[ARM] Enabling range checks on Neon intrinsics' lane arguments

[pratlucas]

Summary

Range checks were not properly performed in the lane arguments of Neon
intrinsics implemented based on splat operations. Calls to those
intrinsics where translated to __builtin__shufflevector calls directly
by the pre-processor through the arm_neon.h macros, missing the chance
for the proper range checks.

This patch enables the range check by introducing an auxiliary splat
instruction in arm_neon.td, delaying the translation to shufflevector
calls to CGBuiltin.cpp in clang after the checks were performed.

[pratlucas]

Adding test for the intrinsics range checks.

[pratlucas]

Merging content from D74617 into this revision.

[dnsampaio]

clang/test/CodeGen/arm-neon-range-checks.c

8

Could we have the valid range in the tests?

[pratlucas]

Adding check for valid range on tests.

[pratlucas]

Clang-format.

[pratlucas]

Fixing missing format issue.

[pratlucas]

The clang-format pre-merge check keeps wanting me to update the indentation for the entire ARMSIMDIntrinsicMap and AArch64SIMDIntrinsicMap maps due to the change in their first entries.
I believe, though, that this change would not only be out of the scope of this patch, but would also bring inconsistencies with the current indentation style of the entire file.

[dnsampaio]

Hi,
thanks for looking into this. The patch LGTM, but regarding the indentation, I don't know what would be the best practice here. We tend to like to preserve the line-git-history, but if we start ignoring the formater check, then it has no sense in they being here.
Perhaps @t.p.northover or @olista01 could share their thoughts here.

[ostannard]

I agree with @dnsampaio here, it's better to match the existing style, and avoid irrelevant churn in the git history.

[leonardchan]

Hi, a bisect seems to show https://reviews.llvm.org/rGf56550cf7f12b581a237b48a7f4d8b6682d45a09 is causing us to see the following error:

error: argument value 1 is outside the valid range [0, 0]
v2 = vdupq_lane_f64(vget_high_f64(a.v), 1);
^                                  ~
../../prebuilt/third_party/clang/linux-x64/lib/clang/11.0.0/include/arm_neon.h:46503:15: note: expanded from macro 'vdupq_lane_f64'
__ret_307 = splatq_lane_f64(__s0_307, __p1_307); \
^                         ~~~~~~~~
../../prebuilt/third_party/clang/linux-x64/lib/clang/11.0.0/include/arm_neon.h:680:25: note: expanded from macro 'splatq_lane_f64'
__ret = (float64x2_t) __builtin_neon_splatq_lane_v((int8x8_t)__s0, __p1, 10); \
^                                            ~~~~
1 error generated.

when building Fuchsia.

I'll see if I can make a minimal reproducer. Just wanted to raise awareness in the meantime to see if you can take a look.

[leonardchan]

Actually this may not need a reproducer since the error seems straightforward with just calling a builtin function. The main issue is that before this patch, something like

#include <arm_neon.h>
float64x2_t func(int8x8_t x) {
  return __builtin_neon_splatq_lane_v(x, 1, 10);
}

would've compiled, but we get the error with this patch. I see that the intent of this patch is to do range checking, but does it make sense for the range shown in the error to be [0, 0]?

[pratlucas]

Hi @leonardchan ,

I've double-checked the Neon intrinsics reference and it indeed confirms that the only allowed value for the lane argument for vdupq_lane_f64 is 0:

Argument Preparation

vec → Vn.1D 
0 << lane << 0

(https://developer.arm.com/architectures/instruction-sets/simd-isas/neon/intrinsics?search=vdupq_lane_f64)

I believe the observed behaviour is expected.

[leonardchan]

In D74619#1936336, @pratlucas wrote:

Hi @leonardchan ,

I've double-checked the Neon intrinsics reference and it indeed confirms that the only allowed value for the lane argument for vdupq_lane_f64 is 0:

Argument Preparation

vec → Vn.1D 
0 << lane << 0

(https://developer.arm.com/architectures/instruction-sets/simd-isas/neon/intrinsics?search=vdupq_lane_f64)

I believe the observed behaviour is expected.

Thanks for the update. Able to confirm this was an "us" issue, not a clang one.

clang/include/clang/Basic/arm_neon.td

 def OP_MUL_N    : Op<(op "*", $p0, (dup $p1))>;
 def OP_MULX_N   : Op<(call "vmulx", $p0, (dup $p1))>;
 def OP_MLA_N    : Op<(op "+", $p0, (op "*", $p1, (dup $p2)))>;
 def OP_MLS_N    : Op<(op "-", $p0, (op "*", $p1, (dup $p2)))>;
 def OP_FMLA_N   : Op<(call "vfma", $p0, $p1, (dup $p2))>;
 def OP_FMLS_N   : Op<(call "vfma", $p0, (op "-", $p1), (dup $p2))>;
 def OP_MLAL_N   : Op<(op "+", $p0, (call "vmull", $p1, (dup $p2)))>;
 def OP_MLSL_N   : Op<(op "-", $p0, (call "vmull", $p1, (dup $p2)))>;
-def OP_MUL_LN   : Op<(op "*", $p0, (splat $p1, $p2))>;
-def OP_MULX_LN  : Op<(call "vmulx", $p0, (splat $p1, $p2))>;
+def OP_MUL_LN   : Op<(op "*", $p0, (call_mangled "splat_lane", $p1, $p2))>;
+def OP_MULX_LN  : Op<(call "vmulx", $p0, (call_mangled "splat_lane", $p1, $p2))>;
 def OP_MULL_N  : Op<(call "vmull", $p0, (dup $p1))>;
-def OP_MULL_LN  : Op<(call "vmull", $p0, (splat $p1, $p2))>;
-def OP_MULLHi_LN: Op<(call "vmull", (call "vget_high", $p0), (splat $p1, $p2))>;
-def OP_MLA_LN   : Op<(op "+", $p0, (op "*", $p1, (splat $p2, $p3)))>;
-def OP_MLS_LN   : Op<(op "-", $p0, (op "*", $p1, (splat $p2, $p3)))>;
-def OP_MLAL_LN  : Op<(op "+", $p0, (call "vmull", $p1, (splat $p2, $p3)))>;
+def OP_MULL_LN  : Op<(call "vmull", $p0, (call_mangled "splat_lane", $p1, $p2))>;
+def OP_MULLHi_LN: Op<(call "vmull", (call "vget_high", $p0), (call_mangled "splat_lane", $p1, $p2))>;
+def OP_MLA_LN   : Op<(op "+", $p0, (op "*", $p1, (call_mangled "splat_lane", $p2, $p3)))>;
+def OP_MLS_LN   : Op<(op "-", $p0, (op "*", $p1, (call_mangled "splat_lane", $p2, $p3)))>;
+def OP_MLAL_LN  : Op<(op "+", $p0, (call "vmull", $p1, (call_mangled "splat_lane", $p2, $p3)))>;
 def OP_MLALHi_LN: Op<(op "+", $p0, (call "vmull", (call "vget_high", $p1),
-                                                  (splat $p2, $p3)))>;
-def OP_MLSL_LN  : Op<(op "-", $p0, (call "vmull", $p1, (splat $p2, $p3)))>;
+                                                  (call_mangled "splat_lane", $p2, $p3)))>;
+def OP_MLSL_LN  : Op<(op "-", $p0, (call "vmull", $p1, (call_mangled "splat_lane", $p2, $p3)))>;
 def OP_MLSLHi_LN : Op<(op "-", $p0, (call "vmull", (call "vget_high", $p1),
-                                                   (splat $p2, $p3)))>;
+                                                   (call_mangled "splat_lane", $p2, $p3)))>;
 def OP_QDMULL_N : Op<(call "vqdmull", $p0, (dup $p1))>;
-def OP_QDMULL_LN : Op<(call "vqdmull", $p0, (splat $p1, $p2))>;
+def OP_QDMULL_LN : Op<(call "vqdmull", $p0, (call_mangled "splat_lane", $p1, $p2))>;
 def OP_QDMULLHi_LN : Op<(call "vqdmull", (call "vget_high", $p0),
-                                         (splat $p1, $p2))>;
+                                         (call_mangled "splat_lane", $p1, $p2))>;
 def OP_QDMLAL_N : Op<(call "vqdmlal", $p0, $p1, (dup $p2))>;
-def OP_QDMLAL_LN : Op<(call "vqdmlal", $p0, $p1, (splat $p2, $p3))>;
+def OP_QDMLAL_LN : Op<(call "vqdmlal", $p0, $p1, (call_mangled "splat_lane", $p2, $p3))>;
 def OP_QDMLALHi_LN : Op<(call "vqdmlal", $p0, (call "vget_high", $p1),
-                                              (splat $p2, $p3))>;
+                                              (call_mangled "splat_lane", $p2, $p3))>;
 def OP_QDMLSL_N : Op<(call "vqdmlsl", $p0, $p1, (dup $p2))>;
-def OP_QDMLSL_LN : Op<(call "vqdmlsl", $p0, $p1, (splat $p2, $p3))>;
+def OP_QDMLSL_LN : Op<(call "vqdmlsl", $p0, $p1, (call_mangled "splat_lane", $p2, $p3))>;
 def OP_QDMLSLHi_LN : Op<(call "vqdmlsl", $p0, (call "vget_high", $p1),
-                                              (splat $p2, $p3))>;
+                                              (call_mangled "splat_lane", $p2, $p3))>;
 def OP_QDMULH_N : Op<(call "vqdmulh", $p0, (dup $p1))>;
-def OP_QDMULH_LN : Op<(call "vqdmulh", $p0, (splat $p1, $p2))>;
-def OP_QRDMULH_LN : Op<(call "vqrdmulh", $p0, (splat $p1, $p2))>;
+def OP_QDMULH_LN : Op<(call "vqdmulh", $p0, (call_mangled "splat_lane", $p1, $p2))>;
+def OP_QRDMULH_LN : Op<(call "vqrdmulh", $p0, (call_mangled "splat_lane", $p1, $p2))>;
 def OP_QRDMULH_N : Op<(call "vqrdmulh", $p0, (dup $p1))>;
 def OP_QRDMLAH : Op<(call "vqadd", $p0, (call "vqrdmulh", $p1, $p2))>;
 def OP_QRDMLSH : Op<(call "vqsub", $p0, (call "vqrdmulh", $p1, $p2))>;
-def OP_QRDMLAH_LN : Op<(call "vqadd", $p0, (call "vqrdmulh", $p1, (splat $p2, $p3)))>;
-def OP_QRDMLSH_LN : Op<(call "vqsub", $p0, (call "vqrdmulh", $p1, (splat $p2, $p3)))>;
+def OP_QRDMLAH_LN : Op<(call "vqadd", $p0, (call "vqrdmulh", $p1, (call_mangled "splat_lane", $p2, $p3)))>;
+def OP_QRDMLSH_LN : Op<(call "vqsub", $p0, (call "vqrdmulh", $p1, (call_mangled "splat_lane", $p2, $p3)))>;
 def OP_FMS_LN   : Op<(call "vfma_lane", $p0, (op "-", $p1), $p2, $p3)>;
 def OP_FMS_LNQ  : Op<(call "vfma_laneq", $p0, (op "-", $p1), $p2, $p3)>;
 def OP_TRN1     : Op<(shuffle $p0, $p1, (interleave (decimate mask0, 2),
                                                     (decimate mask1, 2)))>;
 def OP_ZIP1     : Op<(shuffle $p0, $p1, (lowhalf (interleave mask0, mask1)))>;
 def OP_UZP1     : Op<(shuffle $p0, $p1, (add (decimate mask0, 2),
                                              (decimate mask1, 2)))>;
 def OP_TRN2     : Op<(shuffle $p0, $p1, (interleave

 def OP_ANDN     : Op<(op "&", $p0, (op "~", $p1))>;
 def OP_ORN      : Op<(op "|", $p0, (op "~", $p1))>;
 def OP_CAST     : LOp<[(save_temp $promote, $p0),
                        (cast "R", $promote)]>;
 def OP_HI       : Op<(shuffle $p0, $p0, (highhalf mask0))>;
 def OP_LO       : Op<(shuffle $p0, $p0, (lowhalf mask0))>;
 def OP_CONC     : Op<(shuffle $p0, $p1, (add mask0, mask1))>;
 def OP_DUP      : Op<(dup $p0)>;
-def OP_DUP_LN   : Op<(splat $p0, $p1)>;
+def OP_DUP_LN   : Op<(call_mangled "splat_lane", $p0, $p1)>;
 def OP_SEL      : Op<(cast "R", (op "|",
                                     (op "&", $p0, (cast $p0, $p1)),
                                     (op "&", (op "~", $p0), (cast $p0, $p2))))>;
 def OP_REV16    : Op<(shuffle $p0, $p0, (rev 16, mask0))>;
 def OP_REV32    : Op<(shuffle $p0, $p0, (rev 32, mask0))>;
 def OP_REV64    : Op<(shuffle $p0, $p0, (rev 64, mask0))>;
 def OP_XTN      : Op<(call "vcombine", $p0, (call "vmovn", $p1))>;
 def OP_SQXTUN   : Op<(call "vcombine", (cast $p0, "U", $p0),
def OP_SCALAR_HALF_SET_LN : Op<(bitcast "float16x4_t",
                                          (bitcast "int16x4_t", $p1), $p2))>;
 def OP_SCALAR_HALF_SET_LNQ : Op<(bitcast "float16x8_t",
                                     (call "vset_lane",
                                           (bitcast "int16_t", $p0),
                                           (bitcast "int16x8_t", $p1), $p2))>;
 
 def OP_DOT_LN
     : Op<(call "vdot", $p0, $p1,
-          (bitcast $p1, (splat(bitcast "uint32x2_t", $p2), $p3)))>;
+          (bitcast $p1, (call_mangled "splat_lane", (bitcast "32", $p2), $p3)))>;
 def OP_DOT_LNQ
     : Op<(call "vdot", $p0, $p1,
-          (bitcast $p1, (splat(bitcast "uint32x4_t", $p2), $p3)))>;
+          (bitcast $p1, (call_mangled "splat_lane", (bitcast "32", $p2), $p3)))>;
 
 def OP_FMLAL_LN     : Op<(call "vfmlal_low", $p0, $p1,
                            (dup_typed $p1, (call "vget_lane", $p2, $p3)))>;
 def OP_FMLSL_LN     : Op<(call "vfmlsl_low", $p0, $p1,
                            (dup_typed $p1, (call "vget_lane", $p2, $p3)))>;
 def OP_FMLAL_LN_Hi  : Op<(call "vfmlal_high", $p0, $p1,
                            (dup_typed $p1, (call "vget_lane", $p2, $p3)))>;
 def OP_FMLSL_LN_Hi  : Op<(call "vfmlsl_high", $p0, $p1,
                            (dup_typed $p1, (call "vget_lane", $p2, $p3)))>;
 
 //===----------------------------------------------------------------------===//
-// Instructions
+// Auxiliary Instructions
+//===----------------------------------------------------------------------===//
+
+// Splat operation - performs a range-checked splat over a vector
+def SPLAT  : WInst<"splat_lane", ".(!q)I",
+                   "UcUsUicsilPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUlhdQhQdPlQPl">;
+def SPLATQ : WInst<"splat_laneq", ".(!Q)I",
+                   "UcUsUicsilPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUlhdQhQdPlQPl"> {
+  let isLaneQ = 1;
+}
+
+//===----------------------------------------------------------------------===//
+// Intrinsics
 //===----------------------------------------------------------------------===//
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.1 Addition
 def VADD    : IOpInst<"vadd", "...",
                       "csilfUcUsUiUlQcQsQiQlQfQUcQUsQUiQUl", OP_ADD>;
 def VADDL   : SOpInst<"vaddl", "(>Q)..", "csiUcUsUi", OP_ADDL>;
 def VADDW   : SOpInst<"vaddw", "(>Q)(>Q).", "csiUcUsUi", OP_ADDW>;
let ArchGuard = "defined(__ARM_FEATURE_COMPLEX)" in {
   def VCADD_ROT270  : SInst<"vcadd_rot270", "...", "f">;
   def VCADDQ_ROT90  : SInst<"vcaddq_rot90", "QQQ", "f">;
   def VCADDQ_ROT270 : SInst<"vcaddq_rot270", "QQQ", "f">;
 }
 let ArchGuard = "defined(__ARM_FEATURE_COMPLEX) && defined(__aarch64__)" in {
   def VCADDQ_ROT90_FP64  : SInst<"vcaddq_rot90", "QQQ", "d">;
   def VCADDQ_ROT270_FP64 : SInst<"vcaddq_rot270", "QQQ", "d">;
 }
 No newline at end of file

clang/include/clang/Basic/arm_neon_incl.td

 //          - A typedef string - A NEON or stdint.h type that is then parsed.
 //                               for example: (cast "uint32x4_t", $p0).
 //          - "U" - Make the type unsigned.
 //          - "S" - Make the type signed.
 //          - "H" - Halve the number of lanes in the type.
 //          - "D" - Double the number of lanes in the type.
 //          - "8" - Convert type to an equivalent vector of 8-bit signed
 //                  integers.
+//          - "32" - Convert type to an equivalent vector of 32-bit integers.
 // example: (cast "R", "U", $p0) -> "(uint32x4_t)__p0" (assuming the return
 //           value is of type "int32x4_t".
 //          (cast $p0, "D", "8", $p1) -> "(int8x16_t)__p1" (assuming __p0
 //           has type float64x1_t or any other vector type of 64 bits).
 //          (cast "int32_t", $p2) -> "(int32_t)__p2"
 def cast;
 // bitcast - Same as "cast", except a reinterpret-cast is produced:
 //             (bitcast "T", $p0) -> "*(T*)&__p0".
 //           The VAL argument is saved to a temporary so it can be used
 //           as an l-value.
 def bitcast;
 // dup - Take a scalar argument and create a vector by duplicating it into
 //       all lanes. The type of the vector is the base type of the intrinsic.
 // example: (dup $p1) -> "(uint32x2_t) {__p1, __p1}" (assuming the base type
 //          is uint32x2_t).
 def dup;
 // dup_typed - Take a vector and a scalar argument, and create a new vector of
 //             the same type by duplicating the scalar value into all lanes.
 // example: (dup_typed $p1, $p2) -> "(float16x4_t) {__p2, __p2, __p2, __p2}"
 //          (assuming __p1 is float16x4_t, and __p2 is a compatible scalar).
 def dup_typed;
-// splat - Take a vector and a lane index, and return a vector of the same type
-//         containing repeated instances of the source vector at the lane index.
-// example: (splat $p0, $p1) ->
-//            "__builtin_shufflevector(__p0, __p0, __p1, __p1, __p1, __p1)"
-//          (assuming __p0 has four elements).
-def splat;
 // save_temp - Create a temporary (local) variable. The variable takes a name
 //             based on the zero'th parameter and can be referenced using
 //             using that name in subsequent DAGs in the same
 //             operation. The scope of a temp is the operation. If a variable
 //             with the given name already exists, an error will be given at
 //             tblgen time.
 // example: [(save_temp $var, (call "foo", $p0)),
 //           (op "+", $var, $p1)] ->

clang/lib/CodeGen/CGBuiltin.cpp

case NeonTypeFlags::Int32:
     return llvm::VectorType::get(CGF->FloatTy, (2 << IsQuad));
   case NeonTypeFlags::Int64:
     return llvm::VectorType::get(CGF->DoubleTy, (1 << IsQuad));
   default:
     llvm_unreachable("Type can't be converted to floating-point!");
   }
 }
 
+Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C,
+                                      unsigned NumElements) {
+  Value* SV = llvm::ConstantVector::getSplat(NumElements, C);
+  return Builder.CreateShuffleVector(V, V, SV, "lane");
+}
+
 Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) {
   unsigned nElts = V->getType()->getVectorNumElements();
-  Value* SV = llvm::ConstantVector::getSplat(nElts, C);
-  return Builder.CreateShuffleVector(V, V, SV, "lane");
+  return EmitNeonSplat(V, C, nElts);
 }
 
 Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops,
                                      const char *name,
                                      unsigned shift, bool rightshift) {
   unsigned j = 0;
   for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
        ai != ae; ++ai, ++j)
#define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \
   { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \
       Intrinsic::LLVMIntrinsic, 0, TypeModifier }
 
 #define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \
   { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \
       Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \
       TypeModifier }
 
 static const NeonIntrinsicInfo ARMSIMDIntrinsicMap [] = {
+  NEONMAP0(splat_lane_v),
+  NEONMAP0(splat_laneq_v),
+  NEONMAP0(splatq_lane_v),
+  NEONMAP0(splatq_laneq_v),
   NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts),
   NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts),
   NEONMAP1(vabs_v, arm_neon_vabs, 0),
   NEONMAP1(vabsq_v, arm_neon_vabs, 0),
   NEONMAP0(vaddhn_v),
   NEONMAP1(vaesdq_v, arm_neon_aesd, 0),
   NEONMAP1(vaeseq_v, arm_neon_aese, 0),
   NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0),
static const NeonIntrinsicInfo ARMSIMDIntrinsicMap [] = {
   NEONMAP0(vtstq_v),
   NEONMAP0(vuzp_v),
   NEONMAP0(vuzpq_v),
   NEONMAP0(vzip_v),
   NEONMAP0(vzipq_v)
 };
 
 static const NeonIntrinsicInfo AArch64SIMDIntrinsicMap[] = {
+  NEONMAP0(splat_lane_v),
+  NEONMAP0(splat_laneq_v),
+  NEONMAP0(splatq_lane_v),
+  NEONMAP0(splatq_laneq_v),
   NEONMAP1(vabs_v, aarch64_neon_abs, 0),
   NEONMAP1(vabsq_v, aarch64_neon_abs, 0),
   NEONMAP0(vaddhn_v),
   NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0),
   NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0),
   NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0),
   NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0),
   NEONMAP1(vcadd_rot270_v, aarch64_neon_vcadd_rot270, Add1ArgType),
Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
   };
 
   unsigned Int = LLVMIntrinsic;
   if ((Modifier & UnsignedAlts) && !Usgn)
     Int = AltLLVMIntrinsic;
 
   switch (BuiltinID) {
   default: break;
+  case NEON::BI__builtin_neon_splat_lane_v:
+  case NEON::BI__builtin_neon_splat_laneq_v:
+  case NEON::BI__builtin_neon_splatq_lane_v:
+  case NEON::BI__builtin_neon_splatq_laneq_v: {
+    auto NumElements = VTy->getNumElements();
+    if (BuiltinID == NEON::BI__builtin_neon_splatq_lane_v)
+      NumElements *= 2;
+    if (BuiltinID == NEON::BI__builtin_neon_splat_laneq_v)
+      NumElements /= 2;
+
+    Ops[0] = Builder.CreateBitCast(Ops[0], VTy);
+    return EmitNeonSplat(Ops[0], cast<ConstantInt>(Ops[1]), NumElements);
+  }
   case NEON::BI__builtin_neon_vpadd_v:
   case NEON::BI__builtin_neon_vpaddq_v:
     // We don't allow fp/int overloading of intrinsics.
     if (VTy->getElementType()->isFloatingPointTy() &&
         Int == Intrinsic::aarch64_neon_addp)
       Int = Intrinsic::aarch64_neon_faddp;
     break;
   case NEON::BI__builtin_neon_vabs_v:

clang/lib/CodeGen/CodeGenFunction.h

public:
 
   llvm::Function *LookupNeonLLVMIntrinsic(unsigned IntrinsicID,
                                           unsigned Modifier, llvm::Type *ArgTy,
                                           const CallExpr *E);
   llvm::Value *EmitNeonCall(llvm::Function *F,
                             SmallVectorImpl<llvm::Value*> &O,
                             const char *name,
                             unsigned shift = 0, bool rightshift = false);
+  llvm::Value *EmitNeonSplat(llvm::Value *V, llvm::Constant *Idx,
+                             unsigned NumElements);
   llvm::Value *EmitNeonSplat(llvm::Value *V, llvm::Constant *Idx);
   llvm::Value *EmitNeonShiftVector(llvm::Value *V, llvm::Type *Ty,
                                    bool negateForRightShift);
   llvm::Value *EmitNeonRShiftImm(llvm::Value *Vec, llvm::Value *Amt,
                                  llvm::Type *Ty, bool usgn, const char *name);
   llvm::Value *vectorWrapScalar16(llvm::Value *Op);
   llvm::Value *EmitAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
                                       llvm::Triple::ArchType Arch);

clang/test/CodeGen/aarch64-neon-2velem.c

 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
 // RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -emit-llvm -o - %s | opt -S -mem2reg | FileCheck %s
 
 // Test new aarch64 intrinsics and types
 
 #include <arm_neon.h>
 
 // CHECK-LABEL: @test_vmla_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[ADD]]
 //
 int16x4_t test_vmla_lane_s16(int16x4_t a, int16x4_t b, int16x4_t v) {
   return vmla_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[ADD]]
 //
 int16x8_t test_vmlaq_lane_s16(int16x8_t a, int16x8_t b, int16x4_t v) {
   return vmlaq_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmla_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[ADD]]
 //
 int32x2_t test_vmla_lane_s32(int32x2_t a, int32x2_t b, int32x2_t v) {
   return vmla_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlaq_lane_s32(int32x4_t a, int32x4_t b, int32x2_t v) {
   return vmlaq_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmla_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[ADD]]
 //
 int16x4_t test_vmla_laneq_s16(int16x4_t a, int16x4_t b, int16x8_t v) {
   return vmla_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlaq_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[ADD]]
 //
 int16x8_t test_vmlaq_laneq_s16(int16x8_t a, int16x8_t b, int16x8_t v) {
   return vmlaq_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmla_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[ADD]]
 //
 int32x2_t test_vmla_laneq_s32(int32x2_t a, int32x2_t b, int32x4_t v) {
   return vmla_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlaq_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlaq_laneq_s32(int32x4_t a, int32x4_t b, int32x4_t v) {
   return vmlaq_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmls_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[SUB]]
 //
 int16x4_t test_vmls_lane_s16(int16x4_t a, int16x4_t b, int16x4_t v) {
   return vmls_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[SUB]]
 //
 int16x8_t test_vmlsq_lane_s16(int16x8_t a, int16x8_t b, int16x4_t v) {
   return vmlsq_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmls_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[SUB]]
 //
 int32x2_t test_vmls_lane_s32(int32x2_t a, int32x2_t b, int32x2_t v) {
   return vmls_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsq_lane_s32(int32x4_t a, int32x4_t b, int32x2_t v) {
   return vmlsq_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmls_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[SUB]]
 //
 int16x4_t test_vmls_laneq_s16(int16x4_t a, int16x4_t b, int16x8_t v) {
   return vmls_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlsq_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[SUB]]
 //
 int16x8_t test_vmlsq_laneq_s16(int16x8_t a, int16x8_t b, int16x8_t v) {
   return vmlsq_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmls_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[SUB]]
 //
 int32x2_t test_vmls_laneq_s32(int32x2_t a, int32x2_t b, int32x4_t v) {
   return vmls_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsq_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsq_laneq_s32(int32x4_t a, int32x4_t b, int32x4_t v) {
   return vmlsq_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmul_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i16> [[MUL]]
 //
 int16x4_t test_vmul_lane_s16(int16x4_t a, int16x4_t v) {
   return vmul_lane_s16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <8 x i16> [[MUL]]
 //
 int16x8_t test_vmulq_lane_s16(int16x8_t a, int16x4_t v) {
   return vmulq_lane_s16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmul_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x i32> [[MUL]]
 //
 int32x2_t test_vmul_lane_s32(int32x2_t a, int32x2_t v) {
   return vmul_lane_s32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i32> [[MUL]]
 //
 int32x4_t test_vmulq_lane_s32(int32x4_t a, int32x2_t v) {
   return vmulq_lane_s32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmul_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i16> [[MUL]]
 //
 uint16x4_t test_vmul_lane_u16(uint16x4_t a, uint16x4_t v) {
   return vmul_lane_u16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <8 x i16> [[MUL]]
 //
 uint16x8_t test_vmulq_lane_u16(uint16x8_t a, uint16x4_t v) {
   return vmulq_lane_u16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmul_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x i32> [[MUL]]
 //
 uint32x2_t test_vmul_lane_u32(uint32x2_t a, uint32x2_t v) {
   return vmul_lane_u32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i32> [[MUL]]
 //
 uint32x4_t test_vmulq_lane_u32(uint32x4_t a, uint32x2_t v) {
   return vmulq_lane_u32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i16> [[MUL]]
 //
 int16x4_t test_vmul_laneq_s16(int16x4_t a, int16x8_t v) {
   return vmul_laneq_s16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <8 x i16> [[MUL]]
 //
 int16x8_t test_vmulq_laneq_s16(int16x8_t a, int16x8_t v) {
   return vmulq_laneq_s16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x i32> [[MUL]]
 //
 int32x2_t test_vmul_laneq_s32(int32x2_t a, int32x4_t v) {
   return vmul_laneq_s32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i32> [[MUL]]
 //
 int32x4_t test_vmulq_laneq_s32(int32x4_t a, int32x4_t v) {
   return vmulq_laneq_s32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i16> [[MUL]]
 //
 uint16x4_t test_vmul_laneq_u16(uint16x4_t a, uint16x8_t v) {
   return vmul_laneq_u16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <8 x i16> [[MUL]]
 //
 uint16x8_t test_vmulq_laneq_u16(uint16x8_t a, uint16x8_t v) {
   return vmulq_laneq_u16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x i32> [[MUL]]
 //
 uint32x2_t test_vmul_laneq_u32(uint32x2_t a, uint32x4_t v) {
   return vmul_laneq_u32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i32> [[MUL]]
 //
 uint32x4_t test_vmulq_laneq_u32(uint32x4_t a, uint32x4_t v) {
   return vmulq_laneq_u32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vfma_lane_f32(
 // CHECK-NEXT:  entry:

 // CHECK-NEXT:    ret double [[TMP0]]
 //
 float64_t test_vfmsd_laneq_f64(float64_t a, float64_t b, float64x2_t v) {
   return vfmsd_laneq_f64(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_lane_s16(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vmlal_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_lane_s32(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vmlal_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlal_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_laneq_s16(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vmlal_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlal_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_laneq_s32(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vmlal_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_high_lane_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_high_lane_s16(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vmlal_high_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_high_lane_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_high_lane_s32(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vmlal_high_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlal_high_laneq_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_high_laneq_s16(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vmlal_high_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlal_high_laneq_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_high_laneq_s32(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vmlal_high_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_lane_s16(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vmlsl_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_lane_s32(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vmlsl_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlsl_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_laneq_s16(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vmlsl_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlsl_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_laneq_s32(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vmlsl_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_lane_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_high_lane_s16(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vmlsl_high_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_lane_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_high_lane_s32(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vmlsl_high_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_laneq_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_high_laneq_s16(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vmlsl_high_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_laneq_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_high_laneq_s32(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vmlsl_high_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_lane_u16(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vmlal_lane_u16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_lane_u32(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vmlal_lane_u32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlal_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_laneq_u16(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vmlal_laneq_u16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlal_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_laneq_u32(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vmlal_laneq_u32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_high_lane_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_high_lane_u16(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vmlal_high_lane_u16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_high_lane_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_high_lane_u32(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vmlal_high_lane_u32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlal_high_laneq_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_high_laneq_u16(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vmlal_high_laneq_u16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlal_high_laneq_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_high_laneq_u32(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vmlal_high_laneq_u32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_lane_u16(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vmlsl_lane_u16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_lane_u32(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vmlsl_lane_u32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlsl_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_laneq_u16(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vmlsl_laneq_u16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlsl_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_laneq_u32(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vmlsl_laneq_u32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_lane_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_high_lane_u16(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vmlsl_high_lane_u16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_lane_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_high_lane_u32(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vmlsl_high_lane_u32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_laneq_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_high_laneq_u16(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vmlsl_high_laneq_u16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_laneq_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_high_laneq_u32(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vmlsl_high_laneq_u32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmull_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 int32x4_t test_vmull_lane_s16(int16x4_t a, int16x4_t v) {
   return vmull_lane_s16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmull_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 int64x2_t test_vmull_lane_s32(int32x2_t a, int32x2_t v) {
   return vmull_lane_s32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmull_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 uint32x4_t test_vmull_lane_u16(uint16x4_t a, uint16x4_t v) {
   return vmull_lane_u16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmull_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 uint64x2_t test_vmull_lane_u32(uint32x2_t a, uint32x2_t v) {
   return vmull_lane_u32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmull_high_lane_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 int32x4_t test_vmull_high_lane_s16(int16x8_t a, int16x4_t v) {
   return vmull_high_lane_s16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmull_high_lane_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 int64x2_t test_vmull_high_lane_s32(int32x4_t a, int32x2_t v) {
   return vmull_high_lane_s32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmull_high_lane_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 uint32x4_t test_vmull_high_lane_u16(uint16x8_t a, uint16x4_t v) {
   return vmull_high_lane_u16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmull_high_lane_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 uint64x2_t test_vmull_high_lane_u32(uint32x4_t a, uint32x2_t v) {
   return vmull_high_lane_u32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmull_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 int32x4_t test_vmull_laneq_s16(int16x4_t a, int16x8_t v) {
   return vmull_laneq_s16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vmull_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 int64x2_t test_vmull_laneq_s32(int32x2_t a, int32x4_t v) {
   return vmull_laneq_s32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmull_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 uint32x4_t test_vmull_laneq_u16(uint16x4_t a, uint16x8_t v) {
   return vmull_laneq_u16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vmull_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 uint64x2_t test_vmull_laneq_u32(uint32x2_t a, uint32x4_t v) {
   return vmull_laneq_u32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmull_high_laneq_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 int32x4_t test_vmull_high_laneq_s16(int16x8_t a, int16x8_t v) {
   return vmull_high_laneq_s16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vmull_high_laneq_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 int64x2_t test_vmull_high_laneq_s32(int32x4_t a, int32x4_t v) {
   return vmull_high_laneq_s32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmull_high_laneq_u16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 uint32x4_t test_vmull_high_laneq_u16(uint16x8_t a, uint16x8_t v) {
   return vmull_high_laneq_u16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vmull_high_laneq_u32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 uint64x2_t test_vmull_high_laneq_u32(uint32x4_t a, uint32x4_t v) {
   return vmull_high_laneq_u32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlal_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLAL_V3_I]]
 //
 int32x4_t test_vqdmlal_lane_s16(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vqdmlal_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlal_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqadd.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLAL_V3_I]]
 //
 int64x2_t test_vqdmlal_lane_s32(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vqdmlal_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vqdmlal_high_lane_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLAL_V3_I]]
 //
 int32x4_t test_vqdmlal_high_lane_s16(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vqdmlal_high_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlal_high_lane_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqadd.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLAL_V3_I]]
 //
 int64x2_t test_vqdmlal_high_lane_s32(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vqdmlal_high_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLSL_V3_I]]
 //
 int32x4_t test_vqdmlsl_lane_s16(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vqdmlsl_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I]]
 //
 int64x2_t test_vqdmlsl_lane_s32(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vqdmlsl_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_high_lane_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLSL_V3_I]]
 //
 int32x4_t test_vqdmlsl_high_lane_s16(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vqdmlsl_high_lane_s16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_high_lane_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I]]
 //
 int64x2_t test_vqdmlsl_high_lane_s32(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vqdmlsl_high_lane_s32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vqdmull_lane_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <4 x i32> [[VQDMULL_V2_I]]
 //
 int32x4_t test_vqdmull_lane_s16(int16x4_t a, int16x4_t v) {
   return vqdmull_lane_s16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmull_lane_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <2 x i64> [[VQDMULL_V2_I]]
 //
 int64x2_t test_vqdmull_lane_s32(int32x2_t a, int32x2_t v) {
   return vqdmull_lane_s32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vqdmull_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <4 x i32> [[VQDMULL_V2_I]]
 //
 int32x4_t test_vqdmull_laneq_s16(int16x4_t a, int16x8_t v) {
   return vqdmull_laneq_s16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmull_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <2 x i64> [[VQDMULL_V2_I]]
 //
 int64x2_t test_vqdmull_laneq_s32(int32x2_t a, int32x4_t v) {
   return vqdmull_laneq_s32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmull_high_lane_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <4 x i32> [[VQDMULL_V2_I]]
 //
 int32x4_t test_vqdmull_high_lane_s16(int16x8_t a, int16x4_t v) {
   return vqdmull_high_lane_s16(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmull_high_lane_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <2 x i64> [[VQDMULL_V2_I]]
 //
 int64x2_t test_vqdmull_high_lane_s32(int32x4_t a, int32x2_t v) {
   return vqdmull_high_lane_s32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vqdmull_high_laneq_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <4 x i32> [[VQDMULL_V2_I]]
 //
 int32x4_t test_vqdmull_high_laneq_s16(int16x8_t a, int16x8_t v) {
   return vqdmull_high_laneq_s16(a, v, 7);
 }
 
 // CHECK-LABEL: @test_vqdmull_high_laneq_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <2 x i64> [[VQDMULL_V2_I]]
 //
 int64x2_t test_vqdmull_high_laneq_s32(int32x4_t a, int32x4_t v) {
   return vqdmull_high_laneq_s32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmulh_lane_s16(

 // CHECK-NEXT:    ret <4 x i32> [[VQRDMULHQ_LANE_V2]]
 //
 int32x4_t test_vqrdmulhq_lane_s32(int32x4_t a, int32x2_t v) {
   return vqrdmulhq_lane_s32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmul_lane_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x float> [[MUL]]
 //
 float32x2_t test_vmul_lane_f32(float32x2_t a, float32x2_t v) {
   return vmul_lane_f32(a, v, 1);
 }
 
 
 // CHECK-LABEL: @test_vmul_lane_f64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <1 x double> [[A:%.*]] to <8 x i8>
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast <1 x double> [[V:%.*]] to <8 x i8>
 // CHECK-NEXT:    [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to double
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <1 x double>
 // CHECK-NEXT:    [[EXTRACT:%.*]] = extractelement <1 x double> [[TMP3]], i32 0
 // CHECK-NEXT:    [[TMP4:%.*]] = fmul double [[TMP2]], [[EXTRACT]]
 // CHECK-NEXT:    [[TMP5:%.*]] = bitcast double [[TMP4]] to <1 x double>
 // CHECK-NEXT:    ret <1 x double> [[TMP5]]
 //
 float64x1_t test_vmul_lane_f64(float64x1_t a, float64x1_t v) {
   return vmul_lane_f64(a, v, 0);
 }
 
-
 // CHECK-LABEL: @test_vmulq_lane_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> [[V]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x float> [[MUL]]
 //
 float32x4_t test_vmulq_lane_f32(float32x4_t a, float32x2_t v) {
   return vmulq_lane_f32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <1 x double> [[V:%.*]], <1 x double> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x double> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <1 x double> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x double>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <1 x double> [[TMP1]], <1 x double> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x double> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x double> [[MUL]]
 //
 float64x2_t test_vmulq_lane_f64(float64x2_t a, float64x1_t v) {
   return vmulq_lane_f64(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x float> [[V:%.*]], <4 x float> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x float> [[MUL]]
 //
 float32x2_t test_vmul_laneq_f32(float32x2_t a, float32x4_t v) {
   return vmul_laneq_f32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_f64(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <1 x double> [[A:%.*]] to <8 x i8>
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x double> [[V:%.*]] to <16 x i8>
 // CHECK-NEXT:    [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to double
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <2 x double>
 // CHECK-NEXT:    [[EXTRACT:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
 // CHECK-NEXT:    [[TMP4:%.*]] = fmul double [[TMP2]], [[EXTRACT]]
 // CHECK-NEXT:    [[TMP5:%.*]] = bitcast double [[TMP4]] to <1 x double>
 // CHECK-NEXT:    ret <1 x double> [[TMP5]]
 //
 float64x1_t test_vmul_laneq_f64(float64x1_t a, float64x2_t v) {
   return vmul_laneq_f64(a, v, 1);
 }
 
