code example
TBB Intel Threading Building Blocks
SSE IA SIMD instructions
#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 <xmmintrin.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 half =.5;
   _mm_store_ps(&x[i],
     _mm_mul_ps(_mm_load1_ps(&half),
       _mm_add_ps(_mm_loadu_ps(&y[i+1]),
         _mm_loadu_ps(&y[i-1]))));
   _mm_store_ps(&ve[0],
     _mm_add_ps(_mm_load_ps(&ve[0]),
       _mm_mul_ps(_mm_load_ps(&y[i]),
         _mm_load_ps(&y[i]))));
  }
  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;
}
[start] [references] [download] [install]