Index: llvm/lib/Target/ARM/MVEGatherScatterLowering.cpp
===================================================================
--- llvm/lib/Target/ARM/MVEGatherScatterLowering.cpp
+++ llvm/lib/Target/ARM/MVEGatherScatterLowering.cpp
@@ -84,8 +84,11 @@
 
 static bool isLegalTypeAndAlignment(unsigned NumElements, unsigned ElemSize,
                                     unsigned Alignment) {
-  // Do only allow non-extending v4i32 gathers for now
-  return NumElements == 4 && ElemSize == 32 && Alignment >= 4;
+  // Do only allow non-extending gathers for now
+  return ((NumElements == 4 && ElemSize == 32) ||
+          (NumElements == 8 && ElemSize == 16) ||
+          (NumElements == 16 && ElemSize == 8)) &&
+         ElemSize / 8 <= Alignment;
 }
 
 static bool LowerGather(IntrinsicInst *I) {
@@ -125,18 +128,97 @@
   }
   assert(Ptr->getType()->isVectorTy() && "Unexpected pointer type");
 
-  if (Ty->getVectorNumElements() != 4)
-    // Can't build an intrinsic for this
-    return false;
-  if (match(Mask, m_One()))
-    Load = Builder.CreateIntrinsic(Intrinsic::arm_mve_vldr_gather_base,
-                                   {Ty, Ptr->getType()},
-                                   {Ptr, Builder.getInt32(0)});
-  else
-    Load = Builder.CreateIntrinsic(
-        Intrinsic::arm_mve_vldr_gather_base_predicated,
-        {Ty, Ptr->getType(), Mask->getType()},
-        {Ptr, Builder.getInt32(0), Mask});
+  unsigned Unsigned = 1;
+  Value *BasePtr = Ptr;
+  GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr);
+  if (!GEP) {
+    if (Ty->getVectorNumElements() != 4)
+      // Can't build an intrinsic for this
+      return false;
+
+    LLVM_DEBUG(dbgs() << "masked gathers: no getelementpointer found. Loading"
+                      << " from vector of pointers\n");
+    if (match(Mask, m_One()))
+      Load = Builder.CreateIntrinsic(Intrinsic::arm_mve_vldr_gather_base,
+                                     {Ty, Ptr->getType()},
+                                     {Ptr, Builder.getInt32(0)});
+    else
+      Load = Builder.CreateIntrinsic(
+          Intrinsic::arm_mve_vldr_gather_base_predicated,
+          {Ty, Ptr->getType(), Mask->getType()},
+          {Ptr, Builder.getInt32(0), Mask});
+  } else {
+    LLVM_DEBUG(dbgs() << "masked gathers: getelementpointer found. Loading"
+                      << " from base + vector of offsets\n");
+    // Load from base + vector of offsets
+    Value *Offsets = GEP->getOperand(1);
+    if (ZExtInst *ZextOffs = dyn_cast<ZExtInst>(Offsets))
+      Offsets = ZextOffs->getOperand(0);
+    // SExt offsets inside masked gathers are not permitted by the architecture;
+    // we therefore can't fold them
+    Value *GEPPtr = GEP->getPointerOperand();
+    if (!GEPPtr->getType()->isVectorTy())
+      BasePtr = GEPPtr;
+
+    unsigned FinalIndex = GEP->getNumOperands() - 1;
+    // Ensure all the other indices are 0.
+    for (unsigned i = 1; i < FinalIndex; ++i) {
+      auto *C = dyn_cast<Constant>(GEP->getOperand(i));
+      if (!C) {
+        return false;
+      }
+      if (isa<VectorType>(C->getType()))
+        C = C->getSplatValue();
+      auto *CI = dyn_cast_or_null<ConstantInt>(C);
+      if (!CI || !CI->isZero()) {
+        return false;
+      }
+    }
+
+    Type *OffsType = VectorType::getInteger(cast<VectorType>(Ty));
+    // If the offset we found does not have the type the intrinsic expects,
+    // i.e., the same type as the gather itself, we need to convert it (only i
+    // types) or fall back to expanding the gather
+    if (OffsType != Offsets->getType()) {
+      if (OffsType->getScalarSizeInBits() >=
+          Offsets->getType()->getScalarSizeInBits()) {
+        LLVM_DEBUG(dbgs() << "masked gathers: extending offsets\n");
+        Offsets = Builder.CreateZExt(Offsets, OffsType, "");
+      } else {
+        LLVM_DEBUG(dbgs() << "masked gathers: no correct offset type. can't"
+                          << " create masked gather\n");
+        return false;
+      }
+    }
+    unsigned Scale;
+    LLVMContext &C = I->getContext();
+    if (BasePtr->getType() == Type::getInt32PtrTy(C)
+        && Ty->getScalarSizeInBits() == 32) {
+      Scale = 2;
+    } else if (BasePtr->getType() == Type::getInt16PtrTy(C)
+        && Ty->getScalarSizeInBits() == 16) {
+      Scale = 1;
+    } else if (BasePtr->getType() == Type::getInt8PtrTy(C)) {
+      Scale = 0;
+    } else {
+      return false;
+    }
+
+    if (!match(Mask, m_One()))
+      Load = Builder.CreateIntrinsic(
+          Intrinsic::arm_mve_vldr_gather_offset_predicated,
