[ARM] Pass for Cortex-A57 and Cortex-A72 Fused AES Erratum

[lenary]

Summary

This adds a late Machine Pass to work around a Cortex CPU Erratum
affecting Cortex-A57 and Cortex-A72:

• Cortex-A57 Erratum 1742098
• Cortex-A72 Erratum 1655431

The pass inserts instructions to make the inputs to the fused AES
instruction pairs no longer trigger the erratum. Here the pass errs on
the side of caution, inserting the instructions wherever we cannot prove
that the inputs came from a safe instruction.

As the pass is executed very late in the ARM backend pipeline, it has to
reconstruct the Register Dataflow Graph, for which it uses the RDFGraph
utilities used by other backends.

This initial version will stop at the start of basic block containing
the first AES instruction, but having the full RDF Graph available means
we should be able to be more efficient with AES encryption and
decryption loops in future, if we wish.

The pass is used:

• for Cortex-A57 and Cortex-A72,
• for "generic" cores (which are used when using -march=),
• when the user specifies -mfix-cortex-a57-aes-1742098 or mfix-cortex-a72-aes-1655431 in the command-line arguments to clang.

[lenary]

@kparzysz I've tagged you due to the changes in RDFGraph, which I believe you are the owner of. The asserts are hit in llvm/test/CodeGen/ARM/inlineasm-error-t-toofewregs.ll - the Register Allocator Chokes on the test due to not having enough registers for the inline assembly. llc then reports the error, but continues running passes (including this one), which then core dumps on the asserts I've removed (The first one hit is in DataFlowGraph::buildStmt). When I run the same testcase with -verify-machineinstrs, there is no verification error, but the verifier seems to have decided that analysing physical reg liveness is too hard (see e.g. MachineVerifier.cpp line 2990). I haven't worked out how else to prevent this pass running if the register allocation is known to be wrong, without analysing every instruction before building the RDFGraph. I wasn't sure the right approach, so I wanted to post the patch and then discuss it, rather than the other way around (so you can see the testcase).

[lenary]

Ping

[dmgreen]

I have a high level question regarding RDF, as I've not seen it used in many other places, so it may be under-tested on Arm systems at the moment. This currently, for all code, builds an rdf graph, analyze the rdf graph for a fairly rare instructions, then fixes up the code based on that. It might be best to avoid the (possibly expensive?) rdf graph generation for the common case where the instructions are not present. Check that the instruction exists first.

It might then be simpler to just search back for the def of a register, considering in most code the instruction we are looking for should be fairly rare and we won't expect to need to find def's in bulk. That might be simpler overall, and avoid some of the difficulties of RDF.

llvm/test/CodeGen/ARM/aes-erratum-fix.ll

49

Adding arm_aapcs_vfpcc will make the function "hardfp", which might be useful for testing inputs from argument that don't need to be passed via gpr regs.

[lenary]

In D119720#3339855, @dmgreen wrote:

I have a high level question regarding RDF, as I've not seen it used in many other places, so it may be under-tested on Arm systems at the moment. This currently, for all code, builds an rdf graph, analyze the rdf graph for a fairly rare instructions, then fixes up the code based on that. It might be best to avoid the (possibly expensive?) rdf graph generation for the common case where the instructions are not present. Check that the instruction exists first.

I haven't had time to see what effect this has on compile times, but I don't see the point in iterating over every instruction in the function checking the opcode, to then iterate over them all constructing the rdfgraph, that seems to make the approach a lot more expensive, if I'm then going to iterate over the rdfgraph looking for specific instructions again.

It might then be simpler to just search back for the def of a register, considering in most code the instruction we are looking for should be fairly rare and we won't expect to need to find def's in bulk. That might be simpler overall, and avoid some of the difficulties of RDF.

The reason for using the rdfgraph was to allow us to improve this at a later date if we found the performance issues a problem, i.e. by hoisting the vorr past phis so we can save on executing them in loops. Given the issues with rdfgraph, I'm going to reimplement this as a Block-local analysis and fixup pass without the rdfgraph, which is effectively what the pass does in the current patch.

[lenary]

• Rewrite pass in terms of ReachingDefAnalysis
• Split tests into separate commit, for ease of review.

[lenary]

@kparzysz I have rewritten this to avoid using RDFGraph, so I don't think this needs you to review it any more.

llvm/test/CodeGen/ARM/aes-erratum-fix.ll

49

Yeah, seems I was too zealous with removing some of these attributes.

[lenary]

llvm/lib/CodeGen/RDFGraph.cpp

1096 ↗

(On Diff #419202)

I forgot to remove these comments, will update shortly.

llvm/lib/Target/ARM/ARMSubtarget.h

543 ↗

(On Diff #419202)

I missed this, will update again shortly.

[lenary]

• Remove whitespace change in ARMSubtarget
• Remove commented-out debug lines in RDFGraph

[dmgreen]

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

145

Can/should all these use findFirstPredOperandIdx?

And is it worth checking for more instruction? Anything that sets a Q register, or is that too broad?

250

This needn't scan through checking for the instruction that the loop below is going to check for too.

306

Can you explain more about the IsLiveIn && UnsafeCount==0 case. Am I understanding that correctly that it would be:

function(q0, ...)
  lotsofcode...
  q0 = load
  aes q0

Is there a better way to detect that the live-in doesn't matter in cases like that?

llvm/lib/Target/ARM/ARMTargetMachine.cpp

585

Passes can't insert new instructions (or move things further apart) after ConstantIslandPass. The branches/constant pools it has created may go out of range of the instructions that use them. Would it be OK to move it before that?

[lenary]

Thanks for the review. Lots of comments inline, hopefully Phab doesn't mangle the large one.

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

145

findFirstPredOperandIdx doesn't work as lots of these instructions are not marked isPredicable in the tablegen. I'm not sure if we want to solve that in this work, or as a follow-up (I'd lean towards follow-up).

I believe "anything that sets a Q register" is too broad, as we model subregister insertion as setting the the whole register in LLVM, but I'm not sure that micro-architecturally they are actually doing that. This is why I've tried to only add 64- and 128-bit setting instructions rather than ones that are less wide. Originally I also included the VMOVv2f32 instructions that are now at the bottom of this switch, but I felt that might have been too risky.

250

Ack. Will remove. I think this is vestigal from a previous (unshared) version of the patch which was doing something more complex in the loop.

306

I don't believe there is, and this comes down to issues with the RDA.getGlobalReachingDefs implementation, which I want to fix/enhance, but in a follow-up patch.

To start with, this is actually not a problem, as the pass is intended to be conservative, and we know the clobbers are a no-op at the architectural level (we insert them for their micro-architectural effects). So code will still do the right thing, but maybe with a little too much overhead in the case you showed.

However, this is necessary in some other cases, such as:

function(q0)
   code
   conditional branch to L2
L1:
   q0 = safe_op(…)
   branch to L3
L2:
   code without update to q0
L3:
   aes q0

In this case, AllDefs is a set containing one single defining instruction for Q0, because there is only one within the function (which is all that RDA.getGlobalReachingDefs can report instructions for).

But in my example, we *need* to protect q0 on the other paths, because the safe definitions of q0, when considered as a set, do not entirely dominate the AES use of q0 (this is slightly stretching the conventional definition of dominance, but think of this as "there exists a path from entry to the aes, which does not contain any of the safe instructions". Sadly, MachineDominance doesn't allow us to make this kind of query either!).

In this case though, it is safe to insert the protection at function entry, because that will (by definition) dominate all the AES uses, and the protection doesn't need to be dominated by the safe definitions, as we know they're safe.

I intend to follow-up this initial patch with an enhancement to ReachingDefAnalysis which will also provide the information that you have a set of defs inside the function, and also you're live-in, as this is required info for any conservative pass using the ReachingDefAnalysis. I felt, however, that given the pass is safe as-is, it was good to proceed without this enhancement.

llvm/lib/Target/ARM/ARMTargetMachine.cpp

585

TIL. I'll add a comment about the constant island pass as well.

Should I also look at the restrictions on the Branch Targets pass? I can imagine we also don't want to separate instructions once we've calculated their targets locally?

[dmgreen]

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

200–207

Are these vmov of an immediate? Are they not safe?

I was expecting it to be the lanes sets (VSETLNi8) and other scalar instructions that were unsafe.

306

OK sounds good.

llvm/lib/Target/ARM/ARMTargetMachine.cpp

585

Yeah - It sounds like the BTI would need to remain as the first instruction in the block.

[lenary]

A few comments before I post the next version of the patch.

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

145

I'm wrong about this. The tablegen isPredicable is an override, other code might also set isPredicable to true, so I think findFirstPredOperandIdx should work too.

200–207

I have received the information on what is safe and what is not, and the next version of the patch will have this correct.

306

One note is that the exact problem you describe does show up in the tests (in aese_set64_via_ptr, where the vldr is "safe"), so when the pass is enhanced, we will notice the improvements.

llvm/lib/Target/ARM/ARMTargetMachine.cpp

585

Turns out the AES pass doesn't *have* to come before the BTI pass, because AES instructions are only available on A-profile, and BTI is M-profile. I still will move it to before all these passes anyway, just so it's clear what is going on.

