[M68k] fix call frame destruction elimination when returning structs (#107579)

Fixes #106213.

This adjusts `eliminateCallFramePseudoInstr` to match the behaviour of
the X86 backend.

Author: TechnoElf

llvm/lib/Target/M68k/M68kFrameLowering.cpp

@@ -33,6 +33,8 @@
 
 using namespace llvm;
 
+#define DEBUG_TYPE "m68k-frame"
+
 M68kFrameLowering::M68kFrameLowering(const M68kSubtarget &STI, Align Alignment)
     : TargetFrameLowering(StackGrowsDown, Alignment, -4), STI(STI),
       TII(*STI.getInstrInfo()), TRI(STI.getRegisterInfo()) {
@@ -231,8 +233,8 @@ MachineBasicBlock::iterator M68kFrameLowering::eliminateCallFramePseudoInstr(
   unsigned Opcode = I->getOpcode();
   bool IsDestroy = Opcode == TII.getCallFrameDestroyOpcode();
   DebugLoc DL = I->getDebugLoc();
-  uint64_t Amount = !ReserveCallFrame ? I->getOperand(0).getImm() : 0;
-  uint64_t InternalAmt = (IsDestroy && Amount) ? I->getOperand(1).getImm() : 0;
+  uint64_t Amount = I->getOperand(0).getImm();
+  uint64_t InternalAmt = (IsDestroy || Amount) ? I->getOperand(1).getImm() : 0;
   I = MBB.erase(I);
 
   if (!ReserveCallFrame) {

llvm/test/CodeGen/M68k/multiple-return.ll

@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=m68k-linux -verify-machineinstrs | FileCheck %s
 
-define { i32, i32, i32, i32 } @test() {
-; CHECK-LABEL: test:
+define { i32, i32, i32, i32 } @test0() {
+; CHECK-LABEL: test0:
 ; CHECK:         .cfi_startproc
 ; CHECK-NEXT:  ; %bb.0: ; %start
 ; CHECK-NEXT:    move.l (4,%sp), %a0
@@ -18,3 +18,73 @@ define { i32, i32, i32, i32 } @test() {
 start:
   ret { i32, i32, i32, i32 } { i32 13, i32 17, i32 19, i32 23 }
 }
+
+define void @call_test0() {
+; CHECK-LABEL: call_test0:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0: ; %start
+; CHECK-NEXT:    suba.l #20, %sp
+; CHECK-NEXT:    .cfi_def_cfa_offset -24
+; CHECK-NEXT:    lea (4,%sp), %a0
+; CHECK-NEXT:    move.l %a0, (%sp)
+; CHECK-NEXT:    jsr test0
+; CHECK-NEXT:    adda.l #16, %sp
+; CHECK-NEXT:    rts
+start:
+  %val = call { i32, i32, i32, i32 } @test0()
+  ret void
+}
+
+define void @test1(ptr sret({ i32, i32, i32, i32 }) %ret_val) {
+; CHECK-LABEL: test1:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0: ; %start
+; CHECK-NEXT:    move.l (4,%sp), %d0
+; CHECK-NEXT:    move.l (%sp), %a1
+; CHECK-NEXT:    adda.l #4, %sp
+; CHECK-NEXT:    move.l %a1, (%sp)
+; CHECK-NEXT:    rts
+start:
+  ret void
+}
+
+define void @call_test1() {
+; CHECK-LABEL: call_test1:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0: ; %start
+; CHECK-NEXT:    suba.l #20, %sp
+; CHECK-NEXT:    .cfi_def_cfa_offset -24
+; CHECK-NEXT:    lea (4,%sp), %a0
+; CHECK-NEXT:    move.l %a0, (%sp)
+; CHECK-NEXT:    jsr test1
+; CHECK-NEXT:    adda.l #16, %sp
+; CHECK-NEXT:    rts
+start:
+  %ret_val = alloca { i32, i32, i32, i32 }
+  call void @test1(ptr %ret_val)
+  ret void
+}
+
+define i32 @test2() {
+; CHECK-LABEL: test2:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0: ; %start
+; CHECK-NEXT:    moveq #13, %d0
+; CHECK-NEXT:    rts
+start:
+  ret i32 13
+}
+
+define void @call_test2() {
+; CHECK-LABEL: call_test2:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0: ; %start
+; CHECK-NEXT:    suba.l #4, %sp
+; CHECK-NEXT:    .cfi_def_cfa_offset -8
+; CHECK-NEXT:    jsr test2
+; CHECK-NEXT:    adda.l #4, %sp
+; CHECK-NEXT:    rts
+start:
+  %0 = call i32 @test2()
+  ret void
+}