nudt-compiler-cpp/scripts/compare_ra.sh

#!/usr/bin/env bash
# compare_ra.sh — 对比寄存器分配（新版）与旧版编译器的汇编质量和运行性能
#
# 用法:
#   ./scripts/compare_ra.sh --old feature/mir                          # 对比 feature/mir 分支
#   ./scripts/compare_ra.sh --old 70234dd                              # 对比指定 commit
#   ./scripts/compare_ra.sh --old feature/mir --tests performance       # 仅性能测试
#   ./scripts/compare_ra.sh --old feature/mir --mode asm                # 仅对比汇编
#   ./scripts/compare_ra.sh --old path/to/old/compiler --no-build       # 使用已构建的旧编译器
#
# 输出: 终端表格 + compare_result/ 目录下的详细文件

set -euo pipefail

# ========== 参数解析 ==========
OLD_REF=""
MODE="all"        # asm | run | all
TEST_SET="all"    # functional | performance | all
NO_BUILD=false
OLD_COMPILER_PATH=""
NEW_COMPILER_PATH="./build/bin/compiler"
WORKTREE_DIR=""
KEEP_WORKTREE=false

usage() {
  echo "用法: $0 --old <branch|commit|path> [选项]"
  echo ""
  echo "必选:"
  echo "  --old <ref>        对比基线：git 分支名、commit hash、或旧编译器路径"
  echo ""
  echo "可选:"
  echo "  --mode <mode>      对比模式: asm (仅汇编) | run (仅运行) | all (默认)"
  echo "  --tests <set>      测试集: functional | performance | all (默认)"
  echo "  --no-build         旧编译器已构建好，--old 指向编译器可执行文件路径"
  echo "  --keep-worktree    保留旧的 git worktree（默认会删除）"
  echo ""
  echo "示例:"
  echo "  $0 --old feature/mir                              # 对比 feature/mir 分支"
  echo "  $0 --old 70234dd                                   # 对比特定 commit"
  echo "  $0 --old feature/mir --tests performance --mode run # 仅对比性能测试的运行时间"
  echo "  $0 --old /tmp/old-compiler --no-build               # 使用预构建的旧编译器"
  exit 1
}

while [[ $# -gt 0 ]]; do
  case "$1" in
    --old) OLD_REF="$2"; shift 2 ;;
    --mode) MODE="$2"; shift 2 ;;
    --tests) TEST_SET="$2"; shift 2 ;;
    --no-build) NO_BUILD=true; shift ;;
    --keep-worktree) KEEP_WORKTREE=true; shift ;;
    *) echo "未知参数: $1"; usage ;;
  esac
done

if [[ -z "$OLD_REF" ]]; then
  echo "错误: 必须指定 --old"
  usage
fi

# ========== 路径设置 ==========
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
PROJECT_DIR="$(dirname "$SCRIPT_DIR")"
RESULT_DIR="$PROJECT_DIR/compare_result"
OLD_BUILD_DIR=""
OLD_COMPILER=""

rm -rf "$RESULT_DIR"
mkdir -p "$RESULT_DIR"

# ========== 构建旧版编译器 ==========
setup_old_compiler() {
  if [[ "$NO_BUILD" == true ]]; then
    # 用户直接提供编译器路径
    if [[ ! -x "$OLD_REF" ]]; then
      echo "错误: 旧编译器路径不存在或不可执行: $OLD_REF"
      exit 1
    fi
    OLD_COMPILER="$(realpath "$OLD_REF")"
    echo "使用预构建旧编译器: $OLD_COMPILER"
    return
  fi

  # 检查是否是 git ref
  if ! git rev-parse --verify "$OLD_REF" >/dev/null 2>&1; then
    echo "错误: '$OLD_REF' 不是有效的 git 引用（分支/commit），或使用 --no-build 指定编译器路径"
    exit 1
  fi

  WORKTREE_DIR="$PROJECT_DIR/.worktree-old-$(echo "$OLD_REF" | tr '/' '-')"
  OLD_BUILD_DIR="$WORKTREE_DIR/build"

  echo "=== 准备旧版编译器 ==="
  echo "Git 引用: $OLD_REF"
  echo "Worktree: $WORKTREE_DIR"

  # 清理已有的 worktree
  if [[ -d "$WORKTREE_DIR" ]]; then
    echo "移除已有 worktree..."
    git worktree remove --force "$WORKTREE_DIR" 2>/dev/null || rm -rf "$WORKTREE_DIR"
  fi

  git worktree add "$WORKTREE_DIR" "$OLD_REF"
  echo "Worktree 已创建: $WORKTREE_DIR"

