#include <pslang/jit/arch/linux_x86_64/compiler.hpp>
#include <pslang/jit/arch/linux_x86_64/instruction_builder.hpp>
#include <pslang/jit/helpers.hpp>
#include <pslang/ir/node.hpp>
#include <pslang/ir/compiler.hpp>
#include <pslang/ast/type.hpp>
#include <pslang/ast/struct.hpp>
#include <pslang/ast/function.hpp>
#include <pslang/types/type_visitor.hpp>

namespace pslang::jit::linux_x86_64
{

	namespace
	{

		template <typename ... Regs>
		reg find_free_reg(reg start, Regs ... regs)
		{
			for (int i = (int)start; i < 16; ++i)
			{
				if ((true && ... &&  (regs != reg(i))))
					return reg(i);
			}
			throw std::runtime_error("Unable to find a free register");
		}

		static std::array<reg, 6> integer_arg_reg = {
			reg::rdi,
			reg::rsi,
			reg::rdx,
			reg::rcx,
			reg::r8,
			reg::r9,
		};

		static std::array<reg, 2> integer_ret_reg = {
			reg::rax,
			reg::rdx,
		};

		struct small_struct_data
		{
			struct octet
			{
				bool has_integer = false;
				bool has_floating_point = false;
			};

			std::optional<octet> octets[2];
		};

		struct value_address
		{
			std::optional<reg> base = std::nullopt;
			ir::node_ref global = {};

			std::int32_t offset;
		};

		bool is_global(ir::node_ref it)
		{
			return std::holds_alternative<ir::global>(it->instruction);
		}

		struct local_context
		{
			bool use_frame_pointer = true;

			std::unordered_map<types::type_ptr, small_struct_data> small_structs;

			std::unordered_map<ir::node_ref, std::int32_t> nodes;

			struct resolve_data
			{
				std::int32_t offset;
				ir::node_ref target;
			};

			std::vector<resolve_data> jump_resolve;
			std::vector<resolve_data> cjump_resolve;
			std::vector<resolve_data> node_resolve;
			std::vector<resolve_data> call_resolve;
		};

		std::optional<small_struct_data> classify_small_struct(local_context & lcontext, types::type_ptr type)
		{
			// TODO: unit type?

			if (ast::type_size(*type) > 16)
				return std::nullopt;

			if (auto it = lcontext.small_structs.find(type); it != lcontext.small_structs.end())
				return it->second;

			if (std::holds_alternative<types::struct_type>(*type) || std::holds_alternative<types::array_type>(*type))
			{
				small_struct_data result;

				auto layout = ast::flat_layout(*type);
				for (auto const & field : layout)
				{
					std::size_t octet_index = field.offset / 8;
					auto & octet = result.octets[octet_index].emplace();
					if (types::is_floating_point_type(*field.type))
						octet.has_floating_point = true;
					else
						octet.has_integer = true;
				}

				return (lcontext.small_structs[type] = result);
			}

			if (types::is_floating_point_type(*type))
				return small_struct_data{.octets = {small_struct_data::octet{.has_floating_point = true}, std::nullopt}};

			return small_struct_data{.octets = {small_struct_data::octet{.has_integer = true}, std::nullopt}};
		}

		struct populate_globals_visitor
		{
			compiled_module & result;
			local_context & lcontext;

			template <typename Node>
			void apply(ir::node_ref, Node const &, types::type_ptr const &)
			{}

			void apply(ir::node_ref it, ir::global const & node, types::type_ptr const & type)
			{
				auto size = ast::type_size(*type);
				if (node.initializer.size() > size)
					throw std::runtime_error("global IR node with initializer larger than type size");

				auto alignment = ast::type_alignment(*type);
				auto offset = result.data.size();
				offset = ((offset + (alignment - 1)) / alignment) * alignment;
				result.data.resize(offset + size);
				std::copy(node.initializer.begin(), node.initializer.end(), result.data.begin() + offset);
				result.globals[it] = offset;
			}
		};

		struct literal_visitor
		{
			instruction_builder & builder;

			void operator()(ast::bool_literal const & node)
			{
				if (node.value)
					builder.mov(-1, reg::rax);
				else
					builder.xor_(reg::rax, reg::rax);
			}

			template <typename T>
			requires(std::is_integral_v<T> && !std::is_same_v<T, bool>)
			void operator()(ast::primitive_literal_base<T> const & node)
			{
				if (sizeof(T) <= 4)
				{
					if (std::is_signed_v<T>)
						builder.movsd(std::int32_t(node.value), reg::rax);
					else
						builder.movzd(std::uint32_t(node.value), reg::rax);
				}
				else
					builder.mov(std::uint64_t(node.value), reg::rax);
			}

			void operator()(ast::f16_literal const & node)
			{
				throw std::runtime_error("Not implemented");
			}

			void operator()(ast::f32_literal const & node)
			{
				builder.movzd(*(std::uint32_t const *)(&node.value), reg::rax);
				builder.mov_gpr_to_xmm_32(reg::rax, reg::xmm0);
			}

			void operator()(ast::f64_literal const & node)
			{
				builder.mov(*(std::uint64_t const *)(&node.value), reg::rax);
				builder.mov_gpr_to_xmm(reg::rax, reg::xmm0);
			}
		};

		struct compile_visitor
		{
			compiled_module & module;
			local_context & lcontext;
			instruction_builder & builder;

