Index: lib/Target/AArch64/AArch64FrameLowering.cpp
===================================================================
--- lib/Target/AArch64/AArch64FrameLowering.cpp
+++ lib/Target/AArch64/AArch64FrameLowering.cpp
@@ -1163,7 +1163,7 @@
   // register scavenging. If we already spilled an extra callee-saved register
   // above to keep the number of spills even, we don't need to do anything else
   // here.
-  if (BigStack && !ExtraCSSpill) {
+  if ((BigStack || AFI->hasNonRISpills()) && !ExtraCSSpill) {
     if (UnspilledCSGPR != AArch64::NoRegister) {
       DEBUG(dbgs() << "Spilling " << PrintReg(UnspilledCSGPR, RegInfo)
             << " to get a scratch register.\n");
Index: lib/Target/AArch64/AArch64InstrInfo.cpp
===================================================================
--- lib/Target/AArch64/AArch64InstrInfo.cpp
+++ lib/Target/AArch64/AArch64InstrInfo.cpp
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "AArch64InstrInfo.h"
+#include "AArch64MachineFunctionInfo.h"
 #include "AArch64Subtarget.h"
 #include "MCTargetDesc/AArch64AddressingModes.h"
 #include "Utils/AArch64BaseInfo.h"
@@ -2424,6 +2425,8 @@
 
   if (Offset)
     MI.addImm(0);
+  else
+    MF.getInfo<AArch64FunctionInfo>()->setHasNonRISpills();
   MI.addMemOperand(MMO);
 }
 
@@ -2527,6 +2530,8 @@
                                       .addFrameIndex(FI);
   if (Offset)
     MI.addImm(0);
+  else
+    MF.getInfo<AArch64FunctionInfo>()->setHasNonRISpills();
   MI.addMemOperand(MMO);
 }
 
Index: lib/Target/AArch64/AArch64MachineFunctionInfo.h
===================================================================
--- lib/Target/AArch64/AArch64MachineFunctionInfo.h
+++ lib/Target/AArch64/AArch64MachineFunctionInfo.h
@@ -88,6 +88,10 @@
   /// other stack allocations.
   bool CalleeSaveStackHasFreeSpace = false;
 
+  /// True if the function has spills using instructions with no immediate
+  /// offset field.
+  bool HasNonRISpills = false;
+
 public:
   AArch64FunctionInfo() = default;
 
@@ -179,6 +183,9 @@
     LOHRelated.insert(Args.begin(), Args.end());
   }
 
+  void setHasNonRISpills()    { HasNonRISpills = true; }
+  bool hasNonRISpills() const { return HasNonRISpills; }
+
 private:
   // Hold the lists of LOHs.
   MILOHContainer LOHContainerSet;
Index: test/CodeGen/AArch64/emergency_spill_slot_nonri.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AArch64/emergency_spill_slot_nonri.ll
@@ -0,0 +1,202 @@
+; RUN: llc -mtriple=aarch64-none-linux-gnu -misched-cutoff=0 < %s
+
+; Make sure that the emergency spill slot is created when we spill using an
+; instruction that doesn't have base+imm form.  If we cannot scavenege a spare
+; register when fixing up the frame index, then we need the emergency spill slot
+; to which to spill a register.
+; Test case taken from arm64-spill-lr.ll and modified a little to require a spill
+; using st1, where the estimated stack size doesn't exceed 256 bytes.
