float
ComputeGravitation_SIMD_openmp(
float *force[3],
float *pos[4],
float *mass,
float softeningSquared,
size_t N
)
{
chTimerTimestamp start, end;
chTimerGetTime( &start );
#pragma omp parallel for
for (int i = 0; i < N; i++)
{
v4sf ax = vec_zero;
v4sf ay = vec_zero;
v4sf az = vec_zero;
v4sf *px = (v4sf *) pos[0];
v4sf *py = (v4sf *) pos[1];
v4sf *pz = (v4sf *) pos[2];
v4sf *pmass = (v4sf *) mass;
v4sf x0 = _vec_set_ps1( pos[0][i] );
v4sf y0 = _vec_set_ps1( pos[1][i] );
v4sf z0 = _vec_set_ps1( pos[2][i] );
for ( int j = 0; j < N/4; j++ ) {
bodyBodyInteraction(
ax, ay, az,
x0, y0, z0,
px[j], py[j], pz[j], pmass[j],
_vec_set_ps1( softeningSquared ) );
}
// Accumulate sum of four floats in the AltiVec register
force[0][i] = _vec_sum( ax );
force[1][i] = _vec_sum( ay );
force[2][i] = _vec_sum( az );
}
chTimerGetTime( &end );
return (float) chTimerElapsedTime( &start, &end ) * 1000.0f;
}
void BodySystemCPU::_computeNBodyGravitation()
{
for(int i = 0; i < m_numBodies; ++i)
{
m_force[i*4] = m_force[i*4+1] = m_force[i*4+2] = 0;
for(int j = 0; j < m_numBodies; ++j)
{
float acc[3] = {0, 0, 0};
bodyBodyInteraction(acc, &m_pos[m_currentRead][j*4], &m_pos[m_currentRead][i*4],
m_softeningSquared);
for (int k = 0; k < 3; ++k)
{
m_force[i*4+k] += acc[k];
}
}
}
}
////////////////////////////////////////////////////////////////////////////////
//! Compute reference data set
//! Each element is multiplied with the number of threads / array length
//! @param reference reference data, computed but preallocated
//! @param idata input data as provided to device
//! @param len number of elements in reference / idata
////////////////////////////////////////////////////////////////////////////////
void computeGold( float* force, float* pos, const unsigned int numBodies, float softeningSquared)
{
for(unsigned int i = 0; i < numBodies; ++i)
{
force[i*4 ] = 0;
force[i*4+1] = 0;
force[i*4+2] = 0;
force[i*4+3] = 0;
}
for(unsigned int i = 0; i < numBodies; ++i)
{
for(unsigned int j = 0; j < numBodies; ++j)
{
float f[4];
bodyBodyInteraction(f, &pos[j*4], &pos[i*4], softeningSquared);
for (int k = 0; k < 3; ++k)
{
force[i*4+k] += f[k];
}
}
}
}
