perf(mir): 添加 AddShiftRR/SubShiftRR 支持——lsl+add 合并为 add lsl

- MIR.h: 新增 AddShiftRR/SubShiftRR 操作码（4 操作数: dst, src1, src2, shift）
- RegAlloc: 更新 def/use 分析
- AsmPrinter: 发射 add/sub dst, src1, src2, lsl #shift
- Peephole: 相邻 ShlRR+AddRR/SubRR → AddShiftRR/SubShiftRR

为后续在 Lowering 阶段直接生成移位操作数铺路。

src/include/mir/MIR.h

@@ -147,6 +147,8 @@ namespace mir
     StoreMem,
     AddRR,
     SubRR,
+    AddShiftRR,
+    SubShiftRR,
     MulRR,
     DivRR,
     ModRR,

src/mir/AsmPrinter.cpp

@@ -42,8 +42,10 @@ namespace mir
             case Opcode::StoreStack:
                 return "stur";
             case Opcode::AddRR:
+            case Opcode::AddShiftRR:
                 return "add";
             case Opcode::SubRR:
+            case Opcode::SubShiftRR:
                 return "sub";
             case Opcode::MulRR:
                 return "mul";
@@ -812,6 +814,20 @@ namespace mir
                 }
                 return;
 
+            case Opcode::AddShiftRR:
+            case Opcode::SubShiftRR:
+                if (operands.size() >= 4)
+                {
+                    os << "    " << asm_op << " ";
+                    PrintOperand(operands[0], os);
+                    os << ", ";
+                    PrintOperand(operands[1], os);
+                    os << ", ";
+                    PrintOperand(operands[2], os);
+                    os << ", lsl #" << operands[3].GetImm() << "\n";
+                }
+                return;
+
             default:
                 break;
             }

src/mir/RegAlloc.cpp

@@ -123,6 +123,8 @@ namespace mir
 
       case Opcode::AddRR:
       case Opcode::SubRR:
+      case Opcode::AddShiftRR:
+      case Opcode::SubShiftRR:
       case Opcode::MulRR:
       case Opcode::DivRR:
       case Opcode::ModRR:

src/mir/passes/Peephole.cpp

@@ -256,6 +256,56 @@ namespace mir
         }
       }
 
+      // ShlRR + AddRR/SubRR → AddShiftRR/SubShiftRR (lsl wA,wB,#n; add wC,wB,wA → add wC,wB,wB,lsl#n)
+      if (!changed)
+      {
+        for (auto it = insts.begin(); it != insts.end(); ++it)
+        {
+          if (it->GetOpcode() != Opcode::ShlRR) continue;
+          auto next = std::next(it);
+          if (next == insts.end()) continue;
+          if (next->GetOpcode() != Opcode::AddRR && next->GetOpcode() != Opcode::SubRR)
+            continue;
+
+          const auto &shl_ops = it->GetOperands();
+          const auto &arith_ops = next->GetOperands();
+          if (shl_ops.size() < 3 || arith_ops.size() < 3) continue;
+
+          // shl: [dst(A), src(B), shift_imm]
+          // add/sub: [dst(C), src1, src2]
+          // 检查 shl 的 dst 是否等于 add/sub 的一个 src，且 shl 的 src 等于 add/sub 的另一个 src
+          PhysReg shl_dst = shl_ops[0].GetReg();
+          PhysReg shl_src = shl_ops[1].GetReg();
+          int shift = shl_ops[2].GetImm();
+
+          PhysReg arith_src1 = arith_ops[1].GetReg();
+          PhysReg arith_src2 = arith_ops[2].GetReg();
+
+          if (shl_dst == arith_src1 && shl_src == arith_src2)
+          {
+            // lsl A, B, #n; add C, A, B → add C, B, B, lsl #n
+            auto new_op = (next->GetOpcode() == Opcode::AddRR)
+                              ? Opcode::AddShiftRR : Opcode::SubShiftRR;
+            *next = MachineInstr(new_op, {arith_ops[0], shl_ops[1], shl_ops[1],
+                                           Operand::Imm(shift)});
+            it = insts.erase(it);
+            changed = true;
+            break;
+          }
+          if (shl_dst == arith_src2 && shl_src == arith_src1)
+          {
+            // lsl A, B, #n; add C, B, A → add C, B, B, lsl #n
+            auto new_op = (next->GetOpcode() == Opcode::AddRR)
+                              ? Opcode::AddShiftRR : Opcode::SubShiftRR;
+            *next = MachineInstr(new_op, {arith_ops[0], shl_ops[1], shl_ops[1],
+                                           Operand::Imm(shift)});
+            it = insts.erase(it);
+            changed = true;
+            break;
+          }
+        }
+      }
+
       // 消除冗余 adrp：同一符号、同一寄存器的重复 adrp
       if (!changed)
       {