From 502803c37fd1ba462686c3138f9ee12575412851 Mon Sep 17 00:00:00 2001
From: abnerhexu <20591243+abnerhexu@users.noreply.github.com>
Date: Fri, 26 Jun 2026 08:20:25 +0000
Subject: [PATCH] Initial Chisel core implementation

---
 .gitignore                                    |  25 +
 ARCHITECTURE.md                               | 466 ++++++++++++++++++
 build.sbt                                     |  22 +
 project/build.properties                      |   1 +
 sim/README.md                                 |  98 ++++
 sim/scripts/run_tests.sh                      |  61 +++
 sim/verilator/Makefile                        |  43 ++
 sim/verilator/memory.cpp                      | 111 +++++
 sim/verilator/memory.h                        |  40 ++
 sim/verilator/testbench.cpp                   | 101 ++++
 src/main/scala/Core.scala                     | 173 +++++++
 src/main/scala/commit/CommitStage.scala       |  18 +
 src/main/scala/common/Bundles.scala           |  76 +++
 src/main/scala/common/Consts.scala            |  35 ++
 src/main/scala/common/Parameters.scala        |  28 ++
 src/main/scala/csr/CSRFile.scala              |  99 ++++
 src/main/scala/csr/PrivilegeControl.scala     |  21 +
 src/main/scala/decode/Decoder.scala           | 150 ++++++
 src/main/scala/decode/IDStage.scala           |  20 +
 src/main/scala/execute/ALU.scala              |  36 ++
 src/main/scala/execute/BranchUnit.scala       |  21 +
 src/main/scala/execute/Divider.scala          |  14 +
 src/main/scala/execute/ExecStage.scala        |  29 ++
 src/main/scala/execute/FPU.scala              |  16 +
 src/main/scala/execute/Multiplier.scala       |  11 +
 src/main/scala/frontend/BTB.scala             |  32 ++
 src/main/scala/frontend/BranchPredictor.scala |  38 ++
 src/main/scala/frontend/Frontend.scala        |  43 ++
 src/main/scala/frontend/ICache.scala          |  23 +
 src/main/scala/frontend/ITLB.scala            |  16 +
 src/main/scala/frontend/RAS.scala             |  31 ++
 src/main/scala/issue/IssueQueue.scala         |  24 +
 src/main/scala/issue/IssueStage.scala         |  24 +
 src/main/scala/issue/ReservationStation.scala |  24 +
 src/main/scala/memory/DCache.scala            |  20 +
 src/main/scala/memory/DTLB.scala              |  15 +
 src/main/scala/memory/LSU.scala               |  30 ++
 src/main/scala/memory/LoadQueue.scala         |  23 +
 src/main/scala/memory/MMU.scala               |  15 +
 src/main/scala/memory/MemStage.scala          |  26 +
 src/main/scala/memory/StoreQueue.scala        |  24 +
 src/main/scala/regfile/PhysicalRegFile.scala  |  23 +
 src/main/scala/rename/FreeList.scala          |  26 +
 src/main/scala/rename/ROB.scala               |  71 +++
 src/main/scala/rename/RenameStage.scala       |  49 ++
 src/main/scala/rename/RenameTable.scala       |  32 ++
 src/main/scala/writeback/WriteBackStage.scala |  18 +
 47 files changed, 2342 insertions(+)
 create mode 100644 .gitignore
 create mode 100644 ARCHITECTURE.md
 create mode 100644 build.sbt
 create mode 100644 project/build.properties
 create mode 100644 sim/README.md
 create mode 100755 sim/scripts/run_tests.sh
 create mode 100644 sim/verilator/Makefile
 create mode 100644 sim/verilator/memory.cpp
 create mode 100644 sim/verilator/memory.h
 create mode 100644 sim/verilator/testbench.cpp
 create mode 100644 src/main/scala/Core.scala
 create mode 100644 src/main/scala/commit/CommitStage.scala
 create mode 100644 src/main/scala/common/Bundles.scala
 create mode 100644 src/main/scala/common/Consts.scala
 create mode 100644 src/main/scala/common/Parameters.scala
 create mode 100644 src/main/scala/csr/CSRFile.scala
 create mode 100644 src/main/scala/csr/PrivilegeControl.scala
 create mode 100644 src/main/scala/decode/Decoder.scala
 create mode 100644 src/main/scala/decode/IDStage.scala
 create mode 100644 src/main/scala/execute/ALU.scala
 create mode 100644 src/main/scala/execute/BranchUnit.scala
 create mode 100644 src/main/scala/execute/Divider.scala
 create mode 100644 src/main/scala/execute/ExecStage.scala
 create mode 100644 src/main/scala/execute/FPU.scala
 create mode 100644 src/main/scala/execute/Multiplier.scala
 create mode 100644 src/main/scala/frontend/BTB.scala
 create mode 100644 src/main/scala/frontend/BranchPredictor.scala
 create mode 100644 src/main/scala/frontend/Frontend.scala
 create mode 100644 src/main/scala/frontend/ICache.scala
 create mode 100644 src/main/scala/frontend/ITLB.scala
 create mode 100644 src/main/scala/frontend/RAS.scala
 create mode 100644 src/main/scala/issue/IssueQueue.scala
 create mode 100644 src/main/scala/issue/IssueStage.scala
 create mode 100644 src/main/scala/issue/ReservationStation.scala
 create mode 100644 src/main/scala/memory/DCache.scala
 create mode 100644 src/main/scala/memory/DTLB.scala
 create mode 100644 src/main/scala/memory/LSU.scala
 create mode 100644 src/main/scala/memory/LoadQueue.scala
 create mode 100644 src/main/scala/memory/MMU.scala
 create mode 100644 src/main/scala/memory/MemStage.scala
 create mode 100644 src/main/scala/memory/StoreQueue.scala
 create mode 100644 src/main/scala/regfile/PhysicalRegFile.scala
 create mode 100644 src/main/scala/rename/FreeList.scala
 create mode 100644 src/main/scala/rename/ROB.scala
 create mode 100644 src/main/scala/rename/RenameStage.scala
 create mode 100644 src/main/scala/rename/RenameTable.scala
 create mode 100644 src/main/scala/writeback/WriteBackStage.scala

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..e8e86ac
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,25 @@
+# SBT specific
+target/
+project/target/
+project/project/
+
+# IDE
+.idea/
+.vscode/
+.bsp/
+.metals/
+.bloop/
+metals.sbt
+
+# Scala
+*.class
+*.log
+
+# Generated files
+generated/
+sim/verilator/obj_dir/
+sim/scripts/test_results.txt
+riscv-tests/
+
+# toolchain
+toolchain/
diff --git a/ARCHITECTURE.md b/ARCHITECTURE.md
new file mode 100644
index 0000000..eb84866
--- /dev/null
+++ b/ARCHITECTURE.md
@@ -0,0 +1,466 @@
+# RISC-V RV64G 双发射乱序处理器架构设计
+
+## 目标规格
+- **ISA**: RV64IMAFD (RV64G without C extension)
+- **Issue Width**: 双发射（每周期2条指令）
+- **Physical Registers**: 64个物理寄存器
+- **Branch Predictor**: Gshare
+- **Memory**: Sv39分页，分离的L1 ICache/DCache
+- **Pipeline**: 11级流水线，乱序执行
+
+---
+
+## 1. 流水线架构
+
+### 流水线阶段
+
+```
+┌─────┐   ┌─────┐   ┌────┐   ┌────┐   ┌────┐   ┌────┐   ┌─────┐   ┌─────┐   ┌─────┐   ┌────┐
+│ IF1 │ → │ IF2 │ → │ ID │ → │ RN │ → │ DP │ → │ IS │ → │ EX1 │ → │ EX2 │ → │ MEM │ → │ WB │ → │ CM │
+└─────┘   └─────┘   └────┘   └────┘   └────┘   └────┘   └─────┘   └─────┘   └─────┘   └────┘
+  TLB      ICache    Decode  Rename  Dispatch  Issue    Execute   Execute   DCache    Write   Commit
+  BPU                2-way                                                    Access    Back    ROB
+```
+
+### 各阶段功能
+
+1. **IF1 (Instruction Fetch 1)**
+   - PC生成与选择
+   - ITLB查询（虚拟地址 → 物理地址）
+   - Gshare分支预测
+   - 取两条指令（双发射）
+
+2. **IF2 (Instruction Fetch 2)**
+   - ICache访问（64B缓存行）
+   - 指令对齐与缓冲
+
+3. **ID (Instruction Decode)**
+   - 双路译码器（每周期译码2条指令）
+   - 指令类型识别
+   - 立即数生成
+   - 依赖分析（RAW检测）
+
+4. **RN (Rename)**
+   - 寄存器重命名（32逻辑寄存器 → 64物理寄存器）
+   - 空闲列表(Free List)管理
+   - 重命名表(Rename Map Table)更新
+   - ROB分配
+
+5. **DP (Dispatch)**
+   - 分发到保留站(Reservation Station)
+   - 根据指令类型选择功能单元队列
+
+6. **IS (Issue)**
+   - 从保留站唤醒就绪指令
+   - 乱序发射到执行单元
+   - 读取物理寄存器文件
+
+7. **EX1 (Execute 1)**
+   - ALU运算
+   - 分支计算
+   - 地址生成（load/store）
+   - 浮点运算第一阶段
+
+8. **EX2 (Execute 2)**
+   - 多周期运算完成
+   - 浮点运算完成
+   - 分支结果检查
+
+9. **MEM (Memory)**
+   - DTLB查询
+   - DCache访问
+   - Load/Store执行
+   - Store Buffer管理
+
+10. **WB (Write Back)**
+    - 写回物理寄存器
+    - 唤醒依赖指令
+    - 更新ROB状态
+
+11. **CM (Commit)**
+    - ROB按序提交
+    - 架构状态更新
+    - 异常/中断处理
+    - 物理寄存器释放
+
+---
+
+## 2. 核心部件设计
+
+### 2.1 前端 (Front-End)
+
+#### Gshare 分支预测器
+
+```
+├── Global History Register (GHR): 12-bit
+├── Pattern History Table (PHT): 4096 entries × 2-bit saturating counter
+├── Branch Target Buffer (BTB): 512 entries
+│   ├── Tag: 20-bit
+│   ├── Target Address: 64-bit
+│   └── Valid bit
+└── Return Address Stack (RAS): 16 entries
+```
+
+**Gshare索引计算**: `index = (PC[13:2] XOR GHR[11:0])`
+
+要求：Branch Target Buffer大小可配置；Return Address Stack大小可配置
+
+#### ICache
+```
+├── Capacity: 32 KB
+├── Associativity: 4-way set-associative
+├── Line Size: 64 bytes
+├── Sets: 128
+└── Replacement: Pseudo-LRU
+```
+
+要求：Capacity大小可配置；Associativity大小可配置
+
+#### ITLB
+```
+├── Entries: 32 (Fully associative)
+├── Page Size: 4KB (Sv39)
+├── Replacement: LRU
+└── Support: Mega pages (2MB), Giga pages (1GB)
+```
+
+要求：Entries大小可配置
+
+### 2.2 乱序执行核心
+
+#### 重命名逻辑
+```
+Rename Map Table (RMT)
+├── Arch Regs: 32 (x0-x31)
+├── Phys Regs: 64 (p0-p63)
+└── Mapping: arch_reg → phys_reg
+
+Free List
+├── Available physical registers
+└── FIFO allocation
+
+Committed Map Table (CMT)
+└── For precise exception recovery
+```
+
+#### Reorder Buffer (ROB)
+```
+├── Entries: 64
+├── Fields per entry:
+│   ├── Valid
+│   ├── PC
+│   ├── Instruction Type
+│   ├── Destination Physical Register
+│   ├── Old Physical Register (for freeing)
+│   ├── Exception Info
+│   ├── Completed
+│   └── Branch Mispredict
+├── Head Pointer (commit)
+└── Tail Pointer (allocate)
+```
+
+要求：Entries可配置
+
+#### Reservation Stations
+```
+Integer RS
+├── Entries: 16
+└── ALU, Branch, Integer Multiply/Divide
+
+Load/Store RS
+├── Entries: 12
+└── Address generation, Memory ops
+
+FP RS
+├── Entries: 16
+└── FP Add, FP Multiply, FP Divide
+```
+
+#### Physical Register File
+```
+Integer PRF
+├── Registers: 64 × 64-bit
+└── Read Ports: 4 (dual-issue × 2 operands)
+└── Write Ports: 2 (dual-issue)
+
+FP PRF
+├── Registers: 64 × 64-bit
+└── Read Ports: 4
+└── Write Ports: 2
+```
+
+### 2.3 执行单元
+
+```
+├── ALU0: Integer ALU (ADD, SUB, AND, OR, XOR, SLT, etc.)
