7th Optimization reached 338.219s

doc/Lab6-并行与循环优化.md

@@ -59,8 +59,8 @@ cmake --build build -j "$(nproc)"
 ### 7.1 功能回归
 
 ```bash
-./scripts/verify_ir.sh test/test_case/functional/simple_add.sy test/test_result/function/ir --run
-./scripts/verify_asm.sh test/test_case/functional/simple_add.sy test/test_result/function/asm --run
+./scripts/verify_ir.sh test/test_case/performance/2025-MYO-20.sy test/test_result/performance/ir --run
+./scripts/verify_asm.sh test/test_case/performance/2025-MYO-20.sy test/test_result/performance/asm --run
 ```
 
 `--run` 模式下脚本会自动读取同名 `.in`，并将程序输出与退出码和同名 `.out` 比对。

include/mir/MIR.h

@@ -44,6 +44,8 @@ enum class Opcode {
   LoadStackAddr,     // 加载栈地址：add x9, x29, #offset（用于数组基址）
   LoadIndirect,      // 间接加载：ldr w8, [x9]
   StoreIndirect,     // 间接存储：str w8, [x9]
+  LoadIndirectScaled,   // 间接加载：ldr w8, [x9, w10, uxtw #2]
+  StoreIndirectScaled,  // 间接存储：str w8, [x9, w10, uxtw #2]
   LoadGlobal,
   StoreGlobal,
   LoadGlobalAddr,  // 加载全局变量地址（用于数组）

scripts/run_all_tests.sh

@@ -38,6 +38,20 @@ if [[ ! -x "$compiler" ]]; then
   exit 1
 fi
 
+find_tool() {
+  local name
+  for name in "$@"; do
+    if command -v "$name" >/dev/null 2>&1; then
+      command -v "$name"
+      return 0
+    fi
+  done
+  return 1
+}
+
+LLC_CMD=$(find_tool llc llc-20 || true)
+CLANG_CMD=$(find_tool clang clang-20 || true)
+
 total=0
 passed=0
 failed=0
@@ -76,10 +90,10 @@ run_ir_test() {
   local emit_ns=$(( $(now_ns) - emit_start_ns ))
 
   # 需要 llc + clang
-  if ! command -v llc >/dev/null 2>&1 || ! command -v clang >/dev/null 2>&1; then
+  if [[ -z "$LLC_CMD" || -z "$CLANG_CMD" ]]; then
     RUN_LAST_TOTAL_NS=$(( $(now_ns) - start_ns ))
     RUN_LAST_BREAKDOWN="emit=$(format_ns "$emit_ns")"
-    echo "  [SKIP-IR] $sy (缺少 llc/clang)"
+    echo "  [SKIP-IR] $sy (缺少 llc/llc-20 或 clang/clang-20)"
     return 2
   fi
 
@@ -88,16 +102,16 @@ run_ir_test() {
 
   local lower_link_start_ns
   lower_link_start_ns=$(now_ns)
-  if ! llc -filetype=obj "$out_file" -o "$obj" 2>/dev/null; then
+  if ! "$LLC_CMD" -filetype=obj "$out_file" -o "$obj" 2>/dev/null; then
     RUN_LAST_TOTAL_NS=$(( $(now_ns) - start_ns ))
     RUN_LAST_BREAKDOWN="emit=$(format_ns "$emit_ns")"
-    echo "  [SKIP-IR] $sy (llc 编译失败)"
+    echo "  [SKIP-IR] $sy ($LLC_CMD 编译失败)"
     return 2
   fi
-  if ! clang -no-pie "$obj" sylib/sylib.c -o "$exe" -lm -pthread 2>/dev/null; then
+  if ! "$CLANG_CMD" -no-pie "$obj" sylib/sylib.c -o "$exe" -lm -pthread 2>/dev/null; then
     RUN_LAST_TOTAL_NS=$(( $(now_ns) - start_ns ))
     RUN_LAST_BREAKDOWN="emit=$(format_ns "$emit_ns") lower+link=$(format_ns $(( $(now_ns) - lower_link_start_ns )))"
-    echo "  [SKIP-IR] $sy (clang 链接失败)"
+    echo "  [SKIP-IR] $sy ($CLANG_CMD 链接失败)"
     return 2
   fi
   local lower_link_ns=$(( $(now_ns) - lower_link_start_ns ))
@@ -116,7 +130,7 @@ run_ir_test() {
     timeout $run_timeout "$exe" > "$stdout_file" 2>/dev/null
   fi
   local status=$?
-  set -e
+  set +e
   local run_ns=$(( $(now_ns) - run_start_ns ))
   RUN_LAST_TOTAL_NS=$(( $(now_ns) - start_ns ))
   RUN_LAST_BREAKDOWN="emit=$(format_ns "$emit_ns") lower+link=$(format_ns "$lower_link_ns") run=$(format_ns "$run_ns")"
@@ -216,7 +230,7 @@ run_asm_test() {
     timeout $run_timeout qemu-aarch64 -L /usr/aarch64-linux-gnu "$exe" > "$stdout_file" 2>/dev/null
   fi
   local status=$?
-  set -e
+  set +e
   local run_ns=$(( $(now_ns) - run_start_ns ))
   RUN_LAST_TOTAL_NS=$(( $(now_ns) - start_ns ))
   RUN_LAST_BREAKDOWN="emit=$(format_ns "$emit_ns") asm+link=$(format_ns "$link_ns") run=$(format_ns "$run_ns")"
@@ -251,6 +265,7 @@ run_asm_test() {
 echo "========================================"
 echo "  Lab4 批量回归测试 (mode: $mode)"
 echo "========================================"
+echo "  LLVM tools: $LLC_CMD / $CLANG_CMD"
 echo ""
 
