void CowichanTBB::life(BoolMatrix input, BoolMatrix output) {
GameOfLife game(input, output, nr, nc);
for (index_t i = 0; i < LIFE_ITERATIONS; ++i) {
// update CA simulation
parallel_reduce(Range2D(0, nr, 0, nc), game, auto_partitioner());
// check if there are alive cells
if (!game.isAlive()) {
no_cells_alive();
}
// swap arrays (ping-pong approach)
game.swap();
}
// final result is in input - copy to output
if (LIFE_ITERATIONS % 2 == 0) {
for (index_t r = 0; r < nr; r++) {
for (index_t c = 0; c < nc; c++) {
MATRIX_RECT(output, r, c) = MATRIX_RECT(input, r, c);
}
}
}
}
void CowichanTBB::norm(PointVector pointsIn, PointVector pointsOut) {
MinMaxReducer minmax(pointsIn);
// find min/max coordinates
parallel_reduce(Range(0, n), minmax, auto_partitioner());
Point minPoint = minmax.getMinimum();
Point maxPoint = minmax.getMaximum();
// compute scaling factors
real xfactor = (real)((maxPoint.x == minPoint.x) ?
0.0 : 1.0 / (maxPoint.x - minPoint.x));
real yfactor = (real)((maxPoint.y == minPoint.y) ?
0.0 : 1.0 / (maxPoint.y - minPoint.y));
Normalizer normalizer(pointsIn, pointsOut, minPoint.x, minPoint.y, xfactor,
yfactor);
// normalize the vector
parallel_for(Range(0, n), normalizer, auto_partitioner());
}
void I2D_CoreFMM_AggressiveVel::solve(const Real theta, const Real inv_scaling, BlockInfo * dest, const int nblocks, VelocitySourceParticle * srcparticles, const int nparticles)
{
Profiler profiler;
_THETA = theta;
tBox * rootBox=new tBox;
tick_count start_before_tree = tick_count::now ();
profiler.push_start("tree");
tBoxBuilder::buildBoxes(srcparticles, nparticles, rootBox);
profiler.pop_stop();
tick_count start_before_expansions = tick_count::now ();
profiler.push_start("expansions");
tBoxBuilder::generateExpansions(rootBox);
profiler.pop_stop();
tick_count start_before_evaluations = tick_count::now ();
profiler.push_start("evaluations");
Measurements measurements (nblocks);
VelocityEvaluator evaluator(rootBox, inv_scaling, measurements);
evaluator.destblocks = dest;
tbb::parallel_for(blocked_range<int>(0, nblocks), evaluator, auto_partitioner());
profiler.pop_stop();
tick_count end = tick_count::now ();
if (timestamp++ % 5 == 0) profiler.printSummary();
if (b_verbose) {
const double tree_wallclock_time = (start_before_expansions - start_before_tree).seconds();
const double evaluations_wallclock_time = (end - start_before_evaluations).seconds();
const double expansions_wallclock_time = (start_before_evaluations - start_before_expansions).seconds();
const double num_effective_interactions = (double)nparticles*nblocks*_BLOCKSIZE_*_BLOCKSIZE_;
double effective_direct_gflops, effective_indirect_gflops, effective_total_gflops;
double actual_direct_gflops, actual_indirect_gflops, actual_total_gflops;
double num_direct_evals, num_indirect_evals;
num_direct_evals = num_indirect_evals = 0;
for (int i=0; i<nblocks; ++i) {
num_direct_evals += (double)measurements.num_direct_evals[i];
num_indirect_evals += (double)measurements.num_indirect_evals[i] ;
}
effective_direct_gflops = num_direct_evals * 11./1e9;
actual_direct_gflops = num_direct_evals * 15./1e9;
effective_indirect_gflops = num_indirect_evals * (16 + 14*_ORDER_) * 1./1e9;
actual_indirect_gflops = num_indirect_evals * (20 + 14*_ORDER_) * 1./1e9;
effective_total_gflops = effective_direct_gflops + effective_indirect_gflops;
actual_total_gflops = actual_direct_gflops + actual_indirect_gflops;
//Write data for break even plot
std::fstream file;
file.open ("measurements.txt", std::fstream::app | std::fstream::out);
file << setprecision (6) << scientific << num_effective_interactions << "\t" << effective_total_gflops << "\t" \
<< actual_total_gflops << "\t" << evaluations_wallclock_time << "\t" << \
expansions_wallclock_time << "\t" << tree_wallclock_time << "\n";
file.close ();
}
delete rootBox;
}