[lenary]

• Updated set of safe instructions
• Address reviewer feedback, including reordering passes.

[dmgreen]

Looks OK to me, as far as I can see. If it worth adding a few extra instructions that may come up?

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

147

Perhaps add these, if they are safe:
VBICd/q
VBICi's, VORRi's
VBIF/VBIT/VBSL/VBSP
VCEQ/VCNE/etc?
VDUP? VEXT?
VMVN imm equivalents of VMOV's
VREV's?
VSHL's, VSHR's?
I'm not sure if they will be very useful, but they are the kind of instructions that may come up in aes algorithms.

llvm/lib/Target/ARM/ARMTargetMachine.cpp

582–583

"No new instructions may be inserted" -> "Block sizes cannot be increased"
And maybe "will affect the offsets used for accessing these constants." -> "may push the branch ranges and load offsets of accessing constant pools out of range."

583–585

It's not about "not inserting instructions" exactly - it will replace psuedos with all kinds of new instructions :)
The pseudos needed to have a conservative size through ConstantIslandPass though to allow that though. It does make sure that it will not move instructions further apart from their targets.

[simon_tatham]

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

13–20

This description would leave me still confused if I didn't happen to already know roughly what the plan was. It jumps in half way through the explanation that someone would need if they were coming to this pass cold. (E.g. it talks about "the VORRq" before having even mentioned that there is one, let alone why there is one.)

How about the suggested text as a rewording?

311

nit: repeated word

[lenary]

Ack to all the comment clarifications, will update patch with those soon (probably tomorrow).

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

147

I'm very keen to avoid scope-creep on this patch, so I'm going to push back on this comment.

We know this list as given is safe (and have had internal confirmation). I've sent a new email internally with your list of instructions, to find out of they're safe too, but I'd like any answer to that to be part of a follow-up patch rather than blocking this patch for yet longer.

I believe what I have today is correct, even if the list is not optimal for all expected AES code sequences.

[dmgreen]

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

147

Yeah that sounds OK. So long as you address Simons comments and follow up with the instructions at a later date, this LGTM.

[lenary]

• Address comment nits
• Rebase

[lenary]

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

13–20

I have taken this feedback on board, and reworded the explanation based on your suggestion, with some slight ordering changes and a little more detail about the complex cases and some separation between the erratum itself, and the workaround we have chosen.

[dmgreen]

Thanks. LGTM

[simon_tatham]

Thanks, that explanation looks fine. (And I agree that re-paragraphing it made more sense than my version)

clang/include/clang/Driver/Options.td

 def mgeneral_regs_only : Flag<["-"], "mgeneral-regs-only">, Group<m_Group>,
   HelpText<"Generate code which only uses the general purpose registers (AArch64/x86 only)">;
 def mfix_cmse_cve_2021_35465 : Flag<["-"], "mfix-cmse-cve-2021-35465">,
   Group<m_arm_Features_Group>,
   HelpText<"Work around VLLDM erratum CVE-2021-35465 (ARM only)">;
 def mno_fix_cmse_cve_2021_35465 : Flag<["-"], "mno-fix-cmse-cve-2021-35465">,
   Group<m_arm_Features_Group>,
   HelpText<"Don't work around VLLDM erratum CVE-2021-35465 (ARM only)">;
+def mfix_cortex_a57_aes_1742098 : Flag<["-"], "mfix-cortex-a57-aes-1742098">,
+  Group<m_arm_Features_Group>,
+  HelpText<"Work around Cortex-A57 Erratum 1742098 (ARM only)">;
+def mno_fix_cortex_a57_aes_1742098 : Flag<["-"], "mno-fix-cortex-a57-aes-1742098">,
+  Group<m_arm_Features_Group>,
+  HelpText<"Don't work around Cortex-A57 Erratum 1742098 (ARM only)">;
+def mfix_cortex_a72_aes_1655431 : Flag<["-"], "mfix-cortex-a72-aes-1655431">,
+  Group<m_arm_Features_Group>,
+  HelpText<"Work around Cortex-A72 Erratum 1655431 (ARM only)">,
+  Alias<mfix_cortex_a57_aes_1742098>;
+def mno_fix_cortex_a72_aes_1655431 : Flag<["-"], "mno-fix-cortex-a72-aes-1655431">,
+  Group<m_arm_Features_Group>,
+  HelpText<"Don't work around Cortex-A72 Erratum 1655431 (ARM only)">,
+  Alias<mno_fix_cortex_a57_aes_1742098>;
 def mfix_cortex_a53_835769 : Flag<["-"], "mfix-cortex-a53-835769">,
   Group<m_aarch64_Features_Group>,
   HelpText<"Workaround Cortex-A53 erratum 835769 (AArch64 only)">;
 def mno_fix_cortex_a53_835769 : Flag<["-"], "mno-fix-cortex-a53-835769">,
   Group<m_aarch64_Features_Group>,
   HelpText<"Don't workaround Cortex-A53 erratum 835769 (AArch64 only)">;
 def mmark_bti_property : Flag<["-"], "mmark-bti-property">,
   Group<m_aarch64_Features_Group>,

clang/lib/Driver/ToolChains/Arch/ARM.cpp

if (!Args.getLastArg(options::OPT_mcmse))
           << A->getOption().getName() << "-mcmse";
 
     if (A->getOption().matches(options::OPT_mfix_cmse_cve_2021_35465))
       Features.push_back("+fix-cmse-cve-2021-35465");
     else
       Features.push_back("-fix-cmse-cve-2021-35465");
   }
 
+  // This also handles the -m(no-)fix-cortex-a72-1655431 arguments via aliases.
+  if (Arg *A = Args.getLastArg(options::OPT_mfix_cortex_a57_aes_1742098,
+                               options::OPT_mno_fix_cortex_a57_aes_1742098)) {
+    if (A->getOption().matches(options::OPT_mfix_cortex_a57_aes_1742098)) {
+      Features.push_back("+fix-cortex-a57-aes-1742098");
+    } else {
+      Features.push_back("-fix-cortex-a57-aes-1742098");
+    }
+  }
+
   // Look for the last occurrence of -mlong-calls or -mno-long-calls. If
   // neither options are specified, see if we are compiling for kernel/kext and
   // decide whether to pass "+long-calls" based on the OS and its version.
   if (Arg *A = Args.getLastArg(options::OPT_mlong_calls,
                                options::OPT_mno_long_calls)) {
     if (A->getOption().matches(options::OPT_mlong_calls))
       Features.push_back("+long-calls");
   } else if (KernelOrKext && (!Triple.isiOS() || Triple.isOSVersionLT(6)) &&

clang/test/Driver/arm-fix-cortex-a57-aes-1742098.c

+// RUN: %clang -### %s -target arm-none-none-eabi -march=armv8a -mfix-cortex-a57-aes-1742098 2>&1 | FileCheck %s --check-prefix=FIX
+// RUN: %clang -### %s -target arm-none-none-eabi -march=armv8a -mno-fix-cortex-a57-aes-1742098 2>&1 | FileCheck %s --check-prefix=NO-FIX
+
+// RUN: %clang -### %s -target arm-none-none-eabi -march=armv8a -mfix-cortex-a72-aes-1655431 2>&1 | FileCheck %s --check-prefix=FIX
+// RUN: %clang -### %s -target arm-none-none-eabi -march=armv8a -mno-fix-cortex-a72-aes-1655431 2>&1 | FileCheck %s --check-prefix=NO-FIX
+
+// RUN: %clang -### %s -target arm-none-none-eabi -march=armv8a 2>&1 | FileCheck %s --check-prefix=UNSPEC
+// RUN: %clang -### %s -target arm-none-none-eabi -march=armv8a 2>&1 | FileCheck %s --check-prefix=UNSPEC
+
+// This test checks that "-m(no-)fix-cortex-a57-aes-1742098" and
+// "-m(no-)fix-cortex-a72-aes-1655431" cause the "fix-cortex-a57-aes-1742098"
+// target feature to be passed to `clang -cc1`.
+//
+// This feature is also enabled in the backend for the two affected CPUs and the
+// "generic" cpu (used when only specifying -march), but that won't show up on
+// the `clang -cc1` command line.
+//
+// We do not check whether this option is correctly specified for the CPU: users
+// can specify the "-mfix-cortex-a57-aes-1742098" option with "-mcpu=cortex-a72"
+// and vice-versa, and will still get the fix, as the target feature and the fix
+// is the same in both cases.
+
+// FIX: "-target-feature" "+fix-cortex-a57-aes-1742098"
+// NO-FIX: "-target-feature" "-fix-cortex-a57-aes-1742098"
+// UNSPEC-NOT: "-target-feature" "{[+-]}fix-cortex-a57-aes-1742098"