 # 收集所有测试文件

scripts/verify_ir.sh

@@ -49,6 +49,20 @@ if [[ ! -x "$compiler" ]]; then
   exit 1
 fi
 
+find_tool() {
+  local name
+  for name in "$@"; do
+    if command -v "$name" >/dev/null 2>&1; then
+      command -v "$name"
+      return 0
+    fi
+  done
+  return 1
+}
+
+llc_cmd=$(find_tool llc llc-20 || true)
+clang_cmd=$(find_tool clang clang-20 || true)
+
 mkdir -p "$out_dir"
 base=$(basename "$input")
 stem=${base%.sy}
@@ -63,12 +77,12 @@ echo "IR 已生成: $out_file"
 echo "IR 生成耗时: $(format_ns "$emit_elapsed_ns")"
 
 if [[ "$run_exec" == true ]]; then
-  if ! command -v llc >/dev/null 2>&1; then
-    echo "未找到 llc，无法运行 IR。请安装 LLVM。" >&2
+  if [[ -z "$llc_cmd" ]]; then
+    echo "未找到 llc 或 llc-20，无法运行 IR。请安装 LLVM。" >&2
     exit 1
   fi
-  if ! command -v clang >/dev/null 2>&1; then
-    echo "未找到 clang，无法链接可执行文件。请安装 LLVM/Clang。" >&2
+  if [[ -z "$clang_cmd" ]]; then
+    echo "未找到 clang 或 clang-20，无法链接可执行文件。请安装 Clang。" >&2
     exit 1
   fi
   obj="$out_dir/$stem.o"
@@ -76,8 +90,8 @@ if [[ "$run_exec" == true ]]; then
   stdout_file="$out_dir/$stem.stdout"
   actual_file="$out_dir/$stem.actual.out"
   lower_link_start_ns=$(now_ns)
-  llc -filetype=obj "$out_file" -o "$obj"
-  clang -no-pie "$obj" sylib/sylib.c -o "$exe" -lm -pthread
+  "$llc_cmd" -filetype=obj "$out_file" -o "$obj"
+  "$clang_cmd" -no-pie "$obj" sylib/sylib.c -o "$exe" -lm -pthread
   lower_link_elapsed_ns=$(( $(now_ns) - lower_link_start_ns ))
   echo "IR 落地/链接耗时: $(format_ns "$lower_link_elapsed_ns")"
   echo "运行 $exe ..."

src/ir/BasicBlock.cpp

@@ -100,6 +100,20 @@ PhiInst* BasicBlock::PrependPhi(std::shared_ptr<Type> ty,
 
 void BasicBlock::RemoveInstruction(Instruction* inst) {
   if (!inst) return;
+  if (auto* br = dynamic_cast<BranchInst*>(inst)) {
+    auto* target = br->GetTarget();
+    RemoveSuccessor(target);
+    target->RemovePredecessor(this);
+  } else if (auto* cbr = dynamic_cast<CondBranchInst*>(inst)) {
+    auto* true_bb = cbr->GetTrueBlock();
+    auto* false_bb = cbr->GetFalseBlock();
+    RemoveSuccessor(true_bb);
+    true_bb->RemovePredecessor(this);
+    if (false_bb != true_bb) {
+      RemoveSuccessor(false_bb);
+      false_bb->RemovePredecessor(this);
+    }
+  }
   // 清除该指令所有操作数的 use 关系
   for (size_t i = 0; i < inst->GetNumOperands(); ++i) {
     auto* operand = inst->GetOperand(i);

src/ir/IRPrinter.cpp

@@ -7,6 +7,7 @@
 #include <cstdint>
 #include <cstring>
 #include <ostream>
+#include <sstream>
 #include <stdexcept>
 #include <string>
 
@@ -68,6 +69,24 @@ static const char* OpcodeToString(Opcode op) {
   return "?";
 }
 
+static const char* BinaryOpcodeToString(Opcode op, const Type& ty) {
+  const bool is_float = ty.IsFloat32();
+  switch (op) {
+    case Opcode::Add:
+      return is_float ? "fadd" : "add";
+    case Opcode::Sub:
+      return is_float ? "fsub" : "sub";
+    case Opcode::Mul:
+      return is_float ? "fmul" : "mul";
+    case Opcode::Div:
+      return is_float ? "fdiv" : "sdiv";
+    case Opcode::Mod:
+      return "srem";
+    default:
+      return OpcodeToString(op);
+  }
+}
+
 static const char* CmpOpToString(CmpOp op) {
   switch (op) {
     case CmpOp::Eq:
@@ -86,10 +105,44 @@ static const char* CmpOpToString(CmpOp op) {
   return "?";
 }
 
+static const char* CmpOpcodeToString(const Type& ty) {
+  return ty.IsFloat32() ? "fcmp" : "icmp";
+}
+
+static const char* FloatCmpOpToString(CmpOp op) {
+  switch (op) {
+    case CmpOp::Eq:
+      return "oeq";
+    case CmpOp::Ne:
+      return "one";
+    case CmpOp::Lt:
+      return "olt";
+    case CmpOp::Le:
+      return "ole";
+    case CmpOp::Gt:
+      return "ogt";
+    case CmpOp::Ge:
+      return "oge";
+  }
+  return "?";
+}
+
+static std::string FloatToString(float value) {
+  double widened = static_cast<double>(value);
+  std::uint64_t bits = 0;
+  std::memcpy(&bits, &widened, sizeof(bits));
+  std::ostringstream oss;
+  oss << "0x" << std::uppercase << std::hex << bits;
+  return oss.str();
+}
+
 static std::string ValueToString(const Value* v) {
   if (auto* ci = dynamic_cast<const ConstantInt*>(v)) {
     return std::to_string(ci->GetValue());
   }
+  if (auto* cf = dynamic_cast<const ConstantFloat*>(v)) {
+    return FloatToString(cf->GetValue());
+  }
   if (auto* gv = dynamic_cast<const GlobalVariable*>(v)) {
     return "@" + gv->GetName();
   }
@@ -102,6 +155,20 @@ static std::string ValueToString(const Value* v) {
   return v ? v->GetName() : "<null>";
 }
 