-
 // CHECK-LABEL: @test_vmulq_laneq_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x float> [[V:%.*]], <4 x float> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x float> [[MUL]]
 //
 float32x4_t test_vmulq_laneq_f32(float32x4_t a, float32x4_t v) {
   return vmulq_laneq_f32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x double> [[V:%.*]], <2 x double> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x double> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x double> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <2 x double>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x double> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x double> [[MUL]]
 //
 float64x2_t test_vmulq_laneq_f64(float64x2_t a, float64x2_t v) {
   return vmulq_laneq_f64(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmulx_lane_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x float> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fmulx.v2f32(<2 x float> [[A]], <2 x float> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x float> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fmulx.v2f32(<2 x float> [[A]], <2 x float> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x float> [[VMULX2_I]]
 //
 float32x2_t test_vmulx_lane_f32(float32x2_t a, float32x2_t v) {
   return vmulx_lane_f32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmulxq_lane_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> [[V]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x float> [[SHUFFLE]] to <16 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <4 x float> @llvm.aarch64.neon.fmulx.v4f32(<4 x float> [[A]], <4 x float> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x float> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x float> [[LANE]] to <16 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <4 x float> @llvm.aarch64.neon.fmulx.v4f32(<4 x float> [[A]], <4 x float> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x float> [[VMULX2_I]]
 //
 float32x4_t test_vmulxq_lane_f32(float32x4_t a, float32x2_t v) {
   return vmulxq_lane_f32(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmulxq_lane_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <1 x double> [[V:%.*]], <1 x double> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x double> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x double> [[SHUFFLE]] to <16 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x double> @llvm.aarch64.neon.fmulx.v2f64(<2 x double> [[A]], <2 x double> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <1 x double> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x double>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <1 x double> [[TMP1]], <1 x double> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x double> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x double> [[LANE]] to <16 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x double> @llvm.aarch64.neon.fmulx.v2f64(<2 x double> [[A]], <2 x double> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x double> [[VMULX2_I]]
 //
 float64x2_t test_vmulxq_lane_f64(float64x2_t a, float64x1_t v) {
   return vmulxq_lane_f64(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulx_laneq_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x float> [[V:%.*]], <4 x float> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x float> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fmulx.v2f32(<2 x float> [[A]], <2 x float> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x float> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fmulx.v2f32(<2 x float> [[A]], <2 x float> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x float> [[VMULX2_I]]
 //
 float32x2_t test_vmulx_laneq_f32(float32x2_t a, float32x4_t v) {
   return vmulx_laneq_f32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmulxq_laneq_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x float> [[V:%.*]], <4 x float> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x float> [[SHUFFLE]] to <16 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <4 x float> @llvm.aarch64.neon.fmulx.v4f32(<4 x float> [[A]], <4 x float> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x float> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x float> [[LANE]] to <16 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <4 x float> @llvm.aarch64.neon.fmulx.v4f32(<4 x float> [[A]], <4 x float> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x float> [[VMULX2_I]]
 //
 float32x4_t test_vmulxq_laneq_f32(float32x4_t a, float32x4_t v) {
   return vmulxq_laneq_f32(a, v, 3);
 }
 
 // CHECK-LABEL: @test_vmulxq_laneq_f64(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x double> [[V:%.*]], <2 x double> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x double> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x double> [[SHUFFLE]] to <16 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x double> @llvm.aarch64.neon.fmulx.v2f64(<2 x double> [[A]], <2 x double> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x double> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <2 x double>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x double> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x double> [[LANE]] to <16 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x double> @llvm.aarch64.neon.fmulx.v2f64(<2 x double> [[A]], <2 x double> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x double> [[VMULX2_I]]
 //
 float64x2_t test_vmulxq_laneq_f64(float64x2_t a, float64x2_t v) {
   return vmulxq_laneq_f64(a, v, 1);
 }
 
 // CHECK-LABEL: @test_vmla_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[ADD]]
 //
 int16x4_t test_vmla_lane_s16_0(int16x4_t a, int16x4_t b, int16x4_t v) {
   return vmla_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[ADD]]
 //
 int16x8_t test_vmlaq_lane_s16_0(int16x8_t a, int16x8_t b, int16x4_t v) {
   return vmlaq_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmla_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[ADD]]
 //
 int32x2_t test_vmla_lane_s32_0(int32x2_t a, int32x2_t b, int32x2_t v) {
   return vmla_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlaq_lane_s32_0(int32x4_t a, int32x4_t b, int32x2_t v) {
   return vmlaq_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmla_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[ADD]]
 //
 int16x4_t test_vmla_laneq_s16_0(int16x4_t a, int16x4_t b, int16x8_t v) {
   return vmla_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlaq_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[ADD]]
 //
 int16x8_t test_vmlaq_laneq_s16_0(int16x8_t a, int16x8_t b, int16x8_t v) {
   return vmlaq_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmla_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[ADD]]
 //
 int32x2_t test_vmla_laneq_s32_0(int32x2_t a, int32x2_t b, int32x4_t v) {
   return vmla_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlaq_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlaq_laneq_s32_0(int32x4_t a, int32x4_t b, int32x4_t v) {
   return vmlaq_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmls_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[SUB]]
 //
 int16x4_t test_vmls_lane_s16_0(int16x4_t a, int16x4_t b, int16x4_t v) {
   return vmls_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[SUB]]
 //
 int16x8_t test_vmlsq_lane_s16_0(int16x8_t a, int16x8_t b, int16x4_t v) {
   return vmlsq_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmls_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[SUB]]
 //
 int32x2_t test_vmls_lane_s32_0(int32x2_t a, int32x2_t b, int32x2_t v) {
   return vmls_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsq_lane_s32_0(int32x4_t a, int32x4_t b, int32x2_t v) {
   return vmlsq_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmls_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[SUB]]
 //
 int16x4_t test_vmls_laneq_s16_0(int16x4_t a, int16x4_t b, int16x8_t v) {
   return vmls_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsq_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[SUB]]
 //
 int16x8_t test_vmlsq_laneq_s16_0(int16x8_t a, int16x8_t b, int16x8_t v) {
   return vmlsq_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmls_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[SUB]]
 //
 int32x2_t test_vmls_laneq_s32_0(int32x2_t a, int32x2_t b, int32x4_t v) {
   return vmls_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsq_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsq_laneq_s32_0(int32x4_t a, int32x4_t b, int32x4_t v) {
   return vmlsq_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i16> [[MUL]]
 //
 int16x4_t test_vmul_lane_s16_0(int16x4_t a, int16x4_t v) {
   return vmul_lane_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <8 x i16> [[MUL]]
 //
 int16x8_t test_vmulq_lane_s16_0(int16x8_t a, int16x4_t v) {
   return vmulq_lane_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x i32> [[MUL]]
 //
 int32x2_t test_vmul_lane_s32_0(int32x2_t a, int32x2_t v) {
   return vmul_lane_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i32> [[MUL]]
 //
 int32x4_t test_vmulq_lane_s32_0(int32x4_t a, int32x2_t v) {
   return vmulq_lane_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i16> [[MUL]]
 //
 uint16x4_t test_vmul_lane_u16_0(uint16x4_t a, uint16x4_t v) {
   return vmul_lane_u16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <8 x i16> [[MUL]]
 //
 uint16x8_t test_vmulq_lane_u16_0(uint16x8_t a, uint16x4_t v) {
   return vmulq_lane_u16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x i32> [[MUL]]
 //
 uint32x2_t test_vmul_lane_u32_0(uint32x2_t a, uint32x2_t v) {
   return vmul_lane_u32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i32> [[MUL]]
 //
 uint32x4_t test_vmulq_lane_u32_0(uint32x4_t a, uint32x2_t v) {
   return vmulq_lane_u32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i16> [[MUL]]
 //
 int16x4_t test_vmul_laneq_s16_0(int16x4_t a, int16x8_t v) {
   return vmul_laneq_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <8 x i16> [[MUL]]
 //
 int16x8_t test_vmulq_laneq_s16_0(int16x8_t a, int16x8_t v) {
   return vmulq_laneq_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x i32> [[MUL]]
 //
 int32x2_t test_vmul_laneq_s32_0(int32x2_t a, int32x4_t v) {
   return vmul_laneq_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i32> [[MUL]]
 //
 int32x4_t test_vmulq_laneq_s32_0(int32x4_t a, int32x4_t v) {
   return vmulq_laneq_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i16> [[MUL]]
 //
 uint16x4_t test_vmul_laneq_u16_0(uint16x4_t a, uint16x8_t v) {
   return vmul_laneq_u16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <8 x i16> [[MUL]]
 //
 uint16x8_t test_vmulq_laneq_u16_0(uint16x8_t a, uint16x8_t v) {
   return vmulq_laneq_u16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x i32> [[MUL]]
 //
 uint32x2_t test_vmul_laneq_u32_0(uint32x2_t a, uint32x4_t v) {
   return vmul_laneq_u32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x i32> [[MUL]]
 //
 uint32x4_t test_vmulq_laneq_u32_0(uint32x4_t a, uint32x4_t v) {
   return vmulq_laneq_u32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vfma_lane_f32_0(
 // CHECK-NEXT:  entry:

 // CHECK-NEXT:    ret <2 x double> [[TMP6]]
 //
 float64x2_t test_vfmsq_laneq_f64_0(float64x2_t a, float64x2_t b, float64x2_t v) {
   return vfmsq_laneq_f64(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_lane_s16_0(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vmlal_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_lane_s32_0(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vmlal_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_laneq_s16_0(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vmlal_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_laneq_s32_0(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vmlal_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_high_lane_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_high_lane_s16_0(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vmlal_high_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_high_lane_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_high_lane_s32_0(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vmlal_high_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_high_laneq_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_high_laneq_s16_0(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vmlal_high_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_high_laneq_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_high_laneq_s32_0(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vmlal_high_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_lane_s16_0(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vmlsl_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_lane_s32_0(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vmlsl_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_laneq_s16_0(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vmlsl_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_laneq_s32_0(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vmlsl_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_lane_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_high_lane_s16_0(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vmlsl_high_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_lane_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_high_lane_s32_0(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vmlsl_high_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_laneq_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_high_laneq_s16_0(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vmlsl_high_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_laneq_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_high_laneq_s32_0(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vmlsl_high_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_lane_u16_0(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vmlal_lane_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_lane_u32_0(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vmlal_lane_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_laneq_u16_0(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vmlal_laneq_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_laneq_u32_0(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vmlal_laneq_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_high_lane_u16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_high_lane_u16_0(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vmlal_high_lane_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_high_lane_u32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_high_lane_u32_0(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vmlal_high_lane_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_high_laneq_u16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 int32x4_t test_vmlal_high_laneq_u16_0(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vmlal_high_laneq_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlal_high_laneq_u32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[ADD]]
 //
 int64x2_t test_vmlal_high_laneq_u32_0(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vmlal_high_laneq_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_lane_u16_0(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vmlsl_lane_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_lane_u32_0(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vmlsl_lane_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_laneq_u16_0(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vmlsl_laneq_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_laneq_u32_0(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vmlsl_laneq_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_lane_u16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_high_lane_u16_0(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vmlsl_high_lane_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_lane_u32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_high_lane_u32_0(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vmlsl_high_lane_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_laneq_u16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 int32x4_t test_vmlsl_high_laneq_u16_0(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vmlsl_high_laneq_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsl_high_laneq_u32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK-NEXT:    ret <2 x i64> [[SUB]]
 //
 int64x2_t test_vmlsl_high_laneq_u32_0(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vmlsl_high_laneq_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 int32x4_t test_vmull_lane_s16_0(int16x4_t a, int16x4_t v) {
   return vmull_lane_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 int64x2_t test_vmull_lane_s32_0(int32x2_t a, int32x2_t v) {
   return vmull_lane_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 uint32x4_t test_vmull_lane_u16_0(uint16x4_t a, uint16x4_t v) {
   return vmull_lane_u16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 uint64x2_t test_vmull_lane_u32_0(uint32x2_t a, uint32x2_t v) {
   return vmull_lane_u32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_lane_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 int32x4_t test_vmull_high_lane_s16_0(int16x8_t a, int16x4_t v) {
   return vmull_high_lane_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_lane_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 int64x2_t test_vmull_high_lane_s32_0(int32x4_t a, int32x2_t v) {
   return vmull_high_lane_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_lane_u16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 uint32x4_t test_vmull_high_lane_u16_0(uint16x8_t a, uint16x4_t v) {
   return vmull_high_lane_u16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_lane_u32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 uint64x2_t test_vmull_high_lane_u32_0(uint32x4_t a, uint32x2_t v) {
   return vmull_high_lane_u32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 int32x4_t test_vmull_laneq_s16_0(int16x4_t a, int16x8_t v) {
   return vmull_laneq_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 int64x2_t test_vmull_laneq_s32_0(int32x2_t a, int32x4_t v) {
   return vmull_laneq_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 uint32x4_t test_vmull_laneq_u16_0(uint16x4_t a, uint16x8_t v) {
   return vmull_laneq_u16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 uint64x2_t test_vmull_laneq_u32_0(uint32x2_t a, uint32x4_t v) {
   return vmull_laneq_u32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_laneq_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.smull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 int32x4_t test_vmull_high_laneq_s16_0(int16x8_t a, int16x8_t v) {
   return vmull_high_laneq_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_laneq_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.smull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 int64x2_t test_vmull_high_laneq_s32_0(int32x4_t a, int32x4_t v) {
   return vmull_high_laneq_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_laneq_u16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.umull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VMULL2_I]]
 //
 uint32x4_t test_vmull_high_laneq_u16_0(uint16x8_t a, uint16x8_t v) {
   return vmull_high_laneq_u16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_laneq_u32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VMULL2_I]]
 //
 uint64x2_t test_vmull_high_laneq_u32_0(uint32x4_t a, uint32x4_t v) {
   return vmull_high_laneq_u32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlal_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLAL_V3_I]]
 //
 int32x4_t test_vqdmlal_lane_s16_0(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vqdmlal_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlal_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqadd.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLAL_V3_I]]
 //
 int64x2_t test_vqdmlal_lane_s32_0(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vqdmlal_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlal_high_lane_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLAL_V3_I]]
 //
 int32x4_t test_vqdmlal_high_lane_s16_0(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vqdmlal_high_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlal_high_lane_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqadd.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLAL_V3_I]]
 //
 int64x2_t test_vqdmlal_high_lane_s32_0(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vqdmlal_high_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLSL_V3_I]]
 //
 int32x4_t test_vqdmlsl_lane_s16_0(int32x4_t a, int16x4_t b, int16x4_t v) {
   return vqdmlsl_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I]]
 //
 int64x2_t test_vqdmlsl_lane_s32_0(int64x2_t a, int32x2_t b, int32x2_t v) {
   return vqdmlsl_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_high_lane_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLSL_V3_I]]
 //
 int32x4_t test_vqdmlsl_high_lane_s16_0(int32x4_t a, int16x8_t b, int16x4_t v) {
   return vqdmlsl_high_lane_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_high_lane_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I]]
 //
 int64x2_t test_vqdmlsl_high_lane_s32_0(int64x2_t a, int32x4_t b, int32x2_t v) {
   return vqdmlsl_high_lane_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmull_lane_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <4 x i32> [[VQDMULL_V2_I]]
 //
 int32x4_t test_vqdmull_lane_s16_0(int16x4_t a, int16x4_t v) {
   return vqdmull_lane_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmull_lane_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <2 x i64> [[VQDMULL_V2_I]]
 //
 int64x2_t test_vqdmull_lane_s32_0(int32x2_t a, int32x2_t v) {
   return vqdmull_lane_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmull_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <4 x i32> [[VQDMULL_V2_I]]
 //
 int32x4_t test_vqdmull_laneq_s16_0(int16x4_t a, int16x8_t v) {
   return vqdmull_laneq_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmull_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <2 x i64> [[VQDMULL_V2_I]]
 //
 int64x2_t test_vqdmull_laneq_s32_0(int32x2_t a, int32x4_t v) {
   return vqdmull_laneq_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmull_high_lane_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <4 x i32> [[VQDMULL_V2_I]]
 //
 int32x4_t test_vqdmull_high_lane_s16_0(int16x8_t a, int16x4_t v) {
   return vqdmull_high_lane_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmull_high_lane_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <2 x i64> [[VQDMULL_V2_I]]
 //
 int64x2_t test_vqdmull_high_lane_s32_0(int32x4_t a, int32x2_t v) {
   return vqdmull_high_lane_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmull_high_laneq_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[A:%.*]], <8 x i16> [[A]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <4 x i32> [[VQDMULL_V2_I]]
 //
 int32x4_t test_vqdmull_high_laneq_s16_0(int16x8_t a, int16x8_t v) {
   return vqdmull_high_laneq_s16(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmull_high_laneq_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> [[A]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK-NEXT:    ret <2 x i64> [[VQDMULL_V2_I]]
 //
 int64x2_t test_vqdmull_high_laneq_s32_0(int32x4_t a, int32x4_t v) {
   return vqdmull_high_laneq_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmulh_lane_s16_0(

