#include <ir_emitter.h>

Inherits DfsHloVisitorWithDefault.

Public Member Functions
	IrEmitter (const HloModule &hlo_module, const BufferAssignment &assignment, llvm::Module llvm_module, std::unordered_map< const HloInstruction , int64 > instruction_to_profile_idx, std::unordered_map< const HloComputation , int64 > computation_to_profile_idx, llvm::TargetMachine target_machine, ExternalConstantPool *external_constant_pool)
	Create a new LLVM IR emitter. More...

StatusOr< llvm::Function * >	EmitComputation (HloComputation computation, const string &function_name_prefix, bool is_top_level_computation, std::vector< const HloInstruction > *instruction_order)

Detailed Description

Google docs:

This class is the top-level API for the XLA HLO –> LLVM IR compiler. It implements the DfsHloVisitor interface and emits HLO computations as LLVM IR functions.

Constructor & Destructor Documentation

◆ IrEmitter()

xla::cpu::IrEmitter::IrEmitter	(	const HloModule &	hlo_module,
		const BufferAssignment &	assignment,
		llvm::Module *	llvm_module,
		std::unordered_map< const HloInstruction *, int64 >	instruction_to_profile_idx,
		std::unordered_map< const HloComputation *, int64 >	computation_to_profile_idx,
		llvm::TargetMachine *	target_machine,
		ExternalConstantPool *	external_constant_pool
	)

Create a new LLVM IR emitter.

Parameters

hlo_module	The HLO module we are emitting IR for.
assignment	A BufferAssignment from which we know which temporary buffers are used by the HLO nodes.
llvm_module	The LLVM module to emit IR into.
instruction_to_profile_idx	The mapping from HLO instructions to their index in the profiling array.
computation_to_profile_idx	The mapping from HLO computations to their index in the profiling array.
external_constant_pool	If non-null, points to an ExternalConstantPool instance into which the Ir emitter can spill constants.

Member Function Documentation

◆ EmitComputation()

StatusOr< llvm::Function * > xla::cpu::IrEmitter::EmitComputation	(	HloComputation *	computation,
		const string &	function_name_prefix,
		bool	is_top_level_computation,
		std::vector< const HloInstruction >	instruction_order
	)

The followings are from google docs.

Emit and return the given HLO computation as an LLVM IR function.