[AArch64][GlobalISel] Add G_VECREDUCE fewerElements support for full scalarization.

For some reductions like G_VECREDUCE_OR on AArch64, we need to scalarize
completely if the source is <= 64b. This change adds support for that in
the legalizer. If the source has a pow-2 num elements, then we can do
a tree reduction using the scalar operation in the individual elements.
Otherwise, we just create a sequential chain of operations.

For AArch64, we only need to scalarize if the input is <64b. If it's great than
64b then we can first do a fewElements step to 64b, taking advantage of vector
instructions until we reach the point of scalarization.

I also had to relax the verifier checks for reductions because the intrinsics
support <1 x EltTy> types, which we lower to scalars for GlobalISel.

Differential Revision: https://reviews.llvm.org/D108276

llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h

@@ -403,6 +403,7 @@ public:
   LegalizeResult lowerAbsToAddXor(MachineInstr &MI);
   LegalizeResult lowerAbsToMaxNeg(MachineInstr &MI);
   LegalizeResult lowerIsNaN(MachineInstr &MI);
+  LegalizeResult lowerVectorReduction(MachineInstr &MI);
 };
 
 /// Helper function that creates a libcall to the given \p Name using the given

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

@@ -3489,6 +3489,8 @@ LegalizerHelper::lower(MachineInstr &MI, unsigned TypeIdx, LLT LowerHintTy) {
     return lowerRotate(MI);
   case G_ISNAN:
     return lowerIsNaN(MI);
+  GISEL_VECREDUCE_CASES_NONSEQ
+    return lowerVectorReduction(MI);
   }
 }
 
@@ -4637,35 +4639,7 @@ LegalizerHelper::LegalizeResult LegalizerHelper::fewerElementsVectorShuffle(
   return Legalized;
 }
 
-LegalizerHelper::LegalizeResult LegalizerHelper::fewerElementsVectorReductions(
+static unsigned getScalarOpcForReduction(unsigned Opc) {
-    MachineInstr &MI, unsigned int TypeIdx, LLT NarrowTy) {
-  unsigned Opc = MI.getOpcode();
-  assert(Opc != TargetOpcode::G_VECREDUCE_SEQ_FADD &&
-         Opc != TargetOpcode::G_VECREDUCE_SEQ_FMUL &&
-         "Sequential reductions not expected");
-
-  if (TypeIdx != 1)
-    return UnableToLegalize;
-
-  // The semantics of the normal non-sequential reductions allow us to freely
-  // re-associate the operation.
-  Register SrcReg = MI.getOperand(1).getReg();
-  LLT SrcTy = MRI.getType(SrcReg);
-  Register DstReg = MI.getOperand(0).getReg();
-  LLT DstTy = MRI.getType(DstReg);
-
-  if (SrcTy.getNumElements() % NarrowTy.getNumElements() != 0)
-    return UnableToLegalize;
-
-  SmallVector<Register> SplitSrcs;
-  const unsigned NumParts = SrcTy.getNumElements() / NarrowTy.getNumElements();
-  extractParts(SrcReg, NarrowTy, NumParts, SplitSrcs);
-  SmallVector<Register> PartialReductions;
-  for (unsigned Part = 0; Part < NumParts; ++Part) {
-    PartialReductions.push_back(
-        MIRBuilder.buildInstr(Opc, {DstTy}, {SplitSrcs[Part]}).getReg(0));
-  }
-
   unsigned ScalarOpc;
   switch (Opc) {
   case TargetOpcode::G_VECREDUCE_FADD:
@@ -4708,9 +4682,80 @@ LegalizerHelper::LegalizeResult LegalizerHelper::fewerElementsVectorReductions(
     ScalarOpc = TargetOpcode::G_UMIN;
     break;
   default:
-    LLVM_DEBUG(dbgs() << "Can't legalize: unknown reduction kind.\n");
+    llvm_unreachable("Unhandled reduction");
-    return UnableToLegalize;
   }
+  return ScalarOpc;
+}
+
+LegalizerHelper::LegalizeResult LegalizerHelper::fewerElementsVectorReductions(
+    MachineInstr &MI, unsigned int TypeIdx, LLT NarrowTy) {
+  unsigned Opc = MI.getOpcode();
+  assert(Opc != TargetOpcode::G_VECREDUCE_SEQ_FADD &&
+         Opc != TargetOpcode::G_VECREDUCE_SEQ_FMUL &&
+         "Sequential reductions not expected");
+
+  if (TypeIdx != 1)
+    return UnableToLegalize;
+
+  // The semantics of the normal non-sequential reductions allow us to freely
+  // re-associate the operation.
+  Register SrcReg = MI.getOperand(1).getReg();
+  LLT SrcTy = MRI.getType(SrcReg);
+  Register DstReg = MI.getOperand(0).getReg();
+  LLT DstTy = MRI.getType(DstReg);
+
+  if (NarrowTy.isVector() &&
+      (SrcTy.getNumElements() % NarrowTy.getNumElements() != 0))
+    return UnableToLegalize;
+
+  unsigned ScalarOpc = getScalarOpcForReduction(Opc);
+  SmallVector<Register> SplitSrcs;
+  // If NarrowTy is a scalar then we're being asked to scalarize.
+  const unsigned NumParts =
+      NarrowTy.isVector() ? SrcTy.getNumElements() / NarrowTy.getNumElements()
+                          : SrcTy.getNumElements();
+
+  extractParts(SrcReg, NarrowTy, NumParts, SplitSrcs);
+  if (NarrowTy.isScalar()) {
+    if (DstTy != NarrowTy)
+      return UnableToLegalize; // FIXME: handle implicit extensions.
+
+    if (isPowerOf2_32(NumParts)) {
+      // Generate a tree of scalar operations to reduce the critical path.
+      SmallVector<Register> PartialResults;
+      unsigned NumPartsLeft = NumParts;
+      while (NumPartsLeft > 1) {
+        for (unsigned Idx = 0; Idx < NumPartsLeft - 1; Idx += 2) {
+          PartialResults.emplace_back(
+              MIRBuilder
+                  .buildInstr(ScalarOpc, {NarrowTy},
+                              {SplitSrcs[Idx], SplitSrcs[Idx + 1]})
+                  .getReg(0));
+        }
+        SplitSrcs = PartialResults;
+        PartialResults.clear();
+        NumPartsLeft = SplitSrcs.size();
+      }
+      assert(SplitSrcs.size() == 1);
+      MIRBuilder.buildCopy(DstReg, SplitSrcs[0]);
+      MI.eraseFromParent();
+      return Legalized;
+    }
+    // If we can't generate a tree, then just do sequential operations.
+    Register Acc = SplitSrcs[0];
+    for (unsigned Idx = 1; Idx < NumParts; ++Idx)
+      Acc = MIRBuilder.buildInstr(ScalarOpc, {NarrowTy}, {Acc, SplitSrcs[Idx]})
+                .getReg(0);
+    MIRBuilder.buildCopy(DstReg, Acc);
+    MI.eraseFromParent();
+    return Legalized;
+  }
+  SmallVector<Register> PartialReductions;
+  for (unsigned Part = 0; Part < NumParts; ++Part) {
+    PartialReductions.push_back(
+        MIRBuilder.buildInstr(Opc, {DstTy}, {SplitSrcs[Part]}).getReg(0));
+  }
+
 