+
+@bar = common global i32 0, align 4
+@bar2 = common global i32 0, align 4
+
+define i32 @foo(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e, i32 %f, i32 %g, i32 %h) nounwind {
+entry:
+  %stack = alloca [1 x i32], align 4
+  %0 = bitcast [1 x i32]* %stack to i8*
+  %idxprom = sext i32 %a to i64
+  %arrayidx = getelementptr inbounds [1 x i32], [1 x i32]* %stack, i64 0, i64 %idxprom
+  store i32 %b, i32* %arrayidx, align 4
+  %1 = load volatile i32, i32* @bar, align 4
+  %2 = load volatile i32, i32* @bar, align 4
+  %3 = load volatile i32, i32* @bar, align 4
+  %4 = load volatile i32, i32* @bar, align 4
+  %5 = load volatile i32, i32* @bar, align 4
+  %6 = load volatile i32, i32* @bar, align 4
+  %7 = load volatile i32, i32* @bar, align 4
+  %8 = load volatile i32, i32* @bar, align 4
+  %9 = load volatile i32, i32* @bar, align 4
+  %10 = load volatile i32, i32* @bar, align 4
+  %idx0 = bitcast i32* @bar to <4 x i32>*
+  %vld0 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx0)
+  %idx1temp = getelementptr inbounds i32, i32* @bar, i64 1
+  %idx1 = bitcast i32* %idx1temp to <4 x i32>*
+  %vld1 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx1)
+  %idx2temp = getelementptr inbounds i32, i32* @bar, i64 2
+  %idx2 = bitcast i32* %idx2temp to <4 x i32>*
+  %vld2 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx2)
+  %idx3temp = getelementptr inbounds i32, i32* @bar, i64 3
+  %idx3 = bitcast i32* %idx3temp to <4 x i32>*
+  %vld3 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx3)
+  %idx4temp = getelementptr inbounds i32, i32* @bar, i64 4
+  %idx4 = bitcast i32* %idx4temp to <4 x i32>*
+  %vld4 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx4)
+  %idx5temp = getelementptr inbounds i32, i32* @bar, i64 5
+  %idx5 = bitcast i32* %idx5temp to <4 x i32>*
+  %vld5 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx5)
+  %idx6temp = getelementptr inbounds i32, i32* @bar, i64 6
+  %idx6 = bitcast i32* %idx6temp to <4 x i32>*
+  %vld6 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx6)
+  %idx7temp = getelementptr inbounds i32, i32* @bar, i64 7
+  %idx7 = bitcast i32* %idx7temp to <4 x i32>*
+  %vld7 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx7)
+  %idx8temp = getelementptr inbounds i32, i32* @bar, i64 8
+  %idx8 = bitcast i32* %idx8temp to <4 x i32>*
+  %vld8 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx8)
+  %idx9temp = getelementptr inbounds i32, i32* @bar, i64 9
+  %idx9 = bitcast i32* %idx9temp to <4 x i32>*
+  %vld9 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx9)
+  %idx10temp = getelementptr inbounds i32, i32* @bar, i64 10
+  %idx10 = bitcast i32* %idx10temp to <4 x i32>*
+  %vld10 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx10)
+  %idx11temp = getelementptr inbounds i32, i32* @bar, i64 11
+  %idx11 = bitcast i32* %idx11temp to <4 x i32>*
+  %vld11 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx11)
+  %idx12temp = getelementptr inbounds i32, i32* @bar, i64 12
+  %idx12 = bitcast i32* %idx12temp to <4 x i32>*
+  %vld12 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx12)
+  %idx13temp = getelementptr inbounds i32, i32* @bar, i64 13
+  %idx13 = bitcast i32* %idx13temp to <4 x i32>*
+  %vld13 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx13)
+  %idx14temp = getelementptr inbounds i32, i32* @bar, i64 14
+  %idx14 = bitcast i32* %idx14temp to <4 x i32>*
+  %vld14 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx14)
+  %idx15temp = getelementptr inbounds i32, i32* @bar, i64 15
+  %idx15 = bitcast i32* %idx15temp to <4 x i32>*
+  %vld15 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx15)
+  %idx16temp = getelementptr inbounds i32, i32* @bar, i64 16
+  %idx16 = bitcast i32* %idx16temp to <4 x i32>*
