bool _compareResults(int numBodies)
{
assert(m_nbodyCuda);
bool passed = true;
m_nbody->update(0.001f);
{
m_nbodyCpu = new BodySystemCPU<T>(numBodies);
m_nbodyCpu->setArray(BODYSYSTEM_POSITION, m_hPos);
m_nbodyCpu->setArray(BODYSYSTEM_VELOCITY, m_hVel);
m_nbodyCpu->update(0.001f);
T *cudaPos = m_nbodyCuda->getArray(BODYSYSTEM_POSITION);
T *cpuPos = m_nbodyCpu->getArray(BODYSYSTEM_POSITION);
T tolerance = 0.0005f;
for (int i = 0; i < numBodies; i++)
{
if (fabs(cpuPos[i] - cudaPos[i]) > tolerance)
{
passed = false;
printf("Error: (host)%f != (device)%f\n", cpuPos[i], cudaPos[i]);
}
}
}
return passed;
}
//*****************************************************************************
bool CompareResults(int numBodies)
{
// Run computation on the device/GPU
shrLog(" Computing on the Device / GPU...\n");
nbodyGPU->update(0.001f);
nbodyGPU->synchronizeThreads();
// Write out device/GPU data file for regression analysis
shrLog(" Writing out Device/GPU data file for analysis...\n");
float* fGPUData = nbodyGPU->getArray(BodySystem::BODYSYSTEM_POSITION);
shrWriteFilef( "oclNbody_Regression.dat", fGPUData, numBodies, 0.0, false);
// Run computation on the host CPU
shrLog(" Computing on the Host / CPU...\n\n");
BodySystemCPU* nbodyCPU = new BodySystemCPU(numBodies);
nbodyCPU->setArray(BodySystem::BODYSYSTEM_POSITION, hPos);
nbodyCPU->setArray(BodySystem::BODYSYSTEM_VELOCITY, hVel);
nbodyCPU->update(0.001f);
// Check if result matches
shrBOOL bMatch = shrComparefe(fGPUData,
nbodyGPU->getArray(BodySystem::BODYSYSTEM_POSITION),
numBodies, .001f);
shrLog("Results %s\n\n", (shrTRUE == bMatch) ? "Match" : "do not match!");
// Cleanup local allocation
if(nbodyCPU)delete nbodyCPU;
return (shrTRUE == bMatch);
}