[mlir][Linalg] Revisit the Linalg on tensors abstraction

[nicolasvasilache]

Summary

This revision drops init_tensor arguments from Linalg on tensors and instead uniformizes the output buffers and output tensors to be consistent.
This significantly simplifies the usage of Linalg on tensors and is a stepping stone for
its evolution towards a mixed tensor and shape abstraction discussed in https://llvm.discourse.group/t/linalg-and-shapes/2421/19.

[nicolasvasilache]

Finish the impl.

[nicolasvasilache]

Rebase

[nicolasvasilache]

Drop debug spew.

[nicolasvasilache]

Rebase

[ftynse]

mlir/docs/Dialects/Linalg.md

24

couldn't parse this

25–31

Something went wrong with formatting here.

55

I would have provided at least a summary of the discussion here.

61

67

"tensor" and "write-only" don't belong in the same phrase IMO. Did you mean that there is no data that is expected to be read from the tensor?

205–231

Something wrong happened to the formatting here. Generally, the overall reformat of the document makes it hard to understand what changed.

356–360

bad formatting

376–380

bad formatting

399–413

Even more bad formatting

428–438

and more

507

we have this in langref now

[ftynse]

mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td

115

Nit: getOperands().slice(0,1)?

312–314

The comment mentions nWin but the check is for nPar.

mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOpsInterface.td

148–160

Can we than declare inputs and outputs as interface methods without default implementation? I expect the compiler/linker to complain if they are missing.

318

318

343–347

Nit: isa+cast is an anti-pattern.

413

574–576

I don't see how checking for the number of regions safeguards against manually-defined ops. Do they have no regions?

Also, please add a message into assert.

[ftynse]

This looks mostly mechanical, LGTM after the comments are addressed.

mlir/lib/Dialect/Linalg/IR/LinalgOps.cpp

399–400

This will not catch the case where the ops has N output _buffers_ and N results. Is this expected?

431

This prints the entire type rather than just rank

447–448

The check above (op->getNumResults() > linalgOp.getNumOutputs()) (with my comment) looks redundant to this.

508–509

Nit: I think we use "region arguments" to refer to the arguments of the entry block of the region in LangRef, it's simpler than "region basic block arguments".

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

26

Spurious semicolon

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

348

Spurious semicolon

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

28–38

I'm not convinced this always does the right thing, could you add an explanation as to why it is sufficient to always take the first value with matching type or the first operand?

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

621

It looks weird to call to_vector<4> only to assign the result to SmallVector<, 1>

mlir/test/Dialect/Linalg/invalid.mlir

308–318

Could you elaborate what is the current order and the order that would be required for this to work? The same information (or a reference here) from the assertion is also helpful to debug the cases where it triggers.

[nicolasvasilache]

Rebase on a better Linalg.md

[nicolasvasilache]

mlir/docs/Dialects/Linalg.md

55

it's actually described below in init tensor / shape-only discussion and redundant.
Removed it from here.

205–231

yeah really sorry about that .. I landed a reformat RFC and rebased on top of it, should be better now.

[nicolasvasilache]

mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOpsInterface.td

343–347

Surprisingly annoying to swap map_range and make_filter_range.
Leaving as 2021's problem.

[nicolasvasilache]

Address.

[nicolasvasilache]

clang-tidy

[nicolasvasilache]

Apply fix.

[pifon2a]

Thank you, Nicolas!

[aartbik]

LGTM, with a minor suggestion for unnamed output parameter

mlir/docs/Dialects/Linalg.md

57

For the shape-only case, would it make sense to really just define the shape instead, and not rely on a "dummy" named output argument?

60

nit: built-in

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

470

Thanks for updating this file too! You may have a few merge conflicts due to my latest check in but they should be minor.

543

I suspect this can be simplified now, but I can do that as a follow up

mlir/test/Dialect/Linalg/sparse_1d.mlir

36

e.g. here, arga is really not the output, but it just happens to define the right shape?

[aartbik]

mlir/test/Dialect/Linalg/sparse_1d.mlir

37

here and below, should %s be %argb?

or, otherwise, what does it mean to have an unused scalar incoming argument, and another one that is defined outside the kernel? I see it is also done for some other kernels without scalar invariants?

mlir/docs/Dialects/Linalg.md

 ## Set of Key Transformations<a name="key_transformations"></a>
 
 The following key transformations have been central to driving the design of
 Linalg. They are all implemented in terms of the properties of the
 `linalg.generic` OpInterface and avoid the pitfall of relying on hardcoded
 one-off op knowledge.
 
 The textual form description of these transformations is left for future work.
-Still, it is useful to at least the key transformations that are performed on
-the Linalg IR and that have influenced its design:
+Still, it is useful to list the key transformations that are performed on the
+Linalg IR and that have influenced its design:
 
 1.  Progressive Buffer Allocation.
 1.  Parametric Tiling.
 1.  Promotion to Temporary Buffer in Fast Memory.
 1.  Tiled Producer-Consumer Fusion with Parametric Tile-And-Fuse.
 1.  Map to Parallel and Reduction Loops and Hardware.
 1.  Vectorization: Rewrite in Vector Form.
 1.  Lower to Loops (Affine, Generic, and Parallel).
 1.  Lower to Library Calls or Special Instructions, Intrinsics or ISA.
 1.  Partially Lower to Iterations Over a Finer-Grained Linalg Op.
 
 ## High-Level Description of Linalg Ops<a name="linalg_ops"></a>
 
 Linalg takes at least some inspiration from all previously
 [listed prior art](#prior_art). The design enables the definition of
 ***CustomOps*** with generic properties that enable
 [key transformations](#key_transformations), including lowering to scalar
 load/store and other operations or to external library calls and intrinsics.
 
-These ops can have ***either tensor or buffer operands***, subject to
-[conventions and limitations](#tensors_and_buffers).
+These ops can have ***either tensor or buffer*** as both input and output
+operands. Output tensors operands serve the purpose of providing a unifying
+abstraction and give a shape to the results. Output tensors can come in 2
+flavors and are always associated with a corresponding op result:
+
+1.  an "init tensor" output value which provides an initial value for a tensor
+    that is created by iteratively updating the result (also called "destructive
+    updates"). Such tensor is always materialized in some form. If enough fusion
+    occurs it may end up being materialized only as a register-level SSA value.
+    It is expected (but not required) that the destructive update pattern can be
+    rewritten as an inplace update on buffers.
+
+2.  a "shape-only" tensor output value whose underlying elements are not used in
+    the payload computation and only serves the purpose of carrying shape
+    information to lower levels of abstraction. In the future this will be
+    replaced by an appropriate shape type when it is available as a builtin type
+    (see the discourse discussion
+    [Linalg and Shapes](https://llvm.discourse.group/t/linalg-and-shapes/2421)
+    for more details).
 
 ### Payload-Carrying Ops<a name="payload_ops"></a>
 
 Linalg defines two payload carrying operations that implement the
 [structured ops](https://docs.google.com/presentation/d/1P-j1GrH6Q5gLBjao0afQ-GfvcAeF-QU4GXXeSy0eJ9I/edit#slide=id.p)
 abstraction on tensors and buffers. This is architected as two generic
 operations `linalg.generic` (resp. `linalg.indexed_generic`) that can express
 custom operations with *index-free semantics* (resp. *indexing semantics*). The
 properties of these generic ops are the result of applying the guiding
 principles described in the
 [Rationale Document](../Rationale/RationaleLinalgDialect.md). They are listed
 next, with a brief example and discussion for each.

 }
 ```
 
 The property participates in simplifying analyses and transformations. For
 instance, it guarantees no out-of bounds access can occur by construction
 (assuming dynamic operand dimensions agree with each other, which is the purpose
 of the `assert` runtime check).
 
-Before lowering to loop form, loop induction variables and iterators are *not
-yet materialized*. This is a necessary property if we want an abstraction that
-works on both tensor values and buffers because ***values don’t escape
-loops/nesting***.
-
-The main implications are that: 1. The semantics of the ops are *restricted to
-operate on structured data types*, on which we can define an iterator. 2. This
-does not model arbitrary code with side-effects.
+Before lowering to loop form, loop induction variables and iterators are
+implicit (i.e. *not yet materialized*).
+
+The main implications are that:
+
+1.  The semantics of the ops are *restricted to operate on structured data
+    types*, on which we can define an iterator.
+
+2.  This does not model arbitrary code with side-effects.
 
 We do not think these are serious limitations in practice because MLIR is all
 about mixing different levels of abstractions in the same IR. As long as Linalg
 can progressively lower to the next level of abstraction, it can also be just
 bypassed for things that do not fit.
 
 At the same time, conditioning op semantics on structured data types is a very
 promising path towards extensibility to non-dense tensors as experience with

 }
 ```
 
 The property "*Reversible Mappings Between Control and Data Structures*" is
 materialized by a lowering into a form that will resemble:
 
 ```
 // Run: mlir-opt example2.mlir -allow-unregistered-dialect -convert-linalg-to-loops
 #map0 = affine_map<(d0, d1) -> (d0 * 2 + d1 * 2)>
 
 func @example(%arg0: memref<8x?xf32, #map0>, %arg1: memref<?xvector<4xf32>>) {
   %c8 = constant 8 : index
   %c0 = constant 0 : index
   %c1 = constant 1 : index
   %0 = dim %arg0, %c1 : memref<8x?xf32, #map0>
   scf.for %arg2 = %c0 to %0 step %c1 {
     scf.for %arg3 = %c0 to %c8 step %c1 {
       %1 = load %arg0[%arg3, %arg2] : memref<8x?xf32, #map0>
       %2 = load %arg1[%arg3] : memref<?xvector<4xf32>>
       %3 = "some_compute"(%1, %2) : (f32, vector<4xf32>) -> vector<4xf32>
       store %3, %arg1[%arg3] : memref<?xvector<4xf32>>
     }
   }
   return
 }
 ```
 
 This mapping needs to be reversible because we want to be able to go back and
 forth between the two and answer questions such as:
 
 -   Given a subset of the iteration space, what subset of data does it read and
     write?
 -   Given a subset of data read or written, what subset of the iteration space
     is responsible for this read or write?
 
 Answering these `2` questions is one of the main analyses that Linalg uses to
 implement transformations such as tiling, tiled producer-consumer fusion, and
 promotion to temporary buffers in fast memory.
 
 In the current implementation, `linalg.generic` uses a list of
 [AffineMaps](https://mlir.llvm.org/docs/LangRef/#affinemap-attribute) (see the
 `#indexing_maps` attribute in the previous examples). This is a pragmatic
 short-term solution, but in the longer term note that this property could be
func @example(%A: memref<?x?xf32>, %B: memref<?x?xf32>, %C: memref<?x?xf32>) {
   }: memref<?x?xf32>, memref<?x?xf32>, memref<?x?xf32>
   return
 }
 ```
 
 In the process of lowering to loops and lower-level constructs, similar
 requirements are encountered, as are discussed in the
 [inlined call op proposal](https://llvm.discourse.group/t/introduce-std-inlined-call-op-proposal/282/2).
 We expect to be able to reuse the common lower-level infrastructure provided it
 evolves to support both region arguments and captures.
 
 #### Property 5: May Map To an External Library Call<a name="prop5"></a>
 
 A `linalg.generic` op may map to an external library call by specifying a
 `SymbolAttr`. At this level of abstraction, the important glue is the ability to
 perform transformations that preserve the structure necessary to ***call the
 external library after different transformations have been applied***.
 
 This involves considerations related to preservation of op semantics and
 integration at the ABI level. Regardless of whether one wants to use external
 library calls or a custom ISA, the problem for codegen is similar: preservation
 of a fixed granularity.
 
 Consider the following example that adds an additional attribute
 `library_call="pointwise_add"` that specifies the name of an external library
 call we intend to use:
 
 ```
 // File name: example4.mlir
 #indexing_maps = [
   affine_map<(i, j) -> (i, j)>,
   affine_map<(i, j) -> (i, j)>,
   affine_map<(i, j) -> (i, j)>
 ]
 #attrs = {
   args_in = 2,
   args_out = 1,
   indexing_maps = #indexing_maps,
   iterator_types = ["parallel", "parallel"],
   library_call = "pointwise_add"
 }
 func @example(%A: memref<?x?xf32>, %B: memref<?x?xf32>, %C: memref<?x?xf32>) {
   linalg.generic #attrs %A, %B, %C {
   ^bb0(%a: f32, %b: f32, %c: f32):
     %d = addf %a, %b : f32
     linalg.yield %d : f32
   }: memref<?x?xf32>, memref<?x?xf32>, memref<?x?xf32>
   return
 }
 ```
 
 The property "*Map To an External Library Call*" is materialized by a lowering
 into a form that will resemble:
 
 ```
 // Run: mlir-opt example4.mlir -convert-linalg-to-std
 // Note that we lower the Linalg dialect directly to the Standard dialect.
 // See this doc: https://mlir.llvm.org/docs/Dialects/Standard/
 
 #map0 = affine_map<(d0, d1)[s0, s1, s2] -> (d0 * s1 + s0 + d1 * s2)>
 
 func @example(%arg0: memref<?x?xf32>, %arg1: memref<?x?xf32>, %arg2: memref<?x?xf32>) {
   %0 = memref_cast %arg0 : memref<?x?xf32> to memref<?x?xf32, #map0>
   %1 = memref_cast %arg1 : memref<?x?xf32> to memref<?x?xf32, #map0>
   %2 = memref_cast %arg2 : memref<?x?xf32> to memref<?x?xf32, #map0>
   call @pointwise_add(%0, %1, %2) : (memref<?x?xf32, #map0>, memref<?x?xf32, #map0>, memref<?x?xf32, #map0>) -> ()
   return
 }
 func @pointwise_add(memref<?x?xf32, #map0>, memref<?x?xf32, #map0>, memref<?x?xf32, #map0>) attributes {llvm.emit_c_interface}
 ```
 
 Which, after lowering to LLVM resembles:
 
 ```
 // Run: mlir-opt example4.mlir -convert-linalg-to-std | mlir-opt -convert-std-to-llvm
 // Some generated code are omitted here.
 func @example(%arg0: !llvm<"float*">, ...) {
   ...
   llvm.call @pointwise_add(...) : (!llvm<"float*">, ...) -> ()
   return
 }
 
