GlobalISel: Avoid use of G_INSERT in insertParts

G_INSERT legalization is incomplete and doesn't work very
well. Instead try to use sequences of G_MERGE_VALUES/G_UNMERGE_VALUES
padding with undef values (although this can get pretty large).

For the case of load/store narrowing, this is still performing the
load/stores in irregularly sized pieces. It might be cleaner to split
this down into equal sized pieces, and rely on load/store merging to
optimize it.

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

@@ -220,29 +220,16 @@ void LegalizerHelper::insertParts(Register DstReg,
     return;
   }
 
-  unsigned PartSize = PartTy.getSizeInBits();
-  unsigned LeftoverPartSize = LeftoverTy.getSizeInBits();
+  SmallVector<Register> GCDRegs;
+  LLT GCDTy;
+  for (Register PartReg : PartRegs)
+    GCDTy = extractGCDType(GCDRegs, ResultTy, LeftoverTy, PartReg);
 
-  Register CurResultReg = MRI.createGenericVirtualRegister(ResultTy);
-  MIRBuilder.buildUndef(CurResultReg);
+  for (Register PartReg : LeftoverRegs)
+    extractGCDType(GCDRegs, ResultTy, LeftoverTy, PartReg);
 
-  unsigned Offset = 0;
-  for (Register PartReg : PartRegs) {
-    Register NewResultReg = MRI.createGenericVirtualRegister(ResultTy);
-    MIRBuilder.buildInsert(NewResultReg, CurResultReg, PartReg, Offset);
-    CurResultReg = NewResultReg;
-    Offset += PartSize;
-  }
-
-  for (unsigned I = 0, E = LeftoverRegs.size(); I != E; ++I) {
-    // Use the original output register for the final insert to avoid a copy.
-    Register NewResultReg = (I + 1 == E) ?
-      DstReg : MRI.createGenericVirtualRegister(ResultTy);
-
-    MIRBuilder.buildInsert(NewResultReg, CurResultReg, LeftoverRegs[I], Offset);
-    CurResultReg = NewResultReg;
-    Offset += LeftoverPartSize;
-  }
+  LLT ResultLCMTy = buildLCMMergePieces(ResultTy, LeftoverTy, GCDTy, GCDRegs);
+  buildWidenedRemergeToDst(DstReg, ResultLCMTy, GCDRegs);
 }
 
 /// Append the result registers of G_UNMERGE_VALUES \p MI to \p Regs.

llvm/test/CodeGen/AArch64/GlobalISel/arm64-fallback.ll

@@ -80,7 +80,7 @@ end:
   br label %block
 }
 
-; FALLBACK-WITH-REPORT-ERR: remark: <unknown>:0:0: cannot select: %{{[0-9]+}}:gpr32(s32), %{{[0-9]+}}:gpr(s1) = G_UADDE %{{[0-9]+}}:gpr, %{{[0-9]+}}:gpr, %{{[0-9]+}}:gpr (in function: nonpow2_add_narrowing)
+; FALLBACK-WITH-REPORT-ERR: remark: <unknown>:0:0: cannot select: %{{[0-9]+}}:gpr(s32), %{{[0-9]+}}:gpr(s32) = G_UNMERGE_VALUES %{{[0-9]+}}:gpr(s64) (in function: nonpow2_add_narrowing)
 ; FALLBACK-WITH-REPORT-ERR: warning: Instruction selection used fallback path for nonpow2_add_narrowing
 ; FALLBACK-WITH-REPORT-OUT-LABEL: nonpow2_add_narrowing:
 define void @nonpow2_add_narrowing(i128 %x, i128 %y) {