| 1 |
|
| 2 |
/* $Id: SparseMatrix_MatrixVector.c 1306 2007-09-18 05:51:09Z ksteube $ */ |
| 3 |
|
| 4 |
/******************************************************* |
| 5 |
* |
| 6 |
* Copyright 2003-2007 by ACceSS MNRF |
| 7 |
* Copyright 2007 by University of Queensland |
| 8 |
* |
| 9 |
* http://esscc.uq.edu.au |
| 10 |
* Primary Business: Queensland, Australia |
| 11 |
* Licensed under the Open Software License version 3.0 |
| 12 |
* http://www.opensource.org/licenses/osl-3.0.php |
| 13 |
* |
| 14 |
*******************************************************/ |
| 15 |
|
| 16 |
/**************************************************************/ |
| 17 |
|
| 18 |
/* Paso: raw scaled vector update operation: out = alpha * A * in + beta * out */ |
| 19 |
|
| 20 |
/**************************************************************/ |
| 21 |
|
| 22 |
/* Author: gross@access.edu.au */ |
| 23 |
|
| 24 |
/**************************************************************/ |
| 25 |
|
| 26 |
#include "SparseMatrix.h" |
| 27 |
|
| 28 |
/* CSC format with offset 0*/ |
| 29 |
void Paso_SparseMatrix_MatrixVector_CSC_OFFSET0(double alpha, |
| 30 |
Paso_SparseMatrix* A, |
| 31 |
double* in, |
| 32 |
double beta, |
| 33 |
double* out) { |
| 34 |
|
| 35 |
register index_t ir,icol,iptr,icb,irb,irow,ic; |
| 36 |
register double reg,reg1,reg2,reg3; |
| 37 |
|
| 38 |
if (ABS(beta)>0.) { |
| 39 |
if (beta != 1.) { |
| 40 |
#pragma omp parallel for private(irow) schedule(static) |
| 41 |
for (irow=0;irow < A->numRows * A->row_block_size;irow++) |
| 42 |
out[irow] *= beta; |
| 43 |
} |
| 44 |
} else { |
| 45 |
#pragma omp parallel for private(irow) schedule(static) |
| 46 |
for (irow=0;irow < A->numRows * A->row_block_size;irow++) |
| 47 |
out[irow] = 0; |
| 48 |
} |
| 49 |
if (Paso_Pattern_isEmpty(A->pattern)) return; |
| 50 |
/* do the operation: */ |
| 51 |
if (ABS(alpha)>0) { |
| 52 |
if (A ->col_block_size==1 && A->row_block_size ==1) { |
| 53 |
/* TODO: parallelize (good luck!) */ |
| 54 |
for (icol=0;icol< A->pattern->numOutput;++icol) { |
| 55 |
#pragma ivdep |
| 56 |
for (iptr=A->pattern->ptr[icol];iptr<A->pattern->ptr[icol+1]; ++iptr) { |
| 57 |
out[A->pattern->index[iptr]]+= alpha * A->val[iptr] * in[icol]; |
| 58 |
} |
| 59 |
} |
| 60 |
} else if (A ->col_block_size==2 && A->row_block_size ==2) { |
| 61 |
/* TODO: parallelize */ |
| 62 |
for (ic=0;ic< A->pattern->numOutput;ic++) { |
| 63 |
#pragma ivdep |
| 64 |
for (iptr=A->pattern->ptr[ic];iptr<A->pattern->ptr[ic+1]; iptr++) { |
| 65 |
ir=2*(A->pattern->index[iptr]); |
| 66 |
out[ ir] += alpha * ( A->val[iptr*4 ]*in[ic] + A->val[iptr*4+2]*in[1+ic] ); |