 llvm.func @pointwise_add(%arg0: !llvm<"float*">, ...) attributes {llvm.emit_c_interface} {
   ...
   llvm.call @_mlir_ciface_pointwise_add(%9, %19, %29) : (!llvm<"{ float*, float*, i64, [2 x i64], [2 x i64] }*">, !llvm<"{ float*, float*, i64, [2 x i64], [2 x i64] }*">, !llvm<"{ float*, float*, i64, [2 x i64], [2 x i64] }
 *">) -> ()
   llvm.return
 }
 llvm.func @_mlir_ciface_pointwise_add(!llvm<"{ float*, float*, i64, [2 x i64], [2 x i64] }*">, !llvm<"{ float*, float*, i64, [2 x i64], [2 x i64] }*">, !llvm<"{ float*, float*, i64, [2 x i64], [2 x i64] }*">) attributes {llvm.emit_c_interface}
 ```
 
 ##### Convention For External Library Interoperability
 
 The `linalg` dialect adopts a convention that is similar to `BLAS` when
 offloading operations to fast library implementations: pass a non-owning pointer
 to input and output data with additional metadata. This convention is also found
 in libraries such as `MKL`, `OpenBLAS`, `BLIS`, `cuBLAS`, `cuDNN`, etc.. and
 more generally at interface points across language boundaries (e.g. C++ /
 Python).

 automatically while still maintaining the
 [core guiding principles](#guiding_principles).
 
 For the time being, we have settled on the combination of these properties
 because of empirical evidence building and working on multiple high-level
 compilers. As we lay those down and engage more with the community, we expect
 multiple rounds of discussions and design changes to the original architecture.
 
-### Tensors and Buffers: Conventions and Limitations <a name="tensors_and_buffers"></a>
-
-Tensors are immutable SSA values, buffers are mutable regions of memory subject
-to side-effects and aliasing. As a consequence, output buffers are passed as
-operands whereas output tensors are new SSA values corresponding to op results.
-Inputs can be arbitrary tensors or buffers and are always passed as operands.
-
-The following convention is currently in-flight and is in the process of
-replacing other existing conventions. The following convention currently applies
-to "named" structured ops which are auto-generated by the linalg-ods tool.
-
-The convention adopted is as follows:
-
-1.  A first block of `ins` op operands hold read-only inputs of ShapedType.
-2.  An optional second block of `outs` op operands hold read-write output
-    buffers of MemRefType.
-3.  An optional third block of `init` operands hold initialization tensors of
-    RankedTensorType. Such tensors can appear when the op performs a reduction
-    and returns a tensor.
-
-Structured ops with fully parallel semantics, have empty `init`. They may either
-write in-place into `outs` buffers or return new tensors.
-
-Structured ops with reduction semantics and output tensor(s) however have
-additional restrictions:
-
-1.  They can only return a single tensor for now.
-2.  They cannot have any output buffer operand (i.e. `outs` is empty).
-3.  They have exactly one `init` tensor of the same type as the unique output
-    tensor. Such an `init` tensor does not have an explicit associate indexing
-    map. Instead the map of the result tensor is used to signify that the `init`
-    and the `result` are "tied".
-
-Points 1. and 2. keep complexity of the representation in check by allowing only
-a single result tensor, when reductions are present.
-
-Point 3. is related to the fact that SSA values cannot represent in-place
-updates. Instead, linalg adopts a similar convention that exists in e.g.
-`vector.outerproduct`: the value that is reduced into is passed as an explicit
-argument and a new result of the same shape is produced.
-
-It is expected buffer allocation will fold this last input onto the result in a
-single output buffer argument, which is why the same indexing map is required:
-the last input operand is said to be "tied" to the result.
-
-Alternative, more complex representations, would allow for:
-
-1.  Multiple results and `init` tensors in arbitrary orders, which could be
-    captured by an extra ArrayAttr of position pairs.
-2.  Relaxing the conditions on the indexing map equalities on the each pair and
-    e.g. allow implicit broadcasts of the input.
-
-These representations are deemed unnecessarily complex for now and are left for
-future discussion.
-
-As an illustration, the syntax for a `linalg.matmul` writing into a buffer is:
-
-```
-linalg.matmul ins(%a, %b : memref<?x?xf32>, tensor<?x?xf32>)
-             outs(%c : memref<?x?xf32>)
-```
-
-, whereas the syntax for a `linalg.matmul` returning a new tensor is:
-
-```
-%d = linalg.matmul ins(%a, %b : tensor<?x?xf32>, memref<?x?xf32>)
-                  init(%c : tensor<?x?xf32>)
-                    -> tensor<?x?xf32>
-```
-
 ### Data Representation: Views<a name="views"></a>
 
 The current implementation uses the
 [Strided MemRef (a.k.a View)](https://groups.google.com/a/tensorflow.org/forum/#!topic/mlir/MaL8m2nXuio)
 abstraction. The name *View* is used interchangeably in `linalg` to signify
 *Strided MemRef*. In the future we expect to use other structured data types and
 support ragged, mixed-sparse and other types. We expect to draw on the
 experience from existing LIFT abstractions for
 [sparse](https://www.lift-project.org/publications/2016/harries16sparse.pdf) and
 [position-dependent arrays](https://www.lift-project.org/publications/2019/pizzuti19positiondependentarrays.pdf).
 
 ### Metadata Ops<a name="metadata_ops"></a>

mlir/include/mlir/Dialect/Linalg/Analysis/DependenceAnalysis.h

private:
   DenseMap<Value, Value> aliases;
 };
 
 /// Data structure for holding a dependence graph that operates on LinalgOp and
 /// views as SSA values.
 class LinalgDependenceGraph {
 public:
   enum DependenceType { RAR = 0, RAW, WAR, WAW, NumTypes };
-  struct LinalgOpView {
-    Operation *op;
-    unsigned operandIndex;
-  };
+  // TODO: OpOperand tracks dependencies on buffer operands. Tensor result will
+  // need an extension to use OpResult.
   struct LinalgDependenceGraphElem {
     // dependentOpView may be either:
     //   1. src in the case of dependencesIntoGraphs.
     //   2. dst in the case of dependencesFromDstGraphs.
-    LinalgOpView dependentOpView;
+    OpOperand *dependentOpView;
     // View in the op that is used to index in the graph:
     //   1. src in the case of dependencesFromDstGraphs.
     //   2. dst in the case of dependencesIntoGraphs.
-    LinalgOpView indexingOpView;
+    OpOperand *indexingOpView;
     // Type of the dependence.
     DependenceType dependenceType;
   };
   using LinalgDependences = SmallVector<LinalgDependenceGraphElem, 8>;
   using DependenceGraph = DenseMap<Operation *, LinalgDependences>;
   using dependence_iterator = LinalgDependences::const_iterator;
   using dependence_range = iterator_range<dependence_iterator>;
 
private:
   /// Analyses the aliasing views between `src` and `dst` and inserts the proper
   /// dependences in the graph.
   void addDependencesBetween(LinalgOp src, LinalgOp dst);
 
   // Adds an new dependence unit in the proper graph.
   // Uses std::pair to keep operations and view together and avoid usage errors
   // related to src/dst and producer/consumer terminology in the context of
   // dependences.
-  void addDependenceElem(DependenceType dt, LinalgOpView indexingOpView,
-                         LinalgOpView dependentOpView);
+  void addDependenceElem(DependenceType dt, OpOperand *indexingOpView,
+                         OpOperand *dependentOpView);
 
   /// Implementation detail for findCoveringxxx.
   SmallVector<Operation *, 8>
   findOperationsWithCoveringDependences(LinalgOp srcLinalgOp,
                                         LinalgOp dstLinalgOp, Value view,
                                         ArrayRef<DependenceType> types) const;
 
   Aliases &aliases;
   SmallVector<LinalgOp, 8> linalgOps;
   DenseMap<Operation *, unsigned> linalgOpPositions;
 };
 } // namespace linalg
 } // namespace mlir
 
 #endif // MLIR_DIALECT_LINALG_ANALYSIS_DEPENDENCEANALYSIS_H_

mlir/include/mlir/Dialect/Linalg/EDSC/Builders.h

 
 namespace scf {
 class ParallelOp;
 } // namespace scf
 
 namespace edsc {
 inline void defaultRegionBuilder(ValueRange args) {}
 
-/// Build a `linalg.generic` op with the specified `inputs`, `outputBuffers`,
-/// `initTensors`, `resultTensorsTypes` and `region`.
+/// Build a `linalg.generic` op with the specified `inputs`, `outputs`,
+/// `resultTensorsTypes` and `region`.
 ///
 /// `otherValues` and `otherAttributes` may be passed and will be appended as
 /// operands and attributes respectively.
 ///
 /// Prerequisites:
 /// =============
 ///
 /// 1. `inputs` may contain StructuredIndexed that capture either buffer or
 /// tensor values.
-/// 2. `outputsBuffers` may contain StructuredIndexed that capture buffer
-/// values.
-/// 3. `initTensors` contain tensor values, without indexing maps.
-/// 4. `resultTensorTypes` may contain StructuredIndexed that capture return
-/// tensor types.
+/// 2. `outputs` may contain StructuredIndexed that capture either buffer or
+/// tensor values. In the future this will be extended with ranked shape values.
+/// 4. `resultTensorTypes` may contain return tensor types.
 Operation *makeGenericLinalgOp(
     ArrayRef<IteratorType> iteratorTypes, ArrayRef<StructuredIndexed> inputs,
-    ArrayRef<StructuredIndexed> outputBuffers, ArrayRef<Value> initTensors,
-    ArrayRef<StructuredIndexed> resultTensorTypes,
+    ArrayRef<StructuredIndexed> outputs, TypeRange resultTensorTypes,
     function_ref<void(ValueRange)> regionBuilder = defaultRegionBuilder,
     ArrayRef<Value> otherValues = {}, ArrayRef<Attribute> otherAttributes = {});
 
 namespace ops {
 using edsc::StructuredIndexed;
 
 //===----------------------------------------------------------------------===//
 // EDSC builders for linalg generic operations.

mlir/include/mlir/Dialect/Linalg/EDSC/Intrinsics.h

 
 namespace mlir {
 namespace edsc {
 namespace intrinsics {
 
 using linalg_copy = OperationBuilder<linalg::CopyOp>;
 using linalg_dot = OperationBuilder<linalg::DotOp>;
 using linalg_fill = OperationBuilder<linalg::FillOp>;
+using linalg_init_tensor = ValueBuilder<linalg::InitTensorOp>;
 using linalg_matmul = OperationBuilder<linalg::MatmulOp>;
 using linalg_matvec = OperationBuilder<linalg::MatvecOp>;
 using linalg_vecmat = OperationBuilder<linalg::VecmatOp>;
 using linalg_range = ValueBuilder<linalg::RangeOp>;
 using linalg_reshape = ValueBuilder<linalg::ReshapeOp>;
 using linalg_slice = ValueBuilder<linalg::SliceOp>;
 using linalg_yield = OperationBuilder<linalg::YieldOp>;
 
 } // namespace intrinsics
 } // namespace edsc
 } // namespace mlir
 
 #endif // MLIR_DIALECT_LINALG_EDSC_INTRINSICS_H_

mlir/include/mlir/Dialect/Linalg/IR/LinalgOps.h

 //===- LinalgOps.h - Linalg Operations --------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef MLIR_DIALECT_LINALG_LINALGOPS_H_
 #define MLIR_DIALECT_LINALG_LINALGOPS_H_
 
-#include "mlir/Dialect/Linalg/IR/LinalgTraits.h"
 #include "mlir/Dialect/Linalg/IR/LinalgTypes.h"
 #include "mlir/Dialect/StandardOps/IR/Ops.h"
 #include "mlir/Dialect/Utils/StructuredOpsUtils.h"
 #include "mlir/IR/AffineExpr.h"
 #include "mlir/IR/AffineMap.h"
 #include "mlir/IR/BlockAndValueMapping.h"
 #include "mlir/IR/Builders.h"
 #include "mlir/IR/BuiltinDialect.h"

 SmallVector<AffineExpr, 4> concat(ArrayRef<AffineExpr> a,
                                   ArrayRef<AffineExpr> b);
 
 /// Return the dims that are `iteratorTypeName` loops in the LinalgOp `op`.
 /// Assumes `op` is a LinalgOp.
 void getDimsOfType(Operation *op, StringRef iteratorTypeName,
                    SmallVectorImpl<AffineExpr> &res);
 
+namespace detail {
+LogicalResult verifyStructuredOpInterface(Operation *op);
+} // namespace detail
 } // namespace linalg
 } // namespace mlir
 
+namespace mlir {
+namespace linalg {
+class IndexedGenericOp;
+} // namespace linalg
+} // namespace mlir
 #include "mlir/Dialect/Linalg/IR/LinalgStructuredOpsInterfaces.h.inc"
 
 #define GET_OP_CLASSES
 #include "mlir/Dialect/Linalg/IR/LinalgOps.h.inc"
 
 #define GET_OP_CLASSES
 #include "mlir/Dialect/Linalg/IR/LinalgStructuredOps.h.inc"
 
 #endif // MLIR_DIALECT_LINALG_LINALGOPS_H_

mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td

 #ifndef LINALG_STRUCTURED_OPS
 #define LINALG_STRUCTURED_OPS
 
 include "mlir/Dialect/Linalg/IR/LinalgBase.td"
 include "mlir/Dialect/Linalg/IR/LinalgStructuredOpsInterface.td"
 include "mlir/Interfaces/CopyOpInterface.td"
 include "mlir/Interfaces/SideEffectInterfaces.td"
 
-// The Linalg `NInputs` trait provides the API for ops that are known
-// to have a specified number of inputs, all passed as operands.
-// See Linalg/LinalgTraits.h for implementation details and usage.
-class NInputs<int n> :
-  NativeOpTrait<"linalg::NInputs<" # !cast<string>(n) # ">::Impl"> {}
-
-// The Linalg `ZeroInitTensors` trait provides the API for ops that are known
-// to not have input tensor operands.
-// See Linalg/LinalgTraits.h for implementation details and usage.
-def ZeroInitTensors : NativeOpTrait<"linalg::ZeroInitTensors"> {}
-
-// The Linalg `NOutputs` trait provides the API for ops that are known
-// to have a specified number of outputs, all passed as operands.
-// See Linalg/LinalgTraits.h for implementation details and usage.
-class NOutputs<int n> :
-  NativeOpTrait<"linalg::NOutputs<" # !cast<string>(n) # ">::Impl"> {}
-
-def StructuredOpTraits : NativeOpTrait<"linalg::StructuredOpTraits">;
-def NamedStructuredOpTrait : NativeOpTrait<"linalg::NamedStructuredOpTrait">;
-
 // Base Tablegen class for Linalg ops.
 // Linalg ops that correspond to library calls operate on ShapedType as their
 // first operands. These may be optionally followed by non-view operands
 // depending on the specific Linalg op.
 class LinalgStructuredBase_Op<string mnemonic, list<OpTrait> props>
   : Op<Linalg_Dialect, mnemonic, !listconcat(props, [
        LinalgStructuredInterface])> {}
 
 class LinalgStructured_Op<string mnemonic, list<OpTrait> props>
   : LinalgStructuredBase_Op<mnemonic,
        !listconcat(props, [
-         StructuredOpTraits,
          DeclareOpInterfaceMethods<MemoryEffectsOpInterface>])> {
   code libraryCallName = [{
     std::string getLibraryCallName() {
       return generateLibraryCallName(getOperation());
     }
   }];
   let assemblyFormat = "`(` operands `)` attr-dict `:` type(operands)";
 }
 
 //===----------------------------------------------------------------------===//
 // Named Linalg ops, implemented as special configurations of generic ops.
 //===----------------------------------------------------------------------===//
 // At the moment these are not declarative and require a bunch of C++ code.
 // In the future, these should be migrated to a declarative specification.
-def CopyOp : LinalgStructured_Op<"copy", [
-    CopyOpInterface,
-    NInputs<1>,
-    ZeroInitTensors,
-    NOutputs<1>
-  ]> {
+def CopyOp : LinalgStructured_Op<"copy", [CopyOpInterface]> {
   let description = [{
     Copies the data in the input view into the output view.
 
     Usage:
 
     ```mlir
     linalg.copy(%arg0, %arg1) : memref<?xf32, stride_specification>,
                                 memref<?xf32, stride_specification>
def CopyOp : LinalgStructured_Op<"copy", [CopyOpInterface]> {
   // of builders with default arguments left unspecified.
   let builders = [OpBuilderDAG<(ins "Value":$input, "Value":$output),
     [{
       return build(
         $_builder, $_state, input, output, AffineMapAttr(), AffineMapAttr());
     }]>];
 
   let extraClassDeclaration = libraryCallName # [{
+    ValueRange inputs() { return getOperands().take_front(); }
+    ValueRange outputs() { return getOperands().take_back(); }
+
     // Rank-polymorphic.
     //   filling_value -> O(ivs) with parallel iterators.
     ArrayAttr iterator_types() {
       unsigned nPar = getInputShapedType(0).getRank();
       return Builder(getContext()).getStrArrayAttr(
         SmallVector<StringRef, 8>(nPar, getParallelIteratorTypeName()));
     }
 
let extraClassDeclaration = libraryCallName # [{
     }
   }];
   let verifier = [{ return ::verify(*this); }];
 
   let hasFolder = 1;
   let hasCanonicalizer = 1;
 }
 
-def FillOp : LinalgStructured_Op<"fill", [
-    NInputs<0>,
-    ZeroInitTensors,
-    NOutputs<1>]> {
-
+def FillOp : LinalgStructured_Op<"fill", []> {
   let arguments = (ins AnyStridedMemRef:$output,
                    AnyTypeOf<[AnyFloat, AnySignlessInteger, AnyVector]>:$value);
   let extraClassDeclaration = libraryCallName # [{
+    ValueRange inputs() { return {}; }
+    ValueRange outputs() { return getOperands().take_front(); }
+
     // Rank-polymorphic.
     //   filling_value -> O(ivs) with parallel iterators.
     ArrayAttr iterator_types() {
       unsigned nPar = getOutputShapedType(0).getRank();
       return Builder(getContext()).getStrArrayAttr(
         SmallVector<StringRef, 8>(nPar, getParallelIteratorTypeName()));
     }
 
code commonUtils = libraryCallName # [{
     }
 
     static std::function<void(Block &)> getRegionBuilder() {
       return nullptr;
     }
   }];
 }
 
-def ConvOp : PoolingBase_Op<"conv", [
-    NInputs<2>,
-    // Despite having reductions, this manually defined ConvOp may only take
-    // memref operands and can never have init tensors.
-    ZeroInitTensors,
-    NOutputs<1>]> {
-
+// Only support buffer semantics.
+def ConvOp : PoolingBase_Op<"conv", []> {
   let description = [{
     Generic n-D convolution as described in the TF documentation:
     https://www.tensorflow.org/versions/r2.0/api_docs/python/tf/nn/convolution
 
     ```
       output[b, x[0], ..., x[N-1], k] =
       sum_{z[0], ..., z[N-1], q}
           filter[z[0], ..., z[N-1], q, k] *
def ConvOp : PoolingBase_Op<"conv", []> {
   // into the input before doing the convolution.
   let arguments = (ins AnyStridedMemRef:$filter, AnyStridedMemRef:$input,
                    AnyStridedMemRef:$output,
                    OptionalAttr<I64ArrayAttr>:$strides,
                    OptionalAttr<I64ArrayAttr>:$dilations,
                    OptionalAttr<I64ElementsAttr>:$padding);
 
   let extraClassDeclaration = commonUtils # [{
+    ValueRange inputs() { return getOperands().slice(0, 2); }
+    ValueRange outputs() { return getOperands().take_back(); }
+
     // TODO: extend to support more than 1 dimensions and potentially grouping
     // too.
     unsigned getNumBatchDimensions() { return 1; }
 
     unsigned getNumInputFeatureDimensions() { return 1; }
 
     unsigned getNumOutputFeatureDimensions() { return 1; }
 
     unsigned getNumSpatialDimensions() {
       return getOutputShapedType(0).getRank() - getNumBatchDimensions() -
              getNumOutputFeatureDimensions();
     }
 
     ArrayAttr iterator_types() {
       // Outer parallel loops are always the number of output dimensions; i.e.
       // [b, xs, q] in the TF notation above.
       unsigned nPar = getOutputShapedType(0).getRank();
       unsigned nRed = getNumInputFeatureDimensions();
       // Window loops are a special kind of reduction that is never tiled or
       // parallelized across; i.e. [zs] in the TF notation above whose number
       // match `xs` (i.e. 1 window loop per "image" dimension).
       // This may evolve in the future.
+      // Conditionally check nPar is large enough for cases of ill-formed op: 
+      // this avoids overflows before hitting the verifier.
+      assert(nPar > getNumBatchDimensions() + getNumInputFeatureDimensions() &&
+             "expected at least one window dimension (i.e. memref ranks greater "
+             "than 2). See 'func @conv_rank_limit' in "
+             "mlir/test/Dialect/Linalg/invalid.mlir");
       unsigned nWin =
         nPar - getNumBatchDimensions() - getNumInputFeatureDimensions();
       SmallVector<StringRef, 8> iters(nPar, getParallelIteratorTypeName());
       iters.reserve(nPar + nRed + nWin);
       iters.append(nRed, getReductionIteratorTypeName());
       iters.append(nWin, getWindowIteratorTypeName());
       return Builder(getContext()).getStrArrayAttr(iters);
     }
 
     //   F(z0, ..., zN-1, q, k) *
     //     I(b, x0 + z0 - pad_low_0, ..., xN-1 + zN-1 - pad_low_N-1, q)
     //   ->  O(b, x0, ..., xN-1, k)
     // for N equal to `nWindow`. If there is no padding attribute, it will be
     // ignored.
     ArrayAttr indexing_maps() {
       MLIRContext *context = getContext();
       auto nWin = getNumWindowLoops();
-      assert(nWin > 0 && "expected at least one window dimension");
+      assert(nWin > 0 && "expected at least one window dimension (i.e. memref "
+                         "ranks greater than 2)");
       unsigned idx = 0;
       // In the following, AffineDimExprs are indexed in loop order:
       //   [ b, xs, k,           q,                     zs]
       //    parallels     non-window reductions     windows
       //
       // Parallel dims are exactly the dimensions indexing `output`:
       //     output[b, x[0], ..., x[N-1], k]; i.e.
       //  * batch dimensions (bs with #bs = 1 for now)
def ConvOp : PoolingBase_Op<"conv", []> {
   }];
 
   let verifier = [{ return ::verify(*this); }];
 
   let hasFolder = 1;
   let hasCanonicalizer = 1;
 }
 
+// Only support buffer semantics.
 class SingleInputPoolingBase_Op<string mnemonic>
-  : PoolingBase_Op<mnemonic, [
-    NInputs<2>,
-    // Despite having reductions, this manually defined ConvOp may only take
-    // memref operands and can never have init tensors.
-    ZeroInitTensors,
-    NOutputs<1>]> {
+    : PoolingBase_Op<mnemonic, []> {
   let description = [{
     A base class for single input pooling function.
 
     TODO: Figure out a better way to handle window dimensions, i.e., eliminate
     the fake memref.
     The window dimensions are specified by argument `windowDims`. The i-th
     dimension in the shape of `windowDims` denotes the size of the window along
     dimension i. For example, if the window size is 2x3, then a memref<2x3>
     should be passed to the operation as `windowDims`.
   }];
 
   let arguments = (ins AnyStridedMemRef:$input,
                    AnyStridedMemRef:$windowDims,
                    AnyStridedMemRef:$output,
                    OptionalAttr<I64ArrayAttr>:$strides,
                    OptionalAttr<I64ArrayAttr>:$dilations,
                    OptionalAttr<I64ElementsAttr>:$padding);
 
   let extraClassDeclaration = commonUtils# [{
+    ValueRange inputs() { return getOperands().slice(0, 2); }
+    ValueRange outputs() { return getOperands().take_back(); }
+
     ArrayAttr iterator_types() {
       // Outer parallel loops are always the number of output dimensions.
       unsigned nPar = getOutputShapedType(0).getRank();
       // The window loops has the same number loops with output dimensions.
       unsigned nWin = nPar;
       SmallVector<StringRef, 8> iters(nPar, getParallelIteratorTypeName());
       iters.reserve(nPar + nWin);
       iters.append(nWin, getWindowIteratorTypeName());
class LinalgOperandOfRank<int rank>: Type<
   And<[
     LinalgOperand.predicate,
     CPred<"$_self.cast<ShapedType>().getRank() == " # rank>]
   >>;
 
 class GenericOpBase<string mnemonic> : LinalgStructuredBase_Op<mnemonic, [
     AttrSizedOperandSegments,
     DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
-    NamedStructuredOpTrait,
     SingleBlockImplicitTerminator<"YieldOp">]> {
   let arguments = (ins Variadic<AnyShaped>:$inputs,
-                       Variadic<AnyMemRef>:$output_buffers,
-                       Variadic<AnyRankedTensor>:$init_tensors,
+                       Variadic<AnyShaped>:$outputs,
                        AffineMapArrayAttr:$indexing_maps,
                        ArrayAttr:$iterator_types,
                        OptionalAttr<StrAttr>:$doc,
                        OptionalAttr<StrAttr>:$library_call,
                        // ArrayAttr of StrArrayAttr:
                        OptionalAttr<ArrayAttr>:$sparse);
   let results = (outs Variadic<AnyRankedTensor>:$result_tensors);
   let regions = (region AnyRegion:$region);
let description = [{
 
     To allow progressive lowering from the value world (a.k.a tensor values) to
     the buffer world (a.k.a memref values), a `linalg.generic` op allows mixing
     tensors and buffers operands and tensor results.
 
     ```mlir
     %C = linalg.generic #trait_attribute
       ins(%A, %B : tensor<?x?xf32>, memref<?x?xf32, stride_specification>)
-      init(%C : tensor<?x?xf32>)
+      outs(%C : tensor<?x?xf32>)
       {other-optional-attributes}
       {region}
       -> (tensor<?x?xf32>)
     ```
-
-    The `init` operand and the conventions around mixing tensors and buffers are
-    described in more detail in the "Tensors and Buffers: Conventions and
-    Limitations" section in the [Linalg Document](../docs/Linalg.md)
-
-    Tensor values must be legalized by a buffer allocation pass before most
-    transformations can be applied. Such legalizations move tensor return values
-    into output buffer operands and updates the region arguments accordingly.
   }];
 
   let builders = [
     OpBuilderDAG<(ins "TypeRange":$resultTensorTypes, "ValueRange":$inputs,
-      "ValueRange":$outputBuffers, "ValueRange":$initTensors,
-      "ArrayRef<AffineMap>":$indexingMaps, "ArrayRef<StringRef>":$iteratorTypes,
-      "StringRef":$doc, "StringRef":$libraryCall,
+      "ValueRange":$outputs, "ArrayRef<AffineMap>":$indexingMaps,
+      "ArrayRef<StringRef>":$iteratorTypes, "StringRef":$doc,
+      "StringRef":$libraryCall,
       CArg<"function_ref<void(OpBuilder &, Location, ValueRange)>", "nullptr">)>,
     OpBuilderDAG<(ins "ValueRange":$inputs, "ValueRange":$outputBuffers,
       "ArrayRef<AffineMap>":$indexingMaps, "ArrayRef<StringRef>":$iteratorTypes,
       "StringRef":$doc, "StringRef":$libraryCall,
       CArg<"function_ref<void(OpBuilder &, Location, ValueRange)>", "nullptr">)>,
     OpBuilderDAG<(ins "TypeRange":$resultTensorTypes, "ValueRange":$inputs,
-      "ValueRange":$outputBuffers, "ValueRange":$initTensors,
-      "ArrayRef<AffineMap>":$indexingMaps, "ArrayRef<StringRef>":$iteratorTypes,
+      "ValueRange":$outputs, "ArrayRef<AffineMap>":$indexingMaps,
+      "ArrayRef<StringRef>":$iteratorTypes,
       CArg<"function_ref<void(OpBuilder &, Location, ValueRange)>", "nullptr">)>,
     OpBuilderDAG<(ins "ValueRange":$inputs, "ValueRange":$outputBuffers,
       "ArrayRef<AffineMap>":$indexingMaps, "ArrayRef<StringRef>":$iteratorTypes,
       CArg<"function_ref<void(OpBuilder &, Location, ValueRange)>", "nullptr">)>
   ];
   let verifier = [{ return ::verify(*this); }];
 
   let hasFolder = 1;
#matmul_trait = {
       iterator_types = ["parallel", "parallel", "reduction"]
     }
     ```
 
     And can be reused in multiple places as:
 
     ```mlir
       linalg.indexed_generic #matmul_trait
          ins(%A, %B : memref<?x?xf32, stride_specification>,
                       memref<?x?xf32, stride_specification>)
         outs(%C : memref<?x?xf32, stride_specification>) {
       (%offset_m: index, %offset_n: index, %offset_k: index,
        %a: f32, %b: f32, %c: f32) :
         "some_optional_computation"(%offset_m, %offset_n, %offset_k)
         %d = mulf %a, %b: f32
         %e = addf %c, %d: f32
         linalg_yield %e : f32
     }
let description = [{
     ```
 
     To allow progressive lowering from the value world (a.k.a tensor values) to
     the buffer world (a.k.a memref values), a `linalg.indexed_generic` op
     allows mixing tensors and buffers operands and tensor results.
 
     ```mlir
     %C = linalg.indexed_generic #trait_attribute
        ins(%A, %B : tensor<?x?xf32>, memref<?x?xf32, stride_specification>)
-      init(%C : tensor<?x?xf32>)
+      outs(%C : tensor<?x?xf32>)
       {other-optional-attributes}
       {region_with_index_arguments}
       -> (tensor<?x?xf32>)
     ```
-
-    The `init` operand and the conventions around mixing tensors and buffers are
-    described in more detail in the "Tensors and Buffers: Conventions and
-    Limitations" section in the [Linalg Document](../docs/Linalg.md)
-
-    Tensor values must be legalized by a buffer allocation pass before most
-    transformations can be applied. Such legalizations move tensor return values
-    into output buffer operands and update the region arguments accordingly.
   }];
 
   let builders = [
     OpBuilderDAG<(ins "TypeRange":$resultTensorTypes, "ValueRange":$inputs,
-      "ValueRange":$outputBuffers, "ValueRange":$initTensors,
-      "ArrayRef<AffineMap>":$indexingMaps, "ArrayRef<StringRef>":$iteratorTypes,
-      "StringRef":$doc, "StringRef":$libraryCall,
+      "ValueRange":$outputs, "ArrayRef<AffineMap>":$indexingMaps,
+      "ArrayRef<StringRef>":$iteratorTypes, "StringRef":$doc,
+      "StringRef":$libraryCall,
       CArg<"function_ref<void(OpBuilder &, Location, ValueRange, ValueRange)>",
            "nullptr">)>,
     OpBuilderDAG<(ins "ValueRange":$inputs, "ValueRange":$outputBuffers,
       "ArrayRef<AffineMap>":$indexingMaps, "ArrayRef<StringRef>":$iteratorTypes,
       "StringRef":$doc, "StringRef":$libraryCall,
       CArg<"function_ref<void(OpBuilder &, Location, ValueRange, ValueRange)>",
            "nullptr">)>,
     OpBuilderDAG<(ins "TypeRange":$resultTensorTypes, "ValueRange":$inputs,
-      "ValueRange":$outputBuffers, "ValueRange":$initTensors,
-      "ArrayRef<AffineMap>":$indexingMaps, "ArrayRef<StringRef>":$iteratorTypes,
+      "ValueRange":$outputs, "ArrayRef<AffineMap>":$indexingMaps,
+      "ArrayRef<StringRef>":$iteratorTypes,
       CArg<"function_ref<void(OpBuilder &, Location, ValueRange, ValueRange)>",
            "nullptr">)>,
     OpBuilderDAG<(ins "ValueRange":$inputs, "ValueRange":$outputBuffers,
       "ArrayRef<AffineMap>":$indexingMaps, "ArrayRef<StringRef>":$iteratorTypes,
       CArg<"function_ref<void(OpBuilder &, Location, ValueRange, ValueRange)>",
            "nullptr">)>
   ];
   let verifier = [{ return ::verify(*this); }];

mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOpsInterface.td

 def LinalgStructuredInterface : OpInterface<"LinalgOp"> {
   let cppNamespace = "::mlir::linalg";
   let methods = [
     //===------------------------------------------------------------------===//
     // Loop types handling.
     //===------------------------------------------------------------------===//
     InterfaceMethod<
       /*desc=*/[{
+        Return the number of induction variables in the basic block. This should
+        always be 0 for index-free linalg ops. For IndexedGeneric, this must be
+        equal to numLoops
+      }],
+      /*retTy=*/"unsigned",
+      /*methodName=*/"getNumPayloadInductionVariables",
+      /*args=*/(ins),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        return isa<IndexedGenericOp>(this->getOperation()) ?
+          $_op.getNumLoops() : 0;
+      }]
+    >,
+    //===------------------------------------------------------------------===//
+    // Loop types handling.
+    //===------------------------------------------------------------------===//
+    InterfaceMethod<
+      /*desc=*/[{
         Return the number of parallel loops.
       }],
       /*retTy=*/"unsigned",
       /*methodName=*/"getNumParallelLoops",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
         return getNumIterators(getParallelIteratorTypeName(),
InterfaceMethod<
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
         auto iters = $_op.iterator_types();
         return iters.size() == 1 &&
                getNumIterators(getReductionIteratorTypeName(), iters) == 1;
       }]>,
     //===------------------------------------------------------------------===//
-    // Num input/output/initTensors arguments handling.
+    // Num input/output arguments handling.
     //===------------------------------------------------------------------===//
-    // These special methods must be defined by each op that wants to implement
-    // the LinalgStructuredInterface. For now, this is either:
-    // - Explicitly specified in the op definition.
-    // - Derived from variadic attributes (for "named" ops, linalg.generic and
-    //   linalg.indexed_generic ops).
+    // `inputs` must be defined by each op that wants to implement the
+    // LinalgStructuredInterface.
+    InterfaceMethod<
+      /*desc=*/[{
+        Return the input shape operands.
+      }],
+      /*retTy=*/"ValueRange",
+      /*methodName=*/"inputs",
+      /*args=*/(ins)
+    >,
+    // These special methods rely on `inputs` and `outputs` being defined by
+    // each op that wants to implement the LinalgStructuredInterface.
     InterfaceMethod<
       /*desc=*/[{
         Return the number of inputs.
       }],
       /*retTy=*/"unsigned",
-      /*methodName=*/"getNumInputs"
+      /*methodName=*/"getNumInputs",
+      /*args=*/(ins),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        return $_op.inputs().size();
+      }]
     >,
+    // `outputs` must be defined by each op that wants to implement the
+    // LinalgStructuredInterface.
     InterfaceMethod<
       /*desc=*/[{
-        Return the number of init tensors.
+        Return the output shape operands.
       }],
-      /*retTy=*/"unsigned",
-      /*methodName=*/"getNumInitTensors"
+      /*retTy=*/"ValueRange",
+      /*methodName=*/"outputs",
+      /*args=*/(ins)
     >,
     InterfaceMethod<
       /*desc=*/[{
         Return the number of outputs.
       }],
       /*retTy=*/"unsigned",
-      /*methodName=*/"getNumOutputs"
+      /*methodName=*/"getNumOutputs",
+      /*args=*/(ins),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        return $_op.outputs().size();
+      }]
     >,
     //===------------------------------------------------------------------===//
-    // Input arguments handling.
+    // Input operands handling.
     //===------------------------------------------------------------------===//
     InterfaceMethod<
       /*desc=*/[{
-        Return the `i`-th input value.
-        The `i^th` input argument is always the `i^th` operand regardless of
-        whether we have tensors or buffers.
+        Return the `i`-th input operand.
       }],
       /*retTy=*/"Value",
       /*methodName=*/"getInput",
       /*args=*/(ins "unsigned":$i),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
         assert(i < $_op.getNumInputs());
         return this->getOperation()->getOperand(i);
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the index of the given input value `v`, or `None` if the value is
-        not an input.
-      }],
-      /*retTy=*/"llvm::Optional<unsigned>",
-      /*methodName=*/"getIndexOfInput",
-      /*args=*/(ins "Value":$value),
-      /*methodBody=*/"",
-      /*defaultImplementation=*/[{
-        auto it = llvm::find(getInputs(), value);
-        if (it != getInputs().end())
-          return it - getInputs().begin();
-        return llvm::None;
-      }]
-    >,
-    InterfaceMethod<
-      /*desc=*/[{
-        Return the `i`-th input shaped type, irrespective of buffer or tensor
-        type.
+        Return the `i`-th input shaped type
       }],
       /*retTy=*/"ShapedType",
       /*methodName=*/"getInputShapedType",
       /*args=*/(ins "unsigned":$i),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
         return getInput(i).getType().template cast<ShapedType>();
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the input operands.
+        Return the range of input operands.
       }],
       /*retTy=*/"Operation::operand_range",
       /*methodName=*/"getInputs",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
         auto range = this->getOperation()->getOperands();
         return {range.begin(), range.begin() + $_op.getNumInputs()};
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the range over the input operands that are of buffer type.
+        Return the OpOperands for the input operands.
+      }],
+      /*retTy=*/" MutableArrayRef<OpOperand>",
+      /*methodName=*/"getInputOpOperands",
+      /*args=*/(ins),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        return this->getOperation()->getOpOperands().take_front(getNumInputs());
+      }]
+    >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Return the subset of input operands that are of buffer type.
       }],
       /*retTy=*/"SmallVector<Value, 4>",
       /*methodName=*/"getInputBuffers",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
         return llvm::to_vector<4>(llvm::make_filter_range(
-          getInputs(), [](Value in){ return in.getType().isa<MemRefType>(); }));
+          getInputs(), [](Value in){ return in.getType().template isa<MemRefType>(); }));
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the subset of input operands that are of ranked tensor type.
+        Return the number of input buffer operands.
       }],
-      /*retTy=*/"SmallVector<RankedTensorType, 4>",
-      /*methodName=*/"getInputTensorTypes" ,
+      /*retTy=*/"unsigned",
+      /*methodName=*/"getNumInputBuffers",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        SmallVector<RankedTensorType, 4> res;
-        for (Type type : getInputs().getTypes())
-          if (auto t = type.template dyn_cast<RankedTensorType>())
-            res.push_back(t);
-        return res;
+        return $_op.getInputBuffers().size();
       }]
     >,
-    //===------------------------------------------------------------------===//
-    // Output arguments handling.
-    //===------------------------------------------------------------------===//
     InterfaceMethod<
       /*desc=*/[{
-        Return the output buffer at the given index, asserts that this is a
-        buffer operand and not a tensor result.
-        The `i^th` output argument is an operand (resp. a return value) iff it
-        is a value of buffer type (resp. a return value of tensor type).
+        Return the `index`^th input buffer.
       }],
       /*retTy=*/"Value",
-      /*methodName=*/"getOutputBuffer",
-      /*args=*/(ins "unsigned":$i),
+      /*methodName=*/"getInputBuffer",
+      /*args=*/(ins "unsigned":$index),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        // Output buffers are passed as output buffer operands (side-effecting).
-        // Output tensors are results.
-        // The union of the 2 are all the outputs and we want to ensure i does
-        // not overflow the buffer operands.
-        assert(i + this->getOperation()->getNumResults() < $_op.getNumOutputs()
-               && "overflowing output buffer index");
-        return this->getOperation()->getOperand($_op.getNumInputs() + i);
+        assert(index < getNumInputBuffers());
+        return getInputBuffers()[index];
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the index of the given buffer value, or `None` if the value is
-        not part of the output buffers.
+        Return the subset of input operands that are of buffer type.
       }],
-      /*retTy=*/"llvm::Optional<unsigned>",
-      /*methodName=*/"getIndexOfOutputBuffer",
-      /*args=*/(ins "Value":$value),
+      /*retTy=*/"SmallVector<OpOperand*, 4>",
+      /*methodName=*/"getInputBuffersOpOperands",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        auto it = llvm::find(getOutputBuffers(), value);
-        if (it != getOutputBuffers().end())
-          return it - getOutputBuffers().begin();
-        return llvm::None;
+        SmallVector<OpOperand*, 4> res;
+        res.reserve(getNumInputs());
+        for (OpOperand &o : getInputOpOperands())
+          if (o.get().getType().isa<MemRefType>())
+            res.push_back(&o);
+        return res;
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the type of the output buffer at the given index.
+        Return the subset of input operands that are of tensor type.
       }],
-      /*retTy=*/"MemRefType",
-      /*methodName=*/"getOutputBufferType",
-      /*args=*/(ins "unsigned":$i),
+      /*retTy=*/"SmallVector<Value, 4>",
+      /*methodName=*/"getInputTensors",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        return getOutputBuffer(i).getType().template cast<MemRefType>();
-      }]>,
+        return llvm::to_vector<4>(llvm::make_filter_range(
+          getInputs(),
+          [](Value in){ return in.getType().template isa<RankedTensorType>(); }));
+      }]
+    >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the `i`-th output shaped type, irrespective of buffer or tensor
-        type.
+        Return the subset of op operands that are of tensor type.
       }],
-      /*retTy=*/"ShapedType",
-      /*methodName=*/"getOutputShapedType",
-      /*args=*/(ins "unsigned":$i),
+      /*retTy=*/"SmallVector<OpOperand*, 4>",
+      /*methodName=*/"getInputTensorsOpOperands",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        return getShapedType(i + $_op.getNumInputs());
-      }]>,
+        SmallVector<OpOperand*, 4> res;
+        res.reserve(getNumInputs());
+        for (OpOperand &o : getInputOpOperands())
+          if (o.get().getType().isa<RankedTensorType>())
+            res.push_back(&o);
+        return res;
+      }]
+    >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the results that are of ranked tensor type.
+        Return the types of the subset of input operands that are of buffer type.
       }],
-      /*retTy=*/"SmallVector<RankedTensorType, 4>",
-      /*methodName=*/"getOutputTensorTypes",
+      /*retTy=*/"SmallVector<MemRefType, 4>",
+      /*methodName=*/"getInputBufferTypes" ,
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        SmallVector<RankedTensorType, 4> res;
-        for (Type type : this->getOperation()->getResults().getTypes())
-          res.push_back(type.template cast<RankedTensorType>());
-        return res;
-      }]>,
+        return llvm::to_vector<4>(
+          llvm::map_range(
+            llvm::make_filter_range(
+              ValueRange(getInputs()).getTypes(),
+              [](Type in){ return in.isa<MemRefType>(); }),
+            [](Type in){ return in.cast<MemRefType>(); }));
+      }]
+    >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the output buffers (operands).
+        Return the types of the subset of input operands that are of ranked
+        tensor type.
       }],
-      /*retTy=*/"Operation::operand_range",
-      /*methodName=*/"getOutputBuffers",
+      /*retTy=*/"SmallVector<RankedTensorType, 4>",
+      /*methodName=*/"getInputTensorTypes" ,
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        auto range = this->getOperation()->getOperands();
-        return {range.begin() + $_op.getNumInputs(),
-                range.begin() + getNumInputsAndOutputBuffers()};
+        return llvm::to_vector<4>(
+          llvm::map_range(
+            llvm::make_filter_range(
+              ValueRange(getInputs()).getTypes(),
+              [](Type in){ return in.isa<RankedTensorType>(); }),
+            [](Type in){ return in.cast<RankedTensorType>(); }));
       }]
     >,
 
     //===------------------------------------------------------------------===//
-    // Input and Output arguments handling.
+    // Output operands handling.
     //===------------------------------------------------------------------===//
     InterfaceMethod<
       /*desc=*/[{
-        Return one single buffer at position `$i`.
+        Return the `i`-th output operand.
       }],
       /*retTy=*/"Value",
-      /*methodName=*/"getBuffer",
+      /*methodName=*/"getOutput",
       /*args=*/(ins "unsigned":$i),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        assert(i < getNumInputsAndOutputBuffers() && "overflowing buffers index");
-        return this->getOperation()->getOperand(i);
+        assert(i < $_op.getNumOutputs());
+        return this->getOperation()->getOperand(i + $_op.getNumInputs());
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the number of output buffers
+        Return the `i`-th output shaped type
       }],
-      /*retTy=*/"unsigned",
-      /*methodName=*/"getNumOutputBuffers",
+      /*retTy=*/"ShapedType",
