TorchFEMUtils.h File Reference

Go to the source code of this file.

Namespaces
namespace	TorchFEM
	Small raw-ATen helpers for the batched finite-element assembly used by the Torch FEM kernels.

Functions
at::Tensor	TorchFEM::scatter (const at::Tensor &v, const at::Tensor &dof_map)
	Gather global solution values into element-local DOF values: out[e,a,...] = v[dof_map[e,a,...]].
at::Tensor	TorchFEM::interpolate (const at::Tensor &elem_dofs, const at::Tensor &basis)
	Interpolate element-local DOF values onto quadrature points by contracting over the per-element DOF axis (dim 1) against a basis: out[e,...] = sum_a elem_dofs[e,a] * basis[e,a,...].
at::Tensor	TorchFEM::assemble (const at::Tensor &scattered, const at::Tensor &dof_map, int64_t ndof)
	Scatter-add element-local contributions into a global vector: out[dof_map.flatten()[k]] += scattered.flatten()[k].
at::Tensor	TorchFEM::fullToMandel (const at::Tensor &full)
	Convert a full (symmetric) rank-2 tensor to its 6-component Mandel representation, matching NEML2's SR2 convention: components [00, 11, 22, sqrt(2)*12, sqrt(2)*02, sqrt(2)*01].
at::Tensor	TorchFEM::mandelToFull (const at::Tensor &mandel)
	Inverse of fullToMandel: expand a 6-component Mandel tensor to a full symmetric (..., 3, 3), matching neml2::R2(SR2) (off-diagonals divided by sqrt(2)).