llvm/lib/Target/ARM/ARM.h

FunctionPass *createThumb2SizeReductionPass(
     std::function<bool(const Function &)> Ftor = nullptr);
 InstructionSelector *
 createARMInstructionSelector(const ARMBaseTargetMachine &TM, const ARMSubtarget &STI,
                              const ARMRegisterBankInfo &RBI);
 Pass *createMVEGatherScatterLoweringPass();
 FunctionPass *createARMSLSHardeningPass();
 FunctionPass *createARMIndirectThunks();
 Pass *createMVELaneInterleavingPass();
+FunctionPass *createARMFixCortexA57AES1742098Pass();
 
 void LowerARMMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI,
                                   ARMAsmPrinter &AP);
 
 void initializeARMParallelDSPPass(PassRegistry &);
 void initializeARMLoadStoreOptPass(PassRegistry &);
 void initializeARMPreAllocLoadStoreOptPass(PassRegistry &);
 void initializeARMBranchTargetsPass(PassRegistry &);
 void initializeARMConstantIslandsPass(PassRegistry &);
 void initializeARMExpandPseudoPass(PassRegistry &);
 void initializeThumb2SizeReducePass(PassRegistry &);
 void initializeThumb2ITBlockPass(PassRegistry &);
 void initializeMVEVPTBlockPass(PassRegistry &);
 void initializeMVETPAndVPTOptimisationsPass(PassRegistry &);
 void initializeARMLowOverheadLoopsPass(PassRegistry &);
 void initializeARMBlockPlacementPass(PassRegistry &);
 void initializeMVETailPredicationPass(PassRegistry &);
 void initializeMVEGatherScatterLoweringPass(PassRegistry &);
 void initializeARMSLSHardeningPass(PassRegistry &);
 void initializeMVELaneInterleavingPass(PassRegistry &);
+void initializeARMFixCortexA57AES1742098Pass(PassRegistry &);
 
 } // end namespace llvm
 
 #endif // LLVM_LIB_TARGET_ARM_ARM_H

llvm/lib/Target/ARM/ARM.td

def FeaturePACBTI         : SubtargetFeature<"pacbti", "HasPACBTI", "true",
                                              "Target Identification">;
 
 /// Don't place a BTI instruction after return-twice constructs (setjmp).
 def FeatureNoBTIAtReturnTwice : SubtargetFeature<"no-bti-at-return-twice",
                                                  "NoBTIAtReturnTwice", "true",
                                                  "Don't place a BTI instruction "
                                                  "after a return-twice">;
 
+def FeatureFixCortexA57AES1742098 : SubtargetFeature<"fix-cortex-a57-aes-1742098",
+  "FixCortexA57AES1742098", "true",
+  "Work around Cortex-A57 Erratum 1742098 / Cortex-A72 Erratum 1655431 (AES)">;
+
 //===----------------------------------------------------------------------===//
 // ARM architecture class
 //
 
 // A-series ISA
 def FeatureAClass : SubtargetFeature<"aclass", "ARMProcClass", "AClass",
                                      "Is application profile ('A' series)">;
 

 include "ARMScheduleA57.td"
 include "ARMScheduleM4.td"
 include "ARMScheduleM7.td"
 
 //===----------------------------------------------------------------------===//
 // ARM processors
 //
 // Dummy CPU, used to target architectures
-def : ProcessorModel<"generic",     CortexA8Model,      []>;
+def : ProcessorModel<"generic",     CortexA8Model,      [FeatureFixCortexA57AES1742098]>;
 
 // FIXME: Several processors below are not using their own scheduler
 // model, but one of similar/previous processor. These should be fixed.
 
 def : ProcNoItin<"arm8",                                [ARMv4]>;
 def : ProcNoItin<"arm810",                              [ARMv4]>;
 def : ProcNoItin<"strongarm",                           [ARMv4]>;
 def : ProcNoItin<"strongarm110",                        [ARMv4]>;

 
 def : ProcessorModel<"cortex-a57",  CortexA57Model,     [ARMv8a, ProcA57,
                                                          FeatureHWDivThumb,
                                                          FeatureHWDivARM,
                                                          FeatureCrypto,
                                                          FeatureCRC,
                                                          FeatureFPAO,
                                                          FeatureAvoidPartialCPSR,
-                                                         FeatureCheapPredicableCPSR]>;
+                                                         FeatureCheapPredicableCPSR,
+                                                         FeatureFixCortexA57AES1742098]>;
 
 def : ProcessorModel<"cortex-a72",  CortexA57Model,     [ARMv8a, ProcA72,
                                                          FeatureHWDivThumb,
                                                          FeatureHWDivARM,
                                                          FeatureCrypto,
-                                                         FeatureCRC]>;
+                                                         FeatureCRC,
+                                                         FeatureFixCortexA57AES1742098]>;
 
 def : ProcNoItin<"cortex-a73",                          [ARMv8a, ProcA73,
                                                          FeatureHWDivThumb,
                                                          FeatureHWDivARM,
                                                          FeatureCrypto,
                                                          FeatureCRC]>;
 
