[Transforms][SROA] Promote allocas with mem2reg for scalable types

[c-rhodes]

Summary

Aggregate types containing scalable vectors aren't supported and as far
as I can tell this pass is mostly concerned with optimisations on
aggregate types, so the majority of this pass isn't very useful for
scalable vectors.

This patch modifies SROA such that mem2reg is run on allocas with
scalable types that are promotable, but nothing else such as slicing is
done.

The use of TypeSize in this pass has also been updated to be explicitly
fixed size. When invoking the following methods in DataLayout:

• getTypeSizeInBits
• getTypeStoreSize
• getTypeStoreSizeInBits
• getTypeAllocSize

we now called getFixedSize on the resultant TypeSize. This is quite an
extensive change with around 50 calls to these functions, and also the
first change of this kind (being explicit about fixed vs scalable
size) as far as I'm aware, so feedback welcome.

A test is included containing IR with scalable vectors that this pass is
able to optimise.

[efriedma]

We want SROA to at least run mem2reg on scalable vectors, since we don't run mem2reg separately. This is important for C code using SVE intrinsics.

I agree we don't need to care about slicing etc.; we probably can't slice an alloca of unknown size in most cases.

[c-rhodes]

In D76720#1939896, @efriedma wrote:

We want SROA to at least run mem2reg on scalable vectors, since we don't run mem2reg separately. This is important for C code using SVE intrinsics.

mem2reg isn't run as a separate pass? We have a clang ACLE test for brka intrinsic downstream that was hitting a lot of "Request for a fixed size on a scalable object" asserts in SROA, with this patch applied it gets optimised by mem2reg:

./build/bin/clang -cc1 -internal-isystem ./build/lib/clang/9.0.1/include -nostdsysteminc -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o -  tools/clang/test/CodeGen/AArch64/acle/acle_sve_brka.c -mllvm -print-before-pass=mem2reg -mllvm -print-after-pass=mem2reg
*** IR Dump Before Promote Memory to Register ***
; Function Attrs: nounwind
define <n x 16 x i1> @test_svbrka_b_z(<n x 16 x i1> %pg, <n x 16 x i1> %op) local_unnamed_addr #0 {
entry:
  %pg.addr = alloca <n x 16 x i1>, align 1
  %op.addr = alloca <n x 16 x i1>, align 1
  store <n x 16 x i1> %pg, <n x 16 x i1>* %pg.addr, align 1, !tbaa !2
  store <n x 16 x i1> %op, <n x 16 x i1>* %op.addr, align 1, !tbaa !2
  %0 = load <n x 16 x i1>, <n x 16 x i1>* %pg.addr, align 1, !tbaa !2
  %1 = call <n x 16 x i1> @llvm.aarch64.sve.brka.z.nxv16i1(<n x 16 x i1> %0, <n x 16 x i1> %op)
  ret <n x 16 x i1> %1
}
*** IR Dump After Promote Memory to Register ***
; Function Attrs: nounwind
define <n x 16 x i1> @test_svbrka_b_z(<n x 16 x i1> %pg, <n x 16 x i1> %op) local_unnamed_addr #0 {
entry:
  %0 = call <n x 16 x i1> @llvm.aarch64.sve.brka.z.nxv16i1(<n x 16 x i1> %pg, <n x 16 x i1> %op)
  ret <n x 16 x i1> %0
}

I ran our downstream unit tests with SROA disabled and there's no asm differences.

[c-rhodes]

In D76720#1939896, @efriedma wrote:

We want SROA to at least run mem2reg on scalable vectors, since we don't run mem2reg separately. This is important for C code using SVE intrinsics.

I've tested this on Sander's Clang patch (D76238) adding contiguous loads/stores upstream and mem2reg runs:

$ /home/culrho01/llvm-project/build/bin/clang -cc1 -internal-isystem /home/culrho01/llvm-project/build/lib/clang/11.0.0/include -nostdsysteminc -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - /home/culrho01/llvm-project/clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld1.c -D__ARM_FEATURE_SVE -mllvm -print-before-all -mllvm -print-after-all

*** IR Dump Before Promote Memory to Register ***
; Function Attrs: nounwind
define <vscale x 16 x i8> @test_svld1_s8(<vscale x 16 x i1> %pg, i8* %base) local_unnamed_addr #0 {
entry:
  %pg.addr = alloca <vscale x 16 x i1>, align 2
  store <vscale x 16 x i1> %pg, <vscale x 16 x i1>* %pg.addr, align 2, !tbaa !2
  %0 = bitcast i8* %base to <vscale x 16 x i8>*
  %1 = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* %0, i32 1, <vscale x 16 x i1> %pg, <vscale x 16 x i8> zeroinitializer)
  ret <vscale x 16 x i8> %1
}
*** IR Dump After Promote Memory to Register ***
; Function Attrs: nounwind
define <vscale x 16 x i8> @test_svld1_s8(<vscale x 16 x i1> %pg, i8* %base) local_unnamed_addr #0 {
entry:
  %0 = bitcast i8* %base to <vscale x 16 x i8>*
  %1 = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0nxv16i8(<vscale x 16 x i8>* %0, i32 1, <vscale x 16 x i1> %pg, <vscale x 16 x i8> zeroinitializer)
  ret <vscale x 16 x i8> %1
}``

[efriedma]

Hmm, looking more closely, I guess there's technically one run of mem2reg in the standard pass pipeline. But really, I'm surprised that never got killed off; we should not be relying on it.

[c-rhodes]

• Run mem2reg for scalable types from SROA.
• Fix warning warning: inline function 'llvm::Type::getVectorIsScalable' is not defined reported by Harbor by moving implementation to Type.cpp.

[sdesmalen]

llvm/lib/IR/Type.cpp

161

Can this be inlined in llvm/include/llvm/IR/Type.h ?

llvm/lib/Transforms/Scalar/SROA.cpp

4475

If you base your patch on D76748, you can use DL.getTypeAllocSize(AI.getAllocatedType()).isZero().

[c-rhodes]

llvm/lib/IR/Type.cpp

161

Hm I'm not sure, is there a reason other methods like isVectorTy aren't defined with inline in Type.h?

llvm/lib/Transforms/Scalar/SROA.cpp

4475

Ah nice, thanks for pointing that out I'll update this.

[ctetreau]

llvm/include/llvm/IR/Type.h

233

Personally, I think functions like this are a code smell. Why have a type hierarchy at all if we're just going to do everything through the base type?

However, given the fact that the VectorType hierarchy is going to become more complicated soon, I'd really prefer that this function not be added. I'll propose alternatives inline below, but if you want this function, I'd prefer that:

1. be a static function in SROA.cpp.
2. be implemented in terms of isa<VectorType> instead of isVectorTy.

llvm/lib/Transforms/Scalar/SROA.cpp

4474

This can be rewritten:

{
  auto *AT = AI.getAllocatedType();
  if (AI.isArrayAllocation() || !AT->isSized() ||
      (isa<VectorType>(AT) && cast<VectorType>(AT)->isScalable()) ||
      DL.getTypeAllocSize(AT).getFixedSize() == 0)
    return false;
}

AI.getAllocatedType is used 3 times, might as well give it a name. An while isa<VectorType>(AT) && cast<VectorType>(AT)->isScalable() is a little longer than AT.isScalableVectorTy, it's not that bad. on the positive side, it's more explicit as to what it's doing, and it's also less misleading because there's no such thing as a ScalableVectorTy.

