carbon-lang/toolchain/lowering/lowering.h

// Part of the Carbon Language project, under the Apache License v2.0 with LLVM
// Exceptions. See /LICENSE for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

#ifndef CARBON_TOOLCHAIN_LOWERING_LOWERING_H_
#define CARBON_TOOLCHAIN_LOWERING_LOWERING_H_

#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
#include "toolchain/semantics/semantics_ir.h"
#include "toolchain/semantics/semantics_node.h"

namespace Carbon {

// Use LowerToLLVM rather than calling this directly.
//
// This carries state for lowering. `Run()` should only be called once, and
// handles the main execution.
class Lowering {
 public:
  explicit Lowering(llvm::LLVMContext& llvm_context,
                    llvm::StringRef module_name,
                    const SemanticsIR& semantics_ir);

  // Lowers the SemanticsIR to LLVM IR.
  auto Run() -> std::unique_ptr<llvm::Module>;

 private:
  // Declare handlers for each SemanticsIR node.
#define CARBON_SEMANTICS_NODE_KIND(Name) \
  auto Handle##Name##Node(SemanticsNodeId node_id, SemanticsNode node)->void;
#include "toolchain/semantics/semantics_node_kind.def"

  // Runs lowering for a block.
  auto LowerBlock(SemanticsNodeBlockId block_id) -> void;

  // Returns a type for the given node.
  auto LowerNodeToType(SemanticsNodeId node_id) -> llvm::Type*;

  // State for building the LLVM IR.
  llvm::LLVMContext* llvm_context_;
  std::unique_ptr<llvm::Module> llvm_module_;
  llvm::IRBuilder<> builder_;

  // The input Semantics IR.
  const SemanticsIR* const semantics_ir_;

  // Blocks which we've observed and need to lower.
  llvm::SmallVector<std::pair<llvm::BasicBlock*, SemanticsNodeBlockId>>
      todo_blocks_;

  // Maps nodes in SemanticsIR to a lowered value. This will have one entry per
  // node, and will be non-null when lowered. It's expected to be sparse during
  // execution because while expressions will have entries, statements won't.
  // TODO: This will probably become a PointerUnion of Value and Type.
  // TODO: Long-term, we should examine the practical trade-offs of making this
  // a map; a map may end up lower memory consumption, but a vector offers cache
  // efficiency and better performance. As a consequence, the right choice is
  // unclear.
  llvm::SmallVector<llvm::Value*> lowered_nodes_;
};

}  // namespace Carbon

#endif  // CARBON_TOOLCHAIN_LOWERING_LOWERING_H_