			std::vector<std::int32_t> argument_position;
			std::unordered_map<ir::node_ref, std::int32_t> stack_position;
			std::int32_t stack_size = 0;
			bool return_value_is_large_struct = false;

			void apply(ir::node_ref, ir::label const &, types::type_ptr const &)
			{}

			void apply(ir::node_ref it, ir::literal const & node, types::type_ptr const & type)
			{
				std::visit(literal_visitor{builder}, node.value);
				if (types::is_integer_like_type(*type))
					store(it, reg::rax);
				else if (types::is_floating_point_type(*type))
					store_xmm(it, reg::xmm0, ast::type_size(*type));
			}

			void apply(ir::node_ref it, ir::alloc const & node, types::type_ptr const & type)
			{
				// Nothing to do: alloc just allocates a node of some type,
				// but we already allocated stack space for it
			}

			void apply(ir::node_ref it, ir::global const & node, types::type_ptr const & type)
			{
				// Globals are added in a separate pass before code
			}

			void apply(ir::node_ref it, ir::copy const & node, types::type_ptr const & type)
			{
				auto size = ast::type_size(*type);
				auto dst_address = node_address(it);

				auto src_type = node.source->inferred_type;
				auto src_address = node_address(node.source);
				for (auto field_id : node.path)
				{
					if (auto struct_type = std::get_if<types::struct_type>(src_type.get()))
					{
						auto struct_node = struct_type->node;
						src_type = struct_node->fields[field_id].inferred_type;
						src_address.offset += struct_node->fields[field_id].layout.offset;
					}
					else if (auto array_type = std::get_if<types::array_type>(src_type.get()))
					{
						src_type = array_type->element_type;
						src_address.offset += field_id * ast::type_size(*array_type->element_type);
					}
					else
						throw std::runtime_error("Unknown object type for field copy");
				}

				copy_memory(src_address, dst_address, size);
			}

			void apply(ir::node_ref it, ir::load const & node, types::type_ptr const & type)
			{
				load(node.ptr, reg::rax);
				auto size = ast::type_size(*type);
				auto dst_address = node_address(it);
				copy_memory({.base = reg::rax, .offset = 0}, dst_address, size);
			}

			void apply(ir::node_ref it, ir::store const & node, types::type_ptr const & type)
			{
				load(node.ptr, reg::rax);
				auto size = ast::type_size(*type);
				auto src_address = node_address(node.value);
				copy_memory(src_address, {.base = reg::rax, .offset = 0}, size);
			}

			void apply(ir::node_ref it, ir::unary_operation const & node, types::type_ptr const & type)
			{
				switch (node.type)
				{
				case ast::unary_operation_type::negation:
					if (types::is_integer_type(*type))
					{
						load(node.arg1, reg::rax);
						builder.neg(reg::rax);
						store(it, reg::rax);
					}
					else if (types::is_floating_point_type(*type))
					{
						auto const size = ast::type_size(*type);
						load_xmm(node.arg1, reg::xmm0, size);
						if (size <= 4)
						{
							// f16 or f32
							builder.movzd(0x80000000u, reg::rax);
							builder.mov_gpr_to_xmm_32(reg::rax, reg::xmm1);
							builder.xor_xmm_32(reg::xmm1, reg::xmm0);
						}
						else
						{
							// f64
							builder.mov(0x8000000000000000ull, reg::rax);
							builder.mov_gpr_to_xmm(reg::rax, reg::xmm1);
							builder.xor_xmm(reg::xmm1, reg::xmm0);
						}
						store_xmm(it, reg::xmm0, size);
					}
					break;
				case ast::unary_operation_type::logical_not:
					load(node.arg1, reg::rax);
					builder.not_(reg::rax);
					store(it, reg::rax);
					break;
				case ast::unary_operation_type::address_of:
				case ast::unary_operation_type::mutable_address_of:
					{
						auto const address = node_address(node.arg1);
						if (address.base)
						{
							builder.lea(*address.base, address.offset, reg::rax);
						}
						else
						{
							push_resolve_global(address.global);
							builder.lea_rip_prev(0, reg::rax);
						}
						store(it, reg::rax);
					}
					break;
				case ast::unary_operation_type::dereference:
					throw std::runtime_error("Dereference operator mush not be present in compiled IR");
				}
			}

			void apply(ir::node_ref it, ir::binary_operation const & node, types::type_ptr const & type)
			{
				auto const arg1_type = node.arg1->inferred_type;
				auto const arg2_type = node.arg2->inferred_type;
				bool const is_fp = types::is_floating_point_type(*arg1_type);
				bool const result_is_fp = types::is_floating_point_type(*type);
				auto const arg_size = ast::type_size(*arg1_type);

				if (is_fp)
				{
					load_xmm(node.arg1, reg::xmm0, arg_size);
					load_xmm(node.arg2, reg::xmm1, arg_size);
				}
				else
				{
					load(node.arg1, reg::rax);
					load(node.arg2, reg::rbx);
				}