 def : ProcNoItin<"cortex-a75",                          [ARMv82a, ProcA75,

llvm/lib/Target/ARM/ARMFixCortexA57AES1742098Pass.cpp

+//===-- ARMFixCortexA57AES1742098Pass.cpp ---------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// This pass works around a Cortex Core Fused AES erratum:
+// - Cortex-A57 Erratum 1742098
+// - Cortex-A72 Erratum 1655431
+//
+// The intention is this:
+// - Any 128-bit or 64-bit writes to the neon input register of an AES fused
+//   pair are safe (the inputs are to the AESE/AESD instruction).
+// - Any 32-bit writes to the input register are unsafe, but these may happen
+//   in another function, or only on some control flow paths. In these cases,
+//   conservatively insert the VORRq anyway.
+// - So, analyse both inputs to the AESE/AESD instruction, inserting a VORR if
+//   you cannot prove they're on a list of allowed instructions.
+//===----------------------------------------------------------------------===//
+
+#include "ARM.h"
+#include "ARMBaseInstrInfo.h"
+#include "ARMBaseRegisterInfo.h"
+#include "ARMSubtarget.h"
+#include "Utils/ARMBaseInfo.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineInstrBundleIterator.h"
+#include "llvm/CodeGen/MachineOperand.h"
+#include "llvm/CodeGen/ReachingDefAnalysis.h"
+#include "llvm/CodeGen/Register.h"
+#include "llvm/CodeGen/TargetRegisterInfo.h"
+#include "llvm/IR/DebugLoc.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/MC/MCInstrDesc.h"
+#include "llvm/Pass.h"
+#include "llvm/PassRegistry.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include <assert.h>
+#include <stdint.h>
+
+using namespace llvm;
+
+#define DEBUG_TYPE "arm-fix-cortex-a57-aes-1742098"
+
+//===----------------------------------------------------------------------===//
+
+namespace {
+class ARMFixCortexA57AES1742098 : public MachineFunctionPass {
+public:
+  static char ID;
+  explicit ARMFixCortexA57AES1742098() : MachineFunctionPass(ID) {
+    initializeARMFixCortexA57AES1742098Pass(*PassRegistry::getPassRegistry());
+  }
+
+  bool runOnMachineFunction(MachineFunction &F) override;
+
+  MachineFunctionProperties getRequiredProperties() const override {
+    return MachineFunctionProperties().set(
+        MachineFunctionProperties::Property::NoVRegs);
+  }
+
+  StringRef getPassName() const override {
+    return "ARM fix for Cortex-A57 AES Erratum 1742098";
+  }
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.addRequired<ReachingDefAnalysis>();
+    AU.setPreservesCFG();
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+
+private:
+  // This is the information needed to insert the fixup in the right place.
+  struct AESFixupLocation {
+    MachineBasicBlock *Block;
+    // The fixup instruction will be inserted *before* InsertionPt.
+    MachineInstr *InsertionPt;
+    MachineOperand *MOp;
+  };
+
+  void analyzeMF(MachineFunction &MF, ReachingDefAnalysis &RDA,
+                 const ARMBaseRegisterInfo *TRI,
+                 SmallVectorImpl<AESFixupLocation> &FixupLocsForFn) const;
+
+  void insertAESFixup(AESFixupLocation &FixupLoc, const ARMBaseInstrInfo *TII,
+                      const ARMBaseRegisterInfo *TRI) const;
+
+  static bool isFirstAESPairInstr(MachineInstr &MI);
+  static bool isSafeAESInput(MachineInstr &MI);
+};
+char ARMFixCortexA57AES1742098::ID = 0;
+
+} // end anonymous namespace
+
+INITIALIZE_PASS_BEGIN(ARMFixCortexA57AES1742098, DEBUG_TYPE,
+                      "ARM fix for Cortex-A57 AES Erratum 1742098", false,
+                      false)
+INITIALIZE_PASS_DEPENDENCY(ReachingDefAnalysis);
+INITIALIZE_PASS_END(ARMFixCortexA57AES1742098, DEBUG_TYPE,
+                    "ARM fix for Cortex-A57 AES Erratum 1742098", false, false)
+
+//===----------------------------------------------------------------------===//
+
+bool ARMFixCortexA57AES1742098::isFirstAESPairInstr(MachineInstr &MI) {
+  unsigned Opc = MI.getOpcode();
+  return Opc == ARM::AESD || Opc == ARM::AESE;
+}
+
+bool ARMFixCortexA57AES1742098::isSafeAESInput(MachineInstr &MI) {
+  auto CondCodeIsAL = [&MI](unsigned CCIdx) -> bool {
+    assert(MI.getDesc().OpInfo[CCIdx].isPredicate() &&
+           "CCIdx operand must be a predicate");
+    return MI.getOperand(CCIdx).getImm() == (int64_t)ARMCC::AL;
+  };
+
+  switch (MI.getOpcode()) {
+  // Unknown: Assume not safe.
+  default:
+    return false;
+  // 128-bit wide AES instructions
+  case ARM::AESD:
+  case ARM::AESE:
+  case ARM::AESMC:
+  case ARM::AESIMC:
+    return true;
+  // 128-bit and 64-bit wide bitwise ops (when condition = al)
+  case ARM::VANDd:
+  case ARM::VANDq:
+  case ARM::VORRd:
+  case ARM::VORRq:
+  case ARM::VEORd:
+  case ARM::VEORq:
+  case ARM::VMVNd:
+  case ARM::VMVNq:
+    return CondCodeIsAL(3);
+  // VMOV of 64-bit value between D registers (when condition = al)
+  case ARM::VMOVD:
+    return CondCodeIsAL(2);
+  // VMOV of 64 bit value from GPRs (when condition = al)
+  case ARM::VMOVDRR:
+    return CondCodeIsAL(3);
+  // VMOV of 64-bit immediate into D or Q registers (when condition = al)
+  case ARM::VMOVv2i64:
+  case ARM::VMOVv1i64:
+    return CondCodeIsAL(2);
+  // Loads (when condition = al)
+  // VLD Dn, [Rn, #imm]
+  case ARM::VLDRD:
+    return CondCodeIsAL(3);
+  // VLDM
+  case ARM::VLDMDDB_UPD:
+  case ARM::VLDMDIA_UPD:
+    return CondCodeIsAL(2);
+  case ARM::VLDMDIA:
+    return CondCodeIsAL(1);
+  // VLDn to all lanes (to one single lane is unsafe).
+  case ARM::VLD1d64:
+  case ARM::VLD1q64:
+  case ARM::VLD1d32:
+  case ARM::VLD1q32:
+  case ARM::VLD2b32:
+  case ARM::VLD2d32:
+  case ARM::VLD2q32:
+  case ARM::VLD1d16:
+  case ARM::VLD1q16:
+  case ARM::VLD2d16:
+  case ARM::VLD2q16:
+  case ARM::VLD1d8:
+  case ARM::VLD1q8:
+  case ARM::VLD2b8:
+  case ARM::VLD2d8:
+  case ARM::VLD2q8:
+    return CondCodeIsAL(3);
+  case ARM::VLD3d32:
+  case ARM::VLD3q32:
+  case ARM::VLD3d16:
+  case ARM::VLD3q16:
+  case ARM::VLD3d8:
+  case ARM::VLD3q8:
+    return CondCodeIsAL(5);
+  case ARM::VLD4d32:
+  case ARM::VLD4q32:
+  case ARM::VLD4d16:
+  case ARM::VLD4q16:
+  case ARM::VLD4d8:
+  case ARM::VLD4q8:
+    return CondCodeIsAL(6);
+  // Always Unsafe:
+  // VMOV of smaller immediate into D or Q
+  case ARM::VMOVv2f32:
+  case ARM::VMOVv4f32:
+  case ARM::VMOVv2i32:
+  case ARM::VMOVv4i32:
+  case ARM::VMOVv4i16:
+  case ARM::VMOVv8i16:
+  case ARM::VMOVv8i8:
+  case ARM::VMOVv16i8:
+    return false;
+  };
+
+  return false;
+}
+
+bool ARMFixCortexA57AES1742098::runOnMachineFunction(MachineFunction &F) {
+  LLVM_DEBUG(dbgs() << "***** ARMFixCortexA57AES1742098 *****\n");
+  auto &STI = F.getSubtarget<ARMSubtarget>();
+
+  // Fix not requested or AES instructions not present: skip pass.
+  if (!STI.hasAES() || !STI.fixCortexA57AES1742098())
+    return false;
+
+  const ARMBaseRegisterInfo *TRI = STI.getRegisterInfo();
+  const ARMBaseInstrInfo *TII = STI.getInstrInfo();
+
+  auto &RDA = getAnalysis<ReachingDefAnalysis>();
+
+  // Analyze whole function to find instructions which need fixing up...
+  SmallVector<AESFixupLocation> FixupLocsForFn{};
+  analyzeMF(F, RDA, TRI, FixupLocsForFn);
+
+  // ... and fix the instructions up all at the same time.
+  bool Changed = false;
+  LLVM_DEBUG(dbgs() << "Inserting " << FixupLocsForFn.size() << " fixup(s)\n");
+  for (AESFixupLocation &FixupLoc : FixupLocsForFn) {
+    insertAESFixup(FixupLoc, TII, TRI);
+    Changed |= true;
+  }
+
+  return Changed;
+}
+
+void ARMFixCortexA57AES1742098::analyzeMF(
+    MachineFunction &MF, ReachingDefAnalysis &RDA,
+    const ARMBaseRegisterInfo *TRI,
+    SmallVectorImpl<AESFixupLocation> &FixupLocsForFn) const {
+  unsigned MaxAllowedFixups = 0;
+
+  for (MachineBasicBlock &MBB : MF) {
+    // Early return if no instructions are the start of an AES Pair.
+    if (!llvm::any_of(MBB.instrs(), isFirstAESPairInstr))
+      continue;
+
+    for (MachineInstr &MI : MBB) {
+      if (!isFirstAESPairInstr(MI))
+        continue;
+
+      // Found an instruction to check the operands of.
+      LLVM_DEBUG(dbgs() << "Found AES Pair starting: " << MI);
+      assert(MI.getNumExplicitOperands() == 3 && MI.getNumExplicitDefs() == 1 &&
+             "Unknown AES Instruction Format. Expected 1 def, 2 uses.");
+
+      // A maximum of two fixups should be inserted for each AES pair (one per
+      // register use).
+      MaxAllowedFixups += 2;
+
+      // Inspect all operands, choosing whether to insert a fixup.
+      for (MachineOperand &MOp : MI.uses()) {
+        SmallPtrSet<MachineInstr *, 1> AllDefs{};
+        RDA.getGlobalReachingDefs(&MI, MOp.getReg(), AllDefs);
+
+        // Planned Fixup: This should be added to FixupLocsForFn at most once.
+        AESFixupLocation NewLoc{&MBB, &MI, &MOp};
+
+        // In small functions with loops, this operand may be both a live-in and
+        // have definitions within the function itself. These will need a fixup.
+        bool IsLiveIn = MF.front().isLiveIn(MOp.getReg());
+
+        // If the register doesn't have defining instructions, and is not a
+        // live-in, then something is wrong and the fixup must always be
+        // inserted to be safe.
+        if (!IsLiveIn && AllDefs.size() == 0) {
+          LLVM_DEBUG(dbgs()
+                     << "Fixup Planned: No Defining Instrs found, not live-in: "
+                     << printReg(MOp.getReg(), TRI) << "\n");
+          FixupLocsForFn.emplace_back(NewLoc);
+          continue;
+        }
+
+        auto IsUnsafe = [](MachineInstr *MI) -> bool {
+          return !isSafeAESInput(*MI);
+        };
+        size_t UnsafeCount = llvm::count_if(AllDefs, IsUnsafe);
+
+        // If there are no unsafe unsafe definitions...
+        if (UnsafeCount == 0) {
+          // ... and the register is not live-in ...
+          if (!IsLiveIn) {
+            // ... then skip the fixup.
+            LLVM_DEBUG(dbgs() << "No Fixup: Defining instrs are all safe: "
+                              << printReg(MOp.getReg(), TRI) << "\n");
+            continue;
+          }
+
+          // Otherwise, the only unsafe "definition" is a live-in, so insert the
+          // fixup at the start of the function.
+          LLVM_DEBUG(dbgs()
+                     << "Fixup Planned: Live-In (with safe defining instrs): "
+                     << printReg(MOp.getReg(), TRI) << "\n");
+          NewLoc.Block = &MF.front();
+          NewLoc.InsertionPt = &*NewLoc.Block->begin();
+          LLVM_DEBUG(dbgs() << "Moving Fixup for Live-In to immediately before "
+                            << *NewLoc.InsertionPt);
+          FixupLocsForFn.emplace_back(NewLoc);
+          continue;
+        }
+
+        // If a fixup is needed in more than one place, then the best place to
+        // insert it is adjacent to the use rather than introducing a fixup
+        // adjacent to each def.
+        //
+        // FIXME: It might be better to hoist this to the start of the BB, if
+        // possible.
+        if (IsLiveIn || UnsafeCount > 1) {
+          LLVM_DEBUG(dbgs() << "Fixup Planned: Multiple unsafe defining instrs "
+                               "(including live-ins): "
+                            << printReg(MOp.getReg(), TRI) << "\n");
+          FixupLocsForFn.emplace_back(NewLoc);
+          continue;
+        }
+
+        assert(UnsafeCount == 1 && !IsLiveIn &&
+               "At this point, there should be one unsafe defining instrs "
+               "and the defined register should not be a live-in.");
+        SmallPtrSetIterator<MachineInstr *> It =
+            llvm::find_if(AllDefs, IsUnsafe);
+        assert(It != AllDefs.end() &&
+               "UnsafeCount == 1 but No Unsafe MachineInstr found.");
+        MachineInstr *DefMI = *It;
+
+        LLVM_DEBUG(
+            dbgs() << "Fixup Planned: Found single unsafe defining instrs for "
+                   << printReg(MOp.getReg(), TRI) << ": " << *DefMI);
+
+        // There is one unsafe defining instruction, which needs a fixup. It is
+        // generally good to hoist the fixup to be adjacent to the defining
+        // instruction rather than the using instruction, as the using
+        // instruction may be inside a loop when the defining instruction is
+        // not.
+        MachineBasicBlock::iterator DefIt = DefMI;
+        ++DefIt;
+        if (DefIt != DefMI->getParent()->end()) {
+          LLVM_DEBUG(dbgs() << "Moving Fixup to immediately after " << *DefMI
+                            << "And immediately before " << *DefIt);
+          NewLoc.Block = DefIt->getParent();
+          NewLoc.InsertionPt = &*DefIt;
+        }
+
+        FixupLocsForFn.emplace_back(NewLoc);
+      }
+    }
+  }
+
+  assert(FixupLocsForFn.size() <= MaxAllowedFixups &&
+         "Inserted too many fixups for this function.");
+}
+
+void ARMFixCortexA57AES1742098::insertAESFixup(
+    AESFixupLocation &FixupLoc, const ARMBaseInstrInfo *TII,
+    const ARMBaseRegisterInfo *TRI) const {
+  MachineOperand *OperandToFixup = FixupLoc.MOp;
+
+  assert(OperandToFixup->isReg() && "OperandToFixup must be a register");
+  Register RegToFixup = OperandToFixup->getReg();
+
+  LLVM_DEBUG(dbgs() << "Inserting VORRq of " << printReg(RegToFixup, TRI)
+                    << " before: " << *FixupLoc.InsertionPt);
+
+  // Insert the new `VORRq qN, qN, qN`. There are a few details here:
+  //
+  // The uses are marked as killed, even if the original use of OperandToFixup
+  // is not killed, as the new instruction is clobbering the register. This is
+  // safe even if there are other uses of `qN`, as the VORRq value-wise a no-op
+  // (it is inserted for microarchitectural reasons).
+  //
+  // The def and the uses are still marked as Renamable if the original register
+  // was, to avoid having to rummage through all the other uses and defs and
+  // unset their renamable bits.
+  unsigned Renamable = OperandToFixup->isRenamable() ? RegState::Renamable : 0;
+  BuildMI(*FixupLoc.Block, FixupLoc.InsertionPt, DebugLoc(),
+          TII->get(ARM::VORRq))
+      .addReg(RegToFixup, RegState::Define | Renamable)
+      .addReg(RegToFixup, RegState::Kill | Renamable)
+      .addReg(RegToFixup, RegState::Kill | Renamable)
+      .addImm((uint64_t)ARMCC::AL)
+      .addReg(ARM::NoRegister);
+}
+
+// Factory function used by AArch64TargetMachine to add the pass to
+// the passmanager.
+FunctionPass *llvm::createARMFixCortexA57AES1742098Pass() {
+  return new ARMFixCortexA57AES1742098();
+}

