pub fn execute_compiled(
compiled: &CompiledNode,
inputs: &[Arc<RwLock<Storage>>],
outputs: &mut [Arc<RwLock<Storage>>],
layouts: &[Layout],
) -> Result<CompletionHandle>Expand description
Run a compiled node against the given inputs/outputs. The output
Storages must be pre-allocated (the executor’s responsibility,
using the node’s shape + dtype).
layouts parallels inputs.append(outputs) and carries the
stride/shape/offset metadata that stride-aware kernels consume.
Today’s contiguous-only kernels ignore it; the executor’s
auto-Contiguize pass guarantees every input layout is
Layout::contiguous(shape) before this is called.
Production-correct: surfaces kernel errors as Result; never
panics on dispatch mismatch (the wrapper functions handle that).