run_query_kernel
run_query_kernel is used to launch a device kernel that operates on a mesh element and its local neighborhood. It takes a query type (e.g., Op::EV), which specifies what neighborhood relation to access—such as "incident vertices of an edge" or "adjacent faces of a face". This is the simplest entry point in RXMesh for neighborhood-based computations. The neighborhood is exposed through an Iterator whose type depends on the query's output element—for example, a vertex one-ring query (VV) yields a VertexIterator.
Supported Queries Types
Vertex Queries
| Query | Description |
|---|---|
VV 🟢 ➔ 🟢 |
For vertex V, return its adjacent vertices |
VE 🟢 ➔ ➖ |
For vertex V, return its incident edges |
VF 🟢 ➔ 🔺 |
For vertex V, return its incident faces |
Edge Queries
| Query | Description |
|---|---|
EV ➖ ➔ 🟢 |
For edge E, return its incident vertices |
EF ➖ ➔ 🔺 |
For edge E, return its incident faces |
EVDiamond ➖ ➔ 🟢 |
For edge E, return its incident and opposite vertices |
Face Queries
| Query | Description |
|---|---|
FV 🔺➔ 🟢 |
For face F, return its incident vertices |
FE 🔺➔ ➖ |
For face F, return its incident edges |
FF 🔺➔ 🔺 |
For face F, return its adjacent faces |
Usage
run_query_kernel comes in two variants:
- A simpler overload that automatically prepares the LaunchBox.
- A variant that accepts a user-supplied LaunchBox if you want to manually set up the kernel configuration.
Both require:
- The query type as a template parameter (e.g.,
Op::EV) - The CUDA block size
- A lambda function that accepts a mesh element and an iterator over its neighbors.
Lambdas must be annotated __device__ and capture variables by value.
Example: Computing Edge Length
This example computes the squared rest length for each edge using the EV query.
RXMeshStatic rx("mesh.obj");
auto x = *rx.get_input_vertex_coordinates(); // vertex positions
auto len = *rx.add_edge_attribute<T>("eLength", 1);
constexpr int blockSize = 256;
rx.run_query_kernel<Op::EV, blockSize>(
[=] device(const EdgeHandle& eh, const VertexIterator& iter) {
Eigen::Vector3<T> a = x.to_eigen<3>(iter[0]);
Eigen::Vector3<T> b = x.to_eigen<3>(iter[1]);
len(eh) = (a - b).squaredNorm();
});
See Attribute Management to learn how to create and manipulate attributes like
len.