diff --git a/llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp b/llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp
--- a/llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp
+++ b/llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp
@@ -1631,9 +1631,26 @@
   const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
   const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
   const TargetRegisterClass *RC = is64Bit ? G8RC : GPRC;
-  Register SRegHi = MF.getRegInfo().createVirtualRegister(RC),
-           SReg = MF.getRegInfo().createVirtualRegister(RC);
   unsigned NewOpcode = 0u;
+  bool ScavengingFailed = RS && RS->getRegsAvailable(RC).none() &&
+                          RS->getRegsAvailable(&PPC::VSFRCRegClass).any();
+  Register SRegHi, SReg, VSReg;
+
+  // The register scavenger is unable to get a GPR but can get a VSR. We
+  // need to stash a GPR into a VSR so that we can free one up.
+  if (ScavengingFailed && Subtarget.hasDirectMove()) {
+    // Pick a volatile register and if we are spilling/restoring that
+    // particular one, pick the next one.
+    SRegHi = SReg = is64Bit ? PPC::X4 : PPC::R4;
+    if (MI.getOperand(0).getReg() == SReg)
+      SRegHi = SReg = SReg + 1;
+    VSReg = MF.getRegInfo().createVirtualRegister(&PPC::VSFRCRegClass);
+    BuildMI(MBB, II, dl, TII.get(is64Bit ? PPC::MTVSRD : PPC::MTVSRWZ), VSReg)
+        .addReg(SReg);
+  } else {
+    SRegHi = MF.getRegInfo().createVirtualRegister(RC);
+    SReg = MF.getRegInfo().createVirtualRegister(RC);
+  }
 
   // Insert a set of rA with the full offset value before the ld, st, or add
   if (isInt<16>(Offset))
@@ -1673,6 +1690,12 @@
   MI.getOperand(OperandBase).ChangeToRegister(StackReg, false);
   MI.getOperand(OperandBase + 1).ChangeToRegister(SReg, false, false, true);
 
+  // If we stashed a value from a GPR into a VSR, we need to get it back after
+  // spilling the register.
+  if (ScavengingFailed && Subtarget.hasDirectMove())
+    BuildMI(MBB, ++II, dl, TII.get(is64Bit ? PPC::MFVSRD : PPC::MFVSRWZ), SReg)
+        .addReg(VSReg);
+
   // Since these are not real X-Form instructions, we must
   // add the registers and access 0(NewReg) rather than
   // emitting the X-Form pseudo.
diff --git a/llvm/test/CodeGen/PowerPC/pr52894-32bit.ll b/llvm/test/CodeGen/PowerPC/pr52894-32bit.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/pr52894-32bit.ll
@@ -0,0 +1,290 @@
+; RUN: llc -verify-machineinstrs -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \
+; RUN:   -mcpu=pwr8 -mtriple=powerpcle < %s | FileCheck %s
+; RUN: llc -verify-machineinstrs -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \
+; RUN:   -mcpu=pwr8 -mtriple=powerpc < %s | FileCheck %s
+%struct.d = type { [131072 x i32] }
+
+@a = dso_local local_unnamed_addr global [4096 x i32] zeroinitializer, align 4
+
+; Function Attrs: mustprogress uwtable
+define dso_local void @_Z1g1dILi17EE(%struct.d* nocapture noundef readnone byval(%struct.d) align 4 %0) local_unnamed_addr #0 {
+; CHECK-LABEL: _Z1g1dILi17EE:
+; CHECK:    mtfprwz f0, r4
+; CHECK:    stwx [[REG:r[0-9]+]], r1, r4
+; CHECK:    mffprwz r4, f0
+; CHECK:    mtfprwz f0, r4
+; CHECK:    lwzx [[REG]], r1, r4
+; CHECK:    mffprwz r4, f0
+entry:
+  %c = alloca %struct.d, align 4
+  %1 = bitcast %struct.d* %c to i8*
+  call void @llvm.lifetime.start.p0i8(i64 524288, i8* nonnull %1) #3
+  br label %vector.body
+
+vector.body:                                      ; preds = %vector.body.1, %entry
+  %index = phi i32 [ 0, %entry ], [ %index.next.1, %vector.body.1 ]
+  %vec.ind = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %entry ], [ %vec.ind.next.1, %vector.body.1 ]
+  %step.add = add <4 x i32> %vec.ind, <i32 4, i32 4, i32 4, i32 4>
+  %step.add21 = add <4 x i32> %vec.ind, <i32 8, i32 8, i32 8, i32 8>
+  %step.add22 = add <4 x i32> %vec.ind, <i32 12, i32 12, i32 12, i32 12>
+  %step.add23 = add <4 x i32> %vec.ind, <i32 16, i32 16, i32 16, i32 16>
+  %step.add24 = add <4 x i32> %vec.ind, <i32 20, i32 20, i32 20, i32 20>
+  %step.add25 = add <4 x i32> %vec.ind, <i32 24, i32 24, i32 24, i32 24>
+  %step.add26 = add <4 x i32> %vec.ind, <i32 28, i32 28, i32 28, i32 28>
+  %step.add27 = add <4 x i32> %vec.ind, <i32 32, i32 32, i32 32, i32 32>
