[Generic][AIE2] Combiner for shufflevectors that use build vector

Transforms a shufflevector that uses a build vector or undefined
into just a build vector. This can be done is because a shuffle
vector lowering is an unmerge and then merge. Since build is a
merge, the merge and unmerge cancel each other out and we can just
merge the vector directly.

Example:
```
    %1:_(s32) = COPY $r0
    %3:_(<8 x s32>) = G_IMPLICIT_DEF
    %5:_(s32) = G_IMPLICIT_DEF
    %2:_(<8 x s32>) = G_BUILD_VECTOR %1(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32), %5(s32)
    %0:_(<8 x s32>) = G_SHUFFLE_VECTOR %2(<8 x s32>), %3, shufflemask(0, 0, 0, 0, 0, 0, 0, 0)
    ===>
    %2:_(<8 x s32>) = G_BUILD_VECTOR %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32), %1(s32)
```

llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h

@@ -352,6 +352,15 @@ class CombinerHelper {
   applyCombineUnmergeMergeToPlainValues(MachineInstr &MI,
                                         SmallVectorImpl<Register> &Operands);
 
+  /// Transform <ty, ...> G_SHUFFLE_VECTOR(G_MERGE ty X Y Z) -> G_MERGE
+  bool
+  matchCombineShuffleVectorBuildVector(MachineInstr &MI,
+                                       SmallVectorImpl<Register> &Operands);
+
+  void
+  applyCombineShuffleVectorBuildVector(MachineInstr &MI,
+                                       SmallVectorImpl<Register> &Operands);
+
   /// Transform G_UNMERGE Constant -> Constant1, Constant2, ...
   bool matchCombineUnmergeConstant(MachineInstr &MI,
                                    SmallVectorImpl<APInt> &Csts);

llvm/include/llvm/CodeGen/GlobalISel/GenericMachineInstrs.h

@@ -225,9 +225,9 @@ class GUnmerge : public GenericMachineInstr {
   }
 };
 
-/// Represents G_BUILD_VECTOR, G_CONCAT_VECTORS or G_MERGE_VALUES.
-/// All these have the common property of generating a single value from
-/// multiple sources.
+/// Represents G_BUILD_VECTOR, G_CONCAT_VECTORS, G_SHUFFLE_VECTOR or
+/// G_MERGE_VALUES. All these have the common property of generating a single
+/// value from multiple sources.
 class GMergeLikeInstr : public GenericMachineInstr {
 public:
   /// Returns the number of source registers.
@@ -240,13 +240,22 @@ class GMergeLikeInstr : public GenericMachineInstr {
     case TargetOpcode::G_MERGE_VALUES:
     case TargetOpcode::G_CONCAT_VECTORS:
     case TargetOpcode::G_BUILD_VECTOR:
+    case TargetOpcode::G_SHUFFLE_VECTOR:
       return true;
     default:
       return false;
     }
   }
 };
 
+/// Represents a G_SHUFFLE_VECTOR
+class GShuffleVector : public GMergeLikeInstr {
+public:
+  static bool classof(const MachineInstr *MI) {
+    return MI->getOpcode() == TargetOpcode::G_SHUFFLE_VECTOR;
+  }
+};
+
 /// Represents a G_MERGE_VALUES.
 class GMerge : public GMergeLikeInstr {
 public:

llvm/include/llvm/Target/GlobalISel/Combine.td

@@ -758,6 +758,16 @@ def unmerge_merge : GICombineRule<
   (apply [{ Helper.applyCombineUnmergeMergeToPlainValues(*${d}, ${info}); }])
 >;
 
+// Fold (unmerge(merge x, y, z)) -> z, y, z.
+def shufflevector_merge_matchinfo : GIDefMatchData<"SmallVector<Register, 8>">;
+def shufflevector_merge : GICombineRule<
+  (defs root:$d, shufflevector_merge_matchinfo:$info),
+  (match (wip_match_opcode G_SHUFFLE_VECTOR): $d,
+  [{ return Helper.matchCombineShuffleVectorBuildVector(*${d}, ${info}); }]),
+  (apply [{ Helper.applyCombineShuffleVectorBuildVector(*${d}, ${info}); }])
+>;
+
+
 // Fold merge(unmerge).
 def merge_unmerge : GICombineRule<
   (defs root:$d, register_matchinfo:$matchinfo),
@@ -1324,7 +1334,7 @@ def all_combines : GICombineGroup<[trivial_combines, insert_vec_elt_combines,
     intdiv_combines, mulh_combines, redundant_neg_operands,
     and_or_disjoint_mask, fma_combines, fold_binop_into_select,
     sub_add_reg, select_to_minmax, redundant_binop_in_equality,
-    fsub_to_fneg, commute_constant_to_rhs]>;
+    fsub_to_fneg, commute_constant_to_rhs, shufflevector_merge]>;
 
