[mlir][Vector] Thread 0-d vectors through vector.transfer ops

[nicolasvasilache]

Summary

This revision adds 0-d vector support to vector.transfer ops.
In the process, numerous cleanups are applied, in particular around normalizing
and reducing the number of builders.

[ThomasRaoux]

mlir/test/Dialect/Vector/ops.mlir

7–14

Can 0-D vector be read from non 0-D tensor? It would be good to add a test for it.

[nicolasvasilache]

Address comment.
Use mandatory empty AffineMapAttr instead of optional null to reduce corner cases as suggested offline by @ThomasRaoux.

[springerm]

mlir/include/mlir/Dialect/Vector/VectorOps.td

1157

why did this change?

1329

extract from

1464

insert into

[dcaballe]

Thanks for helping with the 0-d vector problem! Added some minor comments. I have a question about the lowering of 0-d vectors to the scalar world. See comments inline.

mlir/include/mlir/Dialect/Vector/VectorOps.td

1157

+1, same for xfer write

1306–1329

and an?

mlir/lib/Conversion/VectorToROCDL/VectorToROCDL.cpp

69

These checks on GPU and ROCDL mean that 0-d tensors are not expected at all by these backends or it's more a TODO to be implemented in the future? If the latter, we should add a TODO comment staying that it will be implemented in the future.

mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp

230

This looks like we are vectorizing the operand to be stored in place. Shouldn't this call to some generic function that takes care of that (vectorizeOneOp or some other one)?

mlir/lib/Dialect/Vector/VectorOps.cpp

2317

and an?

mlir/lib/Dialect/Vector/VectorTransforms.cpp

2805

Move comment before the if?

2870–2877

load?

2895

Question about this lowering: why are we lowering 0-d vectors to the scalar world? I thought one of the goals discussed in the RFC was to keep everything within the vector world to avoid the scalar<->vector transition that could be very expensive for some targets. Shouldn't we lower 0-d vectors to a `vector<1xtype> in LLVM instead?

[ThomasRaoux]

mlir/lib/Conversion/VectorToGPU/VectorToGPU.cpp

74–76

we check that the rank must be 2 right below so this won't really change anything. I'm assuming you are just adding it from consistency? (same below)

[nicolasvasilache]

mlir/include/mlir/Dialect/Vector/VectorOps.td

1157

Because the comment was off by 1.
The thinking is that if you start counting operands at 0 and:

• plug in rank == 0 you now get [1 .. 1) which is

empty as expected; before it used to be 2 .. 1 which is weird

• plug in rank == 1, you now get [1 .. 2) instead of the more ambiguous 2 .. 2.

1329

dead now

1464

dead

mlir/lib/Conversion/VectorToGPU/VectorToGPU.cpp

74–76

Yes the evolution has been that I broke the API to make sure by way of compiler crashes that I am not missing any entry. Then I just copied this everywhere.

In this very special case I can just remove indeed.

mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp

230

This is a side-effect of the piece below that says:

// Not all ops support 0-d vectors, extract the scalar for now.
// TODO: remove this.
if (readValue.getType().cast<VectorType>().getRank() == 0)
  readValue = b.create<vector::ExtractElementOp>(loc, readValue);

Both should go away once all ops support 0-d vectors.

mlir/lib/Dialect/Vector/VectorTransforms.cpp

2895

Note that this is not changing any behavior here and we were already doing this extraction for vector<1x..x1xtype>.
Your point is very valid but I think is beyond this CL: we need to either:

• go directly to LLVM + bitcast
• introduce a bitcast
• let memref.load/store additionally support the 0-d vector case which itself may be quite fraught with issues

There are still deeper data layout issues lingering even in this trivial case (for architectures for which this matters that is .. ).
For now I don't think I can do better but adding a TODO.

[nicolasvasilache]

Address comments and rebase.

mlir/include/mlir/Dialect/Vector/VectorOps.td

 
 def Vector_TransferReadOp :
   Vector_Op<"transfer_read", [
       DeclareOpInterfaceMethods<VectorTransferOpInterface>,
       DeclareOpInterfaceMethods<VectorUnrollOpInterface, ["getShapeForUnroll"]>,
       DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
       AttrSizedOperandSegments
     ]>,
-    Arguments<(ins AnyShaped:$source, Variadic<Index>:$indices,
-               AffineMapAttr:$permutation_map, AnyType:$padding,
+    Arguments<(ins AnyShaped:$source,
+                   Variadic<Index>:$indices,
+                   AffineMapAttr:$permutation_map,
+                   AnyType:$padding,
                    Optional<VectorOf<[I1]>>:$mask,
                    OptionalAttr<BoolArrayAttr>:$in_bounds)>,
-    Results<(outs AnyVector:$vector)> {
+    Results<(outs AnyVectorOfAnyRank:$vector)> {
 
   let summary = "Reads a supervector from memory into an SSA vector value.";
 
   let description = [{
     The `vector.transfer_read` op performs a read from a slice within a
     [MemRef](../LangRef.md#memref-type) or a Ranked
-    [Tensor](../LangRef.md#tensor-type) supplied as its first operand into a
-    [vector](../LangRef.md#vector-type) of the same base elemental type.
+    [Tensor](../LangRef.md#tensor-type) supplied as its first operand
+    into a [vector](../LangRef.md#vector-type) of the same base elemental type.
 
     A memref/tensor operand with vector element type, must have its vector
     element type match a suffix (shape and element type) of the vector (e.g.
     memref<3x2x6x4x3xf32>, vector<1x1x4x3xf32>).
 
     The slice is further defined by a full-rank index within the MemRef/Tensor,
-    supplied as the operands `2 .. 1 + rank(memref/tensor)`.
+    supplied as the operands `[1 .. 1 + rank(memref/tensor))`.
 
     The permutation_map [attribute](../LangRef.md#attributes) is an
     [affine-map](Affine.md#affine-maps) which specifies the transposition on the
     slice to match the vector shape. The permutation map may be implicit and
     omitted from parsing and printing if it is the canonical minor identity map
     (i.e. if it does not permute or broadcast any dimension).
 
     The size of the slice is specified by the size of the vector, given as the
let description = [{
     // Special encoding for 0-d transfer with 0-d tensor/memref, vector shape
     // {1} and permutation_map () -> (0).
     %0 = vector.transfer_read %arg0[], %f0 {permutation_map = affine_map<()->(0)>} :
       tensor<f32>, vector<1xf32>
     ```
   }];
 
   let builders = [
-    // Builder that sets padding to zero.
-    OpBuilder<(ins "VectorType":$vector, "Value":$source,
-      "ValueRange":$indices, "AffineMap":$permutationMap,
-      CArg<"ArrayRef<bool>", "{}">:$inBounds)>,
-    // Builder that sets permutation map to 'getMinorIdentityMap'.
-    OpBuilder<(ins "VectorType":$vector, "Value":$source,
-      "ValueRange":$indices, "Value":$padding,
-      CArg<"ArrayRef<bool>", "{}">:$inBounds)>,
-    // Builder that sets permutation map (resp. padding) to
-    // 'getMinorIdentityMap' (resp. zero).
-    OpBuilder<(ins "VectorType":$vector, "Value":$source,
-      "ValueRange":$indices, CArg<"ArrayRef<bool>", "{}">:$inBounds)>,
-    // Builder that does not set mask.
-    OpBuilder<(ins "Type":$vector, "Value":$source,
-      "ValueRange":$indices, "AffineMapAttr":$permutationMap, "Value":$padding,
-      "ArrayAttr":$inBounds)>,
-    // Builder that does not set mask.
-    OpBuilder<(ins "Type":$vector, "Value":$source,
-      "ValueRange":$indices, "AffineMap":$permutationMap, "Value":$padding,
-      "ArrayAttr":$inBounds)>
+    /// 1. Builder that sets padding to zero and an empty mask (variant with attrs).
+    OpBuilder<(ins "VectorType":$vectorType,
+                   "Value":$source,
+                   "ValueRange":$indices,
+                   "AffineMapAttr":$permutationMapAttr,
+                   "ArrayAttr":$inBoundsAttr)>,
+    /// 2. Builder that sets padding to zero and an empty mask (variant without attrs).
+    OpBuilder<(ins "VectorType":$vectorType,
+                   "Value":$source,
+                   "ValueRange":$indices,
+                   "AffineMap":$permutationMap,
+                   CArg<"Optional<ArrayRef<bool>>", "::llvm::None">:$inBounds)>,
+    /// 3. Builder that sets permutation map to 'getMinorIdentityMap'.
+    OpBuilder<(ins "VectorType":$vectorType,
+                   "Value":$source,
+                   "ValueRange":$indices,
+                   "Value":$padding,
+                   CArg<"Optional<ArrayRef<bool>>", "::llvm::None">:$inBounds)>,
+    /// 4. Builder that sets padding to zero and permutation map to
+    /// 'getMinorIdentityMap'.
+    OpBuilder<(ins "VectorType":$vectorType,
+                   "Value":$source,
+                   "ValueRange":$indices,
+                   CArg<"Optional<ArrayRef<bool>>", "::llvm::None">:$inBounds)>,
   ];
-
-  let extraClassDeclaration = [{
-    /// Temporary convenience builders to account for the fact that we do not
-    /// have 0-d vectors atm. These create a constant `vector<1xt>` and
-    /// insert/extract into it.
-    // Builder that sets permutation map (resp. padding) to
-    // 'getMinorIdentityMap' (resp. zero).
-    static Value createScalarOp(OpBuilder &builder, Location loc, Value source,
-                                ValueRange indices,
-                                ArrayRef<bool> inBounds = ArrayRef<bool>{});
-  }];
-
   let hasCanonicalizer = 1;
   let hasFolder = 1;
 }
 
 def Vector_TransferWriteOp :
   Vector_Op<"transfer_write", [
       DeclareOpInterfaceMethods<VectorTransferOpInterface>,
       DeclareOpInterfaceMethods<VectorUnrollOpInterface, ["getShapeForUnroll"]>,
       DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
       AttrSizedOperandSegments
   ]>,
-    Arguments<(ins AnyVector:$vector, AnyShaped:$source,
+    Arguments<(ins AnyVectorOfAnyRank:$vector,
+                   AnyShaped:$source,
                    Variadic<Index>:$indices,
                    AffineMapAttr:$permutation_map,
                    Optional<VectorOf<[I1]>>:$mask,
                    OptionalAttr<BoolArrayAttr>:$in_bounds)>,
     Results<(outs Optional<AnyRankedTensor>:$result)> {
 
   let summary = "The vector.transfer_write op writes a supervector to memory.";
 
   let description = [{
     The `vector.transfer_write` op performs a write from a
     [vector](../LangRef.md#vector-type), supplied as its first operand, into a
     slice within a [MemRef](../LangRef.md#memref-type) or a Ranked
     [Tensor](../LangRef.md#tensor-type) of the same base elemental type,
     supplied as its second operand.
 
     A vector memref/tensor operand must have its vector element type match a
     suffix (shape and element type) of the vector (e.g. memref<3x2x6x4x3xf32>,
     vector<1x1x4x3xf32>). If the operand is a tensor, the operation returns a
     new tensor of the same type.
 
     The slice is further defined by a full-rank index within the MemRef/Tensor,
-    supplied as the operands `3 .. 2 + rank(memref/tensor)`.
+    supplied as the operands `[2 .. 2 + rank(memref/tensor))`.
 
     The permutation_map [attribute](../LangRef.md#attributes) is an
     [affine-map](Affine.md#affine-maps) which specifies the transposition on the
     slice to match the vector shape. The permutation map may be implicit and
     omitted from parsing and printing if it is the canonical minor identity map
     (i.e. if it does not permute any dimension). In contrast to `transfer_read`,
     write ops cannot have broadcast dimensions.
 
let description = [{
     // Special encoding for 0-d transfer with 0-d tensor/memref, vector shape
     // {1} and permutation_map () -> (0).
     %1 = vector.transfer_write %0, %arg0[] {permutation_map = affine_map<()->(0)>} :
       vector<1xf32>, tensor<f32>
     ```
   }];
 
   let builders = [
-    // Builder that sets an empty mask.
-    OpBuilder<(ins "Value":$vector, "Value":$source, "ValueRange":$indices,
-      "AffineMap":$permutationMap, CArg<"ArrayRef<bool>", "{}">:$inBounds)>,
-    // Builder that sets permutation map to 'getMinorIdentityMap'.
-    OpBuilder<(ins "Value":$vector, "Value":$source, "ValueRange":$indices,
-      CArg<"ArrayRef<bool>", "{}">:$inBounds)>,
-    OpBuilder<(ins "Value":$vector, "Value":$source, "ValueRange":$indices,
-      "AffineMapAttr":$permutationMap, "ArrayAttr":$inBounds)>,
-    OpBuilder<(ins "Value":$vector, "Value":$source, "ValueRange":$indices,
-      "AffineMap":$permutationMap, "Value":$mask, "ArrayAttr":$inBounds)>,
-    OpBuilder<(ins "Value":$vector, "Value":$source, "ValueRange":$indices,
-      "AffineMap":$permutationMap, "ArrayAttr":$inBounds)>,
+    /// 1. Builder with type inference.
+    OpBuilder<(ins "Value":$vector,
+                   "Value":$dest,
+                   "ValueRange":$indices,
+                   "AffineMapAttr":$permutationMapAttr,
+                   "Value":$mask,
+                   "ArrayAttr":$inBoundsAttr)>,
+    /// 2. Builder with type inference that sets an empty mask (variant with attrs).
+    OpBuilder<(ins "Value":$vector,
+                   "Value":$dest,
+                   "ValueRange":$indices,
+                   "AffineMapAttr":$permutationMapAttr,
+                   "ArrayAttr":$inBoundsAttr)>,
+    /// 3. Builder with type inference that sets an empty mask (variant without attrs).
+    OpBuilder<(ins "Value":$vector,
+                   "Value":$dest,
+                   "ValueRange":$indices,
+                   "AffineMap":$permutationMap,
+                   CArg<"Optional<ArrayRef<bool>>", "::llvm::None">:$inBounds)>,
+    /// 4. Builder with type inference that sets an empty mask and sets permutation
+    /// map to 'getMinorIdentityMap'.
+    OpBuilder<(ins "Value":$vector,
+                   "Value":$dest,
+                   "ValueRange":$indices,
+                   CArg<"Optional<ArrayRef<bool>>", "::llvm::None">:$inBounds)>,
   ];
-
-  let extraClassDeclaration = [{
-    /// Temporary convenience builders to account for the fact that we do not
-    /// have 0-d vectors atm. These create a constant `vector<1xt>` and
-    /// insert/extract into it.
-    // Builder that sets permutation map (resp. padding) to
-    // 'getMinorIdentityMap' (resp. zero).
-    static Operation *createScalarOp(
-      OpBuilder &builder, Location loc, Value value,
-      Value dest, ValueRange indices,
-      ArrayRef<bool> inBounds = ArrayRef<bool>{});
-  }];
-
   let hasFolder = 1;
   let hasCanonicalizer = 1;
 }
 
