[AArch64][GlobalISel] Manually select G_DUP with s8/s16 gpr scalar operands.

These don't get selected by the imported patterns, and avoiding generating them
is a whole load of not-worth-it-hassle (until we have fp types in GlobalISel).

Author: aemerson

llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp

@@ -2971,6 +2971,24 @@ bool AArch64InstructionSelector::select(MachineInstr &I) {
       return constrainSelectedInstRegOperands(*MovMI, TII, TRI, RBI);
     }
   }
+  case AArch64::G_DUP: {
+    // When the scalar of G_DUP is an s8/s16 gpr, they can't be selected by
+    // imported patterns. Do it manually here. Avoiding generating s16 gpr is
+    // difficult because at RBS we may end up pessimizing the fpr case if we
+    // decided to add an anyextend to fix this. Manual selection is the most
+    // robust solution for now.
+    Register SrcReg = I.getOperand(1).getReg();
+    if (RBI.getRegBank(SrcReg, MRI, TRI)->getID() != AArch64::GPRRegBankID)
+      return false; // We expect the fpr regbank case to be imported.
+    LLT SrcTy = MRI.getType(SrcReg);
+    if (SrcTy.getSizeInBits() == 16)
+      I.setDesc(TII.get(AArch64::DUPv8i16gpr));
+    else if (SrcTy.getSizeInBits() == 8)
+      I.setDesc(TII.get(AArch64::DUPv16i8gpr));
+    else
+      return false;
+    return constrainSelectedInstRegOperands(I, TII, TRI, RBI);
+  }
   case TargetOpcode::G_INTRINSIC_TRUNC:
     return selectIntrinsicTrunc(I, MRI);
   case TargetOpcode::G_INTRINSIC_ROUND:

llvm/test/CodeGen/AArch64/GlobalISel/select-dup.mir

@@ -223,6 +223,29 @@ body:             |
     $q0 = COPY %dup(<8 x s16>)
     RET_ReallyLR implicit $q0
 
+...
+---
+name:            DUPv8i16gpr_s16_src
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: $w0
+    ; Checks that we can still select the gpr variant if the scalar is an s16.
+    ; CHECK-LABEL: name: DUPv8i16gpr_s16_src
+    ; CHECK: liveins: $w0
+    ; CHECK: %copy:gpr32 = COPY $w0
+    ; CHECK: %dup:fpr128 = DUPv8i16gpr %copy
+    ; CHECK: $q0 = COPY %dup
+    ; CHECK: RET_ReallyLR implicit $q0
+    %copy:gpr(s32) = COPY $w0
+    %trunc:gpr(s16) = G_TRUNC %copy
+    %dup:fpr(<8 x s16>) = G_DUP %trunc(s16)
+    $q0 = COPY %dup(<8 x s16>)
+    RET_ReallyLR implicit $q0
+
 ...
 ---
 name:            DUPv8i8gpr
@@ -264,3 +287,26 @@ body:             |
     %dup:fpr(<16 x s8>) = G_DUP %copy(s32)
     $q0 = COPY %dup(<16 x s8>)
     RET_ReallyLR implicit $q0
+...
+---
+name:            DUPv16i8gpr_s8_src
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+body:             |
+  bb.0.entry:
+    liveins: $w0
+    ; Check we still select the gpr variant when scalar is an s8.
+    ; CHECK-LABEL: name: DUPv16i8gpr_s8_src
+    ; CHECK: liveins: $w0
+    ; CHECK: %copy:gpr32 = COPY $w0
+    ; CHECK: %dup:fpr128 = DUPv16i8gpr %copy
+    ; CHECK: $q0 = COPY %dup
+    ; CHECK: RET_ReallyLR implicit $q0
+    %copy:gpr(s32) = COPY $w0
+    %trunc:gpr(s8) = G_TRUNC %copy
+    %dup:fpr(<16 x s8>) = G_DUP %trunc(s8)
+    $q0 = COPY %dup(<16 x s8>)
+    RET_ReallyLR implicit $q0
+...