+├── ALU1: Integer ALU
+├── Branch Unit: Branch comparison
+├── AGU: Address Generation Unit (Load/Store)
+├── MDU: Multiply/Divide Unit (3-cycle latency for MUL, iterative for DIV)
+├── FPU0: FP Add/Sub (4-cycle latency)
+├── FPU1: FP Multiply (5-cycle latency)
+└── FPU2: FP Divide/Sqrt (iterative, variable latency)
+```
+
+### 2.4 后端 (Memory Subsystem)
+
+#### DCache
+```
+├── Capacity: 32 KB
+├── Associativity: 8-way set-associative
+├── Line Size: 64 bytes
+├── Sets: 64
+├── Replacement: Pseudo-LRU
+├── Write Policy: Write-back
+└── MSHR: 4 entries (Miss Status Holding Registers)
+```
+
+要求：Capacity大小可配置；Associativity大小可配置
+
+#### DTLB
+```
+├── Entries: 32 (Fully associative)
+├── Page Size: 4KB (Sv39)
+├── Replacement: LRU
+└── Support: Mega pages (2MB), Giga pages (1GB)
+```
+
+要求：Entries大小可配置
+
+#### MMU (Sv39)
+```
+├── Virtual Address: 39-bit
+├── Physical Address: 56-bit (RISC-V spec allows up to 56)
+├── Page Table Walker:
+│   ├── 3-level page table
+│   ├── Hardware page table walk
+│   └── PTE cache (8 entries per level)
+└── CSR Support:
+    ├── satp (Supervisor Address Translation and Protection)
+    ├── sstatus, sie, sip, stvec
+    └── sepc, scause, stval, sscratch
+```
+
+#### Load/Store Queue
+```
+Load Queue (LQ)
+├── Entries: 16
+└── Track in-flight loads
+
+Store Queue (SQ)
+├── Entries: 16
+├── Store-to-Load forwarding
+└── Drain on fence/commit
+```
+
+要求：Entries大小可配置
+
+### 2.5 提交阶段
+
+```
+Commit Logic
+├── Commit Width: 2 instructions per cycle
+├── In-order commit from ROB head
+├── Exception handling:
+│   ├── Flush pipeline
+│   ├── Restore architectural state from CMT
+│   └── Free speculative physical registers
+└── Branch mispredict recovery:
+    ├── Flush younger instructions
+    ├── Restore RMT from ROB checkpoint
+    └── Redirect fetch
+```
+
+---
+
+## 3. 数据通路
+
+### 3.1 指令流
+```
+PC → ITLB → ICache → Instruction Buffer → Decoder (×2) → Rename → 
+Dispatch → Reservation Stations → Issue → Execute → Write Back → ROB → Commit
+```
+
+### 3.2 数据流
+```
+Read PRF → Execute → Write Result → Broadcast (Bypass) → Write PRF → Update ROB
+                                                        └→ Wakeup dependent instructions
+```
+
+### 3.3 控制流
+```
+Branch Predictor → Speculative Fetch → Execute Branch → Compare → 
+[Match: Continue] / [Mispredict: Flush + Redirect]
+```
+
+---
+
+## 4. 关键设计决策
+
+### 4.1 双发射策略
+- **静态配对**: 在ID阶段检查指令间依赖
+- **限制**: 不允许同时发射两条分支指令、两条访存指令
+- **对齐**: 指令对必须来自连续的PC
+
+### 4.2 物理寄存器分配
+- **64个物理寄存器**: 32个用于映射当前架构状态，32个用于投机执行
+- **回收时机**: 指令提交时，释放旧的物理寄存器
+
+### 4.3 分支预测恢复
+- **检查点机制**: 每条分支在RN阶段创建RMT快照
+- **恢复**: 分支错误预测时，恢复到该分支的RMT快照
+
+### 4.4 内存一致性
+- **顺序模型**: RISC-V Weak Memory Order (RVWMO)
+- **Load乱序**: 可以越过Store执行（需要地址消歧）
+- **Store Buffer**: Store指令提交后写入，按序释放到Cache
+
+### 4.5 异常处理
+- **精确异常**: 通过ROB实现
+- **异常类型**: 
+  - Page Fault (ITLB/DTLB miss)
+  - Illegal Instruction
+  - Misaligned Access
+  - Breakpoint
+  - System Call
+
+---
+
+## 5. 性能参数估算
+
+| 指标 | 目标值 |
+|------|--------|
+| IPC (理想) | 1.8-2.0 |
+| IPC (实际) | 1.2-1.5 (考虑分支错误、Cache miss) |
+| 分支预测准确率 | >90% (Gshare) |
+| ICache命中率 | >95% |
+| DCache命中率 | >90% |
+| TLB命中率 | >98% |
+| 频率目标 | 100-200 MHz (FPGA) |
+
+---
+
+## 6. Linux启动需求
+
+要成功启动Linux，CPU必须支持：
+
+### 6.1 特权级
+- **M-mode** (Machine): 最高特权级，处理异常/中断
+- **S-mode** (Supervisor): Linux内核运行在此
+- **U-mode** (User): 用户程序
+
+### 6.2 必需的CSR
+```
+Machine Mode:
+- mstatus, misa, medeleg, mideleg
+- mtvec, mepc, mcause, mtval
+- mip, mie, mtime, mtimecmp
+
+Supervisor Mode:
+- sstatus, stvec, sepc, scause, stval
+- satp (MMU配置)
+- sip, sie
+
+Counters:
+- cycle, time, instret
+```
+
+### 6.3 中断/异常
+- Timer interrupt (CLINT)
+- External interrupt (PLIC)
+- Software interrupt
+- 所有RISC-V异常类型
+
+### 6.4 原子指令 (A扩展)
+- LR/SC (Load-Reserved/Store-Conditional)
+- AMO (Atomic Memory Operations): AMOSWAP, AMOADD, AMOXOR, AMOAND, AMOOR, AMOMIN, AMOMAX
+
+### 6.5 外设接口
+- UART (串口输出)
+- CLINT (核心本地中断器)
+- PLIC (平台级中断控制器)
+
+---
+
+## 7. 验证策略
+
+### 7.1 单元测试
+- 每个模块的功能验证
+- Chisel Unit Tests
+
+### 7.2 ISA测试
+- riscv-tests (官方ISA测试套件)
+- rv64ui, rv64um, rv64ua, rv64uf, rv64ud
+
+### 7.3 随机测试
+- riscv-torture (随机指令生成)
+
+### 7.4 系统测试
+- CoreMark, Dhrystone (性能基准测试)
+- Linux boot (最终目标)
+
+---
+
+## 8. Chisel实现模块划分
+
+```
+src/main/scala/
+├── common/
+│   ├── Consts.scala          // 常量定义
+│   ├── Parameters.scala      // 参数配置
+│   └── Bundles.scala         // 数据结构定义
+├── frontend/
+│   ├── ICache.scala
+│   ├── ITLB.scala
+│   ├── BranchPredictor.scala // Gshare
+│   ├── BTB.scala
+│   ├── RAS.scala
+│   └── Frontend.scala        // 前端顶层
+├── decode/
+│   ├── Decoder.scala
+│   └── IDStage.scala
+├── rename/
+│   ├── RenameTable.scala
+│   ├── FreeList.scala
+│   ├── ROB.scala
+│   └── RenameStage.scala
+├── issue/
+│   ├── ReservationStation.scala
+│   ├── IssueQueue.scala
+│   └── IssueStage.scala
+├── execute/
+│   ├── ALU.scala
+│   ├── BranchUnit.scala
+│   ├── Multiplier.scala
+│   ├── Divider.scala
+│   ├── FPU.scala
+│   └── ExecStage.scala
+├── memory/
+│   ├── DCache.scala
+│   ├── DTLB.scala
+│   ├── MMU.scala
+│   ├── LSU.scala             // Load/Store Unit
+│   ├── LoadQueue.scala
+│   ├── StoreQueue.scala
+│   └── MemStage.scala
+├── writeback/
+│   └── WriteBackStage.scala
+├── commit/
+│   └── CommitStage.scala
+├── regfile/
+│   └── PhysicalRegFile.scala
+├── csr/
+│   ├── CSRFile.scala
+│   └── PrivilegeControl.scala
+└── Core.scala                // 处理器顶层
+```
+
+---
+
+## 9. 参考资料
+
+- **RISC-V Spec**: https://riscv.org/technical/specifications/
+- **Sv39 Paging**: RISC-V Privileged Spec Chapter 4
+- **Gshare Predictor**: "Two-Level Adaptive Training Branch Prediction" (Yeh & Patt, 1991)
+- **BOOM Processor**: Berkeley Out-of-Order Machine (参考设计)
+- **Rocket Chip**: 顺序发射RISC-V处理器（前端参考）
diff --git a/build.sbt b/build.sbt
new file mode 100644
index 0000000..4923c12
--- /dev/null
+++ b/build.sbt
@@ -0,0 +1,22 @@
+ThisBuild / scalaVersion     := "2.13.18"
+ThisBuild / version          := "0.1.0"
+ThisBuild / organization     := "edu.riscv"
+
+val chiselVersion = "7.7.0"
+
+lazy val root = (project in file("."))
+  .settings(
+    name := "riscv-ooo-core",
+    libraryDependencies ++= Seq(
+      "org.chipsalliance" %% "chisel" % chiselVersion,
+      "org.scalatest" %% "scalatest" % "3.2.19" % "test",
+    ),
+    scalacOptions ++= Seq(
+      "-language:reflectiveCalls",
+      "-deprecation",
+      "-feature",
+      "-Xcheckinit",
+      "-Ymacro-annotations",
+    ),
+    addCompilerPlugin("org.chipsalliance" % "chisel-plugin" % chiselVersion cross CrossVersion.full),
+  )
diff --git a/project/build.properties b/project/build.properties
new file mode 100644
index 0000000..e8a1e24
--- /dev/null
+++ b/project/build.properties
@@ -0,0 +1 @@
+sbt.version=1.9.7
diff --git a/sim/README.md b/sim/README.md
new file mode 100644
index 0000000..4489e5e
--- /dev/null
+++ b/sim/README.md
@@ -0,0 +1,98 @@
+# RISC-V Processor Simulation Environment
+
+## 目录结构
+
+```
+sim/
+├── verilator/          # Verilator仿真环境
+│   ├── testbench.cpp   # 主测试平台
+│   ├── memory.cpp/h    # 内存模型
+│   └── Makefile        # 构建脚本
+├── scripts/            # 测试脚本
+│   └── run_tests.sh    # 批量测试运行器
+└── waves/              # 波形文件输出目录
+```
+
+## 快速开始
+
+### 1. 编译测试平台
+
+```bash
+cd sim/verilator
+make compile
+```
+
+这会：
+- 从Chisel生成Verilog代码
+- 使用Verilator编译C++测试平台
+
+### 2. 运行单个测试
+
+```bash
+make run TEST=../../riscv-tests/isa/rv64ui-p-add
+```
+
+或直接运行：
+
+```bash
+./obj_dir/VCore ../../riscv-tests/isa/rv64ui-p-add
+```
+
+### 3. 运行所有测试
+
+```bash
+cd ../scripts
+./run_tests.sh
+```
+
+查看结果：
+
+```bash
+cat test_results.txt
+```
+
+## 测试结果说明
+
+- **PASS**: 测试通过 (tohost写入1)
+- **FAIL**: 测试失败 (tohost写入错误码)
+- **TIMEOUT**: 测试超时 (超过100,000周期)
+
+## tohost/fromhost 协议
+
+- **tohost地址**: 0x80001000
+- **fromhost地址**: 0x80001040
+- **内存基址**: 0x80000000
+- **内存大小**: 64MB
+
+测试程序通过向tohost地址写入来报告结果：
+- `tohost = 1`: 测试通过
+- `tohost = (error_code << 1) | 1`: 测试失败
+
+## 依赖项
+
+- Verilator (>= 4.0)
+- Scala/SBT (Chisel编译)
+- C++14编译器
+- libelf-dev (ELF文件解析)
+
+## 故障排查
+
+### 编译错误
+
+如果Verilator编译失败，检查：
+1. Verilog是否正确生成在 `generated/` 目录
+2. Core模块的IO接口定义
+
+### 测试超时
+
+如果测试一直超时：
+1. 增加 `MAX_CYCLES` 在 testbench.cpp
+2. 检查处理器是否正确执行指令
+3. 验证内存接口握手逻辑
+
+### 内存访问错误
+
+检查：
+1. ELF加载是否成功
+2. 内存地址映射是否正确
+3. 链接脚本中的地址与 `MEM_BASE` 一致
diff --git a/sim/scripts/run_tests.sh b/sim/scripts/run_tests.sh
new file mode 100755
index 0000000..2741a8f
--- /dev/null
+++ b/sim/scripts/run_tests.sh
@@ -0,0 +1,61 @@
+#!/bin/bash
+
+TESTBENCH="../verilator/obj_dir/VCore"
+TESTS_DIR="../../riscv-tests/isa"
+RESULTS_FILE="test_results.txt"
+
+if [ ! -x "$TESTBENCH" ]; then
+    echo "Error: Testbench not found. Run 'make compile' first."