+          {Ty, BasePtr->getType(), Offsets->getType(), Mask->getType()},
+          {BasePtr, Offsets,
+           Builder.getInt32(Ty->getScalarSizeInBits()),
+           Builder.getInt32(Scale), Builder.getInt32(Unsigned), Mask});
+    else
+      Load = Builder.CreateIntrinsic(
+          Intrinsic::arm_mve_vldr_gather_offset,
+          {Ty, BasePtr->getType(), Offsets->getType()},
+          {BasePtr, Offsets,
+           Builder.getInt32(Ty->getScalarSizeInBits()),
+           Builder.getInt32(Scale), Builder.getInt32(Unsigned)});
+  }
 
   if (!isa<UndefValue>(PassThru) && !match(PassThru, m_Zero())) {
     LLVM_DEBUG(dbgs() << "masked gathers: found non-trivial passthru - "
Index: llvm/test/CodeGen/Thumb2/mve-gather-ind16-scaled.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-gather-ind16-scaled.ll
+++ llvm/test/CodeGen/Thumb2/mve-gather-ind16-scaled.ll
@@ -5,39 +5,9 @@
 define arm_aapcs_vfpcc <8 x i16> @scaled_v8i16_i16(i16* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: scaled_v8i16_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r7, lr}
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r2, s2
-; CHECK-NEXT:    vmov r3, s3
-; CHECK-NEXT:    vmov r5, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    vmov r4, s7
-; CHECK-NEXT:    ldrh.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    ldrh.w lr, [r3]
-; CHECK-NEXT:    vmov r3, s6
-; CHECK-NEXT:    ldrh r5, [r5]
-; CHECK-NEXT:    ldrh r0, [r0]
-; CHECK-NEXT:    ldrh r1, [r1]
-; CHECK-NEXT:    ldrh r4, [r4]
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    ldrh r3, [r3]
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    vmov.16 q0[1], r5
-; CHECK-NEXT:    vmov.16 q0[2], r12
-; CHECK-NEXT:    vmov.16 q0[3], lr
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov.16 q0[5], r1
-; CHECK-NEXT:    vmov.16 q0[6], r3
-; CHECK-NEXT:    vmov.16 q0[7], r4
-; CHECK-NEXT:    pop {r4, r5, r7, pc}
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1, uxtw #1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
@@ -50,44 +20,8 @@
 define arm_aapcs_vfpcc <8 x half> @scaled_v8f16_i16(i16* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: scaled_v8f16_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vldr.16 s0, [r2]
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    vldr.16 s0, [r3]
-; CHECK-NEXT:    vmov r3, s0
-; CHECK-NEXT:    vmov.16 q0[0], r3
-; CHECK-NEXT:    vmov.16 q0[1], r2
-; CHECK-NEXT:    vmov r2, s6
-; CHECK-NEXT:    vldr.16 s8, [r2]
-; CHECK-NEXT:    vmov r2, s8
-; CHECK-NEXT:    vmov.16 q0[2], r2
-; CHECK-NEXT:    vmov r2, s7
-; CHECK-NEXT:    vldr.16 s4, [r2]
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vmov.16 q0[3], r2
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vldr.16 s4, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q0[7], r0
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
@@ -250,39 +184,9 @@
 define arm_aapcs_vfpcc <8 x i16> @unsigned_scaled_v8i16_i8(i16* %base, <8 x i8>* %offptr) {
 ; CHECK-LABEL: unsigned_scaled_v8i16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r7, lr}
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r2, s2
-; CHECK-NEXT:    vmov r3, s3
-; CHECK-NEXT:    vmov r5, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    vmov r4, s7
-; CHECK-NEXT:    ldrh.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    ldrh.w lr, [r3]
-; CHECK-NEXT:    vmov r3, s6
-; CHECK-NEXT:    ldrh r5, [r5]
-; CHECK-NEXT:    ldrh r0, [r0]
-; CHECK-NEXT:    ldrh r1, [r1]
-; CHECK-NEXT:    ldrh r4, [r4]
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    ldrh r3, [r3]
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    vmov.16 q0[1], r5
-; CHECK-NEXT:    vmov.16 q0[2], r12
-; CHECK-NEXT:    vmov.16 q0[3], lr
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov.16 q0[5], r1
-; CHECK-NEXT:    vmov.16 q0[6], r3
-; CHECK-NEXT:    vmov.16 q0[7], r4
-; CHECK-NEXT:    pop {r4, r5, r7, pc}
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1, uxtw #1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
@@ -295,44 +199,8 @@
 define arm_aapcs_vfpcc <8 x half> @unsigned_scaled_v8f16_i8(i16* %base, <8 x i8>* %offptr) {
 ; CHECK-LABEL: unsigned_scaled_v8f16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vmov r2, s5
-; CHECK-NEXT:    vldr.16 s0, [r2]
-; CHECK-NEXT:    vmov r3, s4
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    vldr.16 s0, [r3]
-; CHECK-NEXT:    vmov r3, s0
-; CHECK-NEXT:    vmov.16 q0[0], r3
-; CHECK-NEXT:    vmov.16 q0[1], r2
-; CHECK-NEXT:    vmov r2, s6