				switch (node.type)
				{
				case ast::binary_operation_type::addition:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.add_xmm_32(reg::xmm1, reg::xmm0);
						else
							builder.add_xmm(reg::xmm1, reg::xmm0);
					}
					else
						builder.add(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::subtraction:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.sub_xmm_32(reg::xmm1, reg::xmm0);
						else
							builder.sub_xmm(reg::xmm1, reg::xmm0);
					}
					else
						builder.sub(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::multiplication:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.mul_xmm_32(reg::xmm1, reg::xmm0);
						else
							builder.mul_xmm(reg::xmm1, reg::xmm0);
					}
					else
						builder.mul(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::division:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.div_xmm_32(reg::xmm1, reg::xmm0);
						else
							builder.div_xmm(reg::xmm1, reg::xmm0);
					}
					else
					{
						reg_extend(reg::rax, reg::rax, *arg1_type);
						reg_extend(reg::rbx, reg::rbx, *arg1_type);
						if (types::is_unsigned_integer_type(*arg1_type))
						{
							builder.xor_(reg::rdx, reg::rdx);
							builder.udiv(reg::rbx);
						}
						else
						{
							builder.cqo();
							builder.idiv(reg::rbx);
						}
					}
					break;
				case ast::binary_operation_type::remainder:
					reg_extend(reg::rax, reg::rax, *arg1_type);
					reg_extend(reg::rbx, reg::rbx, *arg1_type);
					if (types::is_unsigned_integer_type(*arg1_type))
					{
						builder.xor_(reg::rdx, reg::rdx);
						builder.udiv(reg::rbx);
						builder.mov(reg::rdx, reg::rax);
					}
					else
					{
						builder.cqo();
						builder.idiv(reg::rbx);
						builder.mov(reg::rdx, reg::rax);
					}
					break;
				case ast::binary_operation_type::binary_and:
					builder.and_(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::logical_and:
					throw std::runtime_error("Short-circuiting operators must have been unwrapped in IR compiler");
				case ast::binary_operation_type::binary_or:
					builder.or_(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::logical_or:
					throw std::runtime_error("Short-circuiting operators must have been unwrapped in IR compiler");
				case ast::binary_operation_type::logical_xor:
					builder.xor_(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::left_shift:
					builder.mov(reg::rbx, reg::rcx);
					builder.shl(reg::rax);
					break;
				case ast::binary_operation_type::right_shift:
					reg_extend(reg::rax, reg::rax, *arg1_type);
					reg_extend(reg::rbx, reg::rcx, *arg2_type);
					if (types::is_unsigned_integer_type(*arg1_type))
						builder.shr(reg::rax);
					else
						builder.sar(reg::rax);
					break;
				case ast::binary_operation_type::equals:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.cmp_xmm_32(reg::xmm0, reg::xmm1);
						else
							builder.cmp_xmm(reg::xmm0, reg::xmm1);
					}
					else
					{
						reg_extend(reg::rax, reg::rax, *arg1_type);
						reg_extend(reg::rbx, reg::rbx, *arg1_type);
						builder.cmp(reg::rax, reg::rbx);
					}
					builder.mov(0, reg::rax);
					builder.mov(-1, reg::rbx);
					builder.cmovz(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::not_equals:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.cmp_xmm_32(reg::xmm0, reg::xmm1);
						else
							builder.cmp_xmm(reg::xmm0, reg::xmm1);
					}
					else
					{
						reg_extend(reg::rax, reg::rax, *arg1_type);
						reg_extend(reg::rbx, reg::rbx, *arg1_type);
						builder.cmp(reg::rax, reg::rbx);
					}
					builder.mov(0, reg::rax);
					builder.mov(-1, reg::rbx);
					builder.cmovnz(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::less:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.cmp_xmm_32(reg::xmm0, reg::xmm1);
						else
							builder.cmp_xmm(reg::xmm0, reg::xmm1);
					}
					else
					{
						reg_extend(reg::rax, reg::rax, *arg1_type);
						reg_extend(reg::rbx, reg::rbx, *arg1_type);
						builder.cmp(reg::rax, reg::rbx);
					}
					builder.mov(0, reg::rax);
					builder.mov(-1, reg::rbx);
					if (is_fp || types::is_bool_type(*arg1_type) || types::is_unsigned_integer_type(*arg1_type))
						builder.cmovb(reg::rbx, reg::rax);
					else
						builder.cmovl(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::greater:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.cmp_xmm_32(reg::xmm1, reg::xmm0);
						else
							builder.cmp_xmm(reg::xmm1, reg::xmm0);
					}
					else
					{
						reg_extend(reg::rax, reg::rax, *arg1_type);
						reg_extend(reg::rbx, reg::rbx, *arg1_type);
						builder.cmp(reg::rbx, reg::rax);
					}
					builder.mov(0, reg::rax);
					builder.mov(-1, reg::rbx);
					if (is_fp || types::is_bool_type(*arg1_type) || types::is_unsigned_integer_type(*arg1_type))
						builder.cmovb(reg::rbx, reg::rax);
					else
						builder.cmovl(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::less_equals:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.cmp_xmm_32(reg::xmm1, reg::xmm0);
						else
							builder.cmp_xmm(reg::xmm1, reg::xmm0);
					}
					else
					{
						reg_extend(reg::rax, reg::rax, *arg1_type);
						reg_extend(reg::rbx, reg::rbx, *arg1_type);
						builder.cmp(reg::rbx, reg::rax);
					}
					builder.mov(0, reg::rax);
					builder.mov(-1, reg::rbx);
					if (is_fp || types::is_bool_type(*arg1_type) || types::is_unsigned_integer_type(*arg1_type))
						builder.cmovnb(reg::rbx, reg::rax);
					else
						builder.cmovnl(reg::rbx, reg::rax);
					break;
				case ast::binary_operation_type::greater_equals:
					if (is_fp)
					{
						if (arg_size <= 4)
							builder.cmp_xmm_32(reg::xmm0, reg::xmm1);
						else
							builder.cmp_xmm(reg::xmm0, reg::xmm1);
					}
					else
					{
						reg_extend(reg::rax, reg::rax, *arg1_type);
						reg_extend(reg::rbx, reg::rbx, *arg1_type);
						builder.cmp(reg::rax, reg::rbx);
					}
					builder.mov(0, reg::rax);
					builder.mov(-1, reg::rbx);
					if (is_fp || types::is_bool_type(*arg1_type) || types::is_unsigned_integer_type(*arg1_type))
						builder.cmovnb(reg::rbx, reg::rax);
					else
						builder.cmovnl(reg::rbx, reg::rax);
					break;
				}