 // A combine group used to for prelegalizer combiners at -O0. The combines in
 // this group have been selected based on experiments to balance code size and

llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp

@@ -13,6 +13,7 @@
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallBitVector.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
 #include "llvm/CodeGen/GlobalISel/GISelKnownBits.h"
 #include "llvm/CodeGen/GlobalISel/GenericMachineInstrs.h"
@@ -26,6 +27,7 @@
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterBankInfo.h"
@@ -2147,6 +2149,77 @@ static Register peekThroughBitcast(Register Reg,
   return Reg;
 }
 
+bool CombinerHelper::matchCombineShuffleVectorBuildVector(
+    MachineInstr &MI, SmallVectorImpl<Register> &Operands) {
+  assert(MI.getOpcode() == TargetOpcode::G_SHUFFLE_VECTOR &&
+         "Expected a shuffle vector");
+  auto &ShuffleVector = cast<GShuffleVector>(MI);
+  Register SrcReg1 = peekThroughBitcast(ShuffleVector.getSourceReg(0), MRI);
+  Register SrcReg2 = peekThroughBitcast(ShuffleVector.getSourceReg(1), MRI);
+
+  // Check if the Source registers are either merges or implicit definitions
+  auto *SrcInstr1 = getOpcodeDef<GBuildVector>(SrcReg1, MRI);
+  auto *SrcInstr2 = getOpcodeDef<GBuildVector>(SrcReg2, MRI);
+  auto *IsUndef1 = getOpcodeDef(TargetOpcode::G_IMPLICIT_DEF, SrcReg1, MRI);
+  auto *IsUndef2 = getOpcodeDef(TargetOpcode::G_IMPLICIT_DEF, SrcReg2, MRI);
+
+  // Our inputs need to be either be build vectors or undefined, register inputs
+  // break this optimization. You could maybe do something clever were you
+  // concatenate vectors to save half a build vector.
+  if ((SrcInstr1 == 0 && IsUndef1 == 0) || (SrcInstr2 == 0 && IsUndef2 == 0))
+    return false;
+
+  Register UndefReg;
+  if (SrcInstr1 || SrcInstr2)
+    UndefReg = MRI.createGenericVirtualRegister(MRI.getType(SrcReg1));
+
+  // Since our inputs to shufflevector must be of the same size, we can reuse
+  // the size of the defined register.
+  const unsigned NumElements = (SrcInstr1 != 0) ? SrcInstr1->getNumSources()
+                                                : SrcInstr2->getNumSources();
+  for (unsigned Idx = 0; Idx < NumElements; ++Idx) {
+    const Register Elt =
+        (SrcInstr1 != 0) ? SrcInstr1->getSourceReg(Idx) : UndefReg;
+    Operands.push_back(Elt);
+  }
+
+  for (unsigned Idx = 0; Idx < NumElements; ++Idx) {
+    const Register Elt =
+        (SrcInstr2 != 0) ? SrcInstr2->getSourceReg(Idx) : UndefReg;
+    Operands.push_back(Elt);
+  }
+
+  return true;
+}
+
+void CombinerHelper::applyCombineShuffleVectorBuildVector(
+    MachineInstr &MI, SmallVectorImpl<Register> &Operands) {
+  assert(MI.getOpcode() == TargetOpcode::G_SHUFFLE_VECTOR &&
+         "Expected a shuffle vector");
+  auto &ShuffleVector = cast<GShuffleVector>(MI);
+  Register SrcReg1 = peekThroughBitcast(ShuffleVector.getSourceReg(0), MRI);
+  Register SrcReg2 = peekThroughBitcast(ShuffleVector.getSourceReg(1), MRI);
+
+  // Check if the Source registers are either merges or implicit definitions
+  auto *IsUndef1 = getOpcodeDef(TargetOpcode::G_IMPLICIT_DEF, SrcReg1, MRI);
+  auto *IsUndef2 = getOpcodeDef(TargetOpcode::G_IMPLICIT_DEF, SrcReg2, MRI);
+
+  // If they're both undefined, we will just return an undefined as well.
+  if (IsUndef1 && IsUndef2) {
+    Builder.buildUndef(ShuffleVector.getReg(0));
+    MI.eraseFromParent();
+    return;
+  }
+
+  const ArrayRef<int> ShiftMask = MI.getOperand(3).getShuffleMask();
+  SmallVector<Register, 8> Arguments;
+  for (int Index : ShiftMask)
+    Arguments.push_back(Operands[Index]);
+
+  Builder.buildBuildVector(ShuffleVector.getOperand(0), Arguments);
+  MI.eraseFromParent();
+}
+
 bool CombinerHelper::matchCombineUnmergeMergeToPlainValues(
     MachineInstr &MI, SmallVectorImpl<Register> &Operands) {
   assert(MI.getOpcode() == TargetOpcode::G_UNMERGE_VALUES &&