 def Vector_LoadOp : Vector_Op<"load"> {
   let summary = "reads an n-D slice of memory into an n-D vector";
   let description = [{
     The 'vector.load' operation reads an n-D slice of memory into an n-D

mlir/include/mlir/Interfaces/VectorInterfaces.td

InterfaceMethod<
       /*desc=*/"Return the permutation map.",
       /*retTy=*/"::mlir::AffineMap",
       /*methodName=*/"permutation_map",
       /*args=*/(ins),
       /*methodBody=*/"return $_op.permutation_map();"
       /*defaultImplementation=*/
     >,
     InterfaceMethod<
-      /*desc=*/[{
-        Returns true if op involves a 0-d tensor/memref and a vector
-        of shape {1}. This is temporary until we have 0-d vectors.
-        // TODO: turn this into 0-d vectors + empty permutation_map.
-      }],
-      /*retTy=*/"bool",
-      /*methodName=*/"isZeroD",
-      /*args=*/(ins),
-      /*methodBody=*/"",
-      /*defaultImplementation=*/[{
-        if (getShapedType().getRank() > 0)
-          return false;
-        if (getVectorType().getShape() != ArrayRef<int64_t>{1})
-          return false;
-        AffineMap map = AffineMap::get(
-          /*numDims=*/0, /*numSymbols=*/0,
-          getAffineConstantExpr(0, $_op->getContext()));
-        if ($_op.permutation_map() != map)
-          return false;
-        return true;
-      }]
-    >,
-    InterfaceMethod<
       /*desc=*/[{ Returns true if the specified dimension is a broadcast. }],
       /*retTy=*/"bool",
       /*methodName=*/"isBroadcastDim",
       /*args=*/(ins "unsigned":$idx),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
         auto expr = $_op.permutation_map().getResult(idx);
         return expr.template isa<::mlir::AffineConstantExpr>() &&
                expr.template dyn_cast<::mlir::AffineConstantExpr>().getValue() == 0;
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{ Returns true if at least one of the dimensions in the
                   permutation map is a broadcast.}],
       /*retTy=*/"bool",
       /*methodName=*/"hasBroadcastDim",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        // 0-d transfers are not considered broadcasts but they need to be 
-        // represented with a vector<1xt> until we have 0-d vectors.
-        if ($_op.isZeroD()) return false;
-        for (unsigned i = 0; i < $_op.permutation_map().getNumResults(); ++i) {
+        for (unsigned i = 0, rank = getTransferRank(); i < rank; ++i) {
           if ($_op.isBroadcastDim(i))
             return true;
         }
         return false;
       }]
     >,
     InterfaceMethod<
       /*desc=*/"Return the `in_bounds` boolean ArrayAttr.",

mlir/lib/Conversion/VectorToGPU/VectorToGPU.cpp

if (failed(getStridesAndOffset(memrefType, strides, offset)))
     return llvm::None;
   if (strides[0] == ShapedType::kDynamicStrideOrOffset)
     return llvm::None;
   return strides[0];
 }
 
 // Return true if the transfer op can be converted to a MMA matrix load.
 static bool transferReadSupportsMMAMatrixType(vector::TransferReadOp readOp) {
   if (readOp.mask() || readOp.hasOutOfBoundsDim() ||
       readOp.getVectorType().getRank() != 2)
     return false;
   if (!getMemrefConstantHorizontalStride(readOp.getShapedType()))
     return false;
   AffineMap map = readOp.permutation_map();
   OpBuilder b(readOp.getContext());
   AffineExpr innerDim = b.getAffineDimExpr(map.getNumDims() - 1);
   AffineExpr zero = b.getAffineConstantExpr(0);
   auto broadcastInnerDim = AffineMap::get(map.getNumDims(), 0, {zero, innerDim},
                                           readOp.getContext());
   // TODO: Support transpose once it is added to GPU dialect ops.
   // For now we only support (d0, d1) -> (d0, d1) and (d0, d1) -> (0, d1).
   if (!map.isMinorIdentity() && map != broadcastInnerDim)
     return false;
   return true;
 }
 
 // Return true if the transfer op can be converted to a MMA matrix store.
 static bool
 transferWriteSupportsMMAMatrixType(vector::TransferWriteOp writeOp) {
+  // TODO: support 0-d corner case.
+  if (writeOp.getTransferRank() == 0)
+    return false;
+
   if (writeOp.mask() || writeOp.hasOutOfBoundsDim() ||
       writeOp.getVectorType().getRank() != 2)
     return false;
   if (!getMemrefConstantHorizontalStride(writeOp.getShapedType()))
     return false;
   // TODO: Support transpose once it is added to GPU dialect ops.
   if (!writeOp.permutation_map().isMinorIdentity())
     return false;
struct CombineTransferReadOpTranspose final
     : public OpRewritePattern<vector::TransposeOp> {
   using OpRewritePattern<vector::TransposeOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::TransposeOp op,
                                 PatternRewriter &rewriter) const override {
     auto transferReadOp = op.vector().getDefiningOp<vector::TransferReadOp>();
     if (!transferReadOp)
       return failure();
+
+    // TODO: support 0-d corner case.
+    if (transferReadOp.getTransferRank() == 0)
+      return failure();
+
     if (transferReadOp.mask() || transferReadOp.hasOutOfBoundsDim())
       return failure();
     SmallVector<int64_t, 2> perm;
     op.getTransp(perm);
     SmallVector<unsigned, 2> permU;
     for (int64_t o : perm)
       permU.push_back(unsigned(o));
     AffineMap permutationMap =
         AffineMap::getPermutationMap(permU, op.getContext());
     AffineMap newMap = permutationMap.compose(transferReadOp.permutation_map());
     rewriter.replaceOpWithNewOp<vector::TransferReadOp>(
         op, op.getType(), transferReadOp.source(), transferReadOp.indices(),
-        newMap, transferReadOp.padding(), transferReadOp.mask(),
-        transferReadOp.in_boundsAttr());
+        AffineMapAttr::get(newMap), transferReadOp.padding(),
+        transferReadOp.mask(), transferReadOp.in_boundsAttr());
     return success();
   }
 };
 
 } // namespace
 
 // MMA types have different layout based on how they are used in matmul ops.
 // Figure the right layout to use by looking at op uses.
for (Operation *users : op->getUsers()) {
     if (contract.rhs() == op.getResult())
       return "BOp";
   }
   return "COp";
 }
 
 static void convertTransferReadOp(vector::TransferReadOp op,
                                   llvm::DenseMap<Value, Value> &valueMapping) {
+  assert(op.getTransferRank() > 0 && "unexpected 0-d transfer");
   assert(transferReadSupportsMMAMatrixType(op));
   Optional<int64_t> stride =
       getMemrefConstantHorizontalStride(op.getShapedType());
   AffineMap map = op.permutation_map();
   // Handle broadcast by setting the stride to 0.
   if (map.getResult(0).isa<AffineConstantExpr>()) {
     assert(map.getResult(0).cast<AffineConstantExpr>().getValue() == 0);
     stride = 0;

mlir/lib/Conversion/VectorToROCDL/VectorToROCDL.cpp

 template <typename ConcreteOp>
 class VectorTransferConversion : public ConvertOpToLLVMPattern<ConcreteOp> {
 public:
   using ConvertOpToLLVMPattern<ConcreteOp>::ConvertOpToLLVMPattern;
 
   LogicalResult
   matchAndRewrite(ConcreteOp xferOp, typename ConcreteOp::Adaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (xferOp.getTransferRank() == 0)
+      return failure();
+
     if (xferOp.getVectorType().getRank() > 1 ||
         llvm::size(xferOp.indices()) == 0)
       return failure();
 
     if (!xferOp.permutation_map().isMinorIdentity())
       return failure();
 
     // Have it handled in vector->llvm conversion pass.

mlir/lib/Conversion/VectorToSCF/VectorToSCF.cpp

struct VectorToSCFPattern : public OpRewritePattern<OpTy> {
   VectorTransferToSCFOptions options;
 };
 
 /// Given a vector transfer op, calculate which dimension of the `source`
 /// memref should be unpacked in the next application of TransferOpConversion.
 /// A return value of None indicates a broadcast.
 template <typename OpTy>
 static Optional<int64_t> unpackedDim(OpTy xferOp) {
+  // TODO: support 0-d corner case.
+  assert(xferOp.getTransferRank() > 0 && "unexpected 0-d transfer");
   auto map = xferOp.permutation_map();
   if (auto expr = map.getResult(0).template dyn_cast<AffineDimExpr>()) {
     return expr.getPosition();
   }
   assert(xferOp.isBroadcastDim(0) &&
          "Expected AffineDimExpr or AffineConstantExpr");
   return None;
 }
 
 /// Compute the permutation map for the new (N-1)-D vector transfer op. This
 /// map is identical to the current permutation map, but the first result is
 /// omitted.
 template <typename OpTy>
 static AffineMap unpackedPermutationMap(OpBuilder &b, OpTy xferOp) {
+  // TODO: support 0-d corner case.
+  assert(xferOp.getTransferRank() > 0 && "unexpected 0-d transfer");
   auto map = xferOp.permutation_map();
   return AffineMap::get(map.getNumDims(), 0, map.getResults().drop_front(),
                         b.getContext());
 }
 
 /// Calculate the indices for the new vector transfer op.
 ///
 /// E.g.: transfer_read %A[%a, %b, %c, %d] ... : vector<5x4x3xf32> ...

 /// part of TransferOp1dConversion. Return the memref dimension on which
 /// the transfer is operating. A return value of None indicates a broadcast.
 template <typename OpTy>
 static Optional<int64_t>
 get1dMemrefIndices(OpBuilder &b, OpTy xferOp, Value iv,
                    SmallVector<Value, 8> &memrefIndices) {
   auto indices = xferOp.indices();
   auto map = xferOp.permutation_map();
+  assert(xferOp.getTransferRank() > 0 && "unexpected 0-d transfer");
 
   memrefIndices.append(indices.begin(), indices.end());
   assert(map.getNumResults() == 1 &&
          "Expected 1 permutation map result for 1D transfer");
   if (auto expr = map.getResult(0).template dyn_cast<AffineDimExpr>()) {
     Location loc = xferOp.getLoc();
     auto dim = expr.getPosition();
     AffineExpr d0, d1;

 /// }
 /// ```
 template <typename OpTy>
 struct TransferOp1dConversion : public VectorToSCFPattern<OpTy> {
   using VectorToSCFPattern<OpTy>::VectorToSCFPattern;
 
   LogicalResult matchAndRewrite(OpTy xferOp,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (xferOp.getTransferRank() == 0)
+      return failure();
     auto map = xferOp.permutation_map();
     auto memRefType = xferOp.getShapedType().template dyn_cast<MemRefType>();
 
     if (!memRefType)
       return failure();
     if (xferOp.getVectorType().getRank() != 1)
       return failure();
     if (map.isMinorIdentity() && isLastMemrefDimUnitStride(memRefType))

mlir/lib/Dialect/Linalg/ComprehensiveBufferize/VectorInterfaceImpl.cpp

LogicalResult bufferize(Operation *op, OpBuilder &b,
     // this point.
     assert(writeOp.getShapedType().isa<TensorType>() &&
            "only tensor types expected");
     Value resultBuffer = getResultBuffer(b, op->getResult(0), state);
     if (!resultBuffer)
       return failure();
     b.create<vector::TransferWriteOp>(
         writeOp.getLoc(), writeOp.vector(), resultBuffer, writeOp.indices(),
-        writeOp.permutation_map(),
-        writeOp.in_bounds() ? *writeOp.in_bounds() : ArrayAttr());
+        writeOp.permutation_mapAttr(), writeOp.in_boundsAttr());
     state.mapBuffer(op->getResult(0), resultBuffer);
 
     return success();
   }
 };
 
 } // namespace vector_ext
 } // namespace comprehensive_bufferize

mlir/lib/Dialect/Linalg/Transforms/Vectorization.cpp

struct VectorizationResult {
   /// Replacement behavior is specified by `status`.
   Operation *newOp;
 };
 
 /// Return a vector type of the same shape and element type as the (assumed)
 /// ShapedType of `v`.
 static VectorType extractVectorTypeFromShapedValue(Value v) {
   auto st = v.getType().cast<ShapedType>();
-  if (st.getShape().empty())
-    return VectorType();
   return VectorType::get(st.getShape(), st.getElementType());
 }
 
 static llvm::Optional<vector::CombiningKind>
 getKindForOp(Operation *reductionOp) {
   if (!reductionOp)
     return llvm::None;
   return llvm::TypeSwitch<Operation *, llvm::Optional<vector::CombiningKind>>(
VectorType targetVectorType =
       VectorType::get(shape, getElementTypeOrSelf(value));
   if (vector::isBroadcastableTo(value.getType(), targetVectorType) !=
       vector::BroadcastableToResult::Success)
     return value;
   Location loc = b.getInsertionPoint()->getLoc();
   return b.createOrFold<vector::BroadcastOp>(loc, targetVectorType, value);
 }
 
-/// Build a vector.transfer_read from `source` at indices set to all `0`.
-/// If source has rank zero, build a `vector<1xt> transfer_read + extract`.
-/// Return the produced value.
-static Value buildVectorRead(OpBuilder &b, Value source, Type readType,
-                             AffineMap map) {
-  Location loc = source.getLoc();
-  auto shapedType = source.getType().cast<ShapedType>();
-  SmallVector<Value> indices(shapedType.getRank(),
-                             b.create<arith::ConstantIndexOp>(loc, 0));
-  if (auto vectorType = readType.dyn_cast<VectorType>())
-    return b.create<vector::TransferReadOp>(loc, vectorType, source, indices,
-                                            map);
-  return vector::TransferReadOp::createScalarOp(b, loc, source, indices);
-}
-
 /// Create MultiDimReductionOp to compute the reduction for `reductionOp`. This
 /// assumes that `reductionOp` has two operands and one of them is the reduction
 /// initial value.
 static Value buildMultiDimReduce(OpBuilder &b, Operation *reduceOp,
                                  Value valueToReduce,
                                  const SmallVector<bool> &reductionMask) {
   auto maybeKind = getKindForOp(reduceOp);
   assert(maybeKind && "Failed precondition: could not get reduction kind");