| 67 |
out[1+ir] += alpha * ( A->val[iptr*4+1]*in[ic] + A->val[iptr*4+3]*in[1+ic] ); |
| 68 |
} |
| 69 |
} |
| 70 |
} else if (A ->col_block_size==3 && A->row_block_size ==3) { |
| 71 |
/* TODO: parallelize */ |
| 72 |
for (ic=0;ic< A->pattern->numOutput;ic++) { |
| 73 |
#pragma ivdep |
| 74 |
for (iptr=A->pattern->ptr[ic];iptr<A->pattern->ptr[ic+1]; iptr++) { |
| 75 |
ir=3*(A->pattern->index[iptr]); |
| 76 |
out[ ir] += alpha * ( A->val[iptr*9 ]*in[ic] + A->val[iptr*9+3]*in[1+ic] + A->val[iptr*9+6]*in[2+ic] ); |
| 77 |
out[1+ir] += alpha * ( A->val[iptr*9+1]*in[ic] + A->val[iptr*9+4]*in[1+ic] + A->val[iptr*9+7]*in[2+ic] ); |
| 78 |
out[2+ir] += alpha * ( A->val[iptr*9+2]*in[ic] + A->val[iptr*9+5]*in[1+ic] + A->val[iptr*9+8]*in[2+ic] ); |
| 79 |
} |
| 80 |
} |
| 81 |
} else { |
| 82 |
/* TODO: parallelize */ |
| 83 |
for (ic=0;ic< A->pattern->numOutput;ic++) { |
| 84 |
for (iptr=A->pattern->ptr[ic];iptr<A->pattern->ptr[ic+1]; iptr++) { |
| 85 |
for (irb=0;irb< A->row_block_size;irb++) { |
| 86 |
irow=irb+A->row_block_size*(A->pattern->index[iptr]); |
| 87 |
#pragma ivdep |
| 88 |
for (icb=0;icb< A->col_block_size;icb++) { |
| 89 |
icol=icb+A->col_block_size*ic; |
| 90 |
out[irow] += alpha * A->val[iptr*A->block_size+irb+A->row_block_size*icb] * in[icol]; |
| 91 |
} |
| 92 |
} |
| 93 |
} |
| 94 |
} |
| 95 |
} |
| 96 |
} |
| 97 |
return; |
| 98 |
} |
| 99 |
|
| 100 |
/* CSC format with offset 1*/ |
| 101 |
void Paso_SparseMatrix_MatrixVector_CSC_OFFSET1(double alpha, |
| 102 |
Paso_SparseMatrix* A, |
| 103 |
double* in, |
| 104 |
double beta, |
| 105 |
double* out) { |
| 106 |
|
| 107 |
register index_t ir,icol,iptr,icb,irb,irow,ic; |
| 108 |
register double reg,reg1,reg2,reg3; |
| 109 |
if (ABS(beta)>0.) { |
| 110 |
if (beta != 1.) { |
| 111 |
#pragma omp parallel for private(irow) schedule(static) |
| 112 |
for (irow=0;irow < A->numRows * A->row_block_size;irow++) { |
| 113 |
out[irow] *= beta; |
| 114 |
} |
| 115 |
} |
| 116 |
} else { |
| 117 |
#pragma omp parallel for private(irow) schedule(static) |
| 118 |
for (irow=0;irow < A->numRows * A->row_block_size;irow++) |
| 119 |
out[irow] = 0; |
| 120 |
} |
| 121 |
|
| 122 |
/* do the operation: */ |
| 123 |
if (ABS(alpha)>0) { |
| 124 |
if (A ->col_block_size==1 && A->row_block_size ==1) { |
| 125 |
/* TODO: parallelize (good luck!) */ |
| 126 |
for (icol=0;icol< A->pattern->numOutput;++icol) { |
| 127 |
#pragma ivdep |
| 128 |
for (iptr=A->pattern->ptr[icol]-1;iptr<A->pattern->ptr[icol+1]-1; ++iptr) { |
| 129 |
out[A->pattern->index[iptr]-1]+= alpha * A->val[iptr] * in[icol]; |
| 130 |