+    exit 1
+fi
+
+echo "Running RISC-V tests..." > $RESULTS_FILE
+echo "====================" >> $RESULTS_FILE
+
+PASS=0
+FAIL=0
+TIMEOUT=0
+
+for test in $TESTS_DIR/rv64ui-p-* $TESTS_DIR/rv64um-p-* $TESTS_DIR/rv64ua-p-*; do
+    # Skip dump files
+    [[ $test == *.dump ]] && continue
+    [ ! -f "$test" ] && continue
+
+    testname=$(basename $test)
+    echo -n "Running $testname... "
+
+    # Run with 5 second timeout
+    output=$(timeout 5s $TESTBENCH $test 2>&1)
+    exitcode=$?
+
+    if [ $exitcode -eq 0 ]; then
+        echo "PASS"
+        echo "$testname: PASS" >> $RESULTS_FILE
+        ((PASS++))
+    elif [ $exitcode -eq 124 ]; then
+        echo "TIMEOUT"
+        echo "$testname: TIMEOUT" >> $RESULTS_FILE
+        ((TIMEOUT++))
+    else
+        echo "FAIL (exit code $exitcode)"
+        echo "$testname: FAIL" >> $RESULTS_FILE
+        ((FAIL++))
+    fi
+done
+
+echo "" >> $RESULTS_FILE
+echo "====================" >> $RESULTS_FILE
+echo "Summary:" >> $RESULTS_FILE
+echo "  PASS:    $PASS" >> $RESULTS_FILE
+echo "  FAIL:    $FAIL" >> $RESULTS_FILE
+echo "  TIMEOUT: $TIMEOUT" >> $RESULTS_FILE
+echo "  TOTAL:   $((PASS + FAIL + TIMEOUT))" >> $RESULTS_FILE
+
+echo ""
+echo "Summary:"
+echo "  PASS:    $PASS"
+echo "  FAIL:    $FAIL"
+echo "  TIMEOUT: $TIMEOUT"
+echo "  TOTAL:   $((PASS + FAIL + TIMEOUT))"
+echo ""
+echo "Results saved to $RESULTS_FILE"
diff --git a/sim/verilator/Makefile b/sim/verilator/Makefile
new file mode 100644
index 0000000..2e638f0
--- /dev/null
+++ b/sim/verilator/Makefile
@@ -0,0 +1,43 @@
+VERILATOR = verilator
+VERILATOR_FLAGS = --cc --exe --build -Wall --trace -Wno-fatal
+VERILATOR_FLAGS += -CFLAGS "-std=c++14 -O2"
+SBT = env SBT_OPTS="-Dsbt.boot.directory=/tmp/sbt-boot -Dsbt.ivy.home=/tmp/sbt-ivy" COURSIER_CACHE=/tmp/coursier-cache sbt
+
+CHISEL_DIR = ../..
+GENERATED_DIR = $(CHISEL_DIR)/generated
+SRC_FILES = testbench.cpp memory.cpp
+
+VERILOG_FILES = $(GENERATED_DIR)/Core.sv
+
+TARGET = obj_dir/VCore
+
+.PHONY: all verilog compile run clean
+
+all: compile
+
+verilog:
+	@echo "Generating Verilog from Chisel..."
+	cd $(CHISEL_DIR) && $(SBT) "runMain Core"
+
+compile: verilog
+	@echo "Compiling with Verilator..."
+	$(VERILATOR) $(VERILATOR_FLAGS) \
+		-I$(GENERATED_DIR) \
+		--top-module Core \
+		-o VCore \
+		$(VERILOG_FILES) $(SRC_FILES)
+
+run: compile
+	@if [ -z "$(TEST)" ]; then \
+		echo "Usage: make run TEST=<path/to/test>"; \
+		exit 1; \
+	fi
+	./$(TARGET) $(TEST)
+
+test-simple: compile
+	@echo "Running simple test..."
+	./$(TARGET) ../../riscv-tests/isa/rv64ui-p-simple
+
+clean:
+	rm -rf obj_dir
+	rm -rf $(GENERATED_DIR)
diff --git a/sim/verilator/memory.cpp b/sim/verilator/memory.cpp
new file mode 100644
index 0000000..b62916d
--- /dev/null
+++ b/sim/verilator/memory.cpp
@@ -0,0 +1,111 @@
+#include "memory.h"
+#include <cstring>
+#include <cstdio>
+#include <elf.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <sys/stat.h>
+
+Memory::Memory() : base_addr(MEM_BASE), entry_point(MEM_BASE) {
+    mem.resize(MEM_SIZE, 0);
+}
+
+uint64_t Memory::to_offset(uint64_t addr) {
+    if (addr < base_addr || addr >= base_addr + MEM_SIZE) {
+        fprintf(stderr, "Memory access out of bounds: 0x%lx\n", addr);
+        return 0;
+    }
+    return addr - base_addr;
+}
+
+uint8_t Memory::read8(uint64_t addr) {
+    return mem[to_offset(addr)];
+}
+
+uint16_t Memory::read16(uint64_t addr) {
+    uint64_t off = to_offset(addr);
+    return mem[off] | (mem[off + 1] << 8);
+}
+
+uint32_t Memory::read32(uint64_t addr) {
+    uint64_t off = to_offset(addr);
+    return mem[off] | (mem[off + 1] << 8) | (mem[off + 2] << 16) | (mem[off + 3] << 24);
+}
+
+uint64_t Memory::read64(uint64_t addr) {
+    uint64_t off = to_offset(addr);
+    uint64_t lo = read32(addr);
+    uint64_t hi = read32(addr + 4);
+    return lo | (hi << 32);
+}
+
+void Memory::write8(uint64_t addr, uint8_t data) {
+    mem[to_offset(addr)] = data;
+}
+
+void Memory::write16(uint64_t addr, uint16_t data) {
+    uint64_t off = to_offset(addr);
+    mem[off] = data & 0xFF;
+    mem[off + 1] = (data >> 8) & 0xFF;
+}
+
+void Memory::write32(uint64_t addr, uint32_t data) {
+    uint64_t off = to_offset(addr);
+    mem[off] = data & 0xFF;
+    mem[off + 1] = (data >> 8) & 0xFF;
+    mem[off + 2] = (data >> 16) & 0xFF;
+    mem[off + 3] = (data >> 24) & 0xFF;
+}
+
+void Memory::write64(uint64_t addr, uint64_t data) {
+    write32(addr, data & 0xFFFFFFFF);
+    write32(addr + 4, (data >> 32) & 0xFFFFFFFF);
+}
+
+bool Memory::load_elf(const std::string& filename) {
+    int fd = open(filename.c_str(), O_RDONLY);
+    if (fd < 0) {
+        fprintf(stderr, "Failed to open ELF file: %s\n", filename.c_str());
+        return false;
+    }
+
+    Elf64_Ehdr ehdr;
+    if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr)) {
+        fprintf(stderr, "Failed to read ELF header\n");
+        close(fd);
+        return false;
+    }
+
+    if (memcmp(ehdr.e_ident, ELFMAG, SELFMAG) != 0) {
+        fprintf(stderr, "Not a valid ELF file\n");
+        close(fd);
+        return false;
+    }
+
+    entry_point = ehdr.e_entry;
+
+    // Load program headers
+    for (int i = 0; i < ehdr.e_phnum; i++) {
+        Elf64_Phdr phdr;
+        lseek(fd, ehdr.e_phoff + i * sizeof(phdr), SEEK_SET);
+        if (read(fd, &phdr, sizeof(phdr)) != sizeof(phdr)) {
+            fprintf(stderr, "Failed to read program header %d\n", i);
+            continue;
+        }
+
+        if (phdr.p_type == PT_LOAD) {
+            uint64_t offset = to_offset(phdr.p_paddr);
+            lseek(fd, phdr.p_offset, SEEK_SET);
+            if (read(fd, &mem[offset], phdr.p_filesz) != (ssize_t)phdr.p_filesz) {
+                fprintf(stderr, "Failed to load segment %d\n", i);
+                close(fd);
+                return false;
+            }
+            printf("Loaded segment: paddr=0x%lx size=%lu\n", phdr.p_paddr, phdr.p_filesz);
+        }
+    }
+
+    close(fd);
+    printf("ELF loaded: entry=0x%lx\n", entry_point);
+    return true;
+}
diff --git a/sim/verilator/memory.h b/sim/verilator/memory.h
new file mode 100644
index 0000000..0e9b23d
--- /dev/null
+++ b/sim/verilator/memory.h
@@ -0,0 +1,40 @@
+#ifndef MEMORY_H
+#define MEMORY_H
+
+#include <cstdint>
+#include <vector>
+#include <string>
+
+#define TOHOST_ADDR   0x80001000ULL
+#define FROMHOST_ADDR 0x80001040ULL
+#define MEM_BASE      0x80000000ULL
+#define MEM_SIZE      (64 * 1024 * 1024)  // 64MB
+
+class Memory {
+private:
+    std::vector<uint8_t> mem;
+    uint64_t base_addr;
+
+public:
+    Memory();
+
+    bool load_elf(const std::string& filename);
+
+    uint8_t read8(uint64_t addr);
+    uint16_t read16(uint64_t addr);
+    uint32_t read32(uint64_t addr);
+    uint64_t read64(uint64_t addr);
+
+    void write8(uint64_t addr, uint8_t data);
+    void write16(uint64_t addr, uint16_t data);
+    void write32(uint64_t addr, uint32_t data);
+    void write64(uint64_t addr, uint64_t data);
+
+    uint64_t get_entry_point() const { return entry_point; }
+
+private:
+    uint64_t entry_point;
+    uint64_t to_offset(uint64_t addr);
+};
+
+#endif
diff --git a/sim/verilator/testbench.cpp b/sim/verilator/testbench.cpp
new file mode 100644
index 0000000..352fcf0
--- /dev/null
+++ b/sim/verilator/testbench.cpp
@@ -0,0 +1,101 @@
+#include <verilated.h>
+#include "VCore.h"
+#include "memory.h"
+#include <cstdio>
+#include <cstdlib>
+
+#define MAX_CYCLES 100000
+
+int main(int argc, char** argv) {
+    if (argc < 2) {
+        fprintf(stderr, "Usage: %s <test.elf>\n", argv[0]);
+        return 1;
+    }
+
+    Verilated::commandArgs(argc, argv);
+
+    VCore* core = new VCore;
+    Memory* mem = new Memory();
+
+    if (!mem->load_elf(argv[1])) {
+        fprintf(stderr, "Failed to load test binary\n");
+        return 1;
+    }
+
+    // Reset
+    core->reset = 1;
+    core->clock = 0;
+    core->eval();
+    core->clock = 1;
+    core->eval();
+    core->clock = 0;
+    core->reset = 0;
+    core->eval();
+
+    uint64_t cycle = 0;
+    bool test_done = false;
+    int exit_code = 0;
+
+    while (cycle < MAX_CYCLES && !test_done) {
+        // Handle instruction memory interface
+        if (core->io_imem_req_valid) {
+            uint64_t pc = core->io_imem_req_bits;
+            core->io_imem_resp_valid = 1;
+            core->io_imem_resp_bits_0 = mem->read32(pc);
+            core->io_imem_resp_bits_1 = mem->read32(pc + 4);
+        } else {
+            core->io_imem_resp_valid = 0;
+        }
+
+        // Handle data memory interface
+        if (core->io_dmem_req_valid) {
+            uint64_t addr = core->io_dmem_req_bits_addr;
+
+            // Check for tohost write
+            if (core->io_dmem_req_bits_isStore && addr == TOHOST_ADDR) {
+                uint64_t tohost = core->io_dmem_req_bits_data;
+                if (tohost == 1) {
+                    printf("[%lu] TEST PASSED\n", cycle);
+                    test_done = true;
+                    exit_code = 0;
+                } else if (tohost & 1) {
+                    printf("[%lu] TEST FAILED: error code %lu\n", cycle, tohost >> 1);
+                    test_done = true;
+                    exit_code = 1;
+                }
+            }
+
+            if (core->io_dmem_req_bits_isStore) {
+                switch (core->io_dmem_req_bits_size) {
+                    case 0: mem->write8(addr, core->io_dmem_req_bits_data & 0xff); break;
+                    case 1: mem->write16(addr, core->io_dmem_req_bits_data & 0xffff); break;
+                    case 2: mem->write32(addr, core->io_dmem_req_bits_data & 0xffffffff); break;
+                    default: mem->write64(addr, core->io_dmem_req_bits_data); break;
+                }
+                core->io_dmem_resp_valid = 0;
+            } else {