   // If the types involved are powers of 2, we can generate intermediate vector
   // ops, before generating a final reduction operation.
@@ -7389,3 +7434,22 @@ LegalizerHelper::LegalizeResult LegalizerHelper::lowerIsNaN(MachineInstr &MI) {
   MI.eraseFromParent();
   return Legalized;
 }
+
+LegalizerHelper::LegalizeResult
+LegalizerHelper::lowerVectorReduction(MachineInstr &MI) {
+  Register SrcReg = MI.getOperand(1).getReg();
+  LLT SrcTy = MRI.getType(SrcReg);
+  LLT DstTy = MRI.getType(SrcReg);
+
+  // The source could be a scalar if the IR type was <1 x sN>.
+  if (SrcTy.isScalar()) {
+    if (DstTy.getSizeInBits() > SrcTy.getSizeInBits())
+      return UnableToLegalize; // FIXME: handle extension.
+    // This can be just a plain copy.
+    Observer.changingInstr(MI);
+    MI.setDesc(MIRBuilder.getTII().get(TargetOpcode::COPY));
+    Observer.changedInstr(MI);
+    return Legalized;
+  }
+  return UnableToLegalize;;
+}

llvm/lib/CodeGen/MachineVerifier.cpp

@@ -1589,11 +1589,8 @@ void MachineVerifier::verifyPreISelGenericInstruction(const MachineInstr *MI) {
   case TargetOpcode::G_VECREDUCE_UMAX:
   case TargetOpcode::G_VECREDUCE_UMIN: {
     LLT DstTy = MRI->getType(MI->getOperand(0).getReg());
-    LLT SrcTy = MRI->getType(MI->getOperand(1).getReg());
     if (!DstTy.isScalar())
       report("Vector reduction requires a scalar destination type", MI);
-    if (!SrcTy.isVector())
-      report("Vector reduction requires vector source=", MI);
     break;
   }
 

llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp

@@ -691,6 +691,27 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
       .clampMaxNumElements(1, s32, 4)
       .lower();
 
+  getActionDefinitionsBuilder(G_VECREDUCE_OR)
+      // Try to break down into smaller vectors as long as they're at least 64
+      // bits. This lets us use vector operations for some parts of the
+      // reduction.
+      .fewerElementsIf(
+          [=](const LegalityQuery &Q) {
+            LLT SrcTy = Q.Types[1];
+            if (SrcTy.isScalar())
+              return false;
+            if (!isPowerOf2_32(SrcTy.getNumElements()))
+              return false;
+            // We can usually perform 64b vector operations.
+            return SrcTy.getSizeInBits() > 64;
+          },
+          [=](const LegalityQuery &Q) {
+            LLT SrcTy = Q.Types[1];
+            return std::make_pair(1, SrcTy.divide(2));
+          })
+      .scalarize(1)
+      .lower();
+
   getActionDefinitionsBuilder({G_UADDSAT, G_USUBSAT})
       .lowerIf([=](const LegalityQuery &Q) { return Q.Types[0].isScalar(); });
 

