Hermite kyle davide

src/RadialBasisFunctions.jl

@@ -26,11 +26,11 @@ include("operators/operators.jl")
 export RadialBasisOperator, ScalarValuedOperator, VectorValuedOperator
 export update_weights!, is_cache_valid
 
-include("solve_utils.jl")
-
-# Boundary types needed by solve.jl
+# Boundary types needed by solve.jl and solve_utils.jl
 include("boundary_types.jl")
 
+include("solve_utils.jl")
+
 include("solve.jl")
 
 # New clean Hermite implementation

src/boundary_types.jl

@@ -6,10 +6,10 @@ Boundary condition types and utilities for Hermite interpolation.
 struct BoundaryCondition{T<:Real}
     α::T
     β::T
-    
+
     function BoundaryCondition(α::A, β::B) where {A,B}
         T = promote_type(A, B)
-        new{T}(α, β)
+        return new{T}(α, β)
     end
 end
 
@@ -19,7 +19,7 @@ end
 
 # Predicate functions
 is_dirichlet(bc::BoundaryCondition) = isone(bc.α) && iszero(bc.β)
-is_neumann(bc::BoundaryCondition) = iszero(bc.α) && isone(bc.β) 
+is_neumann(bc::BoundaryCondition) = iszero(bc.α) && isone(bc.β)
 is_robin(bc::BoundaryCondition) = !iszero(bc.α) && !iszero(bc.β)
 
 # Constructor helpers
@@ -28,30 +28,102 @@ Neumann(::Type{T}=Float64) where {T<:Real} = BoundaryCondition(zero(T), one(T))
 Robin(α::Real, β::Real) = BoundaryCondition(α, β)
 
 # Boundary information for a local stencil
-struct HermiteBoundaryInfo{T<:Real}
+"""
+This struct is meant to be used to correctly broadcast the build_stencil() function
+When Hermite scheme is used, it can be given to _build_stencil!() in place of the sole data.
+"""
+struct HermiteStencilData{T<:Real}
+    data::AbstractVector{Vector{T}}  # Coordinates of stencil points
     is_boundary::Vector{Bool}
     boundary_conditions::Vector{BoundaryCondition{T}}
     normals::Vector{Vector{T}}
-
-    function HermiteBoundaryInfo(
+
+    function HermiteStencilData(
+        data::AbstractVector{Vector{T}},
         is_boundary::Vector{Bool},
         boundary_conditions::Vector{BoundaryCondition{T}},
-        normals::Vector{Vector{T}}
+        normals::Vector{Vector{T}},
     ) where {T<:Real}
-        @assert length(is_boundary) == length(boundary_conditions) == length(normals)
-        new{T}(is_boundary, boundary_conditions, normals)
+        @assert length(data) ==
+            length(is_boundary) ==
+            length(boundary_conditions) ==
+            length(normals)
+        return new{T}(data, is_boundary, boundary_conditions, normals)
+    end
+end
+
+#pre-allocation constructor
+function HermiteStencilData{T}(k::Int, dim::Int) where {T<:Real}
+    data = [Vector{T}(undef, dim) for _ in 1:k]  # Pre-allocate with correct dimension
+    is_boundary = falses(k)
+    boundary_conditions = [Dirichlet(T) for _ in 1:k]
+    normals = [Vector{T}(undef, dim) for _ in 1:k]  # Pre-allocate with correct dimension
+    return HermiteStencilData(data, is_boundary, boundary_conditions, normals)
+end
+
+"""
+Populate local boundary information for a specific stencil within a kernel.
+This function extracts boundary data for the neighbors of eval_idx and fills 
+the pre-allocated HermiteBoundaryInfo structure.
+
+# Arguments
+- `boundary_info`: Pre-allocated HermiteBoundaryInfo structure to fill (for this batch)
+- `eval_idx`: Current evaluation point index
+- `adjl`: Adjacency list for eval_idx (the neighbors)  
+- `is_boundary`: Global is_boundary vector for all points
+- `boundary_conditions`: Global boundary_conditions vector for all points
+- `normals`: Global normals vector for all points
+"""
+function update_stencil_data!(
+    hermite_data::HermiteStencilData{T},  # Pre-allocated structure passed in
+    global_data::AbstractVector{Vector{T}},
+    neighbors::Vector{Int},  # adjl[eval_idx]
+    is_boundary::Vector{Bool},
+    boundary_conditions::Vector{BoundaryCondition{T}},
+    normals::Vector{Vector{T}},
+    global_to_boundary::Vector{Int},
+) where {T}
+    k = length(neighbors)
+
+    # Fill local boundary info for each neighbor (in-place, no allocation)
+    @inbounds for local_idx in 1:k
+        global_idx = neighbors[local_idx]
+        hermite_data.data[local_idx] .= global_data[global_idx]
+        hermite_data.is_boundary[local_idx] = is_boundary[global_idx]
+
+        if is_boundary[global_idx]
+            boundary_idx = global_to_boundary[global_idx]
+            hermite_data.boundary_conditions[local_idx] = boundary_conditions[boundary_idx]
+            hermite_data.normals[local_idx] .= normals[boundary_idx]
+        else
+            # Set default Dirichlet for interior points (not used but keeps type consistency)
+            hermite_data.boundary_conditions[local_idx] = Dirichlet(T)
+            fill!(hermite_data.normals[local_idx], zero(T))
+        end
     end
+
+    return nothing
 end
 