+  %step.add28 = add <4 x i32> %vec.ind, <i32 36, i32 36, i32 36, i32 36>
+  %step.add29 = add <4 x i32> %vec.ind, <i32 40, i32 40, i32 40, i32 40>
+  %step.add30 = add <4 x i32> %vec.ind, <i32 44, i32 44, i32 44, i32 44>
+  %2 = getelementptr inbounds [4096 x i32], [4096 x i32]* @a, i32 0, i32 %index
+  %3 = bitcast i32* %2 to <4 x i32>*
+  store <4 x i32> %vec.ind, <4 x i32>* %3, align 4
+  %4 = getelementptr inbounds i32, i32* %2, i32 4
+  %5 = bitcast i32* %4 to <4 x i32>*
+  store <4 x i32> %step.add, <4 x i32>* %5, align 4
+  %6 = getelementptr inbounds i32, i32* %2, i32 8
+  %7 = bitcast i32* %6 to <4 x i32>*
+  store <4 x i32> %step.add21, <4 x i32>* %7, align 4
+  %8 = getelementptr inbounds i32, i32* %2, i32 12
+  %9 = bitcast i32* %8 to <4 x i32>*
+  store <4 x i32> %step.add22, <4 x i32>* %9, align 4
+  %10 = getelementptr inbounds i32, i32* %2, i32 16
+  %11 = bitcast i32* %10 to <4 x i32>*
+  store <4 x i32> %step.add23, <4 x i32>* %11, align 4
+  %12 = getelementptr inbounds i32, i32* %2, i32 20
+  %13 = bitcast i32* %12 to <4 x i32>*
+  store <4 x i32> %step.add24, <4 x i32>* %13, align 4
+  %14 = getelementptr inbounds i32, i32* %2, i32 24
+  %15 = bitcast i32* %14 to <4 x i32>*
+  store <4 x i32> %step.add25, <4 x i32>* %15, align 4
+  %16 = getelementptr inbounds i32, i32* %2, i32 28
+  %17 = bitcast i32* %16 to <4 x i32>*
+  store <4 x i32> %step.add26, <4 x i32>* %17, align 4
+  %18 = getelementptr inbounds i32, i32* %2, i32 32
+  %19 = bitcast i32* %18 to <4 x i32>*
+  store <4 x i32> %step.add27, <4 x i32>* %19, align 4
+  %20 = getelementptr inbounds i32, i32* %2, i32 36
+  %21 = bitcast i32* %20 to <4 x i32>*
+  store <4 x i32> %step.add28, <4 x i32>* %21, align 4
+  %22 = getelementptr inbounds i32, i32* %2, i32 40
+  %23 = bitcast i32* %22 to <4 x i32>*
+  store <4 x i32> %step.add29, <4 x i32>* %23, align 4
+  %24 = getelementptr inbounds i32, i32* %2, i32 44
+  %25 = bitcast i32* %24 to <4 x i32>*
+  store <4 x i32> %step.add30, <4 x i32>* %25, align 4
+  %index.next = add nuw nsw i32 %index, 48
+  %26 = icmp eq i32 %index.next, 4080
+  br i1 %26, label %for.body, label %vector.body.1
+
+vector.body.1:                                    ; preds = %vector.body
+  %vec.ind.next = add <4 x i32> %vec.ind, <i32 48, i32 48, i32 48, i32 48>
+  %step.add.1 = add <4 x i32> %vec.ind, <i32 52, i32 52, i32 52, i32 52>
+  %step.add21.1 = add <4 x i32> %vec.ind, <i32 56, i32 56, i32 56, i32 56>
+  %step.add22.1 = add <4 x i32> %vec.ind, <i32 60, i32 60, i32 60, i32 60>
+  %step.add23.1 = add <4 x i32> %vec.ind, <i32 64, i32 64, i32 64, i32 64>
+  %step.add24.1 = add <4 x i32> %vec.ind, <i32 68, i32 68, i32 68, i32 68>
+  %step.add25.1 = add <4 x i32> %vec.ind, <i32 72, i32 72, i32 72, i32 72>
+  %step.add26.1 = add <4 x i32> %vec.ind, <i32 76, i32 76, i32 76, i32 76>
+  %step.add27.1 = add <4 x i32> %vec.ind, <i32 80, i32 80, i32 80, i32 80>
+  %step.add28.1 = add <4 x i32> %vec.ind, <i32 84, i32 84, i32 84, i32 84>
+  %step.add29.1 = add <4 x i32> %vec.ind, <i32 88, i32 88, i32 88, i32 88>
+  %step.add30.1 = add <4 x i32> %vec.ind, <i32 92, i32 92, i32 92, i32 92>
+  %27 = getelementptr inbounds [4096 x i32], [4096 x i32]* @a, i32 0, i32 %index.next
+  %28 = bitcast i32* %27 to <4 x i32>*
+  store <4 x i32> %vec.ind.next, <4 x i32>* %28, align 4
+  %29 = getelementptr inbounds i32, i32* %27, i32 4
+  %30 = bitcast i32* %29 to <4 x i32>*
+  store <4 x i32> %step.add.1, <4 x i32>* %30, align 4
+  %31 = getelementptr inbounds i32, i32* %27, i32 8
+  %32 = bitcast i32* %31 to <4 x i32>*
+  store <4 x i32> %step.add21.1, <4 x i32>* %32, align 4
+  %33 = getelementptr inbounds i32, i32* %27, i32 12
+  %34 = bitcast i32* %33 to <4 x i32>*
+  store <4 x i32> %step.add22.1, <4 x i32>* %34, align 4
+  %35 = getelementptr inbounds i32, i32* %27, i32 16
+  %36 = bitcast i32* %35 to <4 x i32>*
+  store <4 x i32> %step.add23.1, <4 x i32>* %36, align 4
+  %37 = getelementptr inbounds i32, i32* %27, i32 20
+  %38 = bitcast i32* %37 to <4 x i32>*
+  store <4 x i32> %step.add24.1, <4 x i32>* %38, align 4
+  %39 = getelementptr inbounds i32, i32* %27, i32 24
+  %40 = bitcast i32* %39 to <4 x i32>*
+  store <4 x i32> %step.add25.1, <4 x i32>* %40, align 4