				if (result_is_fp)
					store_xmm(it, reg::xmm0, arg_size);
				else
					store(it, reg::rax);
			}

			void apply(ir::node_ref it, ir::cast_operation const & node, types::type_ptr const & type)
			{
				auto src_type = node.arg1->inferred_type;
				auto dst_type = node.target_type;

				auto u64_type = types::primitive_type{types::u64_type{}};

				if (false
					|| (types::is_builtin_type(*src_type) && types::equal(*src_type, *dst_type))
					|| (types::is_pointer_type(*src_type) && types::is_pointer_type(*dst_type))
					|| (types::equal(*src_type, u64_type) && types::is_pointer_type(*dst_type))
					|| (types::equal(*dst_type, u64_type) && types::is_pointer_type(*src_type))
				)
				{
					load(node.arg1, reg::rax);
					store(it, reg::rax);
					return;
				}

				if (auto array_type = std::get_if<types::array_type>(src_type.get()))
				{
					if (auto pointer_type = std::get_if<types::pointer_type>(dst_type.get()))
					{
						if (types::equal(*array_type->element_type, *pointer_type->referenced_type))
						{
							auto const arg1_address = node_address(node.arg1);
							if (arg1_address.base)
							{
								builder.lea(*arg1_address.base, arg1_address.offset, reg::rax);
							}
							else
							{
								push_resolve_global(arg1_address.global);
								builder.lea_rip_prev(0, reg::rax);
							}
							store(it, reg::rax);
							return;
						}
					}
				}

				if (types::is_numeric_type(*src_type) && types::is_numeric_type(*dst_type))
				{
					if (types::is_integer_type(*src_type))
					{
						load(node.arg1, reg::rax);
						if (types::is_integer_type(*dst_type))
						{
							reg_extend(reg::rax, reg::rax, *src_type);
						}
						else if (types::is_floating_point_type(*dst_type))
						{
							reg_extend(reg::rax, reg::rax, *src_type);

							auto dst_size = ast::type_size(*dst_type);

							if (types::is_unsigned_integer_type(*src_type) && ast::type_size(*src_type) == 8)
							{
								// No u64 -> f32/f64 conversion on x86_64
								// Emulate it by splitting x = 2 * (x >> 1) + (x & 1)

								reg_extend(reg::rax, reg::rax, *src_type);

								builder.test(reg::rax, reg::rax);
								auto first_jump = builder.code.size();
								builder.jump_if_sign(0);

								if (dst_size <= 4)
									builder.gpr_to_xmm_32(reg::rax, reg::xmm0);
								else
									builder.gpr_to_xmm(reg::rax, reg::xmm0);

								auto second_jump = builder.code.size();
								builder.jump(0);

								auto first_jump_target = builder.code.size();

								builder.mov(reg::rax, reg::rcx);
								builder.shr1(reg::rcx);
								builder.and_imm8(reg::rax, 1);
								builder.or_(reg::rax, reg::rcx);

								if (dst_size <= 4)
								{
									builder.gpr_to_xmm_32(reg::rcx, reg::xmm0);
									builder.add_xmm_32(reg::xmm0, reg::xmm0);
								}
								else
								{
									builder.gpr_to_xmm(reg::rcx, reg::xmm0);
									builder.add_xmm(reg::xmm0, reg::xmm0);
								}

								auto second_jump_target = builder.code.size();

								builder.cjump_inject_prev(builder.code.data() + first_jump, first_jump_target - first_jump);
								builder.jump_inject_prev(builder.code.data() + second_jump, second_jump_target - second_jump);
							}
							else
							{
								if (dst_size <= 4)
									builder.gpr_to_xmm_32(reg::rax, reg::xmm0);
								else
									builder.gpr_to_xmm(reg::rax, reg::xmm0);
							}
						}
					}
					else if (types::is_floating_point_type(*src_type))
					{
						if (types::is_integer_type(*dst_type))
						{
							// TODO: correct floating-point to 64-bit unsigned conversion

							auto src_size = ast::type_size(*src_type);
							if (src_size <= 4)
								builder.xmm_32_to_gpr(reg::xmm0, reg::rax);
							else
								builder.xmm_to_gpr(reg::xmm0, reg::rax);
						}
						else if (types::is_floating_point_type(*dst_type))
						{
							auto src_size = ast::type_size(*src_type);
							auto dst_size = ast::type_size(*dst_type);

							if (src_size <= 4 && dst_size == 8)
								builder.xmm_32_to_64(reg::xmm0, reg::xmm0);
							else if (src_size == 8 && dst_size <= 4)
								builder.xmm_64_to_32(reg::xmm0, reg::xmm0);
						}
					}