4596

if (isa<VectorType>(AI->getAllocatedType()) &&
    cast<VectorType>(AI->getAllocatedType())->isScalable() &&
    isAllocaPromotable(AI))

Same as above.

[c-rhodes]

• Removed isScalableVectorTy and replaced uses in SROA with isa<VectorType> and cast<VectorType>(VecTy)->isScalable().

[efriedma]

llvm/include/llvm/IR/Type.h

233

I think for code that specifically cares about the property "would getTypeAllocSize() return a scalable TypeSize", it's not unreasonable to have a function on Type to query that. The key here being that the code doesn't really care whether the type is a vector.

It might make sense to leave it out for now, though, and try to refactor later when we have a better idea what code ends up doing in practice. I think most code that would want to do that calls getTypeAllocSize() somewhere nearby anyway.

[c-rhodes]

Thanks for the comments @ctetreau! I've updated the patch.

llvm/include/llvm/IR/Type.h

233

Thanks for the suggestion, I've removed this. I don't think it's particularly useful as a function in SROA given it isn't (and is unlikely) to be used a great deal.

[efriedma]

llvm/lib/Transforms/Scalar/SROA.cpp

4598

This looks weird; did you mean to write something like this?

if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
  if (isa<VectorType>(AI->getAllocatedType()) &&
      cast<VectorType>(AI->getAllocatedType())->isScalable()) {
    if (isAllocaPromotable(AI))
      PromotableAllocas.push_back(AI);
  } else {
    Worklist.insert(AI);
  }
}

[c-rhodes]

llvm/lib/Transforms/Scalar/SROA.cpp

4598

Oops, yes I did! Good spot, we don't want allocas with scalable types that aren't promotable added to the worklist as runOnAlloca will blow up. Cheers, I'll fix this.

[c-rhodes]

Fix bug in branch where allocas with scalable types that aren’t promotable could still be added to worklist and subsequently passed to runOnAlloca where asserts would be triggered as this is now fixed-width. Thanks for spotting @efriedma.

[ctetreau]

llvm/include/llvm/IR/Type.h

233

Thanks for the suggestion, I've removed this. I don't think it's particularly useful as a function in SROA given it isn't (and is unlikely) to be used a great deal.

Appreciate it.

I think for code that specifically cares about the property "would getTypeAllocSize() return a scalable TypeSize", it's not unreasonable to have a function on Type to query that. The key here being that the code doesn't really care whether the type is a vector.

I'd feel better about this function, but I'd like to see it become a more common issue before we add it; Type is cluttered enough as it is.

[efriedma]

I'd like to see better test coverage here. At least, a case that can't be promoted, and a case where a scalable alloca lands on the worklist despite the check in SROA::runImpl.

[c-rhodes]

• Add test for unpromotable scalable alloca.
• Rename %pg -> %vec in <vscale x 16 x i8> test.

[c-rhodes]

In D76720#1947128, @efriedma wrote:

I'd like to see better test coverage here. At least, a case that can't be promoted, and a case where a scalable alloca lands on the worklist despite the check in SROA::runImpl.

I've added a test for a scalable alloca that can't be promoted. I'm not really sure how test the latter, unless you're thinking about the bug you spotted, in which case the test I've added should cover that?

We shouldn't hit it but we do skip scalable allocas in runOnAlloca.

[efriedma]

LGTM

llvm/include/llvm/IR/Type.h

public:
   bool isPointerTy() const { return getTypeID() == PointerTyID; }
 
   /// Return true if this is a pointer type or a vector of pointer types.
   bool isPtrOrPtrVectorTy() const { return getScalarType()->isPointerTy(); }
 
   /// True if this is an instance of VectorType.
   bool isVectorTy() const { return getTypeID() == VectorTyID; }
 
+  /// True if this an instance of VectorType with the scalable property set.
+  bool isScalableVectorTy() const;
+
   /// Return true if this type could be converted with a lossless BitCast to
   /// type 'Ty'. For example, i8* to i32*. BitCasts are valid for types of the
   /// same size only where no re-interpretation of the bits is done.
   /// Determine if this type could be losslessly bitcast to Ty
   bool canLosslesslyBitCastTo(Type *Ty) const;
 
   /// Return true if this type is empty, that is, it has no elements or all of
   /// its elements are empty.

llvm/lib/IR/Type.cpp

if (auto *ATy = dyn_cast<ArrayType>(this))
     return ATy->getElementType()->isSized(Visited);
 
   if (auto *VTy = dyn_cast<VectorType>(this))
     return VTy->getElementType()->isSized(Visited);
 
   return cast<StructType>(this)->isSized(Visited);
 }
 
+bool Type::isScalableVectorTy() const {
+  return getTypeID() == VectorTyID && getVectorIsScalable();
+}
+
 //===----------------------------------------------------------------------===//
 //                          Primitive 'Type' data
 //===----------------------------------------------------------------------===//
 
 Type *Type::getVoidTy(LLVMContext &C) { return &C.pImpl->VoidTy; }
 Type *Type::getLabelTy(LLVMContext &C) { return &C.pImpl->LabelTy; }
 Type *Type::getHalfTy(LLVMContext &C) { return &C.pImpl->HalfTy; }
 Type *Type::getFloatTy(LLVMContext &C) { return &C.pImpl->FloatTy; }

llvm/lib/Transforms/Scalar/SROA.cpp

class AllocaSlices::SliceBuilder : public PtrUseVisitor<SliceBuilder> {
   SmallDenseMap<Instruction *, uint64_t> PHIOrSelectSizes;
 
   /// Set to de-duplicate dead instructions found in the use walk.
   SmallPtrSet<Instruction *, 4> VisitedDeadInsts;
 
 public:
   SliceBuilder(const DataLayout &DL, AllocaInst &AI, AllocaSlices &AS)
       : PtrUseVisitor<SliceBuilder>(DL),
-        AllocSize(DL.getTypeAllocSize(AI.getAllocatedType())), AS(AS) {}
+        AllocSize(DL.getTypeAllocSize(AI.getAllocatedType()).getFixedSize()),
+        AS(AS) {}
 
 private:
   void markAsDead(Instruction &I) {
     if (VisitedDeadInsts.insert(&I).second)
       AS.DeadUsers.push_back(&I);
   }
 
   void insertUse(Instruction &I, const APInt &Offset, uint64_t Size,
if (SROAStrictInbounds && GEPI.isInBounds()) {
           unsigned ElementIdx = OpC->getZExtValue();
           const StructLayout *SL = DL.getStructLayout(STy);
           GEPOffset +=
               APInt(Offset.getBitWidth(), SL->getElementOffset(ElementIdx));
         } else {
           // For array or vector indices, scale the index by the size of the
           // type.
           APInt Index = OpC->getValue().sextOrTrunc(Offset.getBitWidth());
-          GEPOffset += Index * APInt(Offset.getBitWidth(),
-                                     DL.getTypeAllocSize(GTI.getIndexedType()));
+          GEPOffset +=
+              Index *
+              APInt(Offset.getBitWidth(),
+                    DL.getTypeAllocSize(GTI.getIndexedType()).getFixedSize());
         }
 
