#include "RISCv64Backend.h"
#include "RISCv64ISel.h"
#include "RISCv64RegAlloc.h"
#include "RISCv64AsmPrinter.h"
#include "RISCv64Passes.h"
#include <sstream>

namespace sysy {

// 顶层入口
std::string RISCv64CodeGen::code_gen() {
    return module_gen();
}

void printInitializer(std::stringstream& ss, const ValueCounter& init_values) {
    for (size_t i = 0; i < init_values.getValues().size(); ++i) {
        auto val = init_values.getValues()[i];
        auto count = init_values.getNumbers()[i];
        if (auto constant = dynamic_cast<ConstantValue*>(val)) { 
            for (unsigned j = 0; j < count; ++j) {
                if (constant->isInt()) {
                    ss << "    .word " << constant->getInt() << "\n";
                } else {
                    float f = constant->getFloat();
                    uint32_t float_bits = *(uint32_t*)&f;
                    ss << "    .word " << float_bits << "\n";
                }
            }
        }
    }
}

std::string RISCv64CodeGen::module_gen() {
    std::stringstream ss;
    
    // --- 步骤1：将全局变量(GlobalValue)分为.data和.bss两组 ---
    std::vector<GlobalValue*> data_globals;
    std::vector<GlobalValue*> bss_globals;

    for (const auto& global_ptr : module->getGlobals()) {
        GlobalValue* global = global_ptr.get();
        const auto& init_values = global->getInitValues();
        
        // 判断是否为大型零初始化数组，以便放入.bss段
        bool is_large_zero_array = false;
        if (init_values.getValues().size() == 1) {
            if (auto const_val = dynamic_cast<ConstantValue*>(init_values.getValues()[0])) {
                if (const_val->isInt() && const_val->getInt() == 0 && init_values.getNumbers()[0] > 16) {
                    is_large_zero_array = true;
                }
            }
        }

        if (is_large_zero_array) {
            bss_globals.push_back(global);
        } else {
            data_globals.push_back(global);
        }
    }

    // --- 步骤2：生成 .bss 段的代码 (这部分不变) ---
    if (!bss_globals.empty()) {
        ss << ".bss\n";
        for (GlobalValue* global : bss_globals) {
            unsigned count = global->getInitValues().getNumbers()[0];
            unsigned total_size = count * 4; // 假设元素都是4字节

            ss << "    .align 3\n";
            ss << ".globl " << global->getName() << "\n";
            ss << ".type " << global->getName() << ", @object\n";
            ss << ".size " << global->getName() << ", " << total_size << "\n";
            ss << global->getName() << ":\n";
            ss << "    .space " << total_size << "\n";
        }
    }
    
    // --- [修改] 步骤3：生成 .data 段的代码 ---
    // 我们需要检查 data_globals 和 常量列表是否都为空
    if (!data_globals.empty() || !module->getConsts().empty()) {
        ss << ".data\n";

        // a. 先处理普通的全局变量 (GlobalValue)
        for (GlobalValue* global : data_globals) {
            ss << ".globl " << global->getName() << "\n";
            ss << global->getName() << ":\n";
            printInitializer(ss, global->getInitValues());
        }

        // b. [新增] 再处理全局常量 (ConstantVariable)
        for (const auto& const_ptr : module->getConsts()) {
            ConstantVariable* cnst = const_ptr.get();
            ss << ".globl " << cnst->getName() << "\n";
            ss << cnst->getName() << ":\n";
            printInitializer(ss, cnst->getInitValues());
        }
    }

    // --- 处理函数 (.text段) 的逻辑保持不变 ---
    if (!module->getFunctions().empty()) {
        ss << ".text\n";
        for (const auto& func_pair : module->getFunctions()) {
            if (func_pair.second.get()) {
                ss << function_gen(func_pair.second.get());
            }
        }
    }
    return ss.str();
}

std::string RISCv64CodeGen::function_gen(Function* func) {
    // === 完整的后端处理流水线 ===

    // 阶段 1: 指令选择 (sysy::IR -> LLIR with virtual registers)
    RISCv64ISel isel;
    std::unique_ptr<MachineFunction> mfunc = isel.runOnFunction(func);

    // 第一次调试打印输出
    std::stringstream ss1;
    RISCv64AsmPrinter printer1(mfunc.get());
    printer1.run(ss1, true);

    // 阶段 2: 指令调度 (Instruction Scheduling)
    PreRA_Scheduler scheduler;
    scheduler.runOnMachineFunction(mfunc.get());

    // 阶段 3: 物理寄存器分配 (Register Allocation)
    RISCv64RegAlloc reg_alloc(mfunc.get());
    reg_alloc.run();

    // 阶段 3.1: 处理被调用者保存寄存器
    CalleeSavedHandler callee_handler;
    callee_handler.runOnMachineFunction(mfunc.get());

    // 阶段 4: 窥孔优化 (Peephole Optimization)
    PeepholeOptimizer peephole;
    peephole.runOnMachineFunction(mfunc.get());

    // 阶段 5: 局部指令调度 (Local Scheduling)
    PostRA_Scheduler local_scheduler;
    local_scheduler.runOnMachineFunction(mfunc.get());

    // 阶段 3.2: 插入序言和尾声
    PrologueEpilogueInsertionPass pei_pass;
    pei_pass.runOnMachineFunction(mfunc.get());

    // 阶段 3.3: 清理产生的大立即数
    LegalizeImmediatesPass legalizer;
    legalizer.runOnMachineFunction(mfunc.get());

    // 阶段 6: 代码发射 (Code Emission)
    std::stringstream ss;
    RISCv64AsmPrinter printer(mfunc.get());
    printer.run(ss);
    if (DEBUG) ss << "\n" << ss1.str(); // 将指令选择阶段的结果也包含在最终输出中
    return ss.str();
}

} // namespace sysy