llvm/test/CodeGen/AArch64/GlobalISel/legalize-reduce-or.mir

@@ -0,0 +1,701 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -O0 -march=aarch64 -run-pass=legalizer -global-isel-abort=1 -verify-machineinstrs %s -o - | FileCheck %s
+
+---
+name:            test_redor_v1i1
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$w0' }
+body:             |
+  bb.1:
+    liveins: $w0
+
+    ; CHECK-LABEL: name: test_redor_v1i1
+    ; CHECK: liveins: $w0
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C]]
+    ; CHECK: $w0 = COPY [[AND]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(s32) = COPY $w0
+    %0:_(s1) = G_TRUNC %1(s32)
+    %2:_(s1) = G_VECREDUCE_OR %0(s1)
+    %4:_(s32) = G_ZEXT %2(s1)
+    $w0 = COPY %4(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v2i1
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$d0' }
+body:             |
+  bb.1:
+    liveins: $d0
+
+    ; CHECK-LABEL: name: test_redor_v2i1
+    ; CHECK: liveins: $d0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $d0
+    ; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; CHECK: $w0 = COPY [[AND]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<2 x s32>) = COPY $d0
+    %0:_(<2 x s1>) = G_TRUNC %1(<2 x s32>)
+    %2:_(s1) = G_VECREDUCE_OR %0(<2 x s1>)
+    %4:_(s32) = G_ZEXT %2(s1)
+    $w0 = COPY %4(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v4i1
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$d0' }
+body:             |
+  bb.1:
+    liveins: $d0
+
+    ; CHECK-LABEL: name: test_redor_v4i1
+    ; CHECK: liveins: $d0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $d0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; CHECK: $w0 = COPY [[AND]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<4 x s16>) = COPY $d0
+    %0:_(<4 x s1>) = G_TRUNC %1(<4 x s16>)
+    %2:_(s1) = G_VECREDUCE_OR %0(<4 x s1>)
+    %4:_(s32) = G_ZEXT %2(s1)
+    $w0 = COPY %4(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v8i1
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$d0' }
+body:             |
+  bb.1:
+    liveins: $d0
+
+    ; CHECK-LABEL: name: test_redor_v8i1
+    ; CHECK: liveins: $d0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<8 x s8>) = COPY $d0
+    ; CHECK: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8), [[UV2:%[0-9]+]]:_(s8), [[UV3:%[0-9]+]]:_(s8), [[UV4:%[0-9]+]]:_(s8), [[UV5:%[0-9]+]]:_(s8), [[UV6:%[0-9]+]]:_(s8), [[UV7:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[COPY]](<8 x s8>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s8)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s8)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s8)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s8)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s8)
+    ; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s8)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ANYEXT4]], [[ANYEXT5]]
+    ; CHECK: [[ANYEXT6:%[0-9]+]]:_(s32) = G_ANYEXT [[UV6]](s8)
+    ; CHECK: [[ANYEXT7:%[0-9]+]]:_(s32) = G_ANYEXT [[UV7]](s8)
+    ; CHECK: [[OR3:%[0-9]+]]:_(s32) = G_OR [[ANYEXT6]], [[ANYEXT7]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: [[OR4:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[OR3]](s32)
+    ; CHECK: [[OR5:%[0-9]+]]:_(s32) = G_OR [[COPY3]], [[COPY4]]
+    ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[OR4]](s32)
+    ; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[OR5]](s32)
+    ; CHECK: [[OR6:%[0-9]+]]:_(s32) = G_OR [[COPY5]], [[COPY6]]
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[OR6]](s32)
+    ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C]]
+    ; CHECK: $w0 = COPY [[AND]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<8 x s8>) = COPY $d0
+    %0:_(<8 x s1>) = G_TRUNC %1(<8 x s8>)
+    %2:_(s1) = G_VECREDUCE_OR %0(<8 x s1>)
+    %4:_(s32) = G_ZEXT %2(s1)
+    $w0 = COPY %4(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v16i1
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+frameInfo:
+  maxAlignment:    1
+machineFunctionInfo: {}
+body:             |
+  bb.1:
+    liveins: $q0
+
+    ; CHECK-LABEL: name: test_redor_v16i1
+    ; CHECK: liveins: $q0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<16 x s8>) = COPY $q0
+    ; CHECK: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8), [[UV2:%[0-9]+]]:_(s8), [[UV3:%[0-9]+]]:_(s8), [[UV4:%[0-9]+]]:_(s8), [[UV5:%[0-9]+]]:_(s8), [[UV6:%[0-9]+]]:_(s8), [[UV7:%[0-9]+]]:_(s8), [[UV8:%[0-9]+]]:_(s8), [[UV9:%[0-9]+]]:_(s8), [[UV10:%[0-9]+]]:_(s8), [[UV11:%[0-9]+]]:_(s8), [[UV12:%[0-9]+]]:_(s8), [[UV13:%[0-9]+]]:_(s8), [[UV14:%[0-9]+]]:_(s8), [[UV15:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[COPY]](<16 x s8>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s8)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s8)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s8)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s8)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s8)
+    ; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s8)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ANYEXT4]], [[ANYEXT5]]
+    ; CHECK: [[ANYEXT6:%[0-9]+]]:_(s32) = G_ANYEXT [[UV6]](s8)
+    ; CHECK: [[ANYEXT7:%[0-9]+]]:_(s32) = G_ANYEXT [[UV7]](s8)
+    ; CHECK: [[OR3:%[0-9]+]]:_(s32) = G_OR [[ANYEXT6]], [[ANYEXT7]]
+    ; CHECK: [[ANYEXT8:%[0-9]+]]:_(s32) = G_ANYEXT [[UV8]](s8)
+    ; CHECK: [[ANYEXT9:%[0-9]+]]:_(s32) = G_ANYEXT [[UV9]](s8)
+    ; CHECK: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ANYEXT8]], [[ANYEXT9]]
+    ; CHECK: [[ANYEXT10:%[0-9]+]]:_(s32) = G_ANYEXT [[UV10]](s8)
+    ; CHECK: [[ANYEXT11:%[0-9]+]]:_(s32) = G_ANYEXT [[UV11]](s8)
+    ; CHECK: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ANYEXT10]], [[ANYEXT11]]
+    ; CHECK: [[ANYEXT12:%[0-9]+]]:_(s32) = G_ANYEXT [[UV12]](s8)
+    ; CHECK: [[ANYEXT13:%[0-9]+]]:_(s32) = G_ANYEXT [[UV13]](s8)
+    ; CHECK: [[OR6:%[0-9]+]]:_(s32) = G_OR [[ANYEXT12]], [[ANYEXT13]]