         // If this index has computed an intermediate pointer which is not
         // inbounds, then the result of the GEP is a poison value and we can
         // delete it and all uses.
         if (GEPOffset.ugt(AllocSize))
           return markAsDead(GEPI);
       }
void visitLoadInst(LoadInst &LI) {
 
     if (!IsOffsetKnown)
       return PI.setAborted(&LI);
 
     if (LI.isVolatile() &&
         LI.getPointerAddressSpace() != DL.getAllocaAddrSpace())
       return PI.setAborted(&LI);
 
-    uint64_t Size = DL.getTypeStoreSize(LI.getType());
+    uint64_t Size = DL.getTypeStoreSize(LI.getType()).getFixedSize();
     return handleLoadOrStore(LI.getType(), LI, Offset, Size, LI.isVolatile());
   }
 
   void visitStoreInst(StoreInst &SI) {
     Value *ValOp = SI.getValueOperand();
     if (ValOp == *U)
       return PI.setEscapedAndAborted(&SI);
     if (!IsOffsetKnown)
       return PI.setAborted(&SI);
 
     if (SI.isVolatile() &&
         SI.getPointerAddressSpace() != DL.getAllocaAddrSpace())
       return PI.setAborted(&SI);
 
-    uint64_t Size = DL.getTypeStoreSize(ValOp->getType());
+    uint64_t Size = DL.getTypeStoreSize(ValOp->getType()).getFixedSize();
 
     // If this memory access can be shown to *statically* extend outside the
     // bounds of the allocation, it's behavior is undefined, so simply
     // ignore it. Note that this is more strict than the generic clamping
     // behavior of insertUse. We also try to handle cases which might run the
     // risk of overflow.
     // FIXME: We should instead consider the pointer to have escaped if this
     // function is being instrumented for addressing bugs or race conditions.
if (LI->getParent() != BB)
       return false;
 
     // Ensure that there are no instructions between the PHI and the load that
     // could store.
     for (BasicBlock::iterator BBI(PN); &*BBI != LI; ++BBI)
       if (BBI->mayWriteToMemory())
         return false;
 
-    uint64_t Size = DL.getTypeStoreSize(LI->getType());
+    uint64_t Size = DL.getTypeStoreSize(LI->getType()).getFixedSize();
     MaxAlign = std::max(MaxAlign, MaybeAlign(LI->getAlignment()));
     MaxSize = MaxSize.ult(Size) ? APInt(APWidth, Size) : MaxSize;
     HaveLoad = true;
   }
 
   if (!HaveLoad)
     return false;
 
static Value *getNaturalGEPRecursively(IRBuilderTy &IRB, const DataLayout &DL,
   // We can't recurse through pointer types.
   if (Ty->isPointerTy())
     return nullptr;
 
   // We try to analyze GEPs over vectors here, but note that these GEPs are
   // extremely poorly defined currently. The long-term goal is to remove GEPing
   // over a vector from the IR completely.
   if (VectorType *VecTy = dyn_cast<VectorType>(Ty)) {
-    unsigned ElementSizeInBits = DL.getTypeSizeInBits(VecTy->getScalarType());
+    unsigned ElementSizeInBits =
+        DL.getTypeSizeInBits(VecTy->getScalarType()).getFixedSize();
     if (ElementSizeInBits % 8 != 0) {
       // GEPs over non-multiple of 8 size vector elements are invalid.
       return nullptr;
     }
     APInt ElementSize(Offset.getBitWidth(), ElementSizeInBits / 8);
     APInt NumSkippedElements = Offset.sdiv(ElementSize);
     if (NumSkippedElements.ugt(VecTy->getNumElements()))
       return nullptr;
     Offset -= NumSkippedElements * ElementSize;
     Indices.push_back(IRB.getInt(NumSkippedElements));
     return getNaturalGEPRecursively(IRB, DL, Ptr, VecTy->getElementType(),
                                     Offset, TargetTy, Indices, NamePrefix);
   }
 
   if (ArrayType *ArrTy = dyn_cast<ArrayType>(Ty)) {
     Type *ElementTy = ArrTy->getElementType();
-    APInt ElementSize(Offset.getBitWidth(), DL.getTypeAllocSize(ElementTy));
+    APInt ElementSize(Offset.getBitWidth(),
+                      DL.getTypeAllocSize(ElementTy).getFixedSize());
     APInt NumSkippedElements = Offset.sdiv(ElementSize);
     if (NumSkippedElements.ugt(ArrTy->getNumElements()))
       return nullptr;
 
     Offset -= NumSkippedElements * ElementSize;
     Indices.push_back(IRB.getInt(NumSkippedElements));
     return getNaturalGEPRecursively(IRB, DL, Ptr, ElementTy, Offset, TargetTy,
                                     Indices, NamePrefix);
   }
 
   StructType *STy = dyn_cast<StructType>(Ty);
   if (!STy)
     return nullptr;
 
   const StructLayout *SL = DL.getStructLayout(STy);
   uint64_t StructOffset = Offset.getZExtValue();
   if (StructOffset >= SL->getSizeInBytes())
     return nullptr;
   unsigned Index = SL->getElementContainingOffset(StructOffset);
   Offset -= APInt(Offset.getBitWidth(), SL->getElementOffset(Index));
   Type *ElementTy = STy->getElementType(Index);
-  if (Offset.uge(DL.getTypeAllocSize(ElementTy)))
+  if (Offset.uge(DL.getTypeAllocSize(ElementTy).getFixedSize()))
     return nullptr; // The offset points into alignment padding.
 
   Indices.push_back(IRB.getInt32(Index));
   return getNaturalGEPRecursively(IRB, DL, Ptr, ElementTy, Offset, TargetTy,
                                   Indices, NamePrefix);
 }
 
 /// Get a natural GEP from a base pointer to a particular offset and
static Value *getNaturalGEPWithOffset(IRBuilderTy &IRB, const DataLayout &DL,
   // Don't consider any GEPs through an i8* as natural unless the TargetTy is
   // an i8.
   if (Ty == IRB.getInt8PtrTy(Ty->getAddressSpace()) && TargetTy->isIntegerTy(8))
     return nullptr;
 
   Type *ElementTy = Ty->getElementType();
   if (!ElementTy->isSized())
     return nullptr; // We can't GEP through an unsized element.
-  APInt ElementSize(Offset.getBitWidth(), DL.getTypeAllocSize(ElementTy));
+  APInt ElementSize(Offset.getBitWidth(),
+                    DL.getTypeAllocSize(ElementTy).getFixedSize());
   if (ElementSize == 0)
     return nullptr; // Zero-length arrays can't help us build a natural GEP.
   APInt NumSkippedElements = Offset.sdiv(ElementSize);
 
   Offset -= NumSkippedElements * ElementSize;
   Indices.push_back(IRB.getInt(NumSkippedElements));
   return getNaturalGEPRecursively(IRB, DL, Ptr, ElementTy, Offset, TargetTy,
                                   Indices, NamePrefix);
static bool canConvertValue(const DataLayout &DL, Type *OldTy, Type *NewTy) {
   // issues when in conjunction with loads and stores.
   if (isa<IntegerType>(OldTy) && isa<IntegerType>(NewTy)) {
     assert(cast<IntegerType>(OldTy)->getBitWidth() !=
                cast<IntegerType>(NewTy)->getBitWidth() &&
            "We can't have the same bitwidth for different int types");
     return false;
   }
 