llvm/lib/Target/ARM/ARMTargetMachine.cpp

extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeARMTarget() {
   initializeMVEVPTBlockPass(Registry);
   initializeMVETPAndVPTOptimisationsPass(Registry);
   initializeMVETailPredicationPass(Registry);
   initializeARMLowOverheadLoopsPass(Registry);
   initializeARMBlockPlacementPass(Registry);
   initializeMVEGatherScatterLoweringPass(Registry);
   initializeARMSLSHardeningPass(Registry);
   initializeMVELaneInterleavingPass(Registry);
+  initializeARMFixCortexA57AES1742098Pass(Registry);
 }
 
 static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
   if (TT.isOSBinFormatMachO())
     return std::make_unique<TargetLoweringObjectFileMachO>();
   if (TT.isOSWindows())
     return std::make_unique<TargetLoweringObjectFileCOFF>();
   return std::make_unique<ARMElfTargetObjectFile>();
void ARMPassConfig::addPreEmitPass() {
   if (getOptLevel() != CodeGenOpt::None) {
     addPass(createARMBlockPlacementPass());
     addPass(createARMOptimizeBarriersPass());
   }
 }
 
 void ARMPassConfig::addPreEmitPass2() {
   addPass(createARMBranchTargetsPass());
   addPass(createARMConstantIslandPass());
   addPass(createARMLowOverheadLoopsPass());
 
+  addPass(createARMFixCortexA57AES1742098Pass());
+
   if (TM->getTargetTriple().isOSWindows()) {
     // Identify valid longjmp targets for Windows Control Flow Guard.
     addPass(createCFGuardLongjmpPass());
     // Identify valid eh continuation targets for Windows EHCont Guard.
     addPass(createEHContGuardCatchretPass());
   }
 }

llvm/lib/Target/ARM/CMakeLists.txt

add_llvm_target(ARMCodeGen
   ARMBasicBlockInfo.cpp
   ARMBranchTargets.cpp
   ARMCallingConv.cpp
   ARMCallLowering.cpp
   ARMConstantIslandPass.cpp
   ARMConstantPoolValue.cpp
   ARMExpandPseudoInsts.cpp
   ARMFastISel.cpp
+  ARMFixCortexA57AES1742098Pass.cpp
   ARMFrameLowering.cpp
   ARMHazardRecognizer.cpp
   ARMInstructionSelector.cpp
   ARMISelDAGToDAG.cpp
   ARMISelLowering.cpp
   ARMInstrInfo.cpp
   ARMLegalizerInfo.cpp
   ARMParallelDSP.cpp

llvm/test/CodeGen/ARM/O3-pipeline.ll

 ; CHECK-NEXT:    FunctionPass Manager
 ; CHECK-NEXT:      ARM Branch Targets
 ; CHECK-NEXT:      MachineDominator Tree Construction
 ; CHECK-NEXT:      ARM constant island placement and branch shortening pass
 ; CHECK-NEXT:      MachineDominator Tree Construction
 ; CHECK-NEXT:      Machine Natural Loop Construction
 ; CHECK-NEXT:      ReachingDefAnalysis
 ; CHECK-NEXT:      ARM Low Overhead Loops pass
+; CHECK-NEXT:      ReachingDefAnalysis
+; CHECK-NEXT:      ARM fix for Cortex-A57 AES Erratum 1742098
 ; CHECK-NEXT:      Lazy Machine Block Frequency Analysis
 ; CHECK-NEXT:      Machine Optimization Remark Emitter
 ; CHECK-NEXT:      ARM Assembly Printer
 ; CHECK-NEXT:      Free MachineFunction

llvm/test/CodeGen/ARM/aes-erratum-fix.ll

 declare <16 x i8> @llvm.arm.neon.aesimc(<16 x i8>)
 
 
 define arm_aapcs_vfpcc void @aese_zero(<16 x i8>* %0) nounwind {
 ; CHECK-FIX-LABEL: aese_zero:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r0]
 ; CHECK-FIX-NEXT:    vmov.i32 q9, #0x0
+; CHECK-FIX-NEXT:    vorr q9, q9, q9
 ; CHECK-FIX-NEXT:    aese.8 q9, q8
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q9
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r0]
 ; CHECK-FIX-NEXT:    bx lr
   %2 = load <16 x i8>, <16 x i8>* %0, align 8
   %3 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> zeroinitializer, <16 x i8> %2)
   %4 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %3)
   store <16 x i8> %4, <16 x i8>* %0, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_once_via_ptr(<16 x i8>* %0, <16 x i8>* %1) nounwind {
 ; CHECK-FIX-LABEL: aese_once_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r0]
 ; CHECK-FIX-NEXT:    vld1.64 {d18, d19}, [r1]
 ; CHECK-FIX-NEXT:    aese.8 q9, q8
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q9
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %3 = load <16 x i8>, <16 x i8>* %1, align 8
   %4 = load <16 x i8>, <16 x i8>* %0, align 8
   %5 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %3, <16 x i8> %4)
   %6 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %5)
   store <16 x i8> %6, <16 x i8>* %1, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc <16 x i8> @aese_once_via_val(<16 x i8> %0, <16 x i8> %1) nounwind {
 ; CHECK-FIX-LABEL: aese_once_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q1, q1, q1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q1, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q0, q1
 ; CHECK-FIX-NEXT:    bx lr
   %3 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %1, <16 x i8> %0)
   %4 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %3)
   ret <16 x i8> %4
 }
 