-; CHECK-NEXT:    vldr.16 s8, [r2]
-; CHECK-NEXT:    vmov r2, s8
-; CHECK-NEXT:    vmov.16 q0[2], r2
-; CHECK-NEXT:    vmov r2, s7
-; CHECK-NEXT:    vldr.16 s4, [r2]
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vmov.16 q0[3], r2
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vldr.16 s4, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q0[7], r0
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1, uxtw #1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
@@ -346,39 +214,9 @@
 define arm_aapcs_vfpcc <8 x i16> @scaled_v8i16_i16_passthru0t(i16* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: scaled_v8i16_i16_passthru0t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r7, lr}
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r2, s2
-; CHECK-NEXT:    vmov r3, s3
-; CHECK-NEXT:    vmov r5, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    vmov r4, s7
-; CHECK-NEXT:    ldrh.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    ldrh.w lr, [r3]
-; CHECK-NEXT:    vmov r3, s6
-; CHECK-NEXT:    ldrh r5, [r5]
-; CHECK-NEXT:    ldrh r0, [r0]
-; CHECK-NEXT:    ldrh r1, [r1]
-; CHECK-NEXT:    ldrh r4, [r4]
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    ldrh r3, [r3]
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    vmov.16 q0[1], r5
-; CHECK-NEXT:    vmov.16 q0[2], r12
-; CHECK-NEXT:    vmov.16 q0[3], lr
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov.16 q0[5], r1
-; CHECK-NEXT:    vmov.16 q0[6], r3
-; CHECK-NEXT:    vmov.16 q0[7], r4
-; CHECK-NEXT:    pop {r4, r5, r7, pc}
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1, uxtw #1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
@@ -390,39 +228,9 @@
 define arm_aapcs_vfpcc <8 x i16> @scaled_v8i16_i16_passthru1t(i16* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: scaled_v8i16_i16_passthru1t:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r7, lr}
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r2, s2
-; CHECK-NEXT:    vmov r3, s3
-; CHECK-NEXT:    vmov r5, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    vmov r4, s7
-; CHECK-NEXT:    ldrh.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    ldrh.w lr, [r3]
-; CHECK-NEXT:    vmov r3, s6
-; CHECK-NEXT:    ldrh r5, [r5]
-; CHECK-NEXT:    ldrh r0, [r0]
-; CHECK-NEXT:    ldrh r1, [r1]
-; CHECK-NEXT:    ldrh r4, [r4]
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    ldrh r3, [r3]
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    vmov.16 q0[1], r5
-; CHECK-NEXT:    vmov.16 q0[2], r12
-; CHECK-NEXT:    vmov.16 q0[3], lr
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov.16 q0[5], r1
-; CHECK-NEXT:    vmov.16 q0[6], r3
-; CHECK-NEXT:    vmov.16 q0[7], r4
-; CHECK-NEXT:    pop {r4, r5, r7, pc}
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1, uxtw #1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
@@ -434,38 +242,14 @@
 define arm_aapcs_vfpcc <8 x i16> @scaled_v8i16_i16_passthru1f(i16* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: scaled_v8i16_i16_passthru1f:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r7, lr}
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    movs r5, #1
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r2, s3
-; CHECK-NEXT:    vmov r4, s1
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    vmov r3, s6
-; CHECK-NEXT:    ldrh.w lr, [r0]
-; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    ldrh.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s7
-; CHECK-NEXT:    ldrh r4, [r4]
-; CHECK-NEXT:    ldrh r1, [r1]
-; CHECK-NEXT:    ldrh r3, [r3]
-; CHECK-NEXT:    ldrh r0, [r0]
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    vmov.16 q0[0], r0
-; CHECK-NEXT:    vmov.16 q0[1], r4
-; CHECK-NEXT:    vmov.16 q0[2], r5
-; CHECK-NEXT:    vmov.16 q0[3], r12
-; CHECK-NEXT:    vmov.16 q0[4], lr
-; CHECK-NEXT:    vmov.16 q0[5], r1
-; CHECK-NEXT:    vmov.16 q0[6], r3
-; CHECK-NEXT:    vmov.16 q0[7], r2
-; CHECK-NEXT:    pop {r4, r5, r7, pc}
+; CHECK-NEXT:    movw r2, #65487
+; CHECK-NEXT:    vmov.i16 q0, #0x1
+; CHECK-NEXT:    vmsr p0, r2
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vpst
+; CHECK-NEXT:    vldrht.u16 q2, [r0, q1, uxtw #1]
+; CHECK-NEXT:    vpsel q0, q2, q0
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
@@ -477,38 +261,12 @@
 define arm_aapcs_vfpcc <8 x i16> @scaled_v8i16_i16_passthru0f(i16* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: scaled_v8i16_i16_passthru0f:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r7, lr}
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    movs r5, #0
