[LoongArch] Override TargetFrameLowering::spillCalleeSavedRegisters

When using `llvm.returnaddress` intrinsic, special handling is required
for the spill of the `RA` register. Otherwise it will cause the verifier
fail in some cases (e.g. pr17377.c of the GCC C Torture Suite).

Specifically:
```
*** Bad machine code: Using an undefined physical register ***
- function:    f
- basic block: %bb.0 entry (0xd94d18)
- instruction: ST_D killed $r1, $r22, -40 :: (store (s64) into %stack.2)
- operand 0:   killed $r1
```

Reviewed By: SixWeining

Differential Revision: https://reviews.llvm.org/D137387

Author: wangleiat

llvm/lib/Target/LoongArch/LoongArchFrameLowering.cpp

@@ -406,6 +406,29 @@ LoongArchFrameLowering::eliminateCallFramePseudoInstr(
   return MBB.erase(MI);
 }
 
+bool LoongArchFrameLowering::spillCalleeSavedRegisters(
+    MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
+    ArrayRef<CalleeSavedInfo> CSI, const TargetRegisterInfo *TRI) const {
+  if (CSI.empty())
+    return true;
+
+  MachineFunction *MF = MBB.getParent();
+  const TargetInstrInfo &TII = *MF->getSubtarget().getInstrInfo();
+
+  // Insert the spill to the stack frame.
+  for (auto &CS : CSI) {
+    Register Reg = CS.getReg();
+    // If the register is RA and the return address is taken by method
+    // LoongArchTargetLowering::lowerRETURNADDR, don't set kill flag.
+    bool IsKill =
+        !(Reg == LoongArch::R1 && MF->getFrameInfo().isReturnAddressTaken());
+    const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
+    TII.storeRegToStackSlot(MBB, MI, Reg, IsKill, CS.getFrameIdx(), RC, TRI);
+  }
+
+  return true;
+}
+
 StackOffset LoongArchFrameLowering::getFrameIndexReference(
     const MachineFunction &MF, int FI, Register &FrameReg) const {
   const MachineFrameInfo &MFI = MF.getFrameInfo();

llvm/lib/Target/LoongArch/LoongArchFrameLowering.h

@@ -41,6 +41,10 @@ class LoongArchFrameLowering : public TargetFrameLowering {
   MachineBasicBlock::iterator
   eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
                                 MachineBasicBlock::iterator MI) const override;
+  bool spillCalleeSavedRegisters(MachineBasicBlock &MBB,
+                                 MachineBasicBlock::iterator MI,
+                                 ArrayRef<CalleeSavedInfo> CSI,
+                                 const TargetRegisterInfo *TRI) const override;
 
   StackOffset getFrameIndexReference(const MachineFunction &MF, int FI,
                                      Register &FrameReg) const override;