; CHECK-FIX-NEXT:    bx lr
   %9 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %1, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc <16 x i8> @aese_twice_via_val(<16 x i8> %0, <16 x i8> %1) nounwind {
 ; CHECK-FIX-LABEL: aese_twice_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q1, q1, q1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q1, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q1
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q0, q8
 ; CHECK-FIX-NEXT:    bx lr
   %3 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %1, <16 x i8> %0)
   %4 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %3)
   %5 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %4, <16 x i8> %0)
6:
   %12 = add nuw i32 %7, 1
   %13 = icmp eq i32 %12, %0
   br i1 %13, label %5, label %6
 }
 
 define arm_aapcs_vfpcc <16 x i8> @aese_loop_via_val(i32 %0, <16 x i8> %1, <16 x i8> %2) nounwind {
 ; CHECK-FIX-LABEL: aese_loop_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q1, q1, q1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB6_2
 ; CHECK-FIX-NEXT:  .LBB6_1: @ =>This Inner Loop Header: Depth=1
 ; CHECK-FIX-NEXT:    aese.8 q1, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesmc.8 q1, q1
 ; CHECK-FIX-NEXT:    bne .LBB6_1
 ; CHECK-FIX-NEXT:  .LBB6_2:
7:
   br i1 %13, label %5, label %7
 }
 
 define arm_aapcs_vfpcc void @aese_set8_via_ptr(i8* %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aese_set8_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.8 {d0[0]}, [r0]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = load i8, i8* %0, align 1
   %6 = insertelement <16 x i8> %1, i8 %5, i64 0
   %7 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %4, <16 x i8> %6)
   %8 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %7)
   store <16 x i8> %8, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set8_via_val(i8 zeroext %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aese_set8_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    vmov.8 d16[0], r0
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = insertelement <16 x i8> %4, i8 %0, i64 0
   %6 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %5, <16 x i8> %1)
   %7 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %6)
   store <16 x i8> %7, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set8_cond_via_ptr(i1 zeroext %0, i8* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set8_cond_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB9_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vld1.8 {d0[0]}, [r1]
 ; CHECK-FIX-NEXT:  .LBB9_2:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   br i1 %0, label %6, label %9
 
 6:
   %7 = load i8, i8* %1, align 1
   %8 = insertelement <16 x i8> %2, i8 %7, i64 0
   br label %9
 
 9:
   %10 = phi <16 x i8> [ %8, %6 ], [ %2, %4 ]
   %11 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %5, <16 x i8> %10)
   %12 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %11)
   store <16 x i8> %12, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set8_cond_via_val(i1 zeroext %0, i8 zeroext %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set8_cond_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB10_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vmov.8 d16[0], r1
 ; CHECK-FIX-NEXT:  .LBB10_2: @ %select.end
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   %6 = insertelement <16 x i8> %5, i8 %1, i64 0
   %7 = select i1 %0, <16 x i8> %6, <16 x i8> %5
   %8 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %7, <16 x i8> %2)
   %9 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set8_loop_via_ptr(i32 %0, i8* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set8_loop_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB11_1:
 ; CHECK-FIX-NEXT:    vld1.8 {d0[0]}, [r1]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB11_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB11_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load i8, i8* %1, align 1

 ; CHECK-FIX-LABEL: aese_set8_loop_via_val:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB12_1:
 ; CHECK-FIX-NEXT:    vmov.8 d0[0], r1
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB12_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB12_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = icmp eq i32 %0, 0
11:
   br i1 %17, label %9, label %11
 }
 
 define arm_aapcs_vfpcc void @aese_set16_via_ptr(i16* %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aese_set16_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.16 {d0[0]}, [r0:16]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = load i16, i16* %0, align 2
   %6 = bitcast <16 x i8> %1 to <8 x i16>
   %7 = insertelement <8 x i16> %6, i16 %5, i64 0
   %8 = bitcast <8 x i16> %7 to <16 x i8>
   %9 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %4, <16 x i8> %8)
   %10 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %9)
   store <16 x i8> %10, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set16_via_val(i16 zeroext %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aese_set16_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    vmov.16 d16[0], r0
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = bitcast <16 x i8>* %2 to <8 x i16>*
   %5 = load <8 x i16>, <8 x i16>* %4, align 8
   %6 = insertelement <8 x i16> %5, i16 %0, i64 0
   %7 = bitcast <8 x i16> %6 to <16 x i8>
   %8 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %7, <16 x i8> %1)
   %9 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set16_cond_via_ptr(i1 zeroext %0, i16* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set16_cond_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB15_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vld1.16 {d0[0]}, [r1:16]
 ; CHECK-FIX-NEXT:  .LBB15_2:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   br i1 %0, label %6, label %10
 
 6:
12:
   %16 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %15)
   store <16 x i8> %16, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set16_cond_via_val(i1 zeroext %0, i16 zeroext %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set16_cond_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB16_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vmov.16 d16[0], r1
 ; CHECK-FIX-NEXT:  .LBB16_2: @ %select.end
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = bitcast <16 x i8>* %3 to <8 x i16>*
   %6 = load <8 x i16>, <8 x i16>* %5, align 8
   %7 = insertelement <8 x i16> %6, i16 %1, i64 0
   %8 = select i1 %0, <8 x i16> %7, <8 x i16> %6
   %9 = bitcast <8 x i16> %8 to <16 x i8>
   %10 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %9, <16 x i8> %2)
   %11 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %10)
   store <16 x i8> %11, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set16_loop_via_ptr(i32 %0, i16* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set16_loop_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB17_1:
 ; CHECK-FIX-NEXT:    vld1.16 {d0[0]}, [r1:16]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB17_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB17_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load i16, i16* %1, align 2

 ; CHECK-FIX-LABEL: aese_set16_loop_via_val:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB18_1:
 ; CHECK-FIX-NEXT:    vmov.16 d0[0], r1
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB18_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB18_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = icmp eq i32 %0, 0
13:
   br i1 %19, label %11, label %13
 }
 
 define arm_aapcs_vfpcc void @aese_set32_via_ptr(i32* %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aese_set32_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.32 {d0[0]}, [r0:32]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = load i32, i32* %0, align 4
   %6 = bitcast <16 x i8> %1 to <4 x i32>
   %7 = insertelement <4 x i32> %6, i32 %5, i64 0
   %8 = bitcast <4 x i32> %7 to <16 x i8>
   %9 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %4, <16 x i8> %8)
   %10 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %9)
   store <16 x i8> %10, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set32_via_val(i32 %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aese_set32_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    vmov.32 d16[0], r0
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = bitcast <16 x i8>* %2 to <4 x i32>*
   %5 = load <4 x i32>, <4 x i32>* %4, align 8
   %6 = insertelement <4 x i32> %5, i32 %0, i64 0
   %7 = bitcast <4 x i32> %6 to <16 x i8>
   %8 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %7, <16 x i8> %1)
   %9 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set32_cond_via_ptr(i1 zeroext %0, i32* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set32_cond_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB21_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vld1.32 {d0[0]}, [r1:32]
 ; CHECK-FIX-NEXT:  .LBB21_2:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   br i1 %0, label %6, label %10
 
 6:
12:
   %16 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %15)
   store <16 x i8> %16, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set32_cond_via_val(i1 zeroext %0, i32 %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set32_cond_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB22_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vmov.32 d16[0], r1
 ; CHECK-FIX-NEXT:  .LBB22_2: @ %select.end
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = bitcast <16 x i8>* %3 to <4 x i32>*
   %6 = load <4 x i32>, <4 x i32>* %5, align 8
   %7 = insertelement <4 x i32> %6, i32 %1, i64 0
   %8 = select i1 %0, <4 x i32> %7, <4 x i32> %6
   %9 = bitcast <4 x i32> %8 to <16 x i8>
   %10 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %9, <16 x i8> %2)
   %11 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %10)
   store <16 x i8> %11, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set32_loop_via_ptr(i32 %0, i32* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set32_loop_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB23_1:
 ; CHECK-FIX-NEXT:    vld1.32 {d0[0]}, [r1:32]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB23_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB23_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load i32, i32* %1, align 4