+      /*methodName=*/"getOutputShapedType",
+      /*args=*/(ins "unsigned":$i),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        return getOutput(i).getType().template cast<ShapedType>();
+      }]
+    >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Return the range of output operands.
+      }],
+      /*retTy=*/"Operation::operand_range",
+      /*methodName=*/"getOutputs",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        return $_op.getNumOutputs() - this->getOperation()->getNumResults();
+        auto start =
+          this->getOperation()->getOperands().begin() + $_op.getNumInputs();
+        return {start, start + $_op.getNumOutputs()};
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the number of inputs and outputs, irrespective of their buffer or
-        tensor type.
+        Return the OpOperands for the output operands.
       }],
-      /*retTy=*/"unsigned",
-      /*methodName=*/"getNumInputsAndOutputs",
+      /*retTy=*/" MutableArrayRef<OpOperand>",
+      /*methodName=*/"getOutputOpOperands",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        return $_op.getNumInputs() + $_op.getNumOutputs();
+        return this->getOperation()->getOpOperands().slice(
+          getNumInputs(), getNumOutputs());
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the number of inputs, irrespective of their buffer or tensor type
-        and output buffers
+        Return the subset of output operands that are of buffer type.
       }],
-      /*retTy=*/"unsigned",
-      /*methodName=*/"getNumInputsAndOutputBuffers",
+      /*retTy=*/"SmallVector<Value, 4>",
+      /*methodName=*/"getOutputBuffers",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        return $_op.getNumInputs() + $_op.getNumOutputs() -
-          this->getOperation()->getNumResults();
+        return llvm::to_vector<4>(llvm::make_filter_range(
+          getOutputs(), [](Value in){ return in.getType().template isa<MemRefType>(); }));
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the range over inputs (irrespective of type) and output buffers.
+        Return the `index`^th output buffer.
       }],
-      /*retTy=*/"Operation::operand_range",
-      /*methodName=*/"getInputsAndOutputBuffers",
+      /*retTy=*/"Value",
+      /*methodName=*/"getOutputBuffer",
+      /*args=*/(ins "unsigned":$index),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        assert(index < getNumOutputBuffers());
+        return getOutputBuffers()[index];
+      }]
+    >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Return the subset of output operands that are of buffer type.
+      }],
+      /*retTy=*/"SmallVector<OpOperand*, 4>",
+      /*methodName=*/"getOutputBuffersOpOperands",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        auto range = this->getOperation()->getOperands();
-        return {range.begin(), range.begin() + getNumInputsAndOutputBuffers()};
+        SmallVector<OpOperand*, 4> res;
+        res.reserve(getNumOutputs());
+        for (OpOperand &o : getOutputOpOperands())
+          if (o.get().getType().isa<MemRefType>())
+            res.push_back(&o);
+        return res;
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the range over init tensors.
+        Return the number of output buffer operands.
       }],
-      /*retTy=*/"Operation::operand_range",
-      /*methodName=*/"getInitTensors",
+      /*retTy=*/"unsigned",
+      /*methodName=*/"getNumOutputBuffers",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        auto range = this->getOperation()->getOperands();
-        auto base = range.begin() + getNumInputsAndOutputBuffers();
-        return {base, base + $_op.getNumInitTensors()};
+        return $_op.getOutputBuffers().size();
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return one single init tensor at position `$i`.
+        Return the subset of output operands that are of tensor type.
       }],
-      /*retTy=*/"Value",
-      /*methodName=*/"getInitTensor",
-      /*args=*/(ins "unsigned":$i),
+      /*retTy=*/"SmallVector<Value, 4>",
+      /*methodName=*/"getOutputTensors",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        assert(i < $_op.getNumInitTensors() && "overflowing init tensor index");
-        return getInitTensors()[i];
+        return llvm::to_vector<4>(llvm::make_filter_range(
+          getOutputs(),
+          [](Value in){ return in.getType().template isa<RankedTensorType>(); }));
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return true if the shaped operand index `i` is the index of an init
-        tensor.
+        Return the subset of output operands that are of tensor type.
       }],
-      /*retTy=*/"bool",
-      /*methodName=*/"isIndexOfAnInitTensor",
-      /*args=*/(ins "unsigned":$i),
+      /*retTy=*/"SmallVector<OpOperand*, 4>",
+      /*methodName=*/"getOutputTensorsOpOperands",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        assert(i < $_op.getNumShapedOperands() && "overflowing shaped operand index");
-        return i >= $_op.getNumInputs() + getNumOutputBuffers();
+        SmallVector<OpOperand*, 4> res;
+        res.reserve(getNumOutputs());
+        for (OpOperand &o : getOutputOpOperands())
+          if (o.get().getType().isa<RankedTensorType>())
+            res.push_back(&o);
+        return res;
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the relative init tensor index of the shaped operand index.
+        Return the number of output tensor operands.
       }],
       /*retTy=*/"unsigned",
-      /*methodName=*/"getInitTensorIndexFromShapedIndex",
-      /*args=*/(ins "unsigned":$i),
+      /*methodName=*/"getNumOutputTensors",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        assert(isIndexOfAnInitTensor(i) && "expected an init tensor index");
-        return i - $_op.getNumInputs() - getNumOutputBuffers();
+        return $_op.getOutputTensors().size();
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the index of the given init tensor value, or `None` if the value 
-        is not part of the init tensors.
+        Return the types of the subset of output operands that are of buffer type.
       }],
-      /*retTy=*/"llvm::Optional<unsigned>",
-      /*methodName=*/"getIndexOfInitTensor",
-      /*args=*/(ins "Value":$value),
+      /*retTy=*/"SmallVector<MemRefType, 4>",
+      /*methodName=*/"getOutputBufferTypes" ,
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        auto it = llvm::find(getInitTensors(), value);
-        if (it != getInitTensors().end())
-          return it - getInitTensors().begin();
-        return llvm::None;
+        return llvm::to_vector<4>(
+          llvm::map_range(
+            llvm::make_filter_range(
+              ValueRange(getOutputs()).getTypes(),
+              [](Type in){ return in.isa<MemRefType>(); }),
+            [](Type in){ return in.cast<MemRefType>(); }));
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the number of inputs, output buffers and init tensors operands.
+        Return the types of the subset of output operands that are of ranked
+        tensor type.
       }],
-      /*retTy=*/"unsigned",
-      /*methodName=*/"getNumShapedOperands",
+      /*retTy=*/"SmallVector<RankedTensorType, 4>",
+      /*methodName=*/"getOutputTensorTypes" ,
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        return getNumInputsAndOutputBuffers() + $_op.getNumInitTensors();
+        return llvm::to_vector<4>(
+          llvm::map_range(
+            llvm::make_filter_range(
+              ValueRange(getOutputs()).getTypes(),
+              [](Type in){ return in.isa<RankedTensorType>(); }),
+            [](Type in){ return in.cast<RankedTensorType>(); }));
       }]
     >,
+
+    //===------------------------------------------------------------------===//
+    // Input and Output arguments handling.
+    //===------------------------------------------------------------------===//
     InterfaceMethod<
       /*desc=*/[{
-        Return the `i`-th shaped operand value, which can be an arbitrary input
-        tensor/buffer, init tensor or output buffer.
+        Return true if the payload uses the value loaded from `opOperand`. This
+        is useful to avoid loading from "write-only" memory that may be
+        uninitialized, as well as properly cloning "read-write" operands.
       }],
-      /*retTy=*/"Value",
-      /*methodName=*/"getShapedOperand",
-      /*args=*/(ins "unsigned":$i),
+      /*retTy=*/"bool",
+      /*methodName=*/"payloadUsesValueFromOpOperand",
+      /*args=*/(ins "OpOperand *":$opOperand),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        assert(i < $_op.getNumShapedOperands());
-        return this->getOperation()->getOperand(i);
+        unsigned bbArgNumber =
+          getNumPayloadInductionVariables() + opOperand->getOperandNumber();
+        // Safeguard against the named linalg ops that are manually defined and
+        // that only support buffer semantics: we should not be there.
+        // Such ops have an empty regionBuilder and are not constructed with a
+        // region for now. In the future they are slated to disappear.
+        assert(this->getOperation()->getNumRegions() == 1 && "unexpected "
+               "missing region (calling `payloadUsesValueFromOpOperand` on "
+               "manually defined named Linalg op?)");
+        Block &block = this->getOperation()->getRegion(0).front();
+        // Init tensors have uses.
+        return !block.getArgument(bbArgNumber).use_empty();
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the range over inputs, output buffers and init tensors.
+        Return true if the payload uses the value loaded from input operand
+        `index`.
       }],
-      /*retTy=*/"Operation::operand_range",
-      /*methodName=*/"getShapedOperands",
-      /*args=*/(ins),
+      /*retTy=*/"bool",
+      /*methodName=*/"payloadUsesValueFromInputOperandIndex",
+      /*args=*/(ins "unsigned":$index),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        auto range = this->getOperation()->getOperands();
-        return {range.begin(), range.begin() + getNumShapedOperands()};
+        return payloadUsesValueFromOpOperand(&getInputOpOperands()[index]);
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the `i`-th shaped type, there are 3 cases:
-          1. if `i < $_op.getNumInputs()` then return `getInputShapedType(i)`;
-             otherwise
-          2. if `i < getNumInputsAndOutputBuffers()` then return the
-             `getOutputBufferType(i - $_op.getNumInputs())`; otherwise
-          3. return the `i - getNumInputsAndOutputBuffers()` result type.
+        Return true if the payload uses the value loaded from output operand
+        `index`.
       }],
-      /*retTy=*/"ShapedType",
-      /*methodName=*/"getShapedType",
-      /*args=*/(ins "unsigned":$i),
+      /*retTy=*/"bool",
+      /*methodName=*/"payloadUsesValueFromOutputOperandIndex",
+      /*args=*/(ins "unsigned":$index),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        if (i < $_op.getNumInputs())
-          return getInputShapedType(i);
-        if (i < getNumInputsAndOutputBuffers())
-          return getOutputBufferType(i - $_op.getNumInputs());
-        return this->getOperation()->getResult(
-          i - getNumInputsAndOutputBuffers()).
-          getType().template cast<ShapedType>();
-      }]>,
+        return payloadUsesValueFromOpOperand(&getOutputOpOperands()[index]);
+      }]
+    >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the shaped types for all the inputs and outputs
+        Return true if `opOperand` is an init tensor. This is true when it is
+        an output tensor operand whose value is used in the payload region.
       }],
-      /*retTy=*/"SmallVector<ShapedType, 4>",
-      /*methodName=*/"getInputOutputShapedTypes",
+      /*retTy=*/"bool",
+      /*methodName=*/"isInitTensor",
+      /*args=*/(ins "OpOperand *":$opOperand),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        if (!opOperand->get().getType().template isa<RankedTensorType>())
+          return false;
+        if (opOperand->getOperandNumber() < $_op.getNumInputs())
+          return false;
+        return payloadUsesValueFromOpOperand(opOperand);
+      }]
+    >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Return true if the operand at output index `index` is an init tensor.
+      }],
+      /*retTy=*/"bool",
+      /*methodName=*/"isIndexOfInitTensor",
+      /*args=*/(ins "unsigned":$index),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        assert(index < getNumOutputs());
+        return isInitTensor(
+          &this->getOperation()->getOpOperands()[$_op.getNumInputs() + index]);
+      }]
+    >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Return the output operands that are init tensors.
+      }],
+      /*retTy=*/"SmallVector<Value, 4>",
+      /*methodName=*/"getInitTensors",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        SmallVector<Type, 4> inputOutputTypes(
-            this->getOperation()->operand_type_begin(),
-            this->getOperation()->operand_type_end());
-        inputOutputTypes.append(this->getOperation()->result_type_begin(),
-                                this->getOperation()->result_type_end());
+        auto start =
+          this->getOperation()->getOpOperands().begin() + $_op.getNumInputs();
         return llvm::to_vector<4>(
-            llvm::map_range(inputOutputTypes, [](Type type) -> ShapedType {
-              return type.cast<ShapedType>();
+          llvm::map_range(
+            llvm::make_filter_range(
+              llvm::make_range(start, start + $_op.getNumOutputs()),
+              [&](OpOperand &opOperand) {
+                return $_op.isInitTensor(&opOperand);
+              }),
+              [&](OpOperand &opOperand) {
+                return opOperand.get();
               }));
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Return the first position of the shaped operand in the operand list.
+        Return the number of init tensor operands.
       }],
-      /*retTy=*/"Optional<unsigned>",
-      /*methodName=*/"getIndexOfShapedOperand",
-      /*args=*/(ins "Value":$value),
+      /*retTy=*/"unsigned",
+      /*methodName=*/"getNumInitTensors",
+      /*args=*/(ins),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        return getInitTensors().size();
+      }]
+    >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Return the number of input and output operands.
+      }],
+      /*retTy=*/"unsigned",
+      /*methodName=*/"getNumShapedOperands",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        Optional<unsigned> inputIndex = getIndexOfInput(value);
-        if (inputIndex.hasValue()) return inputIndex.getValue();
-        Optional<unsigned> outputIndex = getIndexOfOutputBuffer(value);
-        if (outputIndex.hasValue())
-          return $_op.getNumInputs() + outputIndex.getValue();
-        Optional<unsigned> initTensorIndex = getIndexOfInitTensor(value);
-        if (initTensorIndex.hasValue())
-          return $_op.getNumInputs() + $_op.getNumOutputBuffers() + initTensorIndex.getValue();
-        return llvm::None;
+        return $_op.getNumInputs() + $_op.getNumOutputs();
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Returns the operand index given the input index. Returns None
-        of the input index is invalid.
+        Return the `i`-th shaped operand value.
       }],
-      /*retTy=*/"Optional<unsigned>",
-      /*methodName=*/"getOperandIndexForInputIndex",
-      /*args=*/(ins "unsigned":$input_index),
+      /*retTy=*/"Value",
+      /*methodName=*/"getShapedOperand",
+      /*args=*/(ins "unsigned":$i),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        if (input_index >= $_op.getNumInputs())
-          return llvm::None;
-        return input_index;
+        assert(i < $_op.getNumShapedOperands());
+        return this->getOperation()->getOperand(i);
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Returns the operand index given the output index. Returns None
-        of the output index is invalid.
+        Return the range over input and output operands.
       }],
-      /*retTy=*/"Optional<unsigned>",
-      /*methodName=*/"getOperandIndexForOutputIndex",
-      /*args=*/(ins "unsigned":$output_index),
+      /*retTy=*/"Operation::operand_range",
+      /*methodName=*/"getShapedOperands",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        if (output_index >= $_op.getNumOutputs())
-          return llvm::None;
-        return output_index + $_op.getNumInputs();
+        auto range = this->getOperation()->getOperands();
+        return {range.begin(), range.begin() + getNumShapedOperands()};
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Returns the input index given the operand index. Return None
-        if the operand index doesnt corresponding to an input.
+        Return the OpOperands for all the shaped operands.
       }],
-      /*retTy=*/"Optional<unsigned>",
-      /*methodName=*/"getInputIndex",
-      /*args=*/(ins "unsigned":$operand_index),
+      /*retTy=*/" MutableArrayRef<OpOperand>",
+      /*methodName=*/"getShapedOpOperands",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-         if (operand_index >= $_op.getNumInputs())
-           return llvm::None;
-         return operand_index;
+        return this->getOperation()->getOpOperands().take_front(
+          getNumShapedOperands());
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
-        Returns the output index given the operand index. Return None
-        if the operand index doesnt corresponding to an output.
+        Return the range over input and output operands.
       }],
-      /*retTy=*/"Optional<unsigned>",
-      /*methodName=*/"getOutputIndex",
-      /*args=*/(ins "unsigned":$operand_index),
+      /*retTy=*/"SmallVector<ShapedType, 4>",
+      /*methodName=*/"getShapedOperandTypes",
+      /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-         if (operand_index < $_op.getNumInputs() ||
-             operand_index >= $_op.getNumInputs() + $_op.getNumOutputs())
-           return llvm::None;
-         return operand_index - $_op.getNumInputs();
+        return llvm::to_vector<4>(
+          llvm::map_range(
+            getShapedOperands(),
+            [](Value v) { return v.getType().cast<ShapedType>(); }));
       }]
     >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Return the `i`-th shaped type
+      }],
+      /*retTy=*/"ShapedType",
+      /*methodName=*/"getShapedType",
+      /*args=*/(ins "unsigned":$i),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        return $_op.getShapedOperand(i).getType().template cast<ShapedType>();
+      }]>,
 
     //===------------------------------------------------------------------===//
     // Other interface methods.
     //===------------------------------------------------------------------===//
     InterfaceMethod<
       /*desc=*/[{
         Return the iterator types attribute within the current operation.
       }],
InterfaceMethod<
       /*desc=*/[{
         Return the input or output indexing map at index `i`.
       }],
       /*retTy=*/"AffineMap",
       /*methodName=*/"getIndexingMap",
       /*args=*/(ins "unsigned":$i),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        assert(i < getNumInputsAndOutputs());
+        assert(i < $_op.getNumShapedOperands());
         return getIndexingMaps()[i];
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
         Return the input indexing map at index `i`.
       }],
       /*retTy=*/"AffineMap",
InterfaceMethod<
         Return whether the op has only MemRef input and outputs.
       }],
       /*retTy=*/"bool",
       /*methodName=*/"hasBufferSemantics",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
         return this->getOperation()->getNumResults() == 0 &&
-          llvm::all_of(getInputs(),
-                       [](Value v) { return v.getType().isa<MemRefType>(); });
+          llvm::all_of(getShapedOperands(), [](Value v) {
+            return v.getType().template isa<MemRefType>(); });
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
         Return whether the op has only RankedTensor input and outputs.
       }],
       /*retTy=*/"bool",
       /*methodName=*/"hasTensorSemantics",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        auto isTensorType = [](Value v) {
-          return v.getType().isa<RankedTensorType>();
-        };
-        return llvm::all_of(getInputs(), isTensorType) &&
-               llvm::all_of(this->getOperation()->getResults(), isTensorType);
+        return llvm::all_of(getShapedOperands(), [](Value v) {
+          return v.getType().template isa<RankedTensorType>();
+        });
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
         Return whether the op has sparse tensor semantics.
       }],
       /*retTy=*/"bool",
       /*methodName=*/"hasSparseSemantics",
       /*args=*/(ins),
       /*methodBody=*/"",
       /*defaultImplementation=*/[{
-        return $_op->getAttr(getSparseAttrName()).template dyn_cast_or_null<ArrayAttr>() != nullptr;
+        return $_op->getAttr(getSparseAttrName()).
+          template dyn_cast_or_null<ArrayAttr>() != nullptr;
       }]
     >,
     InterfaceMethod<
       /*desc=*/[{
         Return the name registered for this op when lowering to an external
         library call.
       }],
       /*retTy=*/"std::string",
StaticInterfaceMethod<
       /*retTy=*/"std::function<void(Block &)>",
       /*methodName=*/"getRegionBuilder",
       (ins),
       [{ return ConcreteOp::getRegionBuilder(); }]
     >
   ];
 
   let extraClassDeclaration = [{
-    /// Return the flat list of all operand dimension sizes in the order they 
+    /// Return the flat list of all operand dimension sizes in the order they
     /// appear in the operands.
     SmallVector<Value, 4> createFlatListOfOperandDims(OpBuilder &, Location);
 
     /// Create the loop ranges to materialize the computation over the current
     /// operands. This is done by applying `getShapesToLoopsMap` to
     /// `createFlatListOfOperandDims`.
     SmallVector<Range, 4> createLoopRanges(OpBuilder &b, Location loc);
 
     /// Returns all the operands past the inputs, output_buffers and
     /// init_tensors operands. Asserts that these operands are value types to
     /// allow transformations like tiling to just use the values when cloning
     /// `linalgOp`.
     SmallVector<Value, 4> getAssumedNonShapedOperands() {
       unsigned numShapedOperands = getNumShapedOperands();
       unsigned nExtraOperands =
         getOperation()->getNumOperands() - numShapedOperands;
       SmallVector<Value, 4> res;
       res.reserve(nExtraOperands);
       for (unsigned i = 0; i < nExtraOperands; ++i) {
         res.push_back(getOperation()->getOperand(numShapedOperands + i));
         assert((res.back().getType().isSignlessIntOrIndexOrFloat()
-                || res.back().getType().isa<VectorType>()) &&
+                || res.back().getType().template isa<VectorType>()) &&
                "expected scalar or vector type");
       }
       return res;
     }
     //========================================================================//
     // Helper functions to mutate the `operand_segment_sizes` attribute.
     // These are useful when cloning and changing operand types.
     //========================================================================//
     void setNumInputs(unsigned num) { setOperandSegmentAt(0, num); }
     void setNumOutputBuffers(unsigned num) { setOperandSegmentAt(1, num); }
-    void setNumInitTensors(unsigned num) { setOperandSegmentAt(2, num); }
 
     private:
     void setOperandSegmentAt(unsigned idx, unsigned val) {
       auto attr = (*this)->getAttr("operand_segment_sizes")
         .cast<DenseIntElementsAttr>();
       unsigned i = 0;
       auto newAttr = attr.mapValues(IntegerType::get(getContext(), 32),
         [&](const APInt &v) { return (i++ == idx) ? APInt(32, val) : v; });
       getOperation()->setAttr("operand_segment_sizes", newAttr);
     }
   }];
+
+  let verify = [{ return detail::verifyStructuredOpInterface($_op); }];
 }
 
 #endif // LINALG_IR_STRUCTURED_OPS_INTERFACE

mlir/include/mlir/Dialect/Linalg/IR/LinalgTraits.h

-//===- LinalgTraits.h - Linalg Traits ---------------------------*- C++ -*-===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef MLIR_DIALECT_LINALG_LINALGTRAITS_H_
-#define MLIR_DIALECT_LINALG_LINALGTRAITS_H_
-
-#include "mlir/Dialect/Linalg/IR/LinalgTypes.h"
-#include "mlir/Dialect/Utils/StructuredOpsUtils.h"
-#include "mlir/IR/AffineMap.h"
-#include "mlir/IR/BuiltinOps.h"
-#include "mlir/IR/BuiltinTypes.h"
-#include "mlir/IR/OpDefinition.h"
-#include "mlir/Support/LLVM.h"
-
-namespace mlir {
-namespace OpTrait {
-namespace linalg {
-
-/// This class provides the API for ops that are known to have a specified
-/// number of inputs, all passed as operands. Use as a trait as follows:
-///
-///   class DotOp : public Op<DotOp, OpTrait::NInputs<2>::Impl> {
-///
-template <unsigned N> class NInputs {
-public:
-  template <typename ConcreteType>
-  class Impl : public OpTrait::TraitBase<ConcreteType, NInputs<N>::Impl> {
-  public:
-    static unsigned getNumInputs() { return N; }
-  };
-};
-
-/// This class provides the API for ops that are known to not have init tensor
-/// operands. Use as a trait as follows:
-///
-///   class CopyOp : public Op<CopyOp, OpTrait::ZeroInitTensors> {
-///
-template <typename ConcreteType>
-class ZeroInitTensors : public TraitBase<ConcreteType, ZeroInitTensors> {
-public:
-  static unsigned getNumInitTensors() { return 0; }
-};
-
-/// This class provides the API for ops that are known to have a specified
-/// number of outputs, all passed as operands. Use as a trait as follows:
-///
-///   class DotOp : public Op<DotOp, OpTrait::NOutputs<2>::Impl> {
-///
-template <unsigned N> class NOutputs {
-public:
-  template <typename ConcreteType>
-  class Impl : public OpTrait::TraitBase<ConcreteType, NOutputs<N>::Impl> {
-  public:
-    static unsigned getNumOutputs() { return N; }
-  };
-};
-
-/// This class provides a verifier for structured ops that are known to operate
-/// on buffers or tensors. This trait must be used in conjunction with an op
-/// definition or a trait that provides the methods `getNumInputs` and
-/// `getNumOutputs`. Use as a trait as follows:
-///
-///   class DotOp : public Op<DotOp, OpTrait::StructuredOpTraits> {
-///
-template <typename ConcreteType>
-class StructuredOpTraits
-    : public OpTrait::TraitBase<ConcreteType, StructuredOpTraits> {
-public:
-  static LogicalResult verifyTrait(Operation *op) {
-    ConcreteType concreteOp = cast<ConcreteType>(op);
-    auto nOperands = concreteOp.getNumInputsAndOutputBuffers();
-    if (failed(OpTrait::impl::verifyAtLeastNOperands(op, nOperands)))
-      return failure();
-    if (op->getNumResults() > concreteOp.getNumOutputs())
-      return op->emitError("unexpected #results > #outputs");
-    return success();
-  }
-};
-
-/// This class provides a verifier for structured ops that are known to operate
-/// on buffers or tensors and that support `ins`, `outs` and `init` arguments.
-/// This trait must be used in conjunction with an op definition or a trait that
-/// provides the methods `getNumInputs` and `getNumOutputs`.
-///
-/// Use as a trait as follows:
-///
-///   class MatmulOp : public Op<MatmulOp, OpTrait::NamedStructuredOpTrait> {
-///
-template <typename ConcreteType>
-class NamedStructuredOpTrait
-    : public OpTrait::TraitBase<ConcreteType, NamedStructuredOpTrait> {
-public:
-  unsigned getNumInputs() {
-    return cast<ConcreteType>(this->getOperation()).inputs().size();
-  }
-  unsigned getNumInitTensors() {
-    return cast<ConcreteType>(this->getOperation()).init_tensors().size();
-  }
-  unsigned getNumOutputs() {
-    ConcreteType concreteOp = cast<ConcreteType>(this->getOperation());
-    return concreteOp.output_buffers().size() +
-           concreteOp.result_tensors().size();
-  }
-  static LogicalResult verifyTrait(Operation *op) {
-    ConcreteType concreteOp = cast<ConcreteType>(op);
-    unsigned nInputAndBufferOperands =
-        concreteOp.getNumInputsAndOutputBuffers();
-    if (failed(
-            OpTrait::impl::verifyAtLeastNOperands(op, nInputAndBufferOperands)))
-      return failure();
-
-    SmallVector<AffineExpr, 4> redDims;
-    concreteOp.getReductionDims(redDims);
-    // If no result and no reduction, only check there is no init tensor and we
-    // are done.
-    if (redDims.empty() || op->getNumResults() == 0) {
-      if (!concreteOp.init_tensors().empty())
-        return op->emitError("expected empty `init` when op has no "
-                             "results or no reduction dims");
-      return success();
-    }
-
-    // Only a single tensor result supported atm.
-    if (op->getNumResults() != 1)
-      return op->emitError(
-          "expected single tensor result when reduction present");
-
-    if (concreteOp.init_tensors().size() != op->getNumResults())
-      return op->emitError(
-          "expected #init tensors to match #results when reduction present");
-
-    for (unsigned idx = 0, e = op->getNumResults(); idx < e; ++idx)
-      if (concreteOp.init_tensors()[idx].getType() != op->getResultTypes()[idx])
-        return op->emitError("expected init tensor #")
-               << idx << " of the same type as result #" << idx;
-
-    // Output tensor indexing map may not depend on reduction index.
-    // TODO: this is not yet tested. Add a test when linalg.generic switches to
-    // this representation.
-    for (unsigned idx = 0, e = concreteOp.getNumOutputs(); idx < e; ++idx) {
-      AffineMap outputMap = concreteOp.getOutputIndexingMap(idx);
-      for (auto expr : outputMap.getResults()) {
-        for (auto dim : redDims) {
-          unsigned pos = dim.cast<AffineDimExpr>().getPosition();
-          if (expr.isFunctionOfDim(pos))
-            return op->emitError(
-                       "unexpected single tensor output indexing map ")
-                   << "is function of reduction dim @" << pos;
-        }
-      }
-    }
-
-    return success();
-  }
-};
-
-} // namespace linalg
-} // namespace OpTrait
-} // namespace mlir
-
-#endif // MLIR_DIALECT_LINALG_LINALGTRAITS_H_

mlir/include/mlir/IR/OpBase.td

 
 def AnyStridedMemRef : StridedMemRefOf<[AnyType]>;
 
 class AnyStridedMemRefOfRank<int rank> :
   Type<And<[AnyStridedMemRef.predicate,
             MemRefRankOf<[AnyType], [rank]>.predicate]>,
        AnyStridedMemRef.description # " of rank " # rank>;
 
+class StridedMemRefRankOf<list<Type> allowedTypes, list<int> ranks> :
+    Type<And<[MemRefOf<allowedTypes>.predicate, HasAnyRankOfPred<ranks>]>,
+         StrJoin<!foreach(rank, ranks, rank # "D"), "/">.result # " " #
+         MemRefOf<allowedTypes>.description>;
+
 // This represents a generic tuple without any constraints on element type.
 def AnyTuple : Type<IsTupleTypePred, "tuple">;
 
 // A container type that has other types embedded in it, but (unlike
 // ContainerType) can hold elements with a mix of types. Requires a call that
 // produces a list of all elements' types.
 class MixedContainerType<Type etype, Pred containerPred, code elementTypesCall,
                          string descr> :

mlir/integration_test/Dialect/Linalg/CPU/test-tensor-matmul.mlir

 func @main() {
   %A = constant dense<[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]> : tensor<2x3xf32>
   %B = constant dense<[[1.0, 2.0, 3.0, 4.0],
                        [5.0, 6.0, 7.0, 8.0],
                        [9.0, 10.0, 11.0, 12.0]]> : tensor<3x4xf32>
   %C = constant dense<1000.0> : tensor<2x4xf32>
 
   %D = linalg.matmul ins(%A, %B: tensor<2x3xf32>, tensor<3x4xf32>)
-                     init(%C: tensor<2x4xf32>) -> tensor<2x4xf32>
+                     outs(%C: tensor<2x4xf32>) -> tensor<2x4xf32>
 
   %unranked = tensor.cast %D : tensor<2x4xf32> to tensor<*xf32>
   call @print_memref_f32(%unranked) : (tensor<*xf32>) -> ()
 
   //      CHECK: Unranked Memref base@ = {{0x[-9a-f]*}}
   // CHECK-SAME: rank = 2 offset = 0 sizes = [2, 4] strides = [4, 1] data =
   // CHECK-NEXT: [1038,   1044,   1050,   1056]
   // CHECK-NEXT: [1083,   1098,   1113,   1128]
 
   return
 }
 
 func private @print_memref_f32(%ptr : tensor<*xf32>)

mlir/lib/Dialect/Linalg/Analysis/DependenceAnalysis.cpp

 //===- DependenceAnalysis.cpp - Dependence analysis on SSA views ----------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file implements view-based alias and dependence analyses.
 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Dialect/Linalg/Analysis/DependenceAnalysis.h"
 #include "mlir/Dialect/Linalg/IR/LinalgOps.h"
 #include "mlir/Dialect/StandardOps/IR/Ops.h"
+#include "mlir/IR/BuiltinOps.h"
 
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 
 #define DEBUG_TYPE "linalg-dependence-analysis"
 
 using namespace mlir;
 using namespace mlir::linalg;
LinalgDependenceGraph::LinalgDependenceGraph(Aliases &aliases,
   for (unsigned i = 0, e = ops.size(); i < e; ++i) {
     for (unsigned j = i + 1; j < e; ++j) {
       addDependencesBetween(ops[i], ops[j]);
     }
   }
 }
 
 void LinalgDependenceGraph::addDependenceElem(DependenceType dt,
-                                              LinalgOpView indexingOpView,
-                                              LinalgOpView dependentOpView) {
+                                              OpOperand *indexingOpView,
+                                              OpOperand *dependentOpView) {
   LLVM_DEBUG(dbgs() << "\nAdd dep type " << getDependenceTypeStr(dt) << ":\t ("
-                    << *indexingOpView.op << ", " << indexingOpView.operandIndex
-                    << ") -> \n\t\t(" << *dependentOpView.op << ", "
-                    << dependentOpView.operandIndex << ")");
-  dependencesFromGraphs[dt][indexingOpView.op].push_back(
+                    << indexingOpView->get() << " @"
+                    << indexingOpView->getOperandNumber() << ") -> \n\t\t("
+                    << dependentOpView->get() << " @"
+                    << dependentOpView->getOperandNumber() << ")");
+  dependencesFromGraphs[dt][indexingOpView->getOwner()].push_back(
       LinalgDependenceGraphElem{dependentOpView, indexingOpView, dt});
-  dependencesIntoGraphs[dt][dependentOpView.op].push_back(
+  dependencesIntoGraphs[dt][dependentOpView->getOwner()].push_back(
       LinalgDependenceGraphElem{indexingOpView, dependentOpView, dt});
 }
 
 LinalgDependenceGraph::dependence_range
 LinalgDependenceGraph::getDependencesFrom(
     LinalgOp src, LinalgDependenceGraph::DependenceType dt) const {
   return getDependencesFrom(src.getOperation(), dt);
 }
LinalgDependenceGraph::getDependencesInto(
     Operation *dst, LinalgDependenceGraph::DependenceType dt) const {
   auto iter = dependencesIntoGraphs[dt].find(dst);
   if (iter == dependencesIntoGraphs[dt].end())
     return llvm::make_range(nullptr, nullptr);
   return llvm::make_range(iter->second.begin(), iter->second.end());
 }
 
 void LinalgDependenceGraph::addDependencesBetween(LinalgOp src, LinalgOp dst) {
-  for (auto srcView : llvm::enumerate(src.getOutputBuffers())) { // W
-    unsigned srcIndex =
-        src.getOperandIndexForOutputIndex(srcView.index()).getValue();
+  for (OpOperand *srcOpOperand : src.getOutputBuffersOpOperands()) { // W
     // RAW graph
-    for (auto dstView : llvm::enumerate(dst.getInputBuffers())) { // R
-      if (aliases.alias(srcView.value(),
-                        dstView.value())) { // if alias, fill RAW
-        unsigned dstIndex =
-            dst.getOperandIndexForInputIndex(dstView.index()).getValue();
-        addDependenceElem(DependenceType::RAW,
-                          LinalgOpView{src.getOperation(), srcIndex},
-                          LinalgOpView{dst.getOperation(), dstIndex});
-      }
-    }
+    for (OpOperand *dstOpOperand : dst.getInputBuffersOpOperands()) // R
+      if (aliases.alias(srcOpOperand->get(), dstOpOperand->get()))  // RAW alias
+        addDependenceElem(DependenceType::RAW, srcOpOperand, dstOpOperand);
     // WAW graph
-    for (auto dstView : llvm::enumerate(dst.getOutputBuffers())) { // W
-      if (aliases.alias(srcView.value(),
-                        dstView.value())) { // if alias, fill WAW
-        unsigned dstIndex =
-            dst.getOperandIndexForOutputIndex(dstView.index()).getValue();
-        addDependenceElem(DependenceType::WAW,
-                          LinalgOpView{src.getOperation(), srcIndex},
-                          LinalgOpView{dst.getOperation(), dstIndex});
-      }
-    }
-  }
-  for (auto srcView : llvm::enumerate(src.getInputBuffers())) { // R
-    unsigned srcIndex =
-        src.getOperandIndexForInputIndex(srcView.index()).getValue();
+    for (OpOperand *dstOpOperand : dst.getOutputBuffersOpOperands()) // W
+      if (aliases.alias(srcOpOperand->get(), dstOpOperand->get())) // WAW alias
+        addDependenceElem(DependenceType::WAW, srcOpOperand, dstOpOperand);
+  for (OpOperand *srcOpOperand : src.getInputBuffersOpOperands()) { // R
     // RAR graph
-    for (auto dstView : llvm::enumerate(dst.getInputBuffers())) { // R
-      if (aliases.alias(srcView.value(),
-                        dstView.value())) { // if alias, fill RAR
-        unsigned dstIndex =
-            dst.getOperandIndexForInputIndex(dstView.index()).getValue();
-        addDependenceElem(DependenceType::RAR,
-                          LinalgOpView{src.getOperation(), srcIndex},
-                          LinalgOpView{dst.getOperation(), dstIndex});
-      }
-    }
+    for (OpOperand *dstOpOperand : dst.getInputBuffersOpOperands()) // R
+      if (aliases.alias(srcOpOperand->get(), dstOpOperand->get()))  // RAR alias
+        addDependenceElem(DependenceType::RAR, srcOpOperand, dstOpOperand);
     // WAR graph
-    for (auto dstView : llvm::enumerate(dst.getOutputBuffers())) { // W
-      if (aliases.alias(srcView.value(),
-                        dstView.value())) { // if alias, fill WAR
-        unsigned dstIndex =
-            dst.getOperandIndexForOutputIndex(dstView.index()).getValue();
-        addDependenceElem(DependenceType::WAR,
-                          LinalgOpView{src.getOperation(), srcIndex},
-                          LinalgOpView{dst.getOperation(), dstIndex});
-      }
-    }
+    for (OpOperand *dstOpOperand : dst.getOutputBuffersOpOperands()) // W
+      if (aliases.alias(srcOpOperand->get(), dstOpOperand->get())) // WAR alias
+        addDependenceElem(DependenceType::WAR, srcOpOperand, dstOpOperand);
   }
 }
 
 SmallVector<Operation *, 8>
 LinalgDependenceGraph::findCoveringDependences(LinalgOp srcLinalgOp,
                                                LinalgOp dstLinalgOp) const {
   return findOperationsWithCoveringDependences(
       srcLinalgOp, dstLinalgOp, nullptr,
LinalgDependenceGraph::findOperationsWithCoveringDependences(
   assert(srcPos < dstPos && "expected dst after src in IR traversal order");
 
   SmallVector<Operation *, 8> res;
   // Consider an intermediate interleaved `interim` op, look for any dependence
   // to an aliasing view on a src -> op -> dst path.
   // TODO: we are not considering paths yet, just interleaved positions.
   for (auto dt : types) {
     for (auto dependence : getDependencesFrom(src, dt)) {
-      auto interimPos = linalgOpPositions.lookup(dependence.dependentOpView.op);
+      auto interimPos =
+          linalgOpPositions.lookup(dependence.dependentOpView->getOwner());
       // Skip if not interleaved.
       if (interimPos >= dstPos || interimPos <= srcPos)
         continue;
-      linalg::LinalgOp consumer =
-          cast<linalg::LinalgOp>(dependence.indexingOpView.op);
-      Value consumerView =
-          consumer.getShapedOperand(dependence.indexingOpView.operandIndex);
+      Value consumerView = dependence.indexingOpView->get();
       if (view && !aliases.alias(view, consumerView))
         continue;
-      auto *op = dependence.dependentOpView.op;
+      auto *op = dependence.dependentOpView->getOwner();
       LLVM_DEBUG(dbgs() << "\n***Found covering dependence of type "
                         << getDependenceTypeStr(dt) << ": " << *src << " -> "
                         << *op << " on " << consumerView);
       res.push_back(op);
     }
   }
   return res;
 }
 
 bool LinalgDependenceGraph::hasDependenceFrom(
     LinalgOp srcLinalgOp, LinalgOp dstLinalgOp,
     ArrayRef<LinalgDependenceGraph::DependenceType> depTypes) const {
-  for (auto dep : depTypes) {
-    for (auto dependence : getDependencesInto(dstLinalgOp, dep)) {
-      if (dependence.dependentOpView.op == srcLinalgOp)
+  for (auto dep : depTypes)
+    for (auto dependence : getDependencesInto(dstLinalgOp, dep))
+      if (dependence.dependentOpView->getOwner() == srcLinalgOp)
         return true;
-    }
-  }
   return false;
 }
 
 bool LinalgDependenceGraph::hasDependentOperationsFrom(
     LinalgOp linalgOp,
     ArrayRef<LinalgDependenceGraph::DependenceType> depTypes) const {
   for (auto dep : depTypes) {
     if (!getDependencesFrom(linalgOp, dep).empty())

mlir/lib/Dialect/Linalg/EDSC/Builders.cpp

 using namespace mlir;
 using namespace mlir::edsc;
 using namespace mlir::edsc::intrinsics;
 using namespace mlir::linalg;
 using namespace mlir::scf;
 
 Operation *mlir::edsc::makeGenericLinalgOp(
     ArrayRef<IteratorType> iteratorTypes, ArrayRef<StructuredIndexed> inputs,
-    ArrayRef<StructuredIndexed> outputBuffers, ArrayRef<Value> initTensors,
-    ArrayRef<StructuredIndexed> resultTensorTypes,
+    ArrayRef<StructuredIndexed> outputs, TypeRange resultTensorTypes,
     function_ref<void(ValueRange)> regionBuilder, ArrayRef<Value> otherValues,
     ArrayRef<Attribute> otherAttributes) {
   OpBuilder &builder = edsc::ScopedContext::getBuilderRef();
 
   // Build maps
   SmallVector<SmallVector<AffineExpr, 4>, 4> exprsList;
-  exprsList.reserve(inputs.size() + outputBuffers.size() + initTensors.size());
-  for (auto container : {inputs, outputBuffers, resultTensorTypes})
+  exprsList.reserve(inputs.size() + outputs.size());
+
+  for (auto container : {inputs, outputs})
     for (const StructuredIndexed &s : container)
       exprsList.emplace_back(s.getExprs().begin(), s.getExprs().end());
   auto maps = AffineMap::inferFromExprList(exprsList);
 
-  SmallVector<Type, 4> types;
-  assert(llvm::all_of(resultTensorTypes, [](const StructuredIndexed &s) {
-    return !s.hasValue();
-  }));
-  std::copy(resultTensorTypes.begin(), resultTensorTypes.end(),
-            std::back_inserter(types));
-
-  SmallVector<Value, 4> inputValues, outputBufferValues, initTensorValues;
+  SmallVector<Value, 4> inputValues, outputValues;
   inputValues.reserve(inputs.size());
-  outputBufferValues.reserve(outputBuffers.size());
-  initTensorValues.reserve(initTensors.size());
+  outputValues.reserve(outputs.size());
   std::copy(inputs.begin(), inputs.end(), std::back_inserter(inputValues));
-  std::copy(outputBuffers.begin(), outputBuffers.end(),
-            std::back_inserter(outputBufferValues));
-  std::copy(initTensors.begin(), initTensors.end(),
-            std::back_inserter(initTensorValues));
+  std::copy(outputs.begin(), outputs.end(), std::back_inserter(outputValues));
 
   auto iteratorStrTypes =
       llvm::to_vector<8>(llvm::map_range(iteratorTypes, toString));
   // clang-format off
   auto *op =
       edsc::ScopedContext::getBuilderRef()
           .create<linalg::GenericOp>(
               edsc::ScopedContext::getLocation(),
-              types,
+              resultTensorTypes,
               inputValues,
-              outputBufferValues,
-              initTensorValues,
+              outputValues,
               builder.getAffineMapArrayAttr(maps),
               builder.getStrArrayAttr(iteratorStrTypes),
               StringAttr() /*doc*/,
               StringAttr() /*library_call*/,