+static std::string CmpBoolName(const CmpInst* cmp) {
+  return cmp->GetName() + ".cmp";
+}
+
+static std::string BranchCondToString(const Value* v) {
+  if (auto* cmp = dynamic_cast<const CmpInst*>(v)) {
+    return CmpBoolName(cmp);
+  }
+  if (auto* ci = dynamic_cast<const ConstantInt*>(v)) {
+    return ci->GetValue() == 0 ? "false" : "true";
+  }
+  return ValueToString(v);
+}
+
 void IRPrinter::Print(const Module& module, std::ostream& os) {
   // 先打印全局变量
   for (const auto& gv : module.GetGlobalVars()) {
@@ -115,7 +182,8 @@ void IRPrinter::Print(const Module& module, std::ostream& os) {
         std::int32_t bits = static_cast<std::int32_t>(gv->GetInitValue());
         float fval = 0.0f;
         std::memcpy(&fval, &bits, sizeof(fval));
-        os << "@" << gv->GetName() << " = global float " << fval << "\n";
+        os << "@" << gv->GetName() << " = global float "
+           << FloatToString(fval) << "\n";
       } else {
         os << "@" << gv->GetName() << " = global i32 " << gv->GetInitValue()
            << "\n";
@@ -163,7 +231,7 @@ void IRPrinter::Print(const Module& module, std::ostream& os) {
           case Opcode::Mod: {
             auto* bin = static_cast<const BinaryInst*>(inst);
             os << "  " << bin->GetName() << " = "
-               << OpcodeToString(bin->GetOpcode()) << " "
+               << BinaryOpcodeToString(bin->GetOpcode(), *bin->GetLhs()->GetType()) << " "
                << TypeToString(*bin->GetLhs()->GetType()) << " "
                << ValueToString(bin->GetLhs()) << ", "
                << ValueToString(bin->GetRhs()) << "\n";
@@ -171,11 +239,16 @@ void IRPrinter::Print(const Module& module, std::ostream& os) {
           }
           case Opcode::Cmp: {
             auto* cmp = static_cast<const CmpInst*>(inst);
-            os << "  " << cmp->GetName() << " = " << OpcodeToString(cmp->GetOpcode())
-               << " " << CmpOpToString(cmp->GetCmpOp()) << " "
+            const bool is_float_cmp = cmp->GetLhs()->GetType()->IsFloat32();
+            os << "  " << CmpBoolName(cmp) << " = "
+               << CmpOpcodeToString(*cmp->GetLhs()->GetType())
+               << " " << (is_float_cmp ? FloatCmpOpToString(cmp->GetCmpOp())
+                                        : CmpOpToString(cmp->GetCmpOp())) << " "
                << TypeToString(*cmp->GetLhs()->GetType()) << " "
                << ValueToString(cmp->GetLhs()) << ", "
                << ValueToString(cmp->GetRhs()) << "\n";
+            os << "  " << cmp->GetName() << " = zext i1 "
+               << CmpBoolName(cmp) << " to i32\n";
             break;
           }
           case Opcode::Cast: {
@@ -222,7 +295,7 @@ void IRPrinter::Print(const Module& module, std::ostream& os) {
           }
           case Opcode::CondBr: {
             auto* cbr = static_cast<const CondBranchInst*>(inst);
-            os << "  br i1 " << ValueToString(cbr->GetCond()) << ", label %"
+            os << "  br i1 " << BranchCondToString(cbr->GetCond()) << ", label %"
                << cbr->GetTrueBlock()->GetName() << ", label %"
                << cbr->GetFalseBlock()->GetName() << "\n";
             break;

src/ir/passes/CFGSimplify.cpp

@@ -99,6 +99,11 @@ bool RunCFGSimplify(Function& func, Context& ctx) {
     // 从后继的前驱列表中移除
     for (auto* succ : bb->GetSuccessors()) {
       succ->RemovePredecessor(bb);
+      for (auto& inst_ptr : succ->MutableInstructions()) {
+        auto* phi = dynamic_cast<PhiInst*>(inst_ptr.get());
+        if (!phi) break;
+        phi->RemoveIncomingBlock(bb);
+      }
     }
     // 清除块中所有指令的 use 关系
     std::vector<Instruction*> all_insts;

src/ir/passes/LICM.cpp

@@ -57,6 +57,15 @@ bool IsSupportedInvariantOpcode(Opcode op) {
   }
 }
 
+bool IsCommutativeExpr(Instruction* inst) {
+  if (!inst || inst->GetNumOperands() != 2) return false;
+  if (inst->GetOpcode() == Opcode::Add || inst->GetOpcode() == Opcode::Mul) {
+    return true;
+  }
+  auto* cmp = dynamic_cast<CmpInst*>(inst);
+  return cmp && (cmp->GetCmpOp() == CmpOp::Eq || cmp->GetCmpOp() == CmpOp::Ne);
+}
+
 bool IsLoopInvariantValue(Value* value, analysis::Loop* loop,
                           const std::unordered_set<Instruction*>& invariant) {
   if (!value) return false;
@@ -126,6 +135,10 @@ ExprKey MakeExprKey(Instruction* inst) {
   for (size_t i = 0; i < inst->GetNumOperands(); ++i) {
     key.operands.push_back(inst->GetOperand(i));
   }
+  if (IsCommutativeExpr(inst) &&
+      std::less<Value*>()(key.operands[1], key.operands[0])) {
+    std::swap(key.operands[0], key.operands[1]);
+  }
   return key;
 }
 

src/ir/passes/LoopFission.cpp

@@ -43,6 +43,36 @@ bool DependsOnAny(Instruction* inst, const std::unordered_set<Instruction*>& def
   return false;
 }
 