+                core->io_dmem_resp_bits = mem->read64(addr);
+                core->io_dmem_resp_valid = 1;
+            }
+        } else {
+            core->io_dmem_resp_valid = 0;
+        }
+
+        // Clock cycle
+        core->clock = 0;
+        core->eval();
+        core->clock = 1;
+        core->eval();
+
+        cycle++;
+    }
+
+    if (!test_done) {
+        printf("[%lu] TEST TIMEOUT\n", cycle);
+        exit_code = 2;
+    }
+
+    delete core;
+    delete mem;
+    return exit_code;
+}
diff --git a/src/main/scala/Core.scala b/src/main/scala/Core.scala
new file mode 100644
index 0000000..a4f63b2
--- /dev/null
+++ b/src/main/scala/Core.scala
@@ -0,0 +1,173 @@
+import chisel3._
+import chisel3.util._
+import _root_.circt.stage.ChiselStage
+
+class Core(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val imem_req_valid = Output(Bool())
+    val imem_req_bits = Output(UInt(p.xlen.W))
+    val imem_resp_valid = Input(Bool())
+    val imem_resp_bits_0 = Input(UInt(32.W))
+    val imem_resp_bits_1 = Input(UInt(32.W))
+
+    val dmem_req_valid = Output(Bool())
+    val dmem_req_bits_addr = Output(UInt(p.xlen.W))
+    val dmem_req_bits_data = Output(UInt(p.xlen.W))
+    val dmem_req_bits_isStore = Output(Bool())
+    val dmem_req_bits_size = Output(UInt(3.W))
+    val dmem_resp_valid = Input(Bool())
+    val dmem_resp_bits = Input(UInt(p.xlen.W))
+  })
+
+  val sFetch :: sExec :: sLoadWait :: Nil = Enum(3)
+  val state = RegInit(sFetch)
+  val pc = RegInit(Consts.ResetVector)
+  val instReg = RegInit(0.U(32.W))
+  val pcReg = RegInit(Consts.ResetVector)
+
+  val regs = RegInit(VecInit(Seq.fill(32)(0.U(p.xlen.W))))
+
+  val decoder = Module(new Decoder(p))
+  decoder.io.pc := pcReg
+  decoder.io.inst := instReg
+  val dec = decoder.io.out
+
+  val alu = Module(new ALU(p))
+  val branch = Module(new BranchUnit(p))
+  val csr = Module(new CSRFile(p))
+
+  def regRead(addr: UInt): UInt = Mux(addr === 0.U, 0.U, regs(addr))
+  def lowLoad(data: UInt, size: UInt, signed: Bool): UInt = {
+    val b = data(7, 0)
+    val h = data(15, 0)
+    val w = data(31, 0)
+    MuxLookup(size, data)(Seq(
+      0.U -> Mux(signed, Consts.signExtend(b, 8), b),
+      1.U -> Mux(signed, Consts.signExtend(h, 16), h),
+      2.U -> Mux(signed, Consts.signExtend(w, 32), w),
+      3.U -> data
+    ))
+  }
+
+  val src1 = regRead(dec.rs1)
+  val src2 = regRead(dec.rs2)
+  val aluB = Mux(dec.isOpImm || dec.isJalr || dec.isLoad, dec.immI, src2)
+
+  alu.io.fn := dec.aluFn
+  alu.io.a := src1
+  alu.io.b := aluB
+  alu.io.isWord := dec.isWord
+  branch.io.funct3 := dec.funct3
+  branch.io.a := src1
+  branch.io.b := src2
+
+  csr.io.cmd.valid := state === sExec && dec.isSystem && dec.funct3 =/= 0.U &&
+    !(dec.funct3(1) && dec.rs1 === 0.U)
+  csr.io.cmd.addr := instReg(31, 20)
+  csr.io.cmd.cmd := dec.funct3
+  csr.io.cmd.rs1 := src1
+  csr.io.cmd.zimm := dec.rs1
+  val isEcall = instReg === "h00000073".U
+  val isEbreak = instReg === "h00100073".U
+  val isMret = instReg === "h30200073".U
+  val takeTrap = state === sExec && (isEcall || isEbreak)
+  csr.io.trap := takeTrap
+  csr.io.trapPc := pcReg
+  csr.io.trapCause := Mux(isEbreak, 3.U, 11.U)
+
+  val loadAddr = src1 + dec.immI
+  val storeAddr = src1 + dec.immS
+  val pendingLoadAddr = Reg(UInt(p.xlen.W))
+  val pendingLoadRd = Reg(UInt(5.W))
+  val pendingLoadSize = Reg(UInt(3.W))
+  val pendingLoadSigned = Reg(Bool())
+
+  val isAmo = instReg(6, 0) === "b0101111".U
+  val amoOp = instReg(31, 27)
+  val amoLoadLike = isAmo && amoOp === "b00010".U
+  val amoStoreLike = isAmo && amoOp === "b00011".U
+
+  val memReqInExec = state === sExec && (dec.isLoad || dec.isStore)
+  io.imem_req_valid := state === sFetch
+  io.imem_req_bits := pc
+  io.dmem_req_valid := memReqInExec || state === sLoadWait
+  io.dmem_req_bits_addr := Mux(state === sLoadWait, pendingLoadAddr, Mux(dec.isStore, storeAddr, loadAddr))
+  io.dmem_req_bits_data := src2
+  io.dmem_req_bits_isStore := state === sExec && Mux(isAmo, !amoLoadLike, dec.isStore)
+  io.dmem_req_bits_size := Mux(state === sLoadWait, pendingLoadSize, dec.memWidth)
+
+  val branchTarget = pcReg + dec.immB
+  val jalTarget = pcReg + dec.immJ
+  val jalrTarget = (src1 + dec.immI) & (~1.U(p.xlen.W))
+  val branchTaken = dec.isBranch && branch.io.taken
+  val nextPc = Mux(dec.isJal, jalTarget,
+    Mux(dec.isJalr, jalrTarget,
+      Mux(branchTaken, branchTarget, pcReg + 4.U)))
+
+  val execWriteData = WireDefault(alu.io.out)
+  when(dec.isLui) {
+    execWriteData := dec.immU
+  }.elsewhen(dec.isAuipc) {
+    execWriteData := pcReg + dec.immU
+  }.elsewhen(dec.isJal || dec.isJalr) {
+    execWriteData := pcReg + 4.U
+  }.elsewhen(dec.isSystem && dec.funct3 =/= 0.U) {
+    execWriteData := csr.io.rdata
+  }
+
+  when(state === sFetch) {
+    when(io.imem_resp_valid) {
+      instReg := io.imem_resp_bits_0
+      pcReg := pc
+      state := sExec
+    }
+  }.elsewhen(state === sExec) {
+    when(takeTrap) {
+      pc := csr.io.mtvec
+      state := sFetch
+    }.elsewhen(isMret) {
+      pc := csr.io.mepc
+      state := sFetch
+    }.elsewhen(dec.isLoad && !amoStoreLike) {
+      when(io.dmem_resp_valid) {
+        when(dec.rd =/= 0.U) {
+          regs(dec.rd) := lowLoad(io.dmem_resp_bits, dec.memWidth, dec.memSigned)
+        }
+        pc := pcReg + 4.U
+        state := sFetch
+      }.otherwise {
+        pendingLoadAddr := loadAddr
+        pendingLoadRd := dec.rd
+        pendingLoadSize := dec.memWidth
+        pendingLoadSigned := dec.memSigned
+        state := sLoadWait
+      }
+    }.elsewhen(dec.isStore || amoStoreLike) {
+      pc := pcReg + 4.U
+      state := sFetch
+    }.otherwise {
+      when(dec.writesRd && dec.rd =/= 0.U) {
+        regs(dec.rd) := execWriteData
+      }
+      pc := nextPc
+      state := sFetch
+    }
+  }.elsewhen(state === sLoadWait) {
+    when(io.dmem_resp_valid) {
+      when(pendingLoadRd =/= 0.U) {
+        regs(pendingLoadRd) := lowLoad(io.dmem_resp_bits, pendingLoadSize, pendingLoadSigned)
+      }
+      pc := pcReg + 4.U
+      state := sFetch
+    }
+  }
+
+}
+
+object Core extends App {
+  ChiselStage.emitSystemVerilogFile(
+    new Core(),
+    args = Array("--target-dir", "generated"),
+    firtoolOpts = Array("-disable-all-randomization", "-strip-debug-info")
+  )
+}
diff --git a/src/main/scala/commit/CommitStage.scala b/src/main/scala/commit/CommitStage.scala
new file mode 100644
index 0000000..031ee93
--- /dev/null
+++ b/src/main/scala/commit/CommitStage.scala
@@ -0,0 +1,18 @@
+import chisel3._
+import chisel3.util._
+
+class CommitStage(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val robValid = Input(Bool())
+    val robEntry = Input(new RobEntry(p))
+    val commitReady = Output(Bool())
+    val freeOldPhys = Output(Bool())
+    val oldPhys = Output(UInt(log2Ceil(p.physRegs).W))
+    val flush = Output(Bool())
+  })
+
+  io.commitReady := io.robValid
+  io.freeOldPhys := io.robValid && io.robEntry.oldDest =/= io.robEntry.dest
+  io.oldPhys := io.robEntry.oldDest
+  io.flush := io.robValid && (io.robEntry.exception || io.robEntry.branchMispredict)
+}
diff --git a/src/main/scala/common/Bundles.scala b/src/main/scala/common/Bundles.scala
new file mode 100644
index 0000000..6a037ac
--- /dev/null
+++ b/src/main/scala/common/Bundles.scala
@@ -0,0 +1,76 @@
+import chisel3._
+import chisel3.util._
+
+class FetchPacket(p: CoreParams = CoreParams()) extends Bundle {
+  val pc = UInt(p.xlen.W)
+  val inst = Vec(p.fetchWidth, UInt(32.W))
+  val predictedTaken = Bool()
+  val predictedTarget = UInt(p.xlen.W)
+}
+
+class BranchUpdate(p: CoreParams = CoreParams()) extends Bundle {
+  val valid = Bool()
+  val pc = UInt(p.xlen.W)
+  val taken = Bool()
+  val target = UInt(p.xlen.W)
+  val isCall = Bool()
+  val isReturn = Bool()
+}
+
+class DecodedInst(p: CoreParams = CoreParams()) extends Bundle {
+  val valid = Bool()
+  val pc = UInt(p.xlen.W)
+  val inst = UInt(32.W)
+  val rs1 = UInt(5.W)
+  val rs2 = UInt(5.W)
+  val rd = UInt(5.W)
+  val funct3 = UInt(3.W)
+  val funct7 = UInt(7.W)
+  val immI = UInt(p.xlen.W)
+  val immS = UInt(p.xlen.W)
+  val immB = UInt(p.xlen.W)
+  val immU = UInt(p.xlen.W)
+  val immJ = UInt(p.xlen.W)
+  val opClass = UInt(Consts.OpClassWidth.W)
+  val aluFn = UInt(5.W)
+  val memWidth = UInt(3.W)
+  val memSigned = Bool()
+  val isLoad = Bool()
+  val isStore = Bool()
+  val isBranch = Bool()
+  val isJal = Bool()
+  val isJalr = Bool()
+  val isLui = Bool()
+  val isAuipc = Bool()
+  val isOpImm = Bool()
+  val isOp = Bool()
+  val isWord = Bool()
+  val isSystem = Bool()
+  val writesRd = Bool()
+  val illegal = Bool()
+}
+
+class RenamePacket(p: CoreParams = CoreParams()) extends Bundle {
+  val valid = Bool()
+  val decoded = new DecodedInst(p)
+  val prs1 = UInt(log2Ceil(p.physRegs).W)
+  val prs2 = UInt(log2Ceil(p.physRegs).W)
+  val prd = UInt(log2Ceil(p.physRegs).W)
+  val oldPrd = UInt(log2Ceil(p.physRegs).W)
+  val robIdx = UInt(log2Ceil(p.robEntries).W)
+}
+
+class MemRequest(p: CoreParams = CoreParams()) extends Bundle {
+  val addr = UInt(p.xlen.W)
+  val data = UInt(p.xlen.W)
+  val isStore = Bool()
+  val size = UInt(3.W)
+}
+
+class CsrCommand(p: CoreParams = CoreParams()) extends Bundle {
+  val valid = Bool()
+  val addr = UInt(12.W)
+  val cmd = UInt(3.W)
+  val rs1 = UInt(p.xlen.W)
+  val zimm = UInt(5.W)
+}
diff --git a/src/main/scala/common/Consts.scala b/src/main/scala/common/Consts.scala
new file mode 100644
index 0000000..0f8ca07
--- /dev/null
+++ b/src/main/scala/common/Consts.scala
@@ -0,0 +1,35 @@
+import chisel3._
+import chisel3.util._
+
+object Consts {
+  val ResetVector = "h80000000".U(64.W)
+
+  val OpClassWidth = 4
+  val OP_NONE = 0.U(OpClassWidth.W)
+  val OP_ALU = 1.U(OpClassWidth.W)
+  val OP_BRANCH = 2.U(OpClassWidth.W)