  # 生成 ANTLR 语法分析器
  echo "生成 ANTLR 语法分析器..."
  java -jar "$WORKTREE_DIR/third_party/antlr-4.13.2-complete.jar" \
    -Dlanguage=Cpp \
    -visitor -no-listener \
    -Xexact-output-dir \
    -o "$OLD_BUILD_DIR/generated/antlr4" \
    "$WORKTREE_DIR/src/antlr4/SysY.g4" > "$RESULT_DIR/build_old.log" 2>&1 || {
    echo "错误: ANTLR 生成失败，日志:"
    tail -20 "$RESULT_DIR/build_old.log"
    exit 1
  }

  # 构建
  echo "构建旧版编译器..."
  cmake -S "$WORKTREE_DIR" -B "$OLD_BUILD_DIR" -DCMAKE_BUILD_TYPE=Release >> "$RESULT_DIR/build_old.log" 2>&1 || {
    echo "错误: 旧版 cmake 配置失败，日志:"
    tail -20 "$RESULT_DIR/build_old.log"
    exit 1
  }
  cmake --build "$OLD_BUILD_DIR" -j"$(nproc 2>/dev/null || echo 4)" >> "$RESULT_DIR/build_old.log" 2>&1 || {
    echo "错误: 旧版构建失败，日志:"
    tail -30 "$RESULT_DIR/build_old.log"
    exit 1
  }

  OLD_COMPILER="$OLD_BUILD_DIR/bin/compiler"
  if [[ ! -x "$OLD_COMPILER" ]]; then
    echo "错误: 旧编译器未生成: $OLD_COMPILER"
    exit 1
  fi
  echo "旧编译器已构建: $OLD_COMPILER"
}

# ========== 确保新版编译器存在 ==========
setup_new_compiler() {
  if [[ ! -x "$NEW_COMPILER_PATH" ]]; then
    echo "错误: 新编译器不存在，请先构建: cmake -B build && cmake --build build -j"
    exit 1
  fi
  NEW_COMPILER="$(realpath "$NEW_COMPILER_PATH")"
  echo "新版编译器: $NEW_COMPILER"
}

# ========== 工具检查 ==========
check_tools() {
  if [[ "$NO_BUILD" == false ]]; then
    if ! command -v java >/dev/null 2>&1; then
      echo "错误: 未找到 java，构建时需要 ANTLR 生成语法分析器"
      exit 1
    fi
  fi
  if ! command -v aarch64-linux-gnu-gcc >/dev/null 2>&1; then
    echo "警告: 未找到 aarch64-linux-gnu-gcc，汇编模式可用但运行模式不可用"
  fi
  if [[ "$MODE" == "run" || "$MODE" == "all" ]]; then
    if ! command -v qemu-aarch64 >/dev/null 2>&1; then
      echo "错误: 未找到 qemu-aarch64，无法运行测试"
      echo "  apt install qemu-user  (Ubuntu/Debian)"
      exit 1
    fi
    if ! command -v aarch64-linux-gnu-gcc >/dev/null 2>&1; then
      echo "错误: 未找到 aarch64-linux-gnu-gcc，无法链接可执行文件"
      exit 1
    fi
  fi
}

# ========== 获取测试列表 ==========
get_tests() {
  local test_dir="$PROJECT_DIR/test/test_case"
  local tests=()

  if [[ "$TEST_SET" == "functional" || "$TEST_SET" == "all" ]]; then
    for f in "$test_dir/functional"/*.sy; do
      [[ -f "$f" ]] && tests+=("$f")
    done
  fi
  if [[ "$TEST_SET" == "performance" || "$TEST_SET" == "all" ]]; then
    for f in "$test_dir/performance"/*.sy; do
      [[ -f "$f" ]] && tests+=("$f")
    done
  fi

  if [[ ${#tests[@]} -eq 0 ]]; then
    echo "错误: 没有找到测试用例"
    exit 1
  fi
  printf '%s\n' "${tests[@]}"
}

# ========== 汇编对比 ==========
compare_asm() {
  local test_file="$1"
  local stem; stem=$(basename "$test_file" .sy)
  local test_dir; test_dir=$(dirname "$test_file")

  local old_asm="$RESULT_DIR/asm/$stem.old.s"
  local new_asm="$RESULT_DIR/asm/$stem.new.s"

  mkdir -p "$RESULT_DIR/asm"