+    ; CHECK: [[ANYEXT14:%[0-9]+]]:_(s32) = G_ANYEXT [[UV14]](s8)
+    ; CHECK: [[ANYEXT15:%[0-9]+]]:_(s32) = G_ANYEXT [[UV15]](s8)
+    ; CHECK: [[OR7:%[0-9]+]]:_(s32) = G_OR [[ANYEXT14]], [[ANYEXT15]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: [[OR8:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[OR3]](s32)
+    ; CHECK: [[OR9:%[0-9]+]]:_(s32) = G_OR [[COPY3]], [[COPY4]]
+    ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[OR4]](s32)
+    ; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[OR5]](s32)
+    ; CHECK: [[OR10:%[0-9]+]]:_(s32) = G_OR [[COPY5]], [[COPY6]]
+    ; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[OR6]](s32)
+    ; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[OR7]](s32)
+    ; CHECK: [[OR11:%[0-9]+]]:_(s32) = G_OR [[COPY7]], [[COPY8]]
+    ; CHECK: [[COPY9:%[0-9]+]]:_(s32) = COPY [[OR8]](s32)
+    ; CHECK: [[COPY10:%[0-9]+]]:_(s32) = COPY [[OR9]](s32)
+    ; CHECK: [[OR12:%[0-9]+]]:_(s32) = G_OR [[COPY9]], [[COPY10]]
+    ; CHECK: [[COPY11:%[0-9]+]]:_(s32) = COPY [[OR10]](s32)
+    ; CHECK: [[COPY12:%[0-9]+]]:_(s32) = COPY [[OR11]](s32)
+    ; CHECK: [[OR13:%[0-9]+]]:_(s32) = G_OR [[COPY11]], [[COPY12]]
+    ; CHECK: [[COPY13:%[0-9]+]]:_(s32) = COPY [[OR12]](s32)
+    ; CHECK: [[COPY14:%[0-9]+]]:_(s32) = COPY [[OR13]](s32)
+    ; CHECK: [[OR14:%[0-9]+]]:_(s32) = G_OR [[COPY13]], [[COPY14]]
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+    ; CHECK: [[COPY15:%[0-9]+]]:_(s32) = COPY [[OR14]](s32)
+    ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY15]], [[C]]
+    ; CHECK: $w0 = COPY [[AND]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<16 x s8>) = COPY $q0
+    %0:_(<16 x s1>) = G_TRUNC %1(<16 x s8>)
+    %2:_(s1) = G_VECREDUCE_OR %0(<16 x s1>)
+    %4:_(s32) = G_ZEXT %2(s1)
+    $w0 = COPY %4(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v1i8
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$d0' }
+body:             |
+  bb.1:
+    liveins: $d0
+
+    ; CHECK-LABEL: name: test_redor_v1i8
+    ; CHECK: liveins: $d0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<8 x s8>) = COPY $d0
+    ; CHECK: [[BITCAST:%[0-9]+]]:_(s64) = G_BITCAST [[COPY]](<8 x s8>)
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[BITCAST]](s64)
+    ; CHECK: $w0 = COPY [[TRUNC]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<8 x s8>) = COPY $d0
+    %11:_(s64) = G_BITCAST %1(<8 x s8>)
+    %0:_(s8) = G_TRUNC %11(s64)
+    %9:_(s8) = G_VECREDUCE_OR %0(s8)
+    %10:_(s32) = G_ANYEXT %9(s8)
+    $w0 = COPY %10(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v3i8
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$w0' }
+  - { reg: '$w1' }
+  - { reg: '$w2' }
+body:             |
+  bb.1:
+    liveins: $w0, $w1, $w2
+
+    ; CHECK-LABEL: name: test_redor_v3i8
+    ; CHECK: liveins: $w0, $w1, $w2
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $w1
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY $w2
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[COPY3]], [[COPY4]]
+    ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[COPY2]](s32)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[COPY5]], [[COPY6]]
+    ; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: $w0 = COPY [[COPY7]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(s32) = COPY $w0
+    %2:_(s32) = COPY $w1
+    %3:_(s32) = COPY $w2
+    %4:_(<3 x s32>) = G_BUILD_VECTOR %1(s32), %2(s32), %3(s32)
+    %0:_(<3 x s8>) = G_TRUNC %4(<3 x s32>)
+    %5:_(s8) = G_VECREDUCE_OR %0(<3 x s8>)
+    %6:_(s32) = G_ANYEXT %5(s8)
+    $w0 = COPY %6(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v4i8
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$d0' }
+body:             |
+  bb.1:
+    liveins: $d0
+
+    ; CHECK-LABEL: name: test_redor_v4i8
+    ; CHECK: liveins: $d0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $d0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: $w0 = COPY [[COPY3]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<4 x s16>) = COPY $d0
+    %0:_(<4 x s8>) = G_TRUNC %1(<4 x s16>)
+    %2:_(s8) = G_VECREDUCE_OR %0(<4 x s8>)
+    %3:_(s32) = G_ANYEXT %2(s8)
+    $w0 = COPY %3(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v8i8
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$d0' }
+body:             |
+  bb.1:
+    liveins: $d0
+
+    ; CHECK-LABEL: name: test_redor_v8i8
+    ; CHECK: liveins: $d0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<8 x s8>) = COPY $d0
+    ; CHECK: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8), [[UV2:%[0-9]+]]:_(s8), [[UV3:%[0-9]+]]:_(s8), [[UV4:%[0-9]+]]:_(s8), [[UV5:%[0-9]+]]:_(s8), [[UV6:%[0-9]+]]:_(s8), [[UV7:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[COPY]](<8 x s8>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s8)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s8)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s8)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s8)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s8)
+    ; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s8)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ANYEXT4]], [[ANYEXT5]]
+    ; CHECK: [[ANYEXT6:%[0-9]+]]:_(s32) = G_ANYEXT [[UV6]](s8)
+    ; CHECK: [[ANYEXT7:%[0-9]+]]:_(s32) = G_ANYEXT [[UV7]](s8)
+    ; CHECK: [[OR3:%[0-9]+]]:_(s32) = G_OR [[ANYEXT6]], [[ANYEXT7]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: [[OR4:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[OR3]](s32)
+    ; CHECK: [[OR5:%[0-9]+]]:_(s32) = G_OR [[COPY3]], [[COPY4]]
+    ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[OR4]](s32)
+    ; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[OR5]](s32)
+    ; CHECK: [[OR6:%[0-9]+]]:_(s32) = G_OR [[COPY5]], [[COPY6]]
+    ; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[OR6]](s32)
+    ; CHECK: $w0 = COPY [[COPY7]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:_(<8 x s8>) = COPY $d0
+    %1:_(s8) = G_VECREDUCE_OR %0(<8 x s8>)