+  val OP_LOAD = 3.U(OpClassWidth.W)
+  val OP_STORE = 4.U(OpClassWidth.W)
+  val OP_SYSTEM = 5.U(OpClassWidth.W)
+  val OP_FP = 6.U(OpClassWidth.W)
+
+  val ALU_ADD = 0.U(5.W)
+  val ALU_SUB = 1.U(5.W)
+  val ALU_SLL = 2.U(5.W)
+  val ALU_SLT = 3.U(5.W)
+  val ALU_SLTU = 4.U(5.W)
+  val ALU_XOR = 5.U(5.W)
+  val ALU_SRL = 6.U(5.W)
+  val ALU_SRA = 7.U(5.W)
+  val ALU_OR = 8.U(5.W)
+  val ALU_AND = 9.U(5.W)
+  val ALU_MUL = 10.U(5.W)
+  val ALU_DIV = 11.U(5.W)
+  val ALU_DIVU = 12.U(5.W)
+  val ALU_REM = 13.U(5.W)
+  val ALU_REMU = 14.U(5.W)
+  val ALU_COPY_B = 15.U(5.W)
+
+  def signExtend(value: UInt, from: Int, to: Int = 64): UInt =
+    Cat(Fill(to - from, value(from - 1)), value(from - 1, 0))
+}
diff --git a/src/main/scala/common/Parameters.scala b/src/main/scala/common/Parameters.scala
new file mode 100644
index 0000000..6ec256f
--- /dev/null
+++ b/src/main/scala/common/Parameters.scala
@@ -0,0 +1,28 @@
+case class CoreParams(
+  xlen: Int = 64,
+  fetchWidth: Int = 2,
+  issueWidth: Int = 2,
+  archRegs: Int = 32,
+  physRegs: Int = 64,
+  robEntries: Int = 64,
+  intRsEntries: Int = 16,
+  lsuRsEntries: Int = 12,
+  fpRsEntries: Int = 16,
+  loadQueueEntries: Int = 16,
+  storeQueueEntries: Int = 16,
+  btbEntries: Int = 512,
+  rasEntries: Int = 16,
+  ghrBits: Int = 12,
+  iCacheBytes: Int = 32 * 1024,
+  iCacheWays: Int = 4,
+  dCacheBytes: Int = 32 * 1024,
+  dCacheWays: Int = 8,
+  cacheLineBytes: Int = 64,
+  itlbEntries: Int = 32,
+  dtlbEntries: Int = 32
+) {
+  require(xlen == 64, "this implementation targets RV64")
+  require(fetchWidth == 2, "frontend is parameterized around dual fetch")
+  require(issueWidth == 2, "backend structures are parameterized around dual issue")
+}
+
diff --git a/src/main/scala/csr/CSRFile.scala b/src/main/scala/csr/CSRFile.scala
new file mode 100644
index 0000000..09ed7c3
--- /dev/null
+++ b/src/main/scala/csr/CSRFile.scala
@@ -0,0 +1,99 @@
+import chisel3._
+import chisel3.util._
+
+class CSRFile(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val cmd = Input(new CsrCommand(p))
+    val rdata = Output(UInt(p.xlen.W))
+    val trap = Input(Bool())
+    val trapPc = Input(UInt(p.xlen.W))
+    val trapCause = Input(UInt(p.xlen.W))
+    val satp = Output(UInt(p.xlen.W))
+    val mtvec = Output(UInt(p.xlen.W))
+    val mepc = Output(UInt(p.xlen.W))
+  })
+
+  val cycle = RegInit(0.U(p.xlen.W))
+  val instret = RegInit(0.U(p.xlen.W))
+  val mstatus = RegInit(0.U(p.xlen.W))
+  val misa = RegInit("h800000000014112d".U(p.xlen.W))
+  val mtvecReg = RegInit(0.U(p.xlen.W))
+  val mepcReg = RegInit(0.U(p.xlen.W))
+  val mcause = RegInit(0.U(p.xlen.W))
+  val mtval = RegInit(0.U(p.xlen.W))
+  val medeleg = RegInit(0.U(p.xlen.W))
+  val mideleg = RegInit(0.U(p.xlen.W))
+  val mie = RegInit(0.U(p.xlen.W))
+  val mip = RegInit(0.U(p.xlen.W))
+  val sstatus = RegInit(0.U(p.xlen.W))
+  val stvec = RegInit(0.U(p.xlen.W))
+  val sepc = RegInit(0.U(p.xlen.W))
+  val scause = RegInit(0.U(p.xlen.W))
+  val stval = RegInit(0.U(p.xlen.W))
+  val sscratch = RegInit(0.U(p.xlen.W))
+  val satpReg = RegInit(0.U(p.xlen.W))
+
+  cycle := cycle + 1.U
+  io.satp := satpReg
+  io.mtvec := mtvecReg
+  io.mepc := mepcReg
+
+  val r = WireDefault(0.U(p.xlen.W))
+  switch(io.cmd.addr) {
+    is("h300".U) { r := mstatus }
+    is("h301".U) { r := misa }
+    is("h302".U) { r := medeleg }
+    is("h303".U) { r := mideleg }
+    is("h304".U) { r := mie }
+    is("h305".U) { r := mtvecReg }
+    is("h341".U) { r := mepcReg }
+    is("h342".U) { r := mcause }
+    is("h343".U) { r := mtval }
+    is("h344".U) { r := mip }
+    is("h100".U) { r := sstatus }
+    is("h105".U) { r := stvec }
+    is("h140".U) { r := sscratch }
+    is("h141".U) { r := sepc }
+    is("h142".U) { r := scause }
+    is("h143".U) { r := stval }
+    is("h180".U) { r := satpReg }
+    is("hf14".U) { r := 0.U }
+    is("hc00".U) { r := cycle }
+    is("hc01".U) { r := 0.U }
+    is("hc02".U) { r := instret }
+  }
+  io.rdata := r
+
+  val operand = Mux(io.cmd.cmd(2), io.cmd.zimm, io.cmd.rs1)
+  val next = MuxLookup(io.cmd.cmd(1, 0), r)(Seq(
+    1.U -> operand,
+    2.U -> (r | operand),
+    3.U -> (r & ~operand)
+  ))
+
+  when(io.cmd.valid && io.cmd.cmd =/= 0.U) {
+    switch(io.cmd.addr) {
+      is("h300".U) { mstatus := next }
+      is("h302".U) { medeleg := next }
+      is("h303".U) { mideleg := next }
+      is("h304".U) { mie := next }
+      is("h305".U) { mtvecReg := next }
+      is("h341".U) { mepcReg := next }
+      is("h342".U) { mcause := next }
+      is("h343".U) { mtval := next }
+      is("h344".U) { mip := next }
+      is("h100".U) { sstatus := next }
+      is("h105".U) { stvec := next }
+      is("h140".U) { sscratch := next }
+      is("h141".U) { sepc := next }
+      is("h142".U) { scause := next }
+      is("h143".U) { stval := next }
+      is("h180".U) { satpReg := next }
+    }
+  }
+
+  when(io.trap) {
+    mepcReg := io.trapPc
+    mcause := io.trapCause
+  }
+}
diff --git a/src/main/scala/csr/PrivilegeControl.scala b/src/main/scala/csr/PrivilegeControl.scala
new file mode 100644
index 0000000..64a5f4d
--- /dev/null
+++ b/src/main/scala/csr/PrivilegeControl.scala
@@ -0,0 +1,21 @@
+import chisel3._
+
+class PrivilegeControl(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val trap = Input(Bool())
+    val mret = Input(Bool())
+    val sret = Input(Bool())
+    val privilege = Output(UInt(2.W))
+  })
+
+  val mode = RegInit(3.U(2.W))
+  when(io.trap) {
+    mode := 3.U
+  }.elsewhen(io.mret) {
+    mode := 0.U
+  }.elsewhen(io.sret) {
+    mode := 0.U
+  }
+  io.privilege := mode
+}
+
diff --git a/src/main/scala/decode/Decoder.scala b/src/main/scala/decode/Decoder.scala
new file mode 100644
index 0000000..8dfa67f
--- /dev/null
+++ b/src/main/scala/decode/Decoder.scala
@@ -0,0 +1,150 @@
+import chisel3._
+import chisel3.util._
+
+class Decoder(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val pc = Input(UInt(p.xlen.W))
+    val inst = Input(UInt(32.W))
+    val out = Output(new DecodedInst(p))
+  })
+
+  val opcode = io.inst(6, 0)
+  val rd = io.inst(11, 7)
+  val funct3 = io.inst(14, 12)
+  val rs1 = io.inst(19, 15)
+  val rs2 = io.inst(24, 20)
+  val funct7 = io.inst(31, 25)
+
+  val immI = Consts.signExtend(io.inst(31, 20), 12)
+  val immS = Consts.signExtend(Cat(io.inst(31, 25), io.inst(11, 7)), 12)
+  val immB = Consts.signExtend(Cat(io.inst(31), io.inst(7), io.inst(30, 25), io.inst(11, 8), 0.U(1.W)), 13)
+  val immU = Consts.signExtend(Cat(io.inst(31, 12), 0.U(12.W)), 32)
+  val immJ = Consts.signExtend(Cat(io.inst(31), io.inst(19, 12), io.inst(20), io.inst(30, 21), 0.U(1.W)), 21)
+
+  val d = WireDefault(0.U.asTypeOf(new DecodedInst(p)))
+  d.valid := true.B
+  d.pc := io.pc
+  d.inst := io.inst
+  d.rs1 := rs1
+  d.rs2 := rs2
+  d.rd := rd
+  d.funct3 := funct3
+  d.funct7 := funct7
+  d.immI := immI
+  d.immS := immS
+  d.immB := immB
+  d.immU := immU
+  d.immJ := immJ
+  d.memWidth := MuxLookup(funct3, 3.U)(Seq(
+    "b000".U -> 0.U,
+    "b001".U -> 1.U,
+    "b010".U -> 2.U,
+    "b011".U -> 3.U,
+    "b100".U -> 0.U,
+    "b101".U -> 1.U,
+    "b110".U -> 2.U
+  ))
+  d.memSigned := !funct3(2)
+  d.illegal := false.B
+
+  switch(opcode) {
+    is("b0110111".U) {
+      d.isLui := true.B
+      d.writesRd := rd =/= 0.U
+      d.opClass := Consts.OP_ALU
+      d.aluFn := Consts.ALU_COPY_B
+    }
+    is("b0010111".U) {
+      d.isAuipc := true.B
+      d.writesRd := rd =/= 0.U
+      d.opClass := Consts.OP_ALU
+      d.aluFn := Consts.ALU_ADD
+    }
+    is("b1101111".U) {
+      d.isJal := true.B
+      d.writesRd := rd =/= 0.U
+      d.opClass := Consts.OP_BRANCH
+    }
+    is("b1100111".U) {
+      d.isJalr := true.B
+      d.writesRd := rd =/= 0.U
+      d.opClass := Consts.OP_BRANCH
+    }
+    is("b1100011".U) {
+      d.isBranch := true.B
+      d.opClass := Consts.OP_BRANCH
+    }
+    is("b0000011".U) {
+      d.isLoad := true.B
+      d.writesRd := rd =/= 0.U
+      d.opClass := Consts.OP_LOAD
+    }
+    is("b0100011".U) {
+      d.isStore := true.B
+      d.opClass := Consts.OP_STORE
+    }
+    is("b0010011".U, "b0011011".U) {
+      d.isOpImm := true.B
+      d.isWord := opcode === "b0011011".U
+      d.writesRd := rd =/= 0.U
+      d.opClass := Consts.OP_ALU
+      d.aluFn := MuxLookup(funct3, Consts.ALU_ADD)(Seq(
+        "b000".U -> Consts.ALU_ADD,
+        "b001".U -> Consts.ALU_SLL,
+        "b010".U -> Consts.ALU_SLT,
+        "b011".U -> Consts.ALU_SLTU,
+        "b100".U -> Consts.ALU_XOR,
+        "b101".U -> Mux(funct7(5), Consts.ALU_SRA, Consts.ALU_SRL),
+        "b110".U -> Consts.ALU_OR,
+        "b111".U -> Consts.ALU_AND
+      ))
+    }
+    is("b0110011".U, "b0111011".U) {
+      d.isOp := true.B
+      d.isWord := opcode === "b0111011".U
+      d.writesRd := rd =/= 0.U
+      d.opClass := Consts.OP_ALU
+      d.aluFn := Mux(funct7 === "b0000001".U, MuxLookup(funct3, Consts.ALU_MUL)(Seq(
+        "b000".U -> Consts.ALU_MUL,
+        "b100".U -> Consts.ALU_DIV,
+        "b101".U -> Consts.ALU_DIVU,
+        "b110".U -> Consts.ALU_REM,
+        "b111".U -> Consts.ALU_REMU
+      )), MuxLookup(funct3, Consts.ALU_ADD)(Seq(
+        "b000".U -> Mux(funct7(5), Consts.ALU_SUB, Consts.ALU_ADD),
+        "b001".U -> Consts.ALU_SLL,
+        "b010".U -> Consts.ALU_SLT,
+        "b011".U -> Consts.ALU_SLTU,
+        "b100".U -> Consts.ALU_XOR,
+        "b101".U -> Mux(funct7(5), Consts.ALU_SRA, Consts.ALU_SRL),
+        "b110".U -> Consts.ALU_OR,
+        "b111".U -> Consts.ALU_AND
+      )))
+    }
+    is("b0001111".U) {
+      d.opClass := Consts.OP_SYSTEM
+    }
+    is("b1110011".U) {
+      d.isSystem := true.B
+      d.writesRd := rd =/= 0.U && funct3 =/= 0.U
+      d.opClass := Consts.OP_SYSTEM
+    }
+    is("b0101111".U) {