 // CHECK-NEXT:    ret <4 x i32> [[VQRDMULHQ_LANE_V2]]
 //
 int32x4_t test_vqrdmulhq_lane_s32_0(int32x4_t a, int32x2_t v) {
   return vqrdmulhq_lane_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_lane_f32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x float> [[MUL]]
 //
 float32x2_t test_vmul_lane_f32_0(float32x2_t a, float32x2_t v) {
   return vmul_lane_f32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_f32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x float> [[MUL]]
 //
 float32x4_t test_vmulq_lane_f32_0(float32x4_t a, float32x2_t v) {
   return vmulq_lane_f32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_f32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x float> [[V:%.*]], <4 x float> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x float> [[MUL]]
 //
 float32x2_t test_vmul_laneq_f32_0(float32x2_t a, float32x4_t v) {
   return vmul_laneq_f32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmul_laneq_f64_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = bitcast <1 x double> [[A:%.*]] to <8 x i8>
 // CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x double> [[V:%.*]] to <16 x i8>
 // CHECK-NEXT:    [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to double
 // CHECK-NEXT:    [[TMP3:%.*]] = bitcast <16 x i8> [[TMP1]] to <2 x double>
 // CHECK-NEXT:    [[EXTRACT:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
 // CHECK-NEXT:    [[TMP4:%.*]] = fmul double [[TMP2]], [[EXTRACT]]
 // CHECK-NEXT:    [[TMP5:%.*]] = bitcast double [[TMP4]] to <1 x double>
 // CHECK-NEXT:    ret <1 x double> [[TMP5]]
 //
 float64x1_t test_vmul_laneq_f64_0(float64x1_t a, float64x2_t v) {
   return vmul_laneq_f64(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_f32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x float> [[V:%.*]], <4 x float> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <4 x float> [[MUL]]
 //
 float32x4_t test_vmulq_laneq_f32_0(float32x4_t a, float32x4_t v) {
   return vmulq_laneq_f32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulq_laneq_f64_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x double> [[V:%.*]], <2 x double> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x double> [[A:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x double> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <2 x double>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = fmul <2 x double> [[A:%.*]], [[LANE]]
 // CHECK-NEXT:    ret <2 x double> [[MUL]]
 //
 float64x2_t test_vmulq_laneq_f64_0(float64x2_t a, float64x2_t v) {
   return vmulq_laneq_f64(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulx_lane_f32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x float> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fmulx.v2f32(<2 x float> [[A]], <2 x float> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x float> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fmulx.v2f32(<2 x float> [[A]], <2 x float> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x float> [[VMULX2_I]]
 //
 float32x2_t test_vmulx_lane_f32_0(float32x2_t a, float32x2_t v) {
   return vmulx_lane_f32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulxq_lane_f32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x float> [[V:%.*]], <2 x float> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x float> [[SHUFFLE]] to <16 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <4 x float> @llvm.aarch64.neon.fmulx.v4f32(<4 x float> [[A]], <4 x float> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x float> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x float> [[LANE]] to <16 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <4 x float> @llvm.aarch64.neon.fmulx.v4f32(<4 x float> [[A]], <4 x float> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x float> [[VMULX2_I]]
 //
 float32x4_t test_vmulxq_lane_f32_0(float32x4_t a, float32x2_t v) {
   return vmulxq_lane_f32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulxq_lane_f64_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <1 x double> [[V:%.*]], <1 x double> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x double> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x double> [[SHUFFLE]] to <16 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x double> @llvm.aarch64.neon.fmulx.v2f64(<2 x double> [[A]], <2 x double> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <1 x double> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x double>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <1 x double> [[TMP1]], <1 x double> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x double> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x double> [[LANE]] to <16 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x double> @llvm.aarch64.neon.fmulx.v2f64(<2 x double> [[A]], <2 x double> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x double> [[VMULX2_I]]
 //
 float64x2_t test_vmulxq_lane_f64_0(float64x2_t a, float64x1_t v) {
   return vmulxq_lane_f64(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulx_laneq_f32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x float> [[V:%.*]], <4 x float> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x float> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fmulx.v2f32(<2 x float> [[A]], <2 x float> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x float> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x float> @llvm.aarch64.neon.fmulx.v2f32(<2 x float> [[A]], <2 x float> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x float> [[VMULX2_I]]
 //
 float32x2_t test_vmulx_laneq_f32_0(float32x2_t a, float32x4_t v) {
   return vmulx_laneq_f32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulxq_laneq_f32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x float> [[V:%.*]], <4 x float> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x float> [[SHUFFLE]] to <16 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <4 x float> @llvm.aarch64.neon.fmulx.v4f32(<4 x float> [[A]], <4 x float> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x float> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x float> [[LANE]] to <16 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <4 x float> @llvm.aarch64.neon.fmulx.v4f32(<4 x float> [[A]], <4 x float> [[LANE]]) #4
 // CHECK-NEXT:    ret <4 x float> [[VMULX2_I]]
 //
 float32x4_t test_vmulxq_laneq_f32_0(float32x4_t a, float32x4_t v) {
   return vmulxq_laneq_f32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmulxq_laneq_f64_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x double> [[V:%.*]], <2 x double> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x double> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x double> [[SHUFFLE]] to <16 x i8>
-// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x double> @llvm.aarch64.neon.fmulx.v2f64(<2 x double> [[A]], <2 x double> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x double> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <2 x double>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x double> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x double> [[LANE]] to <16 x i8>
+// CHECK-NEXT:    [[VMULX2_I:%.*]] = call <2 x double> @llvm.aarch64.neon.fmulx.v2f64(<2 x double> [[A]], <2 x double> [[LANE]]) #4
 // CHECK-NEXT:    ret <2 x double> [[VMULX2_I]]
 //
 float64x2_t test_vmulxq_laneq_f64_0(float64x2_t a, float64x2_t v) {
   return vmulxq_laneq_f64(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmull_high_n_s16(
 // CHECK-NEXT:  entry:

 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I_I]]
 //
 int64x2_t test_vqdmlsl_n_s32(int64x2_t a, int32x2_t b, int32_t c) {
   return vqdmlsl_n_s32(a, b, c);
 }
 
 // CHECK-LABEL: @test_vmla_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[ADD]]
 //
 uint16x4_t test_vmla_lane_u16_0(uint16x4_t a, uint16x4_t b, uint16x4_t v) {
   return vmla_lane_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[ADD]]
 //
 uint16x8_t test_vmlaq_lane_u16_0(uint16x8_t a, uint16x8_t b, uint16x4_t v) {
   return vmlaq_lane_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmla_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[ADD]]
 //
 uint32x2_t test_vmla_lane_u32_0(uint32x2_t a, uint32x2_t b, uint32x2_t v) {
   return vmla_lane_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 uint32x4_t test_vmlaq_lane_u32_0(uint32x4_t a, uint32x4_t b, uint32x2_t v) {
   return vmlaq_lane_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmla_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[ADD]]
 //
 uint16x4_t test_vmla_laneq_u16_0(uint16x4_t a, uint16x4_t b, uint16x8_t v) {
   return vmla_laneq_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlaq_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[ADD]]
 //
 uint16x8_t test_vmlaq_laneq_u16_0(uint16x8_t a, uint16x8_t b, uint16x8_t v) {
   return vmlaq_laneq_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmla_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[ADD]]
 //
 uint32x2_t test_vmla_laneq_u32_0(uint32x2_t a, uint32x2_t b, uint32x4_t v) {
   return vmla_laneq_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlaq_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 uint32x4_t test_vmlaq_laneq_u32_0(uint32x4_t a, uint32x4_t b, uint32x4_t v) {
   return vmlaq_laneq_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlal_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLAL_V3_I]]
 //
 int32x4_t test_vqdmlal_laneq_s16_0(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vqdmlal_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlal_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqadd.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLAL_V3_I]]
 //
 int64x2_t test_vqdmlal_laneq_s32_0(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vqdmlal_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlal_high_laneq_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLAL_V3_I]]
 //
 int32x4_t test_vqdmlal_high_laneq_s16_0(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vqdmlal_high_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlal_high_laneq_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqadd.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLAL_V3_I]]
 //
 int64x2_t test_vqdmlal_high_laneq_s32_0(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vqdmlal_high_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmls_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[SUB]]
 //
 uint16x4_t test_vmls_lane_u16_0(uint16x4_t a, uint16x4_t b, uint16x4_t v) {
   return vmls_lane_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[SUB]]
 //
 uint16x8_t test_vmlsq_lane_u16_0(uint16x8_t a, uint16x8_t b, uint16x4_t v) {
   return vmlsq_lane_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmls_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[SUB]]
 //
 uint32x2_t test_vmls_lane_u32_0(uint32x2_t a, uint32x2_t b, uint32x2_t v) {
   return vmls_lane_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 uint32x4_t test_vmlsq_lane_u32_0(uint32x4_t a, uint32x4_t b, uint32x2_t v) {
   return vmlsq_lane_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmls_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[SUB]]
 //
 uint16x4_t test_vmls_laneq_u16_0(uint16x4_t a, uint16x4_t b, uint16x8_t v) {
   return vmls_laneq_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsq_laneq_u16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[SUB]]
 //
 uint16x8_t test_vmlsq_laneq_u16_0(uint16x8_t a, uint16x8_t b, uint16x8_t v) {
   return vmlsq_laneq_u16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmls_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[SUB]]
 //
 uint32x2_t test_vmls_laneq_u32_0(uint32x2_t a, uint32x2_t b, uint32x4_t v) {
   return vmls_laneq_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vmlsq_laneq_u32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 uint32x4_t test_vmlsq_laneq_u32_0(uint32x4_t a, uint32x4_t b, uint32x4_t v) {
   return vmlsq_laneq_u32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_laneq_s16_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLSL_V3_I]]
 //
 int32x4_t test_vqdmlsl_laneq_s16_0(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vqdmlsl_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_laneq_s32_0(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I]]
 //
 int64x2_t test_vqdmlsl_laneq_s32_0(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vqdmlsl_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_high_laneq_s16_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLSL_V3_I]]
 //
 int32x4_t test_vqdmlsl_high_laneq_s16_0(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vqdmlsl_high_laneq_s16(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_high_laneq_s32_0(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> zeroinitializer
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I]]
 //
 int64x2_t test_vqdmlsl_high_laneq_s32_0(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vqdmlsl_high_laneq_s32(a, b, v, 0);
 }
 
 // CHECK-LABEL: @test_vqdmulh_laneq_s16_0(

 // CHECK-NEXT:    ret <4 x i32> [[VQRDMULHQ_LANEQ_V2]]
 //
 int32x4_t test_vqrdmulhq_laneq_s32_0(int32x4_t a, int32x4_t v) {
   return vqrdmulhq_laneq_s32(a, v, 0);
 }
 
 // CHECK-LABEL: @test_vmla_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[ADD]]
 //
 uint16x4_t test_vmla_lane_u16(uint16x4_t a, uint16x4_t b, uint16x4_t v) {
   return vmla_lane_u16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[ADD]]
 //
 uint16x8_t test_vmlaq_lane_u16(uint16x8_t a, uint16x8_t b, uint16x4_t v) {
   return vmlaq_lane_u16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmla_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[ADD]]
 //
 uint32x2_t test_vmla_lane_u32(uint32x2_t a, uint32x2_t b, uint32x2_t v) {
   return vmla_lane_u32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 uint32x4_t test_vmlaq_lane_u32(uint32x4_t a, uint32x4_t b, uint32x2_t v) {
   return vmlaq_lane_u32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmla_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[ADD]]
 //
 uint16x4_t test_vmla_laneq_u16(uint16x4_t a, uint16x4_t b, uint16x8_t v) {
   return vmla_laneq_u16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlaq_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[ADD]]
 //
 uint16x8_t test_vmlaq_laneq_u16(uint16x8_t a, uint16x8_t b, uint16x8_t v) {
   return vmlaq_laneq_u16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmla_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[ADD]]
 //
 uint32x2_t test_vmla_laneq_u32(uint32x2_t a, uint32x2_t b, uint32x4_t v) {
   return vmla_laneq_u32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlaq_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[ADD]]
 //
 uint32x4_t test_vmlaq_laneq_u32(uint32x4_t a, uint32x4_t b, uint32x4_t v) {
   return vmlaq_laneq_u32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlal_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLAL_V3_I]]
 //
 int32x4_t test_vqdmlal_laneq_s16(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vqdmlal_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vqdmlal_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqadd.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLAL_V3_I]]
 //
 int64x2_t test_vqdmlal_laneq_s32(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vqdmlal_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlal_high_laneq_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLAL_V3_I]]
 //
 int32x4_t test_vqdmlal_high_laneq_s16(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vqdmlal_high_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vqdmlal_high_laneq_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqadd.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLAL_V3_I]]
 //
 int64x2_t test_vqdmlal_high_laneq_s32(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vqdmlal_high_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmls_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[SUB]]
 //
 uint16x4_t test_vmls_lane_u16(uint16x4_t a, uint16x4_t b, uint16x4_t v) {
   return vmls_lane_u16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i16> [[V:%.*]], <4 x i16> [[V]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i16> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[SUB]]
 //
 uint16x8_t test_vmlsq_lane_u16(uint16x8_t a, uint16x8_t b, uint16x4_t v) {
   return vmlsq_lane_u16(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmls_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <2 x i32> <i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[SUB]]
 //
 uint32x2_t test_vmls_lane_u32(uint32x2_t a, uint32x2_t b, uint32x2_t v) {
   return vmls_lane_u32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[V:%.*]], <2 x i32> [[V]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i32> [[V:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 uint32x4_t test_vmlsq_lane_u32(uint32x4_t a, uint32x4_t b, uint32x2_t v) {
   return vmlsq_lane_u32(a, b, v, 1);
 }
 
 // CHECK-LABEL: @test_vmls_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i16> [[SUB]]
 //
 uint16x4_t test_vmls_laneq_u16(uint16x4_t a, uint16x4_t b, uint16x8_t v) {
   return vmls_laneq_u16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmlsq_laneq_u16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <8 x i16> [[SUB]]
 //
 uint16x8_t test_vmlsq_laneq_u16(uint16x8_t a, uint16x8_t b, uint16x8_t v) {
   return vmlsq_laneq_u16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vmls_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <2 x i32> [[SUB]]
 //
 uint32x2_t test_vmls_laneq_u32(uint32x2_t a, uint32x2_t b, uint32x4_t v) {
   return vmls_laneq_u32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vmlsq_laneq_u32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[SHUFFLE]]
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK-NEXT:    [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
 // CHECK-NEXT:    [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK-NEXT:    ret <4 x i32> [[SUB]]
 //
 uint32x4_t test_vmlsq_laneq_u32(uint32x4_t a, uint32x4_t b, uint32x4_t v) {
   return vmlsq_laneq_u32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_laneq_s16(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLSL_V3_I]]
 //
 int32x4_t test_vqdmlsl_laneq_s16(int32x4_t a, int16x4_t b, int16x8_t v) {
   return vqdmlsl_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_laneq_s32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I]]
 //
 int64x2_t test_vqdmlsl_laneq_s32(int64x2_t a, int32x2_t b, int32x4_t v) {
   return vqdmlsl_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_high_laneq_s16(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> [[B]], <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[V:%.*]], <8 x i16> [[V]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <8 x i16> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <4 x i16> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> [[SHUFFLE_I]], <4 x i16> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.aarch64.neon.sqsub.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <4 x i32> [[VQDMLSL_V3_I]]
 //
 int32x4_t test_vqdmlsl_high_laneq_s16(int32x4_t a, int16x8_t b, int16x8_t v) {
   return vqdmlsl_high_laneq_s16(a, b, v, 7);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_high_laneq_s32(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[SHUFFLE_I:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> [[B]], <2 x i32> <i32 2, i32 3>
-// CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[V:%.*]], <4 x i32> [[V]], <2 x i32> <i32 3, i32 3>
-// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
-// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
-// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[SHUFFLE]]) #4
+// CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i32> [[V:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x i32>
+// CHECK-NEXT:    [[LANE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK-NEXT:    [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK-NEXT:    [[TMP3:%.*]] = bitcast <2 x i32> [[SHUFFLE_I]] to <8 x i8>
+// CHECK-NEXT:    [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK-NEXT:    [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqdmull.v2i64(<2 x i32> [[SHUFFLE_I]], <2 x i32> [[LANE]]) #4
 // CHECK-NEXT:    [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.aarch64.neon.sqsub.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #4
 // CHECK-NEXT:    ret <2 x i64> [[VQDMLSL_V3_I]]
 //
 int64x2_t test_vqdmlsl_high_laneq_s32(int64x2_t a, int32x4_t b, int32x4_t v) {
   return vqdmlsl_high_laneq_s32(a, b, v, 3);
 }
 
 // CHECK-LABEL: @test_vqdmulh_laneq_s16(

clang/test/CodeGen/aarch64-neon-fma.c

 // CHECK:   [[MUL_I:%.*]] = fmul <2 x double> %b, [[VECINIT1_I]]
 // CHECK:   [[SUB_I:%.*]] = fsub <2 x double> %a, [[MUL_I]]
 // CHECK:   ret <2 x double> [[SUB_I]]
 float64x2_t test_vmlsq_n_f64(float64x2_t a, float64x2_t b, float64_t c) {
   return vmlsq_n_f64(a, b, c);
 }
 
 // CHECK-LABEL: define <2 x float> @test_vmla_lane_f32_0(<2 x float> %a, <2 x float> %b, <2 x float> %v) #0 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %v, <2 x float> %v, <2 x i32> zeroinitializer
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <2 x float> %a, [[MUL]]
-// CHECK:   ret <2 x float> [[ADD]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK:    [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[ADD:%.*]] = fadd <2 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <2 x float> [[ADD]]
 float32x2_t test_vmla_lane_f32_0(float32x2_t a, float32x2_t b, float32x2_t v) {
   return vmla_lane_f32(a, b, v, 0);
 }
 
 // CHECK-LABEL: define <4 x float> @test_vmlaq_lane_f32_0(<4 x float> %a, <4 x float> %b, <2 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %v, <2 x float> %v, <4 x i32> zeroinitializer
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <4 x float> %a, [[MUL]]
-// CHECK:   ret <4 x float> [[ADD]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK:    [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[ADD:%.*]] = fadd <4 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <4 x float> [[ADD]]
 float32x4_t test_vmlaq_lane_f32_0(float32x4_t a, float32x4_t b, float32x2_t v) {
   return vmlaq_lane_f32(a, b, v, 0);
 }
 
 // CHECK-LABEL: define <2 x float> @test_vmla_laneq_f32_0(<2 x float> %a, <2 x float> %b, <4 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x float> %v, <4 x float> %v, <2 x i32> zeroinitializer
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <2 x float> %a, [[MUL]]
-// CHECK:   ret <2 x float> [[ADD]]
+// CHECK:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK:    [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[ADD:%.*]] = fadd <2 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <2 x float> [[ADD]]
 float32x2_t test_vmla_laneq_f32_0(float32x2_t a, float32x2_t b, float32x4_t v) {
   return vmla_laneq_f32(a, b, v, 0);
 }
 
 // CHECK-LABEL: define <4 x float> @test_vmlaq_laneq_f32_0(<4 x float> %a, <4 x float> %b, <4 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x float> %v, <4 x float> %v, <4 x i32> zeroinitializer
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <4 x float> %a, [[MUL]]
-// CHECK:   ret <4 x float> [[ADD]]
+// CHECK:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK:    [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[ADD:%.*]] = fadd <4 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <4 x float> [[ADD]]
 float32x4_t test_vmlaq_laneq_f32_0(float32x4_t a, float32x4_t b, float32x4_t v) {
   return vmlaq_laneq_f32(a, b, v, 0);
 }
 
 // CHECK-LABEL: define <2 x float> @test_vmls_lane_f32_0(<2 x float> %a, <2 x float> %b, <2 x float> %v) #0 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %v, <2 x float> %v, <2 x i32> zeroinitializer
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <2 x float> %a, [[MUL]]
-// CHECK:   ret <2 x float> [[SUB]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK:    [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[SUB:%.*]] = fsub <2 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <2 x float> [[SUB]]
 float32x2_t test_vmls_lane_f32_0(float32x2_t a, float32x2_t b, float32x2_t v) {
   return vmls_lane_f32(a, b, v, 0);
 }
 
 // CHECK-LABEL: define <4 x float> @test_vmlsq_lane_f32_0(<4 x float> %a, <4 x float> %b, <2 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %v, <2 x float> %v, <4 x i32> zeroinitializer
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <4 x float> %a, [[MUL]]
-// CHECK:   ret <4 x float> [[SUB]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK:    [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[SUB:%.*]] = fsub <4 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <4 x float> [[SUB]]
 float32x4_t test_vmlsq_lane_f32_0(float32x4_t a, float32x4_t b, float32x2_t v) {
   return vmlsq_lane_f32(a, b, v, 0);
 }
 
 // CHECK-LABEL: define <2 x float> @test_vmls_laneq_f32_0(<2 x float> %a, <2 x float> %b, <4 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x float> %v, <4 x float> %v, <2 x i32> zeroinitializer
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <2 x float> %a, [[MUL]]
-// CHECK:   ret <2 x float> [[SUB]]
+// CHECK:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK:    [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[SUB:%.*]] = fsub <2 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <2 x float> [[SUB]]
 float32x2_t test_vmls_laneq_f32_0(float32x2_t a, float32x2_t b, float32x4_t v) {
   return vmls_laneq_f32(a, b, v, 0);
 }
 
 // CHECK-LABEL: define <4 x float> @test_vmlsq_laneq_f32_0(<4 x float> %a, <4 x float> %b, <4 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x float> %v, <4 x float> %v, <4 x i32> zeroinitializer
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <4 x float> %a, [[MUL]]
-// CHECK:   ret <4 x float> [[SUB]]
+// CHECK:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <4 x i32> zeroinitializer
+// CHECK:    [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[SUB:%.*]] = fsub <4 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <4 x float> [[SUB]]
 float32x4_t test_vmlsq_laneq_f32_0(float32x4_t a, float32x4_t b, float32x4_t v) {
   return vmlsq_laneq_f32(a, b, v, 0);
 }
 