               ArrayAttr() /*sparse*/
               /* TODO: other attributes in op */
               )
           .getOperation();
   // clang-format on
 
   using namespace edsc;
   SmallVector<Type, 4> blockTypes;
-  blockTypes.reserve(inputs.size() + outputBuffers.size() + initTensors.size());
-  for (auto container : {inputs, outputBuffers})
+  blockTypes.reserve(inputs.size() + outputs.size());
+  for (auto container : {inputs, outputs})
     for (const StructuredIndexed &s : container)
       blockTypes.push_back(getElementTypeOrSelf(s.getType()));
-  for (Value v : initTensors)
-    blockTypes.push_back(getElementTypeOrSelf(v.getType()));
 
   assert(op->getNumRegions() == 1);
   assert(op->getRegion(0).empty());
   OpBuilder opBuilder(op);
   ScopedContext scope(opBuilder, op->getLoc());
   buildInNewBlock(op->getRegion(0), blockTypes, regionBuilder);
   assert(llvm::hasSingleElement(op->getRegion(0)));
   return op;
Operation *mlir::edsc::ops::linalg_generic_pointwise(
   SmallVector<IteratorType, 4> iterTypes(O.getExprs().size(),
                                          IteratorType::Parallel);
   auto fun = [&unaryOp](ValueRange args) {
     assert(!args.empty() && "expected >= 1 block arguments");
     Value a(args[0]);
     linalg_yield(unaryOp(a));
   };
   if (O.getType().isa<RankedTensorType>())
-    return makeGenericLinalgOp(iterTypes, /*inputs=*/{I}, /*outputBuffers=*/{},
-                               /*initTensors=*/{}, /*resultTensorTypes=*/{O},
-                               fun);
-  return makeGenericLinalgOp(iterTypes, /*inputs=*/{I}, /*outputBuffers=*/{O},
-                             /*initTensors=*/{}, /*resultTensorTypes=*/{}, fun);
+    return makeGenericLinalgOp(iterTypes, /*inputs=*/{I}, /*outputs=*/{O},
+                               /*resultTensorTypes=*/{O}, fun);
+  return makeGenericLinalgOp(iterTypes, /*inputs=*/{I}, /*outputs=*/{O},
+                             /*resultTensorTypes=*/{}, fun);
 }
 
 Operation *mlir::edsc::ops::linalg_generic_pointwise_tanh(StructuredIndexed I,
                                                           StructuredIndexed O) {
   UnaryPointwiseOpBuilder unOp([](Value a) -> Value { return std_tanh(a); });
   return linalg_generic_pointwise(unOp, I, O);
 }
 
 /// Binary pointwise operation (with broadcast) entry point.
 Operation *mlir::edsc::ops::linalg_generic_pointwise(
     BinaryPointwiseOpBuilder binaryOp, StructuredIndexed I1,
     StructuredIndexed I2, StructuredIndexed O) {
   SmallVector<IteratorType, 4> iterTypes(O.getExprs().size(),
                                          IteratorType::Parallel);
   auto fun = [&binaryOp](ValueRange args) {
     assert(args.size() >= 2 && "expected >= 2 block arguments");
     Value a(args[0]), b(args[1]);
     linalg_yield(binaryOp(a, b));
   };
   if (O.getType().isa<RankedTensorType>())
-    return makeGenericLinalgOp(
-        iterTypes, /*inputs=*/{I1, I2}, /*outputBuffers=*/{},
-        /*initTensors=*/{}, /*resultTensorTypes=*/{O}, fun);
+    return makeGenericLinalgOp(iterTypes, /*inputs=*/{I1, I2}, /*outputs=*/{O},
+                               /*resultTensorTypes=*/{O}, fun);
   return makeGenericLinalgOp(iterTypes, /*inputs=*/{I1, I2},
-                             /*outputBuffers=*/{O},
-                             /*initTensors=*/{}, /*resultTensorTypes=*/{}, fun);
+                             /*outputs=*/{O}, /*resultTensorTypes=*/{}, fun);
 }
 
 Operation *mlir::edsc::ops::linalg_generic_pointwise_add(StructuredIndexed I1,
                                                          StructuredIndexed I2,
                                                          StructuredIndexed O) {
   using edsc::op::operator+;
   BinaryPointwiseOpBuilder binOp(
       [](Value a, Value b) -> Value { return a + b; });
mlir::edsc::ops::linalg_generic_matmul(Value vA, Value vB, Value vC,
                                        MatmulRegionBuilder regionBuilder) {
   // clang-format off
   AffineExpr m, n, k;
   bindDims(ScopedContext::getContext(), m, n, k);
   StructuredIndexed A(vA), B(vB), C(vC);
   return makeGenericLinalgOp(
     {IteratorType::Parallel, IteratorType::Parallel, IteratorType::Reduction},
     /*inputs=*/{A({m, k}), B({k, n})},
-    /*outputBuffers=*/{C({m, n})},
-    /*initTensors=*/{},
+    /*outputs=*/{C({m, n})},
     /*resultTensorTypes=*/{},
     regionBuilder);
   // clang-format on
 }
 
 Operation *
 mlir::edsc::ops::linalg_generic_matmul(Value vA, Value vB, Value vC,
                                        RankedTensorType tD,
                                        MatmulRegionBuilder regionBuilder) {
   // clang-format off
   AffineExpr m, n, k;
   bindDims(ScopedContext::getContext(), m, n, k);
   StructuredIndexed A(vA), B(vB), C(vC), D(tD);
   return makeGenericLinalgOp(
     {IteratorType::Parallel, IteratorType::Parallel, IteratorType::Reduction},
     /*inputs=*/{A({m, k}), B({k, n})},
-    /*outputBuffers=*/{},
-    /*initTensors=*/{C({m, n})},
+    /*outputs=*/{C({m, n})},
     /*resultTensorTypes=*/{D({m, n})},
     regionBuilder);
   // clang-format on
 }
 
 Operation *mlir::edsc::ops::linalg_generic_conv_nhwc(Value vI, Value vW,
                                                      Value vO,
                                                      ArrayRef<int> strides,
return makeGenericLinalgOp(
     /*inputs=*/{
       I({b,
          // Roundtrip to flattened form to serve as canonicalization and ensure
          // consistent ordering of subexpressions.
          simplifyAffineExpr(s[0] * h + d[0] * kh, numDims, 0),
          simplifyAffineExpr(s[1] * w + d[1] * kw, numDims, 0),
          c}),
       W({kh, kw, c, f}) },
-    /*outputBuffers=*/{ O({b, h, w, f}) },
-    /*initTensors=*/{},
+    /*outputs=*/{ O({b, h, w, f}) },
     /*resultTensorTypes=*/{},
     macRegionBuilder);
   // clang-format on
 }
 
 Operation *mlir::edsc::ops::linalg_generic_dilated_conv_nhwc(
     Value vI, Value vW, Value vO, int depth_multiplier, ArrayRef<int> strides,
     ArrayRef<int> dilations) {
return makeGenericLinalgOp(
     /*inputs=*/{
       I({b,
          // Roundtrip to flattened form to serve as canonicalization and ensure
          // consistent ordering of subexpressions.
          simplifyAffineExpr(s[0] * h + d[0] * kh, numDims, 0),
          simplifyAffineExpr(s[1] * w + d[1] * kw, numDims, 0),
          c}),
       W({kh, kw, c, dm})},
-    /*outputBuffers=*/{
+    /*outputs=*/{
       O({b, h, w, simplifyAffineExpr(c * depth_multiplier + dm, numDims, 0)})},
-    /*initTensors=*/{},
     /*resultTensorTypes=*/{},
     macRegionBuilder);
   // clang-format on
 }

mlir/lib/Dialect/Linalg/IR/LinalgOps.cpp

if (auto d = result.dyn_cast<AffineDimExpr>()) {
       res[d.getPosition()] = Range{zeroVal, viewSizes[idx], oneVal};
     }
   }
   return res;
 }
 
 /// Forward declarations.
 template <typename NamedStructuredOpType>
-static void buildNamedStructuredOpRegionAndAttributes(
-    OpBuilder &opBuilder, OperationState &result, TypeRange inputTypes,
-    TypeRange outputBufferTypes, TypeRange initTensorTypes,
-    TypeRange resultTypes);
+static void buildNamedStructuredOpRegionAndAttributes(OpBuilder &opBuilder,
+                                                      OperationState &result,
+                                                      TypeRange inputTypes,
+                                                      TypeRange outputTypes);
 
 static ParseResult
 parseCommonStructuredOpParts(OpAsmParser &parser, OperationState &result,
                              SmallVectorImpl<Type> &inputTypes,
-                             SmallVectorImpl<Type> &outputBufferTypes,
-                             SmallVectorImpl<Type> &initTensorTypes);
+                             SmallVectorImpl<Type> &outputTypes);
 
 template <typename NamedStructuredOpType>
 static ParseResult
 parseNamedStructuredOpRegion(OpAsmParser &parser, Region &region,
-                             TypeRange inputTypes, TypeRange outputBufferTypes,
-                             TypeRange initTensorTypes, TypeRange resultTypes);
+                             TypeRange inputTypes, TypeRange outputTypes);
 static ParseResult
 parseNamedStructuredOpResults(OpAsmParser &parser,
                               SmallVectorImpl<Type> &resultTypes);
 
 template <typename NamedStructuredOpType>
 static ParseResult parseNamedStructuredOp(OpAsmParser &parser,
                                           OperationState &result);
 
 template <typename NamedStructuredOpType>
 static void printCommonStructuredOpParts(OpAsmPrinter &p,
                                          NamedStructuredOpType op);
 
 static void printNamedStructuredOpResults(OpAsmPrinter &p,
                                           TypeRange resultTypes);
 
 template <typename NamedStructuredOpType>
 static void printNamedStructuredOp(OpAsmPrinter &p, NamedStructuredOpType op);
 
-template <typename NamedStructuredOpType>
-static LogicalResult verifyNamedStructuredOp(NamedStructuredOpType op);
-
 /// This is a common class used for patterns of the form
 /// ```
 ///    someop(memrefcast) -> someop
 /// ```
 /// It folds the source of the memref_cast into the root operation directly.
 static LogicalResult foldMemRefCast(Operation *op) {
   bool folded = false;
   for (OpOperand &operand : op->getOpOperands()) {

 // LinalgOps.td), we define an overloaded `print` function and a
 // parse`className` function.
 
 //===----------------------------------------------------------------------===//
 // GenericOps
 //===----------------------------------------------------------------------===//
 void GenericOp::build(
     OpBuilder &builder, OperationState &result, TypeRange resultTensorTypes,
-    ValueRange inputs, ValueRange outputBuffers, ValueRange initTensors,
-    ArrayRef<AffineMap> indexingMaps, ArrayRef<StringRef> iteratorTypes,
-    StringRef doc, StringRef libraryCall,
+    ValueRange inputs, ValueRange outputs, ArrayRef<AffineMap> indexingMaps,
+    ArrayRef<StringRef> iteratorTypes, StringRef doc, StringRef libraryCall,
     function_ref<void(OpBuilder &, Location, ValueRange)> bodyBuild) {
-  build(builder, result, resultTensorTypes, inputs, outputBuffers, initTensors,
+  build(builder, result, resultTensorTypes, inputs, outputs,
         builder.getAffineMapArrayAttr(indexingMaps),
         builder.getStrArrayAttr(iteratorTypes),
         doc.empty() ? StringAttr() : builder.getStringAttr(doc),
         libraryCall.empty() ? StringAttr() : builder.getStringAttr(libraryCall),
         ArrayAttr());
   if (!bodyBuild)
     return;
 
   SmallVector<Type, 4> blockArgTypes;
-  for (ValueRange container : {inputs, outputBuffers, initTensors})
+  for (ValueRange container : {inputs, outputs})
     for (Value v : container)
       blockArgTypes.push_back(v.getType().cast<ShapedType>().getElementType());
 
   OpBuilder::InsertionGuard guard(builder);
   auto &region = *result.regions.front();
   Block *bodyBlock = builder.createBlock(&region, region.end(), blockArgTypes);
   bodyBuild(builder, result.location, bodyBlock->getArguments());
 }
 
 void GenericOp::build(
     OpBuilder &builder, OperationState &result, ValueRange inputs,
-    ValueRange outputBuffers, ArrayRef<AffineMap> indexingMaps,
+    ValueRange outputs, ArrayRef<AffineMap> indexingMaps,
     ArrayRef<StringRef> iteratorTypes, StringRef doc, StringRef libraryCall,
     function_ref<void(OpBuilder &, Location, ValueRange)> bodyBuild) {
-  build(builder, result, TypeRange{}, inputs, outputBuffers, ValueRange{},
-        indexingMaps, iteratorTypes, doc, libraryCall, bodyBuild);
+  build(builder, result, TypeRange{}, inputs, outputs, indexingMaps,
+        iteratorTypes, doc, libraryCall, bodyBuild);
 }
 
 void GenericOp::build(
     OpBuilder &builder, OperationState &result, ValueRange inputs,
-    ValueRange outputBuffers, ArrayRef<AffineMap> indexingMaps,
+    ValueRange outputs, ArrayRef<AffineMap> indexingMaps,
     ArrayRef<StringRef> iteratorTypes,
     function_ref<void(OpBuilder &, Location, ValueRange)> bodyBuild) {
-  build(builder, result, inputs, outputBuffers, indexingMaps, iteratorTypes,
+  build(builder, result, inputs, outputs, indexingMaps, iteratorTypes,
         /*doc=*/"",
         /*libraryCall=*/"", bodyBuild);
 }
 
 void GenericOp::build(
     OpBuilder &builder, OperationState &result, TypeRange resultTensorTypes,
-    ValueRange inputs, ValueRange outputBuffers, ValueRange initTensors,
-    ArrayRef<AffineMap> indexingMaps, ArrayRef<StringRef> iteratorTypes,
+    ValueRange inputs, ValueRange outputs, ArrayRef<AffineMap> indexingMaps,
+    ArrayRef<StringRef> iteratorTypes,
     function_ref<void(OpBuilder &, Location, ValueRange)> bodyBuild) {
-  build(builder, result, resultTensorTypes, inputs, outputBuffers, initTensors,
-        indexingMaps, iteratorTypes,
+  build(builder, result, resultTensorTypes, inputs, outputs, indexingMaps,
+        iteratorTypes,
         /*doc=*/"",
         /*libraryCall=*/"", bodyBuild);
 }
 void IndexedGenericOp::build(
     OpBuilder &builder, OperationState &result, TypeRange resultTensorTypes,
-    ValueRange inputs, ValueRange outputBuffers, ValueRange initTensors,
-    ArrayRef<AffineMap> indexingMaps, ArrayRef<StringRef> iteratorTypes,
-    StringRef doc, StringRef libraryCall,
+    ValueRange inputs, ValueRange outputs, ArrayRef<AffineMap> indexingMaps,
+    ArrayRef<StringRef> iteratorTypes, StringRef doc, StringRef libraryCall,
     function_ref<void(OpBuilder &, Location, ValueRange, ValueRange)>
         bodyBuild) {
-  build(builder, result, resultTensorTypes, inputs, outputBuffers, initTensors,
+  build(builder, result, resultTensorTypes, inputs, outputs,
         builder.getAffineMapArrayAttr(indexingMaps),
         builder.getStrArrayAttr(iteratorTypes),
         doc.empty() ? StringAttr() : builder.getStringAttr(doc),
         libraryCall.empty() ? StringAttr() : builder.getStringAttr(libraryCall),
         ArrayAttr());
   if (!bodyBuild)
     return;
 
   unsigned nLoops = iteratorTypes.size();
   SmallVector<Type, 4> blockArgTypes(nLoops, builder.getIndexType());
-  for (ValueRange container : {inputs, outputBuffers, initTensors})
+  for (ValueRange container : {inputs, outputs})
     for (Value v : container)
       blockArgTypes.push_back(v.getType().cast<ShapedType>().getElementType());
 
   OpBuilder::InsertionGuard guard(builder);
   auto &region = *result.regions.front();
   Block *bodyBlock = builder.createBlock(&region, region.end(), blockArgTypes);
   bodyBuild(builder, result.location,
             bodyBlock->getArguments().take_front(nLoops),
             bodyBlock->getArguments().drop_front(nLoops));
 }
 
 void IndexedGenericOp::build(
     OpBuilder &builder, OperationState &result, ValueRange inputs,
-    ValueRange outputBuffers, ArrayRef<AffineMap> indexingMaps,
+    ValueRange outputs, ArrayRef<AffineMap> indexingMaps,
     ArrayRef<StringRef> iteratorTypes, StringRef doc, StringRef libraryCall,
     function_ref<void(OpBuilder &, Location, ValueRange, ValueRange)>
         bodyBuild) {
-  build(builder, result, TypeRange{}, inputs, outputBuffers, ValueRange{},
-        indexingMaps, iteratorTypes, doc, libraryCall, bodyBuild);
+  build(builder, result, TypeRange{}, inputs, outputs, indexingMaps,
+        iteratorTypes, doc, libraryCall, bodyBuild);
 }
 
 void IndexedGenericOp::build(
     OpBuilder &builder, OperationState &result, ValueRange inputs,
-    ValueRange outputBuffers, ArrayRef<AffineMap> indexingMaps,
+    ValueRange outputs, ArrayRef<AffineMap> indexingMaps,
     ArrayRef<StringRef> iteratorTypes,
     function_ref<void(OpBuilder &, Location, ValueRange, ValueRange)>
         bodyBuild) {
-  build(builder, result, inputs, outputBuffers, indexingMaps, iteratorTypes,
+  build(builder, result, inputs, outputs, indexingMaps, iteratorTypes,
         /*doc=*/"", /*libraryCall=*/"", bodyBuild);
 }
 
 void IndexedGenericOp::build(
     OpBuilder &builder, OperationState &result, TypeRange resultTensorTypes,
-    ValueRange inputs, ValueRange outputBuffers, ValueRange initTensors,
-    ArrayRef<AffineMap> indexingMaps, ArrayRef<StringRef> iteratorTypes,
+    ValueRange inputs, ValueRange outputs, ArrayRef<AffineMap> indexingMaps,
+    ArrayRef<StringRef> iteratorTypes,
     function_ref<void(OpBuilder &, Location, ValueRange, ValueRange)>
         bodyBuild) {
-  build(builder, result, resultTensorTypes, inputs, outputBuffers, initTensors,
-        indexingMaps, iteratorTypes,
+  build(builder, result, resultTensorTypes, inputs, outputs, indexingMaps,
+        iteratorTypes,
         /*doc=*/"",
         /*libraryCall=*/"", bodyBuild);
 }
 
 template <typename GenericOpType>
 static void printGenericOp(OpAsmPrinter &p, GenericOpType op) {
   p << op.getOperationName() << " ";
 
static ParseResult parseGenericOp(OpAsmParser &parser, OperationState &result) {
   // The core linalg traits must contain the information necessary to pass the
   // verifier.
   if (parser.parseAttribute(dictAttr, "_", result.attributes))
     return failure();
   result.attributes.assign(dictAttr.getValue().begin(),
                            dictAttr.getValue().end());
 
   // Parsing is shared with named ops, except for the region.
-  SmallVector<Type, 1> inputTypes, outputBufferTypes, initTensorTypes;
-  if (parseCommonStructuredOpParts(parser, result, inputTypes,
-                                   outputBufferTypes, initTensorTypes))
+  SmallVector<Type, 1> inputTypes, outputTypes;
+  if (parseCommonStructuredOpParts(parser, result, inputTypes, outputTypes))
     return failure();
 
   // Optional attributes may be added.
   if (succeeded(parser.parseOptionalKeyword("attrs")))
     if (failed(parser.parseEqual()) ||
         failed(parser.parseOptionalAttrDict(result.attributes)))
       return failure();
 
static ParseResult parseGenericOp(OpAsmParser &parser, OperationState &result) {
   result.addTypes(outputTensorsTypes);
 
   return success();
 }
 
 static void getGenericEffectsImpl(
     SmallVectorImpl<SideEffects::EffectInstance<MemoryEffects::Effect>>
         &effects,
-    ValueRange results, ValueRange inputBuffers, ValueRange outputBuffers) {
+    ValueRange results, ValueRange inputBuffers, ValueRange outputs) {
   for (Value value : results) {
     effects.emplace_back(MemoryEffects::Allocate::get(), value,
                          SideEffects::DefaultResource::get());
   }
   for (Value value : inputBuffers) {
     effects.emplace_back(MemoryEffects::Read::get(), value,
                          SideEffects::DefaultResource::get());
   }
-  for (Value value : outputBuffers) {
+  for (Value value : outputs) {
     effects.emplace_back(MemoryEffects::Read::get(), value,
                          SideEffects::DefaultResource::get());
     effects.emplace_back(MemoryEffects::Write::get(), value,
                          SideEffects::DefaultResource::get());
   }
 }
 
 void GenericOp::getEffects(
     SmallVectorImpl<SideEffects::EffectInstance<MemoryEffects::Effect>>
         &effects) {
   getGenericEffectsImpl(effects, getOperation()->getResults(),
                         getInputBuffers(), getOutputBuffers());
 }
 
 void IndexedGenericOp::getEffects(
     SmallVectorImpl<SideEffects::EffectInstance<MemoryEffects::Effect>>
         &effects) {
   getGenericEffectsImpl(effects, getOperation()->getResults(),
                         getInputBuffers(), getOutputBuffers());
 }
 
-namespace {
+LogicalResult mlir::linalg::detail::verifyStructuredOpInterface(Operation *op) {
+  LinalgOp linalgOp = cast<LinalgOp>(op);
+  // Expect at least one shaped operand.
+  // This means an op that constructs a tensor out of indices cannot be a
+  // LinalgOp at the moment. For now this will have to be a special op until we
+  // have output shape operands that are not tensors.
+  auto nShapedOperands = linalgOp.getNumShapedOperands();
+  if (nShapedOperands == 0)
+    return linalgOp.emitOpError("expected at least 1 Shaped operand");
+  if (failed(OpTrait::impl::verifyAtLeastNOperands(op, nShapedOperands)))
+    return failure();
+  // Should have at least one output tensor per result tensor.
+  // Can also have outbut buffers that do not correspond to results.
+  if (op->getNumResults() > linalgOp.getNumOutputTensors())
+    return op->emitError("unexpected #results > #outputs");
+
+  // All shaped operands must be indexed.
+  if (linalgOp.indexing_maps().size() != linalgOp.getNumShapedOperands())
+    return linalgOp.emitOpError("expected the number of indexing_map (")
+           << linalgOp.indexing_maps().size()
+           << ") to be equal to the number of shaped operands ("
+           << linalgOp.getNumShapedOperands() << ")";
 
-template <typename GenericOpType>
-struct BlockArgsVerifier {
-  static LogicalResult verify(GenericOpType op, Block &block);
-};
+  SmallVector<AffineMap, 4> indexingMaps;
+  indexingMaps.reserve(linalgOp.indexing_maps().size());
+  for (auto en : llvm::enumerate(linalgOp.indexing_maps())) {
+    auto idx = en.index();
+    auto m = en.value().template cast<AffineMapAttr>().getValue();
+    indexingMaps.push_back(m); // Save reference to map for further checks.
+    auto shapedValue = linalgOp.getShapedType(idx);
 
-template <typename GenericOpType>
-LogicalResult BlockArgsVerifier<GenericOpType>::verify(GenericOpType op,
-                                                       Block &block) {
-  auto nOperands = op.getNumOperands();
-  if (block.getNumArguments() != nOperands)
-    return op.emitOpError("expected number of block arguments to match number "
-                          "of operands");
+    // Symbols disallowed.
+    if (m.getNumSymbols() != 0)
+      return linalgOp.emitOpError("unexpected symbols in indexing_map #")
+             << idx;
 
-  // Note: the number and type of yield values are checked in the YieldOp.
-  auto nInputViews = op.getNumInputs();
-  for (unsigned i = 0; i < nOperands; ++i) {
-    auto viewType = op.getShapedType(i);
-    if (viewType.getElementType() != block.getArgument(i).getType())
-      return op.emitOpError("expected block argument ")
-             << (i + 1) << " of the same type as elemental type of "
-             << ((i < nInputViews) ? "input " : "output ")
-             << "operand: " << viewType;
+    // Domain must be consistent.
+    auto nLoops = linalgOp.getNumLoops();
+    if (m.getNumDims() != nLoops)
+      return linalgOp.emitOpError("expected indexing_map #")
+             << idx << " to have " << nLoops
+             << " dim(s) to match the number of loops";
+
+    if (m.getNumResults() != shapedValue.getRank())
+      return linalgOp.emitOpError("expected shaped value rank (")
+             << shapedValue.getRank()
+             << ") to match the result rank of indexing_map #" << idx << " ("
+             << m.getNumResults() << ")";
+  }
+
+  SmallVector<AffineExpr, 4> redDims;
+  linalgOp.getReductionDims(redDims);
+
+  // Simplifying assumption: either full tensor or full buffer mode.
+  // This allows simpler verification of output operands vs result types
+  // without premature tracking of which operand is what in mixed-mode.
+  // TODO: relax when mixed-mode needs to pass verification.
+  if (linalgOp.getNumOutputBuffers() > 0 && linalgOp.getNumOutputTensors() > 0)
+    return op->emitError("expected output operands to all have tensor type or "
+                         "all have buffer type");
+
+  for (auto it :
+       llvm::zip(linalgOp.getOutputOpOperands(), op->getResultTypes())) {
+    if (!std::get<0>(it).get().getType().isa<RankedTensorType>())
+      continue;
+    if (std::get<0>(it).get().getType() != std::get<1>(it))
+      return op->emitError("expected type of operand #")
+             << std::get<0>(it).getOperandNumber() << " ("
+             << std::get<0>(it).get().getType() << ")"
+             << " to match type of corresponding result (" << std::get<1>(it)
+             << ")";
+  }
+
+  // Output tensor indexing map may not depend on reduction indices.
+  for (OpOperand &opOperand : linalgOp.getOutputOpOperands()) {
+    AffineMap outputMap = linalgOp.getIndexingMap(opOperand.getOperandNumber());
+    for (auto expr : outputMap.getResults()) {
+      for (auto dim : redDims) {
+        unsigned pos = dim.cast<AffineDimExpr>().getPosition();
+        if (expr.isFunctionOfDim(pos)) {
+          std::string exprStr;
+          {
+            llvm::raw_string_ostream os(exprStr);
+            os << expr;
+          }
+          return op->emitError(
+                     "unexpected output tensor expression in indexing map #")
+                 << (opOperand.getOperandNumber() - linalgOp.getNumInputs())
+                 << " a.k.a '" << exprStr
+                 << "' is function of reduction iterator 'd" << pos << "'";
+        }
       }
+    }
+  }
+
+  // Named ops that are defined manually have a region builder but no region at
+  // this time. Assume the region is well-formed by specification.
+  // TODO: use linalg-ods-gen for all ops when we have enough expressive power.
+  if (linalgOp->getNumRegions() == 0) {
+    assert(!linalgOp.getRegionBuilder() && "regionBuilder but no region");
     return success();
   }
 
-template <>
-LogicalResult BlockArgsVerifier<IndexedGenericOp>::verify(IndexedGenericOp op,
-                                                          Block &block) {
-  auto nInputViews = op.getNumInputs();
-  auto nLoops = op.getNumLoops();
-  auto nOperands = op.getNumOperands();
-  if (block.getNumArguments() != nOperands + nLoops)
-    return op.emitOpError(
-        "expected number of block arguments to match number of operands + "
-        "number of loops");
+  auto &region = linalgOp->getRegion(0);
+  if (linalgOp->getNumRegions() > 1 || !llvm::hasSingleElement(region))
+    return op->emitOpError("expected 1 region with 1 block");
+
+  if (!linalgOp.getShapesToLoopsMap())
+    return op->emitOpError("expected the shape-to-loops map to be non-null");
+
+  // Simplifying assumption: bbargs match 1-1 with shape operands elemental
+  // types.
+  // TODO: once ranked shape types are plugged in, we may want to drop the
+  // corresponding bbargs, that can never be read from. This will be subject to
+  // consistency discussions (i.e. what to do with output tensors whose bbarg is
+  // not used).
+  Block &block = linalgOp->getRegion(0).front();
+  unsigned numBBIvs = linalgOp.getNumPayloadInductionVariables();
+
+  if (linalgOp.getNumShapedOperands() + numBBIvs != block.getNumArguments())
+    return op->emitError("expected as many non-induction variable region "
+                         "arguments as the number of shaped operands");
 
   // Note: the number and type of yield values are checked in the YieldOp.
-  for (unsigned i = 0; i < nLoops; ++i)
+  for (unsigned i = 0; i < numBBIvs; ++i)
     if (!block.getArgument(i).getType().isIndex())
-      return op.emitOpError("expected block argument ")
-             << (i + 1) << " to be an index";
+      return op->emitOpError("expected index block argument #") << i;
 
-  for (unsigned i = 0; i < nOperands; ++i) {
-    unsigned memrefArgIndex = i + nLoops;
-    auto viewType = op.getShapedType(i);
-    if (viewType.getElementType() !=
-        block.getArgument(memrefArgIndex).getType())
-      return op.emitOpError("expected block argument ")
-             << (memrefArgIndex + 1)
-             << " of the same type as elemental type of "
-             << ((i < nInputViews) ? "input " : "output ")
-             << "operand: " << viewType;
+  unsigned idx = 0;
+  for (auto it : llvm::zip(linalgOp.getShapedOperandTypes(),
+                           block.getArguments().drop_front(numBBIvs))) {
+    if (std::get<0>(it).getElementType() != std::get<1>(it).getType())
+      return op->emitError("expected type of bb argument #")
+             << (idx + numBBIvs) << " (" << std::get<1>(it).getType() << ")"
+             << " to match element type of corresponding shaped operand ("
+             << std::get<0>(it).getElementType() << ")";
+    ++idx;
   }
+
   return success();
 }
 
+namespace {
+
 template <typename GenericOpType>
 struct AnnotationsVerifier {
   static LogicalResult verify(GenericOpType op) { return success(); }
 };
 
 template <>
 LogicalResult AnnotationsVerifier<GenericOp>::verify(GenericOp op) {
   ArrayAttr sparseAttr = op.sparseAttr();
   if (!sparseAttr)
     return success();
   // Verify consistency of sparse annotations.
   if (!op.hasTensorSemantics())
     return op.emitOpError("expected sparse annotations on tensors only");
   if (op.getNumOutputs() != 1)
     return op.emitOpError("expected single output tensor");
-  unsigned numTensors = op.getNumInputsAndOutputs();
+  unsigned numTensors = op.getNumShapedOperands();
   if (sparseAttr.size() != numTensors)
     return op.emitOpError("expected one sparse annotation for each tensor");
   for (unsigned t = 0; t < numTensors; t++) {
     auto dimAttr = sparseAttr[t].dyn_cast_or_null<ArrayAttr>();
     if (!dimAttr)
       return op.emitOpError("expected sparse annotation array for tensor ")
              << t;
     unsigned rank = op.getShapedType(t).getRank();
LogicalResult AnnotationsVerifier<GenericOp>::verify(GenericOp op) {
   }
   return success();
 }
 
 } // namespace
 
 template <typename GenericOpType>
 static LogicalResult verifyGenericOp(GenericOpType op) {
-  auto nLoops = op.getNumLoops();
-
-  if (op.inputs().size() + op.output_buffers().size() +
-          op.init_tensors().size() + op.getNumResults() ==
-      0)
-    return op.emitOpError("expected at least 1 Shaped operand or return");
-
-  auto &region = op.region();
-  if (!llvm::hasSingleElement(region))
-    return op.emitOpError("expected region with 1 block");
-  if (failed(BlockArgsVerifier<GenericOpType>::verify(op, region.front())))
-    return failure();
-
-  if (op.indexing_maps().size() != op.getNumInputsAndOutputs())
-    return op.emitOpError("expected the number of indexing_map (")
-           << op.indexing_maps().size()
-           << ") to be equal to the number of inputs and outputs ("
-           << op.getNumInputsAndOutputs() << ")";
-
-  SmallVector<AffineMap, 4> indexingMaps;
-  indexingMaps.reserve(op.indexing_maps().size());
-  for (auto en : llvm::enumerate(op.indexing_maps())) {
-    auto idx = en.index();
-    auto m = en.value().template cast<AffineMapAttr>().getValue();
-    indexingMaps.push_back(m); // Save reference to map for further checks.
-    auto view = op.getShapedType(idx);
-
-    if (m.getNumSymbols() != 0)
-      return op.emitOpError("unexpected symbols in indexing_map #") << idx;
-
-    if (m.getNumDims() != nLoops)
-      return op.emitOpError("expected indexing_map #")
-             << idx << " to have " << nLoops
-             << " dim(s) to match the number of loops";
-
-    if (m.getNumResults() != view.getRank())
-      return op.emitOpError("expected indexing_map #")
-             << idx << " results to match view rank: " << view;
-  }
-
-  if (!op.getShapesToLoopsMap())
-    return op.emitOpError("expected the shape-to-loops map to be non-null");
-
   if (failed(AnnotationsVerifier<GenericOpType>::verify(op)))
     return failure();
 
   return success();
 }
 
 static LogicalResult verify(GenericOp op) { return verifyGenericOp(op); }
 
static LogicalResult verify(ConvOp op) {
   auto oType = op.output().getType().cast<MemRefType>();
   auto fType = op.filter().getType().cast<MemRefType>();
   auto iType = op.input().getType().cast<MemRefType>();
   if (oType.getElementType() != iType.getElementType() ||
       oType.getElementType() != fType.getElementType())
     return op.emitOpError("expects memref elemental types to match");
   if (oType.getRank() != iType.getRank() || oType.getRank() != fType.getRank())
     return op.emitOpError("expects memref ranks to match");
-  if (oType.getRank() <= 2)
-    return op.emitOpError("expects memref ranks to be greater than 2");
   if (auto strides = op.strides()) {
     if (failed(
             verifyStrideOrDilation(op, strides->getValue(), /*isStride=*/true)))
       return failure();
   }
   if (auto dilations = op.dilations()) {
     if (failed(verifyStrideOrDilation(op, dilations->getValue(),
                                       /*isStride=*/false)))

 
 //===----------------------------------------------------------------------===//
 // Auto-generated Linalg named ops.
 //===----------------------------------------------------------------------===//
 
 template <typename NamedStructuredOpType>
 static void buildNamedStructuredOpRegionAndAttributesImpl(
     OpBuilder &opBuilder, Region &region, TypeRange inputTypes,
-    TypeRange outputBufferTypes, TypeRange initTensorTypes,
-    TypeRange resultTypes,
+    TypeRange outputTypes,
     std::function<void(unsigned, unsigned)> errorHandler) {
   // TODO: atm all operands go through getElementTypeOrSelf,
   // reconsider when we have evidence we need to.
   SmallVector<Type, 8> argTypes;
-  for (auto containers : {inputTypes, outputBufferTypes, resultTypes})
+  for (auto containers : {inputTypes, outputTypes})
     for (auto t : containers)
       argTypes.push_back(getElementTypeOrSelf(t));
 
   // RAII.
   OpBuilder::InsertionGuard guard(opBuilder);
   Block *body = opBuilder.createBlock(&region, {}, argTypes);
   unsigned actual = body->getNumArguments();
   unsigned expected = NamedStructuredOpType::getNumRegionArgs();
   if (expected != actual)
     return errorHandler(expected, actual);
 
   opBuilder.setInsertionPointToStart(body);
   mlir::edsc::ScopedContext scope(opBuilder, opBuilder.getUnknownLoc());
   NamedStructuredOpType::regionBuilder(*body);
 
   // indexing_maps is an auto-generated method.
 
   // iterator_types is an auto-generated method.
 }
 
 template <typename NamedStructuredOpType>
 void buildNamedStructuredOpRegionAndAttributes(OpBuilder &opBuilder,
                                                OperationState &result,
                                                TypeRange inputTypes,
-                                               TypeRange outputBufferTypes,
-                                               TypeRange initTensorTypes,
-                                               TypeRange resultTypes) {
+                                               TypeRange outputTypes) {
   Region &region = *result.addRegion();
   buildNamedStructuredOpRegionAndAttributesImpl<NamedStructuredOpType>(
-      opBuilder, region, inputTypes, outputBufferTypes, initTensorTypes,
-      resultTypes, [&](unsigned expected, unsigned actual) {
+      opBuilder, region, inputTypes, outputTypes,
+      [&](unsigned expected, unsigned actual) {
         llvm::errs() << "region expects " << expected << " args, got "
                      << actual;
         assert(expected != actual && "incorrect number of arguments");
       });
 }
 
 template <typename NamedStructuredOpType>
 static ParseResult
 parseNamedStructuredOpRegion(OpAsmParser &parser, Region &region,
-                             TypeRange inputTypes, TypeRange outputBufferTypes,
-                             TypeRange initTensorTypes, TypeRange resultTypes) {
+                             TypeRange inputTypes, TypeRange outputTypes) {
   ParseResult res = success();
   OpBuilder opBuilder(parser.getBuilder().getContext());
   buildNamedStructuredOpRegionAndAttributesImpl<NamedStructuredOpType>(
-      opBuilder, region, inputTypes, outputBufferTypes, initTensorTypes,
-      resultTypes, [&](unsigned expected, unsigned actual) {
+      opBuilder, region, inputTypes, outputTypes,
+      [&](unsigned expected, unsigned actual) {
         res = parser.emitError(parser.getCurrentLocation(),
                                llvm::formatv("region expects {0} args, got {1}",
                                              expected, actual));
       });
   return res;
 }
 
 static ParseResult
 parseNamedStructuredOpResults(OpAsmParser &parser,
                               SmallVectorImpl<Type> &resultTypes) {
   if (succeeded(parser.parseOptionalArrow()))
     if (parser.parseTypeList(resultTypes))
       return failure();
   return success();
 }
 
 static ParseResult
 parseCommonStructuredOpParts(OpAsmParser &parser, OperationState &result,
                              SmallVectorImpl<Type> &inputTypes,
-                             SmallVectorImpl<Type> &outputBufferTypes,
-                             SmallVectorImpl<Type> &initTensorTypes) {
-  llvm::SMLoc inputsOperandsLoc, outputBuffersOperandsLoc,
-      initTensorsOperandsLoc;
-  SmallVector<OpAsmParser::OperandType, 4> inputsOperands,
-      outputBuffersOperands, initTensorsOperands;
+                             SmallVectorImpl<Type> &outputTypes) {
+  llvm::SMLoc inputsOperandsLoc, outputsOperandsLoc;
+  SmallVector<OpAsmParser::OperandType, 4> inputsOperands, outputsOperands;
 
   parser.parseOptionalAttrDict(result.attributes);
 
   if (succeeded(parser.parseOptionalKeyword("ins"))) {
     if (parser.parseLParen())
       return failure();
 
     inputsOperandsLoc = parser.getCurrentLocation();
     if (parser.parseOperandList(inputsOperands) ||
         parser.parseColonTypeList(inputTypes) || parser.parseRParen())
       return failure();
   }
 
   if (succeeded(parser.parseOptionalKeyword("outs"))) {
-    outputBuffersOperandsLoc = parser.getCurrentLocation();
-    if (parser.parseLParen() ||
-        parser.parseOperandList(outputBuffersOperands) ||
-        parser.parseColonTypeList(outputBufferTypes) || parser.parseRParen())
-      return failure();
-  }
-  if (succeeded(parser.parseOptionalKeyword("init"))) {
-    initTensorsOperandsLoc = parser.getCurrentLocation();
-    if (parser.parseLParen() || parser.parseOperandList(initTensorsOperands) ||
-        parser.parseColonTypeList(initTensorTypes) || parser.parseRParen())
+    outputsOperandsLoc = parser.getCurrentLocation();
+    if (parser.parseLParen() || parser.parseOperandList(outputsOperands) ||
+        parser.parseColonTypeList(outputTypes) || parser.parseRParen())
       return failure();
   }
 
   if (parser.resolveOperands(inputsOperands, inputTypes, inputsOperandsLoc,
                              result.operands) ||
-      parser.resolveOperands(outputBuffersOperands, outputBufferTypes,
-                             outputBuffersOperandsLoc, result.operands) ||
-      parser.resolveOperands(initTensorsOperands, initTensorTypes,
-                             initTensorsOperandsLoc, result.operands))
+      parser.resolveOperands(outputsOperands, outputTypes, outputsOperandsLoc,
+                             result.operands))
     return failure();
 
   result.addAttribute("operand_segment_sizes",
                       parser.getBuilder().getI32VectorAttr(
                           {static_cast<int32_t>(inputsOperands.size()),
-                           static_cast<int32_t>(outputBuffersOperands.size()),
-                           static_cast<int32_t>(initTensorsOperands.size())}));
+                           static_cast<int32_t>(outputsOperands.size())}));
   return success();
 }
 
 template <typename NamedStructuredOpType>
 static ParseResult parseNamedStructuredOp(OpAsmParser &parser,
                                           OperationState &result) {
-  SmallVector<Type, 1> inputTypes, outputBufferTypes, initTensorTypes;
-  if (parseCommonStructuredOpParts(parser, result, inputTypes,
-                                   outputBufferTypes, initTensorTypes))
+  SmallVector<Type, 1> inputTypes, outputTypes;
+  if (parseCommonStructuredOpParts(parser, result, inputTypes, outputTypes))
     return failure();
 
   // TODO: consider merging results parsing into region parsing.
   // Need to wait for declarative assembly resolution to decide.
   SmallVector<Type, 1> outputTensorsTypes;
   if (parseNamedStructuredOpResults(parser, outputTensorsTypes))
     return failure();
   result.addTypes(outputTensorsTypes);
 
   std::unique_ptr<Region> region = std::make_unique<Region>();
   if (parseNamedStructuredOpRegion<NamedStructuredOpType>(
-          parser, *region, inputTypes, outputBufferTypes, initTensorTypes,
-          outputTensorsTypes))
+          parser, *region, inputTypes, outputTypes))
     return failure();
   result.addRegion(std::move(region));
 
   return success();
 }
 
 static void printNamedStructuredOpResults(OpAsmPrinter &p,
                                           TypeRange resultTypes) {
   if (resultTypes.empty())
     return;
   p.printOptionalArrowTypeList(resultTypes);
 }
 
 template <typename NamedStructuredOpType>
 static void printCommonStructuredOpParts(OpAsmPrinter &p,
                                          NamedStructuredOpType op) {
   if (!op.inputs().empty())
     p << " ins(" << op.inputs() << " : " << op.inputs().getTypes() << ")";
-  if (!op.output_buffers().empty())
-    p << " outs(" << op.output_buffers() << " : "
-      << op.output_buffers().getTypes() << ")";
-  if (!op.init_tensors().empty())
-    p << " init(" << op.init_tensors() << " : " << op.init_tensors().getTypes()
-      << ") ";
+  if (!op.outputs().empty())
+    p << " outs(" << op.outputs() << " : " << op.outputs().getTypes() << ")";
 }
 
 template <typename NamedStructuredOpType>
 static void printNamedStructuredOp(OpAsmPrinter &p, NamedStructuredOpType op) {
   p << op.getOperationName();
   p.printOptionalAttrDict(op.getAttrs(),
                           /*elidedAttrs=*/{"operand_segment_sizes"});
 
struct EraseDeadLinalgOp : public RewritePattern {
   EraseDeadLinalgOp(PatternBenefit benefit = 1)
       : RewritePattern(benefit, MatchAnyOpTypeTag()) {}
 
   LogicalResult matchAndRewrite(Operation *op,
                                 PatternRewriter &rewriter) const override {
     auto linalgOp = dyn_cast<LinalgOp>(op);
     if (!linalgOp)
       return failure();
-    for (Value v : linalgOp.getInputsAndOutputBuffers()) {
+    for (Value v : linalgOp.getShapedOperands()) {
       // Linalg "inputs" may be either tensor or memref type.
       // tensor<0xelt_type> is a convention that may not always mean
       // "0 iterations". Only erase in cases we see memref<...x0x...>.
       auto mt = v.getType().dyn_cast<MemRefType>();
       if (!mt)
         continue;
       if (llvm::is_contained(mt.getShape(), 0)) {
         rewriter.eraseOp(linalgOp);
LogicalResult matchAndRewrite(Operation *op,
     SmallVector<Value, 4> newOperands;
     newOperands.reserve(op->getNumOperands());
     // Inputs may fold.
     for (Value v : linalgOp.getInputs()) {
       auto tensorCastOp = v.getDefiningOp<tensor::CastOp>();
       newOperands.push_back(
           canFoldIntoConsumerOp(tensorCastOp) ? tensorCastOp.source() : v);
     }
-    // Output buffers are memrefs, they don't fold.
-    newOperands.append(linalgOp.getOutputBuffers().begin(),
-                       linalgOp.getOutputBuffers().end());
     // Init tensors may fold, in which case the resultType must also change.
-    for (Value v : linalgOp.getInitTensors()) {
+    for (Value v : linalgOp.getOutputs()) {
       auto tensorCastOp = v.getDefiningOp<tensor::CastOp>();
       bool fold = canFoldIntoConsumerOp(tensorCastOp);
       newOperands.push_back(fold ? tensorCastOp.getOperand() : v);
       newResultTypes.push_back(newOperands.back().getType());
     }
     auto extraOperands = linalgOp.getAssumedNonShapedOperands();
     newOperands.append(extraOperands.begin(), extraOperands.end());
     // Clone op.
LogicalResult matchAndRewrite(Operation *op,
     if (canonicalInput.size() == linalgOp.getNumInputs())
       return failure();
 
     // The operands for the newly canonicalized op.
     SmallVector<Value, 6> newOperands;
     for (auto v : llvm::enumerate(linalgOp.getInputs()))
       if (canonicalInputIndices[v.index()] == static_cast<int>(v.index()))
         newOperands.push_back(v.value());
-    llvm::append_range(newOperands, linalgOp.getOutputBuffers());
-    llvm::append_range(newOperands, linalgOp.getInitTensors());
