Programming Guide

This guide explains how to write, compile, and inspect TIRx kernels. Start with the CUDA guide for the core language and native programming model, then use tensor layouts and tile primitives to express reusable tile-level work.

The examples use a single alias for the TIRx TVMScript dialect:

from tvm.script import tirx as Tx

Core language constructs are written as Tx.*. Backend-specific helpers and direct CUDA instructions are available through Tx.cuda.* and Tx.ptx.*. Reusable tile operations are written as Tx.tile.*.

• CUDA Programming Guide
• Tensor Layout
• Tile Primitives

For callable reference entries, see the API reference. For lowering, dispatch, and backend implementation details, see Compiler Internals.