 ; CHECK-FIX-LABEL: aese_set32_loop_via_val:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB24_1:
 ; CHECK-FIX-NEXT:    vmov.32 d0[0], r1
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB24_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB24_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = icmp eq i32 %0, 0
13:
   %18 = add nuw i32 %15, 1
   %19 = icmp eq i32 %18, %0
   br i1 %19, label %11, label %13
 }
 
 define arm_aapcs_vfpcc void @aese_set64_via_ptr(i64* %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-NOSCHED-LABEL: aese_set64_via_ptr:
 ; CHECK-FIX-NOSCHED:       @ %bb.0:
+; CHECK-FIX-NOSCHED-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NOSCHED-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NOSCHED-NEXT:    vldr d0, [r0]
 ; CHECK-FIX-NOSCHED-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NOSCHED-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NOSCHED-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NOSCHED-NEXT:    bx lr
 ;
 ; CHECK-CORTEX-FIX-LABEL: aese_set64_via_ptr:
 ; CHECK-CORTEX-FIX:       @ %bb.0:
+; CHECK-CORTEX-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-CORTEX-FIX-NEXT:    vldr d0, [r0]
 ; CHECK-CORTEX-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-CORTEX-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-CORTEX-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-CORTEX-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-CORTEX-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = load i64, i64* %0, align 8
   %6 = bitcast <16 x i8> %1 to <2 x i64>
   %7 = insertelement <2 x i64> %6, i64 %5, i64 0
   %8 = bitcast <2 x i64> %7 to <16 x i8>
   %9 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %4, <16 x i8> %8)
   %10 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %9)
   store <16 x i8> %10, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set64_via_val(i64 %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aese_set64_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    vmov.32 d16[0], r0
 ; CHECK-FIX-NEXT:    vmov.32 d16[1], r1
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = bitcast <16 x i8>* %2 to <2 x i64>*
   %5 = load <2 x i64>, <2 x i64>* %4, align 8
   %6 = insertelement <2 x i64> %5, i64 %0, i64 0
   %7 = bitcast <2 x i64> %6 to <16 x i8>
   %8 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %7, <16 x i8> %1)
   %9 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set64_cond_via_ptr(i1 zeroext %0, i64* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set64_cond_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    vldrne d0, [r1]
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   br i1 %0, label %6, label %10
 
 6:
12:
   %16 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %15)
   store <16 x i8> %16, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set64_cond_via_val(i1 zeroext %0, i64 %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set64_cond_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    ldr r1, [sp]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    beq .LBB28_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vmov.32 d16[0], r2
 ; CHECK-FIX-NEXT:    vmov.32 d16[1], r3
 ; CHECK-FIX-NEXT:  .LBB28_2: @ %select.end
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = bitcast <16 x i8>* %3 to <2 x i64>*
   %6 = load <2 x i64>, <2 x i64>* %5, align 8
   %7 = insertelement <2 x i64> %6, i64 %1, i64 0
   %8 = select i1 %0, <2 x i64> %7, <2 x i64> %6
   %9 = bitcast <2 x i64> %8 to <16 x i8>
   %10 = call <16 x i8> @llvm.arm.neon.aese(<16 x i8> %9, <16 x i8> %2)
   %11 = call <16 x i8> @llvm.arm.neon.aesmc(<16 x i8> %10)
   store <16 x i8> %11, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aese_set64_loop_via_ptr(i32 %0, i64* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aese_set64_loop_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB29_1:
 ; CHECK-FIX-NEXT:    vldr d0, [r1]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB29_2: @ =>This Inner Loop Header: Depth=1
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1

 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB30_1:
 ; CHECK-FIX-NEXT:    vmov.32 d0[0], r2
 ; CHECK-FIX-NEXT:    ldr r1, [sp]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    vmov.32 d0[1], r3
 ; CHECK-FIX-NEXT:  .LBB30_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aese.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesmc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB30_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = icmp eq i32 %0, 0
13:
   br i1 %19, label %11, label %13
 }
 
 define arm_aapcs_vfpcc void @aesd_zero(<16 x i8>* %0) nounwind {
 ; CHECK-FIX-LABEL: aesd_zero:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r0]
 ; CHECK-FIX-NEXT:    vmov.i32 q9, #0x0
+; CHECK-FIX-NEXT:    vorr q9, q9, q9
 ; CHECK-FIX-NEXT:    aesd.8 q9, q8
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q9
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r0]
 ; CHECK-FIX-NEXT:    bx lr
   %2 = load <16 x i8>, <16 x i8>* %0, align 8
   %3 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> zeroinitializer, <16 x i8> %2)
   %4 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %3)
   store <16 x i8> %4, <16 x i8>* %0, align 8
; CHECK-FIX-NEXT:    bx lr
   %6 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %5)
   store <16 x i8> %6, <16 x i8>* %1, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc <16 x i8> @aesd_once_via_val(<16 x i8> %0, <16 x i8> %1) nounwind {
 ; CHECK-FIX-LABEL: aesd_once_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q1, q1, q1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q1, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q0, q1
 ; CHECK-FIX-NEXT:    bx lr
   %3 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %1, <16 x i8> %0)
   %4 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %3)
   ret <16 x i8> %4
 }
 
; CHECK-FIX-NEXT:    bx lr
   %9 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %1, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc <16 x i8> @aesd_twice_via_val(<16 x i8> %0, <16 x i8> %1) nounwind {
 ; CHECK-FIX-LABEL: aesd_twice_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q1, q1, q1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q1, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q1
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q0, q8
 ; CHECK-FIX-NEXT:    bx lr
   %3 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %1, <16 x i8> %0)
   %4 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %3)
   %5 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %4, <16 x i8> %0)
6:
   %12 = add nuw i32 %7, 1
   %13 = icmp eq i32 %12, %0
   br i1 %13, label %5, label %6
 }
 
 define arm_aapcs_vfpcc <16 x i8> @aesd_loop_via_val(i32 %0, <16 x i8> %1, <16 x i8> %2) nounwind {
 ; CHECK-FIX-LABEL: aesd_loop_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q1, q1, q1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB37_2
 ; CHECK-FIX-NEXT:  .LBB37_1: @ =>This Inner Loop Header: Depth=1
 ; CHECK-FIX-NEXT:    aesd.8 q1, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesimc.8 q1, q1
 ; CHECK-FIX-NEXT:    bne .LBB37_1
 ; CHECK-FIX-NEXT:  .LBB37_2:
7:
   br i1 %13, label %5, label %7
 }
 
 define arm_aapcs_vfpcc void @aesd_set8_via_ptr(i8* %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aesd_set8_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.8 {d0[0]}, [r0]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = load i8, i8* %0, align 1
   %6 = insertelement <16 x i8> %1, i8 %5, i64 0
   %7 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %4, <16 x i8> %6)
   %8 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %7)
   store <16 x i8> %8, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set8_via_val(i8 zeroext %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aesd_set8_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    vmov.8 d16[0], r0
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = insertelement <16 x i8> %4, i8 %0, i64 0
   %6 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %5, <16 x i8> %1)
   %7 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %6)
   store <16 x i8> %7, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set8_cond_via_ptr(i1 zeroext %0, i8* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set8_cond_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB40_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vld1.8 {d0[0]}, [r1]
 ; CHECK-FIX-NEXT:  .LBB40_2:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   br i1 %0, label %6, label %9
 
 6:
   %7 = load i8, i8* %1, align 1
   %8 = insertelement <16 x i8> %2, i8 %7, i64 0
   br label %9
 
 9:
   %10 = phi <16 x i8> [ %8, %6 ], [ %2, %4 ]
   %11 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %5, <16 x i8> %10)
   %12 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %11)
   store <16 x i8> %12, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set8_cond_via_val(i1 zeroext %0, i8 zeroext %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set8_cond_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB41_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vmov.8 d16[0], r1
 ; CHECK-FIX-NEXT:  .LBB41_2: @ %select.end
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   %6 = insertelement <16 x i8> %5, i8 %1, i64 0
   %7 = select i1 %0, <16 x i8> %6, <16 x i8> %5
   %8 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %7, <16 x i8> %2)
   %9 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set8_loop_via_ptr(i32 %0, i8* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set8_loop_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB42_1:
 ; CHECK-FIX-NEXT:    vld1.8 {d0[0]}, [r1]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB42_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB42_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load i8, i8* %1, align 1

 ; CHECK-FIX-LABEL: aesd_set8_loop_via_val:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB43_1:
 ; CHECK-FIX-NEXT:    vmov.8 d0[0], r1
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB43_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB43_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = icmp eq i32 %0, 0
11:
   br i1 %17, label %9, label %11
 }
 
 define arm_aapcs_vfpcc void @aesd_set16_via_ptr(i16* %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aesd_set16_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.16 {d0[0]}, [r0:16]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = load i16, i16* %0, align 2
   %6 = bitcast <16 x i8> %1 to <8 x i16>
   %7 = insertelement <8 x i16> %6, i16 %5, i64 0
   %8 = bitcast <8 x i16> %7 to <16 x i8>
   %9 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %4, <16 x i8> %8)
   %10 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %9)
   store <16 x i8> %10, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set16_via_val(i16 zeroext %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aesd_set16_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    vmov.16 d16[0], r0
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = bitcast <16 x i8>* %2 to <8 x i16>*
   %5 = load <8 x i16>, <8 x i16>* %4, align 8
   %6 = insertelement <8 x i16> %5, i16 %0, i64 0
   %7 = bitcast <8 x i16> %6 to <16 x i8>
   %8 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %7, <16 x i8> %1)
   %9 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set16_cond_via_ptr(i1 zeroext %0, i16* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set16_cond_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB46_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vld1.16 {d0[0]}, [r1:16]
 ; CHECK-FIX-NEXT:  .LBB46_2:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   br i1 %0, label %6, label %10
 