-  if (DL.getTypeSizeInBits(NewTy) != DL.getTypeSizeInBits(OldTy))
+  if (DL.getTypeSizeInBits(NewTy).getFixedSize() !=
+      DL.getTypeSizeInBits(OldTy).getFixedSize())
     return false;
   if (!NewTy->isSingleValueType() || !OldTy->isSingleValueType())
     return false;
 
   // We can convert pointers to integers and vice-versa. Same for vectors
   // of pointers and integers.
   OldTy = OldTy->getScalarType();
   NewTy = NewTy->getScalarType();
static VectorType *isVectorPromotionViable(Partition &P, const DataLayout &DL) {
   SmallVector<VectorType *, 4> CandidateTys;
   Type *CommonEltTy = nullptr;
   bool HaveCommonEltTy = true;
   auto CheckCandidateType = [&](Type *Ty) {
     if (auto *VTy = dyn_cast<VectorType>(Ty)) {
       // Return if bitcast to vectors is different for total size in bits.
       if (!CandidateTys.empty()) {
         VectorType *V = CandidateTys[0];
-        if (DL.getTypeSizeInBits(VTy) != DL.getTypeSizeInBits(V)) {
+        if (DL.getTypeSizeInBits(VTy).getFixedSize() !=
+            DL.getTypeSizeInBits(V).getFixedSize()) {
           CandidateTys.clear();
           return;
         }
       }
       CandidateTys.push_back(VTy);
       if (!CommonEltTy)
         CommonEltTy = VTy->getElementType();
       else if (CommonEltTy != VTy->getElementType())
if (!HaveCommonEltTy) {
     // If there were no integer vector types, give up.
     if (CandidateTys.empty())
       return nullptr;
 
     // Rank the remaining candidate vector types. This is easy because we know
     // they're all integer vectors. We sort by ascending number of elements.
     auto RankVectorTypes = [&DL](VectorType *RHSTy, VectorType *LHSTy) {
       (void)DL;
-      assert(DL.getTypeSizeInBits(RHSTy) == DL.getTypeSizeInBits(LHSTy) &&
+      assert(DL.getTypeSizeInBits(RHSTy).getFixedSize() ==
+                 DL.getTypeSizeInBits(LHSTy).getFixedSize() &&
              "Cannot have vector types of different sizes!");
       assert(RHSTy->getElementType()->isIntegerTy() &&
              "All non-integer types eliminated!");
       assert(LHSTy->getElementType()->isIntegerTy() &&
              "All non-integer types eliminated!");
       return RHSTy->getNumElements() < LHSTy->getNumElements();
     };
     llvm::sort(CandidateTys, RankVectorTypes);
for (VectorType *VTy : CandidateTys) {
              "Different vector types with the same element type!");
     }
 #endif
     CandidateTys.resize(1);
   }
 
   // Try each vector type, and return the one which works.
   auto CheckVectorTypeForPromotion = [&](VectorType *VTy) {
-    uint64_t ElementSize = DL.getTypeSizeInBits(VTy->getElementType());
+    uint64_t ElementSize =
+        DL.getTypeSizeInBits(VTy->getElementType()).getFixedSize();
 
     // While the definition of LLVM vectors is bitpacked, we don't support sizes
     // that aren't byte sized.
     if (ElementSize % 8)
       return false;
-    assert((DL.getTypeSizeInBits(VTy) % 8) == 0 &&
+    assert((DL.getTypeSizeInBits(VTy).getFixedSize() % 8) == 0 &&
            "vector size not a multiple of element size?");
     ElementSize /= 8;
 
     for (const Slice &S : P)
       if (!isVectorPromotionViableForSlice(P, S, VTy, ElementSize, DL))
         return false;
 
     for (const Slice *S : P.splitSliceTails())

 ///
 /// This implements the necessary checking for the \c isIntegerWideningViable
 /// test below on a single slice of the alloca.
 static bool isIntegerWideningViableForSlice(const Slice &S,
                                             uint64_t AllocBeginOffset,
                                             Type *AllocaTy,
                                             const DataLayout &DL,
                                             bool &WholeAllocaOp) {
-  uint64_t Size = DL.getTypeStoreSize(AllocaTy);
+  uint64_t Size = DL.getTypeStoreSize(AllocaTy).getFixedSize();
 
   uint64_t RelBegin = S.beginOffset() - AllocBeginOffset;
   uint64_t RelEnd = S.endOffset() - AllocBeginOffset;
 
   // We can't reasonably handle cases where the load or store extends past
   // the end of the alloca's type and into its padding.
   if (RelEnd > Size)
     return false;
 
   Use *U = S.getUse();
 
   if (LoadInst *LI = dyn_cast<LoadInst>(U->getUser())) {
     if (LI->isVolatile())
       return false;
     // We can't handle loads that extend past the allocated memory.
-    if (DL.getTypeStoreSize(LI->getType()) > Size)
+    if (DL.getTypeStoreSize(LI->getType()).getFixedSize() > Size)
       return false;
     // So far, AllocaSliceRewriter does not support widening split slice tails
     // in rewriteIntegerLoad.
     if (S.beginOffset() < AllocBeginOffset)
       return false;
     // Note that we don't count vector loads or stores as whole-alloca
     // operations which enable integer widening because we would prefer to use
     // vector widening instead.
     if (!isa<VectorType>(LI->getType()) && RelBegin == 0 && RelEnd == Size)
       WholeAllocaOp = true;
     if (IntegerType *ITy = dyn_cast<IntegerType>(LI->getType())) {
-      if (ITy->getBitWidth() < DL.getTypeStoreSizeInBits(ITy))
+      if (ITy->getBitWidth() < DL.getTypeStoreSizeInBits(ITy).getFixedSize())
         return false;
     } else if (RelBegin != 0 || RelEnd != Size ||
                !canConvertValue(DL, AllocaTy, LI->getType())) {
       // Non-integer loads need to be convertible from the alloca type so that
       // they are promotable.
       return false;
     }
   } else if (StoreInst *SI = dyn_cast<StoreInst>(U->getUser())) {
     Type *ValueTy = SI->getValueOperand()->getType();
     if (SI->isVolatile())
       return false;
     // We can't handle stores that extend past the allocated memory.
-    if (DL.getTypeStoreSize(ValueTy) > Size)
+    if (DL.getTypeStoreSize(ValueTy).getFixedSize() > Size)
       return false;
     // So far, AllocaSliceRewriter does not support widening split slice tails
     // in rewriteIntegerStore.
     if (S.beginOffset() < AllocBeginOffset)
       return false;
     // Note that we don't count vector loads or stores as whole-alloca
     // operations which enable integer widening because we would prefer to use
     // vector widening instead.
     if (!isa<VectorType>(ValueTy) && RelBegin == 0 && RelEnd == Size)
       WholeAllocaOp = true;
     if (IntegerType *ITy = dyn_cast<IntegerType>(ValueTy)) {
-      if (ITy->getBitWidth() < DL.getTypeStoreSizeInBits(ITy))
+      if (ITy->getBitWidth() < DL.getTypeStoreSizeInBits(ITy).getFixedSize())
         return false;
     } else if (RelBegin != 0 || RelEnd != Size ||
                !canConvertValue(DL, ValueTy, AllocaTy)) {
       // Non-integer stores need to be convertible to the alloca type so that
       // they are promotable.
       return false;
     }
   } else if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(U->getUser())) {