+  %41 = getelementptr inbounds i32, i32* %27, i32 28
+  %42 = bitcast i32* %41 to <4 x i32>*
+  store <4 x i32> %step.add26.1, <4 x i32>* %42, align 4
+  %43 = getelementptr inbounds i32, i32* %27, i32 32
+  %44 = bitcast i32* %43 to <4 x i32>*
+  store <4 x i32> %step.add27.1, <4 x i32>* %44, align 4
+  %45 = getelementptr inbounds i32, i32* %27, i32 36
+  %46 = bitcast i32* %45 to <4 x i32>*
+  store <4 x i32> %step.add28.1, <4 x i32>* %46, align 4
+  %47 = getelementptr inbounds i32, i32* %27, i32 40
+  %48 = bitcast i32* %47 to <4 x i32>*
+  store <4 x i32> %step.add29.1, <4 x i32>* %48, align 4
+  %49 = getelementptr inbounds i32, i32* %27, i32 44
+  %50 = bitcast i32* %49 to <4 x i32>*
+  store <4 x i32> %step.add30.1, <4 x i32>* %50, align 4
+  %index.next.1 = add nuw nsw i32 %index, 96
+  %vec.ind.next.1 = add <4 x i32> %vec.ind, <i32 96, i32 96, i32 96, i32 96>
+  br label %vector.body
+
+vector.body37:                                    ; preds = %vector.body37.1, %for.body
+  %index38 = phi i32 [ 0, %for.body ], [ %index.next53.1, %vector.body37.1 ]
+  %vec.ind39 = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %for.body ], [ %vec.ind.next52.1, %vector.body37.1 ]
+  %step.add40 = add <4 x i32> %vec.ind39, <i32 4, i32 4, i32 4, i32 4>
+  %step.add41 = add <4 x i32> %vec.ind39, <i32 8, i32 8, i32 8, i32 8>
+  %step.add42 = add <4 x i32> %vec.ind39, <i32 12, i32 12, i32 12, i32 12>
+  %step.add43 = add <4 x i32> %vec.ind39, <i32 16, i32 16, i32 16, i32 16>
+  %step.add44 = add <4 x i32> %vec.ind39, <i32 20, i32 20, i32 20, i32 20>
+  %step.add45 = add <4 x i32> %vec.ind39, <i32 24, i32 24, i32 24, i32 24>
+  %step.add46 = add <4 x i32> %vec.ind39, <i32 28, i32 28, i32 28, i32 28>
+  %step.add47 = add <4 x i32> %vec.ind39, <i32 32, i32 32, i32 32, i32 32>
+  %step.add48 = add <4 x i32> %vec.ind39, <i32 36, i32 36, i32 36, i32 36>
+  %step.add49 = add <4 x i32> %vec.ind39, <i32 40, i32 40, i32 40, i32 40>
+  %step.add50 = add <4 x i32> %vec.ind39, <i32 44, i32 44, i32 44, i32 44>
+  %51 = getelementptr inbounds [4096 x i32], [4096 x i32]* @a, i32 0, i32 %index38
+  %52 = bitcast i32* %51 to <4 x i32>*
+  store <4 x i32> %vec.ind39, <4 x i32>* %52, align 4
+  %53 = getelementptr inbounds i32, i32* %51, i32 4
+  %54 = bitcast i32* %53 to <4 x i32>*
+  store <4 x i32> %step.add40, <4 x i32>* %54, align 4
+  %55 = getelementptr inbounds i32, i32* %51, i32 8
+  %56 = bitcast i32* %55 to <4 x i32>*
+  store <4 x i32> %step.add41, <4 x i32>* %56, align 4
+  %57 = getelementptr inbounds i32, i32* %51, i32 12
+  %58 = bitcast i32* %57 to <4 x i32>*
+  store <4 x i32> %step.add42, <4 x i32>* %58, align 4
+  %59 = getelementptr inbounds i32, i32* %51, i32 16
+  %60 = bitcast i32* %59 to <4 x i32>*
+  store <4 x i32> %step.add43, <4 x i32>* %60, align 4
+  %61 = getelementptr inbounds i32, i32* %51, i32 20
+  %62 = bitcast i32* %61 to <4 x i32>*
+  store <4 x i32> %step.add44, <4 x i32>* %62, align 4
+  %63 = getelementptr inbounds i32, i32* %51, i32 24
+  %64 = bitcast i32* %63 to <4 x i32>*
+  store <4 x i32> %step.add45, <4 x i32>* %64, align 4
+  %65 = getelementptr inbounds i32, i32* %51, i32 28
+  %66 = bitcast i32* %65 to <4 x i32>*
+  store <4 x i32> %step.add46, <4 x i32>* %66, align 4
+  %67 = getelementptr inbounds i32, i32* %51, i32 32
+  %68 = bitcast i32* %67 to <4 x i32>*
+  store <4 x i32> %step.add47, <4 x i32>* %68, align 4
+  %69 = getelementptr inbounds i32, i32* %51, i32 36
+  %70 = bitcast i32* %69 to <4 x i32>*
+  store <4 x i32> %step.add48, <4 x i32>* %70, align 4
+  %71 = getelementptr inbounds i32, i32* %51, i32 40
+  %72 = bitcast i32* %71 to <4 x i32>*
+  store <4 x i32> %step.add49, <4 x i32>* %72, align 4
+  %73 = getelementptr inbounds i32, i32* %51, i32 44
+  %74 = bitcast i32* %73 to <4 x i32>*
+  store <4 x i32> %step.add50, <4 x i32>* %74, align 4
+  %index.next53 = add nuw nsw i32 %index38, 48
+  %75 = icmp eq i32 %index.next53, 4080
+  br i1 %75, label %for.body5, label %vector.body37.1
+
+vector.body37.1:                                  ; preds = %vector.body37
+  %vec.ind.next52 = add <4 x i32> %vec.ind39, <i32 48, i32 48, i32 48, i32 48>
+  %step.add40.1 = add <4 x i32> %vec.ind39, <i32 52, i32 52, i32 52, i32 52>