} |
| 131 |
} |
| 132 |
} else if (A ->col_block_size==2 && A->row_block_size ==2) { |
| 133 |
/* TODO: parallelize */ |
| 134 |
for (ic=0;ic< A->pattern->numOutput;ic++) { |
| 135 |
for (iptr=A->pattern->ptr[ic]-1;iptr<A->pattern->ptr[ic+1]-1; iptr++) { |
| 136 |
ir=2*(A->pattern->index[iptr]-1); |
| 137 |
out[ ir] += alpha * ( A->val[iptr*4 ]*in[ic] + A->val[iptr*4+2]*in[1+ic] ); |
| 138 |
out[1+ir] += alpha * ( A->val[iptr*4+1]*in[ic] + A->val[iptr*4+3]*in[1+ic] ); |
| 139 |
} |
| 140 |
} |
| 141 |
} else if (A ->col_block_size==3 && A->row_block_size ==3) { |
| 142 |
/* TODO: parallelize */ |
| 143 |
for (ic=0;ic< A->pattern->numOutput;ic++) { |
| 144 |
#pragma ivdep |
| 145 |
for (iptr=A->pattern->ptr[ic]-1;iptr<A->pattern->ptr[ic+1]-1; iptr++) { |
| 146 |
ir=3*(A->pattern->index[iptr]-1); |
| 147 |
out[ ir] += alpha * ( A->val[iptr*9 ]*in[ic] + A->val[iptr*9+3]*in[1+ic] + A->val[iptr*9+6]*in[2+ic] ); |
| 148 |
out[1+ir] += alpha * ( A->val[iptr*9+1]*in[ic] + A->val[iptr*9+4]*in[1+ic] + A->val[iptr*9+7]*in[2+ic] ); |
| 149 |
out[2+ir] += alpha * ( A->val[iptr*9+2]*in[ic] + A->val[iptr*9+5]*in[1+ic] + A->val[iptr*9+8]*in[2+ic] ); |
| 150 |
} |
| 151 |
} |
| 152 |
} else { |
| 153 |
/* TODO: parallelize */ |
| 154 |
for (ic=0;ic< A->pattern->numOutput;ic++) { |
| 155 |
for (iptr=A->pattern->ptr[ic]-1;iptr<A->pattern->ptr[ic+1]-1; iptr++) { |
| 156 |
for (irb=0;irb< A->row_block_size;irb++) { |
| 157 |
irow=irb+A->row_block_size*(A->pattern->index[iptr]-1); |
| 158 |
#pragma ivdep |
| 159 |
for (icb=0;icb< A->col_block_size;icb++) { |
| 160 |
icol=icb+A->col_block_size*ic; |
| 161 |
out[irow] += alpha * A->val[iptr*A->block_size+irb+A->row_block_size*icb] * in[icol]; |
| 162 |
} |
| 163 |
} |
| 164 |
} |
| 165 |
} |
| 166 |
} |
| 167 |
} |
| 168 |
return; |
| 169 |
} |
| 170 |
/* CSR format with offset 1*/ |
| 171 |
void Paso_SparseMatrix_MatrixVector_CSR_OFFSET1(double alpha, |
| 172 |
Paso_SparseMatrix* A, |
| 173 |
double* in, |
| 174 |
double beta, |
| 175 |
double* out) { |
| 176 |
|
| 177 |
register index_t ir,icol,iptr,icb,irb,irow,ic; |
| 178 |
register double reg,reg1,reg2,reg3; |
| 179 |
if (ABS(beta)>0.) { |
| 180 |
if (beta != 1.) { |
| 181 |
#pragma omp parallel for private(irow) schedule(static) |
| 182 |
for (irow=0;irow < A->numRows * A->row_block_size;irow++) |
| 183 |
out[irow] *= beta; |
| 184 |
} |
| 185 |
} else { |
| 186 |
#pragma omp parallel for private(irow) schedule(static) |
| 187 |
for (irow=0;irow < A->numRows * A->row_block_size;irow++) |
| 188 |
out[irow] = 0; |
| 189 |
} |
| 190 |
/* do the operation: */ |
| 191 |