+void CollectMemoryBases(const std::vector<Instruction*>& group,
+                        std::unordered_set<Value*>* loads,
+                        std::unordered_set<Value*>* stores) {
+  for (auto* inst : group) {
+    if (auto* load = dynamic_cast<LoadInst*>(inst)) {
+      loads->insert(StripPointerBase(load->GetPtr()));
+    } else if (auto* store = dynamic_cast<StoreInst*>(inst)) {
+      stores->insert(StripPointerBase(store->GetPtr()));
+    }
+  }
+}
+
+bool HasCrossMemoryDependence(const std::vector<Instruction*>& group1,
+                              const std::vector<Instruction*>& group2) {
+  std::unordered_set<Value*> loads1;
+  std::unordered_set<Value*> stores1;
+  std::unordered_set<Value*> loads2;
+  std::unordered_set<Value*> stores2;
+  CollectMemoryBases(group1, &loads1, &stores1);
+  CollectMemoryBases(group2, &loads2, &stores2);
+
+  for (auto* base : stores1) {
+    if (stores2.count(base) != 0 || loads2.count(base) != 0) return true;
+  }
+  for (auto* base : stores2) {
+    if (loads1.count(base) != 0) return true;
+  }
+  return false;
+}
+
 bool RunFissionOnLoop(Function& func, const CanonicalLoopMatch& match,
                       Context& ctx) {
   if (!IsFissionCandidate(match)) return false;
@@ -93,6 +123,7 @@ bool RunFissionOnLoop(Function& func, const CanonicalLoopMatch& match,
   for (auto* inst : group1) {
     if (DependsOnAny(inst, group2_defs)) return false;
   }
+  if (HasCrossMemoryDependence(group1, group2)) return false;
 
   auto* original_exit = match.exit;
   std::string block_suffix = ctx.NextTemp();

src/ir/passes/LoopParallelize.cpp

@@ -349,6 +349,7 @@ bool BuildGuardedFillCandidate(Function& func, analysis::Loop* loop,
 bool BuildParallelCandidate(Function& func, analysis::Loop* loop,
                             ParallelLoopCandidate* out) {
   if (BuildGuardedFillCandidate(func, loop, out)) return true;
+  if (!loop || !loop->IsParallelCandidate()) return false;
 
   auto match = MatchCanonicalLoop(loop);
   if (!match.has_value()) return false;

src/ir/passes/StrengthReduction.cpp

@@ -31,15 +31,21 @@ int ExtractScale(BinaryInst* mul, PhiInst* iv) {
   return 0;
 }
 
-bool ReplaceMulWithRecurrence(Function& func, const CanonicalLoopMatch& match,
-                              BinaryInst* mul, Context& ctx) {
-  (void)func;
-  const int scale = ExtractScale(mul, match.induction.phi);
-  if (scale == 0) return false;
-  if (match.latch != mul->GetParent() &&
-      !match.loop->Contains(mul->GetParent())) {
-    return false;
-  }
+bool HasConstantScale(BinaryInst* mul, PhiInst* iv) {
+  if (!mul || !iv) return false;
+  return (mul->GetLhs() == iv &&
+          dynamic_cast<ConstantInt*>(mul->GetRhs()) != nullptr) ||
+         (mul->GetRhs() == iv &&
+          dynamic_cast<ConstantInt*>(mul->GetLhs()) != nullptr);
+}
+
+Value* GetOrCreateScaledRecurrence(
+    const CanonicalLoopMatch& match, int scale, Context& ctx,
+    std::unordered_map<int, Value*>& recurrence_by_scale) {
+  if (scale == 0) return ctx.GetConstInt(0);
+  if (scale == 1) return match.induction.phi;
+  auto it = recurrence_by_scale.find(scale);
+  if (it != recurrence_by_scale.end()) return it->second;
 
   auto* init_scale =
       InsertBinaryBeforeTerminator(match.preheader, Opcode::Mul,
@@ -49,18 +55,26 @@ bool ReplaceMulWithRecurrence(Function& func, const CanonicalLoopMatch& match,
   auto* sr_phi =
       match.header->PrependPhi(Type::GetInt32Type(), ctx.NextTemp());
 
-  auto* step_scale =
-      InsertBinaryBeforeTerminator(match.latch, Opcode::Mul,
-                                   ctx.GetConstInt(match.induction.step),
-                                   ctx.GetConstInt(scale), ctx.NextTemp());
   auto* sr_next =
-      InsertBinaryBeforeTerminator(match.latch, Opcode::Add, sr_phi, step_scale,
+      InsertBinaryBeforeTerminator(match.latch, Opcode::Add, sr_phi,
+                                   ctx.GetConstInt(match.induction.step * scale),
                                    ctx.NextTemp());
 
   sr_phi->AddIncoming(init_scale, match.preheader);
   sr_phi->AddIncoming(sr_next, match.latch);
+  recurrence_by_scale.emplace(scale, sr_phi);
+  return sr_phi;
+}
+
+bool ReplaceMulWithRecurrence(
+    const CanonicalLoopMatch& match, BinaryInst* mul, Context& ctx,
+    std::unordered_map<int, Value*>& recurrence_by_scale) {
+  const int scale = ExtractScale(mul, match.induction.phi);
+  auto* replacement =
+      GetOrCreateScaledRecurrence(match, scale, ctx, recurrence_by_scale);
+  if (!replacement) return false;
 
-  mul->ReplaceAllUsesWith(sr_phi);
+  mul->ReplaceAllUsesWith(replacement);
   mul->GetParent()->RemoveInstruction(mul);
   return true;
 }
@@ -96,15 +110,16 @@ bool RunStrengthReduction(Function& func, Context& ctx) {
         auto* inst = inst_ptr.get();
         if (IsStrengthReductionCandidate(inst, match->induction.phi)) {
           auto* mul = static_cast<BinaryInst*>(inst);
-          if (ExtractScale(mul, match->induction.phi) != 0) {
+          if (HasConstantScale(mul, match->induction.phi)) {
             candidates.push_back(mul);
           }
         }
       }
     }
 