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r2, s2
-; CHECK-NEXT:    vmov r4, s0
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    vmov r3, s6
-; CHECK-NEXT:    ldrh.w lr, [r0]
-; CHECK-NEXT:    vmov r0, s3
-; CHECK-NEXT:    ldrh.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s7
-; CHECK-NEXT:    ldrh r4, [r4]
-; CHECK-NEXT:    ldrh r1, [r1]
-; CHECK-NEXT:    vmov.16 q0[0], r4
-; CHECK-NEXT:    ldrh r3, [r3]
-; CHECK-NEXT:    vmov.16 q0[1], r5
-; CHECK-NEXT:    vmov.16 q0[2], r12
-; CHECK-NEXT:    ldrh r0, [r0]
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    vmov.16 q0[3], r0
-; CHECK-NEXT:    vmov.16 q0[4], lr
-; CHECK-NEXT:    vmov.16 q0[5], r1
-; CHECK-NEXT:    vmov.16 q0[6], r3
-; CHECK-NEXT:    vmov.16 q0[7], r2
-; CHECK-NEXT:    pop {r4, r5, r7, pc}
+; CHECK-NEXT:    movw r2, #65523
+; CHECK-NEXT:    vmsr p0, r2
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vpst
+; CHECK-NEXT:    vldrht.u16 q0, [r0, q1, uxtw #1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
@@ -520,110 +278,10 @@
 define arm_aapcs_vfpcc <8 x i16> @scaled_v8i16_i16_passthru_icmp0(i16* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: scaled_v8i16_i16_passthru_icmp0:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r7, lr}
-; CHECK-NEXT:    push {r7, lr}
-; CHECK-NEXT:    .pad #8
-; CHECK-NEXT:    sub sp, #8
-; CHECK-NEXT:    vldrh.u16 q0, [r1]
-; CHECK-NEXT:    mov.w r12, #0
-; CHECK-NEXT:    vcmp.s16 gt, q0, zr
-; CHECK-NEXT:    vmov.u16 r2, q0[4]
-; CHECK-NEXT:    vmrs r1, p0
-; CHECK-NEXT:    vmov.32 q1[0], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[5]
-; CHECK-NEXT:    vmov.u16 lr, q0[7]
-; CHECK-NEXT:    vmov.32 q1[1], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[6]
-; CHECK-NEXT:    vmov.32 q1[2], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[0]
-; CHECK-NEXT:    vmov.32 q2[0], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[1]
-; CHECK-NEXT:    vmov.32 q2[1], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[2]
-; CHECK-NEXT:    vmov.32 q2[2], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[3]
-; CHECK-NEXT:    vmov.32 q2[3], r2
-; CHECK-NEXT:    vmovlb.u16 q0, q2
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vadd.i32 q2, q0, r0
-; CHECK-NEXT:    and r3, r1, #1
-; CHECK-NEXT:    rsbs r2, r3, #0
-; CHECK-NEXT:    movs r3, #0
-; CHECK-NEXT:    bfi r3, r2, #0, #1
-; CHECK-NEXT:    ubfx r2, r1, #2, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #1, #1
-; CHECK-NEXT:    ubfx r2, r1, #4, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #2, #1
-; CHECK-NEXT:    ubfx r2, r1, #6, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #3, #1
-; CHECK-NEXT:    ubfx r2, r1, #8, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #4, #1
-; CHECK-NEXT:    ubfx r2, r1, #10, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #5, #1
-; CHECK-NEXT:    ubfx r2, r1, #12, #1
-; CHECK-NEXT:    ubfx r1, r1, #14, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #6, #1
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r3, r1, #7, #1
-; CHECK-NEXT:    uxtb r1, r3
-; CHECK-NEXT:    lsls r2, r3, #31
-; CHECK-NEXT:    beq .LBB11_2
-; CHECK-NEXT:  @ %bb.1: @ %cond.load
-; CHECK-NEXT:    vmov r2, s8
-; CHECK-NEXT:    vdup.16 q0, r12
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    b .LBB11_3
-; CHECK-NEXT:  .LBB11_2:
-; CHECK-NEXT:    vmov.i32 q0, #0x0
-; CHECK-NEXT:  .LBB11_3: @ %else
-; CHECK-NEXT:    vmov.32 q1[3], lr
-; CHECK-NEXT:    lsls r2, r1, #30
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s9
-; CHECK-NEXT:    ldrhmi r2, [r2]
-; CHECK-NEXT:    vmovmi.16 q0[1], r2
-; CHECK-NEXT:    vmovlb.u16 q1, q1
-; CHECK-NEXT:    lsls r2, r1, #29
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s10
-; CHECK-NEXT:    ldrhmi r2, [r2]
-; CHECK-NEXT:    vmovmi.16 q0[2], r2
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    lsls r2, r1, #28
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s11
-; CHECK-NEXT:    ldrhmi r2, [r2]
-; CHECK-NEXT:    vmovmi.16 q0[3], r2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    lsls r0, r1, #27
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r0, s4
-; CHECK-NEXT:    ldrhmi r0, [r0]
-; CHECK-NEXT:    vmovmi.16 q0[4], r0
-; CHECK-NEXT:    lsls r0, r1, #26
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r0, s5
-; CHECK-NEXT:    ldrhmi r0, [r0]
-; CHECK-NEXT:    vmovmi.16 q0[5], r0
-; CHECK-NEXT:    lsls r0, r1, #25
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r0, s6
-; CHECK-NEXT:    ldrhmi r0, [r0]
-; CHECK-NEXT:    vmovmi.16 q0[6], r0