  "$OLD_COMPILER" --emit-asm "$test_file" > "$old_asm" 2>/dev/null || {
    echo "OLD_BUILD_FAIL" > "$RESULT_DIR/asm/$stem.result"
    return
  }
  "$NEW_COMPILER" --emit-asm "$test_file" > "$new_asm" 2>/dev/null || {
    echo "NEW_BUILD_FAIL" > "$RESULT_DIR/asm/$stem.result"
    return
  }

  # 统计指令数（排除标签行、伪指令行、空行）
  local old_inst new_inst old_mem new_mem old_branches new_branches
  old_inst=$(grep -cE '^\s+\w+\s' "$old_asm" 2>/dev/null || echo 0)
  new_inst=$(grep -cE '^\s+\w+\s' "$new_asm" 2>/dev/null || echo 0)
  old_mem=$(grep -cE '\b(ldr|str|ldur|stur|ldp|stp)\b' "$old_asm" 2>/dev/null || echo 0)
  new_mem=$(grep -cE '\b(ldr|str|ldur|stur|ldp|stp)\b' "$new_asm" 2>/dev/null || echo 0)
  old_branches=$(grep -cE '\bb(|\.\w+)\s' "$old_asm" 2>/dev/null || echo 0)
  new_branches=$(grep -cE '\bb(|\.\w+)\s' "$new_asm" 2>/dev/null || echo 0)

  # 计算变化百分比
  local inst_pct mem_pct
  if [[ "$old_inst" -gt 0 ]]; then
    inst_pct=$(echo "scale=1; ($new_inst - $old_inst) * 100 / $old_inst" | bc 2>/dev/null || echo "N/A")
  else
    inst_pct="N/A"
  fi
  if [[ "$old_mem" -gt 0 ]]; then
    mem_pct=$(echo "scale=1; ($new_mem - $old_mem) * 100 / $old_mem" | bc 2>/dev/null || echo "N/A")
  else
    mem_pct="N/A"
  fi

  # 保存结果
  echo "$stem $old_inst $new_inst $inst_pct $old_mem $new_mem $mem_pct $old_branches $new_branches" \
    > "$RESULT_DIR/asm/$stem.result"

  # 生成 diff
  diff -u "$old_asm" "$new_asm" > "$RESULT_DIR/asm/$stem.diff" 2>/dev/null || true
}

# ========== 运行对比 ==========
compare_run() {
  local test_file="$1"
  local stem; stem=$(basename "$test_file" .sy)
  local test_dir; test_dir=$(dirname "$test_file")

  local old_exe="$RESULT_DIR/run/$stem.old"
  local new_exe="$RESULT_DIR/run/$stem.new"
  local old_out="$RESULT_DIR/run/$stem.old.out"
  local new_out="$RESULT_DIR/run/$stem.new.out"
  local stdin_file="$test_dir/$stem.in"
  local expected_file="$test_dir/$stem.out"

  mkdir -p "$RESULT_DIR/run"

  # 生成旧版可执行文件
  local old_asm="$RESULT_DIR/run/$stem.old.s"
  "$OLD_COMPILER" --emit-asm "$test_file" > "$old_asm" 2>/dev/null || {
    echo "OLD_COMPILE_FAIL" > "$RESULT_DIR/run/$stem.result"
    return
  }
  aarch64-linux-gnu-gcc -no-pie "$old_asm" -L"$PROJECT_DIR/sylib" -lsysy -static -o "$old_exe" 2>/dev/null || {
    echo "OLD_LINK_FAIL" > "$RESULT_DIR/run/$stem.result"
    return
  }

  # 生成新版可执行文件
  local new_asm="$RESULT_DIR/run/$stem.new.s"
  "$NEW_COMPILER" --emit-asm "$test_file" > "$new_asm" 2>/dev/null || {
    echo "NEW_COMPILE_FAIL" > "$RESULT_DIR/run/$stem.result"
    return
  }
  aarch64-linux-gnu-gcc -no-pie "$new_asm" -L"$PROJECT_DIR/sylib" -lsysy -static -o "$new_exe" 2>/dev/null || {
    echo "NEW_LINK_FAIL" > "$RESULT_DIR/run/$stem.result"
    return
  }

