tvm_tirx.backend.cuda.tile.exec_context

ExecContext: per-program-point active-thread state.

The active thread set is represented as a TileLayout: active axes live in layout.shard and per-axis lower bounds live in layout.offset. Filters narrow that layout; scope switches derive the current inter/intra view.

Classes

ExecContextError	Raised on structural violations of the ExecContext model.
AxisRange(extent[, offset, stride])	An active slice offset + stride * [0, extent) on one TileLayout axis.
ActiveSet(layout)	Active thread set represented by a TileLayout.
LaneBinding(axis, kind, declared_extent)	Resolution of a user-declared CUDA index binding to one active-set axis.
Split(inter, intra)	A scope_switch split of A.
ExecContext(A, scope_kind, inter, intra)	Per-program-point compiler state: active set + scope kind + split.

Functions

initial_A(*[, lane_ext, cta_ext])	Build A at Function device entry: all threads active, offsets all zero.
filter_narrow(A, binding, lo, hi)	Intersect A's binding axis with [lo, hi).
filter_modulo(A, axis, modulus, residue)	Intersect an active-set axis with axis % modulus == residue.
scope_switch(A, scope_kind)	Split A into (inter, intra) for the target scope kind.