					if (types::is_integer_type(*dst_type))
					{
						store(it, reg::rax);
					}
					else if (types::is_floating_point_type(*dst_type))
					{
						store_xmm(it, reg::xmm0, ast::type_size(*dst_type));
					}

					return;
				}

				throw std::runtime_error("Unknown types for cast instruction");
			}

			void apply(ir::node_ref, ir::argument const &, types::type_ptr const &)
			{
				// Nothing to do: arguments already pushed on stack in function preamble
			}

			void apply(ir::node_ref it, ir::instruction_address const & node, types::type_ptr const & type)
			{
				lcontext.node_resolve.emplace_back(builder.code.size(), node.target);
				builder.lea_rip(0, reg::rax);
				store(it, reg::rax);
			}

			void apply(ir::node_ref it, ir::extern_symbol const & node, types::type_ptr const & type)
			{
				module.external_function_resolve.push_back({.name = node.name, .instruction_offset = (std::int32_t)builder.code.size()});
				builder.mov_read_rip_prev(0, reg::rax);
				store(it, reg::rax);
			}

			void apply(ir::node_ref, ir::assignment const & node, types::type_ptr const & type)
			{
				auto src_address = node_address(node.rhs);

				auto dst_type = node.lhs->inferred_type;
				auto dst_address = node_address(node.lhs);
				for (auto field_id : node.path)
				{
					if (auto struct_type = std::get_if<types::struct_type>(dst_type.get()))
					{
						auto struct_node = struct_type->node;
						dst_type = struct_node->fields[field_id].inferred_type;
						dst_address.offset += struct_node->fields[field_id].layout.offset;
					}
					else if (auto array_type = std::get_if<types::array_type>(dst_type.get()))
					{
						dst_type = array_type->element_type;
						dst_address.offset += field_id * ast::type_size(*array_type->element_type);
					}
					else
						throw std::runtime_error("Unknown object type for field assignment");
				}

				copy_memory(src_address, dst_address, ast::type_size(*dst_type));
			}

			void apply(ir::node_ref, ir::jump const & node, types::type_ptr const & type)
			{
				lcontext.jump_resolve.push_back({(std::int32_t)builder.code.size(), node.target});
				builder.jump(0);
			}

			void apply(ir::node_ref, ir::jump_if_zero const & node, types::type_ptr const & type)
			{
				load(node.condition, reg::rax);
				reg_extend(reg::rax, reg::rax, *node.condition->inferred_type);
				builder.test(reg::rax, reg::rax);
				lcontext.cjump_resolve.push_back({(std::int32_t)builder.code.size(), node.target});
				builder.jump_if_zero(0);
			}

			void apply(ir::node_ref, ir::jump_if_nonzero const & node, types::type_ptr const & type)
			{
				load(node.condition, reg::rax);
				reg_extend(reg::rax, reg::rax, *node.condition->inferred_type);
				builder.test(reg::rax, reg::rax);
				lcontext.cjump_resolve.push_back({(std::int32_t)builder.code.size(), node.target});
				builder.jump_if_nonzero(0);
			}

			template <typename Node, typename DoCall>
			void apply_call(ir::node_ref it, Node const & node, types::type_ptr const & type, DoCall && do_call)
			{
				bool return_value_is_large_struct = false;
				if (std::holds_alternative<types::struct_type>(*type) || std::holds_alternative<types::array_type>(*type))
					return_value_is_large_struct = !classify_small_struct(lcontext, type);

				if (this->return_value_is_large_struct)
				{
					builder.push(reg::rdi);
					stack_size += 8;
				}

				// Compute space required for stack arguments
				std::int32_t arguments_stack_size = 0;
				for (auto const & argument : node.arguments)
				{
					auto size = ast::type_size(*argument->inferred_type);
					auto alignment = ast::type_size(*argument->inferred_type);
					auto struct_type = std::get_if<types::struct_type>(argument->inferred_type.get());
					auto array_type = std::get_if<types::array_type>(argument->inferred_type.get());
					if (struct_type || array_type)
					{
						if (!classify_small_struct(lcontext, argument->inferred_type))
						{
							arguments_stack_size += (alignment - (arguments_stack_size % alignment)) % alignment;
							arguments_stack_size += size;
						}
					}
				}

				// Ensure stack is 16-byte aligned after CALL pushes return address on stack
				// I.e. enforce (stack_size + arguments_stack_size) % 16 == 8
				arguments_stack_size += (24 - ((arguments_stack_size + stack_size) % 16)) % 16;

