Struct TritonLowering 

pub struct TritonLowering { /* private fields */ }

Implementations

impl TritonLowering

pub fn new() -> Self

pub fn with_sm_count(self, sm_count: Option<u32>) -> Self

Enables shape-adaptive tile-size selection for conv kernels, targeting sm_count thread blocks worth of parallelism per launch (see pick_adaptive_block_n). None restores the default fixed-tile-size behavior.

impl TritonLowering

pub fn lower_with_mapping( &self, graph: &Graph, mode: LoweringMode, ) -> Result<(Dag<Box<dyn ExecutableOp>>, Vec<usize>)>

Like lower but also returns the graph-node-index → DAG-node-index mapping. Useful for middleware lowerings that need to patch specific DAG nodes after the base lowering runs.

Trait Implementations

impl Debug for TritonLowering

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

Formats the value using the given formatter.

impl Default for TritonLowering

fn default() -> TritonLowering

Returns the “default value” for a type.

impl<'a> Lowering<'a> for TritonLowering

fn extra_dag_names( &self, graph: &Graph, graph_to_dag: &[usize], ) -> Vec<(usize, String)>

Conv2d-with-bias splits one graph node into two DAG nodes (conv + biasadd). graph_to_dag[graph_idx] already points at the biasadd DAG node; here we propagate the same name to the conv DAG node (biasadd_dag_idx - 1) so that the conv weight parameter can be loaded under the same name prefix.

fn lower( &self, graph: &Graph, mode: LoweringMode, ) -> Result<Dag<Box<dyn ExecutableOp>>>

Lowers graph for mode.

fn lower_with_mapping( &self, graph: &Graph, mode: LoweringMode, ) -> Result<(Dag<Box<dyn ExecutableOp>>, Vec<usize>)>

Like [lower] but also returns a graph_node_idx → dag_node_idx mapping so that graph-level metadata (e.g. names) can be propagated into the compiled DAG.

fn base_lowering(&self) -> Option<&dyn Lowering<'a>>

Returns the next lowering in a middleware chain, or None if this is the final lowering. A custom lowering can call self.base_lowering() to delegate ops it does not handle.