 /// to all `0`; where `outputOperand` is an output operand of the LinalgOp
 /// currently being vectorized. If `dest` has null rank, build an memref.store.
 /// Return the produced value or null if no value is produced.
 static Value buildVectorWrite(OpBuilder &b, Value value,
                               OpOperand *outputOperand) {
   Operation *write;
   Location loc = value.getLoc();
   auto linalgOp = cast<LinalgOp>(outputOperand->getOwner());
-  if (VectorType vectorType =
-          extractVectorTypeFromShapedValue(outputOperand->get())) {
+  ArrayRef<int64_t> shape = linalgOp.getShape(outputOperand);
+  auto vectorType = VectorType::get(
+      shape, getElementTypeOrSelf(outputOperand->get().getType()));
+  if (vectorType.getRank() > 0) {
+    // 0-d case is still special: do not invert the reindexing map.
     AffineMap map =
         reindexIndexingMap(linalgOp.getTiedIndexingMap(outputOperand));
     SmallVector<int64_t> transposeShape =
         applyPermutationMap(inversePermutation(map), vectorType.getShape());
     assert(!transposeShape.empty() && "unexpected empty transpose shape");
     vectorType = VectorType::get(transposeShape, vectorType.getElementType());
     SmallVector<Value> indices(linalgOp.getRank(outputOperand),
                                b.create<arith::ConstantIndexOp>(loc, 0));
     value = broadcastIfNeeded(b, value, vectorType.getShape());
     write = b.create<vector::TransferWriteOp>(loc, value, outputOperand->get(),
                                               indices, map);
   } else {
-    write = vector::TransferWriteOp::createScalarOp(
-        b, loc, value, outputOperand->get(), ValueRange{});
+    if (!value.getType().isa<VectorType>())
+      value = b.create<vector::BroadcastOp>(loc, vectorType, value);
+    assert(value.getType() == vectorType && "incorrect type");
+    write = b.create<vector::TransferWriteOp>(loc, value, outputOperand->get(),
+                                              ValueRange{});
   }
   LDBG("vectorized op: " << *write);
   if (!write->getResults().empty())
     return write->getResult(0);
   return Value();
 }
 
 // Custom vectorization function type. Produce a vector form of Operation*
if (linalgOp.getNumOutputs() == 0)
     return failure();
 
   // TODO: the common vector shape is equal to the static loop sizes only when
   // all indexing maps are projected permutations. For convs and stencils the
   // logic will need to evolve.
   SmallVector<int64_t> commonVectorShape = linalgOp.computeStaticLoopSizes();
 
   // 3. Turn all BBArgs into vector.transfer_read / load.
-  SmallVector<AffineMap> indexings;
+  Location loc = linalgOp.getLoc();
+  Value zero = b.create<arith::ConstantIndexOp>(loc, 0);
   for (OpOperand *opOperand : linalgOp.getInputAndOutputOperands()) {
     BlockArgument bbarg = block->getArgument(opOperand->getOperandNumber());
     if (linalgOp.isScalar(opOperand)) {
       bvm.map(bbarg, opOperand->get());
       continue;
     }
-    // TODO: 0-d vectors.
-    Type readType;
+    VectorType readType;
     AffineMap map;
-    if (linalgOp.getShape(opOperand).empty()) {
-      readType = bbarg.getType();
-    } else {
-      if (opOperand->getOperandNumber() < linalgOp.getNumInputs()) {
-        map = inverseAndBroadcastProjectedPermuation(
-            linalgOp.getTiedIndexingMap(opOperand));
-        readType = VectorType::get(commonVectorShape,
-                                   getElementTypeOrSelf(opOperand->get()));
-      } else {
-        map = inversePermutation(
-            reindexIndexingMap(linalgOp.getTiedIndexingMap(opOperand)));
-        readType = VectorType::get(map.compose(linalgOp.getShape(opOperand)),
-                                   getElementTypeOrSelf(opOperand->get()));
-      }
-    }
-    Value readValue = buildVectorRead(b, opOperand->get(), readType, map);
+    // TODO: can we keep this simplification?
+    // if (linalgOp.getShape(opOperand).empty()) {
+    //   readType = VectorType::get({}, bbarg.getType());
+    // } else {
+    // }
+
+    auto shape = linalgOp.getShape(opOperand);
+    SmallVector<Value> indices(shape.size(), zero);
+    Value readValue = b.create<vector::TransferReadOp>(
+        loc, readType, opOperand->get(), indices, map);
+    // Not all ops support 0-d vectors, extract the scalar for now.
+    // TODO: remove this.
+    if (readValue.getType().cast<VectorType>().getRank() == 0)
+      readValue = b.create<vector::ExtractElementOp>(loc, readValue);
+
     LDBG("new vectorized bbarg(" << bbarg.getArgNumber() << "): " << readValue);
     bvm.map(bbarg, readValue);
     bvm.map(opOperand->get(), readValue);
   }
 
   SmallVector<CustomVectorizationHook> hooks;
   // 4a. Register CustomVectorizationHook for yieldOp.
   CustomVectorizationHook vectorizeYield =
static LogicalResult tryVectorizeCopy(PatternRewriter &rewriter,
     auto vecType = VectorType::get(vecShape, sourceType.getElementType());
 
     // Generate TransferReadOp.
     SmallVector<Value> readIndices(
         vecType.getRank(),
         rewriter.create<arith::ConstantIndexOp>(padOp.getLoc(), 0));
     auto read = rewriter.create<vector::TransferReadOp>(
         padOp.getLoc(), vecType, padOp.source(), readIndices, padValue,
-        readInBounds);
+        ArrayRef<bool>{readInBounds});
 
     // If `dest` is a FillOp and the TransferWriteOp would overwrite the entire
     // tensor, write directly to the FillOp's operand.
     if (llvm::equal(vecShape, resultType.getShape()) &&
         llvm::all_of(writeInBounds, [](bool b) { return b; }))
       if (auto fill = dest.getDefiningOp<FillOp>())
         dest = fill.output();
 
     // Generate TransferWriteOp.
     auto writeIndices =
         ofrToIndexValues(rewriter, padOp.getLoc(), padOp.getMixedLowPad());
     rewriter.replaceOpWithNewOp<vector::TransferWriteOp>(
-        padOp, read, dest, writeIndices, writeInBounds);
+        padOp, read, dest, writeIndices, ArrayRef<bool>{writeInBounds});
 
     return success();
   }
 };
 
 /// Base pattern for rewriting PadTensorOps whose result is consumed by a given
 /// operation type OpTy.
 template <typename OpTy>

 /// - Single, scalar padding value.
 struct PadTensorOpVectorizationWithTransferWritePattern
     : public VectorizePadTensorOpUserPattern<vector::TransferWriteOp> {
   using VectorizePadTensorOpUserPattern<
       vector::TransferWriteOp>::VectorizePadTensorOpUserPattern;
 
   LogicalResult rewriteUser(PatternRewriter &rewriter, PadTensorOp padOp,
                             vector::TransferWriteOp xferOp) const override {
+    // TODO: support 0-d corner case.
+    if (xferOp.getTransferRank() == 0)
+      return failure();
+
     // Low padding must be static 0.
     if (!padOp.hasZeroLowPad())
       return failure();
     // Pad value must be a constant.
     auto padValue = padOp.getConstantPaddingValue();
     if (!padValue)
       return failure();
     // TransferWriteOp result must be directly consumed by an ExtractSliceOp.
LogicalResult rewriteUser(PatternRewriter &rewriter, PadTensorOp padOp,
 
     // Generate TransferWriteOp: Write to InsertSliceOp's dest tensor at
     // specified offsets. Write is fully in-bounds because a InsertSliceOp's
     // source must fit into the destination at the specified offsets.
     auto writeIndices =
         ofrToIndexValues(rewriter, padOp.getLoc(), insertOp.getMixedOffsets());
     SmallVector<bool> inBounds(vecRank, true);
     rewriter.replaceOpWithNewOp<vector::TransferWriteOp>(
-        insertOp, read, insertOp.dest(), writeIndices, inBounds);
+        insertOp, read, insertOp.dest(), writeIndices,
+        ArrayRef<bool>{inBounds});
 
     return success();
   }
 };
 
 void mlir::linalg::populatePadTensorOpVectorizationPatterns(
     RewritePatternSet &patterns, PatternBenefit baseBenefit) {
   patterns.add<GenericPadTensorOpVectorizationPattern>(patterns.getContext(),
static memref::SubViewOp getSubViewUseIfUnique(Value v) {
   return subViewOp;
 }
 
 /// TODO: use interfaces, side-effects and aliasing analysis as appropriate,
 /// when available.
 LogicalResult LinalgCopyVTRForwardingPattern::matchAndRewrite(
     vector::TransferReadOp xferOp, PatternRewriter &rewriter) const {
 
+  // TODO: support mask.
+  if (xferOp.mask())
+    return failure();
+
   // Transfer into `view`.
   Value viewOrAlloc = xferOp.source();
   if (!viewOrAlloc.getDefiningOp<memref::ViewOp>() &&
       !viewOrAlloc.getDefiningOp<memref::AllocOp>())
     return failure();
 
   LDBG(viewOrAlloc);
 
LogicalResult LinalgCopyVTRForwardingPattern::matchAndRewrite(
   Value in = copyOp.input();
 
   // linalg.copy + linalg.fill can be used to create a padded local buffer.
   // The `masked` attribute is only valid on this padded buffer.
   // When forwarding to vector.transfer_read, the attribute must be reset
   // conservatively.
   Value res = rewriter.create<vector::TransferReadOp>(
       xferOp.getLoc(), xferOp.getVectorType(), in, xferOp.indices(),
-      xferOp.permutation_map(), xferOp.padding(), ArrayAttr());
+      xferOp.permutation_mapAttr(), xferOp.padding(), xferOp.mask(),
+      // in_bounds is explicitly reset
+      /*inBoundsAttr=*/ArrayAttr());
 
   if (maybeFillOp)
     rewriter.eraseOp(maybeFillOp);
   rewriter.eraseOp(copyOp);
   rewriter.replaceOp(xferOp, res);
 
   return success();
 }
 
 /// TODO: use interfaces, side-effects and aliasing analysis as appropriate,
 /// when available.
 LogicalResult LinalgCopyVTWForwardingPattern::matchAndRewrite(
     vector::TransferWriteOp xferOp, PatternRewriter &rewriter) const {
+  // TODO: support mask.
+  if (xferOp.mask())
+    return failure();
+
   // Transfer into `viewOrAlloc`.
   Value viewOrAlloc = xferOp.source();
   if (!viewOrAlloc.getDefiningOp<memref::ViewOp>() &&
       !viewOrAlloc.getDefiningOp<memref::AllocOp>())
     return failure();
 
   // Ensure there is exactly one subview of `viewOrAlloc` defining `subView`.
   memref::SubViewOp subViewOp = getSubViewUseIfUnique(viewOrAlloc);
LogicalResult LinalgCopyVTWForwardingPattern::matchAndRewrite(
 
   // Forward vector.transfer into copy.
   // linalg.copy + linalg.fill can be used to create a padded local buffer.
   // The `masked` attribute is only valid on this padded buffer.
   // When forwarding to vector.transfer_write, the attribute must be reset
   // conservatively.
   rewriter.create<vector::TransferWriteOp>(
       xferOp.getLoc(), xferOp.vector(), out, xferOp.indices(),
-      xferOp.permutation_map(), ArrayAttr());
+      xferOp.permutation_mapAttr(), xferOp.mask(),
+      // in_bounds is explicitly reset
+      /*inBoundsAttr=*/ArrayAttr());
 
   rewriter.eraseOp(copyOp);
   rewriter.eraseOp(xferOp);
 
   return success();
 }
 
 //===----------------------------------------------------------------------===//

mlir/lib/Dialect/MemRef/Transforms/FoldSubViewOps.cpp

 
 static Value getMemRefOperand(vector::TransferWriteOp op) {
   return op.source();
 }
 
 /// Given the permutation map of the original
 /// `vector.transfer_read`/`vector.transfer_write` operations compute the
 /// permutation map to use after the subview is folded with it.
-static AffineMap getPermutationMap(MLIRContext *context,
+static AffineMapAttr getPermutationMapAttr(MLIRContext *context,
                                            memref::SubViewOp subViewOp,
                                            AffineMap currPermutationMap) {
   llvm::SmallDenseSet<unsigned> unusedDims = subViewOp.getDroppedDims();
   SmallVector<AffineExpr> exprs;
   int64_t sourceRank = subViewOp.getSourceType().getRank();
   for (auto dim : llvm::seq<int64_t>(0, sourceRank)) {
     if (unusedDims.count(dim))
       continue;
     exprs.push_back(getAffineDimExpr(dim, context));
   }
   auto resultDimToSourceDimMap = AffineMap::get(sourceRank, 0, exprs, context);
-  return currPermutationMap.compose(resultDimToSourceDimMap);
+  return AffineMapAttr::get(
+      currPermutationMap.compose(resultDimToSourceDimMap));
 }
 