+    %2:_(s32) = G_ANYEXT %1(s8)
+    $w0 = COPY %2(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v16i8
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+body:             |
+  bb.1:
+    liveins: $q0
+
+    ; CHECK-LABEL: name: test_redor_v16i8
+    ; CHECK: liveins: $q0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<16 x s8>) = COPY $q0
+    ; CHECK: [[UV:%[0-9]+]]:_(<8 x s8>), [[UV1:%[0-9]+]]:_(<8 x s8>) = G_UNMERGE_VALUES [[COPY]](<16 x s8>)
+    ; CHECK: [[OR:%[0-9]+]]:_(<8 x s8>) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[UV2:%[0-9]+]]:_(s8), [[UV3:%[0-9]+]]:_(s8), [[UV4:%[0-9]+]]:_(s8), [[UV5:%[0-9]+]]:_(s8), [[UV6:%[0-9]+]]:_(s8), [[UV7:%[0-9]+]]:_(s8), [[UV8:%[0-9]+]]:_(s8), [[UV9:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[OR]](<8 x s8>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s8)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s8)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s8)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s8)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV6]](s8)
+    ; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV7]](s8)
+    ; CHECK: [[OR3:%[0-9]+]]:_(s32) = G_OR [[ANYEXT4]], [[ANYEXT5]]
+    ; CHECK: [[ANYEXT6:%[0-9]+]]:_(s32) = G_ANYEXT [[UV8]](s8)
+    ; CHECK: [[ANYEXT7:%[0-9]+]]:_(s32) = G_ANYEXT [[UV9]](s8)
+    ; CHECK: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ANYEXT6]], [[ANYEXT7]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: [[OR5:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR3]](s32)
+    ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[OR4]](s32)
+    ; CHECK: [[OR6:%[0-9]+]]:_(s32) = G_OR [[COPY3]], [[COPY4]]
+    ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[OR5]](s32)
+    ; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[OR6]](s32)
+    ; CHECK: [[OR7:%[0-9]+]]:_(s32) = G_OR [[COPY5]], [[COPY6]]
+    ; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[OR7]](s32)
+    ; CHECK: $w0 = COPY [[COPY7]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:_(<16 x s8>) = COPY $q0
+    %1:_(s8) = G_VECREDUCE_OR %0(<16 x s8>)
+    %2:_(s32) = G_ANYEXT %1(s8)
+    $w0 = COPY %2(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v32i8
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+  - { reg: '$q1' }
+body:             |
+  bb.1:
+    liveins: $q0, $q1
+
+    ; CHECK-LABEL: name: test_redor_v32i8
+    ; CHECK: liveins: $q0, $q1
+    ; CHECK: [[COPY:%[0-9]+]]:_(<16 x s8>) = COPY $q0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<16 x s8>) = COPY $q1
+    ; CHECK: [[OR:%[0-9]+]]:_(<16 x s8>) = G_OR [[COPY]], [[COPY1]]
+    ; CHECK: [[UV:%[0-9]+]]:_(<8 x s8>), [[UV1:%[0-9]+]]:_(<8 x s8>) = G_UNMERGE_VALUES [[OR]](<16 x s8>)
+    ; CHECK: [[OR1:%[0-9]+]]:_(<8 x s8>) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[UV2:%[0-9]+]]:_(s8), [[UV3:%[0-9]+]]:_(s8), [[UV4:%[0-9]+]]:_(s8), [[UV5:%[0-9]+]]:_(s8), [[UV6:%[0-9]+]]:_(s8), [[UV7:%[0-9]+]]:_(s8), [[UV8:%[0-9]+]]:_(s8), [[UV9:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[OR1]](<8 x s8>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s8)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s8)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s8)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s8)
+    ; CHECK: [[OR3:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[ANYEXT4:%[0-9]+]]:_(s32) = G_ANYEXT [[UV6]](s8)
+    ; CHECK: [[ANYEXT5:%[0-9]+]]:_(s32) = G_ANYEXT [[UV7]](s8)
+    ; CHECK: [[OR4:%[0-9]+]]:_(s32) = G_OR [[ANYEXT4]], [[ANYEXT5]]
+    ; CHECK: [[ANYEXT6:%[0-9]+]]:_(s32) = G_ANYEXT [[UV8]](s8)
+    ; CHECK: [[ANYEXT7:%[0-9]+]]:_(s32) = G_ANYEXT [[UV9]](s8)
+    ; CHECK: [[OR5:%[0-9]+]]:_(s32) = G_OR [[ANYEXT6]], [[ANYEXT7]]
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR3]](s32)
+    ; CHECK: [[OR6:%[0-9]+]]:_(s32) = G_OR [[COPY2]], [[COPY3]]
+    ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[OR4]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:_(s32) = COPY [[OR5]](s32)
+    ; CHECK: [[OR7:%[0-9]+]]:_(s32) = G_OR [[COPY4]], [[COPY5]]
+    ; CHECK: [[COPY6:%[0-9]+]]:_(s32) = COPY [[OR6]](s32)
+    ; CHECK: [[COPY7:%[0-9]+]]:_(s32) = COPY [[OR7]](s32)
+    ; CHECK: [[OR8:%[0-9]+]]:_(s32) = G_OR [[COPY6]], [[COPY7]]
+    ; CHECK: [[COPY8:%[0-9]+]]:_(s32) = COPY [[OR8]](s32)
+    ; CHECK: $w0 = COPY [[COPY8]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<16 x s8>) = COPY $q0
+    %2:_(<16 x s8>) = COPY $q1
+    %0:_(<32 x s8>) = G_CONCAT_VECTORS %1(<16 x s8>), %2(<16 x s8>)
+    %3:_(s8) = G_VECREDUCE_OR %0(<32 x s8>)
+    %4:_(s32) = G_ANYEXT %3(s8)
+    $w0 = COPY %4(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v4i16
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$d0' }
+body:             |
+  bb.1:
+    liveins: $d0
+
+    ; CHECK-LABEL: name: test_redor_v4i16
+    ; CHECK: liveins: $d0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $d0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: $w0 = COPY [[COPY3]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:_(<4 x s16>) = COPY $d0
+    %1:_(s16) = G_VECREDUCE_OR %0(<4 x s16>)
+    %2:_(s32) = G_ANYEXT %1(s16)
+    $w0 = COPY %2(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v8i16
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+body:             |
+  bb.1:
+    liveins: $q0
+
+    ; CHECK-LABEL: name: test_redor_v8i16
+    ; CHECK: liveins: $q0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<8 x s16>) = COPY $q0
+    ; CHECK: [[UV:%[0-9]+]]:_(<4 x s16>), [[UV1:%[0-9]+]]:_(<4 x s16>) = G_UNMERGE_VALUES [[COPY]](<8 x s16>)
+    ; CHECK: [[OR:%[0-9]+]]:_(<4 x s16>) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[OR]](<4 x s16>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s16)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s16)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: [[OR3:%[0-9]+]]:_(s32) = G_OR [[COPY1]], [[COPY2]]