+    std::unordered_map<int, Value*> recurrence_by_scale;
     for (auto* mul : candidates) {
-      changed |= ReplaceMulWithRecurrence(func, *match, mul, ctx);
+      changed |= ReplaceMulWithRecurrence(*match, mul, ctx, recurrence_by_scale);
     }
   }
 

src/mir/AsmPrinter.cpp

@@ -242,6 +242,20 @@ void PrintAsm(const MachineModule& module, std::ostream& os) {
              << PhysRegName(ops.at(1).GetReg()) << "]\n";
           break;
         }
+        case Opcode::LoadIndirectScaled: {
+          // ops: wN, xM, wK
+          os << "  ldr " << PhysRegName(ops.at(0).GetReg()) << ", ["
+             << PhysRegName(ops.at(1).GetReg()) << ", "
+             << PhysRegName(ops.at(2).GetReg()) << ", uxtw #2]\n";
+          break;
+        }
+        case Opcode::StoreIndirectScaled: {
+          // ops: wN, xM, wK
+          os << "  str " << PhysRegName(ops.at(0).GetReg()) << ", ["
+             << PhysRegName(ops.at(1).GetReg()) << ", "
+             << PhysRegName(ops.at(2).GetReg()) << ", uxtw #2]\n";
+          break;
+        }
         case Opcode::LoadGlobal: {
           // adrp x9, global_var
           // add x9, x9, :lo12:global_var

src/mir/Lowering.cpp

@@ -15,15 +15,16 @@ namespace {
 
 using ValueSlotMap = std::unordered_map<const ir::Value*, int>;
 
-// GEP 结果：(base_slot_index, byte_offset, global_symbol)
+// GEP 结果：(base_slot_index, byte_offset, global_symbol, index_value)
 // - base_slot >= 0: 本地数组，base_slot 是栈槽索引
 // - base_slot = -1: 全局数组，global_symbol 是全局变量名
 // - byte_offset >= 0: 常量索引
-// - byte_offset < 0: 变量索引，编码为 -1 - index_slot
+// - byte_offset < 0: 变量索引，index_value 是原始 IR 下标值
 struct GepInfo {
   int base_slot;
   int byte_offset;
   std::string global_symbol;
+  const ir::Value* index_value = nullptr;
 };
 using GepMap = std::unordered_map<const ir::Value*, GepInfo>;
 
@@ -176,7 +177,7 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
         if (auto* const_index = dynamic_cast<const ir::ConstantInt*>(index)) {
           // 常量索引：计算地址并存储
           int byte_offset = const_index->GetValue() * 4;
-          geps.emplace(&inst, GepInfo{-1, byte_offset, gv->GetName()});
+          geps.emplace(&inst, GepInfo{-1, byte_offset, gv->GetName(), nullptr});
 
           if (ptr_slot >= 0) {
             // 计算地址：x9 = &global_array + offset
@@ -188,16 +189,11 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
           }
         } else {
           // 变量索引
-          int index_slot = function.CreateFrameIndex();
-          EmitValueToReg(index, PhysReg::W8, slots, block);
-          block.Append(Opcode::StoreStack,
-                       {Operand::Reg(PhysReg::W8), Operand::FrameIndex(index_slot)});
-          geps.emplace(&inst, GepInfo{-1, -1 - index_slot, gv->GetName()});
+          geps.emplace(&inst, GepInfo{-1, -1, gv->GetName(), index});
 
           if (ptr_slot >= 0) {
             // 计算地址：x9 = &global_array + (w10 * 4)
-            block.Append(Opcode::LoadStack,
-                         {Operand::Reg(PhysReg::W10), Operand::FrameIndex(index_slot)});
+            EmitValueToReg(index, PhysReg::W10, slots, block);
             EmitLslBy2(PhysReg::W10, block);
             block.Append(Opcode::LoadGlobalAddr,
                          {Operand::Reg(PhysReg::X9), Operand::Symbol(gv->GetName())});
@@ -238,17 +234,11 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
                        {Operand::Reg(PhysReg::X9), Operand::FrameIndex(ptr_slot)});
         } else {
           // 变量索引
-          int index_slot = function.CreateFrameIndex();
-          EmitValueToReg(index, PhysReg::W8, slots, block);
-          block.Append(Opcode::StoreStack,
-                       {Operand::Reg(PhysReg::W8), Operand::FrameIndex(index_slot)});
-
           // x9 = 从栈加载指针
           block.Append(Opcode::LoadStack,
                        {Operand::Reg(PhysReg::X9), Operand::FrameIndex(base_it->second)});
           // w10 = index * 4
-          block.Append(Opcode::LoadStack,
-                       {Operand::Reg(PhysReg::W10), Operand::FrameIndex(index_slot)});
+          EmitValueToReg(index, PhysReg::W10, slots, block);
           EmitLslBy2(PhysReg::W10, block);
           // x9 = x9 + uxtw(w10)
           block.Append(Opcode::AddRR_UXTW, {Operand::Reg(PhysReg::X9),
@@ -268,7 +258,7 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
       // 检查是否是常量索引
       if (auto* const_index = dynamic_cast<const ir::ConstantInt*>(index)) {
         int byte_offset = const_index->GetValue() * 4;
-        geps.emplace(&inst, GepInfo{base_it->second, byte_offset, ""});
+        geps.emplace(&inst, GepInfo{base_it->second, byte_offset, "", nullptr});
 
         if (ptr_slot >= 0) {
           // 计算地址：x9 = &array_base + byte_offset
@@ -280,16 +270,11 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
         }
       } else {
         // 变量索引
-        int index_slot = function.CreateFrameIndex();
-        EmitValueToReg(index, PhysReg::W8, slots, block);
-        block.Append(Opcode::StoreStack,
-                     {Operand::Reg(PhysReg::W8), Operand::FrameIndex(index_slot)});
-        geps.emplace(&inst, GepInfo{base_it->second, -1 - index_slot, ""});
+        geps.emplace(&inst, GepInfo{base_it->second, -1, "", index});
 