 // CHECK-LABEL: define <2 x float> @test_vmla_lane_f32(<2 x float> %a, <2 x float> %b, <2 x float> %v) #0 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %v, <2 x float> %v, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <2 x float> %a, [[MUL]]
-// CHECK:   ret <2 x float> [[ADD]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:    [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[ADD:%.*]] = fadd <2 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <2 x float> [[ADD]]
 float32x2_t test_vmla_lane_f32(float32x2_t a, float32x2_t b, float32x2_t v) {
   return vmla_lane_f32(a, b, v, 1);
 }
 
 // CHECK-LABEL: define <4 x float> @test_vmlaq_lane_f32(<4 x float> %a, <4 x float> %b, <2 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %v, <2 x float> %v, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <4 x float> %a, [[MUL]]
-// CHECK:   ret <4 x float> [[ADD]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:    [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[ADD:%.*]] = fadd <4 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <4 x float> [[ADD]]
 float32x4_t test_vmlaq_lane_f32(float32x4_t a, float32x4_t b, float32x2_t v) {
   return vmlaq_lane_f32(a, b, v, 1);
 }
 
 // CHECK-LABEL: define <2 x float> @test_vmla_laneq_f32(<2 x float> %a, <2 x float> %b, <4 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x float> %v, <4 x float> %v, <2 x i32> <i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <2 x float> %a, [[MUL]]
-// CHECK:   ret <2 x float> [[ADD]]
+// CHECK:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK:    [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[ADD:%.*]] = fadd <2 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <2 x float> [[ADD]]
 float32x2_t test_vmla_laneq_f32(float32x2_t a, float32x2_t b, float32x4_t v) {
   return vmla_laneq_f32(a, b, v, 3);
 }
 
 // CHECK-LABEL: define <4 x float> @test_vmlaq_laneq_f32(<4 x float> %a, <4 x float> %b, <4 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x float> %v, <4 x float> %v, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <4 x float> %a, [[MUL]]
-// CHECK:   ret <4 x float> [[ADD]]
+// CHECK:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:    [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[ADD:%.*]] = fadd <4 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <4 x float> [[ADD]]
 float32x4_t test_vmlaq_laneq_f32(float32x4_t a, float32x4_t b, float32x4_t v) {
   return vmlaq_laneq_f32(a, b, v, 3);
 }
 
 // CHECK-LABEL: define <2 x float> @test_vmls_lane_f32(<2 x float> %a, <2 x float> %b, <2 x float> %v) #0 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %v, <2 x float> %v, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <2 x float> %a, [[MUL]]
-// CHECK:   ret <2 x float> [[SUB]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:    [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[SUB:%.*]] = fsub <2 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <2 x float> [[SUB]]
 float32x2_t test_vmls_lane_f32(float32x2_t a, float32x2_t b, float32x2_t v) {
   return vmls_lane_f32(a, b, v, 1);
 }
 
 // CHECK-LABEL: define <4 x float> @test_vmlsq_lane_f32(<4 x float> %a, <4 x float> %b, <2 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %v, <2 x float> %v, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <4 x float> %a, [[MUL]]
-// CHECK:   ret <4 x float> [[SUB]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x float> [[V:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:    [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[SUB:%.*]] = fsub <4 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <4 x float> [[SUB]]
+//
 float32x4_t test_vmlsq_lane_f32(float32x4_t a, float32x4_t b, float32x2_t v) {
   return vmlsq_lane_f32(a, b, v, 1);
 }
 // CHECK-LABEL: define <2 x float> @test_vmls_laneq_f32(<2 x float> %a, <2 x float> %b, <4 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x float> %v, <4 x float> %v, <2 x i32> <i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <2 x float> %a, [[MUL]]
-// CHECK:   ret <2 x float> [[SUB]]
+// CHECK:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <2 x i32> <i32 3, i32 3>
+// CHECK:    [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[SUB:%.*]] = fsub <2 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <2 x float> [[SUB]]
 float32x2_t test_vmls_laneq_f32(float32x2_t a, float32x2_t b, float32x4_t v) {
   return vmls_laneq_f32(a, b, v, 3);
 }
 
 // CHECK-LABEL: define <4 x float> @test_vmlsq_laneq_f32(<4 x float> %a, <4 x float> %b, <4 x float> %v) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x float> %v, <4 x float> %v, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <4 x float> %a, [[MUL]]
-// CHECK:   ret <4 x float> [[SUB]]
+// CHECK:    [[TMP0:%.*]] = bitcast <4 x float> [[V:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <4 x float>
+// CHECK:    [[LANE:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:    [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:    [[SUB:%.*]] = fsub <4 x float> [[A:%.*]], [[MUL]]
+// CHECK:    ret <4 x float> [[SUB]]
 float32x4_t test_vmlsq_laneq_f32(float32x4_t a, float32x4_t b, float32x4_t v) {
   return vmlsq_laneq_f32(a, b, v, 3);
 }
 
 // CHECK-LABEL: define <2 x double> @test_vfmaq_n_f64(<2 x double> %a, <2 x double> %b, double %c) #1 {
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <2 x double> undef, double %c, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <2 x double> [[VECINIT_I]], double %c, i32 1
 // CHECK:   [[TMP6:%.*]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %b, <2 x double> [[VECINIT1_I]], <2 x double> %a)

clang/test/CodeGen/aarch64-poly64.c

 // CHECK:   [[VECINIT_I:%.*]] = insertelement <2 x i64> undef, i64 %a, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <2 x i64> [[VECINIT_I]], i64 %a, i32 1
 // CHECK:   ret <2 x i64> [[VECINIT1_I]]
 poly64x2_t test_vmovq_n_p64(poly64_t a) {
   return vmovq_n_p64(a);
 }
 
 // CHECK-LABEL: define <1 x i64> @test_vdup_lane_p64(<1 x i64> %vec) #0 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <1 x i64> %vec, <1 x i64> %vec, <1 x i32> zeroinitializer
-// CHECK:   ret <1 x i64> [[SHUFFLE]]
+// CHECK:    [[TMP0:%.*]] = bitcast <1 x i64> [[VEC:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x i64>
+// CHECK:    [[LANE:%.*]] = shufflevector <1 x i64> [[TMP1]], <1 x i64> [[TMP1]], <1 x i32> zeroinitializer
+// CHECK:    ret <1 x i64> [[LANE]]
 poly64x1_t test_vdup_lane_p64(poly64x1_t vec) {
   return vdup_lane_p64(vec, 0);
 }
 
 // CHECK-LABEL: define <2 x i64> @test_vdupq_lane_p64(<1 x i64> %vec) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <1 x i64> %vec, <1 x i64> %vec, <2 x i32> zeroinitializer
-// CHECK:   ret <2 x i64> [[SHUFFLE]]
+// CHECK:    [[TMP0:%.*]] = bitcast <1 x i64> [[VEC:%.*]] to <8 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x i64>
+// CHECK:    [[LANE:%.*]] = shufflevector <1 x i64> [[TMP1]], <1 x i64> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK:    ret <2 x i64> [[LANE]]
 poly64x2_t test_vdupq_lane_p64(poly64x1_t vec) {
   return vdupq_lane_p64(vec, 0);
 }
 
 // CHECK-LABEL: define <2 x i64> @test_vdupq_laneq_p64(<2 x i64> %vec) #1 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i64> %vec, <2 x i64> %vec, <2 x i32> <i32 1, i32 1>
-// CHECK:   ret <2 x i64> [[SHUFFLE]]
+// CHECK:    [[TMP0:%.*]] = bitcast <2 x i64> [[VEC:%.*]] to <16 x i8>
+// CHECK:    [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <2 x i64>
+// CHECK:    [[LANE:%.*]] = shufflevector <2 x i64> [[TMP1]], <2 x i64> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:    ret <2 x i64> [[LANE]]
 poly64x2_t test_vdupq_laneq_p64(poly64x2_t vec) {
   return vdupq_laneq_p64(vec, 1);
 }
 
 // CHECK-LABEL: define <2 x i64> @test_vcombine_p64(<1 x i64> %low, <1 x i64> %high) #1 {
 // CHECK:   [[SHUFFLE_I:%.*]] = shufflevector <1 x i64> %low, <1 x i64> %high, <2 x i32> <i32 0, i32 1>
 // CHECK:   ret <2 x i64> [[SHUFFLE_I]]
 poly64x2_t test_vcombine_p64(poly64x1_t low, poly64x1_t high) {

clang/test/CodeGen/aarch64-v8.2a-neon-intrinsics.c

 // CHECK: [[EXTR:%.*]] = extractelement <8 x half> %c, i32 7
 // CHECK: [[FMA:%.*]]  = call half @llvm.fma.f16(half [[SUB]], half [[EXTR]], half %a)
 // CHECK: ret half [[FMA]]
 float16_t test_vfmsh_laneq_f16(float16_t a, float16_t b, float16x8_t c) {
   return vfmsh_laneq_f16(a, b, c, 7);
 }
 
 // CHECK-LABEL: test_vmul_lane_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK: [[MUL:%.*]]  = fmul <4 x half> %a, [[TMP0]]
+// CHECK: [[TMP0:%.*]] = bitcast <4 x half> [[B:%.*]] to <8 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK: [[MUL:%.*]] = fmul <4 x half> [[A:%.*]], [[LANE]]
 // CHECK: ret <4 x half> [[MUL]]
 float16x4_t test_vmul_lane_f16(float16x4_t a, float16x4_t b) {
   return vmul_lane_f16(a, b, 3);
 }
 
 // CHECK-LABEL: test_vmulq_lane_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK: [[MUL:%.*]]  = fmul <8 x half> %a, [[TMP0]]
+// CHECK: [[TMP0:%.*]] = bitcast <4 x half> [[B:%.*]] to <8 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK: [[MUL:%.*]] = fmul <8 x half> [[A:%.*]], [[LANE]]
 // CHECK: ret <8 x half> [[MUL]]
 float16x8_t test_vmulq_lane_f16(float16x8_t a, float16x4_t b) {
-  return vmulq_lane_f16(a, b, 7);
+  return vmulq_lane_f16(a, b, 3);
 }
 
 // CHECK-LABEL: test_vmul_laneq_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <8 x half> %b, <8 x half> %b, <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK: [[MUL:%.*]]  = fmul <4 x half> %a, [[TMP0]]
+// CHECK: [[TMP0:%.*]] = bitcast <8 x half> [[B:%.*]] to <16 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP1]], <8 x half> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK: [[MUL:%.*]] = fmul <4 x half> [[A:%.*]], [[LANE]]
 // CHECK: ret <4 x half> [[MUL]]
 float16x4_t test_vmul_laneq_f16(float16x4_t a, float16x8_t b) {
   return vmul_laneq_f16(a, b, 7);
 }
 
 // CHECK-LABEL: test_vmulq_laneq_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <8 x half> %b, <8 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK: [[MUL:%.*]]  = fmul <8 x half> %a, [[TMP0]]
+// CHECK: [[TMP0:%.*]] = bitcast <8 x half> [[B:%.*]] to <16 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP1]], <8 x half> [[TMP1]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK: [[MUL:%.*]] = fmul <8 x half> [[A:%.*]], [[LANE]]
 // CHECK: ret <8 x half> [[MUL]]
 float16x8_t test_vmulq_laneq_f16(float16x8_t a, float16x8_t b) {
   return vmulq_laneq_f16(a, b, 7);
 }
 
 // CHECK-LABEL: test_vmul_n_f16
 // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half %b, i32 0
 // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %b, i32 1

 // CHECK: [[MUL:%.*]]   = fmul float [[CONV0:%.*]], [[CONV0:%.*]]
 // CHECK: [[CONV3:%.*]] = fptrunc float %mul to half
 // CHECK: ret half [[CONV3:%.*]]
 float16_t test_vmulh_laneq_f16(float16_t a, float16x8_t b) {
   return vmulh_laneq_f16(a, b, 7);
 }
 
 // CHECK-LABEL: test_vmulx_lane_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK: [[MUL:%.*]] = call <4 x half> @llvm.aarch64.neon.fmulx.v4f16(<4 x half> %a, <4 x half> [[TMP0]])
-// CHECK: ret <4 x half> [[MUL]]
+// CHECK: [[TMP0:%.*]] = bitcast <4 x half> [[B:%.*]] to <8 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK: [[TMP2:%.*]] = bitcast <4 x half> [[A:%.*]] to <8 x i8>
+// CHECK: [[TMP3:%.*]] = bitcast <4 x half> [[LANE]] to <8 x i8>
+// CHECK: [[VMULX2_I:%.*]] = call <4 x half> @llvm.aarch64.neon.fmulx.v4f16(<4 x half> [[A]], <4 x half> [[LANE]]) #4
+// CHECK: ret <4 x half> [[VMULX2_I]]
 float16x4_t test_vmulx_lane_f16(float16x4_t a, float16x4_t b) {
   return vmulx_lane_f16(a, b, 3);
 }
 
 // CHECK-LABEL: test_vmulxq_lane_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK: [[MUL:%.*]] = call <8 x half> @llvm.aarch64.neon.fmulx.v8f16(<8 x half> %a, <8 x half> [[TMP0]])
-// CHECK: ret <8 x half> [[MUL]]
+// CHECK: [[TMP0:%.*]] = bitcast <4 x half> [[B:%.*]] to <8 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK: [[TMP2:%.*]] = bitcast <8 x half> [[A:%.*]] to <16 x i8>
+// CHECK: [[TMP3:%.*]] = bitcast <8 x half> [[LANE]] to <16 x i8>
+// CHECK: [[VMULX2_I:%.*]] = call <8 x half> @llvm.aarch64.neon.fmulx.v8f16(<8 x half> [[A]], <8 x half> [[LANE]]) #4
+// CHECK: ret <8 x half> [[VMULX2_I]]
 float16x8_t test_vmulxq_lane_f16(float16x8_t a, float16x4_t b) {
-  return vmulxq_lane_f16(a, b, 7);
+  return vmulxq_lane_f16(a, b, 3);
 }
 
 // CHECK-LABEL: test_vmulx_laneq_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <8 x half> %b, <8 x half> %b, <4 x i32> <i32 7, i32 7, i32 7, i32 7>
-// CHECK: [[MUL:%.*]]  = call <4 x half> @llvm.aarch64.neon.fmulx.v4f16(<4 x half> %a, <4 x half> [[TMP0]])
-// CHECK: ret <4 x half> [[MUL]]
+// CHECK: [[TMP0:%.*]] = bitcast <8 x half> [[B:%.*]] to <16 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP1]], <8 x half> [[TMP1]], <4 x i32> <i32 7, i32 7, i32 7, i32 7>
+// CHECK: [[TMP2:%.*]] = bitcast <4 x half> [[A:%.*]] to <8 x i8>
+// CHECK: [[TMP3:%.*]] = bitcast <4 x half> [[LANE]] to <8 x i8>
+// CHECK: [[VMULX2_I:%.*]] = call <4 x half> @llvm.aarch64.neon.fmulx.v4f16(<4 x half> [[A]], <4 x half> [[LANE]]) #4
+// CHECK: ret <4 x half> [[VMULX2_I]]
 float16x4_t test_vmulx_laneq_f16(float16x4_t a, float16x8_t b) {
   return vmulx_laneq_f16(a, b, 7);
 }
 
 // CHECK-LABEL: test_vmulxq_laneq_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <8 x half> %b, <8 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK: [[MUL:%.*]]  = call <8 x half> @llvm.aarch64.neon.fmulx.v8f16(<8 x half> %a, <8 x half> [[TMP0]])
-// CHECK: ret <8 x half> [[MUL]]
+// CHECK: [[TMP0:%.*]] = bitcast <8 x half> [[B:%.*]] to <16 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <8 x half> [[TMP1]], <8 x half> [[TMP1]], <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+// CHECK: [[TMP2:%.*]] = bitcast <8 x half> [[A:%.*]] to <16 x i8>
+// CHECK: [[TMP3:%.*]] = bitcast <8 x half> [[LANE]] to <16 x i8>
+// CHECK: [[VMULX2_I:%.*]] = call <8 x half> @llvm.aarch64.neon.fmulx.v8f16(<8 x half> [[A]], <8 x half> [[LANE]]) #4
+// CHECK: ret <8 x half> [[VMULX2_I]]
 float16x8_t test_vmulxq_laneq_f16(float16x8_t a, float16x8_t b) {
   return vmulxq_laneq_f16(a, b, 7);
 }
 
 // CHECK-LABEL: test_vmulx_n_f16
 // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half %b, i32 0
 // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half %b, i32 1
 // CHECK: [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half %b, i32 2

 // CHECK:   [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half %a, i32 6
 // CHECK:   [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half %a, i32 7
 // CHECK:   ret <8 x half> [[TMP7]]
 float16x8_t test_vdupq_n_f16(float16_t a) {
   return vdupq_n_f16(a);
 }
 
 // CHECK-LABEL: test_vdup_lane_f16
-// CHECK:   [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %a, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   ret <4 x half> [[SHFL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x half> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <4 x half> [[LANE]]
 float16x4_t test_vdup_lane_f16(float16x4_t a) {
   return vdup_lane_f16(a, 3);
 }
 
 // CHECK-LABEL: test_vdupq_lane_f16
-// CHECK:   [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %a, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK:   ret <8 x half> [[SHFL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x half> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <8 x half> [[LANE]]
 float16x8_t test_vdupq_lane_f16(float16x4_t a) {
-  return vdupq_lane_f16(a, 7);
+  return vdupq_lane_f16(a, 3);
 }
 