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR3]](s32)
+    ; CHECK: $w0 = COPY [[COPY3]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:_(<8 x s16>) = COPY $q0
+    %1:_(s16) = G_VECREDUCE_OR %0(<8 x s16>)
+    %2:_(s32) = G_ANYEXT %1(s16)
+    $w0 = COPY %2(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v16i16
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+  - { reg: '$q1' }
+body:             |
+  bb.1:
+    liveins: $q0, $q1
+
+    ; CHECK-LABEL: name: test_redor_v16i16
+    ; CHECK: liveins: $q0, $q1
+    ; CHECK: [[COPY:%[0-9]+]]:_(<8 x s16>) = COPY $q0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<8 x s16>) = COPY $q1
+    ; CHECK: [[OR:%[0-9]+]]:_(<8 x s16>) = G_OR [[COPY]], [[COPY1]]
+    ; CHECK: [[UV:%[0-9]+]]:_(<4 x s16>), [[UV1:%[0-9]+]]:_(<4 x s16>) = G_UNMERGE_VALUES [[OR]](<8 x s16>)
+    ; CHECK: [[OR1:%[0-9]+]]:_(<4 x s16>) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[OR1]](<4 x s16>)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV3]](s16)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[ANYEXT]], [[ANYEXT1]]
+    ; CHECK: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV4]](s16)
+    ; CHECK: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[UV5]](s16)
+    ; CHECK: [[OR3:%[0-9]+]]:_(s32) = G_OR [[ANYEXT2]], [[ANYEXT3]]
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s32) = COPY [[OR3]](s32)
+    ; CHECK: [[OR4:%[0-9]+]]:_(s32) = G_OR [[COPY2]], [[COPY3]]
+    ; CHECK: [[COPY4:%[0-9]+]]:_(s32) = COPY [[OR4]](s32)
+    ; CHECK: $w0 = COPY [[COPY4]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<8 x s16>) = COPY $q0
+    %2:_(<8 x s16>) = COPY $q1
+    %0:_(<16 x s16>) = G_CONCAT_VECTORS %1(<8 x s16>), %2(<8 x s16>)
+    %3:_(s16) = G_VECREDUCE_OR %0(<16 x s16>)
+    %4:_(s32) = G_ANYEXT %3(s16)
+    $w0 = COPY %4(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v2i32
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$d0' }
+body:             |
+  bb.1:
+    liveins: $d0
+
+    ; CHECK-LABEL: name: test_redor_v2i32
+    ; CHECK: liveins: $d0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $d0
+    ; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
+    ; CHECK: $w0 = COPY [[COPY1]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:_(<2 x s32>) = COPY $d0
+    %1:_(s32) = G_VECREDUCE_OR %0(<2 x s32>)
+    $w0 = COPY %1(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v4i32
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+body:             |
+  bb.1:
+    liveins: $q0
+
+    ; CHECK-LABEL: name: test_redor_v4i32
+    ; CHECK: liveins: $q0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $q0
+    ; CHECK: [[UV:%[0-9]+]]:_(<2 x s32>), [[UV1:%[0-9]+]]:_(<2 x s32>) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
+    ; CHECK: [[OR:%[0-9]+]]:_(<2 x s32>) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[OR]](<2 x s32>)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[UV2]], [[UV3]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[OR1]](s32)
+    ; CHECK: $w0 = COPY [[COPY1]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:_(<4 x s32>) = COPY $q0
+    %1:_(s32) = G_VECREDUCE_OR %0(<4 x s32>)
+    $w0 = COPY %1(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v8i32
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+  - { reg: '$q1' }
+body:             |
+  bb.1:
+    liveins: $q0, $q1
+
+    ; CHECK-LABEL: name: test_redor_v8i32
+    ; CHECK: liveins: $q0, $q1
+    ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $q0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $q1
+    ; CHECK: [[OR:%[0-9]+]]:_(<4 x s32>) = G_OR [[COPY]], [[COPY1]]
+    ; CHECK: [[UV:%[0-9]+]]:_(<2 x s32>), [[UV1:%[0-9]+]]:_(<2 x s32>) = G_UNMERGE_VALUES [[OR]](<4 x s32>)
+    ; CHECK: [[OR1:%[0-9]+]]:_(<2 x s32>) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[OR1]](<2 x s32>)
+    ; CHECK: [[OR2:%[0-9]+]]:_(s32) = G_OR [[UV2]], [[UV3]]
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR2]](s32)
+    ; CHECK: $w0 = COPY [[COPY2]](s32)
+    ; CHECK: RET_ReallyLR implicit $w0
+    %1:_(<4 x s32>) = COPY $q0
+    %2:_(<4 x s32>) = COPY $q1
+    %0:_(<8 x s32>) = G_CONCAT_VECTORS %1(<4 x s32>), %2(<4 x s32>)
+    %3:_(s32) = G_VECREDUCE_OR %0(<8 x s32>)
+    $w0 = COPY %3(s32)
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            test_redor_v2i64
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+body:             |
+  bb.1:
+    liveins: $q0
+
+    ; CHECK-LABEL: name: test_redor_v2i64
+    ; CHECK: liveins: $q0
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $q0
+    ; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; CHECK: [[OR:%[0-9]+]]:_(s64) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY [[OR]](s64)
+    ; CHECK: $x0 = COPY [[COPY1]](s64)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %0:_(<2 x s64>) = COPY $q0
+    %1:_(s64) = G_VECREDUCE_OR %0(<2 x s64>)
+    $x0 = COPY %1(s64)
+    RET_ReallyLR implicit $x0
+
+...
+---
+name:            test_redor_v4i64
+alignment:       4
+tracksRegLiveness: true
+liveins:
+  - { reg: '$q0' }
+  - { reg: '$q1' }
+body:             |
+  bb.1:
+    liveins: $q0, $q1
+
+    ; CHECK-LABEL: name: test_redor_v4i64
+    ; CHECK: liveins: $q0, $q1
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $q0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $q1
+    ; CHECK: [[OR:%[0-9]+]]:_(<2 x s64>) = G_OR [[COPY]], [[COPY1]]
+    ; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[OR]](<2 x s64>)
+    ; CHECK: [[OR1:%[0-9]+]]:_(s64) = G_OR [[UV]], [[UV1]]
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY [[OR1]](s64)
+    ; CHECK: $x0 = COPY [[COPY2]](s64)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %1:_(<2 x s64>) = COPY $q0
+    %2:_(<2 x s64>) = COPY $q1
+    %0:_(<4 x s64>) = G_CONCAT_VECTORS %1(<2 x s64>), %2(<2 x s64>)
+    %3:_(s64) = G_VECREDUCE_OR %0(<4 x s64>)
+    $x0 = COPY %3(s64)
+    RET_ReallyLR implicit $x0
+
+...

llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir

@@ -673,8 +673,8 @@
 # DEBUG-NEXT: .. type index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT: G_VECREDUCE_OR (opcode {{[0-9]+}}): 2 type indices, 0 imm indices
-# DEBUG-NEXT: .. type index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT: .. type index coverage check SKIPPED: user-defined predicate detected 
-# DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined
+# DEBUG-NEXT: .. imm index coverage check SKIPPED: user-defined predicate detected
 # DEBUG-NEXT: G_VECREDUCE_XOR (opcode {{[0-9]+}}): 2 type indices, 0 imm indices
 # DEBUG-NEXT: .. type index coverage check SKIPPED: no rules defined
 # DEBUG-NEXT: .. imm index coverage check SKIPPED: no rules defined

llvm/test/CodeGen/AArch64/reduce-or.ll

@@ -1,11 +1,18 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=aarch64-none-linux-gnu -mattr=+neon | FileCheck %s
+; RUN: llc < %s -mtriple=aarch64-none-linux-gnu -mattr=+neon -global-isel -global-isel-abort=1 | FileCheck %s --check-prefix=GISEL
+
 
 define i1 @test_redor_v1i1(<1 x i1> %a) {
 ; CHECK-LABEL: test_redor_v1i1:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    and w0, w0, #0x1
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v1i1:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    and w0, w0, #0x1
+; GISEL-NEXT:    ret
   %or_result = call i1 @llvm.vector.reduce.or.v1i1(<1 x i1> %a)
   ret i1 %or_result
 }
@@ -19,6 +26,16 @@ define i1 @test_redor_v2i1(<2 x i1> %a) {
 ; CHECK-NEXT:    orr w8, w9, w8
 ; CHECK-NEXT:    and w0, w8, #0x1
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v2i1:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    // kill: def $d0 killed $d0 def $q0
+; GISEL-NEXT:    mov s1, v0.s[1]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    and w0, w8, #0x1
+; GISEL-NEXT:    ret
   %or_result = call i1 @llvm.vector.reduce.or.v2i1(<2 x i1> %a)
   ret i1 %or_result
 }
@@ -36,6 +53,22 @@ define i1 @test_redor_v4i1(<4 x i1> %a) {
 ; CHECK-NEXT:    orr w8, w9, w8
 ; CHECK-NEXT:    and w0, w8, #0x1
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v4i1:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    // kill: def $d0 killed $d0 def $q0
+; GISEL-NEXT:    mov h1, v0.h[1]
+; GISEL-NEXT:    mov h2, v0.h[2]
+; GISEL-NEXT:    mov h3, v0.h[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    and w0, w8, #0x1
+; GISEL-NEXT:    ret
   %or_result = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> %a)
   ret i1 %or_result
 }
@@ -61,6 +94,34 @@ define i1 @test_redor_v8i1(<8 x i1> %a) {
 ; CHECK-NEXT:    orr w8, w9, w8
 ; CHECK-NEXT:    and w0, w8, #0x1
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v8i1:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    // kill: def $d0 killed $d0 def $q0
+; GISEL-NEXT:    mov b1, v0.b[1]
+; GISEL-NEXT:    mov b2, v0.b[2]
+; GISEL-NEXT:    mov b3, v0.b[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    mov b4, v0.b[4]
+; GISEL-NEXT:    mov b5, v0.b[5]
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    mov b6, v0.b[6]
+; GISEL-NEXT:    mov b7, v0.b[7]
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    fmov w10, s4
+; GISEL-NEXT:    fmov w11, s5
+; GISEL-NEXT:    orr w10, w10, w11
+; GISEL-NEXT:    fmov w11, s6
+; GISEL-NEXT:    fmov w12, s7
+; GISEL-NEXT:    orr w11, w11, w12
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    orr w9, w10, w11
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    and w0, w8, #0x1
+; GISEL-NEXT:    ret
   %or_result = call i1 @llvm.vector.reduce.or.v8i1(<8 x i1> %a)
   ret i1 %or_result
 }
@@ -87,6 +148,57 @@ define i1 @test_redor_v16i1(<16 x i1> %a) {
 ; CHECK-NEXT:    orr w8, w8, w9
 ; CHECK-NEXT:    and w0, w8, #0x1
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v16i1:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    mov b1, v0.b[1]
+; GISEL-NEXT:    mov b2, v0.b[2]
+; GISEL-NEXT:    mov b3, v0.b[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    mov b4, v0.b[4]
+; GISEL-NEXT:    mov b5, v0.b[5]
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    mov b6, v0.b[6]
+; GISEL-NEXT:    mov b7, v0.b[7]
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    fmov w10, s4
+; GISEL-NEXT:    fmov w11, s5
+; GISEL-NEXT:    mov b16, v0.b[8]
+; GISEL-NEXT:    mov b17, v0.b[9]
+; GISEL-NEXT:    orr w10, w10, w11
+; GISEL-NEXT:    fmov w11, s6
+; GISEL-NEXT:    fmov w12, s7
+; GISEL-NEXT:    mov b18, v0.b[10]
+; GISEL-NEXT:    mov b19, v0.b[11]
+; GISEL-NEXT:    orr w11, w11, w12
+; GISEL-NEXT:    fmov w12, s16
+; GISEL-NEXT:    fmov w13, s17
+; GISEL-NEXT:    mov b20, v0.b[12]
+; GISEL-NEXT:    mov b21, v0.b[13]
+; GISEL-NEXT:    orr w12, w12, w13
+; GISEL-NEXT:    fmov w13, s18
+; GISEL-NEXT:    fmov w14, s19
+; GISEL-NEXT:    mov b22, v0.b[14]
+; GISEL-NEXT:    mov b23, v0.b[15]
+; GISEL-NEXT:    orr w13, w13, w14
+; GISEL-NEXT:    fmov w14, s20
+; GISEL-NEXT:    fmov w15, s21
+; GISEL-NEXT:    orr w14, w14, w15
+; GISEL-NEXT:    fmov w15, s22
+; GISEL-NEXT:    fmov w16, s23
+; GISEL-NEXT:    orr w15, w15, w16
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    orr w9, w10, w11
+; GISEL-NEXT:    orr w10, w12, w13
+; GISEL-NEXT:    orr w11, w14, w15
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    orr w9, w10, w11
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    and w0, w8, #0x1
+; GISEL-NEXT:    ret
   %or_result = call i1 @llvm.vector.reduce.or.v16i1(<16 x i1> %a)
   ret i1 %or_result
 }
@@ -97,6 +209,12 @@ define i8 @test_redor_v1i8(<1 x i8> %a) {
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
 ; CHECK-NEXT:    umov w0, v0.b[0]
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v1i8:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    fmov x0, d0
+; GISEL-NEXT:    // kill: def $w0 killed $w0 killed $x0
+; GISEL-NEXT:    ret
   %or_result = call i8 @llvm.vector.reduce.or.v1i8(<1 x i8> %a)
   ret i8 %or_result
 }
@@ -107,6 +225,12 @@ define i8 @test_redor_v3i8(<3 x i8> %a) {
 ; CHECK-NEXT:    orr w8, w0, w1
 ; CHECK-NEXT:    orr w0, w8, w2
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v3i8:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    orr w8, w0, w1
+; GISEL-NEXT:    orr w0, w8, w2
+; GISEL-NEXT:    ret
   %or_result = call i8 @llvm.vector.reduce.or.v3i8(<3 x i8> %a)
   ret i8 %or_result
 }
@@ -123,6 +247,21 @@ define i8 @test_redor_v4i8(<4 x i8> %a) {
 ; CHECK-NEXT:    orr w9, w10, w9
 ; CHECK-NEXT:    orr w0, w9, w8
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v4i8:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    // kill: def $d0 killed $d0 def $q0
+; GISEL-NEXT:    mov h1, v0.h[1]
+; GISEL-NEXT:    mov h2, v0.h[2]
+; GISEL-NEXT:    mov h3, v0.h[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i8 @llvm.vector.reduce.or.v4i8(<4 x i8> %a)
   ret i8 %or_result
 }
@@ -147,6 +286,33 @@ define i8 @test_redor_v8i8(<8 x i8> %a) {
 ; CHECK-NEXT:    orr w9, w10, w9
 ; CHECK-NEXT:    orr w0, w9, w8
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v8i8:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    // kill: def $d0 killed $d0 def $q0
+; GISEL-NEXT:    mov b1, v0.b[1]
+; GISEL-NEXT:    mov b2, v0.b[2]
+; GISEL-NEXT:    mov b3, v0.b[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    mov b4, v0.b[4]
