 319 |
} |
| 320 |
else if( J < inb ) |
| 321 |
{
|
| 322 |
*JJ = ( MYCOL == *PCOL ? J : 0 );
|
| 323 |
} |
| 324 |
else
|
| 325 |
{
|
| 326 |
nb = NB; |
| 327 |
csrc = *PCOL; |
| 328 |
|
| 329 |
if( MYCOL == csrc )
|
| 330 |
{
|
| 331 |
nblocks = ( J - inb ) / nb + 1;
|
| 332 |
*PCOL += nblocks; |
| 333 |
*PCOL -= ( *PCOL / NPCOL ) * NPCOL; |
| 334 |
|
| 335 |
if( nblocks < NPCOL )
|
| 336 |
{
|
| 337 |
*JJ = inb; |
| 338 |
} |
| 339 |
else
|
| 340 |
{
|
| 341 |
ilocblk = nblocks / NPCOL; |
| 342 |
if( ilocblk * NPCOL >= nblocks )
|
| 343 |
{
|
| 344 |
*JJ = ( ( MYCOL == *PCOL ) ? |
| 345 |
J + ( ilocblk - nblocks ) * nb : |
| 346 |
inb + ( ilocblk - 1 ) * nb );
|
| 347 |
} |
| 348 |
else
|
| 349 |
{
|
| 350 |
*JJ = inb + ilocblk * nb; |
| 351 |
} |
| 352 |
} |
| 353 |
} |
| 354 |
else
|
| 355 |
{
|
| 356 |
nblocks = ( J -= inb ) / nb + 1;
|
| 357 |
*PCOL += nblocks; |
| 358 |
*PCOL -= ( *PCOL / NPCOL ) * NPCOL; |
| 359 |
|
| 360 |
if( ( mydist = MYCOL - csrc ) < 0 ) mydist += NPCOL; |
| 361 |
|
| 362 |
if( nblocks < NPCOL )
|
| 363 |
{
|
| 364 |
mydist -= nblocks; |
| 365 |
*JJ = ( ( mydist < 0 ) ? nb : ( ( MYCOL == *PCOL ) ?
|
| 366 |
J + ( 1 - nblocks )*nb : 0 ) ); |
| 367 |
} |
| 368 |
else
|
| 369 |
{
|
| 370 |
ilocblk = nblocks / NPCOL; |
| 371 |
mydist -= nblocks - ilocblk * NPCOL; |
| 372 |
*JJ = ( ( mydist < 0 ) ? ( ilocblk + 1 ) * nb : |
| 373 |
( ( MYCOL == *PCOL ) ? |
| 374 |
( ilocblk - nblocks + 1 ) * nb + J :
|
| 375 |
ilocblk * nb ) ); |
| 376 |
} |
| 377 |
} |
| 378 |
} |
| 379 |
/*
|
| 380 |
* End of HPL_infog2l
|
| 381 |
*/
|
| 382 |
} |