				// TODO: handle the case when there weren't enough registers
				std::uint8_t reg_index = return_value_is_large_struct ? 1 : 0;
				std::uint8_t fp_reg = 0;
				std::int32_t stack_offset = 0;
				for (auto const & argument : node.arguments)
				{
					auto size = ast::type_size(*argument->inferred_type);
					auto alignment = ast::type_alignment(*argument->inferred_type);
					auto struct_type = std::get_if<types::struct_type>(argument->inferred_type.get());
					auto array_type = std::get_if<types::array_type>(argument->inferred_type.get());
					if (struct_type || array_type)
					{
						if (auto small_struct = classify_small_struct(lcontext, argument->inferred_type))
						{
							for (int o : {0, 1})
							{
								auto & octet = small_struct->octets[o];
								if (!octet) break;

								if (octet->has_integer)
									load(argument, integer_arg_reg[reg_index++], 8 * o);
								else if (octet->has_floating_point)
									load_xmm(argument, (reg)(fp_reg++), 8, 8 * o);
							}
						}
						else
						{
							stack_offset += (alignment - (stack_offset % alignment)) % alignment;
							copy_memory(node_address(argument), {.base = reg::rsp, .offset = stack_offset - arguments_stack_size}, size);
							stack_offset += size;
						}
					}
					else if (types::is_integer_like_type(*argument->inferred_type))
						load(argument, integer_arg_reg[reg_index++]);
					else if (types::is_floating_point_type(*argument->inferred_type))
						load_xmm(argument, (reg)(fp_reg++), size);
					else
						throw std::runtime_error("Unsupported function argument type");
				}

				if (return_value_is_large_struct)
				{
					// Function call node cannot have RIP-relative address
					auto const address = node_address(it);
					builder.lea(address.base.value(), address.offset, reg::rdi);
				}

				if (arguments_stack_size > 0)
				{
					builder.sub_imm(reg::rsp, arguments_stack_size);
					stack_size += arguments_stack_size;
				}

				do_call();

				if (arguments_stack_size > 0)
				{
					builder.add_imm(reg::rsp, arguments_stack_size);
					stack_size -= arguments_stack_size;
				}

				std::size_t const size = ast::type_size(*type);
				auto struct_type = std::get_if<types::struct_type>(type.get());
				auto array_type = std::get_if<types::array_type>(type.get());
				if (size == 0)
				{}
				else if (struct_type || array_type)
				{
					if (auto small_struct = classify_small_struct(lcontext, type))
					{
						std::uint8_t reg_index = 0;
						std::uint8_t fp_reg = 0;
						for (int o : {0, 1})
						{
							auto & octet = small_struct->octets[o];
							if (!octet) break;

							if (octet->has_integer)
								store(it, integer_ret_reg[reg_index++], 8 * o);
							else if (octet->has_floating_point)
								store_xmm(it, (reg)(fp_reg++), 8, 8 * o);
						}
					}
					else
					{
						// Nothing to be done - return value should already be in place
					}
				}
				else if (types::is_integer_like_type(*type))
					store(it, reg::rax);
				else if (types::is_floating_point_type(*type))
					store_xmm(it, reg::xmm0, size);
				else
					throw std::runtime_error("Unsupported return value type");

				if (this->return_value_is_large_struct)
				{
					builder.pop(reg::rdi);
					stack_size -= 8;
				}
			}

			void apply(ir::node_ref it, ir::call const & node, types::type_ptr const & type)
			{
				// TODO: call function from a different module?
				apply_call(it, node, type, [&]{
					lcontext.call_resolve.emplace_back(builder.code.size(), node.target);
					builder.call_imm(0);
				});
			}

			void apply(ir::node_ref it, ir::call_pointer const & node, types::type_ptr const & type)
			{
				apply_call(it, node, type, [&]{
					load(node.pointer, reg::rax);
					builder.call_reg(reg::rax);
				});
			}

			void apply(ir::node_ref, ir::return_value const & node, types::type_ptr const & type)
			{
				if (node.value)
				{
					auto type = (*node.value)->inferred_type;
					auto size = ast::type_size(*type);
					auto struct_type = std::get_if<types::struct_type>(type.get());
					auto array_type = std::get_if<types::array_type>(type.get());
					if (size == 0)
					{}
					else if (struct_type || array_type)
					{
						if (auto small_struct = classify_small_struct(lcontext, type))
						{
							std::uint8_t reg_index = 0;
							std::uint8_t fp_reg = 0;
							for (int o : {0, 1})
							{
								auto & octet = small_struct->octets[o];
								if (!octet) break;

								if (octet->has_integer)
									load(*node.value, integer_ret_reg[reg_index++], 8 * o);
								else if (octet->has_floating_point)
									load_xmm(*node.value, (reg)(fp_reg++), 8, 8 * o);
							}
						}
						else
						{
							copy_memory(node_address(*node.value), {reg::rdi, {}, 0}, size);
						}
					}
					else if (types::is_integer_like_type(*type))
						load(*node.value, reg::rax);
					else if (types::is_floating_point_type(*type))
						load_xmm(*node.value, reg::xmm0, ast::type_size(*type));
					else
						throw std::runtime_error("Unsupported return value type");
				}
				if (stack_size > 0)
					builder.add_imm(reg::rsp, stack_size);

				builder.pop(reg::rbx);
				if (lcontext.use_frame_pointer)
					builder.pop(reg::rbp);
				builder.ret();
			}

			void compile(ast::function_definition const * function_definition, ir::node_ref begin, ir::node_ref end)
			{
				auto result_type = function_definition->inferred_result_type;

				if (result_type)
					if (std::holds_alternative<types::struct_type>(*result_type) || std::holds_alternative<types::array_type>(*result_type))
						return_value_is_large_struct = !classify_small_struct(lcontext, result_type);

				stack_size = 0;

				for (auto const & argument : function_definition->arguments)
				{
					auto size = ast::type_size(*argument.inferred_type);
					// Ensure max alignment for simplicity
					stack_size += ((size + 7) / 8) * 8;
					argument_position.push_back(stack_size);
				}