 //===----------------------------------------------------------------------===//
 // Patterns
 //===----------------------------------------------------------------------===//
 
 namespace {
 /// Merges subview operation with load/transferRead operation.
void LoadOpOfSubViewFolder<memref::LoadOp>::replaceOp(
     memref::LoadOp loadOp, memref::SubViewOp subViewOp,
     ArrayRef<Value> sourceIndices, PatternRewriter &rewriter) const {
   rewriter.replaceOpWithNewOp<memref::LoadOp>(loadOp, subViewOp.source(),
                                               sourceIndices);
 }
 
 template <>
 void LoadOpOfSubViewFolder<vector::TransferReadOp>::replaceOp(
-    vector::TransferReadOp loadOp, memref::SubViewOp subViewOp,
+    vector::TransferReadOp transferReadOp, memref::SubViewOp subViewOp,
     ArrayRef<Value> sourceIndices, PatternRewriter &rewriter) const {
+  // TODO: support 0-d corner case.
+  if (transferReadOp.getTransferRank() == 0)
+    return;
   rewriter.replaceOpWithNewOp<vector::TransferReadOp>(
-      loadOp, loadOp.getVectorType(), subViewOp.source(), sourceIndices,
-      getPermutationMap(rewriter.getContext(), subViewOp,
-                        loadOp.permutation_map()),
-      loadOp.padding(), loadOp.in_boundsAttr());
+      transferReadOp, transferReadOp.getVectorType(), subViewOp.source(),
+      sourceIndices,
+      getPermutationMapAttr(rewriter.getContext(), subViewOp,
+                            transferReadOp.permutation_map()),
+      transferReadOp.padding(),
+      /*mask=*/Value(), transferReadOp.in_boundsAttr());
 }
 
 template <>
 void StoreOpOfSubViewFolder<memref::StoreOp>::replaceOp(
     memref::StoreOp storeOp, memref::SubViewOp subViewOp,
     ArrayRef<Value> sourceIndices, PatternRewriter &rewriter) const {
   rewriter.replaceOpWithNewOp<memref::StoreOp>(
       storeOp, storeOp.value(), subViewOp.source(), sourceIndices);
 }
 
 template <>
 void StoreOpOfSubViewFolder<vector::TransferWriteOp>::replaceOp(
     vector::TransferWriteOp transferWriteOp, memref::SubViewOp subViewOp,
     ArrayRef<Value> sourceIndices, PatternRewriter &rewriter) const {
+  // TODO: support 0-d corner case.
+  if (transferWriteOp.getTransferRank() == 0)
+    return;
   rewriter.replaceOpWithNewOp<vector::TransferWriteOp>(
       transferWriteOp, transferWriteOp.vector(), subViewOp.source(),
       sourceIndices,
-      getPermutationMap(rewriter.getContext(), subViewOp,
+      getPermutationMapAttr(rewriter.getContext(), subViewOp,
                             transferWriteOp.permutation_map()),
       transferWriteOp.in_boundsAttr());
 }
 } // namespace
 
 template <typename OpTy>
 LogicalResult
 LoadOpOfSubViewFolder<OpTy>::matchAndRewrite(OpTy loadOp,
                                              PatternRewriter &rewriter) const {

mlir/lib/Dialect/Vector/VectorDropLeadUnitDim.cpp

 // vector.shape_cast followed by vector.transfer_read on vector without leading
 // 1 dimensions.
 struct CastAwayTransferReadLeadingOneDim
     : public OpRewritePattern<vector::TransferReadOp> {
   using OpRewritePattern::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::TransferReadOp read,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (read.getTransferRank() == 0)
+      return failure();
+
     if (read.mask())
       return failure();
 
     auto shapedType = read.source().getType().cast<ShapedType>();
     if (shapedType.getElementType() != read.getVectorType().getElementType())
       return failure();
 
     VectorType oldType = read.getVectorType();
     VectorType newType = trimLeadingOneDims(oldType);
 
     if (newType == oldType)
       return failure();
 
     AffineMap oldMap = read.permutation_map();
     ArrayRef<AffineExpr> newResults =
         oldMap.getResults().take_back(newType.getRank());
     AffineMap newMap =
         AffineMap::get(oldMap.getNumDims(), oldMap.getNumSymbols(), newResults,
                        rewriter.getContext());
 
-    ArrayAttr inBounds;
+    ArrayAttr inBoundsAttr;
     if (read.in_bounds())
-      inBounds = rewriter.getArrayAttr(
+      inBoundsAttr = rewriter.getArrayAttr(
           read.in_boundsAttr().getValue().take_back(newType.getRank()));
 
     auto newRead = rewriter.create<vector::TransferReadOp>(
-        read.getLoc(), newType, read.source(), read.indices(), newMap,
-        read.padding(), inBounds);
+        read.getLoc(), newType, read.source(), read.indices(),
+        AffineMapAttr::get(newMap), read.padding(), /*mask=*/Value(),
+        inBoundsAttr);
     rewriter.replaceOpWithNewOp<vector::BroadcastOp>(read, oldType, newRead);
 
     return success();
   }
 };
 
 // Turns vector.transfer_write on vector with leading 1 dimensions into
 // vector.shape_cast followed by vector.transfer_write on vector without leading
 // 1 dimensions.
 struct CastAwayTransferWriteLeadingOneDim
     : public OpRewritePattern<vector::TransferWriteOp> {
   using OpRewritePattern::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::TransferWriteOp write,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (write.getTransferRank() == 0)
+      return failure();
+
     if (write.mask())
       return failure();
 
     auto shapedType = write.source().getType().dyn_cast<ShapedType>();
     if (shapedType.getElementType() != write.getVectorType().getElementType())
       return failure();
 
     VectorType oldType = write.getVectorType();
     VectorType newType = trimLeadingOneDims(oldType);
     if (newType == oldType)
       return failure();
     int64_t dropDim = oldType.getRank() - newType.getRank();
 
     AffineMap oldMap = write.permutation_map();
     ArrayRef<AffineExpr> newResults =
         oldMap.getResults().take_back(newType.getRank());
     AffineMap newMap =
         AffineMap::get(oldMap.getNumDims(), oldMap.getNumSymbols(), newResults,
                        rewriter.getContext());
 
-    ArrayAttr inBounds;
+    ArrayAttr inBoundsAttr;
     if (write.in_bounds())
-      inBounds = rewriter.getArrayAttr(
+      inBoundsAttr = rewriter.getArrayAttr(
           write.in_boundsAttr().getValue().take_back(newType.getRank()));
 
     auto newVector = rewriter.create<vector::ExtractOp>(
         write.getLoc(), write.vector(), splatZero(dropDim));
     rewriter.replaceOpWithNewOp<vector::TransferWriteOp>(
-        write, newVector, write.source(), write.indices(), newMap, inBounds);
+        write, newVector, write.source(), write.indices(),
+        AffineMapAttr::get(newMap), inBoundsAttr);
 
     return success();
   }
 };
 
 class CastAwayElementwiseLeadingOneDim : public RewritePattern {
 public:
   CastAwayElementwiseLeadingOneDim(MLIRContext *context)

mlir/lib/Dialect/Vector/VectorOps.cpp

if (positionAttr.size() > static_cast<unsigned>(destVectorType.getRank()))
     return op.emitOpError(
         "expected position attribute of rank smaller than dest vector rank");
   auto srcVectorType = op.getSourceType().dyn_cast<VectorType>();
   if (srcVectorType &&
       (static_cast<unsigned>(srcVectorType.getRank()) + positionAttr.size() !=
        static_cast<unsigned>(destVectorType.getRank())))
     return op.emitOpError("expected position attribute rank + source rank to "
                           "match dest vector rank");
-  if (!srcVectorType && (positionAttr.size() !=
-                              static_cast<unsigned>(destVectorType.getRank())))
+  if (!srcVectorType &&
+      (positionAttr.size() != static_cast<unsigned>(destVectorType.getRank())))
     return op.emitOpError(
         "expected position attribute rank to match the dest vector rank");
   for (auto en : llvm::enumerate(positionAttr)) {
     auto attr = en.value().dyn_cast<IntegerAttr>();
     if (!attr || attr.getInt() < 0 ||
         attr.getInt() >= destVectorType.getDimSize(en.index()))
       return op.emitOpError("expected position attribute #")
              << (en.index() + 1)
void ExtractStridedSliceOp::getCanonicalizationPatterns(
   results.add<StridedSliceConstantMaskFolder, StridedSliceConstantFolder,
               StridedSliceBroadcast, StridedSliceSplat>(context);
 }
 
 //===----------------------------------------------------------------------===//
 // TransferReadOp
 //===----------------------------------------------------------------------===//
 
+/// 1. Builder that sets padding to zero and an empty mask (variant with attrs).
+void TransferReadOp::build(OpBuilder &builder, OperationState &result,
+                           VectorType vectorType, Value source,
+                           ValueRange indices, AffineMapAttr permutationMapAttr,
+                           /*optional*/ ArrayAttr inBoundsAttr) {
+  Type elemType = source.getType().cast<ShapedType>().getElementType();
+  Value padding = builder.create<arith::ConstantOp>(
+      result.location, elemType, builder.getZeroAttr(elemType));
+  build(builder, result, vectorType, source, indices, permutationMapAttr,
+        padding, /*mask=*/Value(), inBoundsAttr);
+}
+
+/// 2. Builder that sets padding to zero an empty mask (variant without attrs).
+void TransferReadOp::build(OpBuilder &builder, OperationState &result,
+                           VectorType vectorType, Value source,
+                           ValueRange indices, AffineMap permutationMap,
+                           Optional<ArrayRef<bool>> inBounds) {
+  auto permutationMapAttr = AffineMapAttr::get(permutationMap);
+  auto inBoundsAttr = (inBounds && !inBounds.getValue().empty())
+                          ? builder.getBoolArrayAttr(inBounds.getValue())
+                          : ArrayAttr();
+  build(builder, result, vectorType, source, indices, permutationMapAttr,
+        inBoundsAttr);
+}
+
+/// 3. Builder that sets permutation map to 'getMinorIdentityMap'.
+void TransferReadOp::build(OpBuilder &builder, OperationState &result,
+                           VectorType vectorType, Value source,
+                           ValueRange indices, Value padding,
+                           Optional<ArrayRef<bool>> inBounds) {
+  AffineMap permutationMap = getTransferMinorIdentityMap(
+      source.getType().cast<ShapedType>(), vectorType);
+  auto permutationMapAttr = AffineMapAttr::get(permutationMap);
+  auto inBoundsAttr = (inBounds && !inBounds.getValue().empty())
+                          ? builder.getBoolArrayAttr(inBounds.getValue())
+                          : ArrayAttr();
+  build(builder, result, vectorType, source, indices, permutationMapAttr,
+        padding,
+        /*mask=*/Value(), inBoundsAttr);
+}
+
+/// 4. Builder that sets padding to zero and permutation map to
+/// 'getMinorIdentityMap'.
+void TransferReadOp::build(OpBuilder &builder, OperationState &result,
+                           VectorType vectorType, Value source,
+                           ValueRange indices,
+                           Optional<ArrayRef<bool>> inBounds) {
+  Type elemType = source.getType().cast<ShapedType>().getElementType();
+  Value padding = builder.create<arith::ConstantOp>(
+      result.location, elemType, builder.getZeroAttr(elemType));
+  build(builder, result, vectorType, source, indices, padding, inBounds);
+}
+
 template <typename EmitFun>
 static LogicalResult verifyPermutationMap(AffineMap permutationMap,
                                           EmitFun emitOpError) {
   SmallVector<bool, 8> seen(permutationMap.getNumInputs(), false);
   for (auto expr : permutationMap.getResults()) {
     auto dim = expr.dyn_cast<AffineDimExpr>();
     auto zero = expr.dyn_cast<AffineConstantExpr>();
     if (zero) {
static LogicalResult verifyPermutationMap(AffineMap permutationMap,
   }
   return success();
 }
 
 static LogicalResult
 verifyTransferOp(VectorTransferOpInterface op, ShapedType shapedType,
                  VectorType vectorType, VectorType maskType,
                  AffineMap permutationMap, ArrayAttr inBounds) {
-  if (shapedType.getRank() == 0 && !op.isZeroD())
-    return op->emitOpError("0-d transfer requires vector<1xt> shape and () -> "
-                           "(0) permutation_map");
-
   if (op->hasAttr("masked")) {
     return op->emitOpError("masked attribute has been removed. "
                            "Use in_bounds instead.");
   }
 
   if (!shapedType.isa<MemRefType, RankedTensorType>())
     return op->emitOpError(
         "requires source to be a memref or ranked tensor type");
+
   auto elementType = shapedType.getElementType();
   DataLayout dataLayout = DataLayout::closest(op);
   if (auto vectorElementType = elementType.dyn_cast<VectorType>()) {
     // Memref or tensor has vector element type.
     unsigned sourceVecSize =
         dataLayout.getTypeSizeInBits(vectorElementType.getElementType()) *
         vectorElementType.getShape().back();
     unsigned resultVecSize =
if (auto vectorElementType = elementType.dyn_cast<VectorType>()) {
     if (permutationMap.getNumResults() != rankOffset)
       return op->emitOpError("requires a permutation_map with result dims of "
                              "the same rank as the vector type");
 
     if (maskType)
       return op->emitOpError("does not support masks with vector element type");
   } else {
     // Memref or tensor has scalar element type.
+    unsigned minorSize =
+        vectorType.getRank() == 0 ? 1 : vectorType.getShape().back();
     unsigned resultVecSize =
-        dataLayout.getTypeSizeInBits(vectorType.getElementType()) *
-        vectorType.getShape().back();
+        dataLayout.getTypeSizeInBits(vectorType.getElementType()) * minorSize;
     if (resultVecSize % dataLayout.getTypeSizeInBits(elementType) != 0)
       return op->emitOpError(
           "requires the bitwidth of the minor 1-D vector to be an integral "
           "multiple of the bitwidth of the source element type");
 
     // Check that permutation map results match rank of vector type.
     if (permutationMap.getNumResults() != vectorType.getRank())
       return op->emitOpError("requires a permutation_map with result dims of "
                              "the same rank as the vector type");
 
