[AArch64] Fix the size passed to __trampoline_setup (#118234)

The trampoline size is 36 bytes on AArch64. The runtime function
__trampoline_setup aborts as it expects the trampoline size of at least 36 
bytes, and the size passed is 20 bytes. Fix the inconsistency in
AArch64TargetLowering::LowerINIT_TRAMPOLINE.

Author: ssijaric-nv

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -7281,9 +7281,16 @@ SDValue AArch64TargetLowering::LowerINIT_TRAMPOLINE(SDValue Op,
   Entry.Ty = IntPtrTy;
   Entry.Node = Trmp;
   Args.push_back(Entry);
-  Entry.Node = DAG.getConstant(20, dl, MVT::i64);
-  Args.push_back(Entry);
 
+  if (auto *FI = dyn_cast<FrameIndexSDNode>(Trmp.getNode())) {
+    MachineFunction &MF = DAG.getMachineFunction();
+    MachineFrameInfo &MFI = MF.getFrameInfo();
+    Entry.Node =
+        DAG.getConstant(MFI.getObjectSize(FI->getIndex()), dl, MVT::i64);
+  } else
+    Entry.Node = DAG.getConstant(36, dl, MVT::i64);
+
+  Args.push_back(Entry);
   Entry.Node = FPtr;
   Args.push_back(Entry);
   Entry.Node = Nest;

llvm/test/CodeGen/AArch64/trampoline.ll

@@ -1,5 +1,7 @@
 ; RUN: llc -mtriple=aarch64-- < %s | FileCheck %s
 
+@trampg = internal global [36 x i8] zeroinitializer, align 8
+
 declare void @llvm.init.trampoline(ptr, ptr, ptr);
 declare ptr @llvm.adjust.trampoline(ptr);
 
@@ -8,12 +10,23 @@ define i64 @f(ptr nest %c, i64 %x, i64 %y) {
   ret i64 %sum
 }
 
-define i64 @main() {
+define i64 @func1() {
   %val = alloca i64
   %nval = bitcast ptr %val to ptr
   %tramp = alloca [36 x i8], align 8
+  ; CHECK:	mov	w1, #36
   ; CHECK:	bl	__trampoline_setup
   call void @llvm.init.trampoline(ptr %tramp, ptr @f, ptr %nval)
   %fp = call ptr @llvm.adjust.trampoline(ptr %tramp)
   ret i64 0
 }
+
+define i64 @func2() {
+  %val = alloca i64
+  %nval = bitcast ptr %val to ptr
+  ; CHECK:	mov	w1, #36
+  ; CHECK:	bl	__trampoline_setup
+  call void @llvm.init.trampoline(ptr @trampg, ptr @f, ptr %nval)
+  %fp = call ptr @llvm.adjust.trampoline(ptr @trampg)
+  ret i64 0
+}