Struct GraphScope 

pub struct GraphScope { /* private fields */ }

Owns a traced graph while you build symbolic tensors.

Implementations

impl GraphScope

pub fn new(name: impl Into<String>) -> Self

pub fn graph(&self) -> Graph

pub fn fresh_symbol(&mut self) -> u32

Fresh dynamic-dimension symbol (for variable batch / sequence length).

pub fn input( &mut self, name: impl Into<String>, shape: impl Into<Shape>, ) -> Tensor

pub fn input_dyn( &mut self, name: impl Into<String>, dims: &[Dim], dtype: DType, ) -> Tensor

Input with a caller-chosen dynamic symbol on one axis.

pub fn param( &mut self, name: impl Into<String>, shape: impl Into<Shape>, ) -> Tensor

pub fn scalar(&mut self, value: f64) -> Tensor

Rank-0 literal broadcastable in binary ops.

pub fn constant(&mut self, value: f64, dtype: DType) -> Tensor

pub fn constant_nd( &mut self, data: Vec<f64>, dims: Vec<usize>, dtype: DType, ) -> Tensor

Materialize an N-dimensional constant tensor from row-major data (given as f64, encoded to dtype’s element width). Backs the rlx! array-const form (const mask = [[0,1],[1,0]] : F32;). data.len() must equal the product of dims. Supports F32/F64/I32/I64.

pub fn scan_block( &mut self, init: &Tensor, bcasts: &[&Tensor], length: u32, body: impl FnOnce(&Tensor, &[Tensor]) -> Tensor, ) -> Tensor

Build a bounded sequential scan (Op::Scan) — a single compact loop node, not an unrolled copy. init is the loop carry; bcasts are values held constant across iterations (typically weights); length is the iteration count. body(carry, bcasts) -> next_carry is traced ONCE into a standalone body graph, so the result is O(1) IR regardless of length (unlike an unrolled repeat). The next carry must match the carry’s shape. Backs the rlx! scan construct.

use rlx_tensor::{graph, shape};
let g = graph("rnn", |s| {
    let h0 = s.input("h0", shape![1, 8]);
    let w  = s.param("w", shape![8, 8]);
    s.scan_block(&h0, &[&w], 12, |h, bc| (h.matmul(&bc[0])).tanh())
});
assert_eq!(g.outputs.len(), 1);

pub fn cat(&mut self, tensors: &[&Tensor], axis: usize) -> Tensor

Stack existing tensors along axis (zero-copy at IR level).

pub fn set_outputs( &mut self, outputs: impl IntoIterator<Item = impl Into<NodeId>>, )

pub fn finish(self) -> Graph

Trait Implementations

impl Debug for GraphScope

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.