if (ABS(alpha)>0) { |
| 192 |
if (A ->col_block_size==1 && A->row_block_size ==1) { |
| 193 |
#pragma omp parallel for private(irow,iptr,reg) schedule(static) |
| 194 |
for (irow=0;irow< A->pattern->numOutput;++irow) { |
| 195 |
reg=0.; |
| 196 |
#pragma ivdep |
| 197 |
for (iptr=(A->pattern->ptr[irow])-1;iptr<(A->pattern->ptr[irow+1])-1; ++iptr) { |
| 198 |
reg += A->val[iptr] * in[A->pattern->index[iptr]-1]; |
| 199 |
} |
| 200 |
out[irow] += alpha * reg; |
| 201 |
} |
| 202 |
} else if (A ->col_block_size==2 && A->row_block_size ==2) { |
| 203 |
#pragma omp parallel for private(ir,reg1,reg2,iptr,ic) schedule(static) |
| 204 |
for (ir=0;ir< A->pattern->numOutput;ir++) { |
| 205 |
reg1=0.; |
| 206 |
reg2=0.; |
| 207 |
#pragma ivdep |
| 208 |
for (iptr=A->pattern->ptr[ir]-1;iptr<A->pattern->ptr[ir+1]-1; iptr++) { |
| 209 |
ic=2*(A->pattern->index[iptr]-1); |
| 210 |
reg1 += A->val[iptr*4 ]*in[ic] + A->val[iptr*4+2]*in[1+ic]; |
| 211 |
reg2 += A->val[iptr*4+1]*in[ic] + A->val[iptr*4+3]*in[1+ic]; |
| 212 |
} |
| 213 |
out[ 2*ir] += alpha * reg1; |
| 214 |
out[1+2*ir] += alpha * reg2; |
| 215 |
} |
| 216 |
} else if (A ->col_block_size==3 && A->row_block_size ==3) { |
| 217 |
#pragma omp parallel for private(ir,reg1,reg2,reg3,iptr,ic) schedule(static) |
| 218 |
for (ir=0;ir< A->pattern->numOutput;ir++) { |
| 219 |
reg1=0.; |
| 220 |
reg2=0.; |
| 221 |
reg3=0.; |
| 222 |
#pragma ivdep |
| 223 |
for (iptr=A->pattern->ptr[ir]-1;iptr<A->pattern->ptr[ir+1]-1; iptr++) { |
| 224 |
ic=3*(A->pattern->index[iptr]-1); |
| 225 |
reg1 += A->val[iptr*9 ]*in[ic] + A->val[iptr*9+3]*in[1+ic] + A->val[iptr*9+6]*in[2+ic]; |
| 226 |
reg2 += A->val[iptr*9+1]*in[ic] + A->val[iptr*9+4]*in[1+ic] + A->val[iptr*9+7]*in[2+ic]; |
| 227 |
reg3 += A->val[iptr*9+2]*in[ic] + A->val[iptr*9+5]*in[1+ic] + A->val[iptr*9+8]*in[2+ic]; |
| 228 |
} |
| 229 |
out[ 3*ir] += alpha * reg1; |
| 230 |
out[1+3*ir] += alpha * reg2; |
| 231 |
out[2+3*ir] += alpha * reg3; |
| 232 |
} |
| 233 |
} else { |
| 234 |
#pragma omp parallel for private(ir,iptr,irb,icb,irow,icol,reg) schedule(static) |
| 235 |
for (ir=0;ir< A->pattern->numOutput;ir++) { |
| 236 |
for (iptr=A->pattern->ptr[ir]-1;iptr<A->pattern->ptr[ir+1]-1; iptr++) { |
| 237 |
for (irb=0;irb< A->row_block_size;irb++) { |
| 238 |
irow=irb+A->row_block_size*ir; |
| 239 |
reg=0.; |
| 240 |
#pragma ivdep |
| 241 |
for (icb=0;icb< A->col_block_size;icb++) { |
| 242 |
icol=icb+A->col_block_size*(A->pattern->index[iptr]-1); |
| 243 |
reg += A->val[iptr*A->block_size+irb+A->row_block_size*icb] * in[icol]; |
| 244 |
} |
| 245 |
out[irow] += alpha * reg; |
| 246 |
} |