 // CHECK-LABEL: @test_vext_f16(
 // CHECK:   [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8>
 // CHECK:   [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8>
 // CHECK:   [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
 // CHECK:   [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half>
 // CHECK:   [[VEXT:%.*]] = shufflevector <4 x half> [[TMP2]], <4 x half> [[TMP3]], <4 x i32> <i32 2, i32 3, i32 4, i32 5>

clang/test/CodeGen/arm-neon-range-checks.c

+// RUN: %clang_cc1 -triple arm64-none-eabi -target-feature +neon -target-feature +dotprod -target-feature +v8.1a -verify %s
+// RUN: %clang_cc1 -triple armv8.1a-none-eabi -target-feature +neon -target-feature +dotprod -target-feature +v8.1a -verify %s
+
+#include <arm_neon.h>
+
+void test_vdot_lane(int32x2_t r, int8x8_t a, int8x8_t b) {
+  vdot_lane_s32(r, a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vdot_lane_s32(r, a, b, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vdot_lane_s32(r, a, b, 0);
+  vdot_lane_s32(r, a, b, 1);
+}
+
+void test_vdotq_lane(int32x4_t r, int8x16_t a, int8x8_t b) {
+  vdotq_lane_s32(r, a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vdotq_lane_s32(r, a, b, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vdotq_lane_s32(r, a, b, 0);
+  vdotq_lane_s32(r, a, b, 1);
+}
+
+#if defined(__aarch64__)
+void test_vdot_laneq(int32x2_t r, int8x8_t a, int8x16_t b) {
+  vdot_laneq_s32(r, a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vdot_laneq_s32(r, a, b, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vdot_laneq_s32(r, a, b, 0);
+  vdot_laneq_s32(r, a, b, 3);
+}
+
+void test_vdotq_laneq(int32x4_t r, int8x16_t a, int8x16_t b) {
+  vdotq_laneq_s32(r, a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vdotq_laneq_s32(r, a, b, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vdotq_laneq_s32(r, a, b, 0);
+  vdotq_laneq_s32(r, a, b, 3);
+}
+#endif
+
+void test_vdup_lane(int32x2_t v) {
+  vdup_lane_s32(v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vdup_lane_s32(v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vdup_lane_s32(v, 0);
+  vdup_lane_s32(v, 1);
+}
+
+void test_vdupq_lane(int32x2_t v) {
+  vdupq_lane_s32(v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vdupq_lane_s32(v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vdupq_lane_s32(v, 0);
+  vdupq_lane_s32(v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vdup_laneq(int32x4_t v) {
+  vdup_laneq_s32(v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vdup_laneq_s32(v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vdup_laneq_s32(v, 0);
+  vdup_laneq_s32(v, 3);
+}
+
+void test_vdupq_laneq(int32x4_t v) {
+  vdupq_laneq_s32(v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vdupq_laneq_s32(v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vdupq_laneq_s32(v, 0);
+  vdupq_laneq_s32(v, 3);
+}
+#endif
+
+void test_vmla_lane(int32x2_t a, int32x2_t b, int32x2_t v) {
+  vmla_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmla_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmla_lane_s32(a, b, v, 0);
+  vmla_lane_s32(a, b, v, 1);
+}
+
+void test_vmlaq_lane(int32x4_t a, int32x4_t b, int32x2_t v) {
+  vmlaq_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmlaq_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmlaq_lane_s32(a, b, v, 0);
+  vmlaq_lane_s32(a, b, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vmla_laneq(int32x2_t a, int32x2_t b, int32x4_t v) {
+  vmla_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmla_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmla_laneq_s32(a, b, v, 0);
+  vmla_laneq_s32(a, b, v, 3);
+}
+
+void test_vmlaq_laneq(int32x4_t a, int32x4_t b, int32x4_t v) {
+  vmlaq_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmlaq_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmlaq_laneq_s32(a, b, v, 0);
+  vmlaq_laneq_s32(a, b, v, 3);
+}
+
+void test_vmlal_high_lane(int64x2_t a, int32x4_t b, int32x2_t v) {
+  vmlal_high_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmlal_high_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmlal_high_lane_s32(a, b, v, 0);
+  vmlal_high_lane_s32(a, b, v, 1);
+}
+
+void test_vmlal_high_laneq(int64x2_t a, int32x4_t b, int32x4_t v) {
+  vmlal_high_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmlal_high_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmlal_high_laneq_s32(a, b, v, 0);
+  vmlal_high_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vmlal_lane(int64x2_t a, int32x2_t b, int32x2_t v) {
+  vmlal_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmlal_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmlal_lane_s32(a, b, v, 0);
+  vmlal_lane_s32(a, b, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vmlal_laneq(int64x2_t a, int32x2_t b, int32x4_t v) {
+  vmlal_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmlal_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmlal_laneq_s32(a, b, v, 0);
+  vmlal_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vmls_lane(int32x2_t a, int32x2_t b, int32x2_t v) {
+  vmls_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmls_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmls_lane_s32(a, b, v, 0);
+  vmls_lane_s32(a, b, v, 1);
+}
+
+void test_vmlsq_lane(int32x4_t a, int32x4_t b, int32x2_t v) {
+  vmlsq_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmlsq_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmlsq_lane_s32(a, b, v, 0);
+  vmlsq_lane_s32(a, b, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vmls_laneq(int32x2_t a, int32x2_t b, int32x4_t v) {
+  vmls_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmls_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmls_laneq_s32(a, b, v, 0);
+  vmls_laneq_s32(a, b, v, 3);
+}
+
+void test_vmlsq_laneq(int32x4_t a, int32x4_t b, int32x4_t v) {
+  vmlsq_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmlsq_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmlsq_laneq_s32(a, b, v, 0);
+  vmlsq_laneq_s32(a, b, v, 3);
+}
+
+void test_vmlsl_high_lane(int64x2_t a, int32x4_t b, int32x2_t v) {
+  vmlsl_high_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmlsl_high_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmlsl_high_lane_s32(a, b, v, 0);
+  vmlsl_high_lane_s32(a, b, v, 1);
+}
+
+void test_vmlsl_high_laneq(int64x2_t a, int32x4_t b, int32x4_t v) {
+  vmlsl_high_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmlsl_high_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmlsl_high_laneq_s32(a, b, v, 0);
+  vmlsl_high_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vmlsl_lane(int64x2_t a, int32x2_t b, int32x2_t v) {
+  vmlsl_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmlsl_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmlsl_lane_s32(a, b, v, 0);
+  vmlsl_lane_s32(a, b, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vmlsl_laneq(int64x2_t a, int32x2_t b, int32x4_t v) {
+  vmlsl_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmlsl_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmlsl_laneq_s32(a, b, v, 0);
+  vmlsl_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vmull_lane(int32x2_t a, int32x2_t b) {
+  vmull_lane_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmull_lane_s32(a, b, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmull_lane_s32(a, b, 0);
+  vmull_lane_s32(a, b, 1);
+}
+
+#if defined(__aarch64__)
+void test_vmull_laneq(int32x2_t a, int32x4_t b) {
+  vmull_laneq_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmull_laneq_s32(a, b, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmull_laneq_s32(a, b, 0);
+  vmull_laneq_s32(a, b, 3);
+}
+
+void test_vmull_high_lane(int32x4_t a, int32x2_t b) {
+  vmull_high_lane_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vmull_high_lane_s32(a, b, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vmull_high_lane_s32(a, b, 0);
+  vmull_high_lane_s32(a, b, 1);
+}
+
+void test_vmull_high_laneq(int32x4_t a, int32x4_t b) {
+  vmull_high_laneq_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vmull_high_laneq_s32(a, b, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vmull_high_laneq_s32(a, b, 0);
+  vmull_high_laneq_s32(a, b, 3);
+}
+
+void test_vqdmlal_high_lane(int64x2_t a, int32x4_t b, int32x2_t v) {
+  vqdmlal_high_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqdmlal_high_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqdmlal_high_lane_s32(a, b, v, 0);
+  vqdmlal_high_lane_s32(a, b, v, 1);
+}
+
+void test_vqdmlal_high_laneq(int64x2_t a, int32x4_t b, int32x4_t v) {
+  vqdmlal_high_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqdmlal_high_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqdmlal_high_laneq_s32(a, b, v, 0);
+  vqdmlal_high_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vqdmlal_lane(int64x2_t a, int32x2_t b, int32x2_t v) {
+  vqdmlal_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqdmlal_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqdmlal_lane_s32(a, b, v, 0);
+  vqdmlal_lane_s32(a, b, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vqdmlal_laneq(int64x2_t a, int32x2_t b, int32x4_t v) {
+  vqdmlal_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqdmlal_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqdmlal_laneq_s32(a, b, v, 0);
+  vqdmlal_laneq_s32(a, b, v, 3);
+}
+
+void test_vqdmlsl_high_lane(int64x2_t a, int32x4_t b, int32x2_t v) {
+  vqdmlsl_high_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqdmlsl_high_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqdmlsl_high_lane_s32(a, b, v, 0);
+  vqdmlsl_high_lane_s32(a, b, v, 1);
+}
+
+void test_vqdmlsl_high_laneq(int64x2_t a, int32x4_t b, int32x4_t v) {
+  vqdmlsl_high_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqdmlsl_high_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqdmlsl_high_laneq_s32(a, b, v, 0);
+  vqdmlsl_high_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vqdmlsl_lane(int64x2_t a, int32x2_t b, int32x2_t v) {
+  vqdmlsl_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqdmlsl_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqdmlsl_lane_s32(a, b, v, 0);
+  vqdmlsl_lane_s32(a, b, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vqdmlsl_laneq(int64x2_t a, int32x2_t b, int32x4_t v) {
+  vqdmlsl_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqdmlsl_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqdmlsl_laneq_s32(a, b, v, 0);
+  vqdmlsl_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vqdmulh_lane(int32x2_t a, int32x2_t b) {
+  vqdmulh_lane_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqdmulh_lane_s32(a, b, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqdmulh_lane_s32(a, b, 0);
+  vqdmulh_lane_s32(a, b, 1);
+}
+
+#if defined(__aarch64__)
+void test_vqdmulh_laneq(int32x2_t a, int32x4_t b) {
+  vqdmulh_laneq_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqdmulh_laneq_s32(a, b, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqdmulh_laneq_s32(a, b, 0);
+  vqdmulh_laneq_s32(a, b, 3);
+}
+
+void test_vqdmulhq_laneq(int32x4_t a, int32x4_t b) {
+  vqdmulhq_laneq_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqdmulhq_laneq_s32(a, b, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqdmulhq_laneq_s32(a, b, 0);
+  vqdmulhq_laneq_s32(a, b, 3);
+}
+
+void test_vqdmull_high_lane(int32x4_t a, int32x2_t b) {
+  vqdmull_high_lane_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqdmull_high_lane_s32(a, b, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqdmull_high_lane_s32(a, b, 0);
+  vqdmull_high_lane_s32(a, b, 1);
+}
+
+void test_vqdmull_high_laneq(int32x4_t a, int32x4_t b) {
+  vqdmull_high_laneq_s32(a, b, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqdmull_high_laneq_s32(a, b, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqdmull_high_laneq_s32(a, b, 0);
+  vqdmull_high_laneq_s32(a, b, 3);
+}
+#endif
+
+void test_vqdmull_lane(int32x2_t a, int32x2_t v) {
+  vqdmull_lane_s32(a, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqdmull_lane_s32(a, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqdmull_lane_s32(a, v, 0);
+  vqdmull_lane_s32(a, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vqdmull_laneq(int32x2_t a, int32x4_t v) {
+  vqdmull_laneq_s32(a, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqdmull_laneq_s32(a, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqdmull_laneq_s32(a, v, 0);
+  vqdmull_laneq_s32(a, v, 3);
+}
+#endif
+
+void test_vqrdmlah_lane(int32x2_t a, int32x2_t b, int32x2_t v) {
+  vqrdmlah_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqrdmlah_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqrdmlah_lane_s32(a, b, v, 0);
+  vqrdmlah_lane_s32(a, b, v, 1);
+}
+
+void test_vqrdmlahq_lane(int32x4_t a, int32x4_t b, int32x2_t v) {
+  vqrdmlahq_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqrdmlahq_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqrdmlahq_lane_s32(a, b, v, 0);
+  vqrdmlahq_lane_s32(a, b, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vqrdmlah_laneq(int32x2_t a, int32x2_t b, int32x4_t v) {
+  vqrdmlah_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqrdmlah_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqrdmlah_laneq_s32(a, b, v, 0);
+  vqrdmlah_laneq_s32(a, b, v, 3);
+}
+
+void test_vqrdmlahq_laneq(int32x4_t a, int32x4_t b, int32x4_t v) {
+  vqrdmlahq_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqrdmlahq_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqrdmlahq_laneq_s32(a, b, v, 0);
+  vqrdmlahq_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vqrdmlsh_lane(int32x2_t a, int32x2_t b, int32x2_t v) {
+  vqrdmlsh_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqrdmlsh_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqrdmlsh_lane_s32(a, b, v, 0);
+  vqrdmlsh_lane_s32(a, b, v, 1);
+}
+
+void test_vqrdmlshq_lane(int32x4_t a, int32x4_t b, int32x2_t v) {
+  vqrdmlshq_lane_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqrdmlshq_lane_s32(a, b, v, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqrdmlshq_lane_s32(a, b, v, 0);
+  vqrdmlshq_lane_s32(a, b, v, 1);
+}
+
+#if defined(__aarch64__)
+void test_vqrdmlsh_laneq(int32x2_t a, int32x2_t b, int32x4_t v) {
+  vqrdmlsh_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqrdmlsh_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqrdmlsh_laneq_s32(a, b, v, 0);
+  vqrdmlsh_laneq_s32(a, b, v, 3);
+}
+
+void test_vqrdmlshq_laneq(int32x4_t a, int32x4_t b, int32x4_t v) {
+  vqrdmlshq_laneq_s32(a, b, v, -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqrdmlshq_laneq_s32(a, b, v, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqrdmlshq_laneq_s32(a, b, v, 0);
+  vqrdmlshq_laneq_s32(a, b, v, 3);
+}
+#endif
+
+void test_vqrdmulh_lane(int32x2_t a, int32x2_t v) {
+  vqrdmulh_lane_s32(a, v,  -1); // expected-error {{argument value -1 is outside the valid range [0, 1]}}
+  vqrdmulh_lane_s32(a, v,  2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  vqrdmulh_lane_s32(a, v,  0);
+  vqrdmulh_lane_s32(a, v,  1);
+}
+
+#if defined(__aarch64__)
+void test_vqrdmulh_laneq(int32x2_t a, int32x4_t v) {
+  vqrdmulh_laneq_s32(a, v,  -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqrdmulh_laneq_s32(a, v,  4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqrdmulh_laneq_s32(a, v,  0);
+  vqrdmulh_laneq_s32(a, v,  3);
+}
+
+void test_vqrdmulhq_laneq(int32x4_t a, int32x4_t v) {
+  vqrdmulhq_laneq_s32(a, v,  -1); // expected-error {{argument value -1 is outside the valid range [0, 3]}}
+  vqrdmulhq_laneq_s32(a, v,  4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  vqrdmulhq_laneq_s32(a, v,  0);
+  vqrdmulhq_laneq_s32(a, v,  3);
+}
+#endif

clang/test/CodeGen/arm-v8.2a-neon-intrinsics.c

 // CHECK:  [[SUB:%.*]] = fneg <8 x half> %b
 // CHECK:  [[ADD:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half> [[SUB]], <8 x half> %c, <8 x half> %a)
 // CHECK:  ret <8 x half> [[ADD]]
 float16x8_t test_vfmsq_f16(float16x8_t a, float16x8_t b, float16x8_t c) {
   return vfmsq_f16(a, b, c);
 }
 
 // CHECK-LABEL: test_vmul_lane_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK: [[MUL:%.*]]  = fmul <4 x half> %a, [[TMP0]]
+// CHECK: [[TMP0:%.*]] = bitcast <4 x half> [[B:%.*]] to <8 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK: [[MUL:%.*]] = fmul <4 x half> [[A:%.*]], [[LANE]]
 // CHECK: ret <4 x half> [[MUL]]
 float16x4_t test_vmul_lane_f16(float16x4_t a, float16x4_t b) {
   return vmul_lane_f16(a, b, 3);
 }
 
 // CHECK-LABEL: test_vmulq_lane_f16
-// CHECK: [[TMP0:%.*]] = shufflevector <4 x half> %b, <4 x half> %b, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK: [[MUL:%.*]]  = fmul <8 x half> %a, [[TMP0]]
+// CHECK: [[TMP0:%.*]] = bitcast <4 x half> [[B:%.*]] to <8 x i8>
+// CHECK: [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK: [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK: [[MUL:%.*]] = fmul <8 x half> [[A:%.*]], [[LANE]]
 // CHECK: ret <8 x half> [[MUL]]
 float16x8_t test_vmulq_lane_f16(float16x8_t a, float16x4_t b) {
-  return vmulq_lane_f16(a, b, 7);
+  return vmulq_lane_f16(a, b, 3);
 }
 
 // CHECK-LABEL: test_vmul_n_f16
 // CHECK: [[TMP0:%.*]] = insertelement <4 x half> undef, half [[b:%.*]], i32 0
 // CHECK: [[TMP1:%.*]] = insertelement <4 x half> [[TMP0]], half [[b]], i32 1
 // CHECK: [[TMP2:%.*]] = insertelement <4 x half> [[TMP1]], half [[b]], i32 2
 // CHECK: [[TMP3:%.*]] = insertelement <4 x half> [[TMP2]], half [[b]], i32 3
 // CHECK: [[MUL:%.*]]  = fmul <4 x half> %a, [[TMP3]]

 // CHECK:   [[TMP6:%.*]] = insertelement <8 x half> [[TMP5]], half [[ARG]], i32 6
 // CHECK:   [[TMP7:%.*]] = insertelement <8 x half> [[TMP6]], half [[ARG]], i32 7
 // CHECK:   ret <8 x half> [[TMP7]]
 float16x8_t test_vdupq_n_f16(float16_t a) {
   return vdupq_n_f16(a);
 }
 
 // CHECK-LABEL: test_vdup_lane_f16
-// CHECK:   [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %a, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   ret <4 x half> [[SHFL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x half> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <4 x half> [[LANE]]
 float16x4_t test_vdup_lane_f16(float16x4_t a) {
   return vdup_lane_f16(a, 3);
 }
 
 // CHECK-LABEL: test_vdupq_lane_f16
-// CHECK:   [[SHFL:%.*]] = shufflevector <4 x half> %a, <4 x half> %a, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-// CHECK:   ret <8 x half> [[SHFL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x half> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x half> [[TMP1]], <4 x half> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <8 x half> [[LANE]]
 float16x8_t test_vdupq_lane_f16(float16x4_t a) {
-  return vdupq_lane_f16(a, 7);
+  return vdupq_lane_f16(a, 3);
 }
 
 // CHECK-LABEL: @test_vext_f16(
 // CHECK:   [[TMP0:%.*]] = bitcast <4 x half> %a to <8 x i8>
 // CHECK:   [[TMP1:%.*]] = bitcast <4 x half> %b to <8 x i8>
 // CHECK:   [[TMP2:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x half>
 // CHECK:   [[TMP3:%.*]] = bitcast <8 x i8> [[TMP1]] to <4 x half>
 // CHECK:   [[VEXT:%.*]] = shufflevector <4 x half> [[TMP2]], <4 x half> [[TMP3]], <4 x i32> <i32 2, i32 3, i32 4, i32 5>

clang/test/CodeGen/arm64_vdupq_n_f64.c

 float32x4_t test_vdupq_n_f32(float32_t w) {
     return vdupq_n_f32(w);
 }
 
 // vdupq_lane_f64 -> dup.2d v0, v0[0]
 // this was in <rdar://problem/11778405>, but had already been implemented,
 // test anyway
 // CHECK-LABEL: define <2 x double> @test_vdupq_lane_f64(<1 x double> %V) #0 {
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <1 x double> %V, <1 x double> %V, <2 x i32> zeroinitializer
+// CHECK:   [[TMP0:%.*]] = bitcast <1 x double> %V to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x double>
+// CHECK:   [[SHUFFLE:%.*]] = shufflevector <1 x double> [[TMP1]], <1 x double> [[TMP1]], <2 x i32> zeroinitializer
 // CHECK:   ret <2 x double> [[SHUFFLE]]
 float64x2_t test_vdupq_lane_f64(float64x1_t V) {
     return vdupq_lane_f64(V, 0);
 }
 
 // vmovq_n_f64 -> dup Vd.2d,X0
 // this wasn't in <rdar://problem/11778405>, but it was between the vdups
 // CHECK-LABEL: define <2 x double> @test_vmovq_n_f64(double %w) #0 {

clang/test/CodeGen/arm_neon_intrinsics.c

 // CHECK-LABEL: @test_vdup_lane_u8(
 // CHECK:   [[SHUFFLE:%.*]] = shufflevector <8 x i8> %a, <8 x i8> %a, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
 // CHECK:   ret <8 x i8> [[SHUFFLE]]
 uint8x8_t test_vdup_lane_u8(uint8x8_t a) {
   return vdup_lane_u8(a, 7);
 }
 
 // CHECK-LABEL: @test_vdup_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %a, <4 x i16> %a, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   ret <4 x i16> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <4 x i16> [[LANE]]
 uint16x4_t test_vdup_lane_u16(uint16x4_t a) {
   return vdup_lane_u16(a, 3);
 }
 
 // CHECK-LABEL: @test_vdup_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %a, <2 x i32> %a, <2 x i32> <i32 1, i32 1>
-// CHECK:   ret <2 x i32> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   ret <2 x i32> [[LANE]]
 uint32x2_t test_vdup_lane_u32(uint32x2_t a) {
   return vdup_lane_u32(a, 1);
 }
 