-; CHECK-NEXT:    lsls r0, r1, #24
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r0, s7
-; CHECK-NEXT:    ldrhmi r0, [r0]
-; CHECK-NEXT:    vmovmi.16 q0[7], r0
-; CHECK-NEXT:    add sp, #8
-; CHECK-NEXT:    pop {r7, pc}
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vpt.s16 gt, q1, zr
+; CHECK-NEXT:    vldrht.u16 q0, [r0, q1, uxtw #1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
@@ -636,110 +294,12 @@
 define arm_aapcs_vfpcc <8 x i16> @scaled_v8i16_i16_passthru_icmp1(i16* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: scaled_v8i16_i16_passthru_icmp1:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r7, lr}
-; CHECK-NEXT:    push {r7, lr}
-; CHECK-NEXT:    .pad #8
-; CHECK-NEXT:    sub sp, #8
-; CHECK-NEXT:    vldrh.u16 q0, [r1]
-; CHECK-NEXT:    movs r3, #0
-; CHECK-NEXT:    vcmp.s16 gt, q0, zr
-; CHECK-NEXT:    vmov.u16 r2, q0[4]
-; CHECK-NEXT:    vmrs r1, p0
-; CHECK-NEXT:    vmov.32 q1[0], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[5]
-; CHECK-NEXT:    vmov.u16 r12, q0[7]
-; CHECK-NEXT:    vmov.32 q1[1], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[6]
-; CHECK-NEXT:    vmov.32 q1[2], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[0]
-; CHECK-NEXT:    vmov.32 q2[0], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[1]
-; CHECK-NEXT:    vmov.32 q2[1], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[2]
-; CHECK-NEXT:    vmov.32 q2[2], r2
-; CHECK-NEXT:    vmov.u16 r2, q0[3]
-; CHECK-NEXT:    vmov.32 q2[3], r2
-; CHECK-NEXT:    vmovlb.u16 q0, q2
-; CHECK-NEXT:    vshl.i32 q0, q0, #1
-; CHECK-NEXT:    vadd.i32 q2, q0, r0
-; CHECK-NEXT:    and lr, r1, #1
-; CHECK-NEXT:    rsb.w r2, lr, #0
-; CHECK-NEXT:    bfi r3, r2, #0, #1
-; CHECK-NEXT:    ubfx r2, r1, #2, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #1, #1
-; CHECK-NEXT:    ubfx r2, r1, #4, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #2, #1
-; CHECK-NEXT:    ubfx r2, r1, #6, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #3, #1
-; CHECK-NEXT:    ubfx r2, r1, #8, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #4, #1
-; CHECK-NEXT:    ubfx r2, r1, #10, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #5, #1
-; CHECK-NEXT:    ubfx r2, r1, #12, #1
-; CHECK-NEXT:    ubfx r1, r1, #14, #1
-; CHECK-NEXT:    rsbs r2, r2, #0
-; CHECK-NEXT:    bfi r3, r2, #6, #1
-; CHECK-NEXT:    rsbs r1, r1, #0
-; CHECK-NEXT:    bfi r3, r1, #7, #1
-; CHECK-NEXT:    uxtb r1, r3
-; CHECK-NEXT:    lsls r2, r3, #31
-; CHECK-NEXT:    beq .LBB12_2
-; CHECK-NEXT:  @ %bb.1: @ %cond.load
-; CHECK-NEXT:    vmov r2, s8
-; CHECK-NEXT:    movs r3, #1
-; CHECK-NEXT:    vdup.16 q0, r3
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    b .LBB12_3
-; CHECK-NEXT:  .LBB12_2:
 ; CHECK-NEXT:    vmov.i16 q0, #0x1
-; CHECK-NEXT:  .LBB12_3: @ %else
-; CHECK-NEXT:    vmov.32 q1[3], r12
-; CHECK-NEXT:    lsls r2, r1, #30
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s9
-; CHECK-NEXT:    ldrhmi r2, [r2]
-; CHECK-NEXT:    vmovmi.16 q0[1], r2
-; CHECK-NEXT:    vmovlb.u16 q1, q1
-; CHECK-NEXT:    lsls r2, r1, #29
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s10
-; CHECK-NEXT:    ldrhmi r2, [r2]
-; CHECK-NEXT:    vmovmi.16 q0[2], r2
-; CHECK-NEXT:    vshl.i32 q1, q1, #1
-; CHECK-NEXT:    lsls r2, r1, #28
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r2, s11
-; CHECK-NEXT:    ldrhmi r2, [r2]
-; CHECK-NEXT:    vmovmi.16 q0[3], r2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    lsls r0, r1, #27
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r0, s4
-; CHECK-NEXT:    ldrhmi r0, [r0]
-; CHECK-NEXT:    vmovmi.16 q0[4], r0
-; CHECK-NEXT:    lsls r0, r1, #26
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r0, s5
-; CHECK-NEXT:    ldrhmi r0, [r0]
-; CHECK-NEXT:    vmovmi.16 q0[5], r0
-; CHECK-NEXT:    lsls r0, r1, #25
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r0, s6
-; CHECK-NEXT:    ldrhmi r0, [r0]
-; CHECK-NEXT:    vmovmi.16 q0[6], r0
-; CHECK-NEXT:    lsls r0, r1, #24
-; CHECK-NEXT:    ittt mi
-; CHECK-NEXT:    vmovmi r0, s7
-; CHECK-NEXT:    ldrhmi r0, [r0]
-; CHECK-NEXT:    vmovmi.16 q0[7], r0
-; CHECK-NEXT:    add sp, #8
-; CHECK-NEXT:    pop {r7, pc}
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vpt.s16 gt, q1, zr
+; CHECK-NEXT:    vldrht.u16 q2, [r0, q1, uxtw #1]
+; CHECK-NEXT:    vpsel q0, q2, q0
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
Index: llvm/test/CodeGen/Thumb2/mve-gather-ind16-unscaled.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-gather-ind16-unscaled.ll
+++ llvm/test/CodeGen/Thumb2/mve-gather-ind16-unscaled.ll
