KOKKOS_INLINE_FUNCTION
void operator() (device_type device) const
{
typename local_view_type::Partition part( device.team_rank() , device.team_size() );
const local_view_type local_x( dev_x , part );
const local_view_type local_y( dev_y , part );
const int element = device.league_rank();
// Apply evaluation function to this thread's fix-sized UQ sample set.
simple_function< value_type >( local_x(element) , local_y(element) );
// Print x and y
if (print) {
device.team_barrier();
if ( ! device.league_rank() && ! device.team_rank() ) {
printf("view_kernel league(%d:%d) team_size(%d) dim(%d) size(%d)\n",
device.league_rank(), device.league_size(),device.team_size(),
int(dev_x.dimension_1()), int(local_x(element).size()) );
}
if ( ! device.team_rank() ) {
printf("x(%i) = { ",element);
for (int sample=0; sample< int(dev_x.dimension_1()); ++sample) {
printf("%g ", dev_x(element,sample));
}
printf("}\n\n");
printf("y(%i) = { ",element);
for (int sample=0; sample< int(dev_y.dimension_1()); ++sample) {
printf("%g ", dev_y(element,sample));
}
printf("}\n\n");
}
device.team_barrier();
}
}
KOKKOS_INLINE_FUNCTION
void operator() (device_type device) const {
int element = device.league_rank();
int num_threads = device.team_size();
int thread = device.team_rank();
int num_samples = dev_x.dimension_1();
int num_samples_per_thread = num_samples / num_threads;
// Initialize x
storage_type x_s(&dev_x(element,thread),
num_samples_per_thread,
num_threads);
storage_type y_s(&dev_y(element,thread),
num_samples_per_thread,
num_threads);
array_vector_type x(x_s), y(y_s);
simple_function<scalar_vector_type>(x,y);
// Print x and y
if (print) {
for (int tidx = 0; tidx<num_threads; tidx++) {
if (thread == tidx) {
printf("x(%i) = [ ",tidx);
for (int sample=0; sample<num_samples_per_thread; sample++)
printf("%g ", x.coeff(sample));
printf("]\n\n");
}
device.team_barrier();
}
for (int tidx = 0; tidx<num_threads; tidx++) {
if (thread == tidx) {
printf("y(%i) = [ ",tidx);
for (int sample=0; sample<num_samples_per_thread; sample++)
printf("%g ", y.coeff(sample));
printf("]\n\n");
}
device.team_barrier();
}
}
}
KOKKOS_INLINE_FUNCTION
void operator() (device_type device) const {
int element = device.league_rank();
int num_threads = device.team_size();
int thread = device.team_rank();
int num_samples = dev_x.dimension_1();
scalar_type x, y;
for (int sample=thread; sample<num_samples; sample+=num_threads) {
// Initialize x
x = dev_x(element, sample);
// Compute function
simple_function<scalar_type>(x,y);
// Return result
dev_y(element, sample) = y;
}
}
KOKKOS_INLINE_FUNCTION
void operator() (device_type device) const {
int element = device.league_rank();
int num_threads = device.team_size();
int thread = device.team_rank();
int num_samples = dev_x.dimension_1();
int num_samples_per_thread = num_samples / num_threads;
// multi-point expansions
array_vector_type x(num_samples_per_thread, 0.0), y(num_samples_per_thread, 0.0);
// Initialize x
if (reset && storage_type::supports_reset) {
storage_type& x_s = x.storage();
storage_type& y_s = y.storage();
x_s.shallowReset(&dev_x(element,thread),
num_samples_per_thread,
num_threads,
false);
y_s.shallowReset(&dev_y(element,thread),
num_samples_per_thread,
num_threads,
false);
}
else {
for (int sample=0; sample<num_samples_per_thread; ++sample)
x.fastAccessCoeff(sample) = dev_x(element,thread+sample*num_threads);
}
simple_function<scalar_vector_type>(x,y);
// Print x and y
if (print) {
for (int tidx = 0; tidx<num_threads; tidx++) {
if (thread == tidx) {
printf("x(%i) = [ ",tidx);
for (int sample=0; sample<num_samples_per_thread; sample++)
printf("%g ", x.coeff(sample));
printf("]\n\n");
}
device.team_barrier();
}
for (int tidx = 0; tidx<num_threads; tidx++) {
if (thread == tidx) {
printf("y(%i) = [ ",tidx);
for (int sample=0; sample<num_samples_per_thread; sample++)
printf("%g ", y.coeff(sample));
printf("]\n\n");
}
device.team_barrier();
}
}
// Return result
if (!(reset && storage_type::supports_reset)) {
for (int sample=0; sample<num_samples_per_thread; ++sample)
dev_y(element,thread+sample*num_threads) = y.fastAccessCoeff(sample);
}
}
