KOKKOS_INLINE_FUNCTION
void OrthPolynomial<1>::generate( /**/ outputViewType output,
const inputViewType input,
const ordinal_type p ) {
typedef outputViewType::value_type value_type;
typedef Sacado::Fad::SFad<value_type,2> fad_type;
constexpr ordinal_type maxCard = (maxOrder+1)*(maxOrder+2)/2;
const ordinal_type
npts = input.dimension(0),
card = output.dimension(0);
// use stack buffer
fad_type inBuf[maxNumPts][2], outBuf[maxCard][maxNumPts];
Kokkos::View<fad_type**, Kokkos::Impl::ActiveExecutionMemorySpace> in(&inBuf[0][0], npts, 2);
Kokkos::View<fad_type***,Kokkos::Impl::ActiveExecutionMemorySpace> out(&outBuf[0][0][0], card, npts);
for (ordinal_type i=0;i<npts;++i)
for (ordinal_type j=0;j<2;++j) {
in(i,j) = Sacado::Fad::SFad<value_type,2>( input(i,j) );
in(i,j).diff(j,2);
}
OrthPolynomial<0>::generate<maxOrder,maxNumPts>(out, in, p);
for (ordinal_type i=0;i<card;++i)
for (ordinal_type j=0;j<npts;++j)
for (ordinal_type k=0;k<2;++k)
output(i,j,k) = out(i,j).dx(k);
}
KOKKOS_INLINE_FUNCTION
void
Basis_HGRAD_LINE_Cn_FEM::Serial<opType>::
getValues( outputViewType output,
const inputViewType input,
workViewType work,
const vinvViewType vinv,
const ordinal_type operatorDn ) {
ordinal_type opDn = operatorDn;
const ordinal_type card = vinv.dimension(0);
const ordinal_type npts = input.dimension(0);
const ordinal_type order = card - 1;
const double alpha = 0.0, beta = 0.0;
typedef typename Kokkos::DynRankView<typename workViewType::value_type, typename workViewType::memory_space> viewType;
auto vcprop = Kokkos::common_view_alloc_prop(work);
switch (opType) {
case OPERATOR_VALUE: {
viewType phis(Kokkos::view_wrap(work.data(), vcprop), card, npts);
Impl::Basis_HGRAD_LINE_Cn_FEM_JACOBI::
Serial<opType>::getValues(phis, input, order, alpha, beta);
for (ordinal_type i=0;i<card;++i)
for (ordinal_type j=0;j<npts;++j) {
output(i,j) = 0.0;
for (ordinal_type k=0;k<card;++k)
output(i,j) += vinv(k,i)*phis(k,j);
}
break;
}
case OPERATOR_GRAD:
case OPERATOR_D1:
case OPERATOR_D2:
case OPERATOR_D3:
case OPERATOR_D4:
case OPERATOR_D5:
case OPERATOR_D6:
case OPERATOR_D7:
case OPERATOR_D8:
case OPERATOR_D9:
case OPERATOR_D10:
opDn = getOperatorOrder(opType);
case OPERATOR_Dn: {
// dkcard is always 1 for 1D element
const ordinal_type dkcard = 1;
viewType phis(Kokkos::view_wrap(work.data(), vcprop), card, npts, dkcard);
Impl::Basis_HGRAD_LINE_Cn_FEM_JACOBI::
Serial<opType>::getValues(phis, input, order, alpha, beta, opDn);
for (ordinal_type i=0;i<card;++i)
for (ordinal_type j=0;j<npts;++j)
for (ordinal_type k=0;k<dkcard;++k) {
output(i,j,k) = 0.0;
for (ordinal_type l=0;l<card;++l)
output(i,j,k) += vinv(l,i)*phis(l,j,k);
}
break;
}
default: {
INTREPID2_TEST_FOR_ABORT( true,