+  %vld16 = tail call { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>* %idx16)
+  %vld0.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld0, 0
+  %vld0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld0, 1
+  %sidx0 = bitcast i32* @bar2 to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld0.fca.0.extract, <4 x i32> %vld0.extract, i64 1, i8* %sidx0)
+  %vld1.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld1, 0
+  %vld1.extract = extractvalue { <4 x i32>, <4 x i32> } %vld1, 1
+  %sidx1temp = getelementptr inbounds i32, i32* @bar2, i64 1
+  %sidx1 = bitcast i32* %sidx1temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld1.fca.0.extract, <4 x i32> %vld1.extract, i64 1, i8* %sidx1)
+  %vld2.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld2, 0
+  %vld2.extract = extractvalue { <4 x i32>, <4 x i32> } %vld2, 1
+  %sidx2temp = getelementptr inbounds i32, i32* @bar2, i64 2
+  %sidx2 = bitcast i32* %sidx2temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld2.fca.0.extract, <4 x i32> %vld2.extract, i64 1, i8* %sidx2)
+  %vld3.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld3, 0
+  %vld3.extract = extractvalue { <4 x i32>, <4 x i32> } %vld3, 1
+  %sidx3temp = getelementptr inbounds i32, i32* @bar2, i64 3
+  %sidx3 = bitcast i32* %sidx3temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld3.fca.0.extract, <4 x i32> %vld3.extract, i64 1, i8* %sidx3)
+  %vld4.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld4, 0
+  %vld4.extract = extractvalue { <4 x i32>, <4 x i32> } %vld4, 1
+  %sidx4temp = getelementptr inbounds i32, i32* @bar2, i64 4
+  %sidx4 = bitcast i32* %sidx4temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld4.fca.0.extract, <4 x i32> %vld4.extract, i64 1, i8* %sidx4)
+  %vld5.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld5, 0
+  %vld5.extract = extractvalue { <4 x i32>, <4 x i32> } %vld5, 1
+  %sidx5temp = getelementptr inbounds i32, i32* @bar2, i64 5
+  %sidx5 = bitcast i32* %sidx5temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld5.fca.0.extract, <4 x i32> %vld5.extract, i64 1, i8* %sidx5)
+  %vld6.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld6, 0
+  %vld6.extract = extractvalue { <4 x i32>, <4 x i32> } %vld6, 1
+  %sidx6temp = getelementptr inbounds i32, i32* @bar2, i64 6
+  %sidx6 = bitcast i32* %sidx6temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld6.fca.0.extract, <4 x i32> %vld6.extract, i64 1, i8* %sidx6)
+  %vld7.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld7, 0
+  %vld7.extract = extractvalue { <4 x i32>, <4 x i32> } %vld7, 1
+  %sidx7temp = getelementptr inbounds i32, i32* @bar2, i64 7
+  %sidx7 = bitcast i32* %sidx7temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld7.fca.0.extract, <4 x i32> %vld7.extract, i64 1, i8* %sidx7)
+  %vld8.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld8, 0
+  %vld8.extract = extractvalue { <4 x i32>, <4 x i32> } %vld8, 1
+  %sidx8temp = getelementptr inbounds i32, i32* @bar2, i64 8
+  %sidx8 = bitcast i32* %sidx8temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld8.fca.0.extract, <4 x i32> %vld8.extract, i64 1, i8* %sidx8)
+  %vld9.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld9, 0
+  %vld9.extract = extractvalue { <4 x i32>, <4 x i32> } %vld9, 1
+  %sidx9temp = getelementptr inbounds i32, i32* @bar2, i64 9
+  %sidx9 = bitcast i32* %sidx9temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld9.fca.0.extract, <4 x i32> %vld9.extract, i64 1, i8* %sidx9)
+  %vld10.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld10, 0
+  %vld10.extract = extractvalue { <4 x i32>, <4 x i32> } %vld10, 1
+  %sidx10temp = getelementptr inbounds i32, i32* @bar2, i64 10