 /// Test whether the given alloca partition's integer operations can be
 /// widened to promotable ones.
 ///
 /// This is a quick test to check whether we can rewrite the integer loads and
 /// stores to a particular alloca into wider loads and stores and be able to
 /// promote the resulting alloca.
 static bool isIntegerWideningViable(Partition &P, Type *AllocaTy,
                                     const DataLayout &DL) {
-  uint64_t SizeInBits = DL.getTypeSizeInBits(AllocaTy);
+  uint64_t SizeInBits = DL.getTypeSizeInBits(AllocaTy).getFixedSize();
   // Don't create integer types larger than the maximum bitwidth.
   if (SizeInBits > IntegerType::MAX_INT_BITS)
     return false;
 
   // Don't try to handle allocas with bit-padding.
-  if (SizeInBits != DL.getTypeStoreSizeInBits(AllocaTy))
+  if (SizeInBits != DL.getTypeStoreSizeInBits(AllocaTy).getFixedSize())
     return false;
 
   // We need to ensure that an integer type with the appropriate bitwidth can
   // be converted to the alloca type, whatever that is. We don't want to force
   // the alloca itself to have an integer type if there is a more suitable one.
   Type *IntTy = Type::getIntNTy(AllocaTy->getContext(), SizeInBits);
   if (!canConvertValue(DL, AllocaTy, IntTy) ||
       !canConvertValue(DL, IntTy, AllocaTy))
static bool isIntegerWideningViable(Partition &P, Type *AllocaTy,
   return WholeAllocaOp;
 }
 
 static Value *extractInteger(const DataLayout &DL, IRBuilderTy &IRB, Value *V,
                              IntegerType *Ty, uint64_t Offset,
                              const Twine &Name) {
   LLVM_DEBUG(dbgs() << "       start: " << *V << "\n");
   IntegerType *IntTy = cast<IntegerType>(V->getType());
-  assert(DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) &&
+  assert(DL.getTypeStoreSize(Ty).getFixedSize() + Offset <=
+             DL.getTypeStoreSize(IntTy).getFixedSize() &&
          "Element extends past full value");
   uint64_t ShAmt = 8 * Offset;
   if (DL.isBigEndian())
-    ShAmt = 8 * (DL.getTypeStoreSize(IntTy) - DL.getTypeStoreSize(Ty) - Offset);
+    ShAmt = 8 * (DL.getTypeStoreSize(IntTy).getFixedSize() -
+                 DL.getTypeStoreSize(Ty).getFixedSize() - Offset);
   if (ShAmt) {
     V = IRB.CreateLShr(V, ShAmt, Name + ".shift");
     LLVM_DEBUG(dbgs() << "     shifted: " << *V << "\n");
   }
   assert(Ty->getBitWidth() <= IntTy->getBitWidth() &&
          "Cannot extract to a larger integer!");
   if (Ty != IntTy) {
     V = IRB.CreateTrunc(V, Ty, Name + ".trunc");
     LLVM_DEBUG(dbgs() << "     trunced: " << *V << "\n");
   }
   return V;
 }
 
 static Value *insertInteger(const DataLayout &DL, IRBuilderTy &IRB, Value *Old,
                             Value *V, uint64_t Offset, const Twine &Name) {
   IntegerType *IntTy = cast<IntegerType>(Old->getType());
   IntegerType *Ty = cast<IntegerType>(V->getType());
   assert(Ty->getBitWidth() <= IntTy->getBitWidth() &&
          "Cannot insert a larger integer!");
   LLVM_DEBUG(dbgs() << "       start: " << *V << "\n");
   if (Ty != IntTy) {
     V = IRB.CreateZExt(V, IntTy, Name + ".ext");
     LLVM_DEBUG(dbgs() << "    extended: " << *V << "\n");
   }
-  assert(DL.getTypeStoreSize(Ty) + Offset <= DL.getTypeStoreSize(IntTy) &&
+  assert(DL.getTypeStoreSize(Ty).getFixedSize() + Offset <=
+             DL.getTypeStoreSize(IntTy).getFixedSize() &&
          "Element store outside of alloca store");
   uint64_t ShAmt = 8 * Offset;
   if (DL.isBigEndian())
-    ShAmt = 8 * (DL.getTypeStoreSize(IntTy) - DL.getTypeStoreSize(Ty) - Offset);
+    ShAmt = 8 * (DL.getTypeStoreSize(IntTy).getFixedSize() -
+                 DL.getTypeStoreSize(Ty).getFixedSize() - Offset);
   if (ShAmt) {
     V = IRB.CreateShl(V, ShAmt, Name + ".shift");
     LLVM_DEBUG(dbgs() << "     shifted: " << *V << "\n");
   }
 
   if (ShAmt || Ty->getBitWidth() < IntTy->getBitWidth()) {
     APInt Mask = ~Ty->getMask().zext(IntTy->getBitWidth()).shl(ShAmt);
     Old = IRB.CreateAnd(Old, Mask, Name + ".mask");
AllocaSliceRewriter(const DataLayout &DL, AllocaSlices &AS, SROA &Pass,
                       uint64_t NewAllocaEndOffset, bool IsIntegerPromotable,
                       VectorType *PromotableVecTy,
                       SmallSetVector<PHINode *, 8> &PHIUsers,
                       SmallSetVector<SelectInst *, 8> &SelectUsers)
       : DL(DL), AS(AS), Pass(Pass), OldAI(OldAI), NewAI(NewAI),
         NewAllocaBeginOffset(NewAllocaBeginOffset),
         NewAllocaEndOffset(NewAllocaEndOffset),
         NewAllocaTy(NewAI.getAllocatedType()),
-        IntTy(IsIntegerPromotable
-                  ? Type::getIntNTy(
-                        NewAI.getContext(),
-                        DL.getTypeSizeInBits(NewAI.getAllocatedType()))
-                  : nullptr),
+        IntTy(
+            IsIntegerPromotable
+                ? Type::getIntNTy(NewAI.getContext(),
+                                  DL.getTypeSizeInBits(NewAI.getAllocatedType())
+                                      .getFixedSize())
         VecTy(PromotableVecTy),
         ElementTy(VecTy ? VecTy->getElementType() : nullptr),
-        ElementSize(VecTy ? DL.getTypeSizeInBits(ElementTy) / 8 : 0),
+        ElementSize(VecTy ? DL.getTypeSizeInBits(ElementTy).getFixedSize() / 8
+                          : 0),
         PHIUsers(PHIUsers), SelectUsers(SelectUsers),
         IRB(NewAI.getContext(), ConstantFolder()) {
     if (VecTy) {
-      assert((DL.getTypeSizeInBits(ElementTy) % 8) == 0 &&
+      assert((DL.getTypeSizeInBits(ElementTy).getFixedSize() % 8) == 0 &&
              "Only multiple-of-8 sized vector elements are viable");
       ++NumVectorized;
     }
     assert((!IntTy && !VecTy) || (IntTy && !VecTy) || (!IntTy && VecTy));
   }
 
   bool visit(AllocaSlices::const_iterator I) {
     bool CanSROA = true;
bool visitLoadInst(LoadInst &LI) {
 