+  %step.add41.1 = add <4 x i32> %vec.ind39, <i32 56, i32 56, i32 56, i32 56>
+  %step.add42.1 = add <4 x i32> %vec.ind39, <i32 60, i32 60, i32 60, i32 60>
+  %step.add43.1 = add <4 x i32> %vec.ind39, <i32 64, i32 64, i32 64, i32 64>
+  %step.add44.1 = add <4 x i32> %vec.ind39, <i32 68, i32 68, i32 68, i32 68>
+  %step.add45.1 = add <4 x i32> %vec.ind39, <i32 72, i32 72, i32 72, i32 72>
+  %step.add46.1 = add <4 x i32> %vec.ind39, <i32 76, i32 76, i32 76, i32 76>
+  %step.add47.1 = add <4 x i32> %vec.ind39, <i32 80, i32 80, i32 80, i32 80>
+  %step.add48.1 = add <4 x i32> %vec.ind39, <i32 84, i32 84, i32 84, i32 84>
+  %step.add49.1 = add <4 x i32> %vec.ind39, <i32 88, i32 88, i32 88, i32 88>
+  %step.add50.1 = add <4 x i32> %vec.ind39, <i32 92, i32 92, i32 92, i32 92>
+  %76 = getelementptr inbounds [4096 x i32], [4096 x i32]* @a, i32 0, i32 %index.next53
+  %77 = bitcast i32* %76 to <4 x i32>*
+  store <4 x i32> %vec.ind.next52, <4 x i32>* %77, align 4
+  %78 = getelementptr inbounds i32, i32* %76, i32 4
+  %79 = bitcast i32* %78 to <4 x i32>*
+  store <4 x i32> %step.add40.1, <4 x i32>* %79, align 4
+  %80 = getelementptr inbounds i32, i32* %76, i32 8
+  %81 = bitcast i32* %80 to <4 x i32>*
+  store <4 x i32> %step.add41.1, <4 x i32>* %81, align 4
+  %82 = getelementptr inbounds i32, i32* %76, i32 12
+  %83 = bitcast i32* %82 to <4 x i32>*
+  store <4 x i32> %step.add42.1, <4 x i32>* %83, align 4
+  %84 = getelementptr inbounds i32, i32* %76, i32 16
+  %85 = bitcast i32* %84 to <4 x i32>*
+  store <4 x i32> %step.add43.1, <4 x i32>* %85, align 4
+  %86 = getelementptr inbounds i32, i32* %76, i32 20
+  %87 = bitcast i32* %86 to <4 x i32>*
+  store <4 x i32> %step.add44.1, <4 x i32>* %87, align 4
+  %88 = getelementptr inbounds i32, i32* %76, i32 24
+  %89 = bitcast i32* %88 to <4 x i32>*
+  store <4 x i32> %step.add45.1, <4 x i32>* %89, align 4
+  %90 = getelementptr inbounds i32, i32* %76, i32 28
+  %91 = bitcast i32* %90 to <4 x i32>*
+  store <4 x i32> %step.add46.1, <4 x i32>* %91, align 4
+  %92 = getelementptr inbounds i32, i32* %76, i32 32
+  %93 = bitcast i32* %92 to <4 x i32>*
+  store <4 x i32> %step.add47.1, <4 x i32>* %93, align 4
+  %94 = getelementptr inbounds i32, i32* %76, i32 36
+  %95 = bitcast i32* %94 to <4 x i32>*
+  store <4 x i32> %step.add48.1, <4 x i32>* %95, align 4
+  %96 = getelementptr inbounds i32, i32* %76, i32 40
+  %97 = bitcast i32* %96 to <4 x i32>*
+  store <4 x i32> %step.add49.1, <4 x i32>* %97, align 4
+  %98 = getelementptr inbounds i32, i32* %76, i32 44
+  %99 = bitcast i32* %98 to <4 x i32>*
+  store <4 x i32> %step.add50.1, <4 x i32>* %99, align 4
+  %index.next53.1 = add nuw nsw i32 %index38, 96
+  %vec.ind.next52.1 = add <4 x i32> %vec.ind39, <i32 96, i32 96, i32 96, i32 96>
+  br label %vector.body37
+
+for.body:                                         ; preds = %vector.body
+  store i32 4080, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4080), align 4
+  store i32 4081, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4081), align 4
+  store i32 4082, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4082), align 4
+  store i32 4083, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4083), align 4
+  store i32 4084, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4084), align 4
+  store i32 4085, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4085), align 4
+  store i32 4086, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4086), align 4
+  store i32 4087, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4087), align 4
+  store i32 4088, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4088), align 4
+  store i32 4089, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4089), align 4
+  store i32 4090, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4090), align 4
+  store i32 4091, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4091), align 4
+  store i32 4092, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4092), align 4
+  store i32 4093, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4093), align 4
+  store i32 4094, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4094), align 4
+  store i32 4095, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4095), align 4