+    llvm::append_range(newOperands, linalgOp.getOutputs());
     llvm::append_range(newOperands, linalgOp.getAssumedNonShapedOperands());
 
     // Clone the old op with new operands.
     Operation *newOp = linalgOp.clone(rewriter, op->getLoc(),
                                       op->getResultTypes(), newOperands);
     auto newLinalgOp = cast<LinalgOp>(newOp);
 
     // Repair the indexing maps by filtering out the ones that have been
     // eliminated.
     SmallVector<AffineMap, 6> newIndexingMaps;
     for (int i = 0, e = newLinalgOp.getNumInputs(); i != e; i++)
       if (canonicalInputIndices[i] == i)
         newIndexingMaps.push_back(newLinalgOp.getIndexingMap(i));
     for (int i = 0, e = newLinalgOp.getNumOutputs(); i != e; i++)
       newIndexingMaps.push_back(newLinalgOp.getOutputIndexingMap(i));
     newOp->setAttr("indexing_maps",
                    rewriter.getAffineMapArrayAttr(newIndexingMaps));
 
     // Set the number of inputs to the new value. The `clone` call above kept
     // the value from the original op.
     newLinalgOp.setNumInputs(canonicalInput.size());
 
     // linalg.indexed_generic payloads have additional arguments prepended to
-    // the block arg list. The number of such args is one per dimension of the
-    // iteration space.
-    int bbArgBaseOffset = 0;
-    if (isa<IndexedGenericOp>(op))
-      bbArgBaseOffset = newIndexingMaps[0].getNumInputs();
+    // the block arg list.
+    int bbArgBaseOffset = newLinalgOp.getNumPayloadInductionVariables();
 
     // Repair the payload entry block by RAUW'ing redundant arguments and
     // erasing them.
     Block &payload = newOp->getRegion(0).front();
     for (int i = 0, e = linalgOp.getNumInputs(); i < e; i++) {
       // Iterate in reverse, so that we erase later args first, preventing the
       // argument list from shifting unexpectedly and invalidating all our
       // indices.

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

 #include "mlir/Dialect/Vector/VectorOps.h"
 #include "mlir/IR/BuiltinDialect.h"
 #include "mlir/IR/Operation.h"
 #include "mlir/Pass/Pass.h"
 
 using namespace ::mlir;
 using namespace ::mlir::linalg;
 
-static Value maybeConvertToIndex(Location loc, Value val, OpBuilder &b) {
-  if (val.getType().isIndex())
-    return val;
-  return b.create<IndexCastOp>(loc, val, b.getIndexType());
-}
-
-static Value cloneMemref(Location loc, Value memref, OpBuilder &b) {
-  auto memrefType = memref.getType().cast<MemRefType>();
+static SmallVector<Value, 4> getDynOperands(Location loc, Value val,
+                                            OpBuilder &b) {
   SmallVector<Value, 4> dynOperands;
-  for (auto dim : llvm::enumerate(memrefType.getShape())) {
+  auto shapedType = val.getType().cast<ShapedType>();
+  for (auto dim : llvm::enumerate(shapedType.getShape())) {
     if (dim.value() == TensorType::kDynamicSize) {
-      dynOperands.push_back(b.create<DimOp>(loc, memref, dim.index()));
+      dynOperands.push_back(b.create<DimOp>(loc, val, dim.index()));
     }
   }
-  auto alloc = b.create<AllocOp>(loc, memrefType, dynOperands);
+  return dynOperands;
+}
+
+static Value cloneMemref(Location loc, Value memref, OpBuilder &b) {
+  auto memrefType = memref.getType().cast<MemRefType>();
+  auto alloc =
+      b.create<AllocOp>(loc, memrefType, getDynOperands(loc, memref, b));
   b.create<linalg::CopyOp>(loc, memref, alloc);
   return alloc;
 }
 
 static LogicalResult
 allocateBuffersForResults(Location loc, LinalgOp linalgOp,
                           linalg::GenericOpAdaptor &adaptor,
                           SmallVectorImpl<Value> &resultBuffers, OpBuilder &b) {
   // Lazily compute loopRanges.
   SmallVector<Range, 4> loopRanges;
 
   // Allocate a buffer for every tensor result.
+  assert(linalgOp.getNumOutputs() == linalgOp->getNumResults());
   for (auto en : llvm::enumerate(linalgOp->getResultTypes())) {
     size_t resultIndex = en.index();
     Type resultType = en.value();
 
     auto tensorType = resultType.dyn_cast<RankedTensorType>();
     if (tensorType == nullptr) {
       linalgOp.emitOpError()
           << "tensor to buffer conversion expects ranked tensor results";
       return failure();
     }
     auto tensorShape = tensorType.getShape();
     auto memrefType = MemRefType::get(tensorShape, tensorType.getElementType());
+    Value resultTensor = adaptor.outputs()[resultIndex];
 
-    // Allocate buffers for init tensors that are assumed to fold onto the first
-    // results.
-    // TODO: update this assumption because the reality is more complex
-    // under linalg on tensor based transformations.
-    bool hasInitTensor = resultIndex < linalgOp.getNumInitTensors();
-    if (hasInitTensor) {
-      resultBuffers.push_back(
-          cloneMemref(loc, adaptor.init_tensors()[resultIndex], b));
+    // Clone output buffers whose value is actually used.
+    if (linalgOp.payloadUsesValueFromOutputOperandIndex(resultIndex)) {
+      resultBuffers.push_back(cloneMemref(loc, resultTensor, b));
       continue;
     }
 
+    if (auto alloc = resultTensor.getDefiningOp<AllocOp>()) {
+      resultBuffers.push_back(resultTensor);
+      continue;
+    }
     // Allocate buffers for statically-shaped results.
     if (memrefType.hasStaticShape()) {
       resultBuffers.push_back(b.create<AllocOp>(loc, memrefType));
       continue;
     }
 
-    // Perform a naive shape inference for the dynamically-shaped results.
-    // Extract the required element out of the vector.
-    SmallVector<Value, 4> dynOperands;
-    auto resultIndexingMap = linalgOp.getOutputIndexingMap(resultIndex);
-    for (auto shapeElement : llvm::enumerate(tensorType.getShape())) {
-      if (loopRanges.empty())
-        loopRanges = linalgOp.createLoopRanges(b, loc);
-      if (shapeElement.value() != ShapedType::kDynamicSize)
-        continue;
-      AffineExpr expr = resultIndexingMap.getResult(shapeElement.index());
-      switch (expr.getKind()) {
-      case AffineExprKind::DimId: {
-        int64_t loopIndex = expr.cast<AffineDimExpr>().getPosition();
-        Value size = maybeConvertToIndex(loc, loopRanges[loopIndex].size, b);
-        dynOperands.push_back(size);
-        break;
-      }
-      default:
-        return failure();
-      }
-    }
-    resultBuffers.push_back(b.create<AllocOp>(loc, memrefType, dynOperands));
+    resultBuffers.push_back(b.create<AllocOp>(
+        loc, memrefType, getDynOperands(loc, resultTensor, b)));
   }
   return success();
 }
 
 /// Specialization for `linalg::GenericOp` and `linalg::IndexedGenericOp`.
 /// A pattern to convert Generic Linalg operations which work on tensors to
 /// use buffers. BufferPlacement pass should be later used to move
 /// Alloc operations to the correct positions and insert the missing Dealloc
 /// operations in the correct places.
 template <typename GenericOpTy>
 static void
 finalizeBufferAllocationForGenericOp(ConversionPatternRewriter &rewriter,
                                      GenericOpTy genericOp, ValueRange inputs,
                                      ValueRange outputs) {
   // Generate a new linalg operation that works on buffers.
   auto newGenericOp = rewriter.create<GenericOpTy>(
       genericOp.getLoc(),
       /*resultTensorTypes=*/llvm::None,
       /*inputs=*/inputs,
-      /*outputBuffers=*/outputs,
-      /*initTensors=*/llvm::None, genericOp.indexing_maps(),
+      /*outputs=*/outputs, genericOp.indexing_maps(),
       genericOp.iterator_types(), genericOp.docAttr(),
       genericOp.library_callAttr(), genericOp.sparseAttr());
 
   // Create a new block in the region of the new Generic Op.
   Block *oldBlock = genericOp.getBody();
   Region &newRegion = newGenericOp.region();
   Block *newBlock = rewriter.createBlock(&newRegion, newRegion.begin(),
                                          oldBlock->getArgumentTypes());
 
-  // Add the result arguments to the new block.
-  for (Value v : ValueRange(outputs).drop_front(genericOp.getNumInitTensors()))
-    newBlock->addArgument(v.getType().cast<MemRefType>().getElementType());
-
   // Clone the body of the old block to the new block.
   BlockAndValueMapping mapping;
   mapping.map(oldBlock->getArguments(), newBlock->getArguments());
 
   OpBuilder::InsertionGuard guard(rewriter);
   rewriter.setInsertionPointToEnd(newBlock);
   for (auto &op : oldBlock->getOperations()) {
     Operation *clonedOp = rewriter.clone(op, mapping);
static void finalizeBufferAllocation(ConversionPatternRewriter &rewriter,
                                      linalg::LinalgOp linalgOp,
                                      ValueRange inputs, ValueRange outputs) {
   assert(!isa<linalg::GenericOp>(linalgOp.getOperation()));
   assert(!isa<linalg::IndexedGenericOp>(linalgOp.getOperation()));
   SmallVector<Value, 8> newOperands = inputs;
   newOperands.append(outputs.begin(), outputs.end());
   auto otherOperands = linalgOp.getAssumedNonShapedOperands();
   newOperands.append(otherOperands.begin(), otherOperands.end());
-  LinalgOp res = cast<LinalgOp>(linalgOp.clone(rewriter, linalgOp.getLoc(),
-                                               /*resultTypes=*/ArrayRef<Type>{},
-                                               newOperands));
-  // Need to mutate the operands_segment_sizes in the resulting op.
-  res.setNumOutputBuffers(outputs.size());
-  res.setNumInitTensors(0);
+  linalgOp.clone(rewriter, linalgOp.getLoc(),
+                 /*resultTypes=*/ArrayRef<Type>{}, newOperands);
   // Replace the results of the old op with the new output buffers.
   rewriter.replaceOp(linalgOp, outputs);
 }
 
 //===----------------------------------------------------------------------===//
 // Bufferization patterns.
 //===----------------------------------------------------------------------===//
 
 namespace {
+
+/// Generic conversion pattern that matches any LinalgOp. This avoids template
+/// instantiating one pattern for each LinalgOp.
+class BufferizeInitTensorOp : public OpConversionPattern<InitTensorOp> {
+public:
+  using OpConversionPattern<InitTensorOp>::OpConversionPattern;
+
+  LogicalResult
+  matchAndRewrite(InitTensorOp op, ArrayRef<Value> operands,
+                  ConversionPatternRewriter &rewriter) const final {
+    linalg::InitTensorOpAdaptor adaptor(operands, op->getAttrDictionary());
+    rewriter.replaceOpWithNewOp<AllocOp>(
+        op, getTypeConverter()->convertType(op.getType()).cast<MemRefType>(),
+        adaptor.sizes());
+    return success();
+  }
+};
+
 /// Generic conversion pattern that matches any LinalgOp. This avoids template
 /// instantiating one pattern for each LinalgOp.
 class BufferizeAnyLinalgOp : public ConversionPattern {
 public:
   BufferizeAnyLinalgOp(TypeConverter &typeConverter)
       : ConversionPattern(/*benefit=*/1, typeConverter, MatchAnyOpTypeTag()) {}
 
   LogicalResult
   matchAndRewrite(Operation *op, ArrayRef<Value> operands,
                   ConversionPatternRewriter &rewriter) const final {
 
     LinalgOp linalgOp = dyn_cast<linalg::LinalgOp>(op);
     if (!linalgOp)
       return failure();
 
     // We abuse the GenericOpAdaptor here.
-    // TODO: Manually create an Adaptor that captures inputs, output_buffers and
-    // init_tensors for all linalg::LinalgOp interface ops.
+    // TODO: Manually create an Adaptor that captures inputs and outputs for all
+    // linalg::LinalgOp interface ops.
     linalg::GenericOpAdaptor adaptor(operands, op->getAttrDictionary());
 
     Location loc = linalgOp.getLoc();
-    SmallVector<Value, 2> newOutputBuffers(adaptor.output_buffers().begin(),
-                                           adaptor.output_buffers().end());
+    SmallVector<Value, 2> newOutputBuffers;
 
     if (failed(allocateBuffersForResults(loc, linalgOp, adaptor,
                                          newOutputBuffers, rewriter))) {
       linalgOp.emitOpError()
           << "Failed to allocate buffers for tensor results.";
       return failure();
     }
 

 struct LinalgBufferizePass : public LinalgBufferizeBase<LinalgBufferizePass> {
   void runOnOperation() override {
     MLIRContext &context = getContext();
     ConversionTarget target(context);
     BufferizeTypeConverter typeConverter;
 
     // Mark all Standard operations legal.
     target.addLegalDialect<AffineDialect, StandardOpsDialect>();
-    target.addIllegalOp<SubTensorOp, SubTensorInsertOp>();
+    target.addIllegalOp<InitTensorOp, SubTensorOp, SubTensorInsertOp>();
 
     // Mark all Linalg operations illegal as long as they work on tensors.
     auto isLegalOperation = [&](Operation *op) {
       return typeConverter.isLegal(op);
     };
     target.addDynamicallyLegalDialect<linalg::LinalgDialect>(isLegalOperation);
     target.addDynamicallyLegalOp<ConstantOp>(isLegalOperation);
 
std::unique_ptr<OperationPass<FuncOp>> mlir::createLinalgBufferizePass() {
   return std::make_unique<LinalgBufferizePass>();
 }
 
 void mlir::linalg::populateLinalgBufferizePatterns(
     MLIRContext *context, BufferizeTypeConverter &typeConverter,
     OwningRewritePatternList &patterns) {
   patterns.insert<BufferizeAnyLinalgOp>(typeConverter);
   // TODO: Drop this once tensor constants work in standard.
-  patterns.insert<
-      // clang-format off
+  patterns.insert<
+      BufferizeInitTensorOp,
       SubTensorOpConverter,
       SubTensorInsertOpConverter
-      // clang-format on
-      >(typeConverter, context);
+  // clang-format on
 }

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

LogicalResult matchAndRewrite(GenericOpTy op,
 
     // Check if any of the iteration dimensions are unit-trip count. They will
     // end up being unit-trip count if they are used to index into a unit-dim
     // tensor/memref.
     AffineMap invertedMap = inversePermutation(concatAffineMaps(indexingMaps));
     if (!invertedMap)
       return failure();
     SmallVector<int64_t, 4> dims;
-    for (ShapedType shapedType : op.getInputOutputShapedTypes())
+    for (ShapedType shapedType : op.getShapedOperandTypes())
       dims.append(shapedType.getShape().begin(), shapedType.getShape().end());
     DenseSet<unsigned> unitDims;
     ArrayAttr iteratorTypes = op.iterator_types();
     for (auto expr : enumerate(invertedMap.getResults())) {
       if (AffineDimExpr dimExpr = expr.value().dyn_cast<AffineDimExpr>())
         if (dims[dimExpr.getPosition()] == 1 &&
             iteratorTypes[expr.index()].dyn_cast<StringAttr>().getValue() ==
                 getParallelIteratorTypeName())

 
 /// Pattern to replace tensors operands/results that are unit extents.
 template <typename GenericOpTy>
 struct ReplaceUnitExtentTensors : public OpRewritePattern<GenericOpTy> {
   using OpRewritePattern<GenericOpTy>::OpRewritePattern;
   LogicalResult matchAndRewrite(GenericOpTy op,
                                 PatternRewriter &rewriter) const override {
     // TODO: support init_tensors and reductions.
-    if (!op.hasTensorSemantics() || !op.init_tensors().empty())
+    if (!op.hasTensorSemantics() || op.getNumInitTensors() != 0)
       return failure();
 
     MLIRContext *context = rewriter.getContext();
     Location loc = op.getLoc();
 
     SmallVector<AffineMap, 4> newIndexingMaps;
     SmallVector<ArrayAttr, 4> reassociationMaps;
     SmallVector<ShapedType, 4> newInputOutputTypes;
     bool doCanonicalization = false;
     for (auto it :
-         llvm::zip(op.getIndexingMaps(), op.getInputOutputShapedTypes())) {
+         llvm::zip(op.getIndexingMaps(), op.getShapedOperandTypes())) {
       auto replacementInfo = replaceUnitExtents(
           std::get<0>(it), std::get<1>(it).template cast<RankedTensorType>(),
           context);
       reassociationMaps.push_back(replacementInfo.reassociation);
       newIndexingMaps.push_back(replacementInfo.indexMap);
       newInputOutputTypes.push_back(replacementInfo.type);
       doCanonicalization |= replacementInfo.type != std::get<1>(it);
     }
auto insertReshapes = [&](ValueRange values) {
               loc, newInputOutputTypes[flattenedIdx], operand.value(),
               reassociationMaps[flattenedIdx]));
         ++flattenedIdx;
       }
       return res;
     };
 
     SmallVector<Value, 4> newInputs = insertReshapes(op.inputs());
-    SmallVector<Value, 4> newOutputBuffers =
-        insertReshapes(op.output_buffers());
-    SmallVector<Value, 4> newInitTensors = insertReshapes(op.init_tensors());
+    SmallVector<Value, 4> newOutputs = insertReshapes(op.outputs());
 
-    // If any result type change, insert a reshape to convert from the original
+    // If any result type changes, insert a reshape to convert from the original
     // type to the new type.
     SmallVector<Type, 4> resultTypes;
     resultTypes.reserve(op.getNumResults());
     for (unsigned i : llvm::seq<unsigned>(0, op.getNumResults()))
       resultTypes.push_back(newInputOutputTypes[i + op.getNumInputs()]);
     GenericOpTy replacementOp = rewriter.create<GenericOpTy>(
-        loc, resultTypes, newInputs, newOutputBuffers, newInitTensors,
-        newIndexingMaps,
+        loc, resultTypes, newInputs, newOutputs, newIndexingMaps,
         llvm::to_vector<4>(
             op.iterator_types().template getAsValueRange<StringAttr>()));
     rewriter.inlineRegionBefore(op.region(), replacementOp.region(),
                                 replacementOp.region().begin());
 
     // If any result tensor has a modified shape, then add reshape to recover
     // the original shape.
     SmallVector<Value, 4> resultReplacements;
     for (auto result : llvm::enumerate(replacementOp.getResults())) {
-      unsigned index = result.index() + replacementOp.getNumOperands();
+      unsigned index = result.index() + replacementOp.getNumInputs();
       RankedTensorType origResultType = op.getResult(result.index())
                                             .getType()
                                             .template cast<RankedTensorType>();
       if (origResultType != result.value().getType())
         resultReplacements.push_back(rewriter.create<linalg::TensorReshapeOp>(
             loc, origResultType, result.value(), reassociationMaps[index]));
       else
         resultReplacements.push_back(result.value());

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

if (!op->hasTrait<OpTrait::ElementwiseMappable>())
     return false;
 
   // TODO: The conversion pattern can be made to work for `any_of` here, but
   // it's more complex as it requires tracking which operands are scalars.
   return llvm::all_of(op->getOperandTypes(),
                       [](Type type) { return type.isa<RankedTensorType>(); });
 }
 
+/// Given `op` assumed `isElementwiseMappableOpOnRankedTensors`, iterate over
+/// the result types and return a list of values such that, for each result type
+/// `t` and value `v` at the same index `idx`:
+///   1. `v.getType() == t`
+///   2. If an operand of `op` has type `t`, let `operand_first` be the first
+///      such operand. Then`v == operand_first`.
+///   3. Otherwise, v is a newly created `linalg::InitTensorOp` with:
+///        a. Static and dynamic dims extracted from the first operand of `op`.
+///        b. Elemental type equal to the elemental type of `t`.
+///
+/// This is sufficient because ElementwiseMappable guarantees that "The static
+/// types of all vector (resp. tensor) operands and results must have the same
+/// shape".
+static SmallVector<Value, 4>
+getOrCreateOperandsMatchingResultTypes(OpBuilder &b, Operation *op) {
+  assert(isElementwiseMappableOpOnRankedTensors(op));
+  Location loc = op->getLoc();
+  ValueRange operands = op->getOperands();
+  TypeRange rankedTensorTypes = op->getResultTypes();
+  SmallVector<Value, 4> res;
+  res.reserve(rankedTensorTypes.size());
+  for (Type t : rankedTensorTypes) {
+    // Try to find an operand with type matching the result tensor.
+    bool found = false;
+    for (Value v : operands) {
+      if (v.getType() == t) {
+        found = true;
+        res.push_back(v);
+        break;
+      }
+    }
+    if (found)
+      continue;
+
+    // Extract static / dynamic shape mix from the first operand.
+    Value firstOperand = operands.front();
+    auto rankedTensorType = t.cast<RankedTensorType>();
+    SmallVector<Value, 8> dynamicShape;
+    SmallVector<int64_t, 8> staticShape;
+    dynamicShape.reserve(rankedTensorType.getRank());
+    staticShape.reserve(rankedTensorType.getRank());
+    unsigned idx = 0;
+    for (auto shape : rankedTensorType.getShape()) {
+      staticShape.push_back(shape);
+      if (rankedTensorType.isDynamicDim(idx))
+        dynamicShape.push_back(b.create<DimOp>(loc, firstOperand, idx));
+      ++idx;
+    }
+    // Create init tensor.
+    res.push_back(b.create<linalg::InitTensorOp>(
+        loc, dynamicShape, staticShape, rankedTensorType.getElementType()));
+  }
+  return res;
+}
+
 namespace {
 struct ConvertAnyElementwiseMappableOpOnRankedTensors : public RewritePattern {
   ConvertAnyElementwiseMappableOpOnRankedTensors()
       : RewritePattern(/*benefit=*/1, MatchAnyOpTypeTag()) {}
   LogicalResult matchAndRewrite(Operation *op,
                                 PatternRewriter &rewriter) const final {
     if (!isElementwiseMappableOpOnRankedTensors(op))
       return rewriter.notifyMatchFailure(
           op, "requires elementwise op on ranked tensors");
 
     auto rank = op->getResult(0).getType().cast<RankedTensorType>().getRank();
     SmallVector<AffineMap, 3> indexingMaps(
         op->getNumResults() + op->getNumOperands(),
         rewriter.getMultiDimIdentityMap(rank));
     SmallVector<StringRef, 6> iteratorTypes(rank,
                                             getParallelIteratorTypeName());
+    auto outputs = getOrCreateOperandsMatchingResultTypes(rewriter, op);
     rewriter.replaceOpWithNewOp<linalg::GenericOp>(
         op, /*resultTensorTypes=*/op->getResultTypes(),
         /*inputs=*/op->getOperands(),
-        /*outputBuffers=*/ValueRange(),
-        /*initTensors=*/ValueRange(),
+        /*outputs=*/outputs,
         /*indexingMaps=*/indexingMaps,
         /*iteratorTypes=*/iteratorTypes,
         /*bodyBuilder=*/
         [&](OpBuilder &builder, Location loc, ValueRange regionArgs) {
           OperationState state(loc, op->getName());
           state.addAttributes(op->getAttrs());
-          state.addOperands(regionArgs);
+          // Only take the input operands in the cloned elementwise op.
+          state.addOperands(regionArgs.take_front(op->getNumOperands()));
           auto resultTypes = llvm::to_vector<6>(
               llvm::map_range(op->getResultTypes(), [](Type type) {
                 return type.cast<TensorType>().getElementType();
               }));
           state.addTypes(resultTypes);
           auto *scalarOp = builder.createOperation(state);
           builder.create<linalg::YieldOp>(loc, scalarOp->getResults());
         });

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

 // they must agree by construction (i.e. have the same size) and we just return
 // the first one.
 static ShapeDimension
 getShapeDefiningLoopRange(LinalgOp op, unsigned loopDepth,
                           bool fromSubViewOpOnly = false) {
   auto maps = op.indexing_maps();
   // Iterate over the inputs and outputs in order.
   // Extract the subranges from the linearized ranges.
-  SmallVector<Value, 8> ios(op.getInputsAndOutputBuffers());
-  for (auto en : llvm::enumerate(ios)) {
+  for (auto en : llvm::enumerate(op.getShapedOperands())) {
     // The method `getRangeFromOperandShape` requires using SubViewOp or
     // SubTensorOps. If the value isnt defined from there continue.
     // todo: The method should be adapted to get the values from
     // `ViewInterface`. The interface needs a `getOrCreateRanges` method which
     // currently returns a `linalg.range`. The fix here is to move this op to
     // `std` dialect and add the method to `ViewInterface`.
     if (fromSubViewOpOnly &&
         !isa_and_nonnull<SubViewOp, SubTensorOp>(en.value().getDefiningOp()))
for (unsigned idx = 1, e = sva.getNumOperands(); idx != e; ++idx)
     if (sva.getOperand(idx) != svb.getOperand(idx))
       return false;
   return true;
 }
 
 static Optional<LinalgDependenceGraph::LinalgDependenceGraphElem>
 findFusableProducer(LinalgOp consumer, unsigned consumerIdx,
                     const LinalgDependenceGraph &dependenceGraph) {
+  assert(consumer.hasBufferSemantics() && "revisit usage of shaped operand");
+
   // Only consider RAW and WAW atm.
   for (auto depType : {
            LinalgDependenceGraph::DependenceType::RAW,
            LinalgDependenceGraph::DependenceType::WAW,
        }) {
     for (auto dependence : llvm::make_filter_range(
              dependenceGraph.getDependencesInto(consumer, depType),
              [consumerIdx](
                  LinalgDependenceGraph::LinalgDependenceGraphElem elem) {
-               return elem.indexingOpView.operandIndex == consumerIdx;
+               return elem.indexingOpView->getOperandNumber() == consumerIdx;
              })) {
-      auto producer = cast<LinalgOp>(dependence.dependentOpView.op);
 
       // Check that the dependence is indeed on the input `consumerIdx` view.
-      auto consumedView =
-          consumer.getBuffer(dependence.indexingOpView.operandIndex);
-      if (!isSameSubView(consumer.getBuffer(consumerIdx), consumedView))
+      Value consumedView = dependence.indexingOpView->get();
+      if (!isSameSubView(consumer.getShapedOperand(consumerIdx), consumedView))
         continue;
 
       // Consumer consumes this view, `isStructurallyFusableProducer` also
       // checks whether it is a strict subview of the producer view.
-      auto producedView =
-          producer.getBuffer(dependence.dependentOpView.operandIndex);
+      auto producer = cast<LinalgOp>(dependence.dependentOpView->getOwner());
+      Value producedView = dependence.dependentOpView->get();
       LLVM_DEBUG(llvm::dbgs()
                  << "\n"
                  << LinalgDependenceGraph::getDependenceTypeStr(depType)
-                 << "producer: " << *producer.getOperation()
-                 << " view: " << producedView << " output index: "
-                 << dependence.dependentOpView.operandIndex -
+                 << "producer: " << *dependence.dependentOpView->getOwner()
+                 << " view: " << dependence.dependentOpView->get()
+                 << " output index: "
+                 << dependence.dependentOpView->getOperandNumber() -
                         producer.getNumInputs()
                  << "\n");
       (void)producedView;
 
       // Simple fusability checks.
       if (!isFusableInto(dependenceGraph, consumer, consumedView, producer))
         continue;
 
       return dependence;
     }
   }
   return {};
 }
 
 Optional<FusionInfo>
 mlir::linalg::fuseProducerOfBuffer(OpBuilder &b, LinalgOp consumer,
                                    unsigned consumerIdx,
                                    const LinalgDependenceGraph &graph) {
   Optional<LinalgDependenceGraph::LinalgDependenceGraphElem> fusableDependence =
       findFusableProducer(consumer, consumerIdx, graph);
   if (!fusableDependence)
     return {};
 
-  LinalgOp producerOp = cast<LinalgOp>(fusableDependence->dependentOpView.op);
+  LinalgOp producerOp =
+      cast<LinalgOp>(fusableDependence->dependentOpView->getOwner());
   // If producer is already in the same block as consumer, we are done.
   if (consumer->getBlock() == producerOp->getBlock())
     return {};
 
-  unsigned producerIdx = fusableDependence->dependentOpView.operandIndex -
-                         producerOp.getNumInputs();
+  unsigned producerIdx =
+      fusableDependence->dependentOpView->getOperandNumber() -
   Value consumerView = consumer.getShapedOperand(consumerIdx);
 
   // Must be a subview or a slice to guarantee there are loops we can fuse
   // into.
   auto subView = consumerView.getDefiningOp<SubViewOp>();
   auto slice = consumerView.getDefiningOp<SliceOp>();
   if (!subView && !slice) {
     LLVM_DEBUG(llvm::dbgs() << "\nNot fusable (not a subview or slice)");

 /// Returns the mapping from iterations in the consumer that write to the same
 /// location as the iterations in the producer. To do so use
 /// - indexing map of the fused view in the consumer : consumerIndexMap
 /// - indexing map of the fused view in the producer : producerIndexMap
 ///     consumerLoopToProducerLoop =
 ///       inverse(producerIndexMap).compose(consumerIndexMap)
 static Optional<AffineMap> getConsumerLoopToProducerLoopMap(
     LinalgDependenceGraph::LinalgDependenceGraphElem dependence) {
-  auto producer = cast<LinalgOp>(dependence.dependentOpView.op);
+  auto producer = cast<LinalgOp>(dependence.dependentOpView->getOwner());
   AffineMap producerIndexingMap =
-      producer.getIndexingMap(dependence.dependentOpView.operandIndex);
-  auto consumer = cast<LinalgOp>(dependence.indexingOpView.op);
+      producer.getIndexingMap(dependence.dependentOpView->getOperandNumber());
+  auto consumer = cast<LinalgOp>(dependence.indexingOpView->getOwner());
   AffineMap consumerIndexingMap =
-      consumer.getIndexingMap(dependence.indexingOpView.operandIndex);
+      consumer.getIndexingMap(dependence.indexingOpView->getOperandNumber());
 
   AffineMap prunedProducerIndexingMap = pruneReductionDimsFromMap(
       producer.iterator_types().getValue(), producerIndexingMap);
   if (!prunedProducerIndexingMap.isPermutation())
     return None;
 
   if (consumerIndexingMap.getNumResults() !=
       prunedProducerIndexingMap.getNumResults())
FusableOpDependencesTy mlir::linalg::findAllFusableDependences(
     ArrayRef<LinalgOp> ops, const LinalgDependenceGraph &dependenceGraph) {
   FusableOpDependencesTy fusableDependences;
   // TODO: Currently fusion would not be legal if the fusable dependence is to
   // the same producer but different indexing map in the consumer. Fix this, but
   // in the meanwhile disallow such a fusion.
   DenseMap<Operation *, AffineMap> fusedProducerIndexingMap;
   for (LinalgOp op : reverse(ops)) {
     for (auto operandIndex :
-         llvm::seq<unsigned>(0, op.getNumInputsAndOutputBuffers())) {
+         llvm::seq<unsigned>(0, op.getNumShapedOperands())) {
       Optional<LinalgDependenceGraph::LinalgDependenceGraphElem>
           fusableDependence =
               findFusableProducer(op, operandIndex, dependenceGraph);
       if (!fusableDependence)
         continue;
       LinalgOp producerOp =
-          cast<LinalgOp>(fusableDependence->dependentOpView.op);
+          cast<LinalgOp>(fusableDependence->dependentOpView->getOwner());
       // Do not fuse dependences that are to operations not in the same basic
       // block. This avoid moving fused operations across loops that might
       // themselves carry dependency making the fusion illegal.
       if (producerOp->getBlock() != op->getBlock()) {
         op.emitRemark("unhandled fusion of ops in different basic blocks");
         return FusableOpDependencesTy{};
       }
       // Make sure that the indexing map of the view used for fusion in the
       // producer is a projected permutation.
-      unsigned producerIdx = fusableDependence->dependentOpView.operandIndex;
+      unsigned producerIdx =
+          fusableDependence->dependentOpView->getOperandNumber();
       AffineMap producerMap = producerOp.getIndexingMap(producerIdx);
       if (!producerMap.isProjectedPermutation()) {
         op.emitRemark(
             "unhandled non permutation indexing map for fused view in "
             "producer for operand at index ")
             << operandIndex;
         return FusableOpDependencesTy{};
       }
 
-      unsigned consumerIdx = fusableDependence->indexingOpView.operandIndex;
+      unsigned consumerIdx =
+          fusableDependence->indexingOpView->getOperandNumber();
       AffineMap consumerMap = op.getIndexingMap(consumerIdx);
       if (!consumerMap.isProjectedPermutation()) {
         op.emitRemark(
             "unhandled case where indexing map for fused view in the consumer "
             "is "
             "not a projected permuration while fusing at index ")
             << operandIndex;
         return FusableOpDependencesTy{};

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

unsigned numFusedOpIndices =
           : 0;
   // Firstly, add all the indices to the block arguments.
   for (unsigned i = 0, e = numFusedOpIndices; i < e; ++i)
     fusedBlock->addArgument(rewriter.getIndexType());
   // Map the arguments for the unmodified args from the consumer.
   for (auto consumerArg : llvm::enumerate(consumerBlock.getArguments())) {
     if (consumerArg.index() == consumerIdx + numConsumerIndices) {
       // Map the arguments for the args from the producer.
-      for (auto producerArg : llvm::enumerate(producerBlock.getArguments())) {
+      for (auto producerArg :
+           llvm::enumerate(producerBlock.getArguments().take_front(
+               producer.getNumInputs() + numProducerIndices))) {
         // If producer is an indexed_generic op, map the indices from consumer
         // loop to producer loop (because the fusedOp is built based on
         // consumer's perspective).
         if (producerArg.index() < numProducerIndices) {
           auto newIndex = rewriter.create<mlir::AffineApplyOp>(
               producer.getLoc(),
               consumerToProducerLoopsMap.getSubMap(producerArg.index()),
               fusedBlock->getArguments().take_front(numFusedOpIndices));
fuseTensorOpsImpl(LinalgOp producer, LinalgOp consumer, unsigned consumerIdx,
   getIndexingMapOfProducerOperandsInFusedOp(
       producer, consumer.getInputIndexingMap(consumerIdx), fusedIndexMaps);
 
   // Append the indexing maps for the remaining consumer operands.
   fusedIndexMaps.append(std::next(consumerIndexMaps.begin(), consumerIdx + 1),
                         consumerIndexMaps.end());
 
   // Generate the fused op.
-  // Tensor-level fusion is only on ops without initTensors and outputBuffers.
   LinalgOp fusedOp;
   if (isa<GenericOp>(producer.getOperation()) &&
       isa<GenericOp>(consumer.getOperation())) {
     fusedOp =
         rewriter
             .create<GenericOp>(consumer.getLoc(), consumer->getResultTypes(),
                                /*inputs=*/fusedOperands,
-                               /*outputBuffers=*/ValueRange{},
-                               /*initTensors=*/ValueRange{},
+                               // TODO: handle outputs.
+                               consumer.getOutputs(),
                                rewriter.getArrayAttr(fusedIndexMaps),
                                consumer.iterator_types(),
                                /*doc=*/nullptr,
                                /*library_call=*/nullptr,
                                /*sparse=*/nullptr)
             .getOperation();
   } else {
-    fusedOp = rewriter
-                  .create<IndexedGenericOp>(
-                      consumer.getLoc(), consumer->getResultTypes(),
-                      /*inputs=*/fusedOperands,
-                      /*outputBuffers=*/ValueRange{},
-                      /*initTensors=*/ValueRange{},
-                      rewriter.getArrayAttr(fusedIndexMaps),
-                      consumer.iterator_types(),
-                      /*doc=*/nullptr,
-                      /*library_call=*/nullptr,
-                      /*sparse=*/nullptr)
-                  .getOperation();
+    fusedOp =
+        rewriter
+                // TODO: handle outputs.
+                consumer.getOutputs(), rewriter.getArrayAttr(fusedIndexMaps),
   }
 
   // Construct an AffineMap from consumer loops to producer loops.
   // consumer loop -> tensor index
   AffineMap consumerResultIndexMap = consumer.getInputIndexingMap(consumerIdx);
   // producer loop -> tensor index
   AffineMap producerResultIndexMap = producer.getOutputIndexingMap(0);
   // tensor index -> producer loop
return isa<GenericOp, IndexedGenericOp>(linalgOp.getOperation()) &&
                       }) &&
          linalgOp.getIndexingMap(fusedTensorIndex).getNumResults() > 0 &&
          llvm::all_of(linalgOp.iterator_types(), [](Attribute attr) {
            return attr.cast<StringAttr>().getValue() ==
                   getParallelIteratorTypeName();
          });
 }
 
+// Get the output tensor to use for the expanded operation. Creates an
+// `linalg.init_tensor` operation to materialize the tensor that carries the
+// shape information.
+static Value getOutputValueForExpansion(
+    OpBuilder &builder, Location loc, AffineMap outputIndexingMap, Value result,
+    ArrayRef<SmallVector<int64_t, 4>> origDimToExpandedShapeMap) {
+  SmallVector<Value, 4> dynamicDims;
+  SmallVector<int64_t, 4> staticDims;
+  ShapedType resultType = result.getType().cast<ShapedType>();
+  ArrayRef<int64_t> origShape = resultType.getShape();
+  for (AffineExpr expr : outputIndexingMap.getResults()) {
+    unsigned origDimPos = expr.cast<AffineDimExpr>().getPosition();
+    ArrayRef<int64_t> expandedShape(origDimToExpandedShapeMap[origDimPos]);
+    bool foundDynamic = false;
+    int64_t linearizedShape = 1;
+    for (int64_t extent : expandedShape) {
+      if (ShapedType::isDynamic(extent)) {
+        assert(!foundDynamic &&
+               "Expanded dimensions of reshape can have only one dynamic dim");
+        staticDims.push_back(ShapedType::kDynamicSize);
+        foundDynamic = true;
+        continue;
+      }
+      staticDims.push_back(extent);
+      linearizedShape *= extent;
+    }
+    if (ShapedType::isDynamic(origShape[origDimPos])) {
+      Value origDim = builder.create<DimOp>(loc, result, origDimPos);
+      dynamicDims.push_back(builder.create<UnsignedDivIOp>(
+          loc, origDim, builder.create<ConstantIndexOp>(loc, linearizedShape)));
+    }
+  }
+  return builder.create<linalg::InitTensorOp>(loc, dynamicDims, staticDims,
+                                              resultType.getElementType());
+}
+
 /// Implements the fusion of a tensor_reshape op and a generic/indexed_generic
 /// op as explained in `isFusableWithReshapeByExpansion`. Assumes that those
 /// conditions have been satisfied.
 static Optional<SmallVector<Value, 1>>
 fuseWithReshapeByExpansion(LinalgOp linalgOp, TensorReshapeOp reshapeOp,
                            unsigned fusedTensorIndex,
                            PatternRewriter &rewriter) {
   assert(isFusableWithReshapeByDimExpansion(linalgOp, fusedTensorIndex) &&
auto getReshapeInfo =
         }
       };
   SmallVector<Value, 4> expandedOpOperands;
   for (auto operand : llvm::enumerate(linalgOp.getInputs())) {
     if (operand.index() == fusedTensorIndex) {
       expandedOpOperands.push_back(reshapeOp.src());
       continue;
     }
-    AffineMap indexingMap = linalgOp.getIndexingMap(operand.index());
+    AffineMap indexingMap = linalgOp.getInputIndexingMap(operand.index());
     SmallVector<ReassociationIndices, 4> reassociation;
     SmallVector<int64_t, 4> expandedOperandShape;
     getReshapeInfo(indexingMap, reassociation, expandedOperandShape);
     Type expandedOperandType = RankedTensorType::get(
         expandedOperandShape,
         operand.value().getType().cast<ShapedType>().getElementType());
     if (expandedOperandType != operand.value().getType()) {
       expandedOpOperands.push_back(rewriter.create<TensorReshapeOp>(
           linalgOp.getLoc(), expandedOperandType, operand.value(),
           reassociation));
     } else {
       expandedOpOperands.push_back(operand.value());
     }
   }
-  SmallVector<Type, 1> resultTypes;
+
+  Location loc = linalgOp.getLoc();
+  SmallVector<Value, 1> outputs;
   SmallVector<SmallVector<ReassociationIndices, 4>, 1> resultReassociation;
-  for (auto result : llvm::enumerate(linalgOp->getResults())) {
-    AffineMap indexingMap =
-        linalgOp.getIndexingMap(linalgOp.getNumInputs() + result.index());
+  for (auto result : llvm::enumerate(linalgOp.getOutputs())) {
+    AffineMap indexingMap = linalgOp.getOutputIndexingMap(result.index());
     SmallVector<ReassociationIndices, 4> reassociation;
     SmallVector<int64_t, 4> expandedResultShape;
     getReshapeInfo(indexingMap, reassociation, expandedResultShape);
-    resultTypes.push_back(RankedTensorType::get(
-        expandedResultShape,
-        result.value().getType().cast<ShapedType>().getElementType()));
+    outputs.push_back(getOutputValueForExpansion(
+        rewriter, loc, indexingMap, result.value(), expandedDimsShape));
     resultReassociation.emplace_back(std::move(reassociation));
   }
 
   // The iterator types of the expanded op are all parallel.
   SmallVector<StringRef, 4> iteratorTypes(remapping.back(),
                                           getParallelIteratorTypeName());
 
+  TypeRange resultTypes = ValueRange(outputs).getTypes();
   LinalgOp fusedOp = createLinalgOpOfSameType(
       linalgOp, rewriter, linalgOp.getLoc(), resultTypes,
-      /*inputs=*/expandedOpOperands,
-      /*outputBuffers=*/ValueRange{},
-      /*initTensors=*/ValueRange{}, expandedOpIndexingMaps, iteratorTypes);
+      /*inputs=*/expandedOpOperands, outputs, expandedOpIndexingMaps,
+      iteratorTypes);
   Region &fusedRegion = fusedOp->getRegion(0);
   Region &originalRegion = linalgOp->getRegion(0);
 
   if (isa<GenericOp>(linalgOp.getOperation())) {
     rewriter.cloneRegionBefore(originalRegion, fusedRegion,
                                fusedRegion.begin());
   } else {
     assert(isa<IndexedGenericOp>(linalgOp.getOperation()));
for (auto result : llvm::enumerate(linalgOp->getResults())) {
     } else {
       resultVals.push_back(fusedOp->getResult(result.index()));
     }
   }
   // Assuming a single result.
   return resultVals;
 }
 
+static Value
+getOutputValueForLinearization(OpBuilder &builder, Location loc,
+                               Value origOutput,
+                               ArrayRef<AffineMap> reassociationMaps) {
+  SmallVector<Value, 4> dynamicDims;
+  SmallVector<int64_t, 4> staticDims;
+  auto shapedType = origOutput.getType().cast<ShapedType>();
+  ArrayRef<int64_t> origShape = shapedType.getShape();
+  for (auto map : reassociationMaps) {
+    Optional<Value> dynamicDim;
+    int64_t staticLinearizedShape = 1;
+    for (AffineDimExpr expr :
+         llvm::map_range(map.getResults(), [](AffineExpr e) {
+           return e.cast<AffineDimExpr>();
+         })) {
+      unsigned pos = expr.getPosition();
+      if (ShapedType::isDynamic(origShape[pos])) {
+        Value dim = builder.create<DimOp>(loc, origOutput, pos);
+        if (dynamicDim) {
+          dynamicDim = builder.create<MulIOp>(loc, dynamicDim.getValue(), dim);
+        } else {
+          dynamicDim = dim;
+        }
+      } else {
+        staticLinearizedShape *= origShape[pos];
+      }
+    }
+    if (dynamicDim) {
+      dynamicDim = builder.create<MulIOp>(
+          loc, dynamicDim.getValue(),
+          builder.create<ConstantIndexOp>(loc, staticLinearizedShape));
+      dynamicDims.push_back(dynamicDim.getValue());
+      staticDims.push_back(ShapedType::kDynamicSize);
+    } else {
+      staticDims.push_back(staticLinearizedShape);
+    }
+  }
+  return builder.create<InitTensorOp>(loc, dynamicDims, staticDims,
+                                      shapedType.getElementType());
+}
+
 namespace {
 
 /// Pattern to fold tensor_reshape op with its consumer by using the source of
 /// the reshape op as the operand in the consumer (instead of the result of the
 /// tensor_reshapeop) when the tensor_reshape op is collapsing. The
 /// corresponding index map in the consumer needs to be modified to linearize
 /// the folded dimension.
 ///
for (auto operand : llvm::enumerate(linalgOp.getInputs())) {
           !isTensorReshapeOpFoldableByLinearization(
               reshapeOp, linalgOp.getInputIndexingMap(operand.index()),
               /*asProducer =*/true))
         continue;
 
       // Compute the fused operands list,
       SmallVector<Value, 2> fusedOperands(linalgOp.getInputs());