     AAMDNodes AATags;
     LI.getAAMetadata(AATags);
 
     unsigned AS = LI.getPointerAddressSpace();
 
     Type *TargetTy = IsSplit ? Type::getIntNTy(LI.getContext(), SliceSize * 8)
                              : LI.getType();
-    const bool IsLoadPastEnd = DL.getTypeStoreSize(TargetTy) > SliceSize;
+    const bool IsLoadPastEnd =
+        DL.getTypeStoreSize(TargetTy).getFixedSize() > SliceSize;
     bool IsPtrAdjusted = false;
     Value *V;
     if (VecTy) {
       V = rewriteVectorizedLoadInst();
     } else if (IntTy && LI.getType()->isIntegerTy()) {
       V = rewriteIntegerLoad(LI);
     } else if (NewBeginOffset == NewAllocaBeginOffset &&
                NewEndOffset == NewAllocaEndOffset &&
if (VecTy) {
       IsPtrAdjusted = true;
     }
     V = convertValue(DL, IRB, V, TargetTy);
 
     if (IsSplit) {
       assert(!LI.isVolatile());
       assert(LI.getType()->isIntegerTy() &&
              "Only integer type loads and stores are split");
-      assert(SliceSize < DL.getTypeStoreSize(LI.getType()) &&
+      assert(SliceSize < DL.getTypeStoreSize(LI.getType()).getFixedSize() &&
              "Split load isn't smaller than original load");
       assert(DL.typeSizeEqualsStoreSize(LI.getType()) &&
              "Non-byte-multiple bit width");
       // Move the insertion point just past the load so that we can refer to it.
       IRB.SetInsertPoint(&*std::next(BasicBlock::iterator(&LI)));
       // Create a placeholder value with the same type as LI to use as the
       // basis for the new value. This allows us to replace the uses of LI with
       // the computed value, and then replace the placeholder with LI, leaving
bool rewriteVectorizedStoreInst(Value *V, StoreInst &SI, Value *OldOp,
 
     LLVM_DEBUG(dbgs() << "          to: " << *Store << "\n");
     return true;
   }
 
   bool rewriteIntegerStore(Value *V, StoreInst &SI, AAMDNodes AATags) {
     assert(IntTy && "We cannot extract an integer from the alloca");
     assert(!SI.isVolatile());
-    if (DL.getTypeSizeInBits(V->getType()) != IntTy->getBitWidth()) {
+    if (DL.getTypeSizeInBits(V->getType()).getFixedSize() !=
+        IntTy->getBitWidth()) {
       Value *Old = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
                                          NewAI.getAlign(), "oldload");
       Old = convertValue(DL, IRB, Old, IntTy);
       assert(BeginOffset >= NewAllocaBeginOffset && "Out of bounds offset");
       uint64_t Offset = BeginOffset - NewAllocaBeginOffset;
       V = insertInteger(DL, IRB, Old, SI.getValueOperand(), Offset, "insert");
     }
     V = convertValue(DL, IRB, V, NewAllocaTy);
bool visitStoreInst(StoreInst &SI) {
     Value *V = SI.getValueOperand();
 
     // Strip all inbounds GEPs and pointer casts to try to dig out any root
     // alloca that should be re-examined after promoting this alloca.
     if (V->getType()->isPointerTy())
       if (AllocaInst *AI = dyn_cast<AllocaInst>(V->stripInBoundsOffsets()))
         Pass.PostPromotionWorklist.insert(AI);
 
-    if (SliceSize < DL.getTypeStoreSize(V->getType())) {
+    if (SliceSize < DL.getTypeStoreSize(V->getType()).getFixedSize()) {
       assert(!SI.isVolatile());
       assert(V->getType()->isIntegerTy() &&
              "Only integer type loads and stores are split");
       assert(DL.typeSizeEqualsStoreSize(V->getType()) &&
              "Non-byte-multiple bit width");
       IntegerType *NarrowTy = Type::getIntNTy(SI.getContext(), SliceSize * 8);
       V = extractInteger(DL, IRB, V, NarrowTy, NewBeginOffset - BeginOffset,
                          "extract");
     }
 
     if (VecTy)
       return rewriteVectorizedStoreInst(V, SI, OldOp, AATags);
     if (IntTy && V->getType()->isIntegerTy())
       return rewriteIntegerStore(V, SI, AATags);
 
-    const bool IsStorePastEnd = DL.getTypeStoreSize(V->getType()) > SliceSize;
+    const bool IsStorePastEnd =
+        DL.getTypeStoreSize(V->getType()).getFixedSize() > SliceSize;
     StoreInst *NewSI;
     if (NewBeginOffset == NewAllocaBeginOffset &&
         NewEndOffset == NewAllocaEndOffset &&
         (canConvertValue(DL, V->getType(), NewAllocaTy) ||
          (IsStorePastEnd && NewAllocaTy->isIntegerTy() &&
           V->getType()->isIntegerTy()))) {
       // If this is an integer store past the end of slice (and thus the bytes
       // past that point are irrelevant or this is unreachable), truncate the
const bool CanContinue = [&]() {
         return false;
       auto *C = cast<ConstantInt>(II.getLength());
       if (C->getBitWidth() > 64)
         return false;
       const auto Len = C->getZExtValue();
       auto *Int8Ty = IntegerType::getInt8Ty(NewAI.getContext());
       auto *SrcTy = VectorType::get(Int8Ty, Len);
       return canConvertValue(DL, SrcTy, AllocaTy) &&
-        DL.isLegalInteger(DL.getTypeSizeInBits(ScalarTy));
+             DL.isLegalInteger(DL.getTypeSizeInBits(ScalarTy).getFixedSize());
     }();
 
     // If this doesn't map cleanly onto the alloca type, and that type isn't
     // a single value type, just emit a memset.
     if (!CanContinue) {
       Type *SizeTy = II.getLength()->getType();
       Constant *Size = ConstantInt::get(SizeTy, NewEndOffset - NewBeginOffset);
       CallInst *New = IRB.CreateMemSet(
if (VecTy) {
       assert(ElementTy == ScalarTy);
 
       unsigned BeginIndex = getIndex(NewBeginOffset);
       unsigned EndIndex = getIndex(NewEndOffset);
       assert(EndIndex > BeginIndex && "Empty vector!");
       unsigned NumElements = EndIndex - BeginIndex;
       assert(NumElements <= VecTy->getNumElements() && "Too many elements!");
 
-      Value *Splat =
-          getIntegerSplat(II.getValue(), DL.getTypeSizeInBits(ElementTy) / 8);
+      Value *Splat = getIntegerSplat(
+          II.getValue(), DL.getTypeSizeInBits(ElementTy).getFixedSize() / 8);
       Splat = convertValue(DL, IRB, Splat, ElementTy);
       if (NumElements > 1)
         Splat = getVectorSplat(Splat, NumElements);
 
       Value *Old = IRB.CreateAlignedLoad(NewAI.getAllocatedType(), &NewAI,
                                          NewAI.getAlign(), "oldload");
       V = insertVector(IRB, Old, Splat, BeginIndex, "vec");
     } else if (IntTy) {
if (VecTy) {
                "Wrong type for an alloca wide integer!");
       }
       V = convertValue(DL, IRB, V, AllocaTy);
     } else {
       // Established these invariants above.
       assert(NewBeginOffset == NewAllocaBeginOffset);
       assert(NewEndOffset == NewAllocaEndOffset);
 
-      V = getIntegerSplat(II.getValue(), DL.getTypeSizeInBits(ScalarTy) / 8);
+      V = getIntegerSplat(II.getValue(),
+                          DL.getTypeSizeInBits(ScalarTy).getFixedSize() / 8);
       if (VectorType *AllocaVecTy = dyn_cast<VectorType>(AllocaTy))
         V = getVectorSplat(V, AllocaVecTy->getNumElements());
 
       V = convertValue(DL, IRB, V, AllocaTy);
     }
 
     StoreInst *New =
         IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlign(), II.isVolatile());
bool visitMemTransferInst(MemTransferInst &II) {
     // memmove with memcpy, and we don't need to worry about all manner of
     // downsides to splitting and transforming the operations.
 