         if (ptr_slot >= 0) {
           // 计算地址：x9 = x29 + base_offset + (w10 * 4)
-          block.Append(Opcode::LoadStack,
-                       {Operand::Reg(PhysReg::W10), Operand::FrameIndex(index_slot)});
+          EmitValueToReg(index, PhysReg::W10, slots, block);
           EmitLslBy2(PhysReg::W10, block);
           block.Append(Opcode::LoadStackAddr,
                        {Operand::Reg(PhysReg::X9), Operand::FrameIndex(base_it->second)});
@@ -330,22 +315,12 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
                          {Operand::Reg(src_reg), Operand::Reg(PhysReg::X9)});
           } else {
             // 变量索引：global_array[var_idx]
-            int index_slot = -1 - gep_info.byte_offset;
-            // 1. 加载 index（4字节 W 寄存器）
-            block.Append(Opcode::LoadStack,
-                         {Operand::Reg(PhysReg::W10), Operand::FrameIndex(index_slot)});
-            // 2. index * 4
-            EmitLslBy2(PhysReg::W10, block);
-            // 3. 获取全局数组基址
+            EmitValueToReg(gep_info.index_value, PhysReg::W10, slots, block);
             block.Append(Opcode::LoadGlobalAddr,
                          {Operand::Reg(PhysReg::X9), Operand::Symbol(gep_info.global_symbol)});
-            // 4. x9 + w10, uxtw（零扩展 W 寄存器后加到 X 寄存器）
-            block.Append(Opcode::AddRR_UXTW, {Operand::Reg(PhysReg::X9),
-                                         Operand::Reg(PhysReg::X9),
-                                         Operand::Reg(PhysReg::W10)});
-            // 5. 存储
-            block.Append(Opcode::StoreIndirect,
-                         {Operand::Reg(src_reg), Operand::Reg(PhysReg::X9)});
+            block.Append(Opcode::StoreIndirectScaled,
+                         {Operand::Reg(src_reg), Operand::Reg(PhysReg::X9),
+                          Operand::Reg(PhysReg::W10)});
           }
         } else if (gep_info.byte_offset >= 0) {
           // 本地数组，常量索引
@@ -355,18 +330,13 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
                         Operand::Imm(gep_info.byte_offset)});
         } else {
           // 本地数组，变量索引
-          int index_slot = -1 - gep_info.byte_offset;
-          block.Append(Opcode::LoadStack,
-                       {Operand::Reg(PhysReg::W10), Operand::FrameIndex(index_slot)});
-          EmitLslBy2(PhysReg::W10, block);
+          EmitValueToReg(gep_info.index_value, PhysReg::W10, slots, block);
           block.Append(Opcode::LoadStackAddr,
                        {Operand::Reg(PhysReg::X9),
                         Operand::FrameIndex(gep_info.base_slot)});
-          block.Append(Opcode::AddRR_UXTW, {Operand::Reg(PhysReg::X9),
-                                       Operand::Reg(PhysReg::X9),
-                                       Operand::Reg(PhysReg::W10)});
-          block.Append(Opcode::StoreIndirect,
-                       {Operand::Reg(src_reg), Operand::Reg(PhysReg::X9)});
+          block.Append(Opcode::StoreIndirectScaled,
+                       {Operand::Reg(src_reg), Operand::Reg(PhysReg::X9),
+                        Operand::Reg(PhysReg::W10)});
         }
         return;
       }
@@ -426,17 +396,12 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
                          {Operand::Reg(value_reg), Operand::Reg(PhysReg::X9)});
           } else {
             // 变量索引
-            int index_slot = -1 - gep_info.byte_offset;
-            block.Append(Opcode::LoadStack,
-                         {Operand::Reg(PhysReg::W10), Operand::FrameIndex(index_slot)});
-            EmitLslBy2(PhysReg::W10, block);
+            EmitValueToReg(gep_info.index_value, PhysReg::W10, slots, block);
             block.Append(Opcode::LoadGlobalAddr,
                          {Operand::Reg(PhysReg::X9), Operand::Symbol(gep_info.global_symbol)});
-            block.Append(Opcode::AddRR_UXTW, {Operand::Reg(PhysReg::X9),
-                                         Operand::Reg(PhysReg::X9),
-                                         Operand::Reg(PhysReg::W10)});
-            block.Append(Opcode::LoadIndirect,
-                         {Operand::Reg(value_reg), Operand::Reg(PhysReg::X9)});
+            block.Append(Opcode::LoadIndirectScaled,
+                         {Operand::Reg(value_reg), Operand::Reg(PhysReg::X9),
+                          Operand::Reg(PhysReg::W10)});
           }
         } else if (gep_info.byte_offset >= 0) {
           // 本地数组，常量索引
@@ -446,18 +411,13 @@ void LowerInstruction(const ir::Instruction& inst, MachineFunction& function,
                         Operand::Imm(gep_info.byte_offset)});
         } else {
           // 本地数组，变量索引
-          int index_slot = -1 - gep_info.byte_offset;
-          block.Append(Opcode::LoadStack,
-                       {Operand::Reg(PhysReg::W10), Operand::FrameIndex(index_slot)});
-          EmitLslBy2(PhysReg::W10, block);
+          EmitValueToReg(gep_info.index_value, PhysReg::W10, slots, block);
           block.Append(Opcode::LoadStackAddr,
                        {Operand::Reg(PhysReg::X9),
                         Operand::FrameIndex(gep_info.base_slot)});
-          block.Append(Opcode::AddRR_UXTW, {Operand::Reg(PhysReg::X9),
-                                       Operand::Reg(PhysReg::X9),
-                                       Operand::Reg(PhysReg::W10)});
-          block.Append(Opcode::LoadIndirect,
-                       {Operand::Reg(value_reg), Operand::Reg(PhysReg::X9)});
+          block.Append(Opcode::LoadIndirectScaled,
+                       {Operand::Reg(value_reg), Operand::Reg(PhysReg::X9),
+                        Operand::Reg(PhysReg::W10)});
         }
 
         block.Append(Opcode::StoreStack,
@@ -1120,7 +1080,10 @@ std::unique_ptr<MachineModule> LowerToMIR(const ir::Module& module) {
         }
         if (!can_emit) continue;  // 跳过无法发射的值
 