     VectorType expectedMaskType =
         vector::detail::transferMaskType(vectorType, permutationMap);
     if (maskType && expectedMaskType != maskType)
       return op->emitOpError("expects mask type consistent with permutation "
                              "map: ")
              << maskType;
   }
 
   if (permutationMap.getNumSymbols() != 0)
     return op->emitOpError("requires permutation_map without symbols");
-  // TODO: implement 0-d vector corner cases.
-  if (!op.isZeroD() && permutationMap.getNumInputs() != shapedType.getRank())
+
+  if (permutationMap.getNumInputs() != shapedType.getRank())
     return op->emitOpError("requires a permutation_map with input dims of the "
                            "same rank as the source type");
 
   if (inBounds) {
     if (permutationMap.getNumResults() != static_cast<int64_t>(inBounds.size()))
       return op->emitOpError("expects the optional in_bounds attr of same rank "
                              "as permutation_map results: ")
-             << AffineMapAttr::get(permutationMap);
+             << AffineMapAttr::get(permutationMap)
+             << " vs inBounds of size: " << inBounds.size();
     for (unsigned int i = 0; i < permutationMap.getNumResults(); ++i)
       if (permutationMap.getResult(i).isa<AffineConstantExpr>() &&
           !inBounds.getValue()[i].cast<BoolAttr>().getValue())
         return op->emitOpError("requires broadcast dimensions to be in-bounds");
   }
 
   return success();
 }
 
-/// Builder that sets padding to zero.
-void TransferReadOp::build(OpBuilder &builder, OperationState &result,
-                           VectorType vectorType, Value source,
-                           ValueRange indices, AffineMap permutationMap,
-                           ArrayRef<bool> inBounds) {
-  Type elemType = source.getType().cast<ShapedType>().getElementType();
-  Value padding = builder.create<arith::ConstantOp>(
-      result.location, elemType, builder.getZeroAttr(elemType));
-  if (inBounds.empty())
-    return build(builder, result, vectorType, source, indices, permutationMap,
-                 padding, ArrayAttr());
-  ArrayAttr inBoundsArrayAttr = builder.getBoolArrayAttr(inBounds);
-  build(builder, result, vectorType, source, indices, permutationMap, padding,
-        inBoundsArrayAttr);
-}
-
-/// Builder that sets permutation map to 'getMinorIdentityMap'.
-void TransferReadOp::build(OpBuilder &builder, OperationState &result,
-                           VectorType vectorType, Value source,
-                           ValueRange indices, Value padding,
-                           ArrayRef<bool> inBounds) {
-  auto permMap = getTransferMinorIdentityMap(
-      source.getType().cast<ShapedType>(), vectorType);
-  if (inBounds.empty())
-    return build(builder, result, vectorType, source, indices, permMap, padding,
-                 ArrayAttr());
-  ArrayAttr inBoundsArrayAttr = builder.getBoolArrayAttr(inBounds);
-  build(builder, result, vectorType, source, indices, permMap, padding,
-        inBoundsArrayAttr);
-}
-
-/// Builder that sets permutation map (resp. padding) to 'getMinorIdentityMap'
-/// (resp. zero).
-void TransferReadOp::build(OpBuilder &builder, OperationState &result,
-                           VectorType vectorType, Value source,
-                           ValueRange indices, ArrayRef<bool> inBounds) {
-  auto permMap = getTransferMinorIdentityMap(
-      source.getType().cast<ShapedType>(), vectorType);
-  build(builder, result, vectorType, source, indices, permMap, inBounds);
-}
-
-/// Builder that does not provide a mask.
-void TransferReadOp::build(OpBuilder &builder, OperationState &result,
-                           Type vectorType, Value source, ValueRange indices,
-                           AffineMap permutationMap, Value padding,
-                           ArrayAttr inBounds) {
-  build(builder, result, vectorType, source, indices, permutationMap, padding,
-        /*mask=*/Value(), inBounds);
-}
-
-/// Builder that does not provide a mask.
-void TransferReadOp::build(OpBuilder &builder, OperationState &result,
-                           Type vectorType, Value source, ValueRange indices,
-                           AffineMapAttr permutationMap, Value padding,
-                           ArrayAttr inBounds) {
-  build(builder, result, vectorType, source, indices, permutationMap, padding,
-        /*mask=*/Value(), inBounds);
-}
-
-Value TransferReadOp::createScalarOp(OpBuilder &builder, Location loc,
-                                     Value source, ValueRange indices,
-                                     ArrayRef<bool> inBounds) {
-  Type elemType = source.getType().cast<ShapedType>().getElementType();
-  auto vectorType = VectorType::get(ArrayRef<int64_t>{1}, elemType);
-  AffineMap map = AffineMap::get(/*numDims=*/0, /*numSymbols=*/0,
-                                 getAffineConstantExpr(0, loc.getContext()));
-  Value read = builder.create<vector::TransferReadOp>(loc, vectorType, source,
-                                                      indices, map, inBounds);
-  return builder.create<vector::ExtractOp>(loc, read, ArrayRef<int64_t>{0});
-}
-
 static void printTransferAttrs(OpAsmPrinter &p, VectorTransferOpInterface op) {
   SmallVector<StringRef, 3> elidedAttrs;
   elidedAttrs.push_back(TransferReadOp::getOperandSegmentSizeAttr());
   if (op.permutation_map().isMinorIdentity())
     elidedAttrs.push_back(op.getPermutationMapAttrName());
   bool elideInBounds = true;
   if (auto inBounds = op.in_bounds()) {
     for (auto attr : *inBounds) {
if (!shapedType || !shapedType.isa<MemRefType, RankedTensorType>())
     return parser.emitError(typesLoc, "requires memref or ranked tensor type");
   VectorType vectorType = types[1].dyn_cast<VectorType>();
   if (!vectorType)
     return parser.emitError(typesLoc, "requires vector type");
   auto permutationAttrName = TransferReadOp::getPermutationMapAttrName();
   Attribute mapAttr = result.attributes.get(permutationAttrName);
   if (!mapAttr) {
     auto permMap = getTransferMinorIdentityMap(shapedType, vectorType);
+    // Update `mapAttr` that is used later to determine mask type.
     mapAttr = AffineMapAttr::get(permMap);
     result.attributes.set(permutationAttrName, mapAttr);
   }
   if (parser.resolveOperand(sourceInfo, shapedType, result.operands) ||
       parser.resolveOperands(indexInfo, indexType, result.operands) ||
       parser.resolveOperand(paddingInfo, shapedType.getElementType(),
                             result.operands))
     return failure();
static bool isInBounds(TransferOp op, int64_t resultIdx, int64_t indicesIdx) {
   int64_t sourceSize = op.getShapedType().getDimSize(indicesIdx);
   int64_t vectorSize = op.getVectorType().getDimSize(resultIdx);
 
   return cstOp.value() + vectorSize <= sourceSize;
 }
 
 template <typename TransferOp>
 static LogicalResult foldTransferInBoundsAttribute(TransferOp op) {
-  // TODO: Be less conservative once we have 0-d vectors.
-  if (op.isZeroD())
+  // TODO: support 0-d corner case.
+  // TODO: Be less conservative.
+  if (op.getTransferRank() == 0)
     return failure();
   AffineMap permutationMap = op.permutation_map();
   bool changed = false;
   SmallVector<bool, 4> newInBounds;
   newInBounds.reserve(op.getTransferRank());
   for (unsigned i = 0; i < op.getTransferRank(); ++i) {
     // Already marked as in-bounds, nothing to see here.
     if (op.isDimInBounds(i)) {

 /// ```
 struct FoldExtractSliceIntoTransferRead
     : public OpRewritePattern<TransferReadOp> {
 public:
   using OpRewritePattern<TransferReadOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(TransferReadOp xferOp,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (xferOp.getTransferRank() == 0)
+      return failure();
     if (xferOp.hasOutOfBoundsDim())
       return failure();
     if (!xferOp.permutation_map().isIdentity())
       return failure();
     if (xferOp.mask())
       return failure();
     auto extractOp = xferOp.source().getDefiningOp<tensor::ExtractSliceOp>();
     if (!extractOp)
for (auto it : llvm::enumerate(xferOp.indices())) {
       OpFoldResult offset =
           extractOp.getMixedOffsets()[it.index() + rankReduced];
       newIndices.push_back(rewriter.create<arith::AddIOp>(
           xferOp->getLoc(), it.value(),
           getValueOrCreateConstantIndexOp(rewriter, extractOp.getLoc(),
                                           offset)));
     }
     SmallVector<bool> inBounds(xferOp.getTransferRank(), true);
-    rewriter.replaceOpWithNewOp<TransferReadOp>(xferOp, xferOp.getVectorType(),
-                                                extractOp.source(), newIndices,
-                                                xferOp.padding(), inBounds);
+    rewriter.replaceOpWithNewOp<TransferReadOp>(
+        xferOp, xferOp.getVectorType(), extractOp.source(), newIndices,
+        xferOp.padding(), ArrayRef<bool>{inBounds});
 
     return success();
   }
 };
 } // namespace
 
 void TransferReadOp::getCanonicalizationPatterns(RewritePatternSet &results,
                                                  MLIRContext *context) {
   results.add<FoldExtractSliceIntoTransferRead>(context);
 }
 
 //===----------------------------------------------------------------------===//
 // TransferWriteOp
 //===----------------------------------------------------------------------===//
 
+/// 1. Builder with type inference.
 void TransferWriteOp::build(OpBuilder &builder, OperationState &result,
                             Value vector, Value dest, ValueRange indices,
-                            AffineMap permutationMap, ArrayRef<bool> inBounds) {
-  if (inBounds.empty())
-    return build(builder, result, vector, dest, indices, permutationMap,
-                 /*mask=*/Value(), ArrayAttr());
-  build(builder, result, vector, dest, indices, permutationMap,
-        /*mask=*/Value(), builder.getBoolArrayAttr(inBounds));
-}
-
-/// Builder that sets permutation map to 'getMinorIdentityMap'.
-void TransferWriteOp::build(OpBuilder &builder, OperationState &result,
-                            Value vector, Value source, ValueRange indices,
-                            ArrayRef<bool> inBounds) {
-  auto vectorType = vector.getType().cast<VectorType>();
-  auto permMap = getTransferMinorIdentityMap(
-      source.getType().cast<ShapedType>(), vectorType);
-  if (inBounds.empty())
-    return build(builder, result, vector, source, indices, permMap,
-                 ArrayAttr());
-  ArrayAttr inBoundsArrayAttr = builder.getBoolArrayAttr(inBounds);
-  build(builder, result, vector, source, indices, permMap, inBoundsArrayAttr);
+                            AffineMapAttr permutationMapAttr,
+                            /*optional*/ Value mask,
+                            /*optional*/ ArrayAttr inBoundsAttr) {
+  Type resultType = dest.getType().dyn_cast<RankedTensorType>();
+  build(builder, result, resultType, vector, dest, indices, permutationMapAttr,
+        mask, inBoundsAttr);
 }
 
+/// 2. Builder with type inference that sets an empty mask (variant with attrs).
 void TransferWriteOp::build(OpBuilder &builder, OperationState &result,
-                            Value vector, Value source, ValueRange indices,
-                            AffineMapAttr permutationMap,
-                            /*optional*/ ArrayAttr inBounds) {
-  Type resultType = source.getType().dyn_cast<RankedTensorType>();
-  build(builder, result, resultType, vector, source, indices, permutationMap,
-        /*mask=*/Value(), inBounds);
+                            Value vector, Value dest, ValueRange indices,
+                            AffineMapAttr permutationMapAttr,
+                            /*optional*/ ArrayAttr inBoundsAttr) {
+  build(builder, result, vector, dest, indices, permutationMapAttr,
+        /*mask=*/Value(), inBoundsAttr);
 }
 
+/// 3. Builder with type inference that sets an empty mask (variant without
+/// attrs)
 void TransferWriteOp::build(OpBuilder &builder, OperationState &result,
-                            Value vector, Value source, ValueRange indices,
+                            Value vector, Value dest, ValueRange indices,
                             AffineMap permutationMap,
-                            /*optional*/ ArrayAttr inBounds) {
-  Type resultType = source.getType().dyn_cast<RankedTensorType>();
-  build(builder, result, resultType, vector, source, indices, permutationMap,
-        /*mask=*/Value(), inBounds);
+                            Optional<ArrayRef<bool>> inBounds) {
+  auto permutationMapAttr = AffineMapAttr::get(permutationMap);
+  auto inBoundsAttr = (inBounds && !inBounds.getValue().empty())
+                          ? builder.getBoolArrayAttr(inBounds.getValue())
+                          : ArrayAttr();
+  build(builder, result, vector, dest, indices, permutationMapAttr,
+        /*mask=*/Value(), inBoundsAttr);
 }
 
+/// 4. Builder with type inference that sets an empty mask and sets permutation
+///    map to 'getMinorIdentityMap'.
 void TransferWriteOp::build(OpBuilder &builder, OperationState &result,
-                            Value vector, Value source, ValueRange indices,
-                            AffineMap permutationMap, /*optional*/ Value mask,
-                            /*optional*/ ArrayAttr inBounds) {
-  Type resultType = source.getType().dyn_cast<RankedTensorType>();
-  build(builder, result, resultType, vector, source, indices, permutationMap,
-        mask, inBounds);
-}
-
-Operation *TransferWriteOp::createScalarOp(OpBuilder &builder, Location loc,
-                                           Value value, Value dest,
-                                           ValueRange indices,
-                                           ArrayRef<bool> inBounds) {
-  Value vectorOfAScalar = value;
-  if (!value.getType().isa<VectorType>())
-    vectorOfAScalar = builder.create<vector::BroadcastOp>(
-        loc, VectorType::get({1}, value.getType()), value);
-  AffineMap map = AffineMap::get(/*numDims=*/0, /*numSymbols=*/0,
-                                 getAffineConstantExpr(0, loc.getContext()));
-  return builder.create<vector::TransferWriteOp>(loc, vectorOfAScalar, dest,
-                                                 indices, map, inBounds);
+                            Value vector, Value dest, ValueRange indices,
+                            Optional<ArrayRef<bool>> inBounds) {
+  auto vectorType = vector.getType().cast<VectorType>();
+  AffineMap permutationMap = getTransferMinorIdentityMap(
+      dest.getType().cast<ShapedType>(), vectorType);
+  build(builder, result, vector, dest, indices, permutationMap, inBounds);
 }
 
 static ParseResult parseTransferWriteOp(OpAsmParser &parser,
                                         OperationState &result) {
   auto &builder = parser.getBuilder();
   llvm::SMLoc typesLoc;
   OpAsmParser::OperandType vectorInfo, sourceInfo;
   SmallVector<OpAsmParser::OperandType, 8> indexInfo;

 ///    %t0
 /// ```
 ///
 /// The producer of t1 may or may not be DCE'd depending on whether it is a
 /// block argument or has side effects.
 static LogicalResult foldReadInitWrite(TransferWriteOp write,
                                        ArrayRef<Attribute>,
                                        SmallVectorImpl<OpFoldResult> &results) {
+  // TODO: support 0-d corner case.
+  if (write.getTransferRank() == 0)
+    return failure();
   auto rankedTensorType = write.source().getType().dyn_cast<RankedTensorType>();
   // If not operating on tensors, bail.
   if (!rankedTensorType)
     return failure();
   // If no read, bail.
   auto read = write.vector().getDefiningOp<vector::TransferReadOp>();
   if (!read)
     return failure();
+  // TODO: support 0-d corner case.
+  if (read.getTransferRank() == 0)
+    return failure();
   // For now, only accept minor identity. Future: composition is minor identity.
   if (!read.permutation_map().isMinorIdentity() ||
       !write.permutation_map().isMinorIdentity())
     return failure();
   // Bail on mismatching ranks.
   if (read.getTransferRank() != write.getTransferRank())
     return failure();
   // Bail on potential out-of-bounds accesses.
struct FoldInsertSliceIntoTransferWrite
     : public OpRewritePattern<tensor::InsertSliceOp> {
 public:
   using OpRewritePattern<tensor::InsertSliceOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(tensor::InsertSliceOp insertOp,
                                 PatternRewriter &rewriter) const override {
     if (!insertOp.hasUnitStride())
       return failure();
+
     auto xferOp = insertOp.source().getDefiningOp<TransferWriteOp>();
     if (!xferOp)
       return failure();
+    // TODO: support 0-d corner case.
+    if (xferOp.getTransferRank() == 0)
+      return failure();
+
     if (xferOp.hasOutOfBoundsDim())
       return failure();
     if (xferOp.getVectorType().getRank() != xferOp.getShapedType().getRank())
       return failure();
     if (xferOp.mask())
       return failure();
     // Fold only if the TransferWriteOp completely overwrites the `source` with
     // a vector. I.e., the result of the TransferWriteOp is a new tensor who's
     // content is the data of the vector.
     if (!llvm::equal(xferOp.getVectorType().getShape(),
                      xferOp.getShapedType().getShape()))
       return failure();
     if (!xferOp.permutation_map().isIdentity())
       return failure();
 