 // CHECK-LABEL: @test_vdup_lane_s8(
 // CHECK:   [[SHUFFLE:%.*]] = shufflevector <8 x i8> %a, <8 x i8> %a, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
 // CHECK:   ret <8 x i8> [[SHUFFLE]]
 int8x8_t test_vdup_lane_s8(int8x8_t a) {
   return vdup_lane_s8(a, 7);
 }
 
 // CHECK-LABEL: @test_vdup_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %a, <4 x i16> %a, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   ret <4 x i16> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <4 x i16> [[LANE]]
 int16x4_t test_vdup_lane_s16(int16x4_t a) {
   return vdup_lane_s16(a, 3);
 }
 
 // CHECK-LABEL: @test_vdup_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %a, <2 x i32> %a, <2 x i32> <i32 1, i32 1>
-// CHECK:   ret <2 x i32> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   ret <2 x i32> [[LANE]]
 int32x2_t test_vdup_lane_s32(int32x2_t a) {
   return vdup_lane_s32(a, 1);
 }
 
 // CHECK-LABEL: @test_vdup_lane_p8(
 // CHECK:   [[SHUFFLE:%.*]] = shufflevector <8 x i8> %a, <8 x i8> %a, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
 // CHECK:   ret <8 x i8> [[SHUFFLE]]
 poly8x8_t test_vdup_lane_p8(poly8x8_t a) {
   return vdup_lane_p8(a, 7);
 }
 
 // CHECK-LABEL: @test_vdup_lane_p16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %a, <4 x i16> %a, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   ret <4 x i16> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <4 x i16> [[LANE]]
 poly16x4_t test_vdup_lane_p16(poly16x4_t a) {
   return vdup_lane_p16(a, 3);
 }
 
 // CHECK-LABEL: @test_vdup_lane_f32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %a, <2 x float> %a, <2 x i32> <i32 1, i32 1>
-// CHECK:   ret <2 x float> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   ret <2 x float> [[LANE]]
 float32x2_t test_vdup_lane_f32(float32x2_t a) {
   return vdup_lane_f32(a, 1);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_u8(
 // CHECK:   [[SHUFFLE:%.*]] = shufflevector <8 x i8> %a, <8 x i8> %a, <16 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
 // CHECK:   ret <16 x i8> [[SHUFFLE]]
 uint8x16_t test_vdupq_lane_u8(uint8x8_t a) {
   return vdupq_lane_u8(a, 7);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %a, <4 x i16> %a, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   ret <8 x i16> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <8 x i16> [[LANE]]
 uint16x8_t test_vdupq_lane_u16(uint16x4_t a) {
   return vdupq_lane_u16(a, 3);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %a, <2 x i32> %a, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   ret <4 x i32> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   ret <4 x i32> [[LANE]]
 uint32x4_t test_vdupq_lane_u32(uint32x2_t a) {
   return vdupq_lane_u32(a, 1);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_s8(
 // CHECK:   [[SHUFFLE:%.*]] = shufflevector <8 x i8> %a, <8 x i8> %a, <16 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
 // CHECK:   ret <16 x i8> [[SHUFFLE]]
 int8x16_t test_vdupq_lane_s8(int8x8_t a) {
   return vdupq_lane_s8(a, 7);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %a, <4 x i16> %a, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   ret <8 x i16> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <8 x i16> [[LANE]]
 int16x8_t test_vdupq_lane_s16(int16x4_t a) {
   return vdupq_lane_s16(a, 3);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %a, <2 x i32> %a, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   ret <4 x i32> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   ret <4 x i32> [[LANE]]
 int32x4_t test_vdupq_lane_s32(int32x2_t a) {
   return vdupq_lane_s32(a, 1);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_p8(
 // CHECK:   [[SHUFFLE:%.*]] = shufflevector <8 x i8> %a, <8 x i8> %a, <16 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
 // CHECK:   ret <16 x i8> [[SHUFFLE]]
 poly8x16_t test_vdupq_lane_p8(poly8x8_t a) {
   return vdupq_lane_p8(a, 7);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_p16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %a, <4 x i16> %a, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   ret <8 x i16> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   ret <8 x i16> [[LANE]]
 poly16x8_t test_vdupq_lane_p16(poly16x4_t a) {
   return vdupq_lane_p16(a, 3);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_f32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %a, <2 x float> %a, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   ret <4 x float> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   ret <4 x float> [[LANE]]
 float32x4_t test_vdupq_lane_f32(float32x2_t a) {
   return vdupq_lane_f32(a, 1);
 }
 
 // CHECK-LABEL: @test_vdup_lane_s64(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <1 x i64> %a, <1 x i64> %a, <1 x i32> zeroinitializer
-// CHECK:   ret <1 x i64> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <1 x i64> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x i64>
+// CHECK:   [[LANE:%.*]] = shufflevector <1 x i64> [[TMP1]], <1 x i64> [[TMP1]], <1 x i32> zeroinitializer
+// CHECK:   ret <1 x i64> [[LANE]]
 int64x1_t test_vdup_lane_s64(int64x1_t a) {
   return vdup_lane_s64(a, 0);
 }
 
 // CHECK-LABEL: @test_vdup_lane_u64(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <1 x i64> %a, <1 x i64> %a, <1 x i32> zeroinitializer
-// CHECK:   ret <1 x i64> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <1 x i64> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x i64>
+// CHECK:   [[LANE:%.*]] = shufflevector <1 x i64> [[TMP1]], <1 x i64> [[TMP1]], <1 x i32> zeroinitializer
+// CHECK:   ret <1 x i64> [[LANE]]
 uint64x1_t test_vdup_lane_u64(uint64x1_t a) {
   return vdup_lane_u64(a, 0);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_s64(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <1 x i64> %a, <1 x i64> %a, <2 x i32> zeroinitializer
-// CHECK:   ret <2 x i64> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <1 x i64> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x i64>
+// CHECK:   [[LANE:%.*]] = shufflevector <1 x i64> [[TMP1]], <1 x i64> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK:   ret <2 x i64> [[LANE]]
 int64x2_t test_vdupq_lane_s64(int64x1_t a) {
   return vdupq_lane_s64(a, 0);
 }
 
 // CHECK-LABEL: @test_vdupq_lane_u64(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <1 x i64> %a, <1 x i64> %a, <2 x i32> zeroinitializer
-// CHECK:   ret <2 x i64> [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <1 x i64> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <1 x i64>
+// CHECK:   [[LANE:%.*]] = shufflevector <1 x i64> [[TMP1]], <1 x i64> [[TMP1]], <2 x i32> zeroinitializer
+// CHECK:   ret <2 x i64> [[LANE]]
 uint64x2_t test_vdupq_lane_u64(uint64x1_t a) {
   return vdupq_lane_u64(a, 0);
 }
 
 // CHECK-LABEL: @test_vdup_n_u8(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <8 x i8> undef, i8 %a, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <8 x i8> [[VECINIT_I]], i8 %a, i32 1
 // CHECK:   [[VECINIT2_I:%.*]] = insertelement <8 x i8> [[VECINIT1_I]], i8 %a, i32 2

 // CHECK:   [[VMULL2_I_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> %b, <2 x i32> %c)
 // CHECK:   [[ADD_I:%.*]] = add <2 x i64> %a, [[VMULL2_I_I]]
 // CHECK:   ret <2 x i64> [[ADD_I]]
 uint64x2_t test_vmlal_u32(uint64x2_t a, uint32x2_t b, uint32x2_t c) {
   return vmlal_u32(a, b, c);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %b to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> %b, <4 x i16> [[SHUFFLE]])
-// CHECK:   [[ADD:%.*]] = add <4 x i32> %a, [[VMULL2_I]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #8
+// CHECK:   [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK:   ret <4 x i32> [[ADD]]
 int32x4_t test_vmlal_lane_s16(int32x4_t a, int16x4_t b, int16x4_t c) {
   return vmlal_lane_s16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %b to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmulls.v2i64(<2 x i32> %b, <2 x i32> [[SHUFFLE]])
-// CHECK:   [[ADD:%.*]] = add <2 x i64> %a, [[VMULL2_I]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmulls.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #8
+// CHECK:   [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK:   ret <2 x i64> [[ADD]]
 int64x2_t test_vmlal_lane_s32(int64x2_t a, int32x2_t b, int32x2_t c) {
   return vmlal_lane_s32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %b to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16> %b, <4 x i16> [[SHUFFLE]])
-// CHECK:   [[ADD:%.*]] = add <4 x i32> %a, [[VMULL2_I]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #8
+// CHECK:   [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK:   ret <4 x i32> [[ADD]]
 uint32x4_t test_vmlal_lane_u16(uint32x4_t a, uint16x4_t b, uint16x4_t c) {
   return vmlal_lane_u16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmlal_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %b to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> %b, <2 x i32> [[SHUFFLE]])
-// CHECK:   [[ADD:%.*]] = add <2 x i64> %a, [[VMULL2_I]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #8
+// CHECK:   [[ADD:%.*]] = add <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK:   ret <2 x i64> [[ADD]]
 uint64x2_t test_vmlal_lane_u32(uint64x2_t a, uint32x2_t b, uint32x2_t c) {
   return vmlal_lane_u32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlal_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %c, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <4 x i16> [[VECINIT_I]], i16 %c, i32 1

 // CHECK:   [[VMULL2_I_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> %b, <2 x i32> [[VECINIT1_I]])
 // CHECK:   [[ADD_I:%.*]] = add <2 x i64> %a, [[VMULL2_I_I]]
 // CHECK:   ret <2 x i64> [[ADD_I]]
 uint64x2_t test_vmlal_n_u32(uint64x2_t a, uint32x2_t b, uint32_t c) {
   return vmlal_n_u32(a, b, c);
 }
 
 // CHECK-LABEL: @test_vmla_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <4 x i16> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = add <4 x i16> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x i16> [[ADD]]
 int16x4_t test_vmla_lane_s16(int16x4_t a, int16x4_t b, int16x4_t c) {
   return vmla_lane_s16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmla_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <2 x i32> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = add <2 x i32> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK:   ret <2 x i32> [[ADD]]
 int32x2_t test_vmla_lane_s32(int32x2_t a, int32x2_t b, int32x2_t c) {
   return vmla_lane_s32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmla_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <4 x i16> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = add <4 x i16> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = add <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x i16> [[ADD]]
 uint16x4_t test_vmla_lane_u16(uint16x4_t a, uint16x4_t b, uint16x4_t c) {
   return vmla_lane_u16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmla_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <2 x i32> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = add <2 x i32> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = add <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK:   ret <2 x i32> [[ADD]]
 uint32x2_t test_vmla_lane_u32(uint32x2_t a, uint32x2_t b, uint32x2_t c) {
   return vmla_lane_u32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmla_lane_f32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %c, <2 x float> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <2 x float> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = fadd <2 x float> [[A:%.*]], [[MUL]]
 // CHECK:   ret <2 x float> [[ADD]]
 float32x2_t test_vmla_lane_f32(float32x2_t a, float32x2_t b, float32x2_t c) {
   return vmla_lane_f32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <8 x i16> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = add <8 x i16> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK:   ret <8 x i16> [[ADD]]
 int16x8_t test_vmlaq_lane_s16(int16x8_t a, int16x8_t b, int16x4_t c) {
   return vmlaq_lane_s16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <4 x i32> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = add <4 x i32> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x i32> [[ADD]]
 int32x4_t test_vmlaq_lane_s32(int32x4_t a, int32x4_t b, int32x2_t c) {
   return vmlaq_lane_s32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <8 x i16> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = add <8 x i16> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = add <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK:   ret <8 x i16> [[ADD]]
 uint16x8_t test_vmlaq_lane_u16(uint16x8_t a, uint16x8_t b, uint16x4_t c) {
   return vmlaq_lane_u16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <4 x i32> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = add <4 x i32> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = add <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x i32> [[ADD]]
 uint32x4_t test_vmlaq_lane_u32(uint32x4_t a, uint32x4_t b, uint32x2_t c) {
   return vmlaq_lane_u32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlaq_lane_f32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %c, <2 x float> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[ADD:%.*]] = fadd <4 x float> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:   [[ADD:%.*]] = fadd <4 x float> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x float> [[ADD]]
 float32x4_t test_vmlaq_lane_f32(float32x4_t a, float32x4_t b, float32x2_t c) {
   return vmlaq_lane_f32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmla_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %c, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <4 x i16> [[VECINIT_I]], i16 %c, i32 1

 // CHECK:   [[VMULL2_I_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> %b, <2 x i32> %c)
 // CHECK:   [[SUB_I:%.*]] = sub <2 x i64> %a, [[VMULL2_I_I]]
 // CHECK:   ret <2 x i64> [[SUB_I]]
 uint64x2_t test_vmlsl_u32(uint64x2_t a, uint32x2_t b, uint32x2_t c) {
   return vmlsl_u32(a, b, c);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %b to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> %b, <4 x i16> [[SHUFFLE]])
-// CHECK:   [[SUB:%.*]] = sub <4 x i32> %a, [[VMULL2_I]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #8
+// CHECK:   [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK:   ret <4 x i32> [[SUB]]
 int32x4_t test_vmlsl_lane_s16(int32x4_t a, int16x4_t b, int16x4_t c) {
   return vmlsl_lane_s16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %b to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmulls.v2i64(<2 x i32> %b, <2 x i32> [[SHUFFLE]])
-// CHECK:   [[SUB:%.*]] = sub <2 x i64> %a, [[VMULL2_I]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmulls.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #8
+// CHECK:   [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK:   ret <2 x i64> [[SUB]]
 int64x2_t test_vmlsl_lane_s32(int64x2_t a, int32x2_t b, int32x2_t c) {
   return vmlsl_lane_s32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %b to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16> %b, <4 x i16> [[SHUFFLE]])
-// CHECK:   [[SUB:%.*]] = sub <4 x i32> %a, [[VMULL2_I]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #8
+// CHECK:   [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[VMULL2_I]]
 // CHECK:   ret <4 x i32> [[SUB]]
 uint32x4_t test_vmlsl_lane_u16(uint32x4_t a, uint16x4_t b, uint16x4_t c) {
   return vmlsl_lane_u16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmlsl_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %b to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> %b, <2 x i32> [[SHUFFLE]])
-// CHECK:   [[SUB:%.*]] = sub <2 x i64> %a, [[VMULL2_I]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #8
+// CHECK:   [[SUB:%.*]] = sub <2 x i64> [[A:%.*]], [[VMULL2_I]]
 // CHECK:   ret <2 x i64> [[SUB]]
 uint64x2_t test_vmlsl_lane_u32(uint64x2_t a, uint32x2_t b, uint32x2_t c) {
   return vmlsl_lane_u32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlsl_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %c, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <4 x i16> [[VECINIT_I]], i16 %c, i32 1

 // CHECK:   [[VMULL2_I_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> %b, <2 x i32> [[VECINIT1_I]])
 // CHECK:   [[SUB_I:%.*]] = sub <2 x i64> %a, [[VMULL2_I_I]]
 // CHECK:   ret <2 x i64> [[SUB_I]]
 uint64x2_t test_vmlsl_n_u32(uint64x2_t a, uint32x2_t b, uint32_t c) {
   return vmlsl_n_u32(a, b, c);
 }
 
 // CHECK-LABEL: @test_vmls_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <4 x i16> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = sub <4 x i16> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x i16> [[SUB]]
 int16x4_t test_vmls_lane_s16(int16x4_t a, int16x4_t b, int16x4_t c) {
   return vmls_lane_s16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmls_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <2 x i32> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = sub <2 x i32> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK:   ret <2 x i32> [[SUB]]
 int32x2_t test_vmls_lane_s32(int32x2_t a, int32x2_t b, int32x2_t c) {
   return vmls_lane_s32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmls_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <4 x i16> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = sub <4 x i16> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <4 x i16> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = sub <4 x i16> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x i16> [[SUB]]
 uint16x4_t test_vmls_lane_u16(uint16x4_t a, uint16x4_t b, uint16x4_t c) {
   return vmls_lane_u16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmls_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <2 x i32> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = sub <2 x i32> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <2 x i32> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = sub <2 x i32> [[A:%.*]], [[MUL]]
 // CHECK:   ret <2 x i32> [[SUB]]
 uint32x2_t test_vmls_lane_u32(uint32x2_t a, uint32x2_t b, uint32x2_t c) {
   return vmls_lane_u32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmls_lane_f32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %c, <2 x float> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <2 x float> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = fmul <2 x float> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = fsub <2 x float> [[A:%.*]], [[MUL]]
 // CHECK:   ret <2 x float> [[SUB]]
 float32x2_t test_vmls_lane_f32(float32x2_t a, float32x2_t b, float32x2_t c) {
   return vmls_lane_f32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <8 x i16> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = sub <8 x i16> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK:   ret <8 x i16> [[SUB]]
 int16x8_t test_vmlsq_lane_s16(int16x8_t a, int16x8_t b, int16x4_t c) {
   return vmlsq_lane_s16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <4 x i32> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = sub <4 x i32> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x i32> [[SUB]]
 int32x4_t test_vmlsq_lane_s32(int32x4_t a, int32x4_t b, int32x2_t c) {
   return vmlsq_lane_s32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <8 x i16> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = sub <8 x i16> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <8 x i16> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = sub <8 x i16> [[A:%.*]], [[MUL]]
 // CHECK:   ret <8 x i16> [[SUB]]
 uint16x8_t test_vmlsq_lane_u16(uint16x8_t a, uint16x8_t b, uint16x4_t c) {
   return vmlsq_lane_u16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <4 x i32> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = sub <4 x i32> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <4 x i32> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = sub <4 x i32> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x i32> [[SUB]]
 uint32x4_t test_vmlsq_lane_u32(uint32x4_t a, uint32x4_t b, uint32x2_t c) {
   return vmlsq_lane_u32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmlsq_lane_f32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %c, <2 x float> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %b, [[SHUFFLE]]
-// CHECK:   [[SUB:%.*]] = fsub <4 x float> %a, [[MUL]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = fmul <4 x float> [[B:%.*]], [[LANE]]
+// CHECK:   [[SUB:%.*]] = fsub <4 x float> [[A:%.*]], [[MUL]]
 // CHECK:   ret <4 x float> [[SUB]]
 float32x4_t test_vmlsq_lane_f32(float32x4_t a, float32x4_t b, float32x2_t c) {
   return vmlsq_lane_f32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vmls_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %c, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <4 x i16> [[VECINIT_I]], i16 %c, i32 1

 // CHECK-LABEL: @test_vmull_p8(
 // CHECK:   [[VMULL_I:%.*]] = call <8 x i16> @llvm.arm.neon.vmullp.v8i16(<8 x i8> %a, <8 x i8> %b)
 // CHECK:   ret <8 x i16> [[VMULL_I]]
 poly16x8_t test_vmull_p8(poly8x8_t a, poly8x8_t b) {
   return vmull_p8(a, b);
 }
 
 // CHECK-LABEL: @test_vmull_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> %a, <4 x i16> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmulls.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #8
 // CHECK:   ret <4 x i32> [[VMULL2_I]]
 int32x4_t test_vmull_lane_s16(int16x4_t a, int16x4_t b) {
   return vmull_lane_s16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vmull_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmulls.v2i64(<2 x i32> %a, <2 x i32> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmulls.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #8
 // CHECK:   ret <2 x i64> [[VMULL2_I]]
 int64x2_t test_vmull_lane_s32(int32x2_t a, int32x2_t b) {
   return vmull_lane_s32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vmull_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16> %a, <4 x i16> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vmullu.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #8
 // CHECK:   ret <4 x i32> [[VMULL2_I]]
 uint32x4_t test_vmull_lane_u16(uint16x4_t a, uint16x4_t b) {
   return vmull_lane_u16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vmull_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> %a, <2 x i32> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VMULL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vmullu.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #8
 // CHECK:   ret <2 x i64> [[VMULL2_I]]
 uint64x2_t test_vmull_lane_u32(uint32x2_t a, uint32x2_t b) {
   return vmull_lane_u32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vmull_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %b, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <4 x i16> [[VECINIT_I]], i16 %b, i32 1

