[SLP]Fix a crash when insert subvector is out of range.

[ABataev]

Summary

If the OffsetBeg + InsertVecSz is greater than VecSz, need to estimate
the cost as shuffle of 2 vector, not as insert of subvector. Otherwise,
the inserted subvector is out of range and compiler may crash.

[ABataev]

Adjust OffsetBeg for better shuffle mask estimation.

[RKSimon]

LGTM

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

case Instruction::InsertElement: {
       unsigned VecSz =
           (1 + OffsetEnd / VecScalarsSz - OffsetBeg / VecScalarsSz) *
           VecScalarsSz;
       unsigned Offset = VecScalarsSz * (OffsetBeg / VecScalarsSz);
       unsigned InsertVecSz = std::min<unsigned>(
           PowerOf2Ceil(OffsetEnd - OffsetBeg + 1),
           ((OffsetEnd - OffsetBeg + VecScalarsSz) / VecScalarsSz) *
               VecScalarsSz);
+      bool IsWholeSubvector =
+          OffsetBeg == Offset && ((OffsetEnd + 1) % VecScalarsSz == 0);
+      // Check if we can safely insert a subvector. If it is not possible, just
+      // generate a whole-sized vector and shuffle the source vector and the new
+      // subvector.
+      if (OffsetBeg + InsertVecSz > VecSz) {
+        // Align OffsetBeg to generate correct mask.
+        OffsetBeg = alignDown(OffsetBeg, VecSz, Offset);
+        InsertVecSz = VecSz;
+      }
 
       APInt DemandedElts = APInt::getZero(NumElts);
       // TODO: Add support for Instruction::InsertValue.
       SmallVector<int> Mask;
       if (!E->ReorderIndices.empty()) {
         inversePermutation(E->ReorderIndices, Mask);
         Mask.append(InsertVecSz - Mask.size(), UndefMaskElem);
       } else {
case Instruction::InsertElement: {
       auto *FirstInsert = cast<Instruction>(*find_if(E->Scalars, [E](Value *V) {
         return !is_contained(E->Scalars, cast<Instruction>(V)->getOperand(0));
       }));
       // Second cost - permutation with subvector, if some elements are from the
       // initial vector or inserting a subvector.
       // TODO: Implement the analysis of the FirstInsert->getOperand(0)
       // subvector of ActualVecTy.
       if (!isUndefVector(FirstInsert->getOperand(0)) && NumScalars != NumElts &&
-          (Offset != OffsetBeg || (OffsetEnd + 1) % VecScalarsSz != 0)) {
+          !IsWholeSubvector) {
         if (InsertVecSz != VecSz) {
           auto *ActualVecTy =
               FixedVectorType::get(SrcVecTy->getElementType(), VecSz);
           Cost += TTI->getShuffleCost(TTI::SK_InsertSubvector, ActualVecTy,
                                       None, OffsetBeg - Offset, InsertVecTy);
         } else {
           for (unsigned I = 0, End = OffsetBeg - Offset; I < End; ++I)
             Mask[I] = I;

llvm/test/Transforms/SLPVectorizer/AArch64/buildvector-vectorize.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -slp-vectorizer -S -mtriple=aarch64 < %s | FileCheck %s
+
+define void @test(ptr %p) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[INC:%.*]] = getelementptr inbounds i16, ptr [[P:%.*]], i64 1
+; CHECK-NEXT:    [[TMP0:%.*]] = load i16, ptr [[INC]], align 4
+; CHECK-NEXT:    [[TMP1:%.*]] = load i16, ptr [[P]], align 2
+; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <8 x i16> zeroinitializer, i16 [[TMP0]], i32 5
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <8 x i16> [[TMP2]], i16 [[TMP1]], i32 7
+; CHECK-NEXT:    ret void
+;
+entry:
+  %inc = getelementptr inbounds i16, ptr %p, i64 1
+  %0 = load i16, ptr %inc, align 4
+  %1 = load i16, ptr %p, align 2
+  %2 = insertelement <8 x i16> zeroinitializer, i16 %0, i32 5
+  %3 = insertelement <8 x i16> %2, i16 %1, i32 7
+  ret void
+}