-        PhysReg tmp = phi_info.is_float ? PhysReg::S8 : PhysReg::W8;
+        // Phi edge stores may be inserted immediately before fused cbz/cbnz
+        // branches, which use w8 as the condition register. Use a different
+        // scratch register so edge materialization does not clobber the branch.
+        PhysReg tmp = phi_info.is_float ? PhysReg::S10 : PhysReg::W10;
         MachineBasicBlock tmp_block("__phi_tmp__");
         EmitValueToReg(val, tmp, slots, tmp_block);
         tmp_block.Append(Opcode::StoreStack,

src/mir/RegAlloc.cpp

@@ -299,10 +299,19 @@ void PromoteToVRegs(MachineFunction& func,
           if (!ops.empty()) PromoteUse(ops[0]);
           if (ops.size() > 1) PromoteUse(ops[1]);
           break;
+        case Opcode::StoreIndirectScaled:
+          if (!ops.empty()) PromoteUse(ops[0]);
+          if (ops.size() > 1) PromoteUse(ops[1]);
+          if (ops.size() > 2) PromoteUse(ops[2]);
+          break;
 
         case Opcode::LoadIndirect:
           if (ops.size() > 1) PromoteUse(ops[1]);
           break;
+        case Opcode::LoadIndirectScaled:
+          if (ops.size() > 1) PromoteUse(ops[1]);
+          if (ops.size() > 2) PromoteUse(ops[2]);
+          break;
 
         case Opcode::AddRI: case Opcode::SubRI:
         case Opcode::LsrRI: case Opcode::LslRI:
@@ -334,6 +343,7 @@ void PromoteToVRegs(MachineFunction& func,
         case Opcode::MovReg: case Opcode::FMovReg:
         case Opcode::LoadStack: case Opcode::LoadStackOffset:
         case Opcode::LoadStackAddr: case Opcode::LoadIndirect:
+        case Opcode::LoadIndirectScaled:
         case Opcode::LoadGlobal: case Opcode::LoadGlobalAddr:
         case Opcode::AddRI: case Opcode::SubRI:
         case Opcode::AddRR: case Opcode::AddRR_UXTW: case Opcode::SubRR:
@@ -382,9 +392,18 @@ void GetVRegDefsUses(const MachineInstr& inst,
       if (!ops.empty() && ops[0].IsVReg()) uses.insert(ops[0].GetVRegId());
       if (ops.size() > 1 && ops[1].IsVReg()) uses.insert(ops[1].GetVRegId());
       break;
+    case Opcode::StoreIndirectScaled:
+      if (!ops.empty() && ops[0].IsVReg()) uses.insert(ops[0].GetVRegId());
+      if (ops.size() > 1 && ops[1].IsVReg()) uses.insert(ops[1].GetVRegId());
+      if (ops.size() > 2 && ops[2].IsVReg()) uses.insert(ops[2].GetVRegId());
+      break;
     case Opcode::LoadIndirect:
       if (ops.size() > 1 && ops[1].IsVReg()) uses.insert(ops[1].GetVRegId());
       break;
+    case Opcode::LoadIndirectScaled:
+      if (ops.size() > 1 && ops[1].IsVReg()) uses.insert(ops[1].GetVRegId());
+      if (ops.size() > 2 && ops[2].IsVReg()) uses.insert(ops[2].GetVRegId());
+      break;
     case Opcode::AddRI: case Opcode::SubRI:
     case Opcode::LsrRI: case Opcode::LslRI:
     case Opcode::FSqrtRR: case Opcode::SIToFP: case Opcode::FPToSI:
@@ -419,6 +438,7 @@ void GetVRegDefsUses(const MachineInstr& inst,
     case Opcode::MovReg: case Opcode::FMovReg:
     case Opcode::LoadStack: case Opcode::LoadStackOffset:
     case Opcode::LoadStackAddr: case Opcode::LoadIndirect:
+    case Opcode::LoadIndirectScaled:
     case Opcode::LoadGlobal: case Opcode::LoadGlobalAddr:
     case Opcode::AddRI: case Opcode::SubRI:
     case Opcode::AddRR: case Opcode::AddRR_UXTW: case Opcode::SubRR:
@@ -514,6 +534,14 @@ void GetPhysRegDefsUses(const MachineInstr& inst,
       if (ops.size() > 1 && ops[1].IsReg() && !ShouldPromote(ops[1].GetReg()))
         phys_uses.insert(ops[1].GetReg());
       break;
+    case Opcode::StoreIndirectScaled:
+      if (!ops.empty() && ops[0].IsReg() && !ShouldPromote(ops[0].GetReg()))
+        phys_uses.insert(ops[0].GetReg());
+      if (ops.size() > 1 && ops[1].IsReg() && !ShouldPromote(ops[1].GetReg()))
+        phys_uses.insert(ops[1].GetReg());
+      if (ops.size() > 2 && ops[2].IsReg() && !ShouldPromote(ops[2].GetReg()))
+        phys_uses.insert(ops[2].GetReg());
+      break;
     case Opcode::AddRI: case Opcode::SubRI:
     case Opcode::LsrRI: case Opcode::LslRI:
     case Opcode::FSqrtRR: case Opcode::SIToFP: case Opcode::FPToSI:
@@ -545,6 +573,12 @@ void GetPhysRegDefsUses(const MachineInstr& inst,
       if (ops.size() > 1 && ops[1].IsReg() && !ShouldPromote(ops[1].GetReg()))
         phys_uses.insert(ops[1].GetReg());
       break;
+    case Opcode::LoadIndirectScaled:
+      if (ops.size() > 1 && ops[1].IsReg() && !ShouldPromote(ops[1].GetReg()))
+        phys_uses.insert(ops[1].GetReg());
+      if (ops.size() > 2 && ops[2].IsReg() && !ShouldPromote(ops[2].GetReg()))
+        phys_uses.insert(ops[2].GetReg());
+      break;
     default:
       break;
   }
@@ -555,6 +589,7 @@ void GetPhysRegDefsUses(const MachineInstr& inst,
     case Opcode::MovReg: case Opcode::FMovReg:
     case Opcode::LoadStack: case Opcode::LoadStackOffset:
     case Opcode::LoadStackAddr: case Opcode::LoadIndirect:
+    case Opcode::LoadIndirectScaled:
     case Opcode::LoadGlobal: case Opcode::LoadGlobalAddr:
     case Opcode::AddRI: case Opcode::SubRI:
     case Opcode::AddRR: case Opcode::AddRR_UXTW: case Opcode::SubRR:

src/mir/passes/Peephole.cpp

@@ -74,6 +74,7 @@ std::optional<PhysReg> GetWrittenReg(const MachineInstr& inst) {
 		case Opcode::LoadStackOffset:
 		case Opcode::LoadStackAddr:
 		case Opcode::LoadIndirect:
+		case Opcode::LoadIndirectScaled:
 		case Opcode::LoadGlobal:
 		case Opcode::LoadGlobalAddr:
 		case Opcode::AddRI:
@@ -119,6 +120,8 @@ bool ReadsReg(const MachineInstr& inst, PhysReg reg) {
 		case Opcode::FPToSI:
 		case Opcode::LslRI:
 			return reads_operand(1);
+		case Opcode::LoadIndirectScaled:
+			return reads_operand(1) || reads_operand(2);
 		case Opcode::AddRR:
 		case Opcode::AddRR_UXTW:
 		case Opcode::SubRR:
@@ -141,6 +144,8 @@ bool ReadsReg(const MachineInstr& inst, PhysReg reg) {
 			return reads_operand(0);
 		case Opcode::StoreIndirect:
 			return reads_operand(0) || reads_operand(1);
+		case Opcode::StoreIndirectScaled:
+			return reads_operand(0) || reads_operand(1) || reads_operand(2);
 		case Opcode::Ret:
 			return false;
 		default:
@@ -158,6 +163,7 @@ bool CanElideIfOverwritten(const MachineInstr& inst) {
 		case Opcode::LoadStackOffset:
 		case Opcode::LoadStackAddr:
 		case Opcode::LoadIndirect:
+		case Opcode::LoadIndirectScaled:
 		case Opcode::LoadGlobal:
 		case Opcode::LoadGlobalAddr:
 		case Opcode::AddRI:
@@ -184,6 +190,7 @@ bool CanElideIfOverwritten(const MachineInstr& inst) {
 bool IsMemoryClobber(const MachineInstr& inst) {
 	switch (inst.GetOpcode()) {
 		case Opcode::StoreIndirect:
+		case Opcode::StoreIndirectScaled:
 		case Opcode::StoreGlobal:
 		case Opcode::Bl:
 			return true;

优化方案.md

@@ -0,0 +1,41 @@
+# Lab6 循环优化推进记录
+
+## 1. 循环不变代码外提（LICM）
+
+- 方案：对可交换的不变表达式（`add`、`mul`、`eq/ne cmp`）做操作数规范化，外提到 preheader 后能复用 `a + b` 与 `b + a` 这类等价表达式。
+- 代码位置：`src/ir/passes/LICM.cpp`，新增 `IsCommutativeExpr`，并在 `MakeExprKey` 中规范化 key。
+
+## 2. 强度削弱
+
+- 方案：同一循环内多个 `iv * C` 共享同一个尺度递推 phi，`iv * 0` 与 `iv * 1` 直接替换为常量或归纳变量；步长增量直接常量化为 `step * C`，减少 latch 中多余乘法。
+- 代码位置：`src/ir/passes/StrengthReduction.cpp`，新增 `GetOrCreateScaledRecurrence`，按 scale 复用递推值。
+
+## 3. 循环展开
+
+- 方案：补强 CFG 边维护的基础设施，删除旧 terminator 时同步清理旧前驱/后继边，避免循环展开、习语替换等 pass 改写分支后留下过期 CFG 边。
+- 代码位置：`src/ir/BasicBlock.cpp`，在 `BasicBlock::RemoveInstruction` 中处理 `br/condbr` 的 CFG 边。
+
+## 4. 循环分裂
+
+- 方案：分裂前额外检查两组语句的 load/store 基址依赖；只要跨组存在 store-store 或 store-load 同基址关系，就放弃分裂，保证循环拆分不改变内存可见顺序。
+- 代码位置：`src/ir/passes/LoopFission.cpp`，新增 `CollectMemoryBases` 与 `HasCrossMemoryDependence`。
+
+## 5. 循环并行化
+
+- 方案：普通点式循环并行化先经过 `LoopInfo::IsParallelCandidate` 过滤，再进入 worker 抽取逻辑；保留已有 guarded fill 特例，减少非候选循环误判和无效分析。
+- 代码位置：`src/ir/passes/LoopParallelize.cpp`，在 `BuildParallelCandidate` 中加入候选过滤。
+
+## 6. 验证中发现的后端正确性修复
+
+- 方案：phi 消除插入边上赋值时不再使用 `w8/s8` 作为临时寄存器，避免覆盖 `cbz/cbnz` 融合分支刚准备好的条件寄存器。
+- 代码位置：`src/mir/Lowering.cpp`，phi elimination 的临时寄存器改为 `w10/s10`。
+
+## 7. ASM 后端数组访存优化
+
+- 方案：GEP 变量下标不再先写入临时栈槽，直接保存原始 IR 下标值，访存时发射到 `w10`；同时把变量下标数组访存改为 AArch64 scaled addressing（`[base, index, uxtw #2]`），减少热循环中的 `store/load index`、`lsl` 与 `add`。
+- 代码位置：`src/mir/Lowering.cpp`、`include/mir/MIR.h`、`src/mir/AsmPrinter.cpp`、`src/mir/RegAlloc.cpp`、`src/mir/passes/Peephole.cpp`。
+
+## 回归编译错误修复
+
+- 方案：IR 文本打印补齐 `ConstantFloat` 的 LLVM 十六进制浮点常量格式，把浮点二元运算/比较输出为 `fadd/fsub/fmul/fdiv`、`fcmp o*`；比较结果打印为 `i1` 后再 `zext` 回内部 `i32` 约定；删除不可达块时同步清理后继 phi 入边；scaled 间接访存补齐 peephole 的读寄存器与内存 clobber 描述。
+- 代码位置：`src/ir/IRPrinter.cpp`、`src/ir/passes/CFGSimplify.cpp`、`src/mir/passes/Peephole.cpp`。