+      d.isLoad := true.B
+      d.isStore := true.B
+      d.writesRd := rd =/= 0.U
+      d.memWidth := Mux(funct3 === "b010".U, 2.U, 3.U)
+      d.opClass := Consts.OP_LOAD
+    }
+  }
+
+  when(opcode =/= "b0110111".U && opcode =/= "b0010111".U && opcode =/= "b1101111".U &&
+       opcode =/= "b1100111".U && opcode =/= "b1100011".U && opcode =/= "b0000011".U &&
+       opcode =/= "b0100011".U && opcode =/= "b0010011".U && opcode =/= "b0011011".U &&
+       opcode =/= "b0110011".U && opcode =/= "b0111011".U && opcode =/= "b0001111".U &&
+       opcode =/= "b1110011".U && opcode =/= "b0101111".U) {
+    d.illegal := true.B
+  }
+
+  io.out := d
+}
diff --git a/src/main/scala/decode/IDStage.scala b/src/main/scala/decode/IDStage.scala
new file mode 100644
index 0000000..8b4b8ce
--- /dev/null
+++ b/src/main/scala/decode/IDStage.scala
@@ -0,0 +1,20 @@
+import chisel3._
+
+class IDStage(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val inValid = Input(Bool())
+    val in = Input(new FetchPacket(p))
+    val outValid = Output(Vec(p.fetchWidth, Bool()))
+    val out = Output(Vec(p.fetchWidth, new DecodedInst(p)))
+  })
+
+  val decoders = Seq.fill(p.fetchWidth)(Module(new Decoder(p)))
+  for (i <- 0 until p.fetchWidth) {
+    decoders(i).io.pc := io.in.pc + (4 * i).U
+    decoders(i).io.inst := io.in.inst(i)
+    io.out(i) := decoders(i).io.out
+    io.out(i).valid := io.inValid
+    io.outValid(i) := io.inValid
+  }
+}
+
diff --git a/src/main/scala/execute/ALU.scala b/src/main/scala/execute/ALU.scala
new file mode 100644
index 0000000..3f8e753
--- /dev/null
+++ b/src/main/scala/execute/ALU.scala
@@ -0,0 +1,36 @@
+import chisel3._
+import chisel3.util._
+
+class ALU(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val fn = Input(UInt(5.W))
+    val a = Input(UInt(p.xlen.W))
+    val b = Input(UInt(p.xlen.W))
+    val isWord = Input(Bool())
+    val out = Output(UInt(p.xlen.W))
+  })
+
+  val shamt = Mux(io.isWord, io.b(4, 0), io.b(5, 0))
+  val raw = WireDefault(0.U(p.xlen.W))
+  switch(io.fn) {
+    is(Consts.ALU_ADD) { raw := io.a + io.b }
+    is(Consts.ALU_SUB) { raw := io.a - io.b }
+    is(Consts.ALU_SLL) { raw := io.a << shamt }
+    is(Consts.ALU_SLT) { raw := (io.a.asSInt < io.b.asSInt).asUInt }
+    is(Consts.ALU_SLTU) { raw := io.a < io.b }
+    is(Consts.ALU_XOR) { raw := io.a ^ io.b }
+    is(Consts.ALU_SRL) { raw := io.a >> shamt }
+    is(Consts.ALU_SRA) { raw := (io.a.asSInt >> shamt).asUInt }
+    is(Consts.ALU_OR) { raw := io.a | io.b }
+    is(Consts.ALU_AND) { raw := io.a & io.b }
+    is(Consts.ALU_MUL) { raw := (io.a * io.b)(p.xlen - 1, 0) }
+    is(Consts.ALU_DIV) { raw := Mux(io.b === 0.U, Fill(p.xlen, 1.U), (io.a.asSInt / io.b.asSInt).asUInt) }
+    is(Consts.ALU_DIVU) { raw := Mux(io.b === 0.U, Fill(p.xlen, 1.U), io.a / io.b) }
+    is(Consts.ALU_REM) { raw := Mux(io.b === 0.U, io.a, (io.a.asSInt % io.b.asSInt).asUInt) }
+    is(Consts.ALU_REMU) { raw := Mux(io.b === 0.U, io.a, io.a % io.b) }
+    is(Consts.ALU_COPY_B) { raw := io.b }
+  }
+
+  io.out := Mux(io.isWord, Consts.signExtend(raw(31, 0), 32), raw)
+}
+
diff --git a/src/main/scala/execute/BranchUnit.scala b/src/main/scala/execute/BranchUnit.scala
new file mode 100644
index 0000000..97c97d0
--- /dev/null
+++ b/src/main/scala/execute/BranchUnit.scala
@@ -0,0 +1,21 @@
+import chisel3._
+import chisel3.util._
+
+class BranchUnit(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val funct3 = Input(UInt(3.W))
+    val a = Input(UInt(p.xlen.W))
+    val b = Input(UInt(p.xlen.W))
+    val taken = Output(Bool())
+  })
+
+  io.taken := MuxLookup(io.funct3, false.B)(Seq(
+    "b000".U -> (io.a === io.b),
+    "b001".U -> (io.a =/= io.b),
+    "b100".U -> (io.a.asSInt < io.b.asSInt),
+    "b101".U -> (io.a.asSInt >= io.b.asSInt),
+    "b110".U -> (io.a < io.b),
+    "b111".U -> (io.a >= io.b)
+  ))
+}
+
diff --git a/src/main/scala/execute/Divider.scala b/src/main/scala/execute/Divider.scala
new file mode 100644
index 0000000..0f111f7
--- /dev/null
+++ b/src/main/scala/execute/Divider.scala
@@ -0,0 +1,14 @@
+import chisel3._
+import chisel3.util._
+
+class Divider(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val signed = Input(Bool())
+    val a = Input(UInt(p.xlen.W))
+    val b = Input(UInt(p.xlen.W))
+    val quotient = Output(UInt(p.xlen.W))
+    val remainder = Output(UInt(p.xlen.W))
+  })
+  io.quotient := Mux(io.b === 0.U, Fill(p.xlen, 1.U), Mux(io.signed, (io.a.asSInt / io.b.asSInt).asUInt, io.a / io.b))
+  io.remainder := Mux(io.b === 0.U, io.a, Mux(io.signed, (io.a.asSInt % io.b.asSInt).asUInt, io.a % io.b))
+}
diff --git a/src/main/scala/execute/ExecStage.scala b/src/main/scala/execute/ExecStage.scala
new file mode 100644
index 0000000..f8bd5e2
--- /dev/null
+++ b/src/main/scala/execute/ExecStage.scala
@@ -0,0 +1,29 @@
+import chisel3._
+
+class ExecStage(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val inValid = Input(Bool())
+    val in = Input(new DecodedInst(p))
+    val src1 = Input(UInt(p.xlen.W))
+    val src2 = Input(UInt(p.xlen.W))
+    val outValid = Output(Bool())
+    val result = Output(UInt(p.xlen.W))
+    val branchTaken = Output(Bool())
+  })
+
+  val alu = Module(new ALU(p))
+  val branch = Module(new BranchUnit(p))
+
+  alu.io.fn := io.in.aluFn
+  alu.io.a := io.src1
+  alu.io.b := io.src2
+  alu.io.isWord := io.in.isWord
+  branch.io.funct3 := io.in.funct3
+  branch.io.a := io.src1
+  branch.io.b := io.src2
+
+  io.outValid := io.inValid
+  io.result := alu.io.out
+  io.branchTaken := branch.io.taken
+}
+
diff --git a/src/main/scala/execute/FPU.scala b/src/main/scala/execute/FPU.scala
new file mode 100644
index 0000000..46509c7
--- /dev/null
+++ b/src/main/scala/execute/FPU.scala
@@ -0,0 +1,16 @@
+import chisel3._
+
+class FPU(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val inValid = Input(Bool())
+    val op = Input(UInt(4.W))
+    val a = Input(UInt(p.xlen.W))
+    val b = Input(UInt(p.xlen.W))
+    val outValid = Output(Bool())
+    val out = Output(UInt(p.xlen.W))
+  })
+
+  io.outValid := io.inValid
+  io.out := 0.U
+}
+
diff --git a/src/main/scala/execute/Multiplier.scala b/src/main/scala/execute/Multiplier.scala
new file mode 100644
index 0000000..e00a586
--- /dev/null
+++ b/src/main/scala/execute/Multiplier.scala
@@ -0,0 +1,11 @@
+import chisel3._
+
+class Multiplier(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val a = Input(UInt(p.xlen.W))
+    val b = Input(UInt(p.xlen.W))
+    val out = Output(UInt(p.xlen.W))
+  })
+  io.out := (io.a * io.b)(p.xlen - 1, 0)
+}
+
diff --git a/src/main/scala/frontend/BTB.scala b/src/main/scala/frontend/BTB.scala
new file mode 100644
index 0000000..58a3280
--- /dev/null
+++ b/src/main/scala/frontend/BTB.scala
@@ -0,0 +1,32 @@
+import chisel3._
+import chisel3.util._
+
+class BTB(p: CoreParams = CoreParams()) extends Module {
+  private val idxBits = log2Ceil(p.btbEntries)
+  private val tagBits = 20
+
+  val io = IO(new Bundle {
+    val pc = Input(UInt(p.xlen.W))
+    val hit = Output(Bool())
+    val target = Output(UInt(p.xlen.W))
+    val update = Input(new BranchUpdate(p))
+  })
+
+  val valid = RegInit(VecInit(Seq.fill(p.btbEntries)(false.B)))
+  val tags = Reg(Vec(p.btbEntries, UInt(tagBits.W)))
+  val targets = Reg(Vec(p.btbEntries, UInt(p.xlen.W)))
+
+  val idx = io.pc(idxBits + 1, 2)
+  val tag = io.pc(idxBits + tagBits + 1, idxBits + 2)
+
+  io.hit := valid(idx) && tags(idx) === tag
+  io.target := targets(idx)
+
+  when(io.update.valid && io.update.taken) {
+    val uidx = io.update.pc(idxBits + 1, 2)
+    valid(uidx) := true.B
+    tags(uidx) := io.update.pc(idxBits + tagBits + 1, idxBits + 2)
+    targets(uidx) := io.update.target
+  }
+}
+
diff --git a/src/main/scala/frontend/BranchPredictor.scala b/src/main/scala/frontend/BranchPredictor.scala
new file mode 100644
index 0000000..9a30e19
--- /dev/null
+++ b/src/main/scala/frontend/BranchPredictor.scala
@@ -0,0 +1,38 @@
+import chisel3._
+import chisel3.util._
+
+class BranchPredictor(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val pc = Input(UInt(p.xlen.W))
+    val taken = Output(Bool())
+    val target = Output(UInt(p.xlen.W))
+    val update = Input(new BranchUpdate(p))
+  })
+
+  val ghr = RegInit(0.U(p.ghrBits.W))
+  val pht = RegInit(VecInit(Seq.fill(1 << p.ghrBits)(1.U(2.W))))
+  val btb = Module(new BTB(p))
+  val ras = Module(new RAS(p))
+
+  val idx = io.pc(p.ghrBits + 1, 2) ^ ghr
+  btb.io.pc := io.pc
+  btb.io.update := io.update
+
+  ras.io.push := io.update.valid && io.update.isCall
+  ras.io.pop := io.update.valid && io.update.isReturn
+  ras.io.pushAddr := io.update.pc + 4.U
+
+  io.taken := pht(idx)(1) && btb.io.hit
+  io.target := Mux(io.update.isReturn && !ras.io.empty, ras.io.top, btb.io.target)
+
+  when(io.update.valid) {
+    val uidx = io.update.pc(p.ghrBits + 1, 2) ^ ghr
+    when(io.update.taken && pht(uidx) =/= 3.U) {
+      pht(uidx) := pht(uidx) + 1.U
+    }.elsewhen(!io.update.taken && pht(uidx) =/= 0.U) {
+      pht(uidx) := pht(uidx) - 1.U
+    }
+    ghr := Cat(ghr(p.ghrBits - 2, 0), io.update.taken)
+  }
+}
+
diff --git a/src/main/scala/frontend/Frontend.scala b/src/main/scala/frontend/Frontend.scala
new file mode 100644
index 0000000..fce25e2
--- /dev/null
+++ b/src/main/scala/frontend/Frontend.scala
@@ -0,0 +1,43 @@
+import chisel3._
+
+class Frontend(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val redirectValid = Input(Bool())
+    val redirectPc = Input(UInt(p.xlen.W))
+    val imemReqValid = Output(Bool())
+    val imemReqAddr = Output(UInt(p.xlen.W))
+    val imemRespValid = Input(Bool())
+    val imemRespBits = Input(Vec(p.fetchWidth, UInt(32.W)))
+    val outValid = Output(Bool())
+    val out = Output(new FetchPacket(p))
+    val branchUpdate = Input(new BranchUpdate(p))
+  })
+
