double transform(double* ptr, size_t size)
{
size_t num_wg = 64;
size_t num_compute_units = 6;
size_t global_size = num_wg * num_compute_units;
size_t workgroup_size = 256;
m_result.resize( global_size);
m_local_dispatch->clearArgs();
FIX_ARGS_STABLE(m_local_dispatch);
m_local_dispatch->pushPointerArg((void*)ptr);
m_local_dispatch->pushIntArg((int)size );
m_local_dispatch->pushPointerArg((void*)&m_result[0]);
size_t global_dims[3] = { std::min(roundUp(size, workgroup_size), global_size*workgroup_size),1,1};
size_t local_dims[3] = {workgroup_size,1,1};
m_local_dispatch->setLaunchAttributes(1, global_dims, local_dims);
m_local_dispatch->dispatchKernelWaitComplete();
if (size < workgroup_size)
{
return m_result[0];
}
return reduceTail(size);
}
// reduce all terms in f by the leading terms of all polynomials in F
// first reduce the leading term completely, then the lower terms
bool reduce(BRP &f, const IntermediateBasis &F, const IntermediateBasis::const_iterator itF) {
bool ret = false;
ret = reduceLt(f, F, itF);
if (!f.isZero() ) {
if ( reduceTail(f, F, itF) ) {
ret = true;
}
}
return ret;
}