 # Trait types for dispatch
 abstract type StencilType end
 struct StandardStencil <: StencilType end
+struct DirichletStencil <: StencilType end
 struct HermiteStencil <: StencilType end
 
-# Trait function to determine stencil type
-stencil_type(boundary_info::Nothing) = StandardStencil()
-stencil_type(boundary_info::HermiteBoundaryInfo) = any(boundary_info.is_boundary) ? HermiteStencil() : StandardStencil()
-
-# Convenience function to check if any boundary points in stencil
-has_boundary_points(boundary_info::Nothing) = false
-has_boundary_points(boundary_info::HermiteBoundaryInfo) = any(boundary_info.is_boundary)
+function stencil_type(
+    is_boundary::Vector{Bool},
+    boundary_conditions::Vector{BoundaryCondition},
+    eval_idx::Int,
+    neighbors::Vector{Int},
+    global_to_boundary::Vector{Int},
+)
+    if sum(is_boundary[neighbors]) == 0
+        return StandardStencil()
+    elseif is_boundary[eval_idx] &&
+        is_dirichlet(boundary_conditions[global_to_boundary[eval_idx]])
+        return DirichletStencil()
+    else
+        return HermiteStencil()
+    end
+end

src/interpolation.jl

@@ -25,7 +25,7 @@ function Interpolator(x, y, basis::B=PHS()) where {B<:AbstractRadialBasis}
     mon = MonomialBasis(dim, basis.poly_deg)
     data_type = promote_type(eltype(first(x)), eltype(y))
     A = Symmetric(zeros(data_type, n, n))
-    _build_collocation_matrix!(A, x, basis, mon, k, StandardStencil())
+    _build_collocation_matrix!(A, x, basis, mon, k)
     b = data_type[i < k ? y[i] : 0 for i in 1:n]
     w = A \ b
     return Interpolator(x, y, w[1:k], w[(k + 1):end], basis, mon)