+  val pc = RegInit(Consts.ResetVector)
+  val predictor = Module(new BranchPredictor(p))
+  val itlb = Module(new ITLB(p))
+  val icache = Module(new ICache(p))
+
+  predictor.io.pc := pc
+  predictor.io.update := io.branchUpdate
+  itlb.io.vaddr := pc
+
+  icache.io.reqValid := true.B
+  icache.io.reqAddr := itlb.io.paddr
+  icache.io.memRespValid := io.imemRespValid
+  icache.io.memRespBits := io.imemRespBits
+
+  io.imemReqValid := icache.io.memReqValid
+  io.imemReqAddr := icache.io.memReqAddr
+  io.outValid := icache.io.respValid
+  io.out := icache.io.resp
+  io.out.predictedTaken := predictor.io.taken
+  io.out.predictedTarget := predictor.io.target
+
+  when(io.redirectValid) {
+    pc := io.redirectPc
+  }.elsewhen(icache.io.respValid) {
+    pc := Mux(predictor.io.taken, predictor.io.target, pc + (4 * p.fetchWidth).U)
+  }
+}
+
diff --git a/src/main/scala/frontend/ICache.scala b/src/main/scala/frontend/ICache.scala
new file mode 100644
index 0000000..09c5873
--- /dev/null
+++ b/src/main/scala/frontend/ICache.scala
@@ -0,0 +1,23 @@
+import chisel3._
+
+class ICache(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val reqValid = Input(Bool())
+    val reqAddr = Input(UInt(p.xlen.W))
+    val memReqValid = Output(Bool())
+    val memReqAddr = Output(UInt(p.xlen.W))
+    val memRespValid = Input(Bool())
+    val memRespBits = Input(Vec(p.fetchWidth, UInt(32.W)))
+    val respValid = Output(Bool())
+    val resp = Output(new FetchPacket(p))
+  })
+
+  io.memReqValid := io.reqValid
+  io.memReqAddr := io.reqAddr
+  io.respValid := io.memRespValid
+  io.resp.pc := io.reqAddr
+  io.resp.inst := io.memRespBits
+  io.resp.predictedTaken := false.B
+  io.resp.predictedTarget := io.reqAddr + (4 * p.fetchWidth).U
+}
+
diff --git a/src/main/scala/frontend/ITLB.scala b/src/main/scala/frontend/ITLB.scala
new file mode 100644
index 0000000..274dcd0
--- /dev/null
+++ b/src/main/scala/frontend/ITLB.scala
@@ -0,0 +1,16 @@
+import chisel3._
+import chisel3.util._
+
+class ITLB(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val vaddr = Input(UInt(p.xlen.W))
+    val paddr = Output(UInt(p.xlen.W))
+    val hit = Output(Bool())
+    val miss = Output(Bool())
+  })
+
+  io.paddr := io.vaddr
+  io.hit := true.B
+  io.miss := false.B
+}
+
diff --git a/src/main/scala/frontend/RAS.scala b/src/main/scala/frontend/RAS.scala
new file mode 100644
index 0000000..ad56b62
--- /dev/null
+++ b/src/main/scala/frontend/RAS.scala
@@ -0,0 +1,31 @@
+import chisel3._
+import chisel3.util._
+
+class RAS(p: CoreParams = CoreParams()) extends Module {
+  private val ptrBits = log2Ceil(p.rasEntries)
+
+  val io = IO(new Bundle {
+    val push = Input(Bool())
+    val pop = Input(Bool())
+    val pushAddr = Input(UInt(p.xlen.W))
+    val top = Output(UInt(p.xlen.W))
+    val empty = Output(Bool())
+  })
+
+  val stack = Reg(Vec(p.rasEntries, UInt(p.xlen.W)))
+  val sp = RegInit(0.U(ptrBits.W))
+  val count = RegInit(0.U(log2Ceil(p.rasEntries + 1).W))
+
+  io.empty := count === 0.U
+  io.top := stack(Mux(sp === 0.U, (p.rasEntries - 1).U, sp - 1.U))
+
+  when(io.push) {
+    stack(sp) := io.pushAddr
+    sp := Mux(sp === (p.rasEntries - 1).U, 0.U, sp + 1.U)
+    when(count =/= p.rasEntries.U) { count := count + 1.U }
+  }.elsewhen(io.pop && count =/= 0.U) {
+    sp := Mux(sp === 0.U, (p.rasEntries - 1).U, sp - 1.U)
+    count := count - 1.U
+  }
+}
+
diff --git a/src/main/scala/issue/IssueQueue.scala b/src/main/scala/issue/IssueQueue.scala
new file mode 100644
index 0000000..3074e44
--- /dev/null
+++ b/src/main/scala/issue/IssueQueue.scala
@@ -0,0 +1,24 @@
+import chisel3._
+
+class IssueQueue(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val enqValid = Input(Bool())
+    val enq = Input(new RenamePacket(p))
+    val enqReady = Output(Bool())
+    val issueValid = Output(Bool())
+    val issue = Output(new RenamePacket(p))
+    val issueReady = Input(Bool())
+    val flush = Input(Bool())
+  })
+
+  val intRs = Module(new ReservationStation(p, p.intRsEntries))
+  intRs.io.enqValid := io.enqValid
+  intRs.io.enq := io.enq
+  intRs.io.issueReady := io.issueReady
+  intRs.io.flush := io.flush
+
+  io.enqReady := intRs.io.enqReady
+  io.issueValid := intRs.io.issueValid
+  io.issue := intRs.io.issue
+}
+
diff --git a/src/main/scala/issue/IssueStage.scala b/src/main/scala/issue/IssueStage.scala
new file mode 100644
index 0000000..4f060d2
--- /dev/null
+++ b/src/main/scala/issue/IssueStage.scala
@@ -0,0 +1,24 @@
+import chisel3._
+
+class IssueStage(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val inValid = Input(Bool())
+    val in = Input(new RenamePacket(p))
+    val inReady = Output(Bool())
+    val outValid = Output(Bool())
+    val out = Output(new RenamePacket(p))
+    val outReady = Input(Bool())
+    val flush = Input(Bool())
+  })
+
+  val queue = Module(new IssueQueue(p))
+  queue.io.enqValid := io.inValid
+  queue.io.enq := io.in
+  queue.io.issueReady := io.outReady
+  queue.io.flush := io.flush
+
+  io.inReady := queue.io.enqReady
+  io.outValid := queue.io.issueValid
+  io.out := queue.io.issue
+}
+
diff --git a/src/main/scala/issue/ReservationStation.scala b/src/main/scala/issue/ReservationStation.scala
new file mode 100644
index 0000000..eec9457
--- /dev/null
+++ b/src/main/scala/issue/ReservationStation.scala
@@ -0,0 +1,24 @@
+import chisel3._
+import chisel3.util._
+
+class ReservationStation(p: CoreParams = CoreParams(), entries: Int = 16) extends Module {
+  val io = IO(new Bundle {
+    val enqValid = Input(Bool())
+    val enq = Input(new RenamePacket(p))
+    val enqReady = Output(Bool())
+    val issueValid = Output(Bool())
+    val issue = Output(new RenamePacket(p))
+    val issueReady = Input(Bool())
+    val flush = Input(Bool())
+  })
+
+  val q = Module(new Queue(new RenamePacket(p), entries, flow = false, pipe = true))
+  q.io.enq.valid := io.enqValid
+  q.io.enq.bits := io.enq
+  q.io.deq.ready := io.issueReady || io.flush
+
+  io.enqReady := q.io.enq.ready
+  io.issueValid := q.io.deq.valid && !io.flush
+  io.issue := q.io.deq.bits
+}
+
diff --git a/src/main/scala/memory/DCache.scala b/src/main/scala/memory/DCache.scala
new file mode 100644
index 0000000..82b89c7
--- /dev/null
+++ b/src/main/scala/memory/DCache.scala
@@ -0,0 +1,20 @@
+import chisel3._
+
+class DCache(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val reqValid = Input(Bool())
+    val req = Input(new MemRequest(p))
+    val memReqValid = Output(Bool())
+    val memReq = Output(new MemRequest(p))
+    val memRespValid = Input(Bool())
+    val memRespData = Input(UInt(p.xlen.W))
+    val respValid = Output(Bool())
+    val respData = Output(UInt(p.xlen.W))
+  })
+
+  io.memReqValid := io.reqValid
+  io.memReq := io.req
+  io.respValid := io.memRespValid
+  io.respData := io.memRespData
+}
+
diff --git a/src/main/scala/memory/DTLB.scala b/src/main/scala/memory/DTLB.scala
new file mode 100644
index 0000000..2a8838a
--- /dev/null
+++ b/src/main/scala/memory/DTLB.scala
@@ -0,0 +1,15 @@
+import chisel3._
+
+class DTLB(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val vaddr = Input(UInt(p.xlen.W))
+    val paddr = Output(UInt(p.xlen.W))
+    val hit = Output(Bool())
+    val miss = Output(Bool())
+  })
+
+  io.paddr := io.vaddr
+  io.hit := true.B
+  io.miss := false.B
+}
+
diff --git a/src/main/scala/memory/LSU.scala b/src/main/scala/memory/LSU.scala
new file mode 100644
index 0000000..0004eb7
--- /dev/null
+++ b/src/main/scala/memory/LSU.scala
@@ -0,0 +1,30 @@
+import chisel3._
+
+class LSU(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val reqValid = Input(Bool())
+    val req = Input(new MemRequest(p))
+    val dmemReqValid = Output(Bool())
+    val dmemReq = Output(new MemRequest(p))
+    val dmemRespValid = Input(Bool())
+    val dmemRespData = Input(UInt(p.xlen.W))
+    val respValid = Output(Bool())
+    val respData = Output(UInt(p.xlen.W))
+  })
+
+  val dtlb = Module(new DTLB(p))
+  val dcache = Module(new DCache(p))
+
+  dtlb.io.vaddr := io.req.addr
+  dcache.io.reqValid := io.reqValid
+  dcache.io.req := io.req
+  dcache.io.req.addr := dtlb.io.paddr
+  dcache.io.memRespValid := io.dmemRespValid
+  dcache.io.memRespData := io.dmemRespData
+
+  io.dmemReqValid := dcache.io.memReqValid
+  io.dmemReq := dcache.io.memReq
+  io.respValid := dcache.io.respValid
+  io.respData := dcache.io.respData
+}
+
diff --git a/src/main/scala/memory/LoadQueue.scala b/src/main/scala/memory/LoadQueue.scala
new file mode 100644
index 0000000..fabbcf7
--- /dev/null
+++ b/src/main/scala/memory/LoadQueue.scala
@@ -0,0 +1,23 @@
+import chisel3._
+import chisel3.util._
+
+class LoadQueue(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val enqValid = Input(Bool())
+    val enqAddr = Input(UInt(p.xlen.W))
+    val enqReady = Output(Bool())
+    val complete = Input(Bool())
+    val flush = Input(Bool())
+  })
+
+  val count = RegInit(0.U(log2Ceil(p.loadQueueEntries + 1).W))
+  io.enqReady := count =/= p.loadQueueEntries.U
+
+  when(io.flush) {
+    count := 0.U
+  }.otherwise {
+    when(io.enqValid && io.enqReady) { count := count + 1.U }
+    when(io.complete && count =/= 0.U) { count := count - 1.U }
+  }
+}
+
diff --git a/src/main/scala/memory/MMU.scala b/src/main/scala/memory/MMU.scala
new file mode 100644
index 0000000..2152f27
--- /dev/null
+++ b/src/main/scala/memory/MMU.scala
@@ -0,0 +1,15 @@
+import chisel3._
+
+class MMU(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val satp = Input(UInt(p.xlen.W))
+    val vaddr = Input(UInt(p.xlen.W))
+    val isStore = Input(Bool())
+    val paddr = Output(UInt(p.xlen.W))
+    val pageFault = Output(Bool())
+  })
+
+  io.paddr := io.vaddr
+  io.pageFault := false.B
+}
+
diff --git a/src/main/scala/memory/MemStage.scala b/src/main/scala/memory/MemStage.scala
new file mode 100644
