static BOOST_FORCEINLINE Vector sincos( input_t const & input, Vector & cosine )
{
typedef typename Vector::value_type scalar_t;
long double const omega_scale( input.omega_scale );
long double const N ( input.N );
int const index ( input.index );
Vector sine;
for ( int i( 0 ); i < Vector::static_size; ++i )
{
//...zzz...should only 'upgrade' float to double and double to long double...
//...zzz...nt2 doesn't seem to support the long double data type...
/*long*/ double const omega( ( index + i ) * omega_scale * full_circle() / N );
/*long*/ double precise_sin;
/*long*/ double precise_cos;
nt2::sincospi( omega, precise_sin, precise_cos );
sine [ i ] = static_cast<scalar_t>( precise_sin );
cosine[ i ] = static_cast<scalar_t>( precise_cos );
}
return sine;
}
// intensity different in node link
double dynamics_mul::intensity_energy(Node_key nid, Vec2 new_pos){
double E = 0.0;
for (auto hew = s_dsc->walker(nid); !hew.full_circle(); hew = hew.circulate_vertex_cw()) {
auto fid = hew.face();
Vec2_array tris;
tris.push_back(new_pos);
tris.push_back(s_dsc->get_pos(hew.vertex()));
tris.push_back(s_dsc->get_pos(hew.next().vertex()));
double ci = mean_inten_[s_dsc->get_label(fid)];
Vec2 min(INFINITY, INFINITY), max(-INFINITY, -INFINITY);
for (auto p: tris){
min[0] = std::min(min[0], p[0]);
min[1] = std::min(min[1], p[1]);
max[0] = std::max(max[0], p[0]);
max[1] = std::max(max[1], p[1]);
}
for (int i = floor(min[0]); i < ceil(max[0]); i++) {
for (int j = floor(min[1]); j < ceil(max[1]); j++) {
if (helper_t::is_point_in_tri(Vec2(i,j), tris)) {
double I = s_img->get_intensity(i, j);
E += (I - ci)*(I-ci);
}
}
}
}
return E;
}
// Interface length in node link
double dynamics_mul::curve_length(Node_key nid, Vec2 new_pos){
double L = 0.0;
for (auto hew = s_dsc->walker(nid); !hew.full_circle(); hew = hew.circulate_vertex_cw()) {
if (s_dsc->is_interface(hew.halfedge())) {
L += (s_dsc->get_pos(hew.vertex()) - new_pos).length();
}
}
return L;
}
static BOOST_FORCEINLINE Vector sincos( input_t const & input, Vector & cosine )
{
Vector sine;
#if defined( _MSC_VER ) && defined( _M_IX86 )
Vector::value_type * const p_sine ( sine .data() );
Vector::value_type * const p_cosine( cosine.data() );
// http://software.intel.com/en-us/forums/showthread.php?t=74354
// http://www.devmaster.net/forums/showthread.php?t=5784
unsigned int const vector_size( Vector::static_size );
long double const omega_scale( input.omega_scale );
long double const N ( input.N );
unsigned int local_index( input.index );
__asm
{
mov ecx, vector_size
mov edx, p_sine
mov edi, p_cosine
sincos_loop:
fldpi
fldpi
fadd
//...zzz...direct version seems to work the same?
//fld [omega_scale]
//fmul
//fild [local_index]
//fmul
//fld [N]
//fdiv
fmul [omega_scale]
fimul [local_index]
fdiv [N]
fsincos
fstp [edi]
add edi, 4
fstp [edx]
add edx, 4
inc [local_index]
dec ecx
jnz sincos_loop
}
#else // 32 bit x86 MSVC
long double const omega_scale( input.omega_scale );
long double const N ( input.N );
int const index ( input.index );
for ( unsigned i( 0 ); i < Vector::static_size; ++i )
{
long double const omega( ( index + i ) * omega_scale * full_circle() / N );
sine [ i ] = std::sin( omega );
cosine[ i ] = std::cos( omega );
}
#endif // 32 bit x86 MSVC
return sine;
}