@@ -92,37 +92,9 @@
 define arm_aapcs_vfpcc <8 x i16> @unscaled_i16_i16(i8* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: unscaled_i16_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r7, lr}
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r2, s2
-; CHECK-NEXT:    vmov r3, s3
-; CHECK-NEXT:    vmov r5, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    vmov r4, s7
-; CHECK-NEXT:    ldrh.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    ldrh.w lr, [r3]
-; CHECK-NEXT:    vmov r3, s6
-; CHECK-NEXT:    ldrh r5, [r5]
-; CHECK-NEXT:    ldrh r0, [r0]
-; CHECK-NEXT:    ldrh r1, [r1]
-; CHECK-NEXT:    ldrh r4, [r4]
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    ldrh r3, [r3]
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    vmov.16 q0[1], r5
-; CHECK-NEXT:    vmov.16 q0[2], r12
-; CHECK-NEXT:    vmov.16 q0[3], lr
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov.16 q0[5], r1
-; CHECK-NEXT:    vmov.16 q0[6], r3
-; CHECK-NEXT:    vmov.16 q0[7], r4
-; CHECK-NEXT:    pop {r4, r5, r7, pc}
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
   %offs.zext = zext <8 x i16> %offs to <8 x i32>
@@ -135,42 +107,8 @@
 define arm_aapcs_vfpcc <8 x half> @unscaled_f16_i16(i8* %base, <8 x i16>* %offptr) {
 ; CHECK-LABEL: unscaled_f16_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vldr.16 s0, [r2]
-; CHECK-NEXT:    vmov r3, s5
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    vldr.16 s0, [r3]
-; CHECK-NEXT:    vmov r3, s0
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    vmov r2, s6
-; CHECK-NEXT:    vmov.16 q0[1], r3
-; CHECK-NEXT:    vldr.16 s8, [r2]
-; CHECK-NEXT:    vmov r2, s8
-; CHECK-NEXT:    vmov.16 q0[2], r2
-; CHECK-NEXT:    vmov r2, s7
-; CHECK-NEXT:    vldr.16 s4, [r2]
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vldrh.u32 q1, [r1, #8]
-; CHECK-NEXT:    vmov.16 q0[3], r2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vldr.16 s4, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q0[7], r0
+; CHECK-NEXT:    vldrh.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i16>, <8 x i16>* %offptr, align 2
@@ -272,37 +210,9 @@
 define arm_aapcs_vfpcc <8 x i16> @unsigned_unscaled_i16_i8(i8* %base, <8 x i8>* %offptr) {
 ; CHECK-LABEL: unsigned_unscaled_i16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r7, lr}
-; CHECK-NEXT:    push {r4, r5, r7, lr}
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r2, s2
-; CHECK-NEXT:    vmov r3, s3
-; CHECK-NEXT:    vmov r5, s1
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s5
-; CHECK-NEXT:    vmov r4, s7
-; CHECK-NEXT:    ldrh.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    ldrh.w lr, [r3]
-; CHECK-NEXT:    vmov r3, s6
-; CHECK-NEXT:    ldrh r5, [r5]
-; CHECK-NEXT:    ldrh r0, [r0]
-; CHECK-NEXT:    ldrh r1, [r1]
-; CHECK-NEXT:    ldrh r4, [r4]
-; CHECK-NEXT:    ldrh r2, [r2]
-; CHECK-NEXT:    ldrh r3, [r3]
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    vmov.16 q0[1], r5
-; CHECK-NEXT:    vmov.16 q0[2], r12
-; CHECK-NEXT:    vmov.16 q0[3], lr
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov.16 q0[5], r1
-; CHECK-NEXT:    vmov.16 q0[6], r3
-; CHECK-NEXT:    vmov.16 q0[7], r4
-; CHECK-NEXT:    pop {r4, r5, r7, pc}
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
   %offs.zext = zext <8 x i8> %offs to <8 x i32>
@@ -315,42 +225,8 @@
 define arm_aapcs_vfpcc <8 x half> @unsigned_unscaled_f16_i8(i8* %base, <8 x i8>* %offptr) {
 ; CHECK-LABEL: unsigned_unscaled_f16_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vldr.16 s0, [r2]
-; CHECK-NEXT:    vmov r3, s5
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    vldr.16 s0, [r3]
-; CHECK-NEXT:    vmov r3, s0
-; CHECK-NEXT:    vmov.16 q0[0], r2
-; CHECK-NEXT:    vmov r2, s6
-; CHECK-NEXT:    vmov.16 q0[1], r3
-; CHECK-NEXT:    vldr.16 s8, [r2]
-; CHECK-NEXT:    vmov r2, s8
-; CHECK-NEXT:    vmov.16 q0[2], r2
-; CHECK-NEXT:    vmov r2, s7
-; CHECK-NEXT:    vldr.16 s4, [r2]
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vldrb.u32 q1, [r1, #4]
-; CHECK-NEXT:    vmov.16 q0[3], r2
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[4], r0
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[5], r0
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vldr.16 s8, [r0]
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov.16 q0[6], r0
-; CHECK-NEXT:    vmov r0, s7
-; CHECK-NEXT:    vldr.16 s4, [r0]
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov.16 q0[7], r0
+; CHECK-NEXT:    vldrb.u16 q1, [r1]
+; CHECK-NEXT:    vldrh.u16 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <8 x i8>, <8 x i8>* %offptr, align 1