+  %sidx10 = bitcast i32* %sidx10temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld10.fca.0.extract, <4 x i32> %vld10.extract, i64 1, i8* %sidx10)
+  %vld11.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld11, 0
+  %vld11.extract = extractvalue { <4 x i32>, <4 x i32> } %vld11, 1
+  %sidx11temp = getelementptr inbounds i32, i32* @bar2, i64 11
+  %sidx11 = bitcast i32* %sidx11temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld11.fca.0.extract, <4 x i32> %vld11.extract, i64 1, i8* %sidx11)
+  %vld12.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld12, 0
+  %vld12.extract = extractvalue { <4 x i32>, <4 x i32> } %vld12, 1
+  %sidx12temp = getelementptr inbounds i32, i32* @bar2, i64 12
+  %sidx12 = bitcast i32* %sidx12temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld12.fca.0.extract, <4 x i32> %vld12.extract, i64 1, i8* %sidx12)
+  %vld13.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld13, 0
+  %vld13.extract = extractvalue { <4 x i32>, <4 x i32> } %vld13, 1
+  %sidx13temp = getelementptr inbounds i32, i32* @bar2, i64 13
+  %sidx13 = bitcast i32* %sidx13temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld13.fca.0.extract, <4 x i32> %vld13.extract, i64 1, i8* %sidx13)
+  %vld14.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld14, 0
+  %vld14.extract = extractvalue { <4 x i32>, <4 x i32> } %vld14, 1
+  %sidx14temp = getelementptr inbounds i32, i32* @bar2, i64 14
+  %sidx14 = bitcast i32* %sidx14temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld14.fca.0.extract, <4 x i32> %vld14.extract, i64 1, i8* %sidx14)
+  %vld15.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld15, 0
+  %vld15.extract = extractvalue { <4 x i32>, <4 x i32> } %vld15, 1
+  %sidx15temp = getelementptr inbounds i32, i32* @bar2, i64 15
+  %sidx15 = bitcast i32* %sidx15temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld15.fca.0.extract, <4 x i32> %vld15.extract, i64 1, i8* %sidx15)
+  %vld16.fca.0.extract = extractvalue { <4 x i32>, <4 x i32> } %vld16, 0
+  %vld16.extract = extractvalue { <4 x i32>, <4 x i32> } %vld16, 1
+  %sidx16temp = getelementptr inbounds i32, i32* @bar2, i64 16
+  %sidx16 = bitcast i32* %sidx16temp to i8*
+  tail call void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32> %vld16.fca.0.extract, <4 x i32> %vld16.extract, i64 1, i8* %sidx16)
+  %idxprom1 = sext i32 %c to i64
+  %arrayidx2 = getelementptr inbounds [1 x i32], [1 x i32]* %stack, i64 0, i64 %idxprom1
+  %11 = load i32, i32* %arrayidx2, align 4
+  %factor = mul i32 %h, -2
+  %factor67 = mul i32 %g, -2
+  %factor68 = mul i32 %f, -2
+  %factor69 = mul i32 %e, -2
+  %factor70 = mul i32 %d, -2
+  %factor71 = mul i32 %c, -2
+  %factor72 = mul i32 %b, -2
+  %sum = add i32 %2, %1
+  %sum73 = add i32 %sum, %3
+  %sum74 = add i32 %sum73, %4
+  %sum75 = add i32 %sum74, %5
+  %sum76 = add i32 %sum75, %6
+  %sum77 = add i32 %sum76, %7
+  %sum78 = add i32 %sum77, %8
+  %sum79 = add i32 %sum78, %9
+  %sum80 = add i32 %sum79, %10
+  %sub15 = sub i32 %11, %sum80
+  %sub16 = add i32 %sub15, %factor
+  %sub17 = add i32 %sub16, %factor67
+  %sub18 = add i32 %sub17, %factor68
+  %sub19 = add i32 %sub18, %factor69
+  %sub20 = add i32 %sub19, %factor70
+  %sub21 = add i32 %sub20, %factor71
+  %add = add i32 %sub21, %factor72
+  ret i32 %add
+}
+
+; Function Attrs: argmemonly nounwind readonly
+declare { <4 x i32>, <4 x i32> } @llvm.aarch64.neon.ld2.v4i32.p0v4i32(<4 x i32>*) #1
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.aarch64.neon.st2lane.v4i32.p0i8(<4 x i32>, <4 x i32>, i64, i8* nocapture) #2
+
+attributes #0 = { nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+neon" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { argmemonly nounwind readonly }
+attributes #2 = { argmemonly nounwind }