     // If this doesn't map cleanly onto the alloca type, and that type isn't
     // a single value type, just emit a memcpy.
     bool EmitMemCpy =
         !VecTy && !IntTy &&
         (BeginOffset > NewAllocaBeginOffset || EndOffset < NewAllocaEndOffset ||
-         SliceSize != DL.getTypeStoreSize(NewAI.getAllocatedType()) ||
+         SliceSize !=
+             DL.getTypeStoreSize(NewAI.getAllocatedType()).getFixedSize() ||
          !NewAI.getAllocatedType()->isSingleValueType());
 
     // If we're just going to emit a memcpy, the alloca hasn't changed, and the
     // size hasn't been shrunk based on analysis of the viable range, this is
     // a no-op.
     if (EmitMemCpy && &OldAI == &NewAI) {
       // Ensure the start lines up.
       assert(NewBeginOffset == BeginOffset);

 ///
 /// This removes no-op aggregate types wrapping an underlying type. It will
 /// strip as many layers of types as it can without changing either the type
 /// size or the allocated size.
 static Type *stripAggregateTypeWrapping(const DataLayout &DL, Type *Ty) {
   if (Ty->isSingleValueType())
     return Ty;
 
-  uint64_t AllocSize = DL.getTypeAllocSize(Ty);
-  uint64_t TypeSize = DL.getTypeSizeInBits(Ty);
+  uint64_t AllocSize = DL.getTypeAllocSize(Ty).getFixedSize();
+  uint64_t TypeSize = DL.getTypeSizeInBits(Ty).getFixedSize();
 
   Type *InnerTy;
   if (ArrayType *ArrTy = dyn_cast<ArrayType>(Ty)) {
     InnerTy = ArrTy->getElementType();
   } else if (StructType *STy = dyn_cast<StructType>(Ty)) {
     const StructLayout *SL = DL.getStructLayout(STy);
     unsigned Index = SL->getElementContainingOffset(0);
     InnerTy = STy->getElementType(Index);
   } else {
     return Ty;
   }
 
-  if (AllocSize > DL.getTypeAllocSize(InnerTy) ||
-      TypeSize > DL.getTypeSizeInBits(InnerTy))
+  if (AllocSize > DL.getTypeAllocSize(InnerTy).getFixedSize() ||
+      TypeSize > DL.getTypeSizeInBits(InnerTy).getFixedSize())
     return Ty;
 
   return stripAggregateTypeWrapping(DL, InnerTy);
 }
 
 /// Try to find a partition of the aggregate type passed in for a given
 /// offset and size.
 ///
 /// This recurses through the aggregate type and tries to compute a subtype
 /// based on the offset and size. When the offset and size span a sub-section
 /// of an array, it will even compute a new array type for that sub-section,
 /// and the same for structs.
 ///
 /// Note that this routine is very strict and tries to find a partition of the
 /// type which produces the *exact* right offset and size. It is not forgiving
 /// when the size or offset cause either end of type-based partition to be off.
 /// Also, this is a best-effort routine. It is reasonable to give up and not
 /// return a type if necessary.
 static Type *getTypePartition(const DataLayout &DL, Type *Ty, uint64_t Offset,
                               uint64_t Size) {
-  if (Offset == 0 && DL.getTypeAllocSize(Ty) == Size)
+  if (Offset == 0 && DL.getTypeAllocSize(Ty).getFixedSize() == Size)
     return stripAggregateTypeWrapping(DL, Ty);
-  if (Offset > DL.getTypeAllocSize(Ty) ||
-      (DL.getTypeAllocSize(Ty) - Offset) < Size)
+  if (Offset > DL.getTypeAllocSize(Ty).getFixedSize() ||
+      (DL.getTypeAllocSize(Ty).getFixedSize() - Offset) < Size)
     return nullptr;
 
   if (SequentialType *SeqTy = dyn_cast<SequentialType>(Ty)) {
     Type *ElementTy = SeqTy->getElementType();
-    uint64_t ElementSize = DL.getTypeAllocSize(ElementTy);
+    uint64_t ElementSize = DL.getTypeAllocSize(ElementTy).getFixedSize();
     uint64_t NumSkippedElements = Offset / ElementSize;
     if (NumSkippedElements >= SeqTy->getNumElements())
       return nullptr;
     Offset -= NumSkippedElements * ElementSize;
 
     // First check if we need to recurse.
     if (Offset > 0 || Size < ElementSize) {
       // Bail if the partition ends in a different array element.
static Type *getTypePartition(const DataLayout &DL, Type *Ty, uint64_t Offset,
   uint64_t EndOffset = Offset + Size;
   if (EndOffset > SL->getSizeInBytes())
     return nullptr;
 
   unsigned Index = SL->getElementContainingOffset(Offset);
   Offset -= SL->getElementOffset(Index);
 
   Type *ElementTy = STy->getElementType(Index);
-  uint64_t ElementSize = DL.getTypeAllocSize(ElementTy);
+  uint64_t ElementSize = DL.getTypeAllocSize(ElementTy).getFixedSize();
   if (Offset >= ElementSize)
     return nullptr; // The offset points into alignment padding.
 
   // See if any partition must be contained by the element.
   if (Offset > 0 || Size < ElementSize) {
     if ((Offset + Size) > ElementSize)
       return nullptr;
     return getTypePartition(DL, ElementTy, Offset, Size);

 AllocaInst *SROA::rewritePartition(AllocaInst &AI, AllocaSlices &AS,
                                    Partition &P) {
   // Try to compute a friendly type for this partition of the alloca. This
   // won't always succeed, in which case we fall back to a legal integer type
   // or an i8 array of an appropriate size.
   Type *SliceTy = nullptr;
   const DataLayout &DL = AI.getModule()->getDataLayout();
   if (Type *CommonUseTy = findCommonType(P.begin(), P.end(), P.endOffset()))
-    if (DL.getTypeAllocSize(CommonUseTy) >= P.size())
+    if (DL.getTypeAllocSize(CommonUseTy).getFixedSize() >= P.size())
       SliceTy = CommonUseTy;
   if (!SliceTy)
     if (Type *TypePartitionTy = getTypePartition(DL, AI.getAllocatedType(),
                                                  P.beginOffset(), P.size()))
       SliceTy = TypePartitionTy;
   if ((!SliceTy || (SliceTy->isArrayTy() &&
                     SliceTy->getArrayElementType()->isIntegerTy())) &&
       DL.isLegalInteger(P.size() * 8))
     SliceTy = Type::getIntNTy(*C, P.size() * 8);
   if (!SliceTy)
     SliceTy = ArrayType::get(Type::getInt8Ty(*C), P.size());
-  assert(DL.getTypeAllocSize(SliceTy) >= P.size());
+  assert(DL.getTypeAllocSize(SliceTy).getFixedSize() >= P.size());
 
   bool IsIntegerPromotable = isIntegerWideningViable(P, SliceTy, DL);
 