  # 运行旧版
  local old_time="N/A" old_rc="N/A"
  set +eo pipefail
  if [[ -f "$stdin_file" ]]; then
    old_time=$( { time qemu-aarch64 -L /usr/aarch64-linux-gnu "$old_exe" < "$stdin_file" > "$old_out" 2>/dev/null; echo $? > "$old_out.rc"; } 2>&1 | grep real | awk '{print $2}' || echo "N/A")
  else
    old_time=$( { time qemu-aarch64 -L /usr/aarch64-linux-gnu "$old_exe" > "$old_out" 2>/dev/null; echo $? > "$old_out.rc"; } 2>&1 | grep real | awk '{print $2}' || echo "N/A")
  fi
  old_rc=$(cat "$old_out.rc" 2>/dev/null || echo "1")

  # 运行新版
  local new_time="N/A" new_rc="N/A"
  if [[ -f "$stdin_file" ]]; then
    new_time=$( { time qemu-aarch64 -L /usr/aarch64-linux-gnu "$new_exe" < "$stdin_file" > "$new_out" 2>/dev/null; echo $? > "$new_out.rc"; } 2>&1 | grep real | awk '{print $2}' || echo "N/A")
  else
    new_time=$( { time qemu-aarch64 -L /usr/aarch64-linux-gnu "$new_exe" > "$new_out" 2>/dev/null; echo $? > "$new_out.rc"; } 2>&1 | grep real | awk '{print $2}' || echo "N/A")
  fi
  new_rc=$(cat "$new_out.rc" 2>/dev/null || echo "1")
  set -eo pipefail

  # 构造实际输出（程序输出 + 退出码），与 verify_asm.sh 格式一致
  local old_actual="$RESULT_DIR/run/$stem.old.actual"
  local new_actual="$RESULT_DIR/run/$stem.new.actual"
  {
    cat "$old_out"
    if [[ -s "$old_out" ]] && (( $(tail -c 1 "$old_out" | wc -l) == 0 )); then
      printf '\n'
    fi
    printf '%s\n' "$old_rc"
  } > "$old_actual"
  {
    cat "$new_out"
    if [[ -s "$new_out" ]] && (( $(tail -c 1 "$new_out" | wc -l) == 0 )); then
      printf '\n'
    fi
    printf '%s\n' "$new_rc"
  } > "$new_actual"

  # 检查输出匹配（与 expected 文件比较，expected 格式为 stdout + exit_code）
  local old_match="N" new_match="N"
  if [[ -f "$expected_file" ]]; then
    diff -w -q "$old_actual" "$expected_file" >/dev/null 2>&1 && old_match="Y"
    diff -w -q "$new_actual" "$expected_file" >/dev/null 2>&1 && new_match="Y"
  fi

  # 速度比
  local speedup="N/A"
  if [[ "$old_time" != "N/A" && "$new_time" != "N/A" ]]; then
    local old_sec new_sec
    old_sec=$(echo "$old_time" | sed 's/m/ /' | awk '{print $1 * 60 + $2}' 2>/dev/null || echo 0)
    new_sec=$(echo "$new_time" | sed 's/m/ /' | awk '{print $1 * 60 + $2}' 2>/dev/null || echo 0)
    if [[ "$(echo "$new_sec > 0" | bc -l 2>/dev/null)" == "1" ]]; then
      speedup=$(echo "scale=2; $old_sec / $new_sec" | bc 2>/dev/null || echo "N/A")
    fi
  fi

  echo "$stem $old_time $new_time $speedup $old_match $new_match $old_rc $new_rc" \
    > "$RESULT_DIR/run/$stem.result"
}

# ========== 主流程 ==========
main() {
  echo "============================================="
  echo "  寄存器分配编译器对比"
  echo "  旧版: $OLD_REF"
  echo "  测试集: $TEST_SET"
  echo "  模式: $MODE"
  echo "============================================="
  echo ""

  check_tools
  setup_old_compiler
  setup_new_compiler

  local tests
  mapfile -t tests < <(get_tests)
  local total=${#tests[@]}
  echo "共 $total 个测试用例"
  echo ""

  # ========== 汇编对比 ==========
  if [[ "$MODE" == "asm" || "$MODE" == "all" ]]; then
    echo "=== 汇编质量对比 ==="
    local count=0
    for test_file in "${tests[@]}"; do
      compare_asm "$test_file"
      count=$((count + 1))
      printf "\r  进度: %d/%d" "$count" "$total"
    done
    echo ""