+  call void @_ZN1dILi17EE1eEv(%struct.d* noundef nonnull align 4 dereferenceable(524288) %c)
+  br label %vector.body37
+
+for.body5:                                        ; preds = %vector.body37
+  store i32 4080, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4080), align 4
+  store i32 4081, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4081), align 4
+  store i32 4082, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4082), align 4
+  store i32 4083, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4083), align 4
+  store i32 4084, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4084), align 4
+  store i32 4085, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4085), align 4
+  store i32 4086, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4086), align 4
+  store i32 4087, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4087), align 4
+  store i32 4088, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4088), align 4
+  store i32 4089, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4089), align 4
+  store i32 4090, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4090), align 4
+  store i32 4091, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4091), align 4
+  store i32 4092, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4092), align 4
+  store i32 4093, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4093), align 4
+  store i32 4094, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4094), align 4
+  store i32 4095, i32* getelementptr inbounds ([4096 x i32], [4096 x i32]* @a, i32 0, i32 4095), align 4
+  call void @_Z1h1dILi17EE(%struct.d* noundef nonnull byval(%struct.d) align 4 %c)
+  call void @llvm.lifetime.end.p0i8(i64 524288, i8* nonnull %1) #3
+  ret void
+}
+
+; Function Attrs: argmemonly mustprogress nocallback nofree nosync nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+; Function Attrs: argmemonly mustprogress nocallback nofree nosync nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+declare dso_local void @_ZN1dILi17EE1eEv(%struct.d* noundef nonnull align 4 dereferenceable(524288)) local_unnamed_addr #2
+
+declare dso_local void @_Z1h1dILi17EE(%struct.d* noundef byval(%struct.d) align 4) local_unnamed_addr #2
+
+attributes #0 = { nounwind }
diff --git a/llvm/test/CodeGen/PowerPC/pr52894.ll b/llvm/test/CodeGen/PowerPC/pr52894.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/pr52894.ll
@@ -0,0 +1,246 @@
+; RUN: llc -verify-machineinstrs -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \
+; RUN:   -mcpu=pwr8 -mtriple=powerpc64le < %s | FileCheck %s
+; RUN: llc -verify-machineinstrs -ppc-asm-full-reg-names -ppc-vsr-nums-as-vr \
+; RUN:   -mcpu=pwr8 -mtriple=powerpc64 < %s | FileCheck %s
+%struct.d = type { [131072 x i32] }
+
+@a = dso_local local_unnamed_addr global [4096 x i32] zeroinitializer, align 4
+
+; Function Attrs: mustprogress uwtable
+define dso_local void @_Z1g1dILi17EE(ptr nocapture noundef readnone byval(%struct.d) align 8 %0) local_unnamed_addr #0 {
+; CHECK-LABEL: _Z1g1dILi17EE:
+; CHECK:    mtfprd f0, r4
+; CHECK:    stdx [[REG:r[0-9]+]], r1, r4
+; CHECK:    mffprd r4, f0
+; CHECK:    mtfprd f0, r4
+; CHECK:    ldx [[REG]], r1, r4
+; CHECK:    mffprd r4, f0
+; CHECK:    mtfprd f0, r4
+; CHECK:    stdx [[REG2:r[0-9]+]], r1, r4
+; CHECK:    mffprd r4, f0
+; CHECK:    mtfprd f0, r4
+; CHECK:    ldx [[REG2]], r1, r4
+; CHECK:    mffprd r4, f0
+entry:
+  %c = alloca %struct.d, align 8
+  call void @llvm.lifetime.start.p0(i64 524288, ptr nonnull %c) #3
+  br label %vector.body
+
+vector.body:                                      ; preds = %vector.body.1, %entry
+  %index = phi i64 [ 0, %entry ], [ %index.next.1, %vector.body.1 ]
+  %vec.ind = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %entry ], [ %vec.ind.next.1, %vector.body.1 ]
+  %step.add = add <4 x i32> %vec.ind, <i32 4, i32 4, i32 4, i32 4>
+  %step.add24 = add <4 x i32> %vec.ind, <i32 8, i32 8, i32 8, i32 8>
+  %step.add25 = add <4 x i32> %vec.ind, <i32 12, i32 12, i32 12, i32 12>
+  %step.add26 = add <4 x i32> %vec.ind, <i32 16, i32 16, i32 16, i32 16>
+  %step.add27 = add <4 x i32> %vec.ind, <i32 20, i32 20, i32 20, i32 20>
+  %step.add28 = add <4 x i32> %vec.ind, <i32 24, i32 24, i32 24, i32 24>