Index: llvm/test/CodeGen/Thumb2/mve-gather-ind32-scaled.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-gather-ind32-scaled.ll
+++ llvm/test/CodeGen/Thumb2/mve-gather-ind32-scaled.ll
@@ -60,10 +60,8 @@
 define arm_aapcs_vfpcc <4 x i32> @scaled_i32_i32(i32* %base, <4 x i32>* %offptr) {
 ; CHECK-LABEL: scaled_i32_i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrw.u32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrw.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
@@ -77,10 +75,8 @@
 define arm_aapcs_vfpcc <4 x float> @scaled_f32_i32(i32* %base, <4 x i32>* %offptr) {
 ; CHECK-LABEL: scaled_f32_i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrw.u32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrw.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
@@ -93,10 +89,8 @@
 define arm_aapcs_vfpcc <4 x i32> @unsigned_scaled_b_i32_i16(i32* %base, <4 x i16>* %offptr) {
 ; CHECK-LABEL: unsigned_scaled_b_i32_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrh.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
@@ -109,10 +103,8 @@
 define arm_aapcs_vfpcc <4 x i32> @signed_scaled_i32_i16(i32* %base, <4 x i16>* %offptr) {
 ; CHECK-LABEL: signed_scaled_i32_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.s32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrh.s32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
@@ -125,10 +117,8 @@
 define arm_aapcs_vfpcc <4 x float> @a_unsigned_scaled_f32_i16(i32* %base, <4 x i16>* %offptr) {
 ; CHECK-LABEL: a_unsigned_scaled_f32_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrh.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
@@ -142,10 +132,8 @@
 define arm_aapcs_vfpcc <4 x float> @b_signed_scaled_f32_i16(i32* %base, <4 x i16>* %offptr) {
 ; CHECK-LABEL: b_signed_scaled_f32_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.s32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrh.s32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
@@ -275,10 +263,8 @@
 define arm_aapcs_vfpcc <4 x i32> @unsigned_scaled_b_i32_i8(i32* %base, <4 x i8>* %offptr) {
 ; CHECK-LABEL: unsigned_scaled_b_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrb.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
@@ -291,10 +277,8 @@
 define arm_aapcs_vfpcc <4 x i32> @signed_scaled_i32_i8(i32* %base, <4 x i8>* %offptr) {
 ; CHECK-LABEL: signed_scaled_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.s32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrb.s32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
@@ -307,10 +291,8 @@
 define arm_aapcs_vfpcc <4 x float> @a_unsigned_scaled_f32_i8(i32* %base, <4 x i8>* %offptr) {
 ; CHECK-LABEL: a_unsigned_scaled_f32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrb.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
@@ -324,10 +306,8 @@
 define arm_aapcs_vfpcc <4 x float> @b_signed_scaled_f32_i8(i32* %base, <4 x i8>* %offptr) {
 ; CHECK-LABEL: b_signed_scaled_f32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.s32 q0, [r1]
-; CHECK-NEXT:    vshl.i32 q0, q0, #2
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrb.s32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1, uxtw #2]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
Index: llvm/test/CodeGen/Thumb2/mve-gather-ind32-unscaled.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-gather-ind32-unscaled.ll
+++ llvm/test/CodeGen/Thumb2/mve-gather-ind32-unscaled.ll
@@ -116,9 +116,8 @@
 define arm_aapcs_vfpcc <4 x i32> @unscaled_i32_i32(i8* %base, <4 x i32>* %offptr) {
 ; CHECK-LABEL: unscaled_i32_i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrw.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrw.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
@@ -131,9 +130,8 @@
 define arm_aapcs_vfpcc <4 x float> @unscaled_f32_i32(i8* %base, <4 x i32>* %offptr) {
 ; CHECK-LABEL: unscaled_f32_i32:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrw.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrw.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i32>, <4 x i32>* %offptr, align 4
@@ -146,9 +144,8 @@
 define arm_aapcs_vfpcc <4 x i32> @unsigned_unscaled_b_i32_i16(i8* %base, <4 x i16>* %offptr) {
 ; CHECK-LABEL: unsigned_unscaled_b_i32_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrh.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
@@ -162,9 +159,8 @@
 define arm_aapcs_vfpcc <4 x i32> @signed_unscaled_i32_i16(i8* %base, <4 x i16>* %offptr) {
