Python API

Tensor-level intermediate representations, compilation, and runtime execution.

Namespaces

Namespace	Purpose
tvm_tirx.runtime	Tensors, devices, data types, and runtime modules.
tvm_tirx.ir	Shared expressions, types, statements, and IR modules.
tvm_tirx.ir.prim	Primitive expressions and scalar operators.
tvm_tirx.ir.transform	Pass infrastructure and instrumentation.
tvm_tirx.sym	Symbolic arithmetic, analysis, and simplification.
tvm_tirx.script	Python syntax registration for IR dialects.
tvm_tirx.script.parser	Translate Python scripts into IR builder operations.
tvm_tirx.script.ir_builder	Shared IR builder contexts and statements.
tvm_tirx.target	Compilation targets and device selection.
tvm_tirx.tirx	Tensor IR construction and compilation.
tvm_tirx.tirx.op	Tensor IR operators and intrinsic calls.
tvm_tirx.tirx.analysis	Tensor IR analysis and verification.
tvm_tirx.tirx.transform	Tensor IR transformation passes.
tvm_tirx.tirx.layout	Tensor layouts and logical-to-physical index mappings.
tvm_tirx.tirx.tile	Tile operations, views, and dispatch.
tvm_tirx.tirx.script	Tensor IR syntax and JIT compilation.
tvm_tirx.codegen	Shared code generation and LLVM helpers.
tvm_tirx.backend	Load and register device backends.
tvm_tirx.backend.cuda	CUDA backend configuration and registration.
tvm_tirx.backend.cuda.op	CUDA intrinsic builders.
tvm_tirx.backend.cuda.script	CUDA kernel authoring namespace.
tvm_tirx.backend.cuda.layout	CUDA shared-memory and tensor-memory layouts.
tvm_tirx.backend.cuda.lang	CUDA pipelines, barriers, storage pools, and scheduling.
tvm_tirx.backend.cuda.transform	CUDA-specific transformation passes.
tvm_tirx.backend.rocm	ROCm backend registration.
tvm_tirx.backend.metal	Metal backend and script namespaces.
tvm_tirx.backend.metal.op	Metal intrinsic builders.
tvm_tirx.backend.opencl	OpenCL backend and script namespaces.
tvm_tirx.backend.adreno	Adreno backend registration.
tvm_tirx.backend.hexagon	Hexagon backend and linker integration.
tvm_tirx.backend.vulkan	Vulkan backend registration.
tvm_tirx.backend.webgpu	WebGPU backend registration.
tvm_tirx.backend.trn	Trainium backend and script namespaces.
tvm_tirx.backend.trn.op	Trainium NKI intrinsic wrappers.
tvm_tirx.backend.trn.transform	Trainium-specific lowering passes.
tvm_tirx.backend.rust	Operators written in Rust.
tvm_tirx.support	Build tools and system utilities.
tvm_tirx.testing	Test fixtures and numerical checks.

TVM-FFI APIs

Page	Purpose
tvm_ffi.structural	Walk, map, visit, and mutate IR.

Classes

Alias	Canonical API
DataTypeCode	tvm_tirx.runtime.DataTypeCode
IRModule	tvm_tirx.ir.IRModule

Functions

Use compile() to build a TIRx function or module into an executable.

compile(mod[, target, pipeline])

Compile a function or module to an executable for the target.

Alias	Canonical API
cpu	tvm_tirx.runtime.cpu
cuda	tvm_tirx.runtime.cuda
opencl	tvm_tirx.runtime.opencl
vulkan	tvm_tirx.runtime.vulkan
metal	tvm_tirx.runtime.metal
vpi	tvm_tirx.runtime.vpi
rocm	tvm_tirx.runtime.rocm
ext_dev	tvm_tirx.runtime.ext_dev
device_from_target	tvm_tirx.target.device_from_target