| 247 |
} |
| 248 |
} |
| 249 |
} |
| 250 |
} |
| 251 |
return; |
| 252 |
} |
| 253 |
/* CSR format with offset 0*/ |
| 254 |
void Paso_SparseMatrix_MatrixVector_CSR_OFFSET0(double alpha, |
| 255 |
Paso_SparseMatrix* A, |
| 256 |
double* in, |
| 257 |
double beta, |
| 258 |
double* out) |
| 259 |
{ |
| 260 |
/*#define PASO_DYNAMIC_SCHEDULING_MVM */ |
| 261 |
|
| 262 |
#if defined PASO_DYNAMIC_SCHEDULING_MVM && defined __OPENMP |
| 263 |
#define USE_DYNAMIC_SCHEDULING |
| 264 |
#endif |
| 265 |
|
| 266 |
char* chksz_chr=NULL; |
| 267 |
dim_t chunk_size=1; |
| 268 |
dim_t nrow=A->numRows; |
| 269 |
dim_t np, len, rest, irow, local_n, p, n_chunks; |
| 270 |
np=omp_get_max_threads(); |
| 271 |
#ifdef USE_DYNAMIC_SCHEDULING |
| 272 |
chksz_chr=getenv("PASO_CHUNK_SIZE_MVM"); |
| 273 |
if (chksz_chr!=NULL) sscanf(chksz_chr, "%d",&chunk_size); |
| 274 |
chunk_size=MIN(MAX(1,chunk_size),nrow/np); |
| 275 |
n_chunks=nrow/chunk_size; |
| 276 |
if (n_chunks*chunk_size<nrow) n_chunks+=1; |
| 277 |
#else |
| 278 |
len=nrow/np; |
| 279 |
rest=nrow-len*np; |
| 280 |
#endif |
| 281 |
|
| 282 |
#pragma omp parallel private(irow, p, local_n) |
| 283 |
{ |
| 284 |
#ifdef USE_DYNAMIC_SCHEDULING |
| 285 |
#pragma omp for schedule(dynamic,1) |
| 286 |
for (p=0; p<n_chunks;p++) { |
| 287 |
irow=chunk_size*p; |
| 288 |
local_n=MIN(chunk_size,nrow-chunk_size*p); |
| 289 |
#else |
| 290 |
#pragma omp for schedule(static) |
| 291 |
for (p=0; p<np;p++) { |
| 292 |
irow=len*p+MIN(p,rest); |
| 293 |
local_n=len+(p<rest ? 1 :0 ); |
| 294 |
#endif |
| 295 |
Paso_SparseMatrix_MatrixVector_CSR_OFFSET0_stripe(alpha, |
| 296 |
local_n, |
| 297 |
A->row_block_size, |
| 298 |
A->col_block_size, |
| 299 |
&(A->pattern->ptr[irow]), |
| 300 |
A->pattern->index, A->val, in, beta, &out[irow*A->row_block_size]); |
| 301 |
#ifdef USE_DYNAMIC_SCHEDULING |
| 302 |
} |
| 303 |
#else |
| 304 |
} |
| 305 |
#endif |
| 306 |
} |
| 307 |
} |
| 308 |
/* CSR format with offset 0*/ |
| 309 |
void Paso_SparseMatrix_MatrixVector_CSR_OFFSET0_stripe(double alpha, |
| 310 |
dim_t nRows, |
| 311 |
dim_t row_block_size, |
| 312 |
dim_t col_block_size, |
| 313 |
index_t* ptr, |
| 314 |
index_t* index, |
| 315 |
double* val, |
| 316 |
double* in, |
| 317 |
double beta, |
| 318 |
double* out) |
| 319 |
{ |
| 320 |
register index_t ir,icol,iptr,icb,irb,irow,ic,Aiptr; |
| 321 |
register double reg,reg1,reg2,reg3,in1,in2,in3,A00,A10,A20,A01,A11,A21,A02,A12,A22; |
| 322 |
dim_t block_size; |
| 323 |
if (ABS(beta)>0.) { |
| 324 |
if (beta != 1.) { |
| 325 |
for (irow=0;irow < nRows * row_block_size;irow++) |
| 326 |