">>> ERROR: (Intrepid2::Basis_HGRAD_LINE_Cn_FEM::Serial::getValues) operator is not supported." );
}
}
}
KOKKOS_INLINE_FUNCTION
void OrthPolynomial<0>::generate( /**/ outputViewType output,
const inputViewType input,
const ordinal_type order ) {
typedef outputViewType::value_type value_type;
const ordinal_type
npts = input.dimension(0);
const auto z = input;
// each point needs to be transformed from Pavel's element
// z(i,0) --> (2.0 * z(i,0) - 1.0)
// z(i,1) --> (2.0 * z(i,1) - 1.0)
auto idx = [](const ordinal_type p,
const ordinal_type q) {
return (p+q)*(p+q+1)/2+q;
};
auto jrc = [](const value_type alpha,
const value_type beta ,
const ordinal_type n ,
/**/ value_type &an,
/**/ value_type &bn,
/**/ value_type &cn) {
an = ( (2.0 * n + 1.0 + alpha + beta) * ( 2.0 * n + 2.0 + alpha + beta ) /
(2.0 * ( n + 1 ) * ( n + 1 + alpha + beta ) ) );
bn = ( (alpha*alpha-beta*beta)*(2.0*n+1.0+alpha+beta) /
(2.0*(n+1.0)*(2.0*n+alpha+beta)*(n+1.0+alpha+beta) ) );
cn = ( (n+alpha)*(n+beta)*(2.0*n+2.0+alpha+beta) /
( (n+1.0)*(n+1.0+alpha+beta)*(2.0*n+alpha+beta) ) );
}
// set D^{0,0} = 1.0
{
const ordinal_type loc = idx(0,0);
for (ordinal_type i=0;i<npts;++i)
output(loc, i) = 1.0 + z(i,0) - z(i,0) + z(i,1) - z(i,1);
}
if (p > 0) {
value_type f1[maxNumPts],f2[maxNumPts],f3[maxNumPts];
for (ordinal_type i=0;i<npts;++i) {
f1[i] = 0.5 * (1.0+2.0*(2.0*z(i,0)-1.0)+(2.0*z(i,1)-1.0));
f2[i] = 0.5 * (1.0-(2.0*z(i,1)-1.0));
f3[i] = f2[i] * f2[i];
}
// set D^{1,0} = f1
{
const ordinal_type loc = idx(1,0);
for (ordinal_type i=0;i<npts;++i)
output(loc, i) = f1[i];
}
// recurrence in p
for (ordinal_type p=1;p<order;p++) {
const ordinal_type
loc = idx(p,0),
loc_p1 = idx(p+1,0),
loc_m1 = idx(p-1,0);
const value_type
a = (2.0*p+1.0)/(1.0+p),
b = p / (p+1.0);
for (ordinal_type i=0;i<npts;++i)
output(loc_p1,i) = ( a * f1[i] * output(loc,i) -
b * f3[i] * output(loc_m1,i) );
}
// D^{p,1}
for (ordinal_type p=0;p<order;++p) {
const ordinal_type
loc_p_0 = idx(p,0),
loc_p_1 = idx(p,1);
for (ordinal_type i=0;i<npts;++i)
output(loc_p_1,i) = output(loc_p_0,i)*0.5*(1.0+2.0*p+(3.0+2.0*p)*(2.0*z(i,1)-1.0));
}
// recurrence in q
for (ordinal_type p=0;p<order-1;++p)
for (ordinal_type q=1;q<order-p;++q) {
const ordinal_type
loc_p_qp1 = idx(p,q+1),
loc_p_q = idx(p,q),
loc_p_qm1 = idx(p,q-1);
value_type a,b,c;
jrc((value_type)(2*p+1),(value_type)0,q,a,b,c);
for (ordinal_type i=0;i<npts;++i)
output(loc_p_qp1,i) = ( (a*(2.0*z(i,1)-1.0)+b)*outputValues(loc_p_q,i)
- c*outputValues(loc_p_qm1,i) );
}
}
// orthogonalize
for (ordinal_type p=0;p<=order;++p)
for (ordinal_type q=0;q<=order-p;++q)
for (ordinal_type i=0;i<npts;++i)
output(idx(p,q),i) *= sqrt( (p+0.5)*(p+q+1.0));
}