 6:
12:
   %16 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %15)
   store <16 x i8> %16, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set16_cond_via_val(i1 zeroext %0, i16 zeroext %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set16_cond_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB47_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vmov.16 d16[0], r1
 ; CHECK-FIX-NEXT:  .LBB47_2: @ %select.end
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = bitcast <16 x i8>* %3 to <8 x i16>*
   %6 = load <8 x i16>, <8 x i16>* %5, align 8
   %7 = insertelement <8 x i16> %6, i16 %1, i64 0
   %8 = select i1 %0, <8 x i16> %7, <8 x i16> %6
   %9 = bitcast <8 x i16> %8 to <16 x i8>
   %10 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %9, <16 x i8> %2)
   %11 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %10)
   store <16 x i8> %11, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set16_loop_via_ptr(i32 %0, i16* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set16_loop_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB48_1:
 ; CHECK-FIX-NEXT:    vld1.16 {d0[0]}, [r1:16]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB48_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB48_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load i16, i16* %1, align 2

 ; CHECK-FIX-LABEL: aesd_set16_loop_via_val:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB49_1:
 ; CHECK-FIX-NEXT:    vmov.16 d0[0], r1
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB49_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB49_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = icmp eq i32 %0, 0
13:
   br i1 %19, label %11, label %13
 }
 
 define arm_aapcs_vfpcc void @aesd_set32_via_ptr(i32* %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aesd_set32_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.32 {d0[0]}, [r0:32]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = load i32, i32* %0, align 4
   %6 = bitcast <16 x i8> %1 to <4 x i32>
   %7 = insertelement <4 x i32> %6, i32 %5, i64 0
   %8 = bitcast <4 x i32> %7 to <16 x i8>
   %9 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %4, <16 x i8> %8)
   %10 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %9)
   store <16 x i8> %10, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set32_via_val(i32 %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aesd_set32_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    vmov.32 d16[0], r0
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = bitcast <16 x i8>* %2 to <4 x i32>*
   %5 = load <4 x i32>, <4 x i32>* %4, align 8
   %6 = insertelement <4 x i32> %5, i32 %0, i64 0
   %7 = bitcast <4 x i32> %6 to <16 x i8>
   %8 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %7, <16 x i8> %1)
   %9 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set32_cond_via_ptr(i1 zeroext %0, i32* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set32_cond_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB52_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vld1.32 {d0[0]}, [r1:32]
 ; CHECK-FIX-NEXT:  .LBB52_2:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   br i1 %0, label %6, label %10
 
 6:
12:
   %16 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %15)
   store <16 x i8> %16, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set32_cond_via_val(i1 zeroext %0, i32 %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set32_cond_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    beq .LBB53_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vmov.32 d16[0], r1
 ; CHECK-FIX-NEXT:  .LBB53_2: @ %select.end
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = bitcast <16 x i8>* %3 to <4 x i32>*
   %6 = load <4 x i32>, <4 x i32>* %5, align 8
   %7 = insertelement <4 x i32> %6, i32 %1, i64 0
   %8 = select i1 %0, <4 x i32> %7, <4 x i32> %6
   %9 = bitcast <4 x i32> %8 to <16 x i8>
   %10 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %9, <16 x i8> %2)
   %11 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %10)
   store <16 x i8> %11, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set32_loop_via_ptr(i32 %0, i32* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set32_loop_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB54_1:
 ; CHECK-FIX-NEXT:    vld1.32 {d0[0]}, [r1:32]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB54_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB54_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load i32, i32* %1, align 4

 ; CHECK-FIX-LABEL: aesd_set32_loop_via_val:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB55_1:
 ; CHECK-FIX-NEXT:    vmov.32 d0[0], r1
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB55_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB55_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = icmp eq i32 %0, 0
13:
   %18 = add nuw i32 %15, 1
   %19 = icmp eq i32 %18, %0
   br i1 %19, label %11, label %13
 }
 
 define arm_aapcs_vfpcc void @aesd_set64_via_ptr(i64* %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-NOSCHED-LABEL: aesd_set64_via_ptr:
 ; CHECK-FIX-NOSCHED:       @ %bb.0:
+; CHECK-FIX-NOSCHED-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NOSCHED-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NOSCHED-NEXT:    vldr d0, [r0]
 ; CHECK-FIX-NOSCHED-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NOSCHED-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NOSCHED-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NOSCHED-NEXT:    bx lr
 ;
 ; CHECK-CORTEX-FIX-LABEL: aesd_set64_via_ptr:
 ; CHECK-CORTEX-FIX:       @ %bb.0:
+; CHECK-CORTEX-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-CORTEX-FIX-NEXT:    vldr d0, [r0]
 ; CHECK-CORTEX-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-CORTEX-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-CORTEX-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-CORTEX-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-CORTEX-FIX-NEXT:    bx lr
   %4 = load <16 x i8>, <16 x i8>* %2, align 8
   %5 = load i64, i64* %0, align 8
   %6 = bitcast <16 x i8> %1 to <2 x i64>
   %7 = insertelement <2 x i64> %6, i64 %5, i64 0
   %8 = bitcast <2 x i64> %7 to <16 x i8>
   %9 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %4, <16 x i8> %8)
   %10 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %9)
   store <16 x i8> %10, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set64_via_val(i64 %0, <16 x i8> %1, <16 x i8>* %2) nounwind {
 ; CHECK-FIX-LABEL: aesd_set64_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    vmov.32 d16[0], r0
 ; CHECK-FIX-NEXT:    vmov.32 d16[1], r1
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %4 = bitcast <16 x i8>* %2 to <2 x i64>*
   %5 = load <2 x i64>, <2 x i64>* %4, align 8
   %6 = insertelement <2 x i64> %5, i64 %0, i64 0
   %7 = bitcast <2 x i64> %6 to <16 x i8>
   %8 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %7, <16 x i8> %1)
   %9 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %8)
   store <16 x i8> %9, <16 x i8>* %2, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set64_cond_via_ptr(i1 zeroext %0, i64* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set64_cond_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    vldrne d0, [r1]
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = load <16 x i8>, <16 x i8>* %3, align 8
   br i1 %0, label %6, label %10
 
 6:
12:
   %16 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %15)
   store <16 x i8> %16, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set64_cond_via_val(i1 zeroext %0, i64 %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set64_cond_via_val:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    ldr r1, [sp]
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    beq .LBB59_2
 ; CHECK-FIX-NEXT:  @ %bb.1:
 ; CHECK-FIX-NEXT:    vmov.32 d16[0], r2
 ; CHECK-FIX-NEXT:    vmov.32 d16[1], r3
 ; CHECK-FIX-NEXT:  .LBB59_2: @ %select.end
+; CHECK-FIX-NEXT:    vorr q8, q8, q8
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = bitcast <16 x i8>* %3 to <2 x i64>*
   %6 = load <2 x i64>, <2 x i64>* %5, align 8
   %7 = insertelement <2 x i64> %6, i64 %1, i64 0
   %8 = select i1 %0, <2 x i64> %7, <2 x i64> %6
   %9 = bitcast <2 x i64> %8 to <16 x i8>
   %10 = call <16 x i8> @llvm.arm.neon.aesd(<16 x i8> %9, <16 x i8> %2)
   %11 = call <16 x i8> @llvm.arm.neon.aesimc(<16 x i8> %10)
   store <16 x i8> %11, <16 x i8>* %3, align 8
   ret void
 }
 
 define arm_aapcs_vfpcc void @aesd_set64_loop_via_ptr(i32 %0, i64* %1, <16 x i8> %2, <16 x i8>* %3) nounwind {
 ; CHECK-FIX-LABEL: aesd_set64_loop_via_ptr:
 ; CHECK-FIX:       @ %bb.0:
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB60_1:
 ; CHECK-FIX-NEXT:    vldr d0, [r1]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r2]
 ; CHECK-FIX-NEXT:  .LBB60_2: @ =>This Inner Loop Header: Depth=1
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1

 ; CHECK-FIX-NEXT:    cmp r0, #0
 ; CHECK-FIX-NEXT:    bxeq lr
 ; CHECK-FIX-NEXT:  .LBB61_1:
 ; CHECK-FIX-NEXT:    vmov.32 d0[0], r2
 ; CHECK-FIX-NEXT:    ldr r1, [sp]
 ; CHECK-FIX-NEXT:    vld1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    vmov.32 d0[1], r3
 ; CHECK-FIX-NEXT:  .LBB61_2: @ =>This Inner Loop Header: Depth=1
+; CHECK-FIX-NEXT:    vorr q0, q0, q0
 ; CHECK-FIX-NEXT:    aesd.8 q8, q0
 ; CHECK-FIX-NEXT:    subs r0, r0, #1
 ; CHECK-FIX-NEXT:    aesimc.8 q8, q8
 ; CHECK-FIX-NEXT:    bne .LBB61_2
 ; CHECK-FIX-NEXT:  @ %bb.3:
 ; CHECK-FIX-NEXT:    vst1.64 {d16, d17}, [r1]
 ; CHECK-FIX-NEXT:    bx lr
   %5 = icmp eq i32 %0, 0