out[irow] *= beta; |
| 327 |
} |
| 328 |
} else { |
| 329 |
for (irow=0;irow < nRows * row_block_size;irow++) |
| 330 |
out[irow] = 0; |
| 331 |
} |
| 332 |
if (ABS(alpha)>0) { |
| 333 |
if (col_block_size==1 && row_block_size ==1) { |
| 334 |
for (irow=0;irow< nRows;++irow) { |
| 335 |
reg=0.; |
| 336 |
#pragma ivdep |
| 337 |
for (iptr=ptr[irow];iptr<ptr[irow+1]; ++iptr) { |
| 338 |
reg += val[iptr] * in[index[iptr]]; |
| 339 |
} |
| 340 |
out[irow] += alpha * reg; |
| 341 |
} |
| 342 |
} else if (col_block_size==2 && row_block_size ==2) { |
| 343 |
for (ir=0;ir< nRows;ir++) { |
| 344 |
reg1=0.; |
| 345 |
reg2=0.; |
| 346 |
#pragma ivdep |
| 347 |
for (iptr=ptr[ir];iptr<ptr[ir+1]; iptr++) { |
| 348 |
ic=2*index[iptr]; |
| 349 |
Aiptr=iptr*4; |
| 350 |
in1=in[ic]; |
| 351 |
in2=in[1+ic]; |
| 352 |
A00=val[Aiptr ]; |
| 353 |
A10=val[Aiptr+1]; |
| 354 |
A01=val[Aiptr+2]; |
| 355 |
A11=val[Aiptr+3]; |
| 356 |
reg1 += A00*in1 + A01*in2; |
| 357 |
reg2 += A10*in1 + A11*in2; |
| 358 |
} |
| 359 |
out[ 2*ir] += alpha * reg1; |
| 360 |
out[1+2*ir] += alpha * reg2; |
| 361 |
} |
| 362 |
} else if (col_block_size==3 && row_block_size ==3) { |
| 363 |
for (ir=0;ir< nRows;ir++) { |
| 364 |
reg1=0.; |
| 365 |
reg2=0.; |
| 366 |
reg3=0.; |
| 367 |
#pragma ivdep |
| 368 |
for (iptr=ptr[ir];iptr<ptr[ir+1]; iptr++) { |
| 369 |
ic=3*index[iptr]; |
| 370 |
Aiptr=iptr*9; |
| 371 |
in1=in[ic]; |
| 372 |
in2=in[1+ic]; |
| 373 |
in3=in[2+ic]; |
| 374 |
A00=val[Aiptr ]; |
| 375 |
A10=val[Aiptr+1]; |
| 376 |
A20=val[Aiptr+2]; |
| 377 |
A01=val[Aiptr+3]; |
| 378 |
A11=val[Aiptr+4]; |
| 379 |
A21=val[Aiptr+5]; |
| 380 |
A02=val[Aiptr+6]; |
| 381 |
A12=val[Aiptr+7]; |
| 382 |
A22=val[Aiptr+8]; |
| 383 |
reg1 += A00*in1 + A01*in2 + A02*in3; |
| 384 |
reg2 += A10*in1 + A11*in2 + A12*in3; |
| 385 |
reg3 += A20*in1 + A21*in2 + A22*in3; |
| 386 |
} |
| 387 |
out[ 3*ir] += alpha * reg1; |
| 388 |
out[1+3*ir] += alpha * reg2; |
| 389 |
out[2+3*ir] += alpha * reg3; |
| 390 |
} |
| 391 |
} else { |
| 392 |
block_size=col_block_size*row_block_size; |
| 393 |
for (ir=0;ir< nRows;ir++) { |
| 394 |
for (iptr=ptr[ir];iptr<ptr[ir+1]; iptr++) { |
| 395 |
for (irb=0;irb< row_block_size;irb++) { |
| 396 |
irow=irb+row_block_size*ir; |
| 397 |
reg=0.; |
| 398 |
#pragma ivdep |
| 399 |
for (icb=0;icb< col_block_size;icb++) { |
| 400 |
icol=icb+col_block_size*index[iptr]; |
| 401 |
reg += val[iptr*block_size+irb+row_block_size*icb] * in[icol]; |
| 402 |
} |
| 403 |
out[irow] += alpha * reg; |
| 404 |
} |
| 405 |
} |
| 406 |
} |
| 407 |
} |
| 408 |
} |
| 409 |
return; |
| 410 |
} |
Parent Directory
| 
Revision Log