 // CHECK-LABEL: @test_vmulq_p8(
 // CHECK:   [[VMULQ_V_I:%.*]] = call <16 x i8> @llvm.arm.neon.vmulp.v16i8(<16 x i8> %a, <16 x i8> %b)
 // CHECK:   ret <16 x i8> [[VMULQ_V_I]]
 poly8x16_t test_vmulq_p8(poly8x16_t a, poly8x16_t b) {
   return vmulq_p8(a, b);
 }
 
 // CHECK-LABEL: @test_vmul_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <4 x i16> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK:   ret <4 x i16> [[MUL]]
 int16x4_t test_vmul_lane_s16(int16x4_t a, int16x4_t b) {
   return vmul_lane_s16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vmul_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <2 x i32> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK:   ret <2 x i32> [[MUL]]
 int32x2_t test_vmul_lane_s32(int32x2_t a, int32x2_t b) {
   return vmul_lane_s32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vmul_lane_f32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %b, <2 x float> %b, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <2 x float> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = fmul <2 x float> [[A:%.*]], [[LANE]]
 // CHECK:   ret <2 x float> [[MUL]]
 float32x2_t test_vmul_lane_f32(float32x2_t a, float32x2_t b) {
   return vmul_lane_f32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vmul_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <4 x i16> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <4 x i16> [[A:%.*]], [[LANE]]
 // CHECK:   ret <4 x i16> [[MUL]]
 uint16x4_t test_vmul_lane_u16(uint16x4_t a, uint16x4_t b) {
   return vmul_lane_u16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vmul_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <2 x i32> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <2 x i32> [[A:%.*]], [[LANE]]
 // CHECK:   ret <2 x i32> [[MUL]]
 uint32x2_t test_vmul_lane_u32(uint32x2_t a, uint32x2_t b) {
   return vmul_lane_u32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <8 x i16> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK:   ret <8 x i16> [[MUL]]
 int16x8_t test_vmulq_lane_s16(int16x8_t a, int16x4_t b) {
   return vmulq_lane_s16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <4 x i32> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK:   ret <4 x i32> [[MUL]]
 int32x4_t test_vmulq_lane_s32(int32x4_t a, int32x2_t b) {
   return vmulq_lane_s32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_f32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x float> %b, <2 x float> %b, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = fmul <4 x float> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x float> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x float>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = fmul <4 x float> [[A:%.*]], [[LANE]]
 // CHECK:   ret <4 x float> [[MUL]]
 float32x4_t test_vmulq_lane_f32(float32x4_t a, float32x2_t b) {
   return vmulq_lane_f32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_u16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[MUL:%.*]] = mul <8 x i16> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[MUL:%.*]] = mul <8 x i16> [[A:%.*]], [[LANE]]
 // CHECK:   ret <8 x i16> [[MUL]]
 uint16x8_t test_vmulq_lane_u16(uint16x8_t a, uint16x4_t b) {
   return vmulq_lane_u16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vmulq_lane_u32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[MUL:%.*]] = mul <4 x i32> %a, [[SHUFFLE]]
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[MUL:%.*]] = mul <4 x i32> [[A:%.*]], [[LANE]]
 // CHECK:   ret <4 x i32> [[MUL]]
 uint32x4_t test_vmulq_lane_u32(uint32x4_t a, uint32x2_t b) {
   return vmulq_lane_u32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vmul_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %b, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <4 x i16> [[VECINIT_I]], i16 %b, i32 1

 // CHECK:   [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> %b, <2 x i32> %c)
 // CHECK:   [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> %a, <2 x i64> [[VQDMLAL2_I]])
 // CHECK:   ret <2 x i64> [[VQDMLAL_V3_I]]
 int64x2_t test_vqdmlal_s32(int64x2_t a, int32x2_t b, int32x2_t c) {
   return vqdmlal_s32(a, b, c);
 }
 
 // CHECK-LABEL: @test_vqdmlal_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i32> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> %b to <8 x i8>
-// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmull.v4i32(<4 x i16> %b, <4 x i16> [[SHUFFLE]])
-// CHECK:   [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> %a, <4 x i32> [[VQDMLAL2_I]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #8
+// CHECK:   [[VQDMLAL_V3_I:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #8
 // CHECK:   ret <4 x i32> [[VQDMLAL_V3_I]]
 int32x4_t test_vqdmlal_lane_s16(int32x4_t a, int16x4_t b, int16x4_t c) {
   return vqdmlal_lane_s16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlal_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i64> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> %b to <8 x i8>
-// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> %b, <2 x i32> [[SHUFFLE]])
-// CHECK:   [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> %a, <2 x i64> [[VQDMLAL2_I]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #8
+// CHECK:   [[VQDMLAL_V3_I:%.*]] = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #8
 // CHECK:   ret <2 x i64> [[VQDMLAL_V3_I]]
 int64x2_t test_vqdmlal_lane_s32(int64x2_t a, int32x2_t b, int32x2_t c) {
   return vqdmlal_lane_s32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vqdmlal_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %c, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <4 x i16> [[VECINIT_I]], i16 %c, i32 1

 // CHECK:   [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> %b, <2 x i32> %c)
 // CHECK:   [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.ssub.sat.v2i64(<2 x i64> %a, <2 x i64> [[VQDMLAL2_I]])
 // CHECK:   ret <2 x i64> [[VQDMLSL_V3_I]]
 int64x2_t test_vqdmlsl_s32(int64x2_t a, int32x2_t b, int32x2_t c) {
   return vqdmlsl_s32(a, b, c);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %c, <4 x i16> %c, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i32> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> %b to <8 x i8>
-// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmull.v4i32(<4 x i16> %b, <4 x i16> [[SHUFFLE]])
-// CHECK:   [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32> %a, <4 x i32> [[VQDMLAL2_I]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP4:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VQDMLAL2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmull.v4i32(<4 x i16> [[B]], <4 x i16> [[LANE]]) #8
+// CHECK:   [[VQDMLSL_V3_I:%.*]] = call <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32> [[A]], <4 x i32> [[VQDMLAL2_I]]) #8
 // CHECK:   ret <4 x i32> [[VQDMLSL_V3_I]]
 int32x4_t test_vqdmlsl_lane_s16(int32x4_t a, int16x4_t b, int16x4_t c) {
   return vqdmlsl_lane_s16(a, b, c, 3);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %c, <2 x i32> %c, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i64> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> %b to <8 x i8>
-// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> %b, <2 x i32> [[SHUFFLE]])
-// CHECK:   [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.ssub.sat.v2i64(<2 x i64> %a, <2 x i64> [[VQDMLAL2_I]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[C:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i64> [[A:%.*]] to <16 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP4:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VQDMLAL2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> [[B]], <2 x i32> [[LANE]]) #8
+// CHECK:   [[VQDMLSL_V3_I:%.*]] = call <2 x i64> @llvm.ssub.sat.v2i64(<2 x i64> [[A]], <2 x i64> [[VQDMLAL2_I]]) #8
 // CHECK:   ret <2 x i64> [[VQDMLSL_V3_I]]
 int64x2_t test_vqdmlsl_lane_s32(int64x2_t a, int32x2_t b, int32x2_t c) {
   return vqdmlsl_lane_s32(a, b, c, 1);
 }
 
 // CHECK-LABEL: @test_vqdmlsl_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %c, i32 0
 // CHECK:   [[VECINIT1_I:%.*]] = insertelement <4 x i16> [[VECINIT_I]], i16 %c, i32 1

 // CHECK:   [[VQDMULHQ_V2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmulh.v4i32(<4 x i32> %a, <4 x i32> %b)
 // CHECK:   [[VQDMULHQ_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULHQ_V2_I]] to <16 x i8>
 // CHECK:   ret <4 x i32> [[VQDMULHQ_V2_I]]
 int32x4_t test_vqdmulhq_s32(int32x4_t a, int32x4_t b) {
   return vqdmulhq_s32(a, b);
 }
 
 // CHECK-LABEL: @test_vqdmulh_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQDMULH_V2_I:%.*]] = call <4 x i16> @llvm.arm.neon.vqdmulh.v4i16(<4 x i16> %a, <4 x i16> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VQDMULH_V2_I:%.*]] = call <4 x i16> @llvm.arm.neon.vqdmulh.v4i16(<4 x i16> [[A]], <4 x i16> [[LANE]]) #8
 // CHECK:   [[VQDMULH_V3_I:%.*]] = bitcast <4 x i16> [[VQDMULH_V2_I]] to <8 x i8>
 // CHECK:   ret <4 x i16> [[VQDMULH_V2_I]]
 int16x4_t test_vqdmulh_lane_s16(int16x4_t a, int16x4_t b) {
   return vqdmulh_lane_s16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vqdmulh_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQDMULH_V2_I:%.*]] = call <2 x i32> @llvm.arm.neon.vqdmulh.v2i32(<2 x i32> %a, <2 x i32> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VQDMULH_V2_I:%.*]] = call <2 x i32> @llvm.arm.neon.vqdmulh.v2i32(<2 x i32> [[A]], <2 x i32> [[LANE]]) #8
 // CHECK:   [[VQDMULH_V3_I:%.*]] = bitcast <2 x i32> [[VQDMULH_V2_I]] to <8 x i8>
 // CHECK:   ret <2 x i32> [[VQDMULH_V2_I]]
 int32x2_t test_vqdmulh_lane_s32(int32x2_t a, int32x2_t b) {
   return vqdmulh_lane_s32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vqdmulhq_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <8 x i16> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <8 x i16> [[SHUFFLE]] to <16 x i8>
-// CHECK:   [[VQDMULHQ_V2_I:%.*]] = call <8 x i16> @llvm.arm.neon.vqdmulh.v8i16(<8 x i16> %a, <8 x i16> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <8 x i16> [[A:%.*]] to <16 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <8 x i16> [[LANE]] to <16 x i8>
+// CHECK:   [[VQDMULHQ_V2_I:%.*]] = call <8 x i16> @llvm.arm.neon.vqdmulh.v8i16(<8 x i16> [[A]], <8 x i16> [[LANE]]) #8
 // CHECK:   [[VQDMULHQ_V3_I:%.*]] = bitcast <8 x i16> [[VQDMULHQ_V2_I]] to <16 x i8>
 // CHECK:   ret <8 x i16> [[VQDMULHQ_V2_I]]
 int16x8_t test_vqdmulhq_lane_s16(int16x8_t a, int16x4_t b) {
   return vqdmulhq_lane_s16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vqdmulhq_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i32> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i32> [[SHUFFLE]] to <16 x i8>
-// CHECK:   [[VQDMULHQ_V2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmulh.v4i32(<4 x i32> %a, <4 x i32> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i32> [[LANE]] to <16 x i8>
+// CHECK:   [[VQDMULHQ_V2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmulh.v4i32(<4 x i32> [[A]], <4 x i32> [[LANE]]) #8
 // CHECK:   [[VQDMULHQ_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULHQ_V2_I]] to <16 x i8>
 // CHECK:   ret <4 x i32> [[VQDMULHQ_V2_I]]
 int32x4_t test_vqdmulhq_lane_s32(int32x4_t a, int32x2_t b) {
   return vqdmulhq_lane_s32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vqdmulh_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %b, i32 0

 // CHECK:   [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> %a, <2 x i32> %b)
 // CHECK:   [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK:   ret <2 x i64> [[VQDMULL_V2_I]]
 int64x2_t test_vqdmull_s32(int32x2_t a, int32x2_t b) {
   return vqdmull_s32(a, b);
 }
 
 // CHECK-LABEL: @test_vqdmull_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmull.v4i32(<4 x i16> %a, <4 x i16> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VQDMULL_V2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqdmull.v4i32(<4 x i16> [[A]], <4 x i16> [[LANE]]) #8
 // CHECK:   [[VQDMULL_V3_I:%.*]] = bitcast <4 x i32> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK:   ret <4 x i32> [[VQDMULL_V2_I]]
 int32x4_t test_vqdmull_lane_s16(int16x4_t a, int16x4_t b) {
   return vqdmull_lane_s16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vqdmull_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> %a, <2 x i32> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VQDMULL_V2_I:%.*]] = call <2 x i64> @llvm.arm.neon.vqdmull.v2i64(<2 x i32> [[A]], <2 x i32> [[LANE]]) #8
 // CHECK:   [[VQDMULL_V3_I:%.*]] = bitcast <2 x i64> [[VQDMULL_V2_I]] to <16 x i8>
 // CHECK:   ret <2 x i64> [[VQDMULL_V2_I]]
 int64x2_t test_vqdmull_lane_s32(int32x2_t a, int32x2_t b) {
   return vqdmull_lane_s32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vqdmull_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %b, i32 0

 // CHECK:   [[VQRDMULHQ_V2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32> %a, <4 x i32> %b)
 // CHECK:   [[VQRDMULHQ_V3_I:%.*]] = bitcast <4 x i32> [[VQRDMULHQ_V2_I]] to <16 x i8>
 // CHECK:   ret <4 x i32> [[VQRDMULHQ_V2_I]]
 int32x4_t test_vqrdmulhq_s32(int32x4_t a, int32x4_t b) {
   return vqrdmulhq_s32(a, b);
 }
 
 // CHECK-LABEL: @test_vqrdmulh_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i16> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQRDMULH_V2_I:%.*]] = call <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16> %a, <4 x i16> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <4 x i32> <i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i16> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i16> [[LANE]] to <8 x i8>
+// CHECK:   [[VQRDMULH_V2_I:%.*]] = call <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16> [[A]], <4 x i16> [[LANE]]) #8
 // CHECK:   [[VQRDMULH_V3_I:%.*]] = bitcast <4 x i16> [[VQRDMULH_V2_I]] to <8 x i8>
 // CHECK:   ret <4 x i16> [[VQRDMULH_V2_I]]
 int16x4_t test_vqrdmulh_lane_s16(int16x4_t a, int16x4_t b) {
   return vqrdmulh_lane_s16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vqrdmulh_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <2 x i32> <i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> %a to <8 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <2 x i32> [[SHUFFLE]] to <8 x i8>
-// CHECK:   [[VQRDMULH_V2_I:%.*]] = call <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32> %a, <2 x i32> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <2 x i32> <i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <2 x i32> [[LANE]] to <8 x i8>
+// CHECK:   [[VQRDMULH_V2_I:%.*]] = call <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32> [[A]], <2 x i32> [[LANE]]) #8
 // CHECK:   [[VQRDMULH_V3_I:%.*]] = bitcast <2 x i32> [[VQRDMULH_V2_I]] to <8 x i8>
 // CHECK:   ret <2 x i32> [[VQRDMULH_V2_I]]
 int32x2_t test_vqrdmulh_lane_s32(int32x2_t a, int32x2_t b) {
   return vqrdmulh_lane_s32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vqrdmulhq_lane_s16(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <4 x i16> %b, <4 x i16> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-// CHECK:   [[TMP0:%.*]] = bitcast <8 x i16> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <8 x i16> [[SHUFFLE]] to <16 x i8>
-// CHECK:   [[VQRDMULHQ_V2_I:%.*]] = call <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16> %a, <8 x i16> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <4 x i16> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <4 x i16>
+// CHECK:   [[LANE:%.*]] = shufflevector <4 x i16> [[TMP1]], <4 x i16> [[TMP1]], <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+// CHECK:   [[TMP2:%.*]] = bitcast <8 x i16> [[A:%.*]] to <16 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <8 x i16> [[LANE]] to <16 x i8>
+// CHECK:   [[VQRDMULHQ_V2_I:%.*]] = call <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16> [[A]], <8 x i16> [[LANE]]) #8
 // CHECK:   [[VQRDMULHQ_V3_I:%.*]] = bitcast <8 x i16> [[VQRDMULHQ_V2_I]] to <16 x i8>
 // CHECK:   ret <8 x i16> [[VQRDMULHQ_V2_I]]
 int16x8_t test_vqrdmulhq_lane_s16(int16x8_t a, int16x4_t b) {
   return vqrdmulhq_lane_s16(a, b, 3);
 }
 
 // CHECK-LABEL: @test_vqrdmulhq_lane_s32(
-// CHECK:   [[SHUFFLE:%.*]] = shufflevector <2 x i32> %b, <2 x i32> %b, <4 x i32> <i32 1, i32 1, i32 1, i32 1>
-// CHECK:   [[TMP0:%.*]] = bitcast <4 x i32> %a to <16 x i8>
-// CHECK:   [[TMP1:%.*]] = bitcast <4 x i32> [[SHUFFLE]] to <16 x i8>
-// CHECK:   [[VQRDMULHQ_V2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32> %a, <4 x i32> [[SHUFFLE]])
+// CHECK:   [[TMP0:%.*]] = bitcast <2 x i32> [[B:%.*]] to <8 x i8>
+// CHECK:   [[TMP1:%.*]] = bitcast <8 x i8> [[TMP0]] to <2 x i32>
+// CHECK:   [[LANE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> [[TMP1]], <4 x i32> <i32 1, i32 1, i32 1, i32 1>
+// CHECK:   [[TMP2:%.*]] = bitcast <4 x i32> [[A:%.*]] to <16 x i8>
+// CHECK:   [[TMP3:%.*]] = bitcast <4 x i32> [[LANE]] to <16 x i8>
+// CHECK:   [[VQRDMULHQ_V2_I:%.*]] = call <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32> [[A]], <4 x i32> [[LANE]]) #8
 // CHECK:   [[VQRDMULHQ_V3_I:%.*]] = bitcast <4 x i32> [[VQRDMULHQ_V2_I]] to <16 x i8>
 // CHECK:   ret <4 x i32> [[VQRDMULHQ_V2_I]]
 int32x4_t test_vqrdmulhq_lane_s32(int32x4_t a, int32x2_t b) {
   return vqrdmulhq_lane_s32(a, b, 1);
 }
 
 // CHECK-LABEL: @test_vqrdmulh_n_s16(
 // CHECK:   [[VECINIT_I:%.*]] = insertelement <4 x i16> undef, i16 %b, i32 0

clang/utils/TableGen/NeonEmitter.cpp

void makeScalar() {
     NumVectors = 0;
   }
 
   void makeOneVector() {
     assert(isVector());
     NumVectors = 1;
   }
 
+  void make32BitElement() {
+    assert_with_loc(Bitwidth > 32, "Not enough bits to make it 32!");
+    ElementBitwidth = 32;
+  }
+
   void doubleLanes() {
     assert_with_loc(Bitwidth != 128, "Can't get bigger than 128!");
     Bitwidth = 128;
   }
 
   void halveLanes() {
     assert_with_loc(Bitwidth != 64, "Can't get smaller than 64!");
     Bitwidth = 64;
if (!DI->getArgNameStr(ArgIdx).empty()) {
       } else if (SI->getAsUnquotedString() == "S") {
         castToType.makeSigned();
       } else if (SI->getAsUnquotedString() == "H") {
         castToType.halveLanes();
       } else if (SI->getAsUnquotedString() == "D") {
         castToType.doubleLanes();
       } else if (SI->getAsUnquotedString() == "8") {
         castToType.makeInteger(8, true);
+      } else if (SI->getAsUnquotedString() == "32") {
+        castToType.make32BitElement();
       } else {
         castToType = Type::fromTypedefName(SI->getAsUnquotedString());
         assert_with_loc(!castToType.isVoid(), "Unknown typedef");
       }
     }
   }
 
   std::string S;