#include <tbb/task_scheduler_init.h>
#include <tbb/blocked_range.h>
#include <tbb/parallel_reduce.h>
#include <tbb/cache_aligned_allocator.h>
using namespace tbb;
#include <altivec.h>
struct thread1 {
float ee;
float *x, *y;
thread1(float *xx, float *yy) : ee(0), x(xx), y(yy) {}
thread1(thread1& s, split) { ee = 0; x = s.x; y = s.y; }
void operator() (const blocked_range<int> & r) {
// do computation
float ve[4] = {0, 0, 0, 0};
float e = ee;
for (int i=r.begin(); i!=r.end(); i+=4) {
float* yp = &y[i+1], y0 = &y[i], ym = &y[i-1];
vec_st(vec_madd(
vec_splats(.5),
vec_add(
vec_perm(vec_ld(0,ym), vec_ld(16,ym),
vec_lvsl(0,ym)),
vec_perm(vec_ld(0,yp), vec_ld(16,yp),
vec_lvsl(0,yp))),
vec_splats(0.)),
0, &x[i]);
vec_st(vec_add(
vec_ld(0,&ve[0]),vec_madd(
vec_ld(0,y0),
vec_ld(0,y0),
vec_splats(0.))),
0, &ve[0]);
}
e += ve[0] + ve[1] + ve[2] + ve[3];
ee = e;
}
void join(thread1& s) { ee += s.ee; }
};
int main(int argc, char *argv[]) {
int n = ...;
float *x, *y;
x = cache_aligned_allocator<float>().allocate(n+1);
y = cache_aligned_allocator<float>().allocate(n+1);
... // fill x, y
thread1 t(x, y);
parallel_reduce(blocked_range<int>(1, n, 1024), t);
float e = t.ee;
... // output x, e
cache_aligned_allocator<float>().deallocate(x, n+1);
cache_aligned_allocator<float>().deallocate(y, n+1);
return 0;
}
|