+  %step.add29 = add <4 x i32> %vec.ind, <i32 28, i32 28, i32 28, i32 28>
+  %step.add30 = add <4 x i32> %vec.ind, <i32 32, i32 32, i32 32, i32 32>
+  %step.add31 = add <4 x i32> %vec.ind, <i32 36, i32 36, i32 36, i32 36>
+  %step.add32 = add <4 x i32> %vec.ind, <i32 40, i32 40, i32 40, i32 40>
+  %step.add33 = add <4 x i32> %vec.ind, <i32 44, i32 44, i32 44, i32 44>
+  %1 = getelementptr inbounds [4096 x i32], ptr @a, i64 0, i64 %index
+  store <4 x i32> %vec.ind, ptr %1, align 4
+  %2 = getelementptr inbounds i32, ptr %1, i64 4
+  store <4 x i32> %step.add, ptr %2, align 4
+  %3 = getelementptr inbounds i32, ptr %1, i64 8
+  store <4 x i32> %step.add24, ptr %3, align 4
+  %4 = getelementptr inbounds i32, ptr %1, i64 12
+  store <4 x i32> %step.add25, ptr %4, align 4
+  %5 = getelementptr inbounds i32, ptr %1, i64 16
+  store <4 x i32> %step.add26, ptr %5, align 4
+  %6 = getelementptr inbounds i32, ptr %1, i64 20
+  store <4 x i32> %step.add27, ptr %6, align 4
+  %7 = getelementptr inbounds i32, ptr %1, i64 24
+  store <4 x i32> %step.add28, ptr %7, align 4
+  %8 = getelementptr inbounds i32, ptr %1, i64 28
+  store <4 x i32> %step.add29, ptr %8, align 4
+  %9 = getelementptr inbounds i32, ptr %1, i64 32
+  store <4 x i32> %step.add30, ptr %9, align 4
+  %10 = getelementptr inbounds i32, ptr %1, i64 36
+  store <4 x i32> %step.add31, ptr %10, align 4
+  %11 = getelementptr inbounds i32, ptr %1, i64 40
+  store <4 x i32> %step.add32, ptr %11, align 4
+  %12 = getelementptr inbounds i32, ptr %1, i64 44
+  store <4 x i32> %step.add33, ptr %12, align 4
+  %index.next = add nuw nsw i64 %index, 48
+  %13 = icmp eq i64 %index.next, 4080
+  br i1 %13, label %for.body, label %vector.body.1
+
+vector.body.1:                                    ; preds = %vector.body
+  %vec.ind.next = add <4 x i32> %vec.ind, <i32 48, i32 48, i32 48, i32 48>
+  %step.add.1 = add <4 x i32> %vec.ind, <i32 52, i32 52, i32 52, i32 52>
+  %step.add24.1 = add <4 x i32> %vec.ind, <i32 56, i32 56, i32 56, i32 56>
+  %step.add25.1 = add <4 x i32> %vec.ind, <i32 60, i32 60, i32 60, i32 60>
+  %step.add26.1 = add <4 x i32> %vec.ind, <i32 64, i32 64, i32 64, i32 64>
+  %step.add27.1 = add <4 x i32> %vec.ind, <i32 68, i32 68, i32 68, i32 68>
+  %step.add28.1 = add <4 x i32> %vec.ind, <i32 72, i32 72, i32 72, i32 72>
+  %step.add29.1 = add <4 x i32> %vec.ind, <i32 76, i32 76, i32 76, i32 76>
+  %step.add30.1 = add <4 x i32> %vec.ind, <i32 80, i32 80, i32 80, i32 80>
+  %step.add31.1 = add <4 x i32> %vec.ind, <i32 84, i32 84, i32 84, i32 84>
+  %step.add32.1 = add <4 x i32> %vec.ind, <i32 88, i32 88, i32 88, i32 88>
+  %step.add33.1 = add <4 x i32> %vec.ind, <i32 92, i32 92, i32 92, i32 92>
+  %14 = getelementptr inbounds [4096 x i32], ptr @a, i64 0, i64 %index.next
+  store <4 x i32> %vec.ind.next, ptr %14, align 4
+  %15 = getelementptr inbounds i32, ptr %14, i64 4
+  store <4 x i32> %step.add.1, ptr %15, align 4
+  %16 = getelementptr inbounds i32, ptr %14, i64 8
+  store <4 x i32> %step.add24.1, ptr %16, align 4
+  %17 = getelementptr inbounds i32, ptr %14, i64 12
+  store <4 x i32> %step.add25.1, ptr %17, align 4
+  %18 = getelementptr inbounds i32, ptr %14, i64 16
+  store <4 x i32> %step.add26.1, ptr %18, align 4
+  %19 = getelementptr inbounds i32, ptr %14, i64 20
+  store <4 x i32> %step.add27.1, ptr %19, align 4
+  %20 = getelementptr inbounds i32, ptr %14, i64 24
+  store <4 x i32> %step.add28.1, ptr %20, align 4
+  %21 = getelementptr inbounds i32, ptr %14, i64 28
+  store <4 x i32> %step.add29.1, ptr %21, align 4
+  %22 = getelementptr inbounds i32, ptr %14, i64 32
+  store <4 x i32> %step.add30.1, ptr %22, align 4
+  %23 = getelementptr inbounds i32, ptr %14, i64 36
+  store <4 x i32> %step.add31.1, ptr %23, align 4
+  %24 = getelementptr inbounds i32, ptr %14, i64 40
+  store <4 x i32> %step.add32.1, ptr %24, align 4
+  %25 = getelementptr inbounds i32, ptr %14, i64 44
+  store <4 x i32> %step.add33.1, ptr %25, align 4
+  %index.next.1 = add nuw nsw i64 %index, 96
+  %vec.ind.next.1 = add <4 x i32> %vec.ind, <i32 96, i32 96, i32 96, i32 96>
+  br label %vector.body
+
+vector.body40:                                    ; preds = %vector.body40.1, %for.body
+  %index41 = phi i64 [ 0, %for.body ], [ %index.next56.1, %vector.body40.1 ]