   VectorType *VecTy =
       IsIntegerPromotable ? nullptr : isVectorPromotionViable(P, DL);
   if (VecTy)
     SliceTy = VecTy;
 
bool SROA::splitAlloca(AllocaInst &AI, AllocaSlices &AS) {
   // Now that we have identified any pre-splitting opportunities,
   // mark loads and stores unsplittable except for the following case.
   // We leave a slice splittable if all other slices are disjoint or fully
   // included in the slice, such as whole-alloca loads and stores.
   // If we fail to split these during pre-splitting, we want to force them
   // to be rewritten into a partition.
   bool IsSorted = true;
 
-  uint64_t AllocaSize = DL.getTypeAllocSize(AI.getAllocatedType());
+  uint64_t AllocaSize =
+      DL.getTypeAllocSize(AI.getAllocatedType()).getFixedSize();
   const uint64_t MaxBitVectorSize = 1024;
   if (AllocaSize <= MaxBitVectorSize) {
     // If a byte boundary is included in any load or store, a slice starting or
     // ending at the boundary is not splittable.
     SmallBitVector SplittableOffset(AllocaSize + 1, true);
     for (Slice &S : AS)
       for (unsigned O = S.beginOffset() + 1;
            O < S.endOffset() && O < AllocaSize; O++)
bool SROA::splitAlloca(AllocaInst &AI, AllocaSlices &AS) {
   SmallVector<Fragment, 4> Fragments;
 
   // Rewrite each partition.
   for (auto &P : AS.partitions()) {
     if (AllocaInst *NewAI = rewritePartition(AI, AS, P)) {
       Changed = true;
       if (NewAI != &AI) {
         uint64_t SizeOfByte = 8;
-        uint64_t AllocaSize = DL.getTypeSizeInBits(NewAI->getAllocatedType());
+        uint64_t AllocaSize =
+            DL.getTypeSizeInBits(NewAI->getAllocatedType()).getFixedSize();
         // Don't include any padding.
         uint64_t Size = std::min(AllocaSize, P.size() * SizeOfByte);
         Fragments.push_back(Fragment(NewAI, P.beginOffset() * SizeOfByte, Size));
       }
     }
     ++NumPartitions;
   }
 
   NumAllocaPartitions += NumPartitions;
   MaxPartitionsPerAlloca.updateMax(NumPartitions);
 
   // Migrate debug information from the old alloca to the new alloca(s)
   // and the individual partitions.
   TinyPtrVector<DbgVariableIntrinsic *> DbgDeclares = FindDbgAddrUses(&AI);
   if (!DbgDeclares.empty()) {
     auto *Var = DbgDeclares.front()->getVariable();
     auto *Expr = DbgDeclares.front()->getExpression();
     auto VarSize = Var->getSizeInBits();
     DIBuilder DIB(*AI.getModule(), /*AllowUnresolved*/ false);
-    uint64_t AllocaSize = DL.getTypeSizeInBits(AI.getAllocatedType());
+    uint64_t AllocaSize =
+        DL.getTypeSizeInBits(AI.getAllocatedType()).getFixedSize();
     for (auto Fragment : Fragments) {
       // Create a fragment expression describing the new partition or reuse AI's
       // expression if there is only one partition.
       auto *FragmentExpr = Expr;
       if (Fragment.Size < AllocaSize || Expr->isFragment()) {
         // If this alloca is already a scalar replacement of a larger aggregate,
         // Fragment.Offset describes the offset inside the scalar.
         auto ExprFragment = Expr->getFragmentInfo();
bool SROA::runOnAlloca(AllocaInst &AI) {
   if (AI.use_empty()) {
     AI.eraseFromParent();
     return true;
   }
   const DataLayout &DL = AI.getModule()->getDataLayout();
 
   // Skip alloca forms that this analysis can't handle.
   if (AI.isArrayAllocation() || !AI.getAllocatedType()->isSized() ||
-      DL.getTypeAllocSize(AI.getAllocatedType()) == 0)
+      AI.getAllocatedType()->isScalableVectorTy() ||
+      DL.getTypeAllocSize(AI.getAllocatedType()).getFixedSize() == 0)
     return false;
 
   bool Changed = false;
 
   // First, split any FCA loads and stores touching this alloca to promote
   // better splitting and promotion opportunities.
   AggLoadStoreRewriter AggRewriter(DL);
   Changed |= AggRewriter.rewrite(AI);
PreservedAnalyses SROA::runImpl(Function &F, DominatorTree &RunDT,
   LLVM_DEBUG(dbgs() << "SROA function: " << F.getName() << "\n");
   C = &F.getContext();
   DT = &RunDT;
   AC = &RunAC;
 
   BasicBlock &EntryBB = F.getEntryBlock();
   for (BasicBlock::iterator I = EntryBB.begin(), E = std::prev(EntryBB.end());
        I != E; ++I) {
-    if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
+    if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
+      if (AI->getAllocatedType()->isScalableVectorTy() &&
+          isAllocaPromotable(AI))
+        PromotableAllocas.push_back(AI);
+      else
         Worklist.insert(AI);
     }
+  }
 
   bool Changed = false;
   // A set of deleted alloca instruction pointers which should be removed from
   // the list of promotable allocas.
   SmallPtrSet<AllocaInst *, 4> DeletedAllocas;
 
   do {
     while (!Worklist.empty()) {

llvm/test/Transforms/SROA/scalable-vectors.ll

+; RUN: opt < %s -sroa -S | FileCheck %s
+; RUN: opt < %s -passes=sroa -S | FileCheck %s
+
+; This test checks that SROA runs mem2reg on scalable vectors.
+
+define <vscale x 16 x i1> @alloca_nxv16i1(<vscale x 16 x i1> %pg) {
+; CHECK-LABEL: alloca_nxv16i1
+; CHECK-NEXT: ret <vscale x 16 x i1> %pg
+  %pg.addr = alloca <vscale x 16 x i1>
+  store <vscale x 16 x i1> %pg, <vscale x 16 x i1>* %pg.addr
+  %1 = load <vscale x 16 x i1>, <vscale x 16 x i1>* %pg.addr
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i8> @alloca_nxv16i8(<vscale x 16 x i8> %pg) {
+; CHECK-LABEL: alloca_nxv16i8
+; CHECK-NEXT: ret <vscale x 16 x i8> %pg
+  %pg.addr = alloca <vscale x 16 x i8>
+  store <vscale x 16 x i8> %pg, <vscale x 16 x i8>* %pg.addr
+  %1 = load <vscale x 16 x i8>, <vscale x 16 x i8>* %pg.addr
+  ret <vscale x 16 x i8> %1
+}