     SmallVector<Value> indices = getValueOrCreateConstantIndexOp(
         rewriter, insertOp.getLoc(), insertOp.getMixedOffsets());
     SmallVector<bool> inBounds(xferOp.getTransferRank(), true);
-    rewriter.replaceOpWithNewOp<TransferWriteOp>(
-        insertOp, xferOp.vector(), insertOp.dest(), indices, inBounds);
+    rewriter.replaceOpWithNewOp<TransferWriteOp>(insertOp, xferOp.vector(),
+                                                 insertOp.dest(), indices,
+                                                 ArrayRef<bool>{inBounds});
     return success();
   }
 };
 } // namespace
 
 void TransferWriteOp::getCanonicalizationPatterns(RewritePatternSet &results,
                                                   MLIRContext *context) {
   results.add<FoldWaw, FoldInsertSliceIntoTransferWrite>(context);

mlir/lib/Dialect/Vector/VectorTransferPermutationMapRewritePatterns.cpp

 transposeInBoundsAttr(OpBuilder &builder, ArrayAttr attr,
                       const SmallVector<unsigned> &permutation) {
   SmallVector<bool> newInBoundsValues;
   for (unsigned pos : permutation)
     newInBoundsValues.push_back(
         attr.getValue()[pos].cast<BoolAttr>().getValue());
   return builder.getBoolArrayAttr(newInBoundsValues);
 }
+
 /// Lower transfer_read op with permutation into a transfer_read with a
 /// permutation map composed of leading zeros followed by a minor identiy +
 /// vector.transpose op.
 /// Ex:
 ///     vector.transfer_read ...
 ///         permutation_map: (d0, d1, d2) -> (0, d1)
 /// into:
 ///     %v = vector.transfer_read ...

 /// Note that an alternative is to transform it to linalg.transpose +
 /// vector.transfer_read to do the transpose in memory instead.
 struct TransferReadPermutationLowering
     : public OpRewritePattern<vector::TransferReadOp> {
   using OpRewritePattern<vector::TransferReadOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::TransferReadOp op,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (op.getTransferRank() == 0)
+      return failure();
+
     SmallVector<unsigned> permutation;
     AffineMap map = op.permutation_map();
     if (map.getNumResults() == 0)
       return failure();
     if (!map.isPermutationOfMinorIdentityWithBroadcasting(permutation))
       return failure();
     AffineMap permutationMap =
         map.getPermutationMap(permutation, op.getContext());
if (op.mask()) {
           maskTransposeIndices.push_back(expr.getPosition());
       }
 
       newMask = rewriter.create<vector::TransposeOp>(op.getLoc(), op.mask(),
                                                      maskTransposeIndices);
     }
 
     // Transpose in_bounds attribute.
-    ArrayAttr newInBounds =
+    ArrayAttr newInBoundsAttr =
         op.in_bounds() ? transposeInBoundsAttr(
                              rewriter, op.in_bounds().getValue(), permutation)
                        : ArrayAttr();
 
     // Generate new transfer_read operation.
     VectorType newReadType =
         VectorType::get(newVectorShape, op.getVectorType().getElementType());
     Value newRead = rewriter.create<vector::TransferReadOp>(
-        op.getLoc(), newReadType, op.source(), op.indices(), newMap,
-        op.padding(), newMask, newInBounds);
+        op.getLoc(), newReadType, op.source(), op.indices(),
+        AffineMapAttr::get(newMap), op.padding(), newMask, newInBoundsAttr);
 
     // Transpose result of transfer_read.
     SmallVector<int64_t> transposePerm(permutation.begin(), permutation.end());
     rewriter.replaceOpWithNewOp<vector::TransposeOp>(op, newRead,
                                                      transposePerm);
     return success();
   }
 };

 ///     %v = vector.transfer_write %tmp ...
 ///         permutation_map: (d0, d1, d2, d3) -> (d2, d3)
 struct TransferWritePermutationLowering
     : public OpRewritePattern<vector::TransferWriteOp> {
   using OpRewritePattern<vector::TransferWriteOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::TransferWriteOp op,
                                 PatternRewriter &rewriter) const override {
-    if (op.isZeroD())
+    // TODO: support 0-d corner case.
+    if (op.getTransferRank() == 0)
       return failure();
 
     SmallVector<unsigned> permutation;
     AffineMap map = op.permutation_map();
     if (map.isMinorIdentity())
       return failure();
     if (!map.isPermutationOfMinorIdentityWithBroadcasting(permutation))
       return failure();
llvm::transform(comp.getResults(), std::back_inserter(indices),
                     });
 
     // Transpose mask operand.
     Value newMask = op.mask() ? rewriter.create<vector::TransposeOp>(
                                     op.getLoc(), op.mask(), indices)
                               : Value();
 
     // Transpose in_bounds attribute.
-    ArrayAttr newInBounds =
+    ArrayAttr newInBoundsAttr =
         op.in_bounds() ? transposeInBoundsAttr(
                              rewriter, op.in_bounds().getValue(), permutation)
                        : ArrayAttr();
 
     // Generate new transfer_write operation.
     Value newVec =
         rewriter.create<vector::TransposeOp>(op.getLoc(), op.vector(), indices);
     auto newMap = AffineMap::getMinorIdentityMap(
         map.getNumDims(), map.getNumResults(), rewriter.getContext());
     rewriter.replaceOpWithNewOp<vector::TransferWriteOp>(
-        op, Type(), newVec, op.source(), op.indices(), newMap, newMask,
-        newInBounds);
+        op, Type(), newVec, op.source(), op.indices(),
+        AffineMapAttr::get(newMap), newMask, newInBoundsAttr);
 
     return success();
   }
 };
 
 /// Lower transfer_read op with broadcast in the leading dimensions into
 /// transfer_read of lower rank + vector.broadcast.
 /// Ex: vector.transfer_read ...
 ///         permutation_map: (d0, d1, d2, d3) -> (0, d1, 0, d3)
 /// into:
 ///     %v = vector.transfer_read ...
 ///         permutation_map: (d0, d1, d2, d3) -> (d1, 0, d3)
 ///     vector.broadcast %v
 struct TransferOpReduceRank : public OpRewritePattern<vector::TransferReadOp> {
   using OpRewritePattern<vector::TransferReadOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::TransferReadOp op,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (op.getTransferRank() == 0)
+      return failure();
+
     AffineMap map = op.permutation_map();
     unsigned numLeadingBroadcast = 0;
     for (auto expr : map.getResults()) {
       auto dimExpr = expr.dyn_cast<AffineConstantExpr>();
       if (!dimExpr || dimExpr.getValue() != 0)
         break;
       numLeadingBroadcast++;
     }
LogicalResult matchAndRewrite(vector::TransferReadOp op,
     }
     SmallVector<int64_t> newShape = llvm::to_vector<4>(
         originalVecType.getShape().take_back(reducedShapeRank));
     // Vector rank cannot be zero. Handled by TransferReadToVectorLoadLowering.
     if (newShape.empty())
       return failure();
     VectorType newReadType =
         VectorType::get(newShape, originalVecType.getElementType());
-    ArrayAttr newInBounds =
+    ArrayAttr newInBoundsAttr =
         op.in_bounds()
             ? rewriter.getArrayAttr(
                   op.in_boundsAttr().getValue().take_back(reducedShapeRank))
             : ArrayAttr();
     Value newRead = rewriter.create<vector::TransferReadOp>(
-        op.getLoc(), newReadType, op.source(), op.indices(), newMap,
-        op.padding(), op.mask(), newInBounds);
+        op.getLoc(), newReadType, op.source(), op.indices(),
+        AffineMapAttr::get(newMap), op.padding(), op.mask(), newInBoundsAttr);
     rewriter.replaceOpWithNewOp<vector::BroadcastOp>(op, originalVecType,
                                                      newRead);
     return success();
   }
 };
 
 void mlir::vector::populateVectorTransferPermutationMapLoweringPatterns(
     RewritePatternSet &patterns) {
   patterns.add<TransferReadPermutationLowering,
                TransferWritePermutationLowering, TransferOpReduceRank>(
       patterns.getContext());
 }

mlir/lib/Dialect/Vector/VectorTransforms.cpp

 struct UnrollTransferReadPattern
     : public OpRewritePattern<vector::TransferReadOp> {
   UnrollTransferReadPattern(MLIRContext *context,
                             const vector::UnrollVectorOptions &options)
       : OpRewritePattern<vector::TransferReadOp>(context, /*benefit=*/1),
         options(options) {}
   LogicalResult matchAndRewrite(vector::TransferReadOp readOp,
                                 PatternRewriter &rewriter) const override {
-
+    // TODO: support 0-d corner case.
+    if (readOp.getTransferRank() == 0)
+      return failure();
     if (readOp.mask())
       return failure();
     auto targetShape = getTargetShape(options, readOp);
     if (!targetShape)
       return failure();
     auto sourceVectorType = readOp.getVectorType();
     SmallVector<int64_t, 4> strides(targetShape->size(), 1);
     Location loc = readOp.getLoc();
     ArrayRef<int64_t> originalSize = readOp.getVectorType().getShape();
     SmallVector<int64_t, 4> ratio = *shapeRatio(originalSize, *targetShape);
     // Compute shape ratio of 'shape' and 'sizes'.
     int64_t sliceCount = computeMaxLinearIndex(ratio);
     // Prepare the result vector;
     Value result = rewriter.create<arith::ConstantOp>(
         loc, sourceVectorType, rewriter.getZeroAttr(sourceVectorType));
     auto targetType =
         VectorType::get(*targetShape, sourceVectorType.getElementType());
     SmallVector<Value, 4> originalIndices(readOp.indices().begin(),
                                           readOp.indices().end());
     for (int64_t i = 0; i < sliceCount; i++) {
       SmallVector<Value, 4> indices =
           sliceTransferIndices(i, originalSize, *targetShape, originalIndices,
                                readOp.permutation_map(), loc, rewriter);
       auto slicedRead = rewriter.create<vector::TransferReadOp>(
-          loc, targetType, readOp.source(), indices, readOp.permutation_map(),
-          readOp.padding(),
-          readOp.in_bounds() ? *readOp.in_bounds() : ArrayAttr());
+          loc, targetType, readOp.source(), indices,
+          readOp.permutation_mapAttr(), readOp.padding(), readOp.mask(),
+          readOp.in_boundsAttr());
 
       SmallVector<int64_t, 4> elementOffsets =
           getVectorOffset(originalSize, *targetShape, i);
       result = rewriter.create<vector::InsertStridedSliceOp>(
           loc, slicedRead, result, elementOffsets, strides);
     }
     rewriter.replaceOp(readOp, result);
     return success();
   }
 
 private:
   vector::UnrollVectorOptions options;
 };
 
 struct UnrollTransferWritePattern
     : public OpRewritePattern<vector::TransferWriteOp> {
   UnrollTransferWritePattern(MLIRContext *context,
                              const vector::UnrollVectorOptions &options)
       : OpRewritePattern<vector::TransferWriteOp>(context, /*benefit=*/1),
         options(options) {}
   LogicalResult matchAndRewrite(vector::TransferWriteOp writeOp,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (writeOp.getTransferRank() == 0)
+      return failure();
+
     if (writeOp.mask())
       return failure();
     auto targetShape = getTargetShape(options, writeOp);
     if (!targetShape)
       return failure();
     auto sourceVectorType = writeOp.getVectorType();
     SmallVector<int64_t, 4> strides(targetShape->size(), 1);
     Location loc = writeOp.getLoc();
for (int64_t i = 0; i < sliceCount; i++) {
       Value slicedVector = rewriter.create<vector::ExtractStridedSliceOp>(
           loc, writeOp.vector(), elementOffsets, *targetShape, strides);
 
       SmallVector<Value, 4> indices =
           sliceTransferIndices(i, originalSize, *targetShape, originalIndices,
                                writeOp.permutation_map(), loc, rewriter);
       Operation *slicedWrite = rewriter.create<vector::TransferWriteOp>(
           loc, slicedVector, resultTensor ? resultTensor : writeOp.source(),
-          indices, writeOp.permutation_map(),
-          writeOp.in_bounds() ? *writeOp.in_bounds() : ArrayAttr());
+          indices, writeOp.permutation_mapAttr(), writeOp.in_boundsAttr());
       // For the tensor case update the destination for the next transfer write.
       if (!slicedWrite->getResults().empty())
         resultTensor = slicedWrite->getResult(0);
     }
     if (resultTensor)
       rewriter.replaceOp(writeOp, resultTensor);
     else
       rewriter.eraseOp(writeOp);

 ///
 /// Preconditions:
 ///  1. `xferOp.permutation_map()` must be a minor identity map
 ///  2. the rank of the `xferOp.memref()` and the rank of the `xferOp.vector()`
 ///  must be equal. This will be relaxed in the future but requires
 ///  rank-reducing subviews.
 static LogicalResult
 splitFullAndPartialTransferPrecondition(VectorTransferOpInterface xferOp) {
+  // TODO: support 0-d corner case.
+  if (xferOp.getTransferRank() == 0)
+    return failure();
+
   // TODO: expand support to these 2 cases.
   if (!xferOp.permutation_map().isMinorIdentity())
     return failure();
   // Must have some out-of-bounds dimension to be a candidate for splitting.
   if (!xferOp.hasOutOfBoundsDim())
     return failure();
   // Don't split transfer operations directly under IfOp, this avoids applying
   // the pattern recursively.