+  %vec.ind42 = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %for.body ], [ %vec.ind.next55.1, %vector.body40.1 ]
+  %step.add43 = add <4 x i32> %vec.ind42, <i32 4, i32 4, i32 4, i32 4>
+  %step.add44 = add <4 x i32> %vec.ind42, <i32 8, i32 8, i32 8, i32 8>
+  %step.add45 = add <4 x i32> %vec.ind42, <i32 12, i32 12, i32 12, i32 12>
+  %step.add46 = add <4 x i32> %vec.ind42, <i32 16, i32 16, i32 16, i32 16>
+  %step.add47 = add <4 x i32> %vec.ind42, <i32 20, i32 20, i32 20, i32 20>
+  %step.add48 = add <4 x i32> %vec.ind42, <i32 24, i32 24, i32 24, i32 24>
+  %step.add49 = add <4 x i32> %vec.ind42, <i32 28, i32 28, i32 28, i32 28>
+  %step.add50 = add <4 x i32> %vec.ind42, <i32 32, i32 32, i32 32, i32 32>
+  %step.add51 = add <4 x i32> %vec.ind42, <i32 36, i32 36, i32 36, i32 36>
+  %step.add52 = add <4 x i32> %vec.ind42, <i32 40, i32 40, i32 40, i32 40>
+  %step.add53 = add <4 x i32> %vec.ind42, <i32 44, i32 44, i32 44, i32 44>
+  %26 = getelementptr inbounds [4096 x i32], ptr @a, i64 0, i64 %index41
+  store <4 x i32> %vec.ind42, ptr %26, align 4
+  %27 = getelementptr inbounds i32, ptr %26, i64 4
+  store <4 x i32> %step.add43, ptr %27, align 4
+  %28 = getelementptr inbounds i32, ptr %26, i64 8
+  store <4 x i32> %step.add44, ptr %28, align 4
+  %29 = getelementptr inbounds i32, ptr %26, i64 12
+  store <4 x i32> %step.add45, ptr %29, align 4
+  %30 = getelementptr inbounds i32, ptr %26, i64 16
+  store <4 x i32> %step.add46, ptr %30, align 4
+  %31 = getelementptr inbounds i32, ptr %26, i64 20
+  store <4 x i32> %step.add47, ptr %31, align 4
+  %32 = getelementptr inbounds i32, ptr %26, i64 24
+  store <4 x i32> %step.add48, ptr %32, align 4
+  %33 = getelementptr inbounds i32, ptr %26, i64 28
+  store <4 x i32> %step.add49, ptr %33, align 4
+  %34 = getelementptr inbounds i32, ptr %26, i64 32
+  store <4 x i32> %step.add50, ptr %34, align 4
+  %35 = getelementptr inbounds i32, ptr %26, i64 36
+  store <4 x i32> %step.add51, ptr %35, align 4
+  %36 = getelementptr inbounds i32, ptr %26, i64 40
+  store <4 x i32> %step.add52, ptr %36, align 4
+  %37 = getelementptr inbounds i32, ptr %26, i64 44
+  store <4 x i32> %step.add53, ptr %37, align 4
+  %index.next56 = add nuw nsw i64 %index41, 48
+  %38 = icmp eq i64 %index.next56, 4080
+  br i1 %38, label %for.body5, label %vector.body40.1
+
+vector.body40.1:                                  ; preds = %vector.body40
+  %vec.ind.next55 = add <4 x i32> %vec.ind42, <i32 48, i32 48, i32 48, i32 48>
+  %step.add43.1 = add <4 x i32> %vec.ind42, <i32 52, i32 52, i32 52, i32 52>
+  %step.add44.1 = add <4 x i32> %vec.ind42, <i32 56, i32 56, i32 56, i32 56>
+  %step.add45.1 = add <4 x i32> %vec.ind42, <i32 60, i32 60, i32 60, i32 60>
+  %step.add46.1 = add <4 x i32> %vec.ind42, <i32 64, i32 64, i32 64, i32 64>
+  %step.add47.1 = add <4 x i32> %vec.ind42, <i32 68, i32 68, i32 68, i32 68>
+  %step.add48.1 = add <4 x i32> %vec.ind42, <i32 72, i32 72, i32 72, i32 72>
+  %step.add49.1 = add <4 x i32> %vec.ind42, <i32 76, i32 76, i32 76, i32 76>
+  %step.add50.1 = add <4 x i32> %vec.ind42, <i32 80, i32 80, i32 80, i32 80>
+  %step.add51.1 = add <4 x i32> %vec.ind42, <i32 84, i32 84, i32 84, i32 84>
+  %step.add52.1 = add <4 x i32> %vec.ind42, <i32 88, i32 88, i32 88, i32 88>
+  %step.add53.1 = add <4 x i32> %vec.ind42, <i32 92, i32 92, i32 92, i32 92>
+  %39 = getelementptr inbounds [4096 x i32], ptr @a, i64 0, i64 %index.next56
+  store <4 x i32> %vec.ind.next55, ptr %39, align 4
+  %40 = getelementptr inbounds i32, ptr %39, i64 4
+  store <4 x i32> %step.add43.1, ptr %40, align 4
+  %41 = getelementptr inbounds i32, ptr %39, i64 8
+  store <4 x i32> %step.add44.1, ptr %41, align 4
+  %42 = getelementptr inbounds i32, ptr %39, i64 12
+  store <4 x i32> %step.add45.1, ptr %42, align 4
+  %43 = getelementptr inbounds i32, ptr %39, i64 16
+  store <4 x i32> %step.add46.1, ptr %43, align 4
+  %44 = getelementptr inbounds i32, ptr %39, i64 20
+  store <4 x i32> %step.add47.1, ptr %44, align 4
+  %45 = getelementptr inbounds i32, ptr %39, i64 24
+  store <4 x i32> %step.add48.1, ptr %45, align 4
+  %46 = getelementptr inbounds i32, ptr %39, i64 28
+  store <4 x i32> %step.add49.1, ptr %46, align 4
+  %47 = getelementptr inbounds i32, ptr %39, i64 32