src/solve.jl

@@ -20,92 +20,20 @@ end
 function _build_weights(
     data, eval_points, adjl, basis, ℒrbf, ℒmon, mon; batch_size=10, device=CPU()
 )
-    TD = eltype(first(data))
-    dim = length(first(data)) # dimension of data
-    nmon = binomial(dim + basis.poly_deg, basis.poly_deg)
-    k = length(first(adjl))  # number of data in influence/support domain
-    sizes = (k, nmon)
-
-    # allocate arrays to build sparse matrix
-    Na = length(adjl)
-    I = zeros(Int, k * Na)
-    J = reduce(vcat, adjl)
-    V = zeros(TD, k * Na, _num_ops(ℒrbf))
-
-    # Calculate number of batches
-    n_batches = ceil(Int, Na / batch_size)
-
-    # Create kernel for building stencils in batches
-    @kernel function build_stencils_kernel(
-        I, J, V, data, eval_points, adjl, basis, ℒrbf, ℒmon, mon, k, batch_size, Na
-    )
-        # Get the batch index for this thread
-        batch_idx = @index(Global)
-
-        # Calculate the range of points for this batch
-        start_idx = (batch_idx - 1) * batch_size + 1
-        end_idx = min(batch_idx * batch_size, Na)
-
-        # Pre-allocate work arrays for this thread
-        n = k + nmon
-        A = Symmetric(zeros(TD, n, n), :U)
-        b = _prepare_b(ℒrbf, TD, n)
-
-        # Process each point in the batch sequentially
-        for i in start_idx:end_idx
-            # Set row indices for sparse matrix
-            for idx in 1:k
-                I[(i - 1) * k + idx] = i
-            end
-
-            # Get data points in the influence domain
-            local_data = [data[j] for j in adjl[i]]
-
-            # Build stencil and store in global weight matrix
-            stencil = _build_stencil!(
-                A, b, ℒrbf, ℒmon, local_data, eval_points[i], basis, mon, k, StandardStencil()
-            )
-
-            # Store the stencil weights in the value array
-            for op in axes(V, 2)
-                for idx in 1:k
-                    V[(i - 1) * k + idx, op] = stencil[idx, op]
-                end
-            end
-        end
-    end
-
-    # Launch kernel with one thread per batch
-    kernel = build_stencils_kernel(device)
-    kernel(
-        I,
-        J,
-        V,
+    # Use the unified kernel infrastructure with standard allocation strategy
+    return _build_weights_unified(
+        StandardAllocation(),
         data,
         eval_points,
         adjl,
         basis,
         ℒrbf,
         ℒmon,
         mon,
-        k,
-        batch_size,
-        Na;
-        ndrange=n_batches,
-        workgroupsize=1,
+        nothing;
+        batch_size=batch_size,
+        device=device,
     )
-
-    # Wait for kernel to complete
-    KernelAbstractions.synchronize(device)
-
-    # Create and return sparse matrix/matrices
-    nrows = length(adjl)
-    ncols = length(data)
-    if size(V, 2) == 1
-        return sparse(I, J, V[:, 1], nrows, ncols)
-    else
-        return ntuple(i -> sparse(I, J, V[:, i], nrows, ncols), size(V, 2))
-    end
 end
 
 function _build_stencil!(
@@ -118,15 +46,14 @@ function _build_stencil!(
     basis::B,
     mon::MonomialBasis,
     k::Int,
-    ::StandardStencil
 ) where {TD,TE,B<:AbstractRadialBasis}
-    _build_collocation_matrix!(A, data, basis, mon, k, StandardStencil())
-    _build_rhs!(b, ℒrbf, ℒmon, data, eval_point, basis, k, StandardStencil())
+    _build_collocation_matrix!(A, data, basis, mon, k)
+    _build_rhs!(b, ℒrbf, ℒmon, data, eval_point, basis, k)
     return (A \ b)[1:k, :]
 end
 
 function _build_collocation_matrix!(
-    A::Symmetric, data::AbstractVector, basis::B, mon::MonomialBasis{Dim,Deg}, k::K, ::StandardStencil
+    A::Symmetric, data::AbstractVector, basis::B, mon::MonomialBasis{Dim,Deg}, k::K
 ) where {B<:AbstractRadialBasis,K<:Int,Dim,Deg}
     # radial basis section
     AA = parent(A)
@@ -147,7 +74,7 @@ function _build_collocation_matrix!(
 end
 
 function _build_rhs!(
-    b, ℒrbf, ℒmon, data::AbstractVector{TD}, eval_point::TE, basis::B, k::K, ::StandardStencil
+    b, ℒrbf, ℒmon, data::AbstractVector{TD}, eval_point::TE, basis::B, k::K
 ) where {TD,TE,B<:AbstractRadialBasis,K<:Int}
     # radial basis section
     @inbounds for i in eachindex(data)
@@ -165,7 +92,7 @@ function _build_rhs!(
 end
 
 function _build_rhs!(
-    b, ℒrbf::Tuple, ℒmon::Tuple, data::AbstractVector{TD}, eval_point::TE, basis::B, k::K, ::StandardStencil
+    b, ℒrbf::Tuple, ℒmon::Tuple, data::AbstractVector{TD}, eval_point::TE, basis::B, k::K
 ) where {TD,TE,B<:AbstractRadialBasis,K<:Int}
     @assert size(b, 2) == length(ℒrbf) == length(ℒmon) "b, ℒrbf, ℒmon must have the same length"
     # radial basis section