				for (auto it = begin; it != end; ++it)
				{
					if (auto argument = std::get_if<ir::argument>(&it->instruction))
					{
						stack_position[it] = argument_position[argument->index];
					}
					else if (is_global(it))
					{
						// stack position doesn't make sense for globals
					}
					else if (ir::is_value_instruction(it->instruction))
					{
						auto size = ast::type_size(*it->inferred_type);
						if (size > 0)
							stack_size += ((size + 7) / 8) * 8;
						stack_position[it] = stack_size;
					}
				}

				if (!std::holds_alternative<ir::label>(begin->instruction))
					throw std::runtime_error("First IR node of a function must be a label");

				auto it = begin;

				lcontext.nodes[it] = builder.code.size();

				if (lcontext.use_frame_pointer)
				{
					builder.push(reg::rbp);
					builder.mov(reg::rsp, reg::rbp);
				}

				builder.push(reg::rbx);

				if (stack_size > 0)
					builder.sub_imm(reg::rsp, stack_size);

				// TODO: handle the case when there weren't enough registers

				// If return value is large struct, RDI stores the pointer to return value
				std::uint8_t reg_index = return_value_is_large_struct ? 1 : 0;
				std::uint8_t fp_reg = 0;
				// 8 for the return address
				// Another 8 for RBP if using frame pointers
				// Another 8 for saved RBX
				std::int32_t stack_offset = stack_size + (lcontext.use_frame_pointer ? 16 : 8) + 8;
				for (std::size_t i = 0; i < function_definition->arguments.size(); ++i)
				{
					auto const & argument = function_definition->arguments[i];
					auto size = ast::type_size(*argument.inferred_type);
					auto alignment = ast::type_alignment(*argument.inferred_type);
					auto struct_type = std::get_if<types::struct_type>(argument.inferred_type.get());
					auto array_type = std::get_if<types::array_type>(argument.inferred_type.get());
					if (size == 0) continue;
					if (struct_type || array_type)
					{
						if (auto small_struct = classify_small_struct(lcontext, argument.inferred_type))
						{
							for (std::size_t o : {0, 1})
							{
								auto & octet = small_struct->octets[o];
								if (!octet) break;
								if (octet->has_integer)
									builder.mov_write(integer_arg_reg[reg_index++], reg::rsp, stack_size - argument_position[i] + 8 * o);
								else if (octet->has_floating_point)
									builder.mov_write_xmm((reg)(fp_reg++), reg::rsp, stack_size - argument_position[i] + 8 * o);
							}
						}
						else
						{
							stack_offset += (alignment - (stack_offset % alignment)) % alignment;
							copy_memory({.base = reg::rsp, .offset = stack_offset}, {.base = reg::rsp, .offset = stack_size - argument_position[i]}, size);
							stack_offset += size;
						}
					}
					else if (types::is_integer_like_type(*argument.inferred_type))
					{
						builder.mov_write(integer_arg_reg[reg_index++], reg::rsp, stack_size - argument_position[i]);
					}
					else if (types::is_floating_point_type(*argument.inferred_type))
					{
						auto const size = ast::type_size(*argument.inferred_type);
						if (size == 4)
							builder.mov_write_xmm_32((reg)(fp_reg++), reg::rsp, stack_size - argument_position[i]);
						else if (size == 8)
							builder.mov_write_xmm((reg)(fp_reg++), reg::rsp, stack_size - argument_position[i]);
						else
							throw std::runtime_error("Bad type size for floating-point argument");
					}
					else
						throw std::runtime_error("Unknown argument type");
				}
				++it;

				for (; it != end; ++it)
				{
					// Globals are added in a separate pass before code
					if (lcontext.nodes.contains(it))
						continue;

					// Uncomment to debug per-node instruction generation:
					builder.nop();

					lcontext.nodes[it] = builder.code.size();
					std::visit([&](auto const & instruction){ apply(it, instruction, it->inferred_type); }, it->instruction);
				}
			}

		private:

			void reg_extend(reg reg_src, reg reg_dst, types::type const & type)
			{
				auto const size = ast::type_size(type);
				if (types::is_unsigned_integer_type(type))
				{
					if (size == 1)
						builder.movzx_8(reg_src, reg_dst);
					else if (size == 2)
						builder.movzx_16(reg_src, reg_dst);
					else if (size == 4)
						builder.movzx_32(reg_src, reg_dst);
					else if (reg_src != reg_dst)
						builder.mov(reg_src, reg_dst);
				}
				else if (types::is_signed_integer_type(type))
				{
					if (size == 1)
						builder.movsx_8(reg_src, reg_dst);
					else if (size == 2)
						builder.movsx_16(reg_src, reg_dst);
					else if (size == 4)
						builder.movsx_32(reg_src, reg_dst);
					else if (reg_src != reg_dst)
						builder.mov(reg_src, reg_dst);
				}
				else if (types::is_bool_type(type))
				{
					builder.movzx_8(reg_src, reg_dst);
				}
			}