 ; CHECK-LABEL: signed_unscaled_i32_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.s32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrh.s32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
@@ -178,9 +174,8 @@
 define arm_aapcs_vfpcc <4 x float> @a_unsigned_unscaled_f32_i16(i8* %base, <4 x i16>* %offptr) {
 ; CHECK-LABEL: a_unsigned_unscaled_f32_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrh.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
@@ -194,9 +189,8 @@
 define arm_aapcs_vfpcc <4 x float> @b_signed_unscaled_f32_i16(i8* %base, <4 x i16>* %offptr) {
 ; CHECK-LABEL: b_signed_unscaled_f32_i16:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrh.s32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrh.s32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i16>, <4 x i16>* %offptr, align 2
@@ -442,9 +436,8 @@
 define arm_aapcs_vfpcc <4 x i32> @unsigned_unscaled_b_i32_i8(i8* %base, <4 x i8>* %offptr) {
 ; CHECK-LABEL: unsigned_unscaled_b_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrb.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
@@ -458,9 +451,8 @@
 define arm_aapcs_vfpcc <4 x i32> @signed_unscaled_i32_i8(i8* %base, <4 x i8>* %offptr) {
 ; CHECK-LABEL: signed_unscaled_i32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.s32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrb.s32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
@@ -474,9 +466,8 @@
 define arm_aapcs_vfpcc <4 x float> @a_unsigned_unscaled_f32_i8(i8* %base, <4 x i8>* %offptr) {
 ; CHECK-LABEL: a_unsigned_unscaled_f32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrb.u32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
@@ -490,9 +481,8 @@
 define arm_aapcs_vfpcc <4 x float> @b_signed_unscaled_f32_i8(i8* %base, <4 x i8>* %offptr) {
 ; CHECK-LABEL: b_signed_unscaled_f32_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vldrb.s32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrw.u32 q0, [q1]
+; CHECK-NEXT:    vldrb.s32 q1, [r1]
+; CHECK-NEXT:    vldrw.u32 q0, [r0, q1]
 ; CHECK-NEXT:    bx lr
 entry:
   %offs = load <4 x i8>, <4 x i8>* %offptr, align 1
Index: llvm/test/CodeGen/Thumb2/mve-gather-ind8-unscaled.ll
===================================================================
--- llvm/test/CodeGen/Thumb2/mve-gather-ind8-unscaled.ll
+++ llvm/test/CodeGen/Thumb2/mve-gather-ind8-unscaled.ll
@@ -5,65 +5,9 @@
 define arm_aapcs_vfpcc <16 x i8> @unscaled_v16i8_i8(i8* %base, <16 x i8>* %offptr) {
 ; CHECK-LABEL: unscaled_v16i8_i8:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r5, r6, lr}
-; CHECK-NEXT:    push {r4, r5, r6, lr}
-; CHECK-NEXT:    vldrb.u32 q0, [r1, #8]
-; CHECK-NEXT:    vldrb.u32 q2, [r1, #4]
-; CHECK-NEXT:    vadd.i32 q1, q0, r0
-; CHECK-NEXT:    vldrb.u32 q0, [r1, #12]
-; CHECK-NEXT:    vmov r2, s6
-; CHECK-NEXT:    vadd.i32 q2, q2, r0
-; CHECK-NEXT:    vadd.i32 q0, q0, r0
-; CHECK-NEXT:    vmov r6, s4
-; CHECK-NEXT:    vmov r3, s2
-; CHECK-NEXT:    vmov r4, s3
-; CHECK-NEXT:    vmov r5, s7
-; CHECK-NEXT:    ldrb.w r12, [r2]
-; CHECK-NEXT:    vmov r2, s0
-; CHECK-NEXT:    ldrb r6, [r6]
-; CHECK-NEXT:    ldrb r3, [r3]
-; CHECK-NEXT:    ldrb r4, [r4]
-; CHECK-NEXT:    ldrb r5, [r5]
-; CHECK-NEXT:    ldrb.w lr, [r2]
-; CHECK-NEXT:    vmov r2, s1
-; CHECK-NEXT:    vldrb.u32 q0, [r1]
-; CHECK-NEXT:    vadd.i32 q3, q0, r0
-; CHECK-NEXT:    vmov r0, s12
-; CHECK-NEXT:    ldrb r2, [r2]
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[0], r0
-; CHECK-NEXT:    vmov r0, s13
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[1], r0
-; CHECK-NEXT:    vmov r0, s14
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[2], r0
-; CHECK-NEXT:    vmov r0, s15
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[3], r0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[4], r0
-; CHECK-NEXT:    vmov r0, s9
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[5], r0
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[6], r0
-; CHECK-NEXT:    vmov r0, s11
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[7], r0
-; CHECK-NEXT:    vmov r0, s5
-; CHECK-NEXT:    vmov.8 q0[8], r6
-; CHECK-NEXT:    ldrb r0, [r0]
-; CHECK-NEXT:    vmov.8 q0[9], r0
-; CHECK-NEXT:    vmov.8 q0[10], r12
-; CHECK-NEXT:    vmov.8 q0[11], r5
-; CHECK-NEXT:    vmov.8 q0[12], lr
-; CHECK-NEXT:    vmov.8 q0[13], r2
-; CHECK-NEXT:    vmov.8 q0[14], r3
-; CHECK-NEXT:    vmov.8 q0[15], r4
-; CHECK-NEXT:    pop {r4, r5, r6, pc}
+; CHECK-NEXT:    vldrb.u8 q1, [r1]
+; CHECK-NEXT:    vldrb.u8 q0, [r0, q1]
+; CHECK-NEXT:    bx lr
 entry:
   %offs = load <16 x i8>, <16 x i8>* %offptr, align 1
   %offs.zext = zext <16 x i8> %offs to <16 x i32>