llvm/test/CodeGen/LoongArch/spill-ra-without-kill.ll

@@ -0,0 +1,102 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -O0 --mtriple=loongarch64 --verify-machineinstrs < %s | FileCheck %s
+
+;; This test case is reduced from pr17377.c of the GCC C Torture Suite using
+;; bugpoint.
+
+@calls = external dso_local global i32, align 4
+declare ptr @llvm.returnaddress(i32 immarg)
+
+define dso_local ptr @f(i32 noundef signext %i) "frame-pointer"="all" {
+; CHECK-LABEL: f:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    addi.d $sp, $sp, -48
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    st.d $ra, $sp, 40 # 8-byte Folded Spill
+; CHECK-NEXT:    st.d $fp, $sp, 32 # 8-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset 1, -8
+; CHECK-NEXT:    .cfi_offset 22, -16
+; CHECK-NEXT:    addi.d $fp, $sp, 48
+; CHECK-NEXT:    .cfi_def_cfa 22, 0
+; CHECK-NEXT:    st.d $ra, $fp, -40 # 8-byte Folded Spill
+; CHECK-NEXT:    move $a1, $a0
+; CHECK-NEXT:    pcalau12i $a0, %pc_hi20(calls)
+; CHECK-NEXT:    addi.d $a3, $a0, %pc_lo12(calls)
+; CHECK-NEXT:    ld.wu $a0, $a3, 0
+; CHECK-NEXT:    addi.d $a2, $a0, 1
+; CHECK-NEXT:    st.w $a2, $a3, 0
+; CHECK-NEXT:    st.w $a1, $fp, -28
+; CHECK-NEXT:    bnez $a0, .LBB0_2
+; CHECK-NEXT:    b .LBB0_1
+; CHECK-NEXT:  .LBB0_1: # %if.then
+; CHECK-NEXT:    ld.d $a0, $fp, -40 # 8-byte Folded Reload
+; CHECK-NEXT:    st.d $a0, $fp, -24
+; CHECK-NEXT:    b .LBB0_7
+; CHECK-NEXT:  .LBB0_2: # %if.end
+; CHECK-NEXT:    ld.wu $a0, $fp, -28
+; CHECK-NEXT:    st.d $a0, $fp, -48 # 8-byte Folded Spill
+; CHECK-NEXT:    beqz $a0, .LBB0_5
+; CHECK-NEXT:    b .LBB0_3
+; CHECK-NEXT:  .LBB0_3: # %if.end
+; CHECK-NEXT:    ld.d $a0, $fp, -48 # 8-byte Folded Reload
+; CHECK-NEXT:    ori $a1, $zero, 1
+; CHECK-NEXT:    bne $a0, $a1, .LBB0_6
+; CHECK-NEXT:    b .LBB0_4
+; CHECK-NEXT:  .LBB0_4: # %sw.bb
+; CHECK-NEXT:    pcalau12i $a0, %pc_hi20(f)
+; CHECK-NEXT:    addi.d $a0, $a0, %pc_lo12(f)
+; CHECK-NEXT:    st.d $a0, $fp, -24
+; CHECK-NEXT:    b .LBB0_7
+; CHECK-NEXT:  .LBB0_5: # %sw.bb1
+; CHECK-NEXT:    ld.d $a0, $fp, -40 # 8-byte Folded Reload
+; CHECK-NEXT:    st.d $a0, $fp, -24
+; CHECK-NEXT:    b .LBB0_7
+; CHECK-NEXT:  .LBB0_6: # %sw.epilog
+; CHECK-NEXT:    move $a0, $zero
+; CHECK-NEXT:    st.d $a0, $fp, -24
+; CHECK-NEXT:    b .LBB0_7
+; CHECK-NEXT:  .LBB0_7: # %return
+; CHECK-NEXT:    ld.d $a0, $fp, -24
+; CHECK-NEXT:    ld.d $fp, $sp, 32 # 8-byte Folded Reload
+; CHECK-NEXT:    ld.d $ra, $sp, 40 # 8-byte Folded Reload
+; CHECK-NEXT:    addi.d $sp, $sp, 48
+; CHECK-NEXT:    ret
+entry:
+  %retval = alloca ptr, align 8
+  %i.addr = alloca i32, align 4
+  store i32 %i, ptr %i.addr, align 4
+  %0 = load i32, ptr @calls, align 4
+  %inc = add nsw i32 %0, 1
+  store i32 %inc, ptr @calls, align 4
+  %cmp = icmp eq i32 %0, 0
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:
+  %1 = call ptr @llvm.returnaddress(i32 0)
+  store ptr %1, ptr %retval, align 8
+  br label %return
+
+if.end:
+  %2 = load i32, ptr %i.addr, align 4
+  switch i32 %2, label %sw.epilog [
+    i32 1, label %sw.bb
+    i32 0, label %sw.bb1
+  ]
+
+sw.bb:
+  store ptr @f, ptr %retval, align 8
+  br label %return
+
+sw.bb1:
+  %3 = call ptr @llvm.returnaddress(i32 0)
+  store ptr %3, ptr %retval, align 8
+  br label %return
+
+sw.epilog:
+  store ptr null, ptr %retval, align 8
+  br label %return
+
+return:
+  %4 = load ptr, ptr %retval, align 8
+  ret ptr %4
+}