			// Must not be called for global nodes - they use RIP-based addressing with future relocation
			value_address node_address(ir::node_ref it)
			{
				if (is_global(it))
					return {.global = it, .offset = 0};
				else
					return {.base = reg::rsp, .offset = stack_size - stack_position.at(it)};
			}

			void load(ir::node_ref it, reg reg_dst, std::int32_t offset = 0)
			{
				auto const address = node_address(it);
				if (address.base)
				{
					builder.mov_read(*address.base, address.offset + offset, reg_dst);
				}
				else
				{
					push_resolve_global(address.global);
					builder.mov_read_rip_prev(offset, reg_dst);
				}
			}

			void store(ir::node_ref it, reg reg_src, std::int32_t offset = 0)
			{
				auto const address = node_address(it);
				if (address.base)
				{
					auto const address = node_address(it);
					builder.mov_write(reg_src, *address.base, address.offset + offset);
				}
				else
				{
					push_resolve_global(address.global);
					builder.mov_write_rip_prev(reg_src, offset);
				}
			}

			void load_xmm(ir::node_ref it, reg reg_dst, std::uint8_t size, std::int32_t offset = 0)
			{
				if (size != 4 && size != 8)
					throw std::runtime_error("Bad type size for load_xmm");

				auto const address = node_address(it);
				if (address.base)
				{
					auto const address = node_address(it);
					if (size == 4)
						builder.mov_read_xmm_32(*address.base, address.offset + offset, reg_dst);
					else
						builder.mov_read_xmm(*address.base, address.offset + offset, reg_dst);
				}
				else
				{
					// TODO
					throw std::runtime_error("RIP-relative XMM read is not supported");
				}
			}

			void store_xmm(ir::node_ref it, reg reg_src, std::uint8_t size, std::int32_t offset = 0)
			{
				if (size != 4 && size != 8)
					throw std::runtime_error("Bad type size for store_xmm");

				auto const address = node_address(it);
				if (address.base)
				{
					if (size == 4)
						builder.mov_write_xmm_32(reg_src, *address.base, address.offset + offset);
					else
						builder.mov_write_xmm(reg_src, *address.base, address.offset + offset);
				}
				else
				{
					// TODO
					throw std::runtime_error("RIP-relative XMM write is not supported");
				}
			}

			void copy_memory(value_address src, value_address dst, std::size_t size)
			{
				auto const storage_size_at_start = (std::int32_t)builder.code.size();

				reg reg_src;
				if (src.base)
					reg_src = *src.base;
				else
				{
					reg_src = find_free_reg(reg::r10, dst.base);
					push_resolve_global(src.global);
					builder.lea_rip_prev(0, reg_src);
				}

				reg reg_dst;
				if (dst.base)
					reg_dst = *dst.base;
				else
				{
					reg_dst = find_free_reg(reg::r10, reg_src);
					push_resolve_global(dst.global);
					builder.lea_rip_prev(0, reg_dst);
				}

				std::int32_t offset = 0;

				auto copy_reg = find_free_reg(reg::r10, reg_src, reg_dst);

				while (size > 0)
				{
					auto check_step = [&](std::size_t step)
					{
						return size >= step;
					};

					if (check_step(8))
					{
						builder.mov_read(reg_src, src.offset + offset, copy_reg);
						builder.mov_write(copy_reg, reg_dst, dst.offset + offset);
						size -= 8;
						offset += 8;
					}
					else if (check_step(4))
					{
						builder.mov_read_32(reg_src, src.offset + offset, copy_reg);
						builder.mov_write_32(copy_reg, reg_dst, dst.offset + offset);
						size -= 4;
						offset += 4;
					}
					else if (check_step(2))
					{
						builder.mov_read_16(reg_src, src.offset + offset, copy_reg);
						builder.mov_write_16(copy_reg, reg_dst, dst.offset + offset);
						size -= 2;
						offset += 2;
					}
					else
					{
						builder.mov_read_8(reg_src, src.offset + offset, copy_reg);
						builder.mov_write_8(copy_reg, reg_dst, dst.offset + offset);
						size -= 1;
						offset += 1;
					}
				}
			}

			void push_resolve_global(ir::node_ref global)
			{
				module.internal_global_resolve.push_back({.global = global, .instruction_offset = (std::int32_t)builder.code.size()});
			}
		};

	}

	compiled_module compile(ir::compiled_module const & module_in)
	{
		compiled_module result;

		local_context lcontext;

		{
			populate_globals_visitor visitor{result, lcontext};
			for (auto it = module_in.nodes->begin(); it != module_in.nodes->end(); ++it)
				std::visit([&](auto const & instruction){ visitor.apply(it, instruction, it->inferred_type); }, it->instruction);
		}

		instruction_builder builder{result.code};

		for (auto const & function : module_in.functions)
		{
			result.functions[function.second.begin] = result.code.size();
			compile_visitor visitor{result, lcontext, builder};
			visitor.compile(function.first, function.second.begin, function.second.end);
		}

		if (module_in.entry_point)
			result.entry_point = module_in.functions.at(module_in.entry_point).begin;

		for (auto const & resolve : lcontext.jump_resolve)
			builder.jump_inject_prev(result.code.data() + resolve.offset, lcontext.nodes.at(resolve.target) - resolve.offset);

		for (auto const & resolve : lcontext.cjump_resolve)
			builder.cjump_inject_prev(result.code.data() + resolve.offset, lcontext.nodes.at(resolve.target) - resolve.offset);

		for (auto const & resolve : lcontext.node_resolve)
			builder.lea_rip_inject_prev(result.code.data() + resolve.offset, lcontext.nodes.at(resolve.target) - resolve.offset);

		for (auto const & resolve : lcontext.call_resolve)
			builder.call_imm_inject_prev(result.code.data() + resolve.offset, lcontext.nodes.at(resolve.target) - resolve.offset);

		return result;
	}

}