    # 输出汇编对比表
    echo ""
    printf "%-30s %8s %8s %8s %8s %8s %8s\n" \
      "测试用例" "旧指令数" "新指令数" "变化%" "旧访存" "新访存" "变化%"
    printf "%-30s %8s %8s %8s %8s %8s %8s\n" \
      "------------------------------" "--------" "--------" "--------" "--------" "--------" "--------"

    local total_old_inst=0 total_new_inst=0 total_old_mem=0 total_new_mem=0 valid_count=0
    for f in "$RESULT_DIR/asm"/*.result; do
      [[ -f "$f" ]] || continue
      local result
      result=$(cat "$f")
      if [[ "$result" == *"FAIL"* ]]; then
        printf "%-30s %8s\n" "$(basename "$f" .result)" "编译失败"
        continue
      fi
      read -r stem old_inst new_inst inst_pct old_mem new_mem mem_pct _ _ <<< "$result"
      printf "%-30s %8d %8d %7s%% %8d %8d %7s%%\n" \
        "$stem" "$old_inst" "$new_inst" "$inst_pct" "$old_mem" "$new_mem" "$mem_pct"
      total_old_inst=$((total_old_inst + old_inst))
      total_new_inst=$((total_new_inst + new_inst))
      total_old_mem=$((total_old_mem + old_mem))
      total_new_mem=$((total_new_mem + new_mem))
      valid_count=$((valid_count + 1))
    done

    if [[ "$valid_count" -gt 0 ]]; then
      local avg_inst_pct avg_mem_pct
      avg_inst_pct=$(echo "scale=1; ($total_new_inst - $total_old_inst) * 100 / $total_old_inst" | bc 2>/dev/null || echo "N/A")
      avg_mem_pct=$(echo "scale=1; ($total_new_mem - $total_old_mem) * 100 / $total_old_mem" | bc 2>/dev/null || echo "N/A")
      printf "%-30s %8d %8d %7s%% %8d %8d %7s%%\n" \
        "--- 合计 ---" "$total_old_inst" "$total_new_inst" "$avg_inst_pct" \
        "$total_old_mem" "$total_new_mem" "$avg_mem_pct"
    fi
    echo ""
    echo "详细 diff 文件: $RESULT_DIR/asm/*.diff"
  fi

  # ========== 运行对比 ==========
  if [[ "$MODE" == "run" || "$MODE" == "all" ]]; then
    echo "=== 运行结果对比 ==="
    local count=0
    for test_file in "${tests[@]}"; do
      compare_run "$test_file"
      count=$((count + 1))
      printf "\r  进度: %d/%d" "$count" "$total"
    done
    echo ""

    echo ""
    printf "%-30s %10s %10s %8s %6s %6s %8s %8s\n" \
      "测试用例" "旧耗时" "新耗时" "加速比" "旧匹配" "新匹配" "旧退出码" "新退出码"
    printf "%-30s %10s %10s %8s %6s %6s %8s %8s\n" \
      "------------------------------" "----------" "----------" "--------" "------" "------" "--------" "--------"

    local pass_old=0 pass_new=0 total_valid=0
    for f in "$RESULT_DIR/run"/*.result; do
      [[ -f "$f" ]] || continue
      local result
      result=$(cat "$f")
      if [[ "$result" == *"FAIL"* ]]; then
        printf "%-30s %10s\n" "$(basename "$f" .result)" "$result"
        continue
      fi
      read -r stem old_time new_time speedup old_match new_match old_status new_status <<< "$result"
      printf "%-30s %10s %10s %8s %6s %6s %8s %8s\n" \
        "$stem" "$old_time" "$new_time" "$speedup" "$old_match" "$new_match" "$old_status" "$new_status"
      [[ "$old_match" == "Y" ]] && pass_old=$((pass_old + 1))
      [[ "$new_match" == "Y" ]] && pass_new=$((pass_new + 1))
      total_valid=$((total_valid + 1))
    done

    if [[ "$total_valid" -gt 0 ]]; then
      echo ""
      printf "输出匹配率: 旧版 %d/%d, 新版 %d/%d\n" "$pass_old" "$total_valid" "$pass_new" "$total_valid"
    fi
  fi

  # ========== 清理 ==========
  if [[ -n "$WORKTREE_DIR" && "$KEEP_WORKTREE" == false ]]; then
    echo ""
    echo "清理 worktree..."
    git worktree remove --force "$WORKTREE_DIR" 2>/dev/null || true
  fi

  echo ""
  echo "对比结果保存在: $RESULT_DIR"
}

main