index 0000000..a7a798d
--- /dev/null
+++ b/src/main/scala/memory/MemStage.scala
@@ -0,0 +1,26 @@
+import chisel3._
+
+class MemStage(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val reqValid = Input(Bool())
+    val req = Input(new MemRequest(p))
+    val dmemReqValid = Output(Bool())
+    val dmemReq = Output(new MemRequest(p))
+    val dmemRespValid = Input(Bool())
+    val dmemRespData = Input(UInt(p.xlen.W))
+    val respValid = Output(Bool())
+    val respData = Output(UInt(p.xlen.W))
+  })
+
+  val lsu = Module(new LSU(p))
+  lsu.io.reqValid := io.reqValid
+  lsu.io.req := io.req
+  lsu.io.dmemRespValid := io.dmemRespValid
+  lsu.io.dmemRespData := io.dmemRespData
+
+  io.dmemReqValid := lsu.io.dmemReqValid
+  io.dmemReq := lsu.io.dmemReq
+  io.respValid := lsu.io.respValid
+  io.respData := lsu.io.respData
+}
+
diff --git a/src/main/scala/memory/StoreQueue.scala b/src/main/scala/memory/StoreQueue.scala
new file mode 100644
index 0000000..ceefdf1
--- /dev/null
+++ b/src/main/scala/memory/StoreQueue.scala
@@ -0,0 +1,24 @@
+import chisel3._
+import chisel3.util._
+
+class StoreQueue(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val enqValid = Input(Bool())
+    val enq = Input(new MemRequest(p))
+    val enqReady = Output(Bool())
+    val drainValid = Output(Bool())
+    val drain = Output(new MemRequest(p))
+    val drainReady = Input(Bool())
+    val flush = Input(Bool())
+  })
+
+  val q = Module(new Queue(new MemRequest(p), p.storeQueueEntries))
+  q.io.enq.valid := io.enqValid && !io.flush
+  q.io.enq.bits := io.enq
+  q.io.deq.ready := io.drainReady || io.flush
+
+  io.enqReady := q.io.enq.ready
+  io.drainValid := q.io.deq.valid && !io.flush
+  io.drain := q.io.deq.bits
+}
+
diff --git a/src/main/scala/regfile/PhysicalRegFile.scala b/src/main/scala/regfile/PhysicalRegFile.scala
new file mode 100644
index 0000000..e363f1b
--- /dev/null
+++ b/src/main/scala/regfile/PhysicalRegFile.scala
@@ -0,0 +1,23 @@
+import chisel3._
+import chisel3.util._
+
+class PhysicalRegFile(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val raddr = Input(Vec(4, UInt(log2Ceil(p.physRegs).W)))
+    val rdata = Output(Vec(4, UInt(p.xlen.W)))
+    val wen = Input(Vec(2, Bool()))
+    val waddr = Input(Vec(2, UInt(log2Ceil(p.physRegs).W)))
+    val wdata = Input(Vec(2, UInt(p.xlen.W)))
+  })
+
+  val regs = RegInit(VecInit(Seq.fill(p.physRegs)(0.U(p.xlen.W))))
+  for (i <- 0 until 4) {
+    io.rdata(i) := Mux(io.raddr(i) === 0.U, 0.U, regs(io.raddr(i)))
+  }
+  for (i <- 0 until 2) {
+    when(io.wen(i) && io.waddr(i) =/= 0.U) {
+      regs(io.waddr(i)) := io.wdata(i)
+    }
+  }
+}
+
diff --git a/src/main/scala/rename/FreeList.scala b/src/main/scala/rename/FreeList.scala
new file mode 100644
index 0000000..5ea0400
--- /dev/null
+++ b/src/main/scala/rename/FreeList.scala
@@ -0,0 +1,26 @@
+import chisel3._
+import chisel3.util._
+
+class FreeList(p: CoreParams = CoreParams()) extends Module {
+  private val physBits = log2Ceil(p.physRegs)
+  val io = IO(new Bundle {
+    val alloc = Input(Bool())
+    val allocPhys = Output(UInt(physBits.W))
+    val canAlloc = Output(Bool())
+    val free = Input(Bool())
+    val freePhys = Input(UInt(physBits.W))
+  })
+
+  val freeBits = RegInit(VecInit((0 until p.physRegs).map(i => (i >= p.archRegs).B)))
+  val chosen = PriorityEncoder(freeBits)
+  io.canAlloc := freeBits.asUInt.orR
+  io.allocPhys := chosen
+
+  when(io.alloc && io.canAlloc) {
+    freeBits(chosen) := false.B
+  }
+  when(io.free && io.freePhys >= p.archRegs.U) {
+    freeBits(io.freePhys) := true.B
+  }
+}
+
diff --git a/src/main/scala/rename/ROB.scala b/src/main/scala/rename/ROB.scala
new file mode 100644
index 0000000..cc0d6db
--- /dev/null
+++ b/src/main/scala/rename/ROB.scala
@@ -0,0 +1,71 @@
+import chisel3._
+import chisel3.util._
+
+class RobEntry(p: CoreParams = CoreParams()) extends Bundle {
+  val valid = Bool()
+  val pc = UInt(p.xlen.W)
+  val opClass = UInt(Consts.OpClassWidth.W)
+  val dest = UInt(log2Ceil(p.physRegs).W)
+  val oldDest = UInt(log2Ceil(p.physRegs).W)
+  val completed = Bool()
+  val exception = Bool()
+  val branchMispredict = Bool()
+}
+
+class ROB(p: CoreParams = CoreParams()) extends Module {
+  private val idxBits = log2Ceil(p.robEntries)
+  val io = IO(new Bundle {
+    val allocate = Input(Bool())
+    val allocatePc = Input(UInt(p.xlen.W))
+    val allocateClass = Input(UInt(Consts.OpClassWidth.W))
+    val allocateDest = Input(UInt(log2Ceil(p.physRegs).W))
+    val allocateOldDest = Input(UInt(log2Ceil(p.physRegs).W))
+    val allocateIdx = Output(UInt(idxBits.W))
+    val canAllocate = Output(Bool())
+    val complete = Input(Bool())
+    val completeIdx = Input(UInt(idxBits.W))
+    val commitValid = Output(Bool())
+    val commit = Output(new RobEntry(p))
+    val commitReady = Input(Bool())
+    val flush = Input(Bool())
+  })
+
+  val entries = RegInit(VecInit(Seq.fill(p.robEntries)(0.U.asTypeOf(new RobEntry(p)))))
+  val head = RegInit(0.U(idxBits.W))
+  val tail = RegInit(0.U(idxBits.W))
+  val count = RegInit(0.U(log2Ceil(p.robEntries + 1).W))
+
+  io.canAllocate := count =/= p.robEntries.U
+  io.allocateIdx := tail
+  io.commit := entries(head)
+  io.commitValid := count =/= 0.U && entries(head).valid && entries(head).completed
+
+  when(io.flush) {
+    entries.foreach(_.valid := false.B)
+    head := 0.U
+    tail := 0.U
+    count := 0.U
+  }.otherwise {
+    when(io.allocate && io.canAllocate) {
+      entries(tail).valid := true.B
+      entries(tail).pc := io.allocatePc
+      entries(tail).opClass := io.allocateClass
+      entries(tail).dest := io.allocateDest
+      entries(tail).oldDest := io.allocateOldDest
+      entries(tail).completed := false.B
+      entries(tail).exception := false.B
+      entries(tail).branchMispredict := false.B
+      tail := tail + 1.U
+      count := count + 1.U
+    }
+    when(io.complete) {
+      entries(io.completeIdx).completed := true.B
+    }
+    when(io.commitValid && io.commitReady) {
+      entries(head).valid := false.B
+      head := head + 1.U
+      count := count - 1.U
+    }
+  }
+}
+
diff --git a/src/main/scala/rename/RenameStage.scala b/src/main/scala/rename/RenameStage.scala
new file mode 100644
index 0000000..69544f5
--- /dev/null
+++ b/src/main/scala/rename/RenameStage.scala
@@ -0,0 +1,49 @@
+import chisel3._
+import chisel3.util._
+
+class RenameStage(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val inValid = Input(Bool())
+    val in = Input(new DecodedInst(p))
+    val outValid = Output(Bool())
+    val out = Output(new RenamePacket(p))
+    val commitFree = Input(Bool())
+    val commitOldPhys = Input(UInt(log2Ceil(p.physRegs).W))
+    val flush = Input(Bool())
+  })
+
+  val table = Module(new RenameTable(p))
+  val freeList = Module(new FreeList(p))
+  val rob = Module(new ROB(p))
+
+  table.io.rs1 := io.in.rs1
+  table.io.rs2 := io.in.rs2
+  table.io.rd := io.in.rd
+  table.io.newPhys := freeList.io.allocPhys
+  table.io.wen := io.inValid && io.in.writesRd && freeList.io.canAlloc && rob.io.canAllocate
+  table.io.recover := io.flush
+
+  freeList.io.alloc := table.io.wen
+  freeList.io.free := io.commitFree
+  freeList.io.freePhys := io.commitOldPhys
+
+  rob.io.allocate := io.inValid && rob.io.canAllocate && (!io.in.writesRd || freeList.io.canAlloc)
+  rob.io.allocatePc := io.in.pc
+  rob.io.allocateClass := io.in.opClass
+  rob.io.allocateDest := Mux(io.in.writesRd, freeList.io.allocPhys, table.io.oldPrd)
+  rob.io.allocateOldDest := table.io.oldPrd
+  rob.io.complete := false.B
+  rob.io.completeIdx := 0.U
+  rob.io.commitReady := false.B
+  rob.io.flush := io.flush
+
+  io.outValid := rob.io.allocate
+  io.out.valid := io.outValid
+  io.out.decoded := io.in
+  io.out.prs1 := table.io.prs1
+  io.out.prs2 := table.io.prs2
+  io.out.prd := Mux(io.in.writesRd, freeList.io.allocPhys, table.io.oldPrd)
+  io.out.oldPrd := table.io.oldPrd
+  io.out.robIdx := rob.io.allocateIdx
+}
+
diff --git a/src/main/scala/rename/RenameTable.scala b/src/main/scala/rename/RenameTable.scala
new file mode 100644
index 0000000..92007b3
--- /dev/null
+++ b/src/main/scala/rename/RenameTable.scala
@@ -0,0 +1,32 @@
+import chisel3._
+import chisel3.util._
+
+class RenameTable(p: CoreParams = CoreParams()) extends Module {
+  private val physBits = log2Ceil(p.physRegs)
+  val io = IO(new Bundle {
+    val rs1 = Input(UInt(5.W))
+    val rs2 = Input(UInt(5.W))
+    val rd = Input(UInt(5.W))
+    val newPhys = Input(UInt(physBits.W))
+    val wen = Input(Bool())
+    val prs1 = Output(UInt(physBits.W))
+    val prs2 = Output(UInt(physBits.W))
+    val oldPrd = Output(UInt(physBits.W))
+    val recover = Input(Bool())
+  })
+
+  val init = VecInit((0 until p.archRegs).map(_.U(physBits.W)))
+  val speculative = RegInit(init)
+  val committed = RegInit(init)
+
+  io.prs1 := speculative(io.rs1)
+  io.prs2 := speculative(io.rs2)
+  io.oldPrd := speculative(io.rd)
+
+  when(io.recover) {
+    speculative := committed
+  }.elsewhen(io.wen && io.rd =/= 0.U) {
+    speculative(io.rd) := io.newPhys
+  }
+}
+
diff --git a/src/main/scala/writeback/WriteBackStage.scala b/src/main/scala/writeback/WriteBackStage.scala
new file mode 100644
index 0000000..79ca0b9
--- /dev/null
+++ b/src/main/scala/writeback/WriteBackStage.scala
@@ -0,0 +1,18 @@
+import chisel3._
+import chisel3.util._
+
+class WriteBackStage(p: CoreParams = CoreParams()) extends Module {
+  val io = IO(new Bundle {
+    val valid = Input(Bool())
+    val physDest = Input(UInt(log2Ceil(p.physRegs).W))
+    val data = Input(UInt(p.xlen.W))
+    val wen = Output(Bool())
+    val waddr = Output(UInt(log2Ceil(p.physRegs).W))
+    val wdata = Output(UInt(p.xlen.W))
+  })
+
+  io.wen := io.valid
+  io.waddr := io.physDest
+  io.wdata := io.data
+}
+