+; GISEL-NEXT:    mov b5, v0.b[5]
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    mov b6, v0.b[6]
+; GISEL-NEXT:    mov b7, v0.b[7]
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    fmov w10, s4
+; GISEL-NEXT:    fmov w11, s5
+; GISEL-NEXT:    orr w10, w10, w11
+; GISEL-NEXT:    fmov w11, s6
+; GISEL-NEXT:    fmov w12, s7
+; GISEL-NEXT:    orr w11, w11, w12
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    orr w9, w10, w11
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i8 @llvm.vector.reduce.or.v8i8(<8 x i8> %a)
   ret i8 %or_result
 }
@@ -172,6 +338,34 @@ define i8 @test_redor_v16i8(<16 x i8> %a) {
 ; CHECK-NEXT:    umov w9, v0.b[7]
 ; CHECK-NEXT:    orr w0, w8, w9
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v16i8:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    mov d1, v0.d[1]
+; GISEL-NEXT:    orr v0.8b, v0.8b, v1.8b
+; GISEL-NEXT:    mov b1, v0.b[1]
+; GISEL-NEXT:    mov b2, v0.b[2]
+; GISEL-NEXT:    mov b3, v0.b[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    mov b4, v0.b[4]
+; GISEL-NEXT:    mov b5, v0.b[5]
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    mov b6, v0.b[6]
+; GISEL-NEXT:    mov b7, v0.b[7]
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    fmov w10, s4
+; GISEL-NEXT:    fmov w11, s5
+; GISEL-NEXT:    orr w10, w10, w11
+; GISEL-NEXT:    fmov w11, s6
+; GISEL-NEXT:    fmov w12, s7
+; GISEL-NEXT:    orr w11, w11, w12
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    orr w9, w10, w11
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i8 @llvm.vector.reduce.or.v16i8(<16 x i8> %a)
   ret i8 %or_result
 }
@@ -198,6 +392,35 @@ define i8 @test_redor_v32i8(<32 x i8> %a) {
 ; CHECK-NEXT:    umov w9, v0.b[7]
 ; CHECK-NEXT:    orr w0, w8, w9
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v32i8:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    orr v0.16b, v0.16b, v1.16b
+; GISEL-NEXT:    mov d1, v0.d[1]
+; GISEL-NEXT:    orr v0.8b, v0.8b, v1.8b
+; GISEL-NEXT:    mov b1, v0.b[1]
+; GISEL-NEXT:    mov b2, v0.b[2]
+; GISEL-NEXT:    mov b3, v0.b[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    mov b4, v0.b[4]
+; GISEL-NEXT:    mov b5, v0.b[5]
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    mov b6, v0.b[6]
+; GISEL-NEXT:    mov b7, v0.b[7]
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    fmov w10, s4
+; GISEL-NEXT:    fmov w11, s5
+; GISEL-NEXT:    orr w10, w10, w11
+; GISEL-NEXT:    fmov w11, s6
+; GISEL-NEXT:    fmov w12, s7
+; GISEL-NEXT:    orr w11, w11, w12
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    orr w9, w10, w11
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i8 @llvm.vector.reduce.or.v32i8(<32 x i8> %a)
   ret i8 %or_result
 }
@@ -214,6 +437,21 @@ define i16 @test_redor_v4i16(<4 x i16> %a) {
 ; CHECK-NEXT:    orr w9, w10, w9
 ; CHECK-NEXT:    orr w0, w9, w8
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v4i16:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    // kill: def $d0 killed $d0 def $q0
+; GISEL-NEXT:    mov h1, v0.h[1]
+; GISEL-NEXT:    mov h2, v0.h[2]
+; GISEL-NEXT:    mov h3, v0.h[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i16 @llvm.vector.reduce.or.v4i16(<4 x i16> %a)
   ret i16 %or_result
 }
@@ -231,6 +469,22 @@ define i16 @test_redor_v8i16(<8 x i16> %a) {
 ; CHECK-NEXT:    umov w9, v0.h[3]
 ; CHECK-NEXT:    orr w0, w8, w9
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v8i16:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    mov d1, v0.d[1]
+; GISEL-NEXT:    orr v0.8b, v0.8b, v1.8b
+; GISEL-NEXT:    mov h1, v0.h[1]
+; GISEL-NEXT:    mov h2, v0.h[2]
+; GISEL-NEXT:    mov h3, v0.h[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i16 @llvm.vector.reduce.or.v8i16(<8 x i16> %a)
   ret i16 %or_result
 }
@@ -249,6 +503,23 @@ define i16 @test_redor_v16i16(<16 x i16> %a) {
 ; CHECK-NEXT:    umov w9, v0.h[3]
 ; CHECK-NEXT:    orr w0, w8, w9
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v16i16:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    orr v0.16b, v0.16b, v1.16b
+; GISEL-NEXT:    mov d1, v0.d[1]
+; GISEL-NEXT:    orr v0.8b, v0.8b, v1.8b
+; GISEL-NEXT:    mov h1, v0.h[1]
+; GISEL-NEXT:    mov h2, v0.h[2]
+; GISEL-NEXT:    mov h3, v0.h[3]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w8, w8, w9
+; GISEL-NEXT:    fmov w9, s2
+; GISEL-NEXT:    fmov w10, s3
+; GISEL-NEXT:    orr w9, w9, w10
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i16 @llvm.vector.reduce.or.v16i16(<16 x i16> %a)
   ret i16 %or_result
 }
@@ -261,6 +532,15 @@ define i32 @test_redor_v2i32(<2 x i32> %a) {
 ; CHECK-NEXT:    fmov w9, s0
 ; CHECK-NEXT:    orr w0, w9, w8
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v2i32:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    // kill: def $d0 killed $d0 def $q0
+; GISEL-NEXT:    mov s1, v0.s[1]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> %a)
   ret i32 %or_result
 }
@@ -274,6 +554,16 @@ define i32 @test_redor_v4i32(<4 x i32> %a) {
 ; CHECK-NEXT:    fmov w9, s0
 ; CHECK-NEXT:    orr w0, w9, w8
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v4i32:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    mov d1, v0.d[1]
+; GISEL-NEXT:    orr v0.8b, v0.8b, v1.8b
+; GISEL-NEXT:    mov s1, v0.s[1]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> %a)
   ret i32 %or_result
 }
@@ -288,6 +578,17 @@ define i32 @test_redor_v8i32(<8 x i32> %a) {
 ; CHECK-NEXT:    fmov w9, s0
 ; CHECK-NEXT:    orr w0, w9, w8
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v8i32:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    orr v0.16b, v0.16b, v1.16b
+; GISEL-NEXT:    mov d1, v0.d[1]
+; GISEL-NEXT:    orr v0.8b, v0.8b, v1.8b
+; GISEL-NEXT:    mov s1, v0.s[1]
+; GISEL-NEXT:    fmov w8, s0
+; GISEL-NEXT:    fmov w9, s1
+; GISEL-NEXT:    orr w0, w8, w9
+; GISEL-NEXT:    ret
   %or_result = call i32 @llvm.vector.reduce.or.v8i32(<8 x i32> %a)
   ret i32 %or_result
 }
@@ -299,6 +600,14 @@ define i64 @test_redor_v2i64(<2 x i64> %a) {
 ; CHECK-NEXT:    orr v0.8b, v0.8b, v1.8b
 ; CHECK-NEXT:    fmov x0, d0
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v2i64:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    mov d1, v0.d[1]
+; GISEL-NEXT:    fmov x8, d0
+; GISEL-NEXT:    fmov x9, d1
+; GISEL-NEXT:    orr x0, x8, x9
+; GISEL-NEXT:    ret
   %or_result = call i64 @llvm.vector.reduce.or.v2i64(<2 x i64> %a)
   ret i64 %or_result
 }
@@ -311,6 +620,15 @@ define i64 @test_redor_v4i64(<4 x i64> %a) {
 ; CHECK-NEXT:    orr v0.8b, v0.8b, v1.8b
 ; CHECK-NEXT:    fmov x0, d0
 ; CHECK-NEXT:    ret
+;
+; GISEL-LABEL: test_redor_v4i64:
+; GISEL:       // %bb.0:
+; GISEL-NEXT:    orr v0.16b, v0.16b, v1.16b
+; GISEL-NEXT:    mov d1, v0.d[1]
+; GISEL-NEXT:    fmov x8, d0
+; GISEL-NEXT:    fmov x9, d1
+; GISEL-NEXT:    orr x0, x8, x9
+; GISEL-NEXT:    ret
   %or_result = call i64 @llvm.vector.reduce.or.v4i64(<4 x i64> %a)
   ret i64 %or_result
 }

llvm/test/MachineVerifier/test_vector_reductions.mir

@@ -30,6 +30,4 @@ body:             |
     %dst:_(s64) = G_VECREDUCE_SEQ_FADD %scalar_s64, %scalar_s64
     ; CHECK: Bad machine code: Sequential FADD/FMUL vector reduction must have a vector 2nd operand
 
-    %dst2:_(s64) = G_VECREDUCE_MUL %scalar_s64
-    ; CHECK: Bad machine code: Vector reduction requires vector source
 ...