std::vector<float3> gpu_stgauss2_path( int ix, int iy, const cpu_image<float4>& st, float sigma, float max_angle,
bool adaptive, bool st_linear, int order, float step_size )
{
cpu_sampler<float4> st_sampler(st, st_linear? cudaFilterModeLinear : cudaFilterModePoint);
float2 p0 = make_float2(ix + 0.5f, iy + 0.5f);
if (adaptive) {
float A = st2A(st(p0.x, p0.y));
sigma *= 0.25f * (1 + A)*(1 + A);
}
std::deque<float3> C;
float cos_max = cosf(radians(max_angle));
stgauss2_path f(C, sigma);
if (order == 1) st_integrate_euler(p0, st_sampler, f, cos_max, st.w(), st.h(), step_size);
if (order == 2) st_integrate_rk2(p0, st_sampler, f, cos_max, st.w(), st.h(), step_size);
if (order == 4) st_integrate_rk4(p0, st_sampler, f, cos_max, st.w(), st.h(), step_size);
return std::vector<float3>(C.begin(), C.end());
}
std::vector<float3> stgauss3_path_( int ix, int iy, const cpu_image& st, float sigma,
bool st_linear, bool adaptive, bool ustep,
int order, float step_size )
{
cpu_sampler<float3> st_sampler(st, st_linear? cudaFilterModeLinear : cudaFilterModePoint);
float2 p0 = make_float2(ix + 0.5f, iy + 0.5f);
filter_path f(2 * sigma);
if (!ustep) {
if (order == 1) {
if (adaptive)
st3_int<cpu_sampler<float3>,filter_path,1,true>(p0, st_sampler, f, st.w(), st.h(), step_size);
else
st3_int<cpu_sampler<float3>,filter_path,1,false>(p0, st_sampler, f, st.w(), st.h(), step_size);
} else {
if (adaptive)
st3_int<cpu_sampler<float3>,filter_path,2,true>(p0, st_sampler, f, st.w(), st.h(), step_size);
else
st3_int<cpu_sampler<float3>,filter_path,2,false>(p0, st_sampler, f, st.w(), st.h(), step_size);
}
} else {
if (order == 1) {
st3_int_ustep<cpu_sampler<float3>,filter_path,1>(p0, st_sampler, f, st.w(), st.h(), step_size);
} else {
st3_int_ustep<cpu_sampler<float3>,filter_path,2>(p0, st_sampler, f, st.w(), st.h(), step_size);
}
}
return f.path();
}