       fusedOperands[operand.index()] = reshapeOp.src();
+      fusedOperands.append(linalgOp.getOutputs().begin(),
+                           linalgOp.getOutputs().end());
 
       // Compute indexing_maps for the fused operation. The indexing_maps for
       // the operands of the consumers that arent fused are the same.
       SmallVector<AffineMap, 4> fusedIndexMaps = llvm::to_vector<4>(
           op.indexing_maps().template getAsValueRange<AffineMapAttr>());
 
       // Accepted consumer maps are either identity or permutation.
       auto invMap = inversePermutation(fusedIndexMaps[operand.index()]);
for (auto operand : llvm::enumerate(linalgOp.getInputs())) {
       rewriter.startRootUpdate(op);
       op->setOperands(fusedOperands);
       op.indexing_mapsAttr(rewriter.getAffineMapArrayAttr(fusedIndexMaps));
       rewriter.finalizeRootUpdate(op);
       if (reshapeOp.use_empty())
         rewriter.eraseOp(reshapeOp);
       return success();
     }
-    return op.emitRemark("no fusion candidates found");
+    return failure();
   }
 };
 
 /// Pattern to fuse a tensor_reshape op with its consumer
 /// generic/indexed_generic op, when the reshape op is collapsing
 /// dimensions. The dimensionality of the loop in the consumer is expanded.
 template <typename GenericOpTy>
 struct FoldWithProducerReshapeOpByExpansion
LogicalResult matchAndRewrite(TensorReshapeOp reshapeOp,
     }
     fusedIndexMaps.back() = modifiedMap;
 
     // Further check that the resulting index maps can be fused and
     // inverted. Without this the resultant op is not legal.
     if (!inversePermutation(concatAffineMaps(fusedIndexMaps)))
       return reshapeOp.emitRemark("fused op loop bound computation failed");
 
+    Location loc = producer.getLoc();
+    Value output =
+        getOutputValueForLinearization(rewriter, loc, producer.getOutputs()[0],
+                                       reshapeOp.getReassociationMaps());
     LinalgOp fusedOp = createLinalgOpOfSameType(
-        producer, rewriter, rewriter.getUnknownLoc(), reshapeOp.getResultType(),
+        producer, rewriter, loc, reshapeOp.getResultType(),
         /*inputs=*/producer.getInputs(),
-        /*outputBuffers=*/ValueRange{},
-        /*initTensors=*/ValueRange{}, // no init tensors for now.
-        rewriter.getAffineMapArrayAttr(fusedIndexMaps),
+        // TODO: handle outputs.
+        /*outputs=*/output, rewriter.getAffineMapArrayAttr(fusedIndexMaps),
         producer.iterator_types(),
         /*doc=*/nullptr,
         /*library_call=*/nullptr,
         /*sparse=*/nullptr);
     auto &fusedRegion = fusedOp->getRegion(0);
     rewriter.cloneRegionBefore(producer->getRegion(0), fusedRegion,
                                fusedRegion.begin());
     rewriter.replaceOp(reshapeOp, fusedOp->getResults());
for (auto operand : llvm::enumerate(linalgOp.getInputs())) {
       Value scalarConstant = rewriter.create<ConstantOp>(
           constantOp.getLoc(),
           constantOp.value().cast<DenseElementsAttr>().getSplatValue());
 
       LinalgOp fusedOp = createLinalgOpOfSameType(
           linalgOp, rewriter, rewriter.getUnknownLoc(),
           linalgOp->getResultTypes(),
           /*inputs=*/fusedOperands,
-          /*outputBuffers=*/ValueRange{},
-          /*initTensors=*/ValueRange{}, // no init tensors for now.
+          /*outputs=*/linalgOp.getOutputs(),
           rewriter.getAffineMapArrayAttr(fusedIndexMaps),
           linalgOp.iterator_types(),
           /*doc=*/nullptr,
           /*library_call=*/nullptr,
           /*sparse=*/nullptr);
 
       // Map the block argument corresponding to the replaced argument with the
       // scalar constant.
       Region &linalgOpRegion = linalgOp->getRegion(0);
       Block &entryBlock = *linalgOpRegion.begin();
       unsigned argIndex = entryBlock.getNumArguments() -
-                          linalgOp.getNumInputs() + operand.index();
+                          linalgOp.getNumShapedOperands() + operand.index();
       BlockAndValueMapping mapping;
       mapping.map(entryBlock.getArgument(argIndex), scalarConstant);
       Region &fusedRegion = fusedOp->getRegion(0);
       rewriter.cloneRegionBefore(linalgOpRegion, fusedRegion,
                                  fusedRegion.begin(), mapping);
       rewriter.replaceOp(linalgOp, fusedOp->getResults());
       if (constantOp.use_empty())
         rewriter.eraseOp(constantOp);

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

static linalg::GenericOp createGenericOpFromNamedOp(linalg::LinalgOp namedOp,
 
   SmallVector<AffineMap, 4> indexingMaps = namedOp.getIndexingMaps();
   auto iterators = llvm::to_vector<4>(
       namedOp.iterator_types().getAsValueRange<StringAttr>());
   auto resultTypes = namedOp.getOutputTensorTypes();
   SmallVector<Type, 4> types(resultTypes.begin(), resultTypes.end());
 
   return builder.create<linalg::GenericOp>(
-      namedOp.getLoc(), types, namedOp.getInputs(), namedOp.getOutputBuffers(),
-      namedOp.getInitTensors(), indexingMaps, iterators,
+      namedOp.getLoc(), types, namedOp.getInputs(), namedOp.getOutputs(),
+      indexingMaps, iterators,
       [&regionBuilder](OpBuilder &bodyBuilder, Location loc, ValueRange) {
         edsc::ScopedContext scope(bodyBuilder, loc);
         regionBuilder(*bodyBuilder.getBlock());
       });
 }
 
 namespace {
 

 
 linalg::GenericOp GeneralizeConvOp::createGenericOp(linalg::ConvOp convOp,
                                                     OpBuilder &builder) const {
   SmallVector<AffineMap, 4> indexingMaps = convOp.getIndexingMaps();
   auto iterators =
       llvm::to_vector<4>(convOp.iterator_types().getAsValueRange<StringAttr>());
   return builder.create<linalg::GenericOp>(
       convOp.getLoc(), /*resultTensorTypes=*/ArrayRef<Type>(),
-      convOp.getInputBuffers(), convOp.getOutputBuffers(),
-      /*initTensors=*/ValueRange(), indexingMaps, iterators,
+      convOp.getInputBuffers(), convOp.getOutputBuffers(), indexingMaps,
+      iterators,
       [](OpBuilder &bodyBuilder, Location bodyLoc, ValueRange bodyArgs) {
         Value mul =
             bodyBuilder.create<MulFOp>(bodyLoc, bodyArgs[0], bodyArgs[1]);
         Value add = bodyBuilder.create<AddFOp>(bodyLoc, mul, bodyArgs[2]);
         bodyBuilder.create<linalg::YieldOp>(bodyLoc, add);
       });
 }
 

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

if (interchangeVector.empty())
     return op;
 
   MLIRContext *context = op.getContext();
   auto permutationMap = inversePermutation(
       AffineMap::getPermutationMap(interchangeVector, context));
   assert(permutationMap && "expected permutation to be invertible");
   SmallVector<Attribute, 4> newIndexingMaps;
   auto indexingMaps = op.indexing_maps().getValue();
-  for (unsigned i = 0, e = op.getNumInputsAndOutputs(); i != e; ++i) {
+  for (unsigned i = 0, e = op.getNumShapedOperands(); i != e; ++i) {
     AffineMap m = indexingMaps[i].cast<AffineMapAttr>().getValue();
     if (!permutationMap.isEmpty())
       m = m.compose(permutationMap);
     newIndexingMaps.push_back(AffineMapAttr::get(m));
   }
   auto itTypes = op.iterator_types().getValue();
   SmallVector<Attribute, 4> itTypesVector;
   for (unsigned i = 0, e = itTypes.size(); i != e; ++i)

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

struct LinalgOpInstancePromotionOptions {
   Optional<unsigned> alignment;
 };
 } // namespace
 
 LinalgOpInstancePromotionOptions::LinalgOpInstancePromotionOptions(
     LinalgOp linalgOp, const LinalgPromotionOptions &options)
     : subViews(), dynamicBuffers(options.dynamicBuffers),
       alignment(options.alignment) {
-  unsigned nBuffers = linalgOp.getNumInputsAndOutputBuffers();
+  assert(linalgOp.hasBufferSemantics() && "revisit usage of shaped operand");
+  unsigned nBuffers = linalgOp.getNumShapedOperands();
   auto vUseFullTileBuffers =
       options.useFullTileBuffers.getValueOr(llvm::SmallBitVector());
   vUseFullTileBuffers.resize(nBuffers, options.useFullTileBuffersDefault);
 
   for (unsigned idx = 0; idx != nBuffers; ++idx) {
     if (options.operandsToPromote && !options.operandsToPromote->count(idx))
       continue;
-    auto *op = linalgOp.getBuffer(idx).getDefiningOp();
+    auto *op = linalgOp.getShapedOperand(idx).getDefiningOp();
     if (auto sv = dyn_cast_or_null<SubViewOp>(op)) {
       subViews[idx] = sv;
       useFullTileBuffers[sv] = vUseFullTileBuffers[idx];
     }
   }
 
   allocationFn =
       (options.allocationFn ? *(options.allocationFn)
promoteSubViews(OpBuilder &b, LinalgOp op,
   if (!promotedBuffersAndViews ||
       promotedBuffersAndViews->size() != options.subViews.size())
     return {};
 
   // 2. Append all other operands as they appear, this enforces that such
   // operands are not views. This is to support cases such as FillOp taking
   // extra scalars etc.  Keep a reference to output buffers;
   SmallVector<Value, 8> opViews;
-  opViews.reserve(op.getNumInputsAndOutputs());
+  opViews.reserve(op.getNumShapedOperands());
   SmallVector<std::pair<Value, Value>, 8> writebackViews;
   writebackViews.reserve(promotedBuffersAndViews->size());
-  for (auto view : llvm::enumerate(op.getInputsAndOutputBuffers())) {
+  for (auto view : llvm::enumerate(op.getShapedOperands())) {
     if (options.subViews.count(view.index()) != 0) {
       if (options.useFullTileBuffers[view.value()])
         opViews.push_back(
             (*promotedBuffersAndViews)[view.index()].fullLocalView);
       else
         opViews.push_back(
             (*promotedBuffersAndViews)[view.index()].partialLocalView);
       if (view.index() >= op.getNumInputs())

 LogicalResult
 mlir::linalg::promoteSubviewsPrecondition(Operation *op,
                                           LinalgPromotionOptions options) {
   LinalgOp linOp = dyn_cast<LinalgOp>(op);
   // Transformation applies to buffers only.
   if (!linOp || !linOp.hasBufferSemantics())
     return failure();
   // Check that at least one of the requested operands is indeed a subview.
-  for (auto en : llvm::enumerate(linOp.getInputsAndOutputBuffers())) {
+  for (auto en : llvm::enumerate(linOp.getShapedOperands())) {
     auto sv = isa_and_nonnull<SubViewOp>(en.value().getDefiningOp());
     if (sv) {
       if (!options.operandsToPromote.hasValue() ||
           options.operandsToPromote->count(en.index()))
         return success();
     }
   }
   // TODO: Check all subviews requested are bound by a static constant.

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

struct CodeGen {
   Value redVal;
 };
 
 } // namespace
 
 /// Helper method to inspect sparse annotations in the linalg operation.
 /// Fills the per-dimension sparsity information for all tensors.
 static void findSparseAnnotations(Merger &merger, linalg::GenericOp op) {
-  unsigned numTensors = op.getNumInputsAndOutputs();
+  unsigned numTensors = op.getNumShapedOperands();
   ArrayAttr sparseAttr = op.sparseAttr();
   for (unsigned t = 0; t < numTensors; t++) {
     auto map = op.getIndexingMap(t);
     auto dimAttr = sparseAttr[t].cast<ArrayAttr>();
     // For each tensor, we accept a per-dimension Sparse or Dense annotation.
     // This is translated to the loop index that indexes that dimension.
     unsigned rank = op.getShapedType(t).getRank();
     for (unsigned d = 0; d < rank; d++) {
static unsigned buildLattices(Merger &merger, linalg::GenericOp op,
                               unsigned exp, unsigned idx) {
   Kind kind = merger.exp(exp).kind;
   if (kind == Kind::kTensor || kind == Kind::kInvariant) {
     // Either the index is really used in the tensor expression, or it is
     // set to the undefined index in that dimension. An invariant expression
     // is set to a synthetic tensor with undefined indices only.
     unsigned s = merger.addSet();
     unsigned t = kind == Kind::kTensor ? merger.exp(exp).e0
-                                       : op.getNumInputsAndOutputs();
+                                       : op.getNumShapedOperands() - 1;
     merger.set(s).push_back(merger.addLat(t, idx, exp));
     return s;
   }
   unsigned s0 = buildLattices(merger, op, merger.exp(exp).e0, idx);
   unsigned s1 = buildLattices(merger, op, merger.exp(exp).e1, idx);
   switch (kind) {
   case Kind::kTensor:
   case Kind::kInvariant:

 }
 
 /// Local bufferization of all dense and sparse data structures.
 /// This code enables testing the first prototype sparse compiler.
 // TODO: replace this with a proliferated bufferization strategy
 static void genBuffers(Merger &merger, CodeGen &codegen,
                        PatternRewriter &rewriter, linalg::GenericOp op) {
   Location loc = op.getLoc();
-  unsigned numTensors = op.getNumInputsAndOutputs();
+  unsigned numTensors = op.getNumShapedOperands();
   unsigned numInputs = op.getNumInputs();
   assert(numTensors == numInputs + 1);
 
   // For now, set all unknown dimensions to 999.
   // TODO: compute these values (using sparsity or by reading tensor)
   Value unknown = rewriter.create<ConstantIndexOp>(loc, 999);
 
   // For every tensor, find lower and upper bound on dimensions, set the
for (unsigned d = 0, rank = shape.size(); d < rank; d++) {
         codegen.pointers[t][i] = rewriter.create<AllocaOp>(loc, ptrTp, unknown);
         codegen.indices[t][i] = rewriter.create<AllocaOp>(loc, indTp, unknown);
       }
       // Find lower and upper bound in current dimension.
       Value up;
       if (shape[d] == TensorType::kDynamicSize) {
         // For the output tensor, we may need to infer the upper bound.
         // For all others, we look at the incoming argument.
         if (t == numInputs && !op.getNumInitTensors()) {
           up = codegen.sizes[i];
           assert(up); // TODO: what else?
         } else {
-          Value arg = t < numInputs ? op.getInput(t) : op.getInitTensor(0);
+          Value arg = t < numInputs ? op.getInput(t) : op.getInitTensors()[0];
           up = rewriter.create<DimOp>(loc, arg, d);
         }
         args.push_back(up);
       } else {
         up = rewriter.create<ConstantIndexOp>(loc, shape[d]);
       }
       codegen.sizes[i] = codegen.highs[t][i] = up;
     }
static Value genTensorLoad(Merger &merger, CodeGen &codegen,
   return rewriter.create<LoadOp>(loc, ptr, args);
 }
 
 /// Generates a store on a dense tensor.
 static void genTensorStore(Merger &merger, CodeGen &codegen,
                            PatternRewriter &rewriter, linalg::GenericOp op,
                            unsigned tensor, Value rhs) {
   // Test if this is a scalarized reduction.
-  unsigned lhs = op.getNumInputsAndOutputs() - 1;
+  unsigned lhs = op.getNumShapedOperands() - 1;
   if (lhs == tensor && codegen.redVal) {
     codegen.redVal = rhs;
     return;
   }
   // Actual load.
   SmallVector<Value, 4> args;
   auto map = op.getIndexingMap(tensor);
   for (unsigned i = 0, m = map.getNumResults(); i < m; ++i) {
if (merger.exp(exp).kind == Kind::kTensor) {
     for (unsigned i = 0, m = map.getNumResults(); i < m; ++i) {
       unsigned idx = map.getDimPosition(i);
       if (!codegen.loops[idx])
         return; // still in play
       else if (idx == ldx)
         atLevel = true;
     }
     // All exhausted at this level (atLevel denotes exactly at this level).
-    unsigned lhs = op.getNumInputsAndOutputs() - 1;
+    unsigned lhs = op.getNumShapedOperands() - 1;
     if (lhs == tensor) {
       codegen.redExp = hoist ? exp : -1u;
     } else if (atLevel) {
       merger.exp(exp).val =
           hoist ? genTensorLoad(merger, codegen, rewriter, op, exp) : Value();
     }
   } else if (merger.exp(exp).kind != Kind::kInvariant) {
     // Traverse into the binary operations. Note that we only hoist

 /// Recursively generates code while computing iteration lattices in order
 /// to manage the complexity of implementing co-iteration over unions
 /// and intersections of sparse iterations spaces.
 static void genStmt(Merger &merger, CodeGen &codegen, PatternRewriter &rewriter,
                     linalg::GenericOp op, std::vector<unsigned> &topSort,
                     unsigned exp, unsigned at) {
   // At each leaf, assign remaining tensor (sub)expression to output tensor.
   if (at == topSort.size()) {
-    unsigned lhs = op.getNumInputsAndOutputs() - 1;
+    unsigned lhs = op.getNumShapedOperands() - 1;
     Value rhs = genExp(merger, codegen, rewriter, op, exp);
     genTensorStore(merger, codegen, rewriter, op, lhs, rhs);
     return;
   }
 
   // Construct iteration lattices for current loop index, with L0 at top.
   // Then emit initialization code for the loop sequence at this level.
   // We maintain the universal dense index if dense indices are still
for (unsigned i = 0; i < lsize; i++) {
     }
     rewriter.setInsertionPointAfter(loop);
   }
 
   // Wrap-up loop sequence.
   Value red = codegen.redVal;
   if (red) {
     codegen.redVal = merger.exp(codegen.redExp).val = Value(); // end chain
-    unsigned lhs = op.getNumInputsAndOutputs() - 1;
+    unsigned lhs = op.getNumShapedOperands() - 1;
     genTensorStore(merger, codegen, rewriter, op, lhs, red);
   }
   codegen.loops[idx] = Value();
   genInvariants(merger, codegen, rewriter, op, exp, ldx, /*hoist=*/false);
 }
 
 namespace {
 
 /// Sparse rewriting rule for generic Lingalg operation.
 struct GenericOpSparsifier : public OpRewritePattern<linalg::GenericOp> {
 public:
   GenericOpSparsifier(MLIRContext *context, linalg::SparsificationOptions o)
       : OpRewritePattern<linalg::GenericOp>(context), options(o) {}
 
   LogicalResult matchAndRewrite(linalg::GenericOp op,
                                 PatternRewriter &rewriter) const override {
     // Detects sparse annotations and translate the per-dimension sparsity
     // information for all tensors to loop indices in the kernel.
     if (!op.hasSparseSemantics())
       return failure();
     assert(op.getNumOutputs() == 1);
-    unsigned numTensors = op.getNumInputsAndOutputs();
+    unsigned numTensors = op.getNumShapedOperands();
     unsigned numLoops = op.iterator_types().getValue().size();
     Merger merger(numTensors, numLoops);
     findSparseAnnotations(merger, op);
 
     // Computes a topologically sorted iteration graph to ensure
     // tensors are visited in natural index order. Fails on cycles.
     // This assumes that higher-level passes have already put the
     // tensors in each tensor expression in a feasible order.

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

if (!options.interchangeVector.empty()) {
     assert(invPermutationMap);
     applyPermutationToVector(loopRanges, interchangeVector);
     applyPermutationToVector(iteratorTypes, interchangeVector);
   }
 
   // 2. Create the tiled loops.
   LinalgOp res = op;
   SmallVector<Value, 4> ivs, tensorResults;
-  auto initTensors = op.getInitTensors();
+  auto outputTensors = op.getOutputTensors();
   GenerateLoopNest<LoopTy>::doit(
-      loopRanges, /*iterArgInitValues*/ initTensors, iteratorTypes,
+      loopRanges, /*iterArgInitValues*/ outputTensors, iteratorTypes,
       [&](ValueRange localIvs, ValueRange iterArgs) -> scf::ValueVector {
         auto &b = ScopedContext::getBuilderRef();
         auto loc = ScopedContext::getLocation();
         ivs.assign(localIvs.begin(), localIvs.end());
 
         // When an `interchangeVector` is present, it has been applied to the
         // loop ranges and the iterator types. Apply its inverse to the
         // resulting loop `ivs` to match the op definition.
         SmallVector<Value, 4> interchangedIvs;
         if (!options.interchangeVector.empty())
           interchangedIvs = applyMapToValues(b, loc, invPermutationMap, ivs);
         else
           interchangedIvs.assign(ivs.begin(), ivs.end());
 
-        assert(op.getNumInitTensors() == iterArgs.size() &&
-               "num init tensors must match number of loop iter arguments");
-        // This uses knowledge about position of the init tensor in the list
-        // of operands.
-        auto operands = llvm::to_vector<4>(op.getShapedOperands());
-        std::copy(iterArgs.begin(), iterArgs.end(),
-                  operands.begin() + op.getNumInputsAndOutputBuffers());
+        assert(op.getNumOutputTensors() == iterArgs.size() &&
+               "num output tensors must match number of loop iter arguments");
 
+        auto operands = llvm::to_vector<4>(op.getInputs());
+        SmallVector<Value, 4> outputBuffers = op.getOutputBuffers();
+        // TODO: thanks to simplifying assumption we do not need to worry about
+        // order of output buffers and tensors: there is only ever one kind.
+        assert(outputBuffers.empty() || iterArgs.empty());
+        operands.append(outputBuffers.begin(), outputBuffers.end());
+        operands.append(iterArgs.begin(), iterArgs.end());
         SmallVector<Value, 4> tiledOperands =
             makeTiledShapes(b, loc, op, operands, shapeSizesToLoopsMap,
                             interchangedIvs, tileSizes, allShapeSizes);
         auto nonShapedOperands = op.getAssumedNonShapedOperands();
         tiledOperands.append(nonShapedOperands.begin(),
                              nonShapedOperands.end());
 
-        // If LinalgOp has results, they must all be tied to init tensors.
-        // We enforce this to ensure all tiled ops have been rewritten in
-        // "init tensor" form. This ensures tiling has anchor values into which
-        // to subtensor / subtensor_insert. Otherwise tiling would need to
-        // allocate which is not acceptable.
-        // This would not be the case with a special terminator op that
-        // generates the whole tensor (instead of inserting a subtensor). But
-        // the generator-based abstraction has other issues.
-        assert(op.getNumInitTensors() == op->getNumResults() &&
-               "expected same number of init tensors as number of results");
-
-        // Handle init tensor operands.
-        // This uses knowledge about position of the init tensor in the list
-        // of operands.
-        // TODO: InterfaceAdaptor ?
+        // TODO: use an interface/adaptor to avoid leaking position in
+        // `tiledOperands`.
         SmallVector<Type, 4> resultTensorTypes;
-        for (auto idx : llvm::seq<unsigned>(0, op.getNumInitTensors()))
+        for (OpOperand *opOperand : op.getOutputTensorsOpOperands())
           resultTensorTypes.push_back(
-              tiledOperands[op.getNumInputsAndOutputBuffers() + idx].getType());
+              tiledOperands[opOperand->getOperandNumber()].getType());
 
         res = op.clone(b, loc, resultTensorTypes, tiledOperands);
 
-        // Insert a subtensor_insert for each init subtensor.
-        for (unsigned idx = 0, e = op.getNumInitTensors(); idx != e; ++idx) {
-          Value initTensor =
-              tiledOperands[op.getNumInputsAndOutputBuffers() + idx];
-          if (auto subtensor = initTensor.getDefiningOp<SubTensorOp>()) {
+        // Insert a subtensor_insert for each output tensor.
+        unsigned resultIdx = 0;
+        for (OpOperand *opOperand : op.getOutputTensorsOpOperands()) {
+          // TODO: use an interface/adaptor to avoid leaking position in
+          // `tiledOperands`.
+          Value outputTensor = tiledOperands[opOperand->getOperandNumber()];
+          if (auto subtensor = outputTensor.getDefiningOp<SubTensorOp>()) {
             tensorResults.push_back(b.create<SubTensorInsertOp>(
-                loc, subtensor.source().getType(), res->getResult(idx),
+                loc, subtensor.source().getType(), res->getResult(resultIdx),
                 subtensor.source(), subtensor.offsets(), subtensor.sizes(),
                 subtensor.strides(), subtensor.static_offsets(),
                 subtensor.static_sizes(), subtensor.static_strides()));
           } else {
-            tensorResults.push_back(res->getResult(idx));
+            tensorResults.push_back(res->getResult(resultIdx));
           }
+          ++resultIdx;
         }
         return scf::ValueVector(tensorResults.begin(), tensorResults.end());
       },
       options.distribution);
 
   // 3. Transforms index arguments of `linalg.generic` w.r.t. to the tiling.
   transformIndexedGenericOpIndices(b, res, ivs, loopIndexToRangeIndex);
 

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

LogicalResult mlir::linalg::LinalgBaseTilingPattern::matchAndRewriteBase(
     Operation *op, PatternRewriter &rewriter,
     SmallVectorImpl<Value> &tensorResults) const {
   LinalgOp linalgOp = dyn_cast<LinalgOp>(op);
   if (!linalgOp)
     return failure();
   if (failed(marker.checkAndNotify(rewriter, linalgOp)))
     return failure();
 
-  // If LinalgOp has results, they must all be tied to init tensors.
-  // We enforce this to ensure all tiled ops have been rewritten in
-  // "init tensor" form. This ensures tiling has anchor values into which to
-  // subtensor / subtensor_insert. Otherwise tiling would need to allocate which
-  // is not acceptable.
-  // This would not be the case with a special terminator op that generates the
-  // whole tensor (instead of inserting a subtensor). But the generator-based
-  // abstraction has other issues.
-  if (linalgOp.getNumInitTensors() != linalgOp->getNumResults())
-    return failure();
-
   Optional<TiledLinalgOp> res = tileLinalgOp(rewriter, linalgOp, options);
 
   if (!res)
     return failure();
 
   // Return relevant information to derived pattern.
   tensorResults = res->tensorResults;
 
if (failed(marker.checkAndNotify(rewriter, linalgOp)))
     return failure();
   if (!linalgOp.hasBufferSemantics())
     return failure();
 
   DenseSet<Operation *> producers;
   producers.insert(linalgOp);
   for (auto dependence : dependenceGraph.getDependentOperations(linalgOp)) {
     if (!fusionOptions.indicesToFuse.count(
-            dependence.indexingOpView.operandIndex))
+            dependence.indexingOpView->getOperandNumber()))
       continue;
-    if (isa<LinalgOp>(dependence.dependentOpView.op))
-      producers.insert(dependence.dependentOpView.op);
+    if (isa<LinalgOp>(dependence.dependentOpView->getOwner()))
+      producers.insert(dependence.dependentOpView->getOwner());
   }
 
   SmallVector<LinalgOp, 1> fusionOps;
   for (auto it = op->getBlock()->begin(), ie = Block::iterator(op); it != ie;
        ++it) {
     auto producerLinalgOp = dyn_cast<LinalgOp>(&(*it));
     if (producerLinalgOp && producers.count(producerLinalgOp))
       fusionOps.push_back(producerLinalgOp);

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

Value vectorize(Value scalarValue) {
     auto vectorValueIt = valueCache.find(scalarValue);
     if (vectorValueIt != valueCache.end())
       return vectorValueIt->second;
 
     // If the value is from the region but not in the cache it means it is a
     // block argument.
     auto scalarArg = scalarValue.cast<BlockArgument>();
     assert(scalarArg.getOwner() == &generic.region().front());
-    Value vector_arg =
-        generic.getInputsAndOutputBuffers()[scalarArg.getArgNumber()];
-    Value vectorResult = transferReadVector(builder, vector_arg);
+    Value vectorArg = generic.getShapedOperand(scalarArg.getArgNumber());
+    Value vectorResult = transferReadVector(builder, vectorArg);
     valueCache[scalarArg] = vectorResult;
     return vectorResult;
   }
 
   // Return the largest shape of all the given values. Return an empty
   // SmallVector if there are no vector value.
   static SmallVector<int64_t, 4> getLargestShape(ArrayRef<Value> values) {
     SmallVector<int64_t, 4> largestShape;
static void vectorizeElementwise(linalg::GenericOp op, OpBuilder &builder) {
   for (Operation &scalarOp : op.region().front()) {
     vectorizer.vectorize(scalarOp);
   }
 }
 
 LogicalResult mlir::linalg::vectorizeLinalgOpPrecondition(Operation *op) {
   auto linalgOp = cast<linalg::LinalgOp>(op);
   // All types must be static shape to go to vector.
-  for (Value operand : linalgOp.getInputsAndOutputBuffers())
+  for (Value operand : linalgOp.getShapedOperands())
     if (!operand.getType().cast<ShapedType>().hasStaticShape())
       return failure();
   for (Type outputTensorType : linalgOp.getOutputTensorTypes())
     if (!outputTensorType.cast<ShapedType>().hasStaticShape())
       return failure();
 
   if (isa<linalg::FillOp, linalg::CopyOp>(op))
     return success();

mlir/lib/Dialect/Linalg/Utils/Utils.cpp

 namespace linalg {
 
 SmallVector<int64_t, 8> getStaticShape(LinalgOp linalgOp) {
   SmallVector<int64_t, 8> res;
   for (Value v : linalgOp.getShapedOperands()) {
     auto shape = v.getType().cast<ShapedType>().getShape();
     res.append(shape.begin(), shape.end());
   }
-  if (linalgOp.getNumInitTensors())
-    return res;
-  for (Value v : linalgOp.getOperation()->getResults()) {
-    auto shape = v.getType().cast<ShapedType>().getShape();
-    res.append(shape.begin(), shape.end());
-  }
   return res;
 }
 
 Optional<SmallVector<int64_t, 4>> getStaticLoopRanges(LinalgOp linalgOp) {
   SmallVector<int64_t, 8> viewSizes = getStaticShape(linalgOp);
   AffineMap invertedMap = linalgOp.getShapesToLoopsMap();
   if (!invertedMap)
     return {};

mlir/lib/Dialect/StandardOps/IR/Ops.cpp

LogicalResult matchAndRewrite(DimOp dimOp,
     auto castOp = dimOp.memrefOrTensor().getDefiningOp<CastOpTy>();
     if (!castOp)
       return failure();
     Value newSource = castOp.getOperand();
     rewriter.replaceOpWithNewOp<DimOp>(dimOp, newSource, dimOp.index());
     return success();
   }
 };
-
 } // end anonymous namespace.
 
 void DimOp::getCanonicalizationPatterns(OwningRewritePatternList &results,
                                         MLIRContext *context) {
-  results.insert<DimOfMemRefReshape, DimOfCastOp<tensor::CastOp>>(context);
+  results.insert<DimOfMemRefReshape, DimOfCastOp<TensorToMemrefOp>,
+                 DimOfCastOp<tensor::CastOp>>(context);
 }
 
 // ---------------------------------------------------------------------------
 // DmaStartOp
 // ---------------------------------------------------------------------------
 
 void DmaStartOp::build(OpBuilder &builder, OperationState &result,
                        Value srcMemRef, ValueRange srcIndices, Value destMemRef,

mlir/test/Dialect/Linalg/bufferize.mlir

-// RUN: mlir-opt -linalg-bufferize -split-input-file %s | FileCheck %s
+// RUN: mlir-opt -linalg-bufferize  -canonicalize -cse -split-input-file %s | FileCheck %s
 
 #map0 = affine_map<(d0) -> (d0)>
 
 // In-depth checking of a basic case, this is testing
 // - tensor_to_memref / tensor_load materializations are properly inserted
 // - payload is correctly carried over
 // - affine maps are correctly carried over
 // Later tests will not check all these details.

 // CHECK:             linalg.yield %[[DIM1]] : f32
 // CHECK:           }
 // CHECK:           %[[RESULT:.*]] = tensor_load %[[RESULT_MEMREF]] : memref<4xf32>
 // CHECK:           return %[[RESULT]] : tensor<4xf32>
 func @basic(%arg0: tensor<4xf32>) -> tensor<4xf32> {
     %0 = linalg.generic {
       indexing_maps = [#map0, #map0],
       iterator_types = ["parallel"]
-    } ins(%arg0 : tensor<4xf32>) {
-      ^bb0(%gen_arg1: f32):
+    } ins(%arg0 : tensor<4xf32>)
+      outs(%arg0 : tensor<4xf32>) {
+      ^bb0(%gen_arg1: f32, %out: f32):
         %tmp1 = exp %gen_arg1 : f32
         linalg.yield %tmp1 : f32
     } -> tensor<4xf32>
     return %0 : tensor<4xf32>
 }
 
 
 // -----
 
 #map0 = affine_map<(d0) -> (d0)>
 
+// Same as above but with linalg.init_tensor op.
+
+// CHECK: #map = affine_map<(d0) -> (d0)>
+// CHECK-LABEL: func @init_tensor(
+// CHECK-SAME:      %[[IN:.*]]: tensor<?xf32>, %[[SIZE:.*]]: index)
+// CHECK:         %[[OUT_BUF:.*]] = alloc(%[[SIZE]]) : memref<?xf32>
+// CHECK:         %[[MEMREF:.*]] = tensor_to_memref %[[IN]] : memref<?xf32>
+// CHECK:         linalg.generic
+// CHECK-SAME:    ins(%[[MEMREF]] : memref<?xf32>)
+// CHECK-SAME:    outs(%[[OUT_BUF]] : memref<?xf32>) {
+func @init_tensor(%in : tensor<?xf32>, %size: index) -> tensor<?xf32> {
+  %init = linalg.init_tensor [%size] : tensor<?xf32>
+  %0 = linalg.generic {
+    indexing_maps = [#map0, #map0],
+    iterator_types = ["parallel"]
+  } ins(%in : tensor<?xf32>)
+    outs(%init : tensor<?xf32>) {
+    ^bb0(%gen_arg1: f32, %out: f32):
+      %tmp1 = exp %gen_arg1 : f32
+      linalg.yield %tmp1 : f32
+  } -> tensor<?xf32>
+  return %0 : tensor<?xf32>
+}
+
+
+// -----
+
+#map0 = affine_map<(d0) -> (d0)>
+
 // CHECK-LABEL:   func @multiple_results
 // CHECK:           %[[RESULT0:.*]] = alloc() : memref<4xf32>
 // CHECK:           %[[RESULT1:.*]] = alloc() : memref<4xf32>
 // CHECK:           linalg.generic
 // CHECK-SAME:      ins(%{{.*}} : memref<4xf32>)
 // CHECK-SAME:      outs(%[[RESULT0]], %[[RESULT1]] : memref<4xf32>, memref<4xf32>)
 // CHECK-NEXT: ^bb0(%{{.*}}: f32, %{{.*}}: f32, %{{.*}}: f32):
 func @multiple_results(%arg0: tensor<4xf32>) -> (tensor<4xf32>, tensor<4xf32>) {
     %0, %1 = linalg.generic {
       indexing_maps = [#map0, #map0, #map0],
       iterator_types = ["parallel"]
-    } ins(%arg0 : tensor<4xf32>) {
-      ^bb0(%gen_arg1: f32):
+    } ins(%arg0 : tensor<4xf32>)
+      outs (%arg0, %arg0 : tensor<4xf32>, tensor<4xf32>) {
+      ^bb0(%gen_arg1: f32, %out1: f32, %out2: f32):
         %tmp1 = exp %gen_arg1 : f32
         linalg.yield %tmp1, %tmp1 : f32, f32
     } -> tensor<4xf32>, tensor<4xf32>
     return %0, %1 : tensor<4xf32>, tensor<4xf32>
 }
 
 // -----
 
 #map0 = affine_map<(d0) -> (d0)>
 
 // CHECK-LABEL:   func @multiple_results_indexed
 // CHECK:           %[[RESULT0:.*]] = alloc() : memref<4xi32>
 // CHECK:           %[[RESULT1:.*]] = alloc() : memref<4xi32>
 // CHECK:           linalg.indexed_generic
 // CHECK-SAME:      ins(%{{.*}} : memref<4xi32>)
 // CHECK-SAME:      outs(%[[RESULT0]], %[[RESULT1]] : memref<4xi32>, memref<4xi32>)
 // CHECK-NEXT: ^bb0(%{{.*}}: index, %{{.*}}: i32, %{{.*}}: i32, %{{.*}}: i32):
 func @multiple_results_indexed(%arg0: tensor<4xi32>)
         -> (tensor<4xi32>, tensor<4xi32>) {
     %0, %1 = linalg.indexed_generic {
       indexing_maps = [#map0, #map0, #map0],
       iterator_types = ["parallel"]
-    } ins(%arg0 : tensor<4xi32>) {
-      ^bb0(%i: index, %gen_arg1: i32):
+    } ins(%arg0 : tensor<4xi32>)
+      outs (%arg0, %arg0 : tensor<4xi32>, tensor<4xi32>) {
+      ^bb0(%i: index, %gen_arg1: i32, %out1: i32, %out2: i32):
         %i_i32 = index_cast %i : index to i32
         %tmp1 = addi %gen_arg1, %i_i32 : i32
         linalg.yield %tmp1, %tmp1 : i32, i32
     } -> tensor<4xi32>, tensor<4xi32>
     return %0, %1 : tensor<4xi32>, tensor<4xi32>
 }
 
 // -----
 
 #map_2d = affine_map<(d0, d1) -> (d0, d1)>
-#map_2d_inv = affine_map<(d0, d1) -> (d1, d0)>
 
 // Check that the allocs properly consider the different shapes of the output
 // operands. The permuted indexing maps translate to different output shapes.
 
-// CHECK: #map0 = affine_map<(d0, d1) -> (d0, d1)>
-// CHECK: #map1 = affine_map<(d0, d1) -> (d1, d0)>
 // CHECK-LABEL:   func @dynamic_results(
 // CHECK-SAME:                          %[[ARG:.*]]: tensor<?x?xf32>
-// CHECK:           %[[MEMREF_ARG:.*]] = tensor_to_memref %[[ARG]] : memref<?x?xf32>
 // CHECK:           %[[C0:.*]] = constant 0 : index
-// CHECK:           %[[DIM0:.*]] = dim %[[ARG]], %[[C0]] : tensor<?x?xf32>
 // CHECK:           %[[C1:.*]] = constant 1 : index
+// CHECK:           %[[MEMREF_ARG:.*]] = tensor_to_memref %[[ARG]] : memref<?x?xf32>
+// CHECK:           %[[DIM0:.*]] = dim %[[ARG]], %[[C0]] : tensor<?x?xf32>
 // CHECK:           %[[DIM1:.*]] = dim %[[ARG]], %[[C1]] : tensor<?x?xf32>
 // CHECK:           %[[RESULT0:.*]] = alloc(%[[DIM0]], %[[DIM1]]) : memref<?x?xf32>
-// CHECK:           %[[RESULT1:.*]] = alloc(%[[DIM1]], %[[DIM0]]) : memref<?x?xf32>
-// CHECK:           linalg.generic {indexing_maps = [#map0, #map0, #map1]
+// CHECK:           %[[RESULT1:.*]] = alloc(%[[DIM0]], %[[DIM1]]) : memref<?x?xf32>
+// CHECK:           linalg.generic
 // CHECK-SAME:      ins(%[[MEMREF_ARG]] : memref<?x?xf32>)
 // CHECK-SAME:      outs(%[[RESULT0]], %[[RESULT1]] : memref<?x?xf32>, memref<?x?xf32>)
 func @dynamic_results(%arg0: tensor<?x?xf32>)
          -> (tensor<?x?xf32>, tensor<?x?xf32>) {
     %0, %1 = linalg.generic {
-      indexing_maps = [#map_2d, #map_2d, #map_2d_inv],
+      indexing_maps = [#map_2d, #map_2d, #map_2d],
       iterator_types = ["parallel", "parallel"]
-    } ins(%arg0 : tensor<?x?xf32>) {
-      ^bb0(%gen_arg1: f32):
+    } ins(%arg0 : tensor<?x?xf32>)
+      outs (%arg0, %arg0 : tensor<?x?xf32>, tensor<?x?xf32>) {
+      ^bb0(%gen_arg1: f32, %out1: f32, %out2: f32):
         %tmp1 = exp %gen_arg1 : f32
         linalg.yield %tmp1, %tmp1 : f32, f32
     } -> tensor<?x?xf32>, tensor<?x?xf32>
     return %0, %1 : tensor<?x?xf32>, tensor<?x?xf32>
 }
 
 // -----
 

 // CHECK:           linalg.generic
 // CHECK-SAME:      ins(%[[ARG0_MEMREF]] : memref<2x3x4xvector<3x4xi4>>)
 // CHECK-SAME:      outs(%[[INIT_BUFFER]] : memref<3x2xf32>) {
 func @generic_with_init_tensor(%arg0: tensor<2x3x4xvector<3x4xi4>>,
   %arg1: tensor<3x2xf32>) -> (tensor<3x2xf32>) {
 
   %0 = linalg.generic #trait
     ins(%arg0 : tensor<2x3x4xvector<3x4xi4>>)
-   init(%arg1 : tensor<3x2xf32>) {
+   outs(%arg1 : tensor<3x2xf32>) {
     ^bb(%v0: vector<3x4xi4>, %v1: f32) :
-      %f0 = constant 0.0 : f32
-      linalg.yield %f0 : f32
+      linalg.yield %v1 : f32
   } -> tensor<3x2xf32>
 
   return %0 : tensor<3x2xf32>
 }
 
 // -----
 
 // CHECK-DAG: #[[$MAP0:[0-9a-z]*]] = affine_map<(d0, d1)[s0, s1] -> (d0 * s1 + s0 + d1)>

 // CHECK-LABEL: func @bufferize_subtensor_insert(
 //  CHECK-SAME:   %[[T:[0-9a-z]*]]: tensor<?x?xf32>
 //  CHECK-SAME:   %[[ST0:[0-9a-z]*]]: tensor<2x3xf32>
 //  CHECK-SAME:   %[[ST1:[0-9a-z]*]]: tensor<2x?xf32>
 func @bufferize_subtensor_insert(%t : tensor<?x?xf32>, %st0 : tensor<2x3xf32>, %st1 : tensor<2x?xf32>) ->
     (tensor<?x?xf32>, tensor<?x?xf32>) {
   %c0 = constant 0 : index
   %c1 = constant 1 : index
-  //      CHECK: %[[IDX:.*]] = call @make_index() : () -> index
+  // CHECK-NEXT: %[[C0:.*]] = constant 0 : index
+  // CHECK-NEXT: %[[C1:.*]] = constant 1 : index
   %i0 = call @make_index() : () -> index
+  // CHECK: %[[IDX:.*]] = call @make_index() : () -> index
 
 
-  //  CHECK-DAG: %[[M0:.*]] = tensor_to_memref %[[T]] : memref<?x?xf32>
-  //  CHECK-DAG: %[[SM0:.*]] = tensor_to_memref %[[ST0]] : memref<2x3xf32>
-  // CHECK-NEXT: %[[C0:.*]] = constant 0 : index
-  // CHECK-NEXT: %[[DIM0:.*]] = dim %[[M0]], %[[C0]] : memref<?x?xf32>
-  // CHECK-NEXT: %[[C1:.*]] = constant 1 : index
-  // CHECK-NEXT: %[[DIM1:.*]] = dim %[[M0]], %[[C1]] : memref<?x?xf32>
+  // CHECK-DAG: %[[M0:.*]] = tensor_to_memref %[[T]] : memref<?x?xf32>
+  // CHECK-DAG: %[[SM0:.*]] = tensor_to_memref %[[ST0]] : memref<2x3xf32>
+  // CHECK-NEXT: %[[DIM0:.*]] = dim %[[T]], %[[C0]] : tensor<?x?xf32>
+  // CHECK-NEXT: %[[DIM1:.*]] = dim %[[T]], %[[C1]] : tensor<?x?xf32>
   // CHECK-NEXT: %[[M0_COPY:.*]] = alloc(%[[DIM0]], %[[DIM1]]) : memref<?x?xf32>
   // CHECK-NEXT: linalg.copy(%[[M0]], %[[M0_COPY]]) : memref<?x?xf32>, memref<?x?xf32>
   // CHECK-NEXT: %[[SUBVIEW0:.*]] = subview %[[M0_COPY]][0, 0] [2, 3] [1, 1]
   // CHECK-SAME:   memref<?x?xf32> to memref<2x3xf32, #[[$MAP0]]>
   // CHECK-NEXT: linalg.copy(%[[SM0]], %[[SUBVIEW0]]) : memref<2x3xf32>, memref<2x3xf32, #[[$MAP0]]>
   // CHECK-NEXT: %[[RT0:.*]] = tensor_load %[[M0_COPY]] : memref<?x?xf32>
   %t0 = subtensor_insert %st0 into %t[0, 0][2, 3][1, 1] : tensor<2x3xf32> into tensor<?x?xf32>
 
   //  CHECK-DAG: %[[M1:.*]] = tensor_to_memref %[[T]] : memref<?x?xf32>
   //  CHECK-DAG: %[[SM1:.*]] = tensor_to_memref %[[ST1]] : memref<2x?xf32>
-  // CHECK-NEXT: %[[C0:.*]] = constant 0 : index
-  // CHECK-NEXT: %[[DIM0:.*]] = dim %[[M1]], %[[C0]] : memref<?x?xf32>
-  // CHECK-NEXT: %[[C1:.*]] = constant 1 : index
-  // CHECK-NEXT: %[[DIM1:.*]] = dim %[[M1]], %[[C1]] : memref<?x?xf32>
   // CHECK-NEXT: %[[M1_COPY:.*]] = alloc(%[[DIM0]], %[[DIM1]]) : memref<?x?xf32>
   // CHECK-NEXT: linalg.copy(%[[M1]], %[[M1_COPY]]) : memref<?x?xf32>, memref<?x?xf32>
   // CHECK-NEXT: %[[SUBVIEW1:.*]] = subview %[[M1_COPY]][0, %[[IDX]]] [2, %[[IDX]]] [1, 2]
   // CHECK-SAME:   memref<?x?xf32> to memref<2x?xf32, #[[$MAP1]]>
   // CHECK-NEXT: linalg.copy(%[[SM1]], %[[SUBVIEW1]]) : memref<2x?xf32>, memref<2x?xf32, #[[$MAP1]]>
   // CHECK-NEXT: %[[RT1:.*]] = tensor_load %[[M1_COPY]] : memref<?x?xf32>
   %t1 = subtensor_insert %st1 into %t[0, %i0][2, %i0][1, 2] : tensor<2x?xf32> into tensor<?x?xf32>
 
   //     CHECK: return %[[RT0]], %[[RT1]]
   return %t0, %t1: tensor<?x?xf32>, tensor<?x?xf32>
 }
+

mlir/test/Dialect/Linalg/canonicalize-duplicate-inputs.mlir

 // RUN: mlir-opt %s -split-input-file -canonicalize | FileCheck %s
 
 // Test case: Most basic case. Adding a vector to itself.
 
 #map = affine_map<(d0) -> (d0)>
 
 // CHECK: #[[$MAP:.*]] = affine_map<(d0) -> (d0)>
 // CHECK-LABEL: @basic
 func @basic(%arg0: tensor<?xf32>) -> tensor<?xf32> {
   // CHECK: linalg.generic{{.*}}[#[[$MAP]], #[[$MAP]]]
-  // CHECK:   ^bb0(%[[BBARG:.*]]: f32):
+  // CHECK:   ^bb0(%[[BBARG:.*]]: f32, %{{.*}}: f32):
   // CHECK:     addf %[[BBARG]], %[[BBARG]]
-  %0 = linalg.generic {indexing_maps = [#map, #map, #map], iterator_types = ["parallel"]} ins(%arg0, %arg0 : tensor<?xf32>, tensor<?xf32>) {
-  ^bb0(%arg1: f32, %arg2: f32):
+  %0 = linalg.generic {indexing_maps = [#map, #map, #map], iterator_types = ["parallel"]}
+     ins(%arg0, %arg0 : tensor<?xf32>, tensor<?xf32>)
+    outs(%arg0 : tensor<?xf32>) {
+  ^bb0(%arg1: f32, %arg2: f32, %arg3: f32):
     %1 = addf %arg1, %arg2 : f32
     linalg.yield %1 : f32
   } -> tensor<?xf32>
   return %0 : tensor<?xf32>
 }
 
 // -----
 
 // Test case: Different indexing maps mean that args are not redundant, despite
 // being the same Value.
 
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 #map1 = affine_map<(d0, d1) -> (d1, d0)>
 
 // CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1) -> (d1, d0)>
 // CHECK-LABEL: @distinct_affine_maps
 func @distinct_affine_maps(%arg0: tensor<?x?xf32>) -> tensor<?x?xf32> {
   // CHECK: linalg.generic{{.*}}[#[[$MAP0]], #[[$MAP1]], #[[$MAP0]]]
-  %0 = linalg.generic {indexing_maps = [#map0, #map1, #map0], iterator_types = ["parallel", "parallel"]} ins(%arg0, %arg0 : tensor<?x?xf32>, tensor<?x?xf32>) {
-  ^bb0(%arg1: f32, %arg2: f32):
+  %0 = linalg.generic {indexing_maps = [#map0, #map1, #map0], iterator_types = ["parallel", "parallel"]}
+     ins(%arg0, %arg0 : tensor<?x?xf32>, tensor<?x?xf32>)
+    outs(%arg0 : tensor<?x?xf32>) {
+  ^bb0(%arg1: f32, %arg2: f32, %arg3: f32):
     %1 = addf %arg1, %arg2 : f32
     linalg.yield %1 : f32
   } -> tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }
 
 // -----
 
 // Test case: Check rewriting mechanics for mixed redundant and
 // non-redundant args.
 
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 #map1 = affine_map<(d0, d1) -> (d1, d0)>
 
 // CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1) -> (d1, d0)>
 // CHECK-LABEL: @mixed_redundant_non_redundant
 func @mixed_redundant_non_redundant(%arg0: tensor<?x?xf32>) -> tensor<?x?xf32> {
   // CHECK: linalg.generic{{.*}}[#[[$MAP0]], #[[$MAP1]], #[[$MAP0]]]
-  // CHECK:   ^bb0(%[[BBARG0:.*]]: f32, %[[BBARG1:.*]]: f32):
+  // CHECK:   ^bb0(%[[BBARG0:.*]]: f32, %[[BBARG1:.*]]: f32, %{{[a-zA-Z0-9]+}}: f32):
   // CHECK:     "test.elementwise_mappable"(%[[BBARG0]], %[[BBARG1]], %[[BBARG0]])
-  %0 = linalg.generic {indexing_maps = [#map0, #map1, #map0, #map0], iterator_types = ["parallel", "parallel"]} ins(%arg0, %arg0, %arg0 : tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>) {
-  ^bb0(%arg1: f32, %arg2: f32, %arg3: f32):
+  %0 = linalg.generic {indexing_maps = [#map0, #map1, #map0, #map0], iterator_types = ["parallel", "parallel"]}
+     ins(%arg0, %arg0, %arg0 : tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>)
+    outs(%arg0 : tensor<?x?xf32>) {
+  ^bb0(%arg1: f32, %arg2: f32, %arg3: f32, %arg4: f32):
     %1 = "test.elementwise_mappable"(%arg1, %arg2, %arg3) : (f32, f32, f32) -> f32
     linalg.yield %1 : f32
