317 {

318  p_fetch = 0;

319  int * __restrict a = (int*) addr;

320  const int inc = CLSIZE/ssize; // number of elements in one cache line

321 

322  // Grab every 64th address

323  for(int i=0; i<total; i+=inc)

324  {

325  p_fetch += a[i];

326  }

327 }

328 

329 

330 

331 template <class T, class ITYPE>

332 void Sym<T, ITYPE>::Transpose()

333 {

334  // when we jump to the next block in the same block-column, we move leaddim positions inside "top" array

335  // leadim ~= sqrt(n) => number of blocks in each block-row

336  ITYPE leaddim = lowcolmask+1;

337  Sym symT(nz, m, n); // create empty transposed object

338 

339  ITYPE k = 0;

340  ITYPE cnz = 0;

341 

342  for(ITYPE j = 0; j < leaddim; ++j) // scan columns of top-level structure (~sqrt(n) iterations)

343  {

344  for(ITYPE i = j; i < ntop ; i += leaddim) // iterates ~ sqrt(m) times within the block column

345  {

346  symT.top[k++] = cnz;

347  cnz += top[i+1]-top[i];

348  }

349  }

350  symT.top[k] = cnz;

351 

352  // Embarrassingly parallel sort of indices to get new bottom array

353  // ITYPE nindex = (highmask & csc.ir [i]) | ((highmask & bot) >> 4);

354 }

355