+  store <4 x i32> %step.add50.1, ptr %47, align 4
+  %48 = getelementptr inbounds i32, ptr %39, i64 36
+  store <4 x i32> %step.add51.1, ptr %48, align 4
+  %49 = getelementptr inbounds i32, ptr %39, i64 40
+  store <4 x i32> %step.add52.1, ptr %49, align 4
+  %50 = getelementptr inbounds i32, ptr %39, i64 44
+  store <4 x i32> %step.add53.1, ptr %50, align 4
+  %index.next56.1 = add nuw nsw i64 %index41, 96
+  %vec.ind.next55.1 = add <4 x i32> %vec.ind42, <i32 96, i32 96, i32 96, i32 96>
+  br label %vector.body40
+
+for.body:                                         ; preds = %vector.body
+  store i32 4080, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4080), align 4
+  store i32 4081, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4081), align 4
+  store i32 4082, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4082), align 4
+  store i32 4083, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4083), align 4
+  store i32 4084, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4084), align 4
+  store i32 4085, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4085), align 4
+  store i32 4086, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4086), align 4
+  store i32 4087, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4087), align 4
+  store i32 4088, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4088), align 4
+  store i32 4089, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4089), align 4
+  store i32 4090, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4090), align 4
+  store i32 4091, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4091), align 4
+  store i32 4092, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4092), align 4
+  store i32 4093, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4093), align 4
+  store i32 4094, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4094), align 4
+  store i32 4095, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4095), align 4
+  call void @_ZN1dILi17EE1eEv(ptr noundef nonnull align 4 dereferenceable(524288) %c)
+  br label %vector.body40
+
+for.body5:                                        ; preds = %vector.body40
+  store i32 4080, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4080), align 4
+  store i32 4081, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4081), align 4
+  store i32 4082, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4082), align 4
+  store i32 4083, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4083), align 4
+  store i32 4084, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4084), align 4
+  store i32 4085, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4085), align 4
+  store i32 4086, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4086), align 4
+  store i32 4087, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4087), align 4
+  store i32 4088, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4088), align 4
+  store i32 4089, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4089), align 4
+  store i32 4090, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4090), align 4
+  store i32 4091, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4091), align 4
+  store i32 4092, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4092), align 4
+  store i32 4093, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4093), align 4
+  store i32 4094, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4094), align 4
+  store i32 4095, ptr getelementptr inbounds ([4096 x i32], ptr @a, i64 0, i64 4095), align 4
+  call void @_Z1h1dILi17EE(ptr noundef nonnull byval(%struct.d) align 8 %c)
+  call void @llvm.lifetime.end.p0(i64 524288, ptr nonnull %c) #3
+  ret void
+}
+
+; Function Attrs: argmemonly mustprogress nocallback nofree nosync nounwind willreturn
+declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #1
+
+; Function Attrs: argmemonly mustprogress nocallback nofree nosync nounwind willreturn
+declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #1
+
+declare void @_ZN1dILi17EE1eEv(ptr noundef nonnull align 4 dereferenceable(524288)) local_unnamed_addr #2
+
+declare void @_Z1h1dILi17EE(ptr noundef byval(%struct.d) align 8) local_unnamed_addr #2
+attributes #0 = { nounwind }