[AArch64] Fix selection of G_UNMERGE <2 x s16>

[aemerson]

Thanks for fixing this, LGTM.

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

   // If our elements fit into exactly 128 bits, then we can copy from the source
   // directly. Otherwise, we need to do a bit of setup with some subregister
   // inserts.
   if (NarrowTy.getSizeInBits() * NumElts == 128) {
     InsertRegs = SmallVector<Register, 4>(NumInsertRegs, SrcReg);
   } else {
     // No. We have to perform subregister inserts. For each insert, create an
     // implicit def and a subregister insert, and save the register we create.
+    const TargetRegisterClass *RC =
+        getMinClassForRegBank(*RBI.getRegBank(SrcReg, MRI, TRI),
+                              WideTy.getScalarSizeInBits() * NumElts);
+    unsigned SubReg = 0;
+    bool Found = getSubRegForClass(RC, TRI, SubReg);
+    (void)Found;
+    assert(Found && "expected to find last operand's subeg idx");
     for (unsigned Idx = 0; Idx < NumInsertRegs; ++Idx) {
       Register ImpDefReg = MRI.createVirtualRegister(&AArch64::FPR128RegClass);
       MachineInstr &ImpDefMI =
           *BuildMI(MBB, I, I.getDebugLoc(), TII.get(TargetOpcode::IMPLICIT_DEF),
                    ImpDefReg);
 
       // Now, create the subregister insert from SrcReg.
       Register InsertReg = MRI.createVirtualRegister(&AArch64::FPR128RegClass);
       MachineInstr &InsMI =
           *BuildMI(MBB, I, I.getDebugLoc(),
                    TII.get(TargetOpcode::INSERT_SUBREG), InsertReg)
                .addUse(ImpDefReg)
                .addUse(SrcReg)
-               .addImm(AArch64::dsub);
+               .addImm(SubReg);
 
       constrainSelectedInstRegOperands(ImpDefMI, TII, TRI, RBI);
       constrainSelectedInstRegOperands(InsMI, TII, TRI, RBI);
 
       // Save the register so that we can copy from it after.
       InsertRegs.push_back(InsertReg);
     }
   }

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

 
 # RUN: llc -O0 -mattr=-fullfp16 -mtriple=aarch64-- \
 # RUN: -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s
 
 --- |
   define <2 x double> @test_v2s64_unmerge(<2 x double> %a) {
     ret <2 x double> %a
   }
 
   define <4 x float> @test_v4s32_unmerge(<4 x float> %a) {
     ret <4 x float> %a
   }
 
+  define <2 x half> @test_v2s16_unmerge(<2 x half> %a) {
+    ret <2 x half> %a
+  }
+
   define <4 x half> @test_v4s16_unmerge(<4 x half> %a) {
     ret <4 x half> %a
   }
 
   define <8 x half> @test_v8s16_unmerge(<8 x half> %a) {
     ret <8 x half> %a
   }
 
bb.1 (%ir-block.0):
     ; CHECK: %5:fpr32 = CPYi32 %0, 3
     %2:fpr(s32), %3:fpr(s32), %4:fpr(s32), %5:fpr(s32) = G_UNMERGE_VALUES %0(<4 x s32>)
 
     %1:fpr(<4 x s32>) = G_BUILD_VECTOR %2(s32), %3(s32), %4(s32), %5(s32)
     $q0 = COPY %1(<4 x s32>)
     RET_ReallyLR implicit $q0
 ...
 ---
+name:            test_v2s16_unmerge
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: fpr }
+  - { id: 1, class: fpr }
+  - { id: 2, class: fpr }
+  - { id: 3, class: fpr }
+  - { id: 4, class: fpr }
+  - { id: 5, class: fpr }
+body:             |
+  bb.1 (%ir-block.0):
+    liveins: $s0
+    ; CHECK-LABEL: name: test_v2s16_unmerge
+
+    ; CHECK: [[COPY:%[0-9]+]]:fpr32 = COPY $s0
+    %0:fpr(<2 x s16>) = COPY $s0
+
+    ; Since 2 * 16 != 128, we need to widen using implicit defs.
+    ; Note that we expect to reuse one of the INSERT_SUBREG results, as CPYi16
+    ; expects a lane > 0.
+    ; CHECK: [[DEF:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK: [[INSERT_SUBREG:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF]], [[COPY]], %subreg.ssub
+    ; CHECK: [[COPY1:%[0-9]+]]:fpr16 = COPY [[INSERT_SUBREG]].hsub
+    ; CHECK: [[CPYi16_:%[0-9]+]]:fpr16 = CPYi16 [[INSERT_SUBREG]], 1
+    %2:fpr(s16), %3:fpr(s16) = G_UNMERGE_VALUES %0(<2 x s16>)
+
+    ; CHECK: [[DEF1:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK: [[INSERT_SUBREG1:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF1]], [[COPY1]], %subreg.hsub
+    ; CHECK: [[DEF2:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK: [[INSERT_SUBREG2:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF2]], [[CPYi16_]], %subreg.hsub
+    ; CHECK: [[INSvi16lane:%[0-9]+]]:fpr128 = INSvi16lane [[INSERT_SUBREG1]], 1, [[INSERT_SUBREG2]], 0
+    ; CHECK: [[COPY2:%[0-9]+]]:fpr32 = COPY [[INSvi16lane]].ssub
+    %1:fpr(<2 x s16>) = G_BUILD_VECTOR %2(s16), %3(s16)
+
+    ; CHECK: $s0 = COPY [[COPY2]]
+    $s0 = COPY %1(<2 x s16>)
+
+    ; CHECK: RET_ReallyLR implicit $s0
+    RET_ReallyLR implicit $s0
+...
+---
 name:            test_v4s16_unmerge
 alignment:       4
 legalized:       true
 regBankSelected: true
 tracksRegLiveness: true
 registers:
   - { id: 0, class: fpr }
   - { id: 1, class: fpr }