 ///   memref<64x64x64xf32>, vector<2x4x1xf32>
 /// ```
 struct TransferReadExtractPattern
     : public OpRewritePattern<vector::TransferReadOp> {
   TransferReadExtractPattern(MLIRContext *context)
       : OpRewritePattern<vector::TransferReadOp>(context) {}
   LogicalResult matchAndRewrite(vector::TransferReadOp read,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (read.getTransferRank() == 0)
+      return failure();
+
     if (!read.getResult().hasOneUse())
       return failure();
     auto extract =
         dyn_cast<vector::ExtractMapOp>(*read.getResult().getUsers().begin());
     if (!extract)
       return failure();
     if (read.mask())
       return failure();
for (auto it :
       unsigned vectorPos = std::get<1>(it).cast<AffineDimExpr>().getPosition();
       auto scale = getAffineConstantExpr(
           extract.getResultType().getDimSize(vectorPos), read.getContext());
       indices[indexPos] = makeComposedAffineApply(
           rewriter, read.getLoc(), d0 + scale * d1,
           {indices[indexPos], extract.ids()[idCount++]});
     }
     Value newRead = lb.create<vector::TransferReadOp>(
-        extract.getType(), read.source(), indices, read.permutation_map(),
-        read.padding(), read.in_boundsAttr());
+        extract.getType(), read.source(), indices, read.permutation_mapAttr(),
+        read.padding(), read.mask(), read.in_boundsAttr());
     Value dest = lb.create<arith::ConstantOp>(
         read.getType(), rewriter.getZeroAttr(read.getType()));
     newRead = lb.create<vector::InsertMapOp>(newRead, dest, extract.ids());
     rewriter.replaceOp(read, newRead);
     return success();
   }
 };
 
 struct TransferWriteInsertPattern
     : public OpRewritePattern<vector::TransferWriteOp> {
   TransferWriteInsertPattern(MLIRContext *context)
       : OpRewritePattern<vector::TransferWriteOp>(context) {}
   LogicalResult matchAndRewrite(vector::TransferWriteOp write,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (write.getTransferRank() == 0)
+      return failure();
+
     auto insert = write.vector().getDefiningOp<vector::InsertMapOp>();
     if (!insert)
       return failure();
     if (write.mask())
       return failure();
     SmallVector<Value, 4> indices(write.indices().begin(),
                                   write.indices().end());
     AffineMap indexMap = insert.map().compose(write.permutation_map());
for (auto it :
       auto scale = getAffineConstantExpr(
           insert.getSourceVectorType().getDimSize(vectorPos),
           write.getContext());
       indices[indexPos] =
           makeComposedAffineApply(rewriter, loc, d0 + scale * d1,
                                   {indices[indexPos], insert.ids()[idCount++]});
     }
     rewriter.create<vector::TransferWriteOp>(
-        loc, insert.vector(), write.source(), indices, write.permutation_map(),
-        write.in_boundsAttr());
+        loc, insert.vector(), write.source(), indices,
+        write.permutation_mapAttr(), write.in_boundsAttr());
     rewriter.eraseOp(write);
     return success();
   }
 };
 
 /// Progressive lowering of transfer_read. This pattern supports lowering of
 /// `vector.transfer_read` to a combination of `vector.load` and
 /// `vector.broadcast` if all of the following hold:
 /// - Stride of most minor memref dimension must be 1.
 /// - Out-of-bounds masking is not required.
 /// - If the memref's element type is a vector type then it coincides with the
 ///   result type.
 /// - The permutation map doesn't perform permutation (broadcasting is allowed).
 struct TransferReadToVectorLoadLowering
     : public OpRewritePattern<vector::TransferReadOp> {
   TransferReadToVectorLoadLowering(MLIRContext *context,
                                    llvm::Optional<unsigned> maxRank)
       : OpRewritePattern<vector::TransferReadOp>(context),
         maxTransferRank(maxRank) {}
 
   LogicalResult matchAndRewrite(vector::TransferReadOp read,
                                 PatternRewriter &rewriter) const override {
     if (maxTransferRank && read.getVectorType().getRank() > *maxTransferRank)
       return failure();
+
     SmallVector<unsigned, 4> broadcastedDims;
     // Permutations are handled by VectorToSCF or
     // populateVectorTransferPermutationMapLoweringPatterns.
+    // We let the 0-d corner case pass-through as it is supported.
     if (!read.permutation_map().isMinorIdentityWithBroadcasting(
             &broadcastedDims))
       return failure();
+
     auto memRefType = read.getShapedType().dyn_cast<MemRefType>();
     if (!memRefType)
       return failure();
+
     // Non-unit strides are handled by VectorToSCF.
     if (!vector::isLastMemrefDimUnitStride(memRefType))
       return failure();
 
     // If there is broadcasting involved then we first load the unbroadcasted
     // vector, and then broadcast it with `vector.broadcast`.
     ArrayRef<int64_t> vectorShape = read.getVectorType().getShape();
     SmallVector<int64_t, 4> unbroadcastedVectorShape(vectorShape.begin(),
                                                      vectorShape.end());
     for (unsigned i : broadcastedDims)
       unbroadcastedVectorShape[i] = 1;
     VectorType unbroadcastedVectorType = VectorType::get(
         unbroadcastedVectorShape, read.getVectorType().getElementType());
 
     // `vector.load` supports vector types as memref's elements only when the
     // resulting vector type is the same as the element type.
     auto memrefElTy = memRefType.getElementType();
     if (memrefElTy.isa<VectorType>() && memrefElTy != unbroadcastedVectorType)
       return failure();
+
     // Otherwise, element types of the memref and the vector must match.
     if (!memrefElTy.isa<VectorType>() &&
         memrefElTy != read.getVectorType().getElementType())
       return failure();
 
     // Out-of-bounds dims are handled by MaterializeTransferMask.
     if (read.hasOutOfBoundsDim())
       return failure();
LogicalResult matchAndRewrite(vector::TransferReadOp read,
     }
 
     return success();
   }
 
   llvm::Optional<unsigned> maxTransferRank;
 };
 
-/// Replace a scalar vector.load with a memref.load.
+/// Replace a 0-d vector.load with a memref.load + vector.broadcast.
+// TODO: we shouldn't cross the vector/scalar domains just for this
+// but atm we lack the infra to avoid it. Possible solutions include:
+// - go directly to LLVM + bitcast
+// - introduce a bitcast op and likely a new pointer dialect
+// - let memref.load/store additionally support the 0-d vector case
+// There are still deeper data layout issues lingering even in this
+// trivial case (for architectures for which this matters).
 struct VectorLoadToMemrefLoadLowering
     : public OpRewritePattern<vector::LoadOp> {
   using OpRewritePattern<vector::LoadOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::LoadOp loadOp,
                                 PatternRewriter &rewriter) const override {
     auto vecType = loadOp.getVectorType();
     if (vecType.getNumElements() != 1)
       return failure();
     auto memrefLoad = rewriter.create<memref::LoadOp>(
         loadOp.getLoc(), loadOp.base(), loadOp.indices());
-    rewriter.replaceOpWithNewOp<vector::BroadcastOp>(
-        loadOp, VectorType::get({1}, vecType.getElementType()), memrefLoad);
+    rewriter.replaceOpWithNewOp<vector::BroadcastOp>(loadOp, vecType,
+                                                     memrefLoad);
     return success();
   }
 };
 
-/// Replace a scalar vector.store with a memref.store.
+/// Replace a 0-d vector.store with a vector.extractelement + memref.store.
 struct VectorStoreToMemrefStoreLowering
     : public OpRewritePattern<vector::StoreOp> {
   using OpRewritePattern<vector::StoreOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::StoreOp storeOp,
                                 PatternRewriter &rewriter) const override {
     auto vecType = storeOp.getVectorType();
     if (vecType.getNumElements() != 1)
       return failure();
+    Value extracted;
+    if (vecType.getRank() == 0) {
+      // TODO: Unifiy once ExtractOp supports 0-d vectors.
+      extracted = rewriter.create<vector::ExtractElementOp>(
+          storeOp.getLoc(), storeOp.valueToStore());
+    } else {
       SmallVector<int64_t> indices(vecType.getRank(), 0);
-    Value extracted = rewriter.create<vector::ExtractOp>(
+      extracted = rewriter.create<vector::ExtractOp>(
           storeOp.getLoc(), storeOp.valueToStore(), indices);
+    }
+
     rewriter.replaceOpWithNewOp<memref::StoreOp>(
         storeOp, extracted, storeOp.base(), storeOp.indices());
     return success();
   }
 };
 
 /// Progressive lowering of transfer_write. This pattern supports lowering of
 /// `vector.transfer_write` to `vector.store` if all of the following hold:
TransferWriteToVectorStoreLowering(MLIRContext *context,
                                      llvm::Optional<unsigned> maxRank)
       : OpRewritePattern<vector::TransferWriteOp>(context),
         maxTransferRank(maxRank) {}
 
   LogicalResult matchAndRewrite(vector::TransferWriteOp write,
                                 PatternRewriter &rewriter) const override {
     if (maxTransferRank && write.getVectorType().getRank() > *maxTransferRank)
       return failure();
+
     // Permutations are handled by VectorToSCF or
     // populateVectorTransferPermutationMapLoweringPatterns.
-    if (!write.isZeroD() && !write.permutation_map().isMinorIdentity())
+    if ( // pass-through for the 0-d corner case.
+        !write.permutation_map().isMinorIdentity())
       return failure();
+
     auto memRefType = write.getShapedType().dyn_cast<MemRefType>();
     if (!memRefType)
       return failure();
+
     // Non-unit strides are handled by VectorToSCF.
     if (!vector::isLastMemrefDimUnitStride(memRefType))
       return failure();
+
     // `vector.store` supports vector types as memref's elements only when the
     // type of the vector value being written is the same as the element type.
     auto memrefElTy = memRefType.getElementType();
     if (memrefElTy.isa<VectorType>() && memrefElTy != write.getVectorType())
       return failure();
+
     // Otherwise, element types of the memref and the vector must match.
     if (!memrefElTy.isa<VectorType>() &&
         memrefElTy != write.getVectorType().getElementType())
       return failure();
+
     // Out-of-bounds dims are handled by MaterializeTransferMask.
     if (write.hasOutOfBoundsDim())
       return failure();
     if (write.mask()) {
       rewriter.replaceOpWithNewOp<vector::MaskedStoreOp>(
           write, write.source(), write.indices(), write.mask(), write.vector());
     } else {
       rewriter.replaceOpWithNewOp<vector::StoreOp>(

 };
 
 // Drop inner most contiguous unit dimensions from transfer_read operand.
 class DropInnerMostUnitDims : public OpRewritePattern<vector::TransferReadOp> {
   using OpRewritePattern<vector::TransferReadOp>::OpRewritePattern;
 
   LogicalResult matchAndRewrite(vector::TransferReadOp readOp,
                                 PatternRewriter &rewriter) const override {
+    // TODO: support 0-d corner case.
+    if (readOp.getTransferRank() == 0)
+      return failure();
+
+    // TODO: support mask.
+    if (readOp.mask())
+      return failure();
+
     auto srcType = readOp.source().getType().dyn_cast<MemRefType>();
     if (!srcType || !srcType.hasStaticShape())
       return failure();
 
     if (!readOp.permutation_map().isMinorIdentity())
       return failure();
 
     auto targetType = readOp.getVectorType();
if (srcType.getLayout().getAffineMap().isIdentity()) {
           srcType.getShape().drop_back(dimsToDrop), srcType.getElementType(),
           map, srcType.getMemorySpaceAsInt());
     }
 
     auto loc = readOp.getLoc();
     SmallVector<int64_t> offsets(srcType.getRank(), 0);
     SmallVector<int64_t> strides(srcType.getRank(), 1);
 
-    ArrayAttr inBounds =
+    ArrayAttr inBoundsAttr =
         readOp.in_bounds()
             ? rewriter.getArrayAttr(
                   readOp.in_boundsAttr().getValue().drop_back(dimsToDrop))
             : ArrayAttr();
     Value rankedReducedView = rewriter.create<memref::SubViewOp>(
         loc, resultMemrefType, readOp.source(), offsets, srcType.getShape(),
         strides);
     auto permMap = getTransferMinorIdentityMap(
         rankedReducedView.getType().cast<ShapedType>(), resultTargetVecType);
     Value result = rewriter.create<vector::TransferReadOp>(
         loc, resultTargetVecType, rankedReducedView,
-        readOp.indices().drop_back(dimsToDrop), permMap, readOp.padding(),
-        inBounds);
+        readOp.indices().drop_back(dimsToDrop), AffineMapAttr::get(permMap),
+        readOp.padding(),
+        // TODO: support mask.
+        /*mask=*/Value(), inBoundsAttr);
     rewriter.replaceOpWithNewOp<vector::ShapeCastOp>(readOp, targetType,
                                                      result);
     return success();
   }
 };
 
 void mlir::vector::populateVectorMaskMaterializationPatterns(
     RewritePatternSet &patterns, bool indexOptimizations) {

mlir/lib/Interfaces/VectorInterfaces.cpp

VectorType mlir::vector::detail::transferMaskType(VectorType vecType,
   auto i1Type = IntegerType::get(map.getContext(), 1);
   SmallVector<int64_t, 8> shape;
   for (int64_t i = 0; i < vecType.getRank(); ++i) {
     // Only result dims have a corresponding dim in the mask.
     if (map.getResult(i).template isa<AffineDimExpr>()) {
       shape.push_back(vecType.getDimSize(i));
     }
   }
-  return shape.empty() ? VectorType() : VectorType::get(shape, i1Type);
+  return VectorType::get(shape, i1Type);
 }
 
 //===----------------------------------------------------------------------===//
 // VectorUnroll Interfaces
 //===----------------------------------------------------------------------===//
 
 /// Include the definitions of the VectorUnroll interfaces.
 #include "mlir/Interfaces/VectorInterfaces.cpp.inc"

mlir/test/Conversion/VectorToSCF/vector-to-scf.mlir

 // RUN: mlir-opt %s -convert-vector-to-scf -split-input-file -allow-unregistered-dialect | FileCheck %s
 // RUN: mlir-opt %s -convert-vector-to-scf=full-unroll=true -split-input-file -allow-unregistered-dialect | FileCheck %s --check-prefix=FULL-UNROLL
 
 // CHECK-LABEL: func @vector_transfer_ops_0d(
-//  CHECK-SAME:   %[[MEM:.*]]: memref<f32>) {
 func @vector_transfer_ops_0d(%M: memref<f32>) {
-    %f0 = arith.constant 0.0 : f32
 
-//  CHECK: %[[V0:.*]] = arith.constant dense<0{{.*}}> : vector<1xf32>
-//  CHECK: %[[R0:.*]] = scf.for %[[I:.*]] = {{.*}} iter_args(%[[V0_ITER:.*]] = %[[V0]]) -> (vector<1xf32>) {
-//  CHECK:   %[[S:.*]] = memref.load %[[MEM]][] : memref<f32>
-//  CHECK:   %[[R_ITER:.*]] = vector.insertelement %[[S]], %[[V0_ITER]][%[[I]] : index] : vector<1xf32>
-//  CHECK:   scf.yield %[[R_ITER]] : vector<1xf32>
-    %0 = vector.transfer_read %M[], %f0 {permutation_map = affine_map<()->(0)>} :
-      memref<f32>, vector<1xf32>
-
-//  CHECK: scf.for %[[J:.*]] = %{{.*}}
-//  CHECK:   %[[SS:.*]] = vector.extractelement %[[R0]][%[[J]] : index] : vector<1xf32>
-//  CHECK:   memref.store %[[SS]], %[[MEM]][] : memref<f32>
-    vector.transfer_write %0, %M[] {permutation_map = affine_map<()->(0)>} :
-      vector<1xf32>, memref<f32>
+  // 0-d transfers are left untouched by vector-to-scf.
+  // They are independently lowered to the proper memref.load/store.
+  //  CHECK: vector.transfer_read {{.*}}: memref<f32>, vector<f32>
+  %0 = vector.transfer_read %M[], %f0 {permutation_map = affine_map<()->()>} :
+    memref<f32>, vector<f32>
+
+  //  CHECK: vector.transfer_write {{.*}}: vector<f32>, memref<f32>
+  vector.transfer_write %0, %M[] {permutation_map = affine_map<()->()>} :
+    vector<f32>, memref<f32>
 
-    return
 }
 
 // -----
 
 // CHECK-LABEL: func @materialize_read_1d() {
 func @materialize_read_1d() {
   %f0 = arith.constant 0.0: f32
   %A = memref.alloc () : memref<7x42xf32>

mlir/test/Dialect/Linalg/vectorization.mlir

func @test_vectorize_fill(%A : memref<8x16xf32>, %arg0 : f32) {
   return
 }
 
 // -----
 
 // CHECK-LABEL: func @test_vectorize_fill
 func @test_vectorize_fill_scalar(%A : memref<f32>, %arg0 : f32) {
   // CHECK-SAME: (%[[M:.*]]: memref<f32>, %[[val:.*]]: f32)
-  //      CHECK:   %[[VEC:.*]] = vector.broadcast %[[val]] : f32 to vector<1xf32>
-  //      CHECK:   vector.transfer_write %[[VEC]], %[[M]][] {{.*}} : vector<1xf32>, memref<f32>
+  //      CHECK:   %[[VEC:.*]] = vector.broadcast %[[val]] : f32 to vector<f32>
+  //      CHECK:   vector.transfer_write %[[VEC]], %[[M]][] : vector<f32>, memref<f32>
   linalg.fill(%arg0, %A) : f32, memref<f32>
   return
 }
 
 // -----
 
 // CHECK-LABEL: func @test_vectorize_copy
 func @test_vectorize_copy(%A : memref<8x16xf32>, %B : memref<8x16xf32>) {
   //       CHECK: %[[V:.*]] = vector.transfer_read {{.*}} : memref<8x16xf32>, vector<8x16xf32>
   //       CHECK: vector.transfer_write %[[V]], {{.*}} : vector<8x16xf32>, memref<8x16xf32>
   linalg.copy(%A, %B) :  memref<8x16xf32>, memref<8x16xf32>
   return
 }
 
 // -----
 
 // CHECK-LABEL: func @test_vectorize_copy_scalar
 func @test_vectorize_copy_scalar(%A : memref<f32>, %B : memref<f32>) {
   //  CHECK-SAME: (%[[A:.*]]: memref<f32>, %[[B:.*]]: memref<f32>)
-  //       CHECK:   %[[V:.*]] = vector.transfer_read %[[A]][]{{.*}} : memref<f32>, vector<1xf32>
-  //       CHECK:   %[[val:.*]] = vector.extract %[[V]][0] : vector<1xf32>
-  //       CHECK:   %[[VV:.*]] = vector.broadcast %[[val]] : f32 to vector<1xf32>
-  //       CHECK:   vector.transfer_write %[[VV]], %[[B]][] {{.*}} : vector<1xf32>, memref<f32>
+  //       CHECK:   %[[V:.*]] = vector.transfer_read %[[A]][]{{.*}} : memref<f32>, vector<f32>
+  //       CHECK:   %[[val:.*]] = vector.extractelement %[[V]][] : vector<f32>
+  //       CHECK:   %[[VV:.*]] = vector.broadcast %[[val]] : f32 to vector<f32>
+  //       CHECK:   vector.transfer_write %[[VV]], %[[B]][] : vector<f32>, memref<f32>
   linalg.copy(%A, %B) :  memref<f32>, memref<f32>
   return
 }
 
 // -----
 
 // CHECK-LABEL: func @test_vectorize_trailing_index
   //  CHECK-SAME: (%[[ARG0:.*]]: memref<1x2x4x8xindex>)
func @fused_broadcast_red_2d(%arg0: tensor<4x4xf32>, %arg1: tensor<4x1xf32>) -> tensor<4xf32> {
   return %red : tensor<4xf32>
 }
 
 // -----
 
 //  CHECK-LABEL: func @reduce_1d(
 //   CHECK-SAME:   %[[A:.*]]: tensor<32xf32>
 func @reduce_1d(%arg0: tensor<32xf32>) -> tensor<f32> {
-  //  CHECK-DAG: %[[F0_v1:.*]] = arith.constant dense<0.000000e+00> : vector<1xf32>
+  //  CHECK-DAG: %[[vF0:.*]] = arith.constant dense<0.000000e+00> : vector<f32>
   //  CHECK-DAG: %[[F0:.*]] = arith.constant 0.000000e+00 : f32
   //  CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
   %f0 = arith.constant 0.000000e+00 : f32
 
   //      CHECK: %[[init:.*]] = linalg.init_tensor [] : tensor<f32>
   %0 = linalg.init_tensor [] : tensor<f32>
 
-  //      CHECK: %[[f:.*]] = vector.transfer_write %[[F0_v1]], %[[init]][]
-  // CHECK-SAME:   : vector<1xf32>, tensor<f32>
+  //      CHECK: %[[f:.*]] = vector.transfer_write %[[vF0]], %[[init]][]
+  // CHECK-SAME:   : vector<f32>, tensor<f32>
   %1 = linalg.fill(%f0, %0) : f32, tensor<f32> -> tensor<f32>
   //      CHECK: %[[r:.*]] = vector.transfer_read %[[A]][%[[C0]]]
   // CHECK-SAME:   : tensor<32xf32>, vector<32xf32>
+  //      CHECK: %[[f0:.*]] = vector.extractelement %[[vF0]][] : vector<f32>
   //      CHECK: %[[red:.*]] = vector.multi_reduction #vector.kind<add>, %[[r]] [0]
   // CHECK-SAME:   : vector<32xf32> to f32
-  //      CHECK: %[[a:.*]] = arith.addf %[[red]], %[[F0]] : f32
-  //      CHECK: %[[red_v1:.*]] = vector.broadcast %[[a]] : f32 to vector<1xf32>
+  //      CHECK: %[[a:.*]] = arith.addf %[[red]], %[[f0]] : f32
+  //      CHECK: %[[red_v1:.*]] = vector.broadcast %[[a]] : f32 to vector<f32>
   //      CHECK: %[[res:.*]] = vector.transfer_write %[[red_v1]], %[[f]][]
-  // CHECK-SAME:   : vector<1xf32>, tensor<f32>
+  // CHECK-SAME:   : vector<f32>, tensor<f32>
   %2 = linalg.generic {
          indexing_maps = [affine_map<(d0) -> (d0)>,
                           affine_map<(d0) -> ()>],
          iterator_types = ["reduction"]}
          ins(%arg0 : tensor<32xf32>)
          outs(%1 : tensor<f32>) {
     ^bb0(%a: f32, %b: f32):  // no predecessors
       %3 = arith.addf %a, %b : f32
       linalg.yield %3 : f32
     } -> tensor<f32>
 
   return %2 : tensor<f32>
 }

mlir/test/Dialect/Vector/invalid.mlir

 
 // -----
 
 func @insert_map_id(%v: vector<2x1xf32>, %v1: vector<4x32xf32>, %id : index) {
   // expected-error@+1 {{'vector.insert_map' op expected number of ids must match the number of dimensions distributed}}
   %0 = vector.insert_map %v, %v1[%id] : vector<2x1xf32> into vector<4x32xf32>
 }
 
-// -----
-
-func @vector_transfer_ops_0d(%arg0: tensor<f32>)
-  -> tensor<f32> {
-    %f0 = arith.constant 0.0 : f32
-    // expected-error@+1 {{0-d transfer requires vector<1xt> shape and () -> (0) permutation_map}}
-    %0 = vector.transfer_read %arg0[], %f0 {permutation_map = affine_map<(d0)->(d0)>} :
-      tensor<f32>, vector<1xf32>
-    %1 = vector.transfer_write %0, %arg0[] {permutation_map = affine_map<()->(0)>} :
-      vector<1xf32>, tensor<f32>
-    return %1: tensor<f32>
-}

mlir/test/Dialect/Vector/ops.mlir

 // RUN: mlir-opt %s | mlir-opt | FileCheck %s
 
 // CHECK-LABEL: func @vector_transfer_ops_0d(
 func @vector_transfer_ops_0d(%arg0: tensor<f32>, %arg1: memref<f32>)
   -> tensor<f32> {
     %f0 = arith.constant 0.0 : f32
-    %0 = vector.transfer_read %arg0[], %f0 {permutation_map = affine_map<()->(0)>} :
-      tensor<f32>, vector<1xf32>
-    %1 = vector.transfer_write %0, %arg0[] {permutation_map = affine_map<()->(0)>} :
-      vector<1xf32>, tensor<f32>
-    %2 = vector.transfer_read %arg1[], %f0 {permutation_map = affine_map<()->(0)>} :
-      memref<f32>, vector<1xf32>
-    vector.transfer_write %2, %arg1[] {permutation_map = affine_map<()->(0)>} :
-      vector<1xf32>, memref<f32>
+    %0 = vector.transfer_read %arg0[], %f0 {permutation_map = affine_map<()->()>} :
+      tensor<f32>, vector<f32>
+    %1 = vector.transfer_write %0, %arg0[] {permutation_map = affine_map<()->()>} :
+      vector<f32>, tensor<f32>
+    %2 = vector.transfer_read %arg1[], %f0 {permutation_map = affine_map<()->()>} :
+      memref<f32>, vector<f32>
+    vector.transfer_write %2, %arg1[] {permutation_map = affine_map<()->()>} :
+      vector<f32>, memref<f32>
     return %1: tensor<f32>
 }
 
+// CHECK-LABEL: func @vector_transfer_ops_0d_from_higher_d(
+func @vector_transfer_ops_0d_from_higher_d(%arg0: tensor<?xf32>, %arg1: memref<?x?xf32>)
+  -> tensor<?xf32> {
+    %c0 = arith.constant 0 : index
+    %f0 = arith.constant 0.0 : f32
+    %0 = vector.transfer_read %arg0[%c0], %f0 {permutation_map = affine_map<(d0)->()>} :
+      tensor<?xf32>, vector<f32>
+    %1 = vector.transfer_write %0, %arg0[%c0] {permutation_map = affine_map<(d0)->()>} :
+      vector<f32>, tensor<?xf32>
+    %2 = vector.transfer_read %arg1[%c0, %c0], %f0 {permutation_map = affine_map<(d0, d1)->()>} :
+      memref<?x?xf32>, vector<f32>
+    vector.transfer_write %2, %arg1[%c0, %c0] {permutation_map = affine_map<(d0, d1)->()>} :
+      vector<f32>, memref<?x?xf32>
+    return %1: tensor<?xf32>
+}
+
 // CHECK-LABEL: func @vector_transfer_ops(
 func @vector_transfer_ops(%arg0: memref<?x?xf32>,
                           %arg1 : memref<?x?xvector<4x3xf32>>,
                           %arg2 : memref<?x?xvector<4x3xi32>>,
                           %arg3 : memref<?x?xvector<4x3xindex>>,
                           %arg4 : memref<?x?x?xf32>) {
   // CHECK: %[[C3:.*]] = arith.constant 3 : index
   %c3 = arith.constant 3 : index

mlir/test/Dialect/Vector/vector-transfer-to-vector-load-store.mlir

 // RUN: mlir-opt %s -test-vector-transfer-lowering-patterns -canonicalize -split-input-file | FileCheck %s
 
 // CHECK-LABEL: func @vector_transfer_ops_0d_memref(
 //  CHECK-SAME:   %[[MEM:.*]]: memref<f32>
 //  CHECK-SAME:   %[[VV:.*]]: vector<1x1x1xf32>
 func @vector_transfer_ops_0d_memref(%M: memref<f32>, %v: vector<1x1x1xf32>) {
     %f0 = arith.constant 0.0 : f32
 
-//  CHECK-NEXT:   %[[V:.*]] = memref.load %[[MEM]][] : memref<f32>
-    %0 = vector.transfer_read %M[], %f0 {permutation_map = affine_map<()->(0)>} :
-      memref<f32>, vector<1xf32>
-
-//  CHECK-NEXT:   memref.store %[[V]], %[[MEM]][] : memref<f32>
-    vector.transfer_write %0, %M[] {permutation_map = affine_map<()->(0)>} :
-      vector<1xf32>, memref<f32>
+//  CHECK-NEXT:   %[[s:.*]] = memref.load %[[MEM]][] : memref<f32>
+//  CHECK-NEXT:   %[[V:.*]] = vector.broadcast %[[s]] : f32 to vector<f32>
+    %0 = vector.transfer_read %M[], %f0 : memref<f32>, vector<f32>
+
+//  CHECK-NEXT:   %[[ss:.*]] = vector.extractelement %[[V]][] : vector<f32>
+//  CHECK-NEXT:   memref.store %[[ss]], %[[MEM]][] : memref<f32>
+    vector.transfer_write %0, %M[] : vector<f32>, memref<f32>
 
 //  CHECK-NEXT:   %[[VV:.*]] = vector.extract %arg1[0, 0, 0] : vector<1x1x1xf32>
 //  CHECK-NEXT:   memref.store %[[VV]], %[[MEM]][] : memref<f32>
     vector.store %v, %M[] : memref<f32>, vector<1x1x1xf32>
 
     return
 }