diff --git a/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp b/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
--- a/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
@@ -3429,7 +3429,11 @@
   Register BaseReg = FrameReg;
   int64_t BaseRegOffsetBytes = FrameRegOffset.getFixed();
   if (BaseRegOffsetBytes < kMinOffset ||
-      BaseRegOffsetBytes + (Size - Size % 32) > kMaxOffset) {
+      BaseRegOffsetBytes + (Size - Size % 32) > kMaxOffset ||
+      // BaseReg can be FP, which is not necessarily aligned to 16-bytes. In
+      // that case, BaseRegOffsetBytes will not be aligned to 16 bytes, which
+      // is required for the offset of ST2G.
+      BaseRegOffsetBytes % 16 != 0) {
     Register ScratchReg = MRI->createVirtualRegister(&AArch64::GPR64RegClass);
     emitFrameOffset(*MBB, InsertI, DL, ScratchReg, BaseReg,
                     StackOffset::getFixed(BaseRegOffsetBytes), TII);
@@ -3444,6 +3448,7 @@
         InstrSize == 16
             ? (ZeroData ? AArch64::STZGOffset : AArch64::STGOffset)
             : (ZeroData ? AArch64::STZ2GOffset : AArch64::ST2GOffset);
+    assert(BaseRegOffsetBytes % 16 == 0);
     MachineInstr *I = BuildMI(*MBB, InsertI, DL, TII->get(Opcode))
                           .addReg(AArch64::SP)
                           .addReg(BaseReg)
diff --git a/llvm/test/CodeGen/AArch64/settag-merge-nonaligned-fp.ll b/llvm/test/CodeGen/AArch64/settag-merge-nonaligned-fp.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/settag-merge-nonaligned-fp.ll
@@ -0,0 +1,55 @@
+; RUN: llc < %s -aarch64-order-frame-objects=0 | FileCheck %s
+; Regression test for bug that occured with FP that was not 16-byte aligned.
+; We would miscalculate the offset for the st2g.
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-unknown-linux-android10000"
+
+%class.testcls = type <{ %class.testsubcls.1, %class.testsubcls.6, %class.testsubcls, %class.testsubcls.7, %class.testsubcls.8, %class.testsubcls.9, %struct.teststruct, float, float, %union.anon, [4 x i8] }>
+%class.testsubcls.1 = type { ptr }
+%class.testsubcls.6 = type { ptr }
+%class.testsubcls = type { ptr }
+%class.testsubcls.7 = type { ptr }
+%class.testsubcls.8 = type { ptr }
+%class.testsubcls.9 = type { ptr }
+%struct.teststruct = type { float, float, float, float }
+%union.anon = type { %struct.anon }
+%struct.anon = type { i32 }
+%class.testcls2 = type { %class.testcls3, %class.testsubcls, %class.testsubcls.1, %class.testsubcls.2 }
+%class.testcls3 = type <{ ptr, [124 x i8], [4 x i8] }>
+%class.testsubcls.2 = type { ptr }
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #0
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #0
+
+; Function Attrs: sanitize_memtag sspstrong
+define void @test(ptr %agg.result, float %call) #1 personality ptr null {
+entry:
+  %pathPaint = alloca %class.testcls, align 8
+  %ref.tmp = alloca %class.testcls2, align 8
+  call void @test1(ptr %pathPaint)
+  call void @llvm.lifetime.start.p0(i64 1, ptr %ref.tmp)
+  store i32 0, ptr %ref.tmp, align 4
+  store float %call, ptr %agg.result, align 8
+  call void @llvm.lifetime.end.p0(i64 0, ptr %ref.tmp)
+  %0 = atomicrmw add ptr null, i32 0 monotonic, align 4
+  ret void
+}
+
+; CHECK-LABEL: test
+; CHECK: sub	x8, x29, #104
+; CHECK: st2g	sp, [x8, #32]
+; CHECK: stg	sp, [x8, #64]
+; CHECK: st2g	sp, [x8]
+
+
+declare void @test1(ptr)
+
+; uselistorder directives
+uselistorder ptr null, { 1, 2, 3, 0 }
+
+attributes #0 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
+attributes #1 = { sanitize_memtag sspstrong "frame-pointer"="non-leaf" "target-features"="+crc,+crypto,+dotprod,+fix-cortex-a53-835769,+fp-armv8,+fullfp16,+lse,+mte,+neon,+outline-atomics,+ras,+rcpc,+rdm,+v8a" }