   } -> tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }
 
 // -----
 
 // Test case: Check rewriting mechanics for multiple different redundant args.
 
 #map = affine_map<(d0) -> (d0)>
 
 // CHECK: #[[$MAP:.*]] = affine_map<(d0) -> (d0)>
 // CHECK-LABEL: @multiple_different_redundant_args
 func @multiple_different_redundant_args(%arg0: tensor<?xf32>, %arg1: tensor<?xf32>) -> tensor<?xf32> {
   // CHECK: linalg.generic{{.*}}[#[[$MAP]], #[[$MAP]], #[[$MAP]]]
-  // CHECK:   ^bb0(%[[BBARG0:.*]]: f32, %[[BBARG1:.*]]: f32):
+  // CHECK:   ^bb0(%[[BBARG0:.*]]: f32, %[[BBARG1:.*]]: f32, %{{[a-zA-Z0-9]+}}: f32):
   // CHECK:     "test.elementwise_mappable"(%[[BBARG0]], %[[BBARG1]], %[[BBARG0]], %[[BBARG1]])
-  %0 = linalg.generic {indexing_maps = [#map, #map, #map, #map, #map], iterator_types = ["parallel"]} ins(%arg0, %arg1, %arg0, %arg1 : tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>) {
-  ^bb0(%arg2: f32, %arg3: f32, %arg4: f32, %arg5: f32): 
+  %0 = linalg.generic {indexing_maps = [#map, #map, #map, #map, #map], iterator_types = ["parallel"]}
+     ins(%arg0, %arg1, %arg0, %arg1 : tensor<?xf32>, tensor<?xf32>, tensor<?xf32>, tensor<?xf32>)
+    outs(%arg0 : tensor<?xf32>) {
+  ^bb0(%arg2: f32, %arg3: f32, %arg4: f32, %arg5: f32, %arg6: f32):
     %1 = "test.elementwise_mappable"(%arg2, %arg3, %arg4, %arg5) : (f32, f32, f32, f32) -> f32
     linalg.yield %1 : f32
   } -> tensor<?xf32>
   return %0 : tensor<?xf32>
 }
 
 // -----
 
 // Test case: linalg.indexed_generic.
 // Other than the payload argument handling, everything else is the same.
 
 #map = affine_map<(d0) -> (d0)>
 
 // CHECK: #[[$MAP:.*]] = affine_map<(d0) -> (d0)>
 // CHECK-LABEL: @indexed_generic
 func @indexed_generic(%arg0: tensor<?xf32>) -> tensor<?xf32> {
   // CHECK: linalg.indexed_generic
-  // CHECK:   ^bb0(%{{.*}}: index, %[[BBARG:.*]]: f32):
+  // CHECK:   ^bb0(%{{.*}}: index, %[[BBARG:.*]]: f32, %{{[a-zA-Z0-9]+}}: f32):
   // CHECK:     addf %[[BBARG]], %[[BBARG]]
-  %0 = linalg.indexed_generic {indexing_maps = [#map, #map, #map], iterator_types = ["parallel"]} ins(%arg0, %arg0 : tensor<?xf32>, tensor<?xf32>) {
-  ^bb0(%index: index, %arg1: f32, %arg2: f32):
+  %0 = linalg.indexed_generic {indexing_maps = [#map, #map, #map], iterator_types = ["parallel"]}
+      ins(%arg0, %arg0 : tensor<?xf32>, tensor<?xf32>)
+     outs(%arg0 : tensor<?xf32>) {
+  ^bb0(%index: index, %arg1: f32, %arg2: f32, %arg3: f32):
     %1 = addf %arg1, %arg2 : f32
     linalg.yield %1 : f32
   } -> tensor<?xf32>
   return %0 : tensor<?xf32>
 }

mlir/test/Dialect/Linalg/canonicalize.mlir

 }
 // CHECK-LABEL: @no_fold_memref_reshape
 //       CHECK:   linalg.reshape
 //       CHECK:   linalg.reshape
 
 // -----
 
 #accesses = [
-  affine_map<(i) -> (i)>,
   affine_map<(i) -> (i)>
 ]
 
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"]
 }
 
 func @dce_zero_memref(%arg0 : memref<0xf32>, %arg1: tensor<0xf32>) -> tensor<0xf32> {
   // memref<0x32> is expected to be dce'ed
   linalg.copy(%arg0, %arg0): memref<0xf32>, memref<0xf32>
 
   // tensor<0xf32> cannot be dce'ed
-  %1 = linalg.generic #trait ins(%arg1 : tensor<0xf32>) {
+  %1 = linalg.generic #trait outs(%arg1 : tensor<0xf32>) {
   ^bb(%0: f32) :
     linalg.yield %0 : f32
   } -> tensor<0xf32>
 
   return %1: tensor<0xf32>
 }
 // CHECK-LABEL: @dce_zero_memref
 //   CHECK-NOT:   linalg.copy

 func @tensor.cast(%a : tensor<3x4xf32>, %b : tensor<4x?xf32>, %c : tensor<3x?xf32>)
   -> tensor<3x?xf32>
 {
   %ta = tensor.cast %a : tensor<3x4xf32> to tensor<?x?xf32>
   %tb = tensor.cast %b : tensor<4x?xf32> to tensor<?x?xf32>
   %tc = tensor.cast %c : tensor<3x?xf32> to tensor<?x?xf32>
 
   //      CHECK:  linalg.matmul ins({{.*}}tensor<3x4xf32>, tensor<4x?xf32>)
-  // CHECK-SAME:    init({{.*}}tensor<3x?xf32>) -> tensor<3x?xf32>
+  // CHECK-SAME:    outs({{.*}}tensor<3x?xf32>) -> tensor<3x?xf32>
   %0 = linalg.matmul ins(%ta, %tb: tensor<?x?xf32>, tensor<?x?xf32>)
-               init(%tc: tensor<?x?xf32>) -> tensor<?x?xf32>
+                    outs(%tc: tensor<?x?xf32>) -> tensor<?x?xf32>
 
   %1 = tensor.cast %0 : tensor<?x?xf32> to tensor<3x?xf32>
 
   return %1: tensor<3x?xf32>
 }
 
 // -----
 
 // CHECK-LABEL: func @linalg_effects(
 //  CHECK-SAME:     %[[A:[a-z0-9]*]]: tensor<?x?xf32>
 //  CHECK-SAME:     %[[B:[a-z0-9]*]]: memref<?x?xf32>
 //  CHECK-SAME:     %[[C:[a-z0-9]*]]: tensor<?x?xf32>
 func @linalg_effects(%a : tensor<?x?xf32>, %b : memref<?x?xf32>, %c : tensor<?x?xf32>) {
   // CHECK-NOT:   %{{.*}} = linalg.matmul
   %t = linalg.matmul ins(%a, %b : tensor<?x?xf32>, memref<?x?xf32>)
-                    init(%c : tensor<?x?xf32>) -> tensor<?x?xf32>
+                    outs(%c : tensor<?x?xf32>) -> tensor<?x?xf32>
 
   // CHECK-NOT:   %{{.*}} = linalg.matmul
   linalg.matmul ins(%a, %c : tensor<?x?xf32>, tensor<?x?xf32>)
                outs(%b : memref<?x?xf32>)
   return
 }
 // -----
 

mlir/test/Dialect/Linalg/convert-elementwise-to-linalg.mlir

 // RUN: mlir-opt -convert-elementwise-to-linalg -split-input-file %s | FileCheck %s
 
 // In-depth checking of the linalg.generic op for a very trivial case.
-// CHECK: #map = affine_map<() -> ()>
-// CHECK-LABEL:   func @addf_rank0
+// CHECK: #[[$MAP:.*]] = affine_map<() -> ()>
+// CHECK-LABEL: func @addf_rank0
+//  CHECK-SAME:   %[[ARG0:[0-9a-zA-Z]*]]: tensor<f32>
+//  CHECK-SAME:   %[[ARG1:[0-9a-zA-Z]*]]: tensor<f32>
 func @addf_rank0(%arg0: tensor<f32>, %arg1: tensor<f32>) -> tensor<f32> {
-  // CHECK: %{{.*}} = linalg.generic {indexing_maps = [#map, #map, #map], iterator_types = []} ins(%{{.*}}, %{{.*}} : tensor<f32>, tensor<f32>) {
-  // CHECK: ^bb0(%[[LHS:.*]]: f32, %[[RHS:.*]]: f32):
-  // CHECK:   %[[YIELD:.*]] = addf %[[LHS]], %[[RHS]] : f32
-  // CHECK:   linalg.yield %[[YIELD]] : f32
-  // CHECK: } -> tensor<f32>
+  //      CHECK: %{{.*}} = linalg.generic
+  // CHECK-SAME: indexing_maps = [#[[$MAP]], #[[$MAP]], #[[$MAP]]]
+  // CHECK-SAME: iterator_types = []
+  // CHECK-SAME:  ins(%[[ARG0]], %[[ARG1]]
+  // CHECK-SAME: outs(%[[ARG0]]
+  //      CHECK: ^bb0(%[[LHS:.*]]: f32, %[[RHS:.*]]: f32, %{{.*}}: f32):
+  //      CHECK:   %[[YIELD:.*]] = addf %[[LHS]], %[[RHS]] : f32
+  //      CHECK:   linalg.yield %[[YIELD]] : f32
+  //      CHECK: } -> tensor<f32>
   %0 = addf %arg0, %arg1 : tensor<f32>
   return %0 : tensor<f32>
 }
 
 // -----
 
 // Check indexing maps and iterator types for the rank > 0 case.
-// CHECK: #map = affine_map<(d0) -> (d0)>
 // CHECK-LABEL: func @addf_rank1
+//  CHECK-SAME:   %[[ARG0:[0-9a-zA-Z]*]]: tensor<?xf32>
+//  CHECK-SAME:   %[[ARG1:[0-9a-zA-Z]*]]: tensor<?xf32>
 func @addf_rank1(%arg0: tensor<?xf32>, %arg1: tensor<?xf32>) -> tensor<?xf32> {
-  // CHECK: linalg.generic{{.*}}indexing_maps = [#map, #map, #map], iterator_types = ["parallel"]
+  // CHECK: linalg.generic
+  // CHECK-SAME: iterator_types = ["parallel"]
+  // CHECK-SAME:  ins(%[[ARG0]], %[[ARG1]]
+  // CHECK-SAME: outs(%[[ARG0]]
   %0 = addf %arg0, %arg1 : tensor<?xf32>
   return %0 : tensor<?xf32>
 }
 
 // -----
 
 // Check a unary op.
 // CHECK-LABEL: func @exp
+//  CHECK-SAME:   %[[ARG0:[0-9a-zA-Z]*]]: tensor<f32>
 func @exp(%arg0: tensor<f32>) -> tensor<f32> {
   // CHECK: linalg.generic
-  // CHECK: ^bb0(%[[SCALAR:.*]]: f32):
+  // CHECK-SAME:  ins(%[[ARG0]]
+  // CHECK-SAME: outs(%[[ARG0]]
+  // CHECK: ^bb0(%[[SCALAR:.*]]: f32, %{{.*}}: f32):
   // CHECK:   %[[YIELD:.*]] = exp %[[SCALAR]] : f32
   // CHECK:   linalg.yield %[[YIELD]] : f32
   %0 = exp %arg0 : tensor<f32>
   return %0 : tensor<f32>
 }
 
 // -----
 
 // Check a case with varying operand types.
 // CHECK-LABEL: func @select
+//  CHECK-SAME:   %[[ARG0:[0-9a-zA-Z]*]]: tensor<i1>
+//  CHECK-SAME:   %[[ARG1:[0-9a-zA-Z]*]]: tensor<i32>
+//  CHECK-SAME:   %[[ARG2:[0-9a-zA-Z]*]]: tensor<i32>
 func @select(%arg0: tensor<i1>, %arg1: tensor<i32>, %arg2: tensor<i32>) -> tensor<i32> {
   // CHECK: linalg.generic
-  // CHECK: ^bb0(%[[PRED:.*]]: i1, %[[TRUE_VAL:.*]]: i32, %[[FALSE_VAL:.*]]: i32):
+  // CHECK-SAME:  ins(%[[ARG0]], %[[ARG1]], %[[ARG2]]
+  // CHECK-SAME: outs(%[[ARG1]]
+  // CHECK: ^bb0(%[[PRED:.*]]: i1, %[[TRUE_VAL:.*]]: i32, %[[FALSE_VAL:.*]]: i32, %{{.*}}: i32):
   // CHECK:   select %[[PRED]], %[[TRUE_VAL]], %[[FALSE_VAL]] : i32
   %0 = select %arg0, %arg1, %arg2 : tensor<i1>, tensor<i32>
   return %0 : tensor<i32>
 }
 
 // -----
 
 // Spot-check an op that requires copying attributes properly to the created scalar op.
+// Also checks proper init_tensor usage.
 // CHECK-LABEL: func @cmpf(
+//  CHECK-SAME:   %[[ARG0:[0-9a-zA-Z]*]]: tensor<f32>
+//  CHECK-SAME:   %[[ARG1:[0-9a-zA-Z]*]]: tensor<f32>
 func @cmpf(%arg0: tensor<f32>, %arg1: tensor<f32>) -> tensor<i1> {
+  // CHECK: %[[INIT:.*]] = linalg.init_tensor [] : tensor<i1>
+  // CHECK: linalg.generic
+  // CHECK-SAME:  ins(%[[ARG0]], %[[ARG1]]
+  // CHECK-SAME: outs(%[[INIT]]
+  // CHECK: ^bb0(%{{.*}}: f32, %{{.*}}: f32, %{{.*}}: i1):
   // CHECK: cmpf "olt", %{{.*}}, %{{.*}} : f32
   %0 = cmpf "olt", %arg0, %arg1 : tensor<f32>
   return %0 : tensor<i1>
 }
+
+// -----
+
+// Check proper init_tensor usage in a mixed case.
+// CHECK-LABEL: func @cmpf(
+//  CHECK-SAME:   %[[ARG0:[0-9a-zA-Z]*]]: tensor<4x?x?x8x2x?xf32>
+//  CHECK-SAME:   %[[ARG1:[0-9a-zA-Z]*]]: tensor<4x?x?x8x2x?xf32>
+func @cmpf(%arg0: tensor<4x?x?x8x2x?xf32>, %arg1: tensor<4x?x?x8x2x?xf32>) -> tensor<4x?x?x8x2x?xi1> {
+  // CHECK: %[[C1:.*]] = constant 1 : index
+  // CHECK: %[[D1:.*]] = dim %[[ARG0]], %[[C1]] : tensor<4x?x?x8x2x?xf32>
+  // CHECK: %[[C2:.*]] = constant 2 : index
+  // CHECK: %[[D2:.*]] = dim %[[ARG0]], %[[C2]] : tensor<4x?x?x8x2x?xf32>
+  // CHECK: %[[C5:.*]] = constant 5 : index
+  // CHECK: %[[D5:.*]] = dim %[[ARG0]], %[[C5]] : tensor<4x?x?x8x2x?xf32>
+  // CHECK: %[[INIT:.*]] = linalg.init_tensor [4, %[[D1]], %[[D2]], 8, 2, %[[D5]]] : tensor<4x?x?x8x2x?xi1>
+  // CHECK: linalg.generic
+  // CHECK-SAME:  ins(%[[ARG0]], %[[ARG1]]
+  // CHECK-SAME: outs(%[[INIT]]
+  // CHECK: ^bb0(%{{.*}}: f32, %{{.*}}: f32, %{{.*}}: i1):
+  // CHECK: cmpf "olt", %{{.*}}, %{{.*}} : f32
+  %0 = cmpf "olt", %arg0, %arg1 : tensor<4x?x?x8x2x?xf32>
+  return %0 : tensor<4x?x?x8x2x?xi1>
+}
+

mlir/test/Dialect/Linalg/drop-unit-extent-dims.mlir

-// RUN: mlir-opt %s -linalg-fold-unit-extent-dims -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -split-input-file -linalg-fold-unit-extent-dims | FileCheck %s
 
 #accesses = [
   affine_map<(i, j, k, l, m) -> (i, k, m)>,
   affine_map<(i, j, k, l, m) -> (i, k, j, l, m)>
 ]
 
 #trait = {
   iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel"],
   indexing_maps = #accesses,
   library_call = "some_external_func"
 }
 
-func @drop_one_trip_loops(%arg0 : tensor<?x1x?xf32>) -> tensor<?x1x?x1x?xf32>
-{
+func @drop_one_trip_loops(%arg0 : tensor<?x1x?xf32>, %shape: tensor<?x1x?x1x?xf32>) -> tensor<?x1x?x1x?xf32> {
   %0 = linalg.generic #trait
-    ins(%arg0 : tensor<?x1x?xf32>) {
-       ^bb0(%arg1 : f32) :
-         linalg.yield %arg1 : f32
+     ins(%arg0 : tensor<?x1x?xf32>)
+    outs(%shape : tensor<?x1x?x1x?xf32>) {
+       ^bb0(%arg2 : f32, %arg3 : f32) :
+         linalg.yield %arg2 : f32
        } -> tensor<?x1x?x1x?xf32>
   return %0 : tensor<?x1x?x1x?xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2) -> (d0, d1)>
 //   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2) -> (d2)>
 //   CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0, d1, d2) -> (d0, d2)>
 //   CHECK-DAG: #[[$MAP3:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 //   CHECK-DAG: #[[$MAP4:.*]] = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1)>

 
 #trait = {
   iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel"],
   indexing_maps = #accesses,
   library_call = "some_external_func"
 }
 
 func @drop_one_trip_loops_indexed_generic
-  (%arg0 : tensor<?x1x?xi32>) -> tensor<?x1x?x1x?xi32>
+  (%arg0 : tensor<?x1x?xi32>, %shape: tensor<?x1x?x1x?xi32>) -> tensor<?x1x?x1x?xi32>
 {
   %0 = linalg.indexed_generic #trait
-    ins(%arg0 : tensor<?x1x?xi32>) {
+     ins(%arg0 : tensor<?x1x?xi32>)
+    outs(%shape: tensor<?x1x?x1x?xi32>) {
        ^bb0(%arg1 : index, %arg2 : index, %arg3 : index, %arg4 : index,
-            %arg5 : index, %arg6 : i32) :
+            %arg5 : index, %arg6 : i32, %arg7 : i32) :
    %1 = addi %arg1, %arg2 : index
    %2 = addi %1, %arg3 : index
    %3 = addi %2, %arg4 : index
    %4 = addi %3, %arg5 : index
    %5 = index_cast %4 : index to i32
    %6 = addi %5, %arg6 : i32
          linalg.yield %6 : i32
        } -> tensor<?x1x?x1x?xi32>
   return %0 : tensor<?x1x?x1x?xi32>
 }
 // CHECK-LABEL: func @drop_one_trip_loops_indexed_generic
 //       CHECK:   linalg.indexed_generic
 //       CHECK:   ^{{.+}}(
 //  CHECK-SAME:     %[[ARG1:[a-zA-Z0-9]+]]: index, %[[ARG2:[a-zA-Z0-9]+]]: index
-//  CHECK-SAME:     %[[ARG3:[a-zA-Z0-9]+]]: index, %[[ARG4:[a-zA-Z0-9]+]]: i32)
+//  CHECK-SAME:     %[[ARG3:[a-zA-Z0-9]+]]: index, %[[ARG4:[a-zA-Z0-9]+]]: i32, %{{.*}}: i32)
 //       CHECK:     %[[T3:.+]] = addi %[[ARG1]], %[[ARG2]]
 //       CHECK:     %[[T4:.+]] = addi %[[T3]], %[[ARG3]]
 //       CHECK:     %[[T5:.+]] = index_cast %[[T4]] : index to i32
 //       CHECK:     %[[T6:.+]] = addi %[[T5]], %[[ARG4]] : i32
 //       CHECK:     linalg.yield %[[T6]] : i32
 
 // -----
 
 #map0 = affine_map<(i, j) -> (i, j)>
 #access = [#map0, #map0]
 #trait = {
   iterator_types = ["parallel", "parallel"],
   indexing_maps = #access,
   library_call = "some_external_func"
 }
 
 func @drop_all_loops(%arg0 : tensor<1x1xf32>) -> tensor<1x1xf32>
 {
   %0 = linalg.generic #trait
-    ins(%arg0 : tensor<1x1xf32>) {
-       ^bb0(%arg1: f32) :
+     ins(%arg0 : tensor<1x1xf32>)
+    outs(%arg0 : tensor<1x1xf32>) {
+       ^bb0(%arg1: f32, %arg2: f32) :
          linalg.yield %arg1 : f32
        } -> tensor<1x1xf32>
   return %0 : tensor<1x1xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<() -> ()>
 // CHECK-LABEL: func @drop_all_loops
 //       CHECK:   linalg.tensor_reshape %{{.*}} []
 //       CHECK:   linalg.generic
 //  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP0]]]
 //  CHECK-SAME:     iterator_types = []
 
 // -----
 
 #map0 = affine_map<(i, j) -> (i, j)>
 #access = [#map0, #map0]
 #trait = {
   iterator_types = ["parallel", "parallel"],
   indexing_maps = #access,
   library_call = "some_external_func"
 }
 
 func @drop_all_loops_indexed_generic
-  (%arg0 : tensor<1x1xi32>) -> tensor<1x1xi32>
-{
+  (%arg0 : tensor<1x1xi32>) -> tensor<1x1xi32>{
   %0 = linalg.indexed_generic #trait
-    ins(%arg0 : tensor<1x1xi32>) {
-       ^bb0(%arg1 : index, %arg2 : index, %arg3: i32) :
+     ins(%arg0 : tensor<1x1xi32>)
+    outs(%arg0 : tensor<1x1xi32>) {
+       ^bb0(%arg1 : index, %arg2 : index, %arg3: i32, %arg4: i32) :
          %1 = addi %arg1, %arg2 : index
    %2 = index_cast %1 : index to i32
    %3 = addi %2, %arg3 : i32
          linalg.yield %3 : i32
        } -> tensor<1x1xi32>
   return %0 : tensor<1x1xi32>
 }
 
 // CHECK-LABEL: func @drop_all_loops_indexed_generic
 //       CHECK:   linalg.indexed_generic
-//       CHECK:   ^{{.+}}(%[[ARG1:.+]]: i32)
+//       CHECK:   ^{{.+}}(%[[ARG1:.+]]: i32, %[[ARG2:.+]]: i32)
 //       CHECK:     linalg.yield %[[ARG1]] : i32
 
 // -----
 
 #accesses = [
   affine_map<(d0) -> (0, d0)>,
   affine_map<(d0) -> (d0)>
 ]
 
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel"],
   library_call = "some_external_fn"
 }
 
-func @leading_dim_1_canonicalization(%arg0: tensor<1x5xf32>) -> tensor<5xf32> {
+func @leading_dim_1_canonicalization(%arg0: tensor<1x5xf32>, %shape: tensor<5xf32>) -> tensor<5xf32> {
   %0 = linalg.generic #trait
-    ins(%arg0 : tensor<1x5xf32>) {
-  ^bb0(%arg2: f32):     // no predecessors
+     ins(%arg0 : tensor<1x5xf32>)
+    outs(%shape : tensor<5xf32>) {
+  ^bb0(%arg2: f32, %arg3: f32):     // no predecessors
     linalg.yield %arg2 : f32
   } -> tensor<5xf32>
   return %0 : tensor<5xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-LABEL: func @leading_dim_1_canonicalization
 //       CHECK:   linalg.tensor_reshape %{{.*}} [#[[$MAP0]]]
 //       CHECK:   linalg.generic

 ]
 
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel", "parallel"],
   library_call = "some_external_fn"
 }
 
-func @broadcast_test(%arg0 : tensor<5xf32>, %arg1 : tensor<5xf32>) -> tensor<5x5xf32>
+func @broadcast_test(%arg0 : tensor<5xf32>, %arg1 : tensor<5xf32>, %shape : tensor<5x5xf32>) -> tensor<5x5xf32>
 {
   %0 = linalg.tensor_reshape %arg0 [affine_map<(d0, d1) -> (d0, d1)>] :
        tensor<5xf32> into tensor<1x5xf32>
   %1 = linalg.tensor_reshape %arg1 [affine_map<(d0, d1) -> (d0, d1)>] :
        tensor<5xf32> into tensor<5x1xf32>
   %2 = linalg.generic #trait
-    ins(%0, %1 : tensor<1x5xf32>, tensor<5x1xf32>) {
-       ^bb0(%arg2: f32, %arg3: f32):
-         %3 = addf %arg2, %arg3 : f32
+     ins(%0, %1 : tensor<1x5xf32>, tensor<5x1xf32>)
+    outs(%shape : tensor<5x5xf32>) {
+       ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):
+         %3 = addf %arg3, %arg4 : f32
          linalg.yield %3 : f32
        } -> tensor<5x5xf32>
   return %2 : tensor<5x5xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d1)>
 //   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1) -> (d0)>
 //   CHECK-DAG: #[[$MAP2:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-LABEL: func @broadcast_test

 ]
 
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel", "parallel"],
   library_call = "some_external_fn"
 }
 
-func @broadcast_scalar(%arg0 : tensor<1x1xf32>) -> tensor<?x?xf32>
+func @broadcast_scalar(%arg0 : tensor<1x1xf32>, %shape : tensor<?x?xf32>) -> tensor<?x?xf32>
 {
    %0 = linalg.generic #trait
-    ins(%arg0 : tensor<1x1xf32>) {
-      ^bb0(%arg1 : f32):
-        linalg.yield %arg1 : f32
+     ins(%arg0 : tensor<1x1xf32>)
+    outs(%shape : tensor<?x?xf32>) {
+      ^bb0(%arg2 : f32, %arg3 : f32):
+        linalg.yield %arg2 : f32
    } -> tensor<?x?xf32>
    return %0 : tensor<?x?xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> ()>
 //   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-LABEL: func @broadcast_scalar
 //  CHECK-SAME:   %[[ARG0:.*]]: tensor<1x1xf32>
 //       CHECK:   %[[A:.*]] = linalg.tensor_reshape %[[ARG0]] []

mlir/test/Dialect/Linalg/fold-unit-trip-loops.mlir

-// RUN: mlir-opt %s -linalg-fold-unit-extent-dims="fold-one-trip-loops-only" -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -split-input-file -linalg-fold-unit-extent-dims="fold-one-trip-loops-only" | FileCheck %s
 
 #accesses = [
   affine_map<(i, j, k, l, m) -> (i, k, m)>,
   affine_map<(i, j, k, l, m) -> (i, k, j, l, m)>
 ]
 
 #trait = {
   iterator_types = ["parallel", "parallel", "parallel", "parallel", "parallel"],
   indexing_maps = #accesses,
   library_call = "some_external_func"
 }
 
-func @drop_one_trip_loops(%arg0 : tensor<?x1x?xf32>) -> tensor<?x1x?x1x?xf32>
+func @drop_one_trip_loops(%arg0 : tensor<?x1x?xf32>, %shape: tensor<?x1x?x1x?xf32>) -> tensor<?x1x?x1x?xf32>
 {
   %0 = linalg.generic #trait
-    ins(%arg0 : tensor<?x1x?xf32>) {
-       ^bb0(%arg1 : f32) :
+    ins(%arg0 : tensor<?x1x?xf32>)
+    outs(%shape : tensor<?x1x?x1x?xf32>) {
+       ^bb0(%arg1 : f32, %arg2 : f32) :
          linalg.yield %arg1 : f32
        } -> tensor<?x1x?x1x?xf32>
   return %0 : tensor<?x1x?x1x?xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2) -> (d0, 0, d2)>
 //   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2) -> (d0, 0, d1, 0, d2)>
 // CHECK-LABEL: func @drop_one_trip_loops
 //       CHECK:   linalg.generic
 //  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
 //  CHECK-SAME:     iterator_types = ["parallel", "parallel", "parallel"]
 
 // -----
 
 #map0 = affine_map<(i, j) -> (i, j)>
 #access = [#map0, #map0]
 #trait = {
   iterator_types = ["parallel", "parallel"],
   indexing_maps = #access,
   library_call = "some_external_func"
 }
 
 func @drop_all_loops(%arg0 : tensor<1x1xf32>) -> tensor<1x1xf32>
 {
   %0 = linalg.generic #trait
-    ins(%arg0 : tensor<1x1xf32>) {
-       ^bb0(%arg1: f32) :
+     ins(%arg0 : tensor<1x1xf32>)
+    outs(%arg0 : tensor<1x1xf32>) {
+       ^bb0(%arg1: f32, %arg2: f32) :
          linalg.yield %arg1 : f32
        } -> tensor<1x1xf32>
   return %0 : tensor<1x1xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<() -> (0, 0)>
 // CHECK-LABEL: func @drop_all_loops
 //       CHECK:   linalg.generic
 //  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP0]]]

 ]
 
 #trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel", "parallel"],
   library_call = "some_external_fn"
 }
 
-func @leading_dim_1_canonicalization(%arg0: tensor<1x5xf32>) -> tensor<5xf32> {
+func @leading_dim_1_canonicalization(%arg0: tensor<1x5xf32>, %shape: tensor<5xf32>) -> tensor<5xf32> {
   %0 = linalg.generic #trait
-      ins(%arg0 : tensor<1x5xf32>) {
-    ^bb0(%arg2: f32):     // no predecessors
+       ins(%arg0 : tensor<1x5xf32>)
+      outs(%shape : tensor<5xf32>) {
+    ^bb0(%arg2: f32, %arg3: f32):     // no predecessors
       linalg.yield %arg2 : f32
   } -> tensor<5xf32>
   return %0 : tensor<5xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0) -> (0, d0)>
 //   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0) -> (d0)>
 // CHECK-LABEL: func @leading_dim_1_canonicalization
 //       CHECK:   linalg.generic
 //  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
 //  CHECK-SAME:     iterator_types = ["parallel"]

mlir/test/Dialect/Linalg/fusion-tensor.mlir

 // RUN: mlir-opt %s -linalg-fusion-for-tensor-ops -split-input-file | FileCheck %s
 
 // CHECK-DAG: [[$MAP0:#[a-zA-Z0-9_]*]] = affine_map<(d0, d1) -> (d0, d1)>
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 
 // CHECK-LABEL: @add_mul_fusion
 func @add_mul_fusion(%arg0: tensor<?x?xf32>, %arg1 : tensor<?x?xf32>, %arg2 : tensor<?x?xf32>) -> tensor<?x?xf32>
 {
-  %0 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = ["parallel", "parallel"]}
-      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
-      %1 = addf %arg3, %arg4 : f32
-      linalg.yield %1 : f32
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %0 = dim %arg0, %c0 : tensor<?x?xf32>
+  %1 = dim %arg0, %c1 : tensor<?x?xf32>
+  %2 = linalg.init_tensor [%0, %1] : tensor<?x?xf32>
+  %3 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = ["parallel", "parallel"]}
+      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%2 : tensor<?x?xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
+      %4 = addf %arg3, %arg4 : f32
+      linalg.yield %4 : f32
   } -> tensor<?x?xf32>
   // CHECK: linalg.generic {
   // CHECK-SAME: indexing_maps = {{\[}}[[$MAP0]], [[$MAP0]], [[$MAP0]], [[$MAP0]]{{\]}}
-  %2 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = ["parallel", "parallel"]}
-      ins(%0, %arg2 : tensor<?x?xf32>, tensor<?x?xf32>) {
+  %4 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = ["parallel", "parallel"]}
+      ins(%3, %arg2 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%2 : tensor<?x?xf32>) {
     // CHECK: ^{{[a-zA-Z0-9_]*}}
     // CHECK-SAME: [[ARG0:%[a-zA-Z0-9_]*]]
     // CHECK-SAME: [[ARG1:%[a-zA-Z0-9_]*]]
     // CHECK-SAME: [[ARG2:%[a-zA-Z0-9_]*]]
-    ^bb0(%arg5: f32, %arg6: f32):       // no predecessors
+    ^bb0(%arg5: f32, %arg6: f32, %arg7: f32):       // no predecessors
       // CHECK: [[T1:%[a-zA-Z0-9_]*]] = addf [[ARG0]], [[ARG1]]
       // CHECK-NOT: linalg.yield
       // CHECK: mulf [[T1]], [[ARG2]]
       // CHECK: linalg.yield
-      %3 = mulf %arg5, %arg6 : f32
-      linalg.yield %3 : f32
+      %5 = mulf %arg5, %arg6 : f32
+      linalg.yield %5 : f32
     } -> tensor<?x?xf32>
-  return %2 : tensor<?x?xf32>
+  return %4 : tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK-DAG: [[$MAP0:#[a-zA-Z0-9_]*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-DAG: [[$MAP1:#[a-zA-Z0-9_]*]] = affine_map<(d0, d1) -> (d1, d0)>
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 #map1 = affine_map<(d0, d1) -> (d1, d0)>
 
 // CHECK-LABEL: @transpose_add_mul_fusion
 func @transpose_add_mul_fusion(%arg0: tensor<?x?xf32>, %arg1 : tensor<?x?xf32>, %arg2 : tensor<?x?xf32>) -> tensor<?x?xf32>
 {
-  %0 = linalg.generic {indexing_maps = [#map0, #map1, #map0], iterator_types = ["parallel", "parallel"]}
-      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
-      %1 = addf %arg3, %arg4 : f32
-      linalg.yield %1 : f32
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %0 = dim %arg0, %c0 : tensor<?x?xf32>
+  %1 = dim %arg0, %c1 : tensor<?x?xf32>
+  %2 = linalg.init_tensor [%0, %1] : tensor<?x?xf32>
+  %3 = linalg.generic {indexing_maps = [#map0, #map1, #map0], iterator_types = ["parallel", "parallel"]}
+      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%2 : tensor<?x?xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
+      %4 = addf %arg3, %arg4 : f32
+      linalg.yield %4 : f32
   } -> tensor<?x?xf32>
   // CHECK: linalg.generic {
   // CHECK-SAME: indexing_maps = {{\[}}[[$MAP0]], [[$MAP1]], [[$MAP0]], [[$MAP0]]{{\]}}
-  %2 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = ["parallel", "parallel"]}
-      ins(%0, %arg2 : tensor<?x?xf32>, tensor<?x?xf32>) {
-    ^bb0(%arg5: f32, %arg6: f32):       // no predecessors
-      %3 = mulf %arg5, %arg6 : f32
-      linalg.yield %3 : f32
+  %4 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = ["parallel", "parallel"]}
+      ins(%3, %arg2 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%2 : tensor<?x?xf32>) {
+    ^bb0(%arg5: f32, %arg6: f32, %arg7: f32):       // no predecessors
+      %5 = mulf %arg5, %arg6 : f32
+      linalg.yield %5 : f32
     } -> tensor<?x?xf32>
-  return %2 : tensor<?x?xf32>
+  return %4 : tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK-DAG: [[$MAP0:#[a-zA-Z0-9_]*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-DAG: [[$MAP1:#[a-zA-Z0-9_]*]] = affine_map<(d0, d1) -> (d1, d0)>
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 #map1 = affine_map<(d0, d1) -> (d1, d0)>
 
 // CHECK-LABEL: @add_transpose_mul_fusion
 func @add_transpose_mul_fusion(%arg0: tensor<?x?xf32>, %arg1 : tensor<?x?xf32>, %arg2 : tensor<?x?xf32>) -> tensor<?x?xf32>
 {
-  %0 = linalg.generic {indexing_maps = [#map0, #map1, #map0], iterator_types = ["parallel", "parallel"]}
-      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
-      %1 = addf %arg3, %arg4 : f32
-      linalg.yield %1 : f32
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %0 = dim %arg0, %c0 : tensor<?x?xf32>
+  %1 = dim %arg0, %c1 : tensor<?x?xf32>
+  %2 = linalg.init_tensor [%0, %1] : tensor<?x?xf32>
+  %3 = linalg.generic {indexing_maps = [#map0, #map1, #map0], iterator_types = ["parallel", "parallel"]}
+      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%2 : tensor<?x?xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
+      %4 = addf %arg3, %arg4 : f32
+      linalg.yield %4 : f32
   } -> tensor<?x?xf32>
   // CHECK: linalg.generic {
   // CHECK-SAME: indexing_maps = {{\[}}[[$MAP1]], [[$MAP0]], [[$MAP0]], [[$MAP0]]{{\]}}
-  %2 = linalg.generic {indexing_maps = [#map1, #map0, #map0], iterator_types = ["parallel", "parallel"]}
-      ins(%0, %arg2 : tensor<?x?xf32>, tensor<?x?xf32>) {
-    ^bb0(%arg5: f32, %arg6: f32):       // no predecessors
-      %3 = mulf %arg5, %arg6 : f32
-      linalg.yield %3 : f32
+  %4 = linalg.generic {indexing_maps = [#map1, #map0, #map0], iterator_types = ["parallel", "parallel"]}
+      ins(%3, %arg2 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%2 : tensor<?x?xf32>){
+    ^bb0(%arg5: f32, %arg6: f32, %arg7: f32):       // no predecessors
+      %5= mulf %arg5, %arg6 : f32
+      linalg.yield %5 : f32
     } -> tensor<?x?xf32>
-  return %2 : tensor<?x?xf32>
+  return %4 : tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK-DAG: [[$MAP0:#[a-zA-Z0-9_]*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-DAG: [[$MAP1:#[a-zA-Z0-9_]*]] = affine_map<(d0, d1) -> (d0)>
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 #map1 = affine_map<(d0, d1) -> (d0)>
 #map2 = affine_map<(d0) -> (d0)>
 
 // CHECK-LABEL: @add_broadcast_mul_fusion
 func @add_broadcast_mul_fusion(%arg0: tensor<?xf32>, %arg1 : tensor<?xf32>, %arg2 : tensor<?x?xf32>) -> tensor<?x?xf32>
 {
-  %0 = linalg.generic {indexing_maps = [#map2, #map2, #map2], iterator_types = ["parallel"]}
-      ins(%arg0, %arg1 : tensor<?xf32>, tensor<?xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
-      %1 = addf %arg3, %arg4 : f32
-      linalg.yield %1 : f32
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %0 = dim %arg0, %c0 : tensor<?xf32>
+  %1 = linalg.init_tensor [%0] : tensor<?xf32>
+  %2 = linalg.generic {indexing_maps = [#map2, #map2, #map2], iterator_types = ["parallel"]}
+      ins(%arg0, %arg1 : tensor<?xf32>, tensor<?xf32>)
+      outs(%1 : tensor<?xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
+      %3 = addf %arg3, %arg4 : f32
+      linalg.yield %3 : f32
   } -> tensor<?xf32>
   // CHECK: linalg.generic {
   // CHECK-SAME: indexing_maps = {{\[}}[[$MAP1]], [[$MAP1]], [[$MAP0]], [[$MAP0]]
-  %2 = linalg.generic {indexing_maps = [#map1, #map0, #map0], iterator_types = ["parallel", "parallel"]}
-      ins(%0, %arg2 : tensor<?xf32>, tensor<?x?xf32>) {
-    ^bb0(%arg5: f32, %arg6: f32):       // no predecessors
-      %3 = mulf %arg5, %arg6 : f32
-      linalg.yield %3 : f32
+  %3 = dim %arg2, %c1 : tensor<?x?xf32>
+  %4 = linalg.init_tensor [%0, %3] : tensor<?x?xf32>
+  %5 = linalg.generic {indexing_maps = [#map1, #map0, #map0], iterator_types = ["parallel", "parallel"]}
+      ins(%2, %arg2 : tensor<?xf32>, tensor<?x?xf32>)
+      outs(%4 : tensor<?x?xf32>){
+    ^bb0(%arg5: f32, %arg6: f32, %arg7: f32):       // no predecessors
+      %6 = mulf %arg5, %arg6 : f32
+      linalg.yield %6 : f32
     } -> tensor<?x?xf32>
-  return %2 : tensor<?x?xf32>
+  return %5 : tensor<?x?xf32>
 }
 
 // -----
 
 // CHECK: #[[$MAP0:.*]] = affine_map<() -> ()>
 #map0 = affine_map<() -> ()>
 
 // CHECK-LABEL: @add_mul_scalar_fusion
 func @add_mul_scalar_fusion(%arg0: tensor<f32>, %arg1: tensor<f32>, %arg2: tensor<f32>) -> tensor<f32>
 {
-  %0 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = []}
-      ins(%arg0, %arg1 : tensor<f32>, tensor<f32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
-      %1 = addf %arg3, %arg4 : f32
-      linalg.yield %1 : f32
+  %0 = linalg.init_tensor [] : tensor<f32>
+  %1 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = []}
+      ins(%arg0, %arg1 : tensor<f32>, tensor<f32>)
+      outs(%0 : tensor<f32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
+      %2 = addf %arg3, %arg4 : f32
+      linalg.yield %2 : f32
   } -> tensor<f32>
   // CHECK: linalg.generic {
   // CHECK: addf
   // CHECK: mulf
-  %1 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = []}
-      ins(%0, %arg2 : tensor<f32>, tensor<f32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
-      %1 = mulf %arg3, %arg4 : f32
-      linalg.yield %1 : f32
+  %2 = linalg.generic {indexing_maps = [#map0, #map0, #map0], iterator_types = []}
+      ins(%1, %arg2 : tensor<f32>, tensor<f32>)
+      outs(%0 : tensor<f32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
+      %3 = mulf %arg3, %arg4 : f32
+      linalg.yield %3 : f32
   } -> tensor<f32>
 
-  return %1 : tensor<f32>
+  return %2 : tensor<f32>
 }
 
 // -----
 
 #map0 = affine_map<(d0, d1, d2) -> (d0)>
 #map1 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 func @generic_op_constant_fusion(%arg0 : tensor<5x?x?xf32>) -> tensor<5x?x?xf32>
 {
-  %0 = constant dense<42.0> : tensor<5xf32>
-  %1 = linalg.generic {
-         indexing_maps = [#map0, #map1, #map1],
-         iterator_types = ["parallel", "parallel", "parallel"]}
-         ins(%0, %arg0 : tensor<5xf32>, tensor<5x?x?xf32>) {
-       ^bb0(%arg1: f32, %arg2: f32):
-         %2 = mulf %arg1, %arg2 : f32
-         linalg.yield %2 : f32
-       } -> tensor<5x?x?xf32>
-  return %1 : tensor<5x?x?xf32>
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %c2 = constant 2 : index
+  %cst = constant dense<42.0> : tensor<5xf32>
+  %0 = dim %arg0, %c1 : tensor<5x?x?xf32>
+  %1 = dim %arg0, %c2 : tensor<5x?x?xf32>
+  %2 = linalg.init_tensor [5, %0, %1] : tensor<5x?x?xf32>
+  %3 = linalg.generic {
+    ins(%cst, %arg0 : tensor<5xf32>, tensor<5x?x?xf32>)
+    outs(%2 : tensor<5x?x?xf32>) {
+    ^bb0(%arg1: f32, %arg2: f32, %arg3: f32):
+      %4 = mulf %arg1, %arg2 : f32
+      linalg.yield %4 : f32
+  return %3 : tensor<5x?x?xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 // CHECK-LABEL: func @generic_op_constant_fusion
 //       CHECK:   %[[CST:.*]] = constant {{.*}} : f32
 //       CHECK:   linalg.generic
-//       CHECK:   ^{{.*}}(%[[ARG1:.*]]: f32)
+//       CHECK:   ^{{.+}}(%[[ARG1:[a-zA-Z0-9_]+]]: f32, %{{.+}}: f32):
 //       CHECK:     mulf %[[CST]], %[[ARG1]]
 
 // -----
 
 #map0 = affine_map<(d0, d1, d2) -> (d0)>
 #map1 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 func @indexed_generic_op_constant_fusion(%arg0 : tensor<5x?x?xf32>)
                                          -> tensor<5x?x?xf32>
 {
-  %0 = constant dense<42.0> : tensor<5xf32>
-  %1 = linalg.indexed_generic {
-         indexing_maps = [#map0, #map1, #map1],
-         iterator_types = ["parallel", "parallel", "parallel"]}
-         ins(%0, %arg0 : tensor<5xf32>, tensor<5x?x?xf32>) {
-       ^bb0(%arg1: index, %arg2: index, %arg3: index, %arg4: f32, %arg5 : f32):
-         %2 = mulf %arg4, %arg5 : f32
-         linalg.yield %2 : f32
-       } -> tensor<5x?x?xf32>
-  return %1 : tensor<5x?x?xf32>
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %c2 = constant 2 : index
+  %cst = constant dense<42.0> : tensor<5xf32>
+  %0 = dim %arg0, %c1 : tensor<5x?x?xf32>
+  %1 = dim %arg0, %c2 : tensor<5x?x?xf32>
+  %2 = linalg.init_tensor [5, %0, %1] : tensor<5x?x?xf32>
+  %3 = linalg.indexed_generic {
+    ins(%cst, %arg0 : tensor<5xf32>, tensor<5x?x?xf32>)
+    outs(%2 : tensor<5x?x?xf32>) {
+    ^bb0(%arg1: index, %arg2: index, %arg3: index, %arg4: f32, %arg5 : f32, %arg6 : f32):
+      %4 = mulf %arg4, %arg5 : f32
+      linalg.yield %4 : f32
+  return %3 : tensor<5x?x?xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 // CHECK-LABEL: func @indexed_generic_op_constant_fusion
 //       CHECK:   %[[CST:.*]] = constant {{.*}} : f32
 //       CHECK:   linalg.indexed_generic
 //       CHECK:   ^{{[a-zA-Z0-9_]*}}
 //  CHECK-SAME:     %[[ARG1:[a-zA-Z0-9]*]]: index
 //  CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]*]]: index
 //  CHECK-SAME:     %[[ARG3:[a-zA-Z0-9]*]]: index
-//  CHECK-SAME:     %[[ARG4:.*]]: f32)
+//  CHECK-SAME:     %[[ARG4:[a-zA-Z0-9_]*]]: f32, %{{.*}}: f32)
 //       CHECK:     mulf %[[CST]], %[[ARG4]]
 
 // -----
 
 #map0 = affine_map<(d0, d1, d2) -> ()>
 #map1 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 func @generic_op_zero_dim_constant_fusion(%arg0 : tensor<5x?x?xf32>)
   -> tensor<5x?x?xf32>
 {
-  %0 = constant dense<42.0> : tensor<f32>
-  %1 = linalg.generic {
-         indexing_maps = [#map0, #map1, #map1],
-         iterator_types = ["parallel", "parallel", "parallel"]}
-         ins(%0, %arg0 : tensor<f32>, tensor<5x?x?xf32>) {
-       ^bb0(%arg1: f32, %arg2: f32):
-         %2 = mulf %arg1, %arg2 : f32
-         linalg.yield %2 : f32
-       } -> tensor<5x?x?xf32>
-  return %1 : tensor<5x?x?xf32>
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %c2 = constant 2 : index
+  %cst = constant dense<42.0> : tensor<f32>
+  %0 = dim %arg0, %c1 : tensor<5x?x?xf32>
+  %1 = dim %arg0, %c2 : tensor<5x?x?xf32>
+  %2 = linalg.init_tensor [5, %0, %1] : tensor<5x?x?xf32>
+  %3 = linalg.generic {
+    ins(%cst, %arg0 : tensor<f32>, tensor<5x?x?xf32>)
+    outs(%2 : tensor<5x?x?xf32>) {
+    ^bb0(%arg1: f32, %arg2: f32, %arg3: f32):
+      %4 = mulf %arg1, %arg2 : f32
+      linalg.yield %4 : f32
+  return %3 : tensor<5x?x?xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 // CHECK-LABEL: func @generic_op_zero_dim_constant_fusion
 //       CHECK:   %[[CST:.*]] = constant {{.*}} : f32
 //       CHECK:   linalg.generic
-//       CHECK:   ^{{.*}}(%[[ARG1:.*]]: f32)
+//       CHECK:   ^{{.*}}(%[[ARG1:[a-zA-Z0-9_]*]]: f32, %{{.*}}: f32)
 //       CHECK:     mulf %[[CST]], %[[ARG1]]
 
 // -----
 
 #map0 = affine_map<(d0, d1, d2) -> ()>
 #map1 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 func @indexed_generic_op_zero_dim_constant_fusion
   (%arg0 : tensor<5x?x?xf32>) -> tensor<5x?x?xf32>
 {
-  %0 = constant dense<42.0> : tensor<f32>
-  %1 = linalg.indexed_generic {
-         indexing_maps = [#map0, #map1, #map1],
-         iterator_types = ["parallel", "parallel", "parallel"]}
-         ins(%0, %arg0 : tensor<f32>, tensor<5x?x?xf32>) {
-       ^bb0(%arg1 : index, %arg2 : index, %arg3 : index, %arg4: f32, %arg5: f32):
-         %2 = mulf %arg4, %arg5 : f32
-         linalg.yield %2 : f32
-       } -> tensor<5x?x?xf32>
-  return %1 : tensor<5x?x?xf32>
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %c2 = constant 2 : index
+  %cst = constant dense<42.0> : tensor<f32>
+  %0 = dim %arg0, %c1 : tensor<5x?x?xf32>
+  %1 = dim %arg0, %c2 : tensor<5x?x?xf32>
+  %2 = linalg.init_tensor [5, %0, %1] : tensor<5x?x?xf32>
+  %3 = linalg.indexed_generic {
+    ins(%cst, %arg0 : tensor<f32>, tensor<5x?x?xf32>)
+    outs(%2 : tensor<5x?x?xf32>) {
+    ^bb0(%arg1 : index, %arg2 : index, %arg3 : index, %arg4: f32, %arg5: f32, %arg6: f32):
+      %4 = mulf %arg4, %arg5 : f32
+      linalg.yield %4 : f32
+  return %3 : tensor<5x?x?xf32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 // CHECK-LABEL: func @indexed_generic_op_zero_dim_constant_fusion
 //       CHECK:   %[[CST:.*]] = constant {{.*}} : f32
 //       CHECK:   linalg.indexed_generic
 //       CHECK:   ^{{[a-zA-Z0-9_]*}}
 //  CHECK-SAME:     %[[ARG1:[a-zA-Z0-9]*]]: index
 //  CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]*]]: index
 //  CHECK-SAME:     %[[ARG3:[a-zA-Z0-9]*]]: index
-//  CHECK-SAME:     %[[ARG4:.*]]: f32)
+//  CHECK-SAME:     %[[ARG4:[a-zA-Z0-9_]*]]: f32, %{{.*}}: f32)
 //       CHECK:     mulf %[[CST]], %[[ARG4]]
 
 // -----
 
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 func @generic_op_indexed_generic_op_fusion(%arg0: tensor<?x?xi32>,
                                            %arg1: tensor<?x?xi32>) -> tensor<?x?xi32> {
-    %0 = linalg.generic {
-      indexing_maps = [#map0, #map0, #map0],
-      iterator_types = ["parallel", "parallel"] }
-      ins(%arg0, %arg1  : tensor<?x?xi32>, tensor<?x?xi32>) {
-    ^bb0(%arg2: i32, %arg3: i32):       // no predecessors
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %0 = dim %arg0, %c0 : tensor<?x?xi32>
+  %1 = dim %arg0, %c1 : tensor<?x?xi32>
+  %2 = linalg.init_tensor [%0, %1] : tensor<?x?xi32>
+  %3 = linalg.generic {
+    ins(%arg0, %arg1  : tensor<?x?xi32>, tensor<?x?xi32>)
+    outs(%2 : tensor<?x?xi32>) {
+    ^bb0(%arg2: i32, %arg3: i32, %arg4: i32):       // no predecessors
       %10 = addi %arg2, %arg3 : i32
       linalg.yield %10 : i32
     } -> tensor<?x?xi32>
-    %1 = linalg.indexed_generic {
-      indexing_maps = [#map0, #map0],
-      iterator_types = ["parallel", "parallel"] }
-      ins(%0 : tensor<?x?xi32>) {
-    ^bb0(%arg2: index, %arg3: index, %arg4: i32):       // no predecessors
-      %2 = index_cast %arg2 : index to i32
-      %3 = index_cast %arg3 : index to i32
-      %4 = addi %arg4, %2 : i32
-      %5 = subi %4, %3 : i32
-      linalg.yield %5 : i32
+  %4 = linalg.indexed_generic {
+    ins(%3 : tensor<?x?xi32>)
+    outs(%2 : tensor<?x?xi32>) {
+    ^bb0(%arg2: index, %arg3: index, %arg4: i32, %arg5: i32):       // no predecessors
+      %5 = index_cast %arg2 : index to i32
+      %6 = index_cast %arg3 : index to i32
+      %7 = addi %arg4, %5 : i32
+      %8 = subi %7, %6 : i32
+      linalg.yield %8 : i32
     } -> tensor<?x?xi32>
-  return %1 : tensor<?x?xi32>
+  return %4 : tensor<?x?xi32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-LABEL: func @generic_op_indexed_generic_op_fusion
 //   CHECK-NOT: linalg.generic
 //       CHECK: linalg.indexed_generic
 // CHECK-SAME:    indexing_maps = [#[[$MAP0]], #[[$MAP0]], #[[$MAP0]]]
 //      CHECK: ^{{[a-zA-Z0-9_]*}}
 // CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]*]]: index
 // CHECK-SAME: %[[ARG1:[a-zA-Z0-9_]*]]: index
 // CHECK-SAME: %[[ARG2:[a-zA-Z0-9_]*]]: i32
 // CHECK-SAME: %[[ARG3:[a-zA-Z0-9_]*]]: i32
 //      CHECK:   %[[VAL1:.+]] = addi %[[ARG2]], %[[ARG3]] : i32
 //      CHECK:   %[[ADD_OPERAND:.+]] = index_cast %[[ARG0]] : index to i32
 //      CHECK:   %[[SUB_OPERAND:.+]] = index_cast %[[ARG1]] : index to i32
 //      CHECK:   %[[VAL2:.+]] = addi %[[VAL1]], %[[ADD_OPERAND]] : i32
 //      CHECK:   %[[VAL3:.+]] = subi %[[VAL2]], %[[SUB_OPERAND]] : i32
 //      CHECK:   linalg.yield %[[VAL3]] : i32
 
 // -----
 
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 func @indexed_generic_op_generic_op_fusion(%arg0: tensor<?x?xi32>,
                                            %arg1: tensor<?x?xi32>) -> tensor<?x?xi32> {
-  %0 = linalg.indexed_generic {
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %0 = dim %arg0, %c0 : tensor<?x?xi32>
+  %1 = dim %arg0, %c1 : tensor<?x?xi32>
+  %2 = linalg.init_tensor [%0, %1] : tensor<?x?xi32>
+  %3 = linalg.indexed_generic {
     indexing_maps = [#map0, #map0],
     iterator_types = ["parallel", "parallel"] }
-    ins(%arg0 : tensor<?x?xi32>) {
-  ^bb0(%arg2: index, %arg3: index, %arg4: i32):       // no predecessors
-    %2 = index_cast %arg2 : index to i32
-    %3 = index_cast %arg3 : index to i32
-    %4 = addi %arg4, %2 : i32
-    %5 = subi %4, %3 : i32
-    linalg.yield %5 : i32
+    ins(%arg0 : tensor<?x?xi32>)
+    outs(%2 : tensor<?x?xi32>) {
+    ^bb0(%arg2: index, %arg3: index, %arg4: i32, %arg5: i32):       // no predecessors
+      %4 = index_cast %arg2 : index to i32
+      %5 = index_cast %arg3 : index to i32
+      %6 = addi %arg4, %4 : i32
+      %7 = subi %6, %5 : i32
+      linalg.yield %7 : i32
     } -> tensor<?x?xi32>
-  %1 = linalg.generic {
+  %4 = linalg.generic {
     indexing_maps = [#map0, #map0, #map0],
     iterator_types = ["parallel", "parallel"] }
-    ins(%0, %arg1 : tensor<?x?xi32>, tensor<?x?xi32>) {
-  ^bb0(%arg2: i32, %arg3: i32):       // no predecessors
+    ins(%3, %arg1 : tensor<?x?xi32>, tensor<?x?xi32>)
+    outs(%2 : tensor<?x?xi32>) {
+    ^bb0(%arg2: i32, %arg3: i32, %arg4: i32):       // no predecessors
       %10 = addi %arg2, %arg3 : i32
       linalg.yield %10 : i32
     } -> tensor<?x?xi32>
-  return %1 : tensor<?x?xi32>
+  return %4 : tensor<?x?xi32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-LABEL: func @indexed_generic_op_generic_op_fusion
 //       CHECK: linalg.indexed_generic
 // CHECK-SAME:    indexing_maps = [#[[$MAP0]], #[[$MAP0]], #[[$MAP0]]]
 //      CHECK: ^{{[a-zA-Z0-9_]*}}
 // CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]*]]: index
 // CHECK-SAME: %[[ARG1:[a-zA-Z0-9_]*]]: index
 // CHECK-SAME: %[[ARG2:[a-zA-Z0-9_]*]]: i32
 // CHECK-SAME: %[[ARG3:[a-zA-Z0-9_]*]]: i32
 //      CHECK:   %[[ADD_OPERAND:.+]] = index_cast %[[ARG0]] : index to i32
 //      CHECK:   %[[SUB_OPERAND:.+]] = index_cast %[[ARG1]] : index to i32
 //      CHECK:   %[[VAL1:.+]] = addi %[[ARG2]], %[[ADD_OPERAND]] : i32
 //      CHECK:   %[[VAL2:.+]] = subi %[[VAL1]], %[[SUB_OPERAND]] : i32
 //      CHECK:   %[[VAL3:.+]] = addi %[[VAL2]], %[[ARG3]] : i32
 //      CHECK:   linalg.yield %[[VAL3]] : i32
 //   CHECK-NOT: linalg.generic
 
 // -----
 
 // The indices of the first indexed_generic op are swapped after fusion.
 #map0 = affine_map<(d0, d1) -> (d1, d0)>
 #map1 = affine_map<(d0, d1) -> (d0, d1)>
 func @indexed_generic_op_fusion(%arg0: tensor<?x?xi32>) -> tensor<?x?xi32> {
-    %0 = linalg.indexed_generic {
-      indexing_maps = [#map0, #map0],
-      iterator_types = ["parallel", "parallel"] }
-      ins(%arg0 : tensor<?x?xi32>) {
-    ^bb0(%arg2: index, %arg3: index, %arg4: i32):       // no predecessors
-      %2 = index_cast %arg2 : index to i32
-      %3 = index_cast %arg3 : index to i32
-      %4 = addi %arg4, %2 : i32
-      %5 = subi %4, %3 : i32
-      linalg.yield %5 : i32
+  %c0 = constant 0 : index
+  %c1 = constant 1 : index
+  %0 = dim %arg0, %c0 : tensor<?x?xi32>
+  %1 = dim %arg0, %c1 : tensor<?x?xi32>
+  %2 = linalg.init_tensor [%0, %1] : tensor<?x?xi32>
+  %3 = linalg.indexed_generic {
+    ins(%arg0 : tensor<?x?xi32>)
+    outs(%2 : tensor<?x?xi32>) {
+    ^bb0(%arg2: index, %arg3: index, %arg4: i32, %arg5: i32):       // no predecessors
+      %4 = index_cast %arg2 : index to i32
+      %5 = index_cast %arg3 : index to i32
+      %6 = addi %arg4, %4 : i32
+      %7 = subi %5, %6 : i32
+      linalg.yield %7 : i32
     } -> tensor<?x?xi32>
-    %1 = linalg.indexed_generic {
-      indexing_maps = [#map1, #map1],
-      iterator_types = ["parallel", "parallel"] }
-      ins(%0 : tensor<?x?xi32>) {
-    ^bb0(%arg2: index, %arg3: index, %arg4: i32):       // no predecessors
-      %2 = index_cast %arg2 : index to i32
-      %3 = index_cast %arg3 : index to i32
-      %4 = addi %arg4, %2 : i32
-      %5 = subi %4, %3 : i32
-      linalg.yield %5 : i32
+  %4= linalg.indexed_generic {
+    ins(%3 : tensor<?x?xi32>)
+    outs(%2 : tensor<?x?xi32>) {
+    ^bb0(%arg2: index, %arg3: index, %arg4: i32, %arg5: i32):       // no predecessors
+      %5 = index_cast %arg2 : index to i32
+      %6 = index_cast %arg3 : index to i32
+      %7 = addi %arg4, %5 : i32
+      %8 = subi %7, %6 : i32
+      linalg.yield %8 : i32
     } -> tensor<?x?xi32>
-  return %1 : tensor<?x?xi32>
+  return %4 : tensor<?x?xi32>
 }
 //   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1) -> (d0, d1)>
 // CHECK-LABEL: func @indexed_generic_op_fusion
 //       CHECK: linalg.indexed_generic
 // CHECK-SAME:    indexing_maps = [#[[$MAP0]], #[[$MAP0]]]
 //      CHECK: ^{{[a-zA-Z0-9_]*}}
 // CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]*]]: index
 // CHECK-SAME: %[[ARG1:[a-zA-Z0-9_]*]]: index
 // CHECK-SAME: %[[ARG2:[a-zA-Z0-9_]*]]: i32
 //      CHECK:   %[[ADD_OPERAND1:.+]] = index_cast %[[ARG1]] : index to i32
 //      CHECK:   %[[SUB_OPERAND1:.+]] = index_cast %[[ARG0]] : index to i32
 //      CHECK:   %[[VAL1:.+]] = addi %[[ARG2]], %[[ADD_OPERAND1]] : i32
-//      CHECK:   %[[VAL2:.+]] = subi %[[VAL1]], %[[SUB_OPERAND1]] : i32
+//      CHECK:   %[[VAL2:.+]] = subi %[[SUB_OPERAND1]], %[[VAL1]] : i32
 //      CHECK:   %[[ADD_OPERAND2:.+]] = index_cast %[[ARG0]] : index to i32
 //      CHECK:   %[[SUB_OPERAND2:.+]] = index_cast %[[ARG1]] : index to i32
 //      CHECK:   %[[VAL3:.+]] = addi %[[VAL2]], %[[ADD_OPERAND2]] : i32
 //      CHECK:   %[[VAL4:.+]] = subi %[[VAL3]], %[[SUB_OPERAND2]] : i32
 //      CHECK:   linalg.yield %[[VAL4]] : i32
 //   CHECK-NOT: linalg.indexed_generic
 
 // -----
 
 func @scalar_indexed_generic_fusion
   (%arg0: tensor<5x1x1xf32>, %arg1 : tensor<i32>) -> tensor<10xf32>
 {
   %c0 = constant 0 : index  
   %cst = constant dense<1.000000e+00> : tensor<10xf32>
-  %0 = linalg.indexed_generic
+  %0 = linalg.init_tensor [] : tensor<f32>
+  %1 = linalg.indexed_generic
     {indexing_maps = [affine_map<() -> ()>, affine_map<() -> ()>],
      iterator_types = []}
-    ins(%arg1 : tensor<i32>) {
-    ^bb0(%arg2: i32):  // no predecessors
+    ins(%arg1 : tensor<i32>) outs(%0 : tensor<f32>) {
+    ^bb0(%arg2: i32, %arg3: f32):  // no predecessors
       %3 = index_cast %arg2 : i32 to index
       %4 = tensor.extract %arg0[%3, %c0, %c0] : tensor<5x1x1xf32>
       linalg.yield %4 : f32
     } -> tensor<f32>
-  %1 = linalg.generic
+  %2 = linalg.init_tensor [10] : tensor<10xf32>
+  %3 = linalg.generic
    {indexing_maps = [affine_map<(d0) -> ()>, affine_map<(d0) -> (d0)>,
                      affine_map<(d0) -> (d0)>],
     iterator_types = ["parallel"]}
-    ins(%0, %cst : tensor<f32>, tensor<10xf32>) {
-    ^bb0(%arg2: f32, %arg3: f32):  // no predecessors
-      %3 = mulf %arg2, %arg3 : f32
-      linalg.yield %3 : f32
+    ins(%1, %cst : tensor<f32>, tensor<10xf32>) outs(%2 : tensor<10xf32>) {
+    ^bb0(%arg2: f32, %arg3: f32, %arg4: f32):  // no predecessors
+      %4 = mulf %arg2, %arg3 : f32
+      linalg.yield %4 : f32
     } -> tensor<10xf32>
-  return %1 : tensor<10xf32>
+  return %3 : tensor<10xf32>
 }
 //   CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0) -> ()>
 //   CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0) -> (d0)>
 //       CHECK: func @scalar_indexed_generic_fusion
 //  CHECK-SAME:   %[[ARG0:[a-zA-Z0-9]+]]: tensor<5x1x1xf32>
 //  CHECK-SAME:   %[[ARG1:[a-zA-Z0-9]+]]: tensor<i32>
 //       CHECK:   %[[T0:.+]] = linalg.indexed_generic
 //  CHECK-SAME:     indexing_maps = [#[[MAP0]], #[[MAP1]]]
 //  CHECK-SAME:     iterator_types = ["parallel"]
 //  CHECK-SAME:     ins(%[[ARG1]] : tensor<i32>)
 //       CHECK:     tensor.extract %[[ARG0]]
 //       CHECK:     linalg.yield
 //       CHECK   return %[[T0]]
+
+// -----
+
+func @constant_fusion(%arg0 : tensor<4xf32>) -> (tensor<4xf32>) {
+  %cst = constant dense<1.0> : tensor<4xf32>
+  %1 = linalg.init_tensor [4] : tensor<4xf32>
+  %2 = linalg.generic
+    {indexing_maps = [affine_map<(d0) -> (d0)>, affine_map<(d0) -> (d0)>,
+                      affine_map<(d0) -> (d0)>],
+     iterator_types = ["parallel"]}
+    ins (%arg0, %cst : tensor<4xf32>, tensor<4xf32>)
+    outs (%1 : tensor<4xf32>) {
+    ^bb0(%arg1: f32, %arg2: f32, %arg3: f32):
+      %3 = addf %arg1, %arg2 : f32
+      linalg.yield %3 : f32
+    } -> tensor<4xf32>
+  return %2 : tensor<4xf32>
+}
+
+//  CHECK-DAG: #[[MAP:.+]] = affine_map<(d0) -> (d0)>
+//      CHECK: func @constant_fusion(%[[ARG0:.+]]: tensor<4xf32>)
+//  CHECK-DAG:   %[[CST:.+]] = constant 1.000000e+00 : f32
+//  CHECK-DAG:   %[[T0:.+]] = linalg.init_tensor [4] : tensor<4xf32>
+//      CHECK:   %[[T1:.+]] = linalg.generic
+// CHECK-SAME:     indexing_maps = [#[[MAP]], #[[MAP]]]
+// CHECK-SAME:     ins(%[[ARG0]] : tensor<4xf32>)
+// CHECK-SAME:     outs(%[[T0]] : tensor<4xf32>)
+//      CHECK:   ^{{.+}}(
+// CHECK-SAME:     %[[ARG1:[a-zA-Z0-9_]+]]: f32, %[[ARG2:[a-zA-Z0-9_]+]]: f32)
+//      CHECK:     %[[T2:.+]] = addf %[[ARG1]], %[[CST]]
+//      CHECK:     linalg.yield %[[T2]]
+//      CHECK:   return %[[T1]]

mlir/test/Dialect/Linalg/generalize-named-ops.mlir

 // CHECK: ^{{.*}}(%[[FILTER_ARG:.+]]: f32, %[[INPUT_ARG:.+]]: f32, %[[OUTPUT_ARG:.+]]: f32)
 // CHECK:   %[[MUL:.+]] = mulf %[[FILTER_ARG]], %[[INPUT_ARG]]
 // CHECK:   %[[ADD:.+]] = addf %[[MUL]], %[[OUTPUT_ARG]]
 // CHECK:   linalg.yield %[[ADD]]
 
 // -----
 
 func @generalize_matmul_buffer(%A : memref<16x8xf32>, %B: memref<8x32xf32>, %C: memref<16x32xf32>) {
-  linalg.matmul ins(%A, %B: memref<16x8xf32>, memref<8x32xf32>) outs(%C: memref<16x32xf32>)
+  linalg.matmul ins(%A, %B: memref<16x8xf32>, memref<8x32xf32>)
+               outs(%C: memref<16x32xf32>)
   return
 }
 
 
 // CHECK: #[[A_MAP:.+]] = affine_map<(d0, d1, d2) -> (d0, d2)>
 // CHECK: #[[B_MAP:.+]] = affine_map<(d0, d1, d2) -> (d2, d1)>
 // CHECK: #[[C_MAP:.+]] = affine_map<(d0, d1, d2) -> (d0, d1)>
 
 // CHECK: func @generalize_matmul_buffer
 // CHECK-SAME: %[[A:.+]]: memref<16x8xf32>
 // CHECK-SAME: %[[B:.+]]: memref<8x32xf32>
 // CHECK-SAME: %[[C:.+]]: memref<16x32xf32>
 
 // CHECK: linalg.generic
 // CHECK-SAME: indexing_maps = [#[[A_MAP]], #[[B_MAP]], #[[C_MAP]]]
 // CHECK-SAME: iterator_types = ["parallel", "parallel", "reduction"]
-// CHECK-SAME: ins(%[[A]], %[[B]]
+// CHECK-SAME:  ins(%[[A]], %[[B]]
 // CHECK-SAME: outs(%[[C]]
 
 // CHECK: ^{{.*}}(%[[A_ARG:.+]]: f32, %[[B_ARG:.+]]: f32, %[[C_ARG:.+]]: f32)
 // CHECK:   %[[MUL:.+]] = mulf %[[A_ARG]], %[[B_ARG]] : f32
 // CHECK:   %[[ADD:.+]] = addf %[[C_ARG]], %[[MUL]] : f32
 // CHECK:   linalg.yield %[[ADD]] : f32
 
 // -----
 
 func @generalize_matmul_tensor(%A : tensor<16x8xf32>, %B: tensor<8x32xf32>, %C: tensor<16x32xf32>) -> tensor<16x32xf32> {
-  %0 = linalg.matmul ins(%A, %B: tensor<16x8xf32>, tensor<8x32xf32>) init(%C: tensor<16x32xf32>) -> tensor<16x32xf32>
+  %0 = linalg.matmul ins(%A, %B: tensor<16x8xf32>, tensor<8x32xf32>)
+                    outs(%C: tensor<16x32xf32>) -> tensor<16x32xf32>
   return %0: tensor<16x32xf32>
 }
 
 // CHECK: func @generalize_matmul_tensor
 
 // CHECK: linalg.generic
-// CHECK-SAME: ins(%{{.+}}, %{{.+}} : tensor<16x8xf32>, tensor<8x32xf32>)
-// CHECK-SAME: init(%{{.+}} : tensor<16x32xf32>)
+// CHECK-SAME:  ins(%{{.+}}, %{{.+}} : tensor<16x8xf32>, tensor<8x32xf32>)
+// CHECK-SAME: outs(%{{.+}} : tensor<16x32xf32>)
 
 // CHECK:      ^{{.*}}(%[[A_ARG:.+]]: f32, %[[B_ARG:.+]]: f32, %[[C_ARG:.+]]: f32)
 // CHECK-NEXT:   %[[MUL:.+]] = mulf %[[A_ARG]], %[[B_ARG]] : f32
 // CHECK-NEXT:   %[[ADD:.+]] = addf %[[C_ARG]], %[[MUL]] : f32
 // CHECK-NEXT:   linalg.yield %[[ADD]] : f32
 // CHECK-NEXT: -> tensor<16x32xf32>

mlir/test/Dialect/Linalg/invalid.mlir

linalg.generic {
     ^bb(%i : i32):
     linalg.yield %i : i32
   }
 }
 
 // -----
 
 func @generic_one_d_view(%arg0: memref<?xf32, affine_map<(i)[off]->(off + i)>>) {
-  // expected-error @+1 {{op expected indexing_map #0 results to match view rank: 'memref<?xf32, affine_map<(d0)[s0] -> (d0 + s0)>>'}}
+  // expected-error @+1 {{expected shaped value rank (1) to match the result rank of indexing_map #0 (2)}}
   linalg.generic {
     indexing_maps =  [ affine_map<() -> (0, 0)> ],
     iterator_types = []}
       outs(%arg0 : memref<?xf32, affine_map<(i)[off]->(off + i)>>) {
     ^bb(%f : f32):
       linalg.yield %f: f32
   }
 }
^bb2:
       linalg.yield %f0: f32
   }
 }
 
 // -----
 
 func @generic_empty_region(%arg0: memref<f32>) {
   %f0 = constant 0.0: f32
-  // expected-error @+1 {{linalg.generic' op expected region with 1 block}}
+  // expected-error @+1 {{linalg.generic' op expected 1 region with 1 block}}
   linalg.generic {
-    indexing_maps =  [ affine_map<() -> (0)> ],
+    indexing_maps =  [ affine_map<() -> ()> , affine_map<() -> ()> ],
     iterator_types = []}
     ins(%arg0 : memref<f32>)
    outs(%arg0 : memref<f32>) {
   }
 }
 
 // -----
 
 func @generic_mismatched_num_arguments(%arg0: memref<f32>) {
-  // expected-error @+1 {{op expected number of block arguments to match number of operands}}
+  // expected-error @+1 {{expected as many non-induction variable region arguments as the number of shaped operands}}
   linalg.generic {
-      indexing_maps =  [ affine_map<() -> (0)> ],
+      indexing_maps =  [ affine_map<() -> ()>, affine_map<() -> ()> ],
       iterator_types = []}
-      outs(%arg0 : memref<f32>) {
-    ^bb(%f: f32, %g: f32):
+      outs(%arg0, %arg0 : memref<f32>, memref<f32>) {
+    ^bb(%f: f32):
       linalg.yield %f: f32
   }
 }
 
 // -----
 
 func @generic_block_arg_type(%arg0: memref<f32>) {
-  // expected-error @+1 {{op expected block argument 1 of the same type as elemental type of output operand: 'memref<f32>'}}
+  // expected-error @+1 {{expected type of bb argument #0 ('i1') to match element type of corresponding shaped operand ('f32')}}
   linalg.generic {
-    indexing_maps =  [ affine_map<() -> (0)> ],
+    indexing_maps =  [ affine_map<() -> ()> ],
     iterator_types = []}
       outs(%arg0 : memref<f32>) {
     ^bb(%i: i1):
     linalg.yield %i : i1
   }
 }
 
 // -----
 
-func @indexed_generic_block_arg_count(%arg0: memref<f32>) {
-  // expected-error @+1 {{op expected number of block arguments to match number of operands + number of loops}}
+func @indexed_generic_block_arg_count(%arg0: memref<?xf32>) {
+  // expected-error @+1 {{expected as many non-induction variable region arguments as the number of shaped operands}}
   linalg.indexed_generic {
-    indexing_maps =  [ affine_map<(d0) -> (d0)> ],
+    indexing_maps =  [ affine_map<(i) -> (i)> ],
     iterator_types = ["parallel"]}
-      outs(%arg0 : memref<f32>) {
+      outs(%arg0 : memref<?xf32>) {
     ^bb(%f: f32):
       linalg.yield %f : f32
   }
 }
 
 // -----
 
-func @indexed_generic_block_induction_var_arg_type(%arg0: memref<f32>) {
-  // expected-error @+1 {{op expected block argument 1 to be an index}}
+func @indexed_generic_block_induction_var_arg_type(%arg0: memref<?xf32>) {
+  // expected-error @+1 {{op expected index block argument #0}}
   linalg.indexed_generic {
     indexing_maps =  [ affine_map<(d0) -> (d0)> ],
     iterator_types = ["parallel"]}
-      outs(%arg0 : memref<f32>) {
+      outs(%arg0 : memref<?xf32>) {
     ^bb(%i: f64, %f: f32):
     linalg.yield %f: f32
   }
 }
 
 // -----
 
-func @indexed_generic_block_arg_type(%arg0: memref<f32>) {
-  // expected-error @+1 {{op expected block argument 2 of the same type as elemental type of output operand: 'memref<f32>'}}
+func @indexed_generic_block_arg_type(%arg0: memref<?xf32>) {
+  // expected-error @+1 {{expected type of bb argument #1 ('i1') to match element type of corresponding shaped operand ('f32')}}
   linalg.indexed_generic {
     indexing_maps =  [ affine_map<(d0) -> (d0)> ],
     iterator_types = ["parallel"]}
-      outs(%arg0 : memref<f32>) {
+      outs(%arg0 : memref<?xf32>) {
     ^bb(%i: index, %f: i1):
     linalg.yield %i: index
   }
 }
 
 // -----
 
 func @indexed_generic_arg_count(%arg0: memref<f32>) {
-  // expected-error @+1 {{op expected number of block arguments to match number of operands + number of loops}}
+  // expected-error @+1 {{expected as many non-induction variable region arguments as the number of shaped operands}}
   linalg.indexed_generic {
     indexing_maps =  [ affine_map<()[] -> ()> ],
     iterator_types = []}
       outs(%arg0 : memref<f32>) {
     ^bb(%0: index, %1: f32):
       linalg.yield %1: f32
   }
   return
 }
 
 // -----
 
-func @indexed_generic_induction_var_arg_type(%arg0: memref<f32>) {
-  // expected-error @+1 {{op expected block argument 1 to be an index}}
-  linalg.indexed_generic {
-    iterator_types = ["parallel"],
-    indexing_maps = [ affine_map<(i) -> (i)> ]}
-      outs(%arg0 : memref<f32>) {
-    ^bb(%0: i32, %1: f32):
-      linalg.yield %1: f32
-  }
-}
-
-// -----
-
 func @indexed_generic_result_count(%arg0: memref<?xf32>) {
   // expected-error @+6 {{op expected number of yield values (1) to match the number of operands of the enclosing LinalgOp (2)}}
   linalg.indexed_generic {
     indexing_maps =  [ affine_map<(d0) -> (d0)> ],
     iterator_types = ["parallel"]}
       outs(%arg0 : memref<?xf32>) {
     ^bb(%i: index, %val: f32):
       linalg.yield %val, %val: f32, f32
linalg.generic {
     ^bb(%i: f32):
       %0 = constant 0: i1
       linalg.yield %0: i1
   }
 }
 
 // -----
 
-func @generic_result_tensor_type(%arg0: memref<?xf32, affine_map<(i)[off]->(off + i)>>) {
-  // expected-error @+1 {{op result #0 must be ranked tensor of any type values, but got 'f32'}}
+func @generic_result_tensor_type(%arg0: memref<?xf32, affine_map<(i)[off]->(off + i)>>,
+                                 %arg1: tensor<?xf32>) {
+  // expected-error @+1 {{expected type of operand #1 ('tensor<?xf32>') to match type of corresponding result ('f32')}}
   %0 = linalg.generic {
-    indexing_maps = [ affine_map<(i) -> (i)> ],
+    indexing_maps = [ affine_map<(i) -> (i)> , affine_map<(i) -> (i)> ],
     iterator_types = ["parallel"]}
-      ins(%arg0 : memref<?xf32, affine_map<(i)[off]->(off + i)>>) {
-    ^bb(%i: f32):
+       ins(%arg0 : memref<?xf32, affine_map<(i)[off]->(off + i)>>)
+      outs(%arg1 : tensor<?xf32>) {
+    ^bb(%i: f32, %j: f32):
       linalg.yield %i: f32
   } -> f32
 }
 
 // -----
 
+func @generic_result_tensor_type(%arg0: memref<?xf32, affine_map<(i)[off]->(off + i)>>,
+                                 %arg1: tensor<?xf32>) {
+  // expected-error @+1 {{unexpected output tensor expression in indexing map #0 a.k.a 'd0' is function of reduction iterator 'd0'}}
+  %0 = linalg.generic {
+    indexing_maps = [ affine_map<(i) -> (i)> , affine_map<(i) -> (i)> ],
+    iterator_types = ["reduction"]}
+       ins(%arg0 : memref<?xf32, affine_map<(i)[off]->(off + i)>>)
+      outs(%arg1 : tensor<?xf32>) {
+    ^bb(%i: f32, %j: f32):
+      linalg.yield %i: f32
+  } -> tensor<?xf32>
+}
+
+// -----
+
 func @generic(%arg0: memref<?x?xi4>) {
   // expected-error @+2 {{op expects regions to end with 'linalg.yield', found 'std.addf'}}
   // expected-note @+1 {{in custom textual format, the absence of terminator implies 'linalg.yield'}}
   linalg.generic  {
     indexing_maps = [ affine_map<(i) -> (i)> ],
     iterator_types = ["parallel"]}
       outs(%arg0 : memref<?x?xi4>) {
     ^bb(%0: i4) :
       %1 = std.addf %0, %0: i4
   }
   return
 }
 
 // -----
 
-func @conv_rank_limit(%arg0: memref<?xf32>, %arg1: memref<?xf32>, %arg2: memref<?xf32>) {
-  // expected-error @+1 {{expects memref ranks to be greater than 2}}
-  linalg.conv(%arg0, %arg1, %arg2) : memref<?xf32>, memref<?xf32>, memref<?xf32>
-}
-
-// -----
+// This test is currently disabled: subject to verifier ordering issues.
+// Instead, when the ranks are not greater than 2, an assertion will be triggered
+// in LinalgStructuredOps.td::ConvOp::iterator_types() for now because the
+// verifier inspects the iterator_types. This is slated to become an
+// autogenerated op in the future, alleviating the issue.
+// func @conv_rank_limit(%arg0: memref<?xf32>, %arg1: memref<?xf32>, %arg2: memref<?xf32>) {
+//   // DISABLED_expected -error @+1 {{expects memref ranks to be greater than 2}}
+//   linalg.conv(%arg0, %arg1, %arg2) : memref<?xf32>, memref<?xf32>, memref<?xf32>
+// }
+//
+// // -----
 
 // expected-error @+1 {{unknown Linalg type}}
 !invalid_type = type !linalg.unknown
 
 // -----
 
 // expected-error @+1 {{expected valid keyword}}
 !invalid_type = type !linalg<"?">
%0 = linalg.reshape %arg0 [affine_map<(i, j, k) -> (i, j)>, affine_map<(i, j, k) -> (k)>] :
     memref<?x?x?xf32> into memref<?x?xf32, affine_map<(d0, d1)[s0] -> (d0 * s0 + d1)>>
 }
 
 // -----
 
 func @pooling_rank_mismatch(%arg0: memref<?x?x?xf32>,
                             %arg1: memref<2x3xf32>,
                             %arg2: memref<?x?x?xf32>) {
-  // expected-error @+1 {{expects memref ranks to match}}
+  // expected-error @+1 {{expected shaped value rank (2) to match the result rank of indexing_map #1 (3)}}
   linalg.pooling_max(%arg0, %arg1, %arg2) {strides = [2, 1, 2]}:
     memref<?x?x?xf32>, memref<2x3xf32>, memref<?x?x?xf32>
   return
 }
 
 // -----
 
 func @named_ops(%a3: memref<?x?x?xf32>, %b3: memref<?x?xf32>, %c3: memref<?x?x?xf32>) {
-  // expected-error @+1 {{op expected indexing_map #1 results to match view rank: 'memref<?x?xf32>'}}
+  // expected-error @+1 {{expected shaped value rank (2) to match the result rank of indexing_map #1 (3)}}
   linalg.batch_matmul ins(%a3, %b3: memref<?x?x?xf32>, memref<?x?xf32>)
                      outs(%c3 : memref<?x?x?xf32>)
   return
 }
 
 // -----
 
-func @empty_init_expected(%m: memref<?x?xf32>, %t: tensor<?x?xf32>) {
-  // expected-error @+1 {{expected empty `init` when op has no results or no reduction dims}}
-  linalg.matmul ins(%m, %m: memref<?x?xf32>, memref<?x?xf32>)
-               outs(%m : memref<?x?xf32>)
-               init(%t : tensor<?x?xf32>)
-  return
-}
-
-// -----
-
 func @incorrect_region_arg_count(%m: memref<?x?xf32>) {
-  // expected-error @+3 {{region expects 3 args, got 4}}
+  // expected-error @+3 {{region expects 3 args, got 2}}
   %res = linalg.matmul ins(%m, %m : memref<?x?xf32>, memref<?x?xf32>)
                        -> tensor<?x?xf32>, tensor<?x?xf32>
   return
 }
 
 // -----
 
-func @single_tensor_result(%m: memref<?x?xf32>, %t: tensor<?x?xf32>) {
-  // expected-error @+1 {{expected single tensor result when reduction present}}
-  %res:2 = linalg.matmul ins(%m : memref<?x?xf32>)
-                        init(%t, %t : tensor<?x?xf32>, tensor<?x?xf32>)
-                          -> tensor<?x?xf32>, tensor<?x?xf32>
-  return
-}
-
-// -----
-
-func @matching_inits(%m: memref<?x?xf32>, %t: tensor<?x?xf32>) {
-  // expected-error @+1 {{expected #init tensors to match #results when reduction present}}
-  %res = linalg.matmul ins(%m, %m : memref<?x?xf32>, memref<?x?xf32>)
-                      init(%t, %t : tensor<?x?xf32>, tensor<?x?xf32>)
-                        -> tensor<?x?xf32>
-  return
-}
-
-// -----
-
 func @matching_inits(%m: memref<?x?xf32>, %t: tensor<?x?xf32>) {
-  // expected-error @+1 {{expected init tensor #0 of the same type as result #0}}
+  // expected-error @+1 {{expected type of operand #2 ('tensor<?x?xf32>') to match type of corresponding result ('tensor<?xf32>')}}
   %res = linalg.matmul ins(%m, %m : memref<?x?xf32>, memref<?x?xf32>)
-                      init(%t : tensor<?x?xf32>)
+                      outs(%t : tensor<?x?xf32>)
                         -> tensor<?xf32>
   return
 }

mlir/test/Dialect/Linalg/parallel-loops.mlir

 //       CHECK:     scf.parallel (%[[IV3:.*]]) = (%[[C0]]) to (%[[D3]]) step (%[[C1]])
 //       CHECK:       load %{{.*}}[%[[IV0]], %[[IV1]], %[[IV2]], %[[IV3]]]
 //       CHECK:       store %{{.*}}, %{{.*}}[%[[IV0]], %[[IV1]], %[[IV3]]]
 
 // -----
 
 #accesses = [
   affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2, d3, d4, d5)>,
-  affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d2, d4, d5)>
+  affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d4, d3)>
 ]
 #trait = {
   iterator_types = ["parallel", "parallel", "reduction", "parallel", "parallel", "reduction"],
   indexing_maps = #accesses
 }
 
 func @lower_mixed_parallel(%A: memref<?x?x?x?x?x?xf32>, %B: memref<?x?x?x?xf32>) {
   linalg.generic #trait

 //   CHECK-DAG: %[[D3:.*]] = dim %{{.*}}, %c3
 //   CHECK-DAG: %[[D4:.*]] = dim %{{.*}}, %c4
 //   CHECK-DAG: %[[D5:.*]] = dim %{{.*}}, %c5
 //       CHECK: scf.parallel (%[[IV0:.*]], %[[IV1:.*]]) = (%[[C0]], %[[C0]]) to (%[[D0]], %[[D1]]) step (%[[C1]], %[[C1]])
 //       CHECK:   scf.for %[[IV2:.*]] = %[[C0]] to %[[D2]] step %[[C1]]
 //       CHECK:     scf.parallel (%[[IV3:.*]], %[[IV4:.*]]) = (%[[C0]], %[[C0]]) to (%[[D3]], %[[D4]]) step (%[[C1]], %[[C1]])
 //       CHECK:       scf.for %[[IV5:.*]] = %[[C0]] to %[[D5]] step %[[C1]]
 //       CHECK:       load %{{.*}}[%[[IV0]], %[[IV1]], %[[IV2]], %[[IV3]], %[[IV4]], %[[IV5]]]
-//       CHECK:       store %{{.*}}, %{{.*}}[%[[IV0]], %[[IV2]], %[[IV4]], %[[IV5]]]
+//       CHECK:       store %{{.*}}, %{{.*}}[%[[IV0]], %[[IV1]], %[[IV4]], %[[IV3]]]

mlir/test/Dialect/Linalg/reshape_fusion.mlir

-// RUN: mlir-opt %s -linalg-fusion-for-tensor-ops -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -linalg-fusion-for-tensor-ops -split-input-file -verify-each=0 | FileCheck %s
 
 #map0 = affine_map<(d0, d1, d2) -> (d2, d0, d1)>
 #map1 = affine_map<(d0, d1, d2) -> (d1, d2, d0)>
-func @generic_op_reshape_producer_fusion(%arg0 : tensor<?x?x?x?xf32>,
+func @generic_op_reshape_producer_fusion(%arg0 : tensor<?x?x4x?xf32>,
                                          %arg1 : tensor<?x?x?xf32>) ->
                                          tensor<?x?x?xf32>
 {
   %0 = linalg.tensor_reshape %arg0 [affine_map<(i, j, k, l) -> (i)>,
                                     affine_map<(i, j, k, l) -> (j, k)>,
                                     affine_map<(i, j, k, l) -> (l)>] :
-    tensor<?x?x?x?xf32> into tensor<?x?x?xf32>
+    tensor<?x?x4x?xf32> into tensor<?x?x?xf32>
   %1 = linalg.generic {
      indexing_maps = [#map0, #map1, #map1],
      iterator_types = ["parallel", "parallel", "parallel"]}
-      ins(%0, %arg1 : tensor<?x?x?xf32>, tensor<?x?x?xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
+       ins(%0, %arg1 : tensor<?x?x?xf32>, tensor<?x?x?xf32>)
+      outs(%0 : tensor<?x?x?xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %s: f32):       // no predecessors
       %1 = mulf %arg3, %arg4 : f32
       linalg.yield %1 : f32
   } -> tensor<?x?x?xf32>
   return %1 : tensor<?x?x?xf32>
 }
 
 //  CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2, d3) -> (d0)>
-//  CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2, d3) -> (d1)>
-//  CHECK-DAG: #[[MAP2:.+]] = affine_map<(d0, d1, d2, d3) -> (d2, d3)>
-//  CHECK-DAG: #[[MAP3:.+]] = affine_map<(d0, d1, d2, d3) -> (d3, d0, d1, d2)>
-//  CHECK-DAG: #[[MAP4:.+]] = affine_map<(d0, d1, d2, d3) -> (d2, d3, d0, d1)>
+//  CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2, d3) -> (d1, d2)>
+//  CHECK-DAG: #[[MAP2:.+]] = affine_map<(d0, d1, d2, d3) -> (d3)>
+//  CHECK-DAG: #[[MAP3:.+]] = affine_map<(d0, d1, d2, d3) -> (d1)>
+//  CHECK-DAG: #[[MAP4:.+]] = affine_map<(d0, d1, d2, d3) -> (d2, d3)>
+//  CHECK-DAG: #[[MAP5:.+]] = affine_map<(d0, d1, d2, d3) -> (d3, d0, d1, d2)>
+//  CHECK-DAG: #[[MAP6:.+]] = affine_map<(d0, d1, d2, d3) -> (d2, d3, d0, d1)>
 //      CHECK: func @generic_op_reshape_producer_fusion
-// CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: tensor<?x?x?x?xf32>
+// CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: tensor<?x?x4x?xf32>
 // CHECK-SAME:   %[[ARG1:[a-zA-Z0-9_]+]]: tensor<?x?x?xf32>
-//      CHECK:   %[[T0:.+]] = linalg.tensor_reshape %[[ARG1]]
+//  CHECK-DAG:   %[[C0:.+]] = constant 0 : index
+//  CHECK-DAG:   %[[C1:.+]] = constant 1 : index
+//  CHECK-DAG:   %[[C2:.+]] = constant 2 : index
+//  CHECK-DAG:   %[[C4:.+]] = constant 4 : index
+//      CHECK:   %[[T0:.+]] = linalg.tensor_reshape %[[ARG0]]
 // CHECK-SAME:     [#[[MAP0]], #[[MAP1]], #[[MAP2]]]
-// CHECK-SAME:     tensor<?x?x?xf32> into tensor<?x?x?x?xf32>
-//      CHECK:   %[[T1:.+]] = linalg.generic
-// CHECK-SAME:     indexing_maps = [#[[MAP3]], #[[MAP4]], #[[MAP4]]]
+//      CHECK:   %[[T1:.+]] = linalg.tensor_reshape %[[ARG1]]
+// CHECK-SAME:     [#[[MAP0]], #[[MAP3]], #[[MAP4]]]
+//  CHECK-DAG:   %[[D0:.+]] = dim %[[T0]], %[[C0]]
+//  CHECK-DAG:   %[[D1:.+]] = dim %[[T0]], %[[C1]]
+//  CHECK-DAG:   %[[D2:.+]] = dim %[[T0]], %[[C2]]
+//      CHECK:   %[[D3:.+]] = divi_unsigned %[[D0]], %[[C4]]
+//      CHECK:   %[[T2:.+]] = linalg.init_tensor [%[[D1]], %[[D2]], %[[D3]], 4]
+//      CHECK:   %[[T3:.+]] = linalg.generic
+// CHECK-SAME:     indexing_maps = [#[[MAP5]], #[[MAP6]], #[[MAP6]]]
 // CHECK-SAME:     ["parallel", "parallel", "parallel", "parallel"]
-// CHECK-SAME:     ins(%[[ARG0]], %[[T0]] : tensor<?x?x?x?xf32>, tensor<?x?x?x?xf32>)
-//      CHECK:   %[[T2:.+]] = linalg.tensor_reshape
-// CHECK-SAME:     [#[[MAP0]], #[[MAP1]], #[[MAP2]]]
-// CHECK-SAME:     tensor<?x?x?x?xf32> into tensor<?x?x?xf32>
-//      CHECK:   return %[[T2]]
+// CHECK-SAME:     ins(%[[ARG0]], %[[T1]] : tensor<?x?x4x?xf32>, tensor<?x?x?x4xf32>)
+// CHECK-SAME:     outs(%[[T2]] : tensor<?x?x?x4xf32>)
+//      CHECK:   %[[T4:.+]] = linalg.tensor_reshape %[[T3]]
+// CHECK-SAME:     [#[[MAP0]], #[[MAP3]], #[[MAP4]]]
+// CHECK-SAME:     tensor<?x?x?x4xf32> into tensor<?x?x?xf32>
+//      CHECK:   return %[[T4]]
 
 // -----
 
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 func @generic_op_reshape_consumer_fusion(%arg0 : tensor<?x?xf32>,
                                          %arg1 : tensor<?x?xf32>) ->
-                                         tensor<?x?x4x5xf32>
+                                         tensor<?x4x?x5xf32>
 {
   %0 = linalg.generic {
      indexing_maps = [#map0, #map0, #map0],
      iterator_types = ["parallel", "parallel"]}
-      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
+       ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%arg0 : tensor<?x?xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %s: f32):       // no predecessors
       %1 = mulf %arg3, %arg4 : f32
       linalg.yield %1 : f32
   } -> tensor<?x?xf32>
   %1 = linalg.tensor_reshape %0 [affine_map<(i, j, k, l) -> (i)>,
                                  affine_map<(i, j, k, l) -> (j, k, l)>] :
-    tensor<?x?xf32> into tensor<?x?x4x5xf32>
-  return %1 : tensor<?x?x4x5xf32>
+    tensor<?x?xf32> into tensor<?x4x?x5xf32>
+  return %1 : tensor<?x4x?x5xf32>
 }
 
 //  CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2, d3) -> (d0)>
 //  CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2, d3) -> (d1, d2, d3)>
 //  CHECK-DAG: #[[MAP2:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 //      CHECK: func @generic_op_reshape_consumer_fusion
 // CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: tensor<?x?xf32>
 // CHECK-SAME:   %[[ARG1:[a-zA-Z0-9_]+]]: tensor<?x?xf32>
+//  CHECK-DAG:   %[[C0:.+]] = constant 0 : index
+//  CHECK-DAG:   %[[C1:.+]] = constant 1 : index
+//  CHECK-DAG:   %[[C20:.+]] = constant 20 : index
 //      CHECK:   %[[T0:.+]] = linalg.tensor_reshape %[[ARG0]]
 // CHECK-SAME:     [#[[MAP0]], #[[MAP1]]]
-// CHECK-SAME:     tensor<?x?xf32> into tensor<?x?x4x5xf32>
+// CHECK-SAME:     tensor<?x?xf32> into tensor<?x4x?x5xf32>
 //      CHECK:   %[[T1:.+]] = linalg.tensor_reshape %[[ARG1]]
 // CHECK-SAME:     [#[[MAP0]], #[[MAP1]]]
-// CHECK-SAME:     tensor<?x?xf32> into tensor<?x?x4x5xf32>
-//      CHECK:   %[[T2:.+]] = linalg.generic
+// CHECK-SAME:     tensor<?x?xf32> into tensor<?x4x?x5xf32>
+//  CHECK-DAG:   %[[D0:.+]] = dim %[[ARG0]], %[[C0]]
+//  CHECK-DAG:   %[[D1:.+]] = dim %[[ARG0]], %[[C1]]
+//      CHECK:   %[[D2:.+]] = divi_unsigned %[[D1]], %[[C20]]
+//      CHECK:   %[[T2:.+]] = linalg.init_tensor [%[[D0]], 4, %[[D2]], 5]
+//      CHECK:   %[[T3:.+]] = linalg.generic
 // CHECK-SAME:     indexing_maps = [#[[MAP2]], #[[MAP2]], #[[MAP2]]]
 // CHECK-SAME:     ["parallel", "parallel", "parallel", "parallel"]
-// CHECK-SAME:     ins(%[[T0]], %[[T1]] : tensor<?x?x4x5xf32>, tensor<?x?x4x5xf32>)
-//      CHECK:   return %[[T2]] : tensor<?x?x4x5xf32>
+// CHECK-SAME:     ins(%[[T0]], %[[T1]] : tensor<?x4x?x5xf32>, tensor<?x4x?x5xf32>)
+// CHECK-SAME:     outs(%[[T2]] : tensor<?x4x?x5xf32>)
+//      CHECK:   return %[[T3]] : tensor<?x4x?x5xf32>
 
 
 // -----
 
 func @reshape_as_consumer_permutation
   (%a : tensor<?x?x?xf32>, %b : tensor<?x?xf32>)
-    -> tensor<?x?x?x?x?x?xf32> {
+    -> tensor<?x2x?x3x4x?xf32> {
   %c = linalg.generic {
          indexing_maps = [affine_map<(d0, d1, d2) -> (d1, d0, d2)>,
                           affine_map<(d0, d1, d2) -> (d1, d2)>,
                           affine_map<(d0, d1, d2) -> (d0, d2, d1)>],
          iterator_types = ["parallel", "parallel", "parallel"]}
-         ins(%a, %b : tensor<?x?x?xf32>, tensor<?x?xf32>) {
-       ^bb0(%arg0 : f32, %arg1: f32):
+          ins(%a, %b : tensor<?x?x?xf32>, tensor<?x?xf32>)
+         outs(%a : tensor<?x?x?xf32>) {
+       ^bb0(%arg0 : f32, %arg1: f32, %s: f32):
          %1 = addf %arg0, %arg1 : f32
          linalg.yield %1 : f32
        } -> tensor<?x?x?xf32>
   %d = linalg.tensor_reshape %c
          [affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1)>,
           affine_map<(d0, d1, d2, d3, d4, d5) -> (d2)>,
           affine_map<(d0, d1, d2, d3, d4, d5) -> (d3, d4, d5)>]
-       : tensor<?x?x?xf32> into tensor<?x?x?x?x?x?xf32>
-  return %d : tensor<?x?x?x?x?x?xf32>
+       : tensor<?x?x?xf32> into tensor<?x2x?x3x4x?xf32>
+  return %d : tensor<?x2x?x3x4x?xf32>
 }
 //  CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d2)>
 //  CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d3, d4)>
 //  CHECK-DAG: #[[MAP2:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d5)>
 //  CHECK-DAG: #[[MAP3:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2)>
 //  CHECK-DAG: #[[MAP4:.+]] = affine_map<(d0, d1, d2, d3) -> (d3)>
 //  CHECK-DAG: #[[MAP5:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d2, d3, d4, d0, d1, d5)>
 //  CHECK-DAG: #[[MAP6:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d2, d3, d4, d5)>
 //  CHECK-DAG: #[[MAP7:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d5, d2, d3, d4)>
 //      CHECK: func @reshape_as_consumer_permutation
 // CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: tensor<?x?x?xf32>
 // CHECK-SAME:   %[[ARG1:[a-zA-Z0-9_]+]]: tensor<?x?xf32>
+//  CHECK-DAG:   %[[C0:.+]] = constant 0 : index
+//  CHECK-DAG:   %[[C1:.+]] = constant 1 : index
+//  CHECK-DAG:   %[[C2:.+]] = constant 2 : index
+//  CHECK-DAG:   %[[C12:.+]] = constant 12 : index
 //      CHECK:   %[[T0:.+]] = linalg.tensor_reshape %[[ARG0]]
 // CHECK-SAME:     [#[[MAP0]], #[[MAP1]], #[[MAP2]]]
-// CHECK-SAME:     tensor<?x?x?xf32> into tensor<?x?x?x?x?x?xf32>
+// CHECK-SAME:     tensor<?x?x?xf32> into tensor<3x4x?x?x2x?xf32>
 //      CHECK:   %[[T1:.+]] = linalg.tensor_reshape %[[ARG1]]
 // CHECK-SAME:     [#[[MAP3]], #[[MAP4]]]
-// CHECK-SAME:     tensor<?x?xf32> into tensor<?x?x?x?xf32>
-//      CHECK:   %[[T2:.+]] = linalg.generic
+// CHECK-SAME:     tensor<?x?xf32> into tensor<3x4x?x?xf32>
+//  CHECK-DAG:   %[[D0:.+]] = dim %[[ARG0]], %[[C0]]
+//      CHECK:   %[[D1:.+]] = divi_unsigned %[[D0]], %[[C2]]
+//  CHECK-DAG:   %[[D2:.+]] = dim %[[ARG0]], %[[C2]]
+//  CHECK-DAG:   %[[D3:.+]] = dim %[[ARG0]], %[[C1]]
+//  CHECK-DAG:   %[[D4:.+]] = divi_unsigned %[[D3]], %[[C12]]
+//      CHECK:   %[[T2:.+]] = linalg.init_tensor [%[[D1]], 2, %[[D2]], 3, 4, %[[D4]]]
+//      CHECK:   %[[T3:.+]] = linalg.generic
 // CHECK-SAME:     indexing_maps = [#[[MAP5]], #[[MAP6]], #[[MAP7]]]
 // CHECK-SAME:     ["parallel", "parallel", "parallel", "parallel", "parallel", "parallel"]
-// CHECK-SAME:     ins(%[[T0]], %[[T1]] : tensor<?x?x?x?x?x?xf32>, tensor<?x?x?x?xf32>)
-//      CHECK:   return %[[T2]] : tensor<?x?x?x?x?x?xf32>
+// CHECK-SAME:     ins(%[[T0]], %[[T1]] : tensor<3x4x?x?x2x?xf32>, tensor<3x4x?x?xf32>)
+// CHECK-SAME:     outs(%[[T2]] : tensor<?x2x?x3x4x?xf32>)
+//      CHECK:   return %[[T3]] : tensor<?x2x?x3x4x?xf32>
 
 // -----
 
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 #map1 = affine_map<(d0, d1, d2) -> (d0, d1)>
 #map2 = affine_map<(d0, d1, d2) -> (d2)>
 
 func @generic_op_reshape_consumer_static(%arg0: tensor<264x4xf32>)
                                             -> tensor<8x33x4xf32> {
   %cst = constant dense<2.000000e+00> : tensor<264x4xf32>
   %0 = linalg.generic {
      indexing_maps = [#map0, #map0, #map0],
      iterator_types = ["parallel", "parallel"]}
-      ins(%arg0, %cst : tensor<264x4xf32>, tensor<264x4xf32>) {
-    ^bb0(%arg1: f32, %arg2: f32):  // no predecessors
+       ins(%arg0, %cst : tensor<264x4xf32>, tensor<264x4xf32>)
+      outs(%arg0 : tensor<264x4xf32>) {
+    ^bb0(%arg1: f32, %arg2: f32, %s: f32):  // no predecessors
       %2 = mulf %arg1, %arg2 : f32
       linalg.yield %2 : f32
     } -> tensor<264x4xf32>
   %1 = linalg.tensor_reshape %0 [#map1, #map2] :
     tensor<264x4xf32> into tensor<8x33x4xf32>
   return %1 : tensor<8x33x4xf32>
 }
 
 //  CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d0, d1)>
 //  CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d2)>
 //  CHECK-DAG: #[[MAP2:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 //      CHECK: func @generic_op_reshape_consumer_static
 // CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: tensor<264x4xf32>
 //      CHECK:   %[[T0:.+]] = linalg.tensor_reshape %[[ARG0]]
 // CHECK-SAME:     [#[[MAP0]], #[[MAP1]]]
 // CHECK-SAME:     tensor<264x4xf32> into tensor<8x33x4xf32>
-//      CHECK:   %[[T1:.+]] = linalg.generic
+//      CHECK:   %[[T1:.+]] = linalg.init_tensor [8, 33, 4] : tensor<8x33x4xf32>
+//      CHECK:   %[[T2:.+]] = linalg.generic
 // CHECK-SAME:     indexing_maps = [#[[MAP2]], #[[MAP2]]]
 // CHECK-SAME:     ["parallel", "parallel", "parallel"]
 // CHECK-SAME:     ins(%[[T0]] : tensor<8x33x4xf32>)
-//      CHECK:   return %[[T1]] : tensor<8x33x4xf32>
+// CHECK-SAME:     outs(%[[T1]] : tensor<8x33x4xf32>)
+//      CHECK:   return %[[T2]] : tensor<8x33x4xf32>
 
 // -----
 
-func @scalar_reshape(%arg0 : tensor<1x10xf32>, %arg1 : tensor<1xf32>)
-                     -> tensor<1x10xf32> {
+func @scalar_reshape(
+  %arg0 : tensor<1x10xf32>, %arg1 : tensor<1xf32>, %shape : tensor<10xf32>)
+    -> tensor<1x10xf32>
+{
   %0 = linalg.tensor_reshape %arg1 [] : tensor<1xf32> into tensor<f32>
   %1 = linalg.generic
     {indexing_maps = [affine_map<(d0) -> ()>, affine_map<(d0) -> (d0)>],
-     iterator_types = ["parallel"]} ins(%0 : tensor<f32>) {
-  ^bb0(%arg2: f32):  // no predecessors
+     iterator_types = ["parallel"]}
+     ins(%0 : tensor<f32>)
+    outs(%shape : tensor<10xf32>) {
+  ^bb0(%arg2: f32, %s: f32):  // no predecessors
     linalg.yield %arg2 : f32
   } -> tensor<10xf32>
   %2 = linalg.tensor_reshape %1 [affine_map<(d0, d1) -> (d0, d1)>]
     : tensor<10xf32> into tensor<1x10xf32>
   return %2 : tensor<1x10xf32>
 }
 
 //  CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1) -> ()>
 //  CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1) -> (d0, d1)>
 //      CHECK: func @scalar_reshape
 // CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: tensor<1x10xf32>
 // CHECK-SAME:   %[[ARG1:[a-zA-Z0-9_]+]]: tensor<1xf32>
 //      CHECK:   %[[T0:.+]] = linalg.tensor_reshape %[[ARG1]] []
 // CHECK-SAME:     tensor<1xf32> into tensor<f32>
-//      CHECK:   %[[T1:.+]] = linalg.generic
+//      CHECK:   %[[T1:.+]] = linalg.init_tensor [1, 10] : tensor<1x10xf32>
+//      CHECK:   %[[T2:.+]] = linalg.generic
 // CHECK-SAME:     indexing_maps = [#[[MAP0]], #[[MAP1]]]
 // CHECK-SAME:     iterator_types = ["parallel", "parallel"]
 // CHECK-SAME:     ins(%[[T0]] : tensor<f32>)
-//      CHECK:   return %[[T1]] : tensor<1x10xf32>
+// CHECK-SAME:     outs(%[[T1]] : tensor<1x10xf32>)
+//      CHECK:   return %[[T2]] : tensor<1x10xf32>
 
 // -----
 
 #map0 = affine_map<(d0, d1, d2) -> (d2, d0, d1)>
 #map1 = affine_map<(d0, d1, d2) -> (d1, d2, d0)>
 func @indexed_generic_op_reshape_producer_fusion(%arg0 : tensor<?x?x4x?xi32>,
                                          %arg1 : tensor<?x?x?xi32>) ->
                                          tensor<?x?x?xi32>
 {
   %0 = linalg.tensor_reshape %arg0 [affine_map<(i, j, k, l) -> (i)>,
                                     affine_map<(i, j, k, l) -> (j, k)>,
                                     affine_map<(i, j, k, l) -> (l)>] :
     tensor<?x?x4x?xi32> into tensor<?x?x?xi32>
   %1 = linalg.indexed_generic {
      indexing_maps = [#map0, #map1, #map1],
      iterator_types = ["parallel", "parallel", "parallel"]}
-      ins(%0, %arg1 : tensor<?x?x?xi32>, tensor<?x?x?xi32>) {
-    ^bb0(%arg3 : index, %arg4 : index, %arg5 : index, %arg6: i32, %arg7: i32):
+       ins(%0, %arg1 : tensor<?x?x?xi32>, tensor<?x?x?xi32>)
+      outs(%0 : tensor<?x?x?xi32>) {
+    ^bb0(%arg3 : index, %arg4 : index, %arg5 : index, %arg6: i32, %arg7: i32, %s: i32):
       %1 = muli %arg6, %arg7 : i32
       %2 = index_cast %arg3 : index to i32
       %3 = addi %1, %2 : i32
       %4 = index_cast %arg4 : index to i32
       %5 = addi %3, %4 : i32
       %6 = index_cast %arg5 : index to i32
       %7 = addi %5, %6 : i32
       linalg.yield %7 : i32
   } -> tensor<?x?x?xi32>
   return %1 : tensor<?x?x?xi32>
 }
 
 // The generic op version of the test check for the op structure. Only
 // checking the op body here.
 //       CHECK: #[[MAP:.+]] =  affine_map<(d0, d1) -> (d0 * 4 + d1)>
 //       CHECK: func @indexed_generic_op_reshape_producer_fusion
 //       CHECK:   linalg.indexed_generic
 //       CHECK:   ^{{.*}}(
 //  CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]+]]: index, %[[ARG3:[a-zA-Z0-9]+]]: index,
 //  CHECK-SAME:     %[[ARG4:[a-zA-Z0-9]+]]: index, %[[ARG5:[a-zA-Z0-9]+]]: index,
-//  CHECK-SAME:     %[[ARG6:[a-zA-Z0-9]+]]: i32, %[[ARG7:[a-zA-Z0-9]+]]: i32)
+//  CHECK-SAME:     %[[ARG6:[a-zA-Z0-9]+]]: i32, %[[ARG7:[a-zA-Z0-9]+]]: i32,
+//  CHECK-SAME:     %[[ARG8:[a-zA-Z0-9]+]]: i32)
 //       CHECK:     %[[T3:.+]] = affine.apply #[[MAP]](%[[ARG2]], %[[ARG3]])
 //       CHECK:     %[[T4:.+]] = muli %[[ARG6]], %[[ARG7]]
 //       CHECK:     %[[T5:.+]] = index_cast %[[T3]]
 //       CHECK:     %[[T6:.+]] = addi %[[T4]], %[[T5]]
 //       CHECK:     %[[T7:.+]] = index_cast %[[ARG4]]
 //       CHECK:     %[[T8:.+]] = addi %[[T6]], %[[T7]]
 //       CHECK:     %[[T9:.+]] = index_cast %[[ARG5]]
 //       CHECK:     %[[T10:.+]] = addi %[[T8]], %[[T9]]
 //       CHECK:     linalg.yield %[[T10]]
 
 // -----
 
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 func @indexed_generic_op_reshape_consumer_fusion(%arg0 : tensor<?x?xi32>,
                                          %arg1 : tensor<?x?xi32>) ->
                                          tensor<?x?x4x5xi32>
 {
   %0 = linalg.indexed_generic {
      indexing_maps = [#map0, #map0, #map0],
      iterator_types = ["parallel", "parallel"]}
-      ins(%arg0, %arg1 : tensor<?x?xi32>, tensor<?x?xi32>) {
-    ^bb0(%arg3 : index, %arg4 : index, %arg5: i32, %arg6: i32):       // no predecessors
+       ins(%arg0, %arg1 : tensor<?x?xi32>, tensor<?x?xi32>)
+      outs(%arg0 : tensor<?x?xi32>) {
+    ^bb0(%arg3 : index, %arg4 : index, %arg5: i32, %arg6: i32, %s: i32):       // no predecessors
       %1 = muli %arg5, %arg6 : i32
       %2 = index_cast %arg3 : index to i32
       %3 = addi %1, %2 : i32
       %4 = index_cast %arg4 : index to i32
       %5 = addi %3, %4 : i32
       linalg.yield %5 : i32
   } -> tensor<?x?xi32>
   %1 = linalg.tensor_reshape %0 [affine_map<(i, j, k, l) -> (i)>,
                                  affine_map<(i, j, k, l) -> (j, k, l)>] :
     tensor<?x?xi32> into tensor<?x?x4x5xi32>
   return %1 : tensor<?x?x4x5xi32>
 }
 // The generic op version of the test check for the op structure. Only
 // checking the op body here.
 //       CHECK: #[[MAP:.+]] = affine_map<(d0, d1, d2) -> (d0 * 20 + d1 * 5 + d2)>
 //       CHECK: func @indexed_generic_op_reshape_consumer_fusion
 //       CHECK:   linalg.indexed_generic
 //       CHECK:   ^{{.*}}(
 //  CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]+]]: index, %[[ARG3:[a-zA-Z0-9]+]]: index,
 //  CHECK-SAME:     %[[ARG4:[a-zA-Z0-9]+]]: index, %[[ARG5:[a-zA-Z0-9]+]]: index,
-//  CHECK-SAME:     %[[ARG6:[a-zA-Z0-9]+]]: i32, %[[ARG7:[a-zA-Z0-9]+]]: i32)
+//  CHECK-SAME:     %[[ARG6:[a-zA-Z0-9]+]]: i32, %[[ARG7:[a-zA-Z0-9]+]]: i32,
+//  CHECK-SAME:     %[[ARG8:[a-zA-Z0-9]+]]: i32)
 //       CHECK:     %[[T3:.+]] = affine.apply #[[MAP]](%[[ARG3]], %[[ARG4]], %[[ARG5]])
 //       CHECK:     %[[T4:.+]] = muli %[[ARG6]], %[[ARG7]]
 //       CHECK:     %[[T5:.+]] = index_cast %[[ARG2]]
 //       CHECK:     %[[T6:.+]] = addi %[[T4]], %[[T5]]
 //       CHECK:     %[[T7:.+]] = index_cast %[[T3]]
 //       CHECK:     %[[T8:.+]] = addi %[[T6]], %[[T7]]
 //       CHECK:     linalg.yield %[[T8]]
 
 // -----
 
 func @reshape_as_consumer_permutation
-  (%a : tensor<210x6x4xi32>, %b : tensor<210x4xi32>)
+  (%a : tensor<210x6x4xi32>, %b : tensor<210x4xi32>, %shape : tensor<6x4x210xi32>)
     -> tensor<2x3x4x5x6x7xi32> {
   %c = linalg.indexed_generic {
          indexing_maps = [affine_map<(d0, d1, d2) -> (d1, d0, d2)>,
                           affine_map<(d0, d1, d2) -> (d1, d2)>,
                           affine_map<(d0, d1, d2) -> (d0, d2, d1)>],
          iterator_types = ["parallel", "parallel", "parallel"]}
-         ins(%a, %b : tensor<210x6x4xi32>, tensor<210x4xi32>) {
-       ^bb0(%arg0 : index, %arg1 : index, %arg2 : index, %arg3 : i32, %arg4: i32):
+          ins(%a, %b : tensor<210x6x4xi32>, tensor<210x4xi32>)
+         outs(%shape : tensor<6x4x210xi32>) {
+       ^bb0(%arg0 : index, %arg1 : index, %arg2 : index, %arg3 : i32, %arg4: i32, %s: i32):
          %1 = addi %arg3, %arg4 : i32
          %2 = index_cast %arg0 : index to i32
          %3 = addi %1, %2 : i32
          %4 = index_cast %arg1 : index to i32
          %5 = addi %3, %4 : i32
          %6 = index_cast %arg2 : index to i32
          %7 = addi %5, %6 : i32
    linalg.yield %7 : i32

 //   CHECK-DAG: #[[MAP9:.+]] = affine_map<(d0, d1, d2, d3, d4, d5) -> (d0, d1, d5, d2, d3, d4)>
 //       CHECK: func @reshape_as_consumer_permutation
 //  CHECK-SAME:   %[[ARG0:.+]]: tensor<210x6x4xi32>
 //  CHECK-SAME:   %[[ARG1:.+]]: tensor<210x4xi32>
 //   CHECK-DAG:   %[[T0:.+]] = linalg.tensor_reshape %[[ARG0]]
 //  CHECK-SAME:     [#[[MAP0]], #[[MAP1]], #[[MAP2]]]
 //   CHECK-DAG:   %[[T1:.+]] = linalg.tensor_reshape %[[ARG1]]
 //  CHECK-SAME:     [#[[MAP3]], #[[MAP4]]]
-//       CHECK:   %[[T2:.+]] = linalg.indexed_generic
+//       CHECK:   %[[T2:.+]] = linalg.init_tensor [2, 3, 4, 5, 6, 7]
+//       CHECK:   %[[T3:.+]] = linalg.indexed_generic
 //  CHECK-SAME:     indexing_maps = [#[[MAP7]], #[[MAP8]], #[[MAP9]]]
-//  CHECK-SAME:     ins(%[[T0]], %[[T1]] : tensor<{{.+}}>, tensor<{{.+}}>)
+//  CHECK-SAME:     ins(%[[T0]], %[[T1]] : tensor<5x6x7x2x3x4xi32>, tensor<5x6x7x4xi32>)
+//  CHECK-SAME:     outs(%[[T2]] : tensor<2x3x4x5x6x7xi32>)
 //       CHECK:   ^{{.+}}(
 //  CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]+]]: index, %[[ARG3:[a-zA-Z0-9]+]]: index,
 //  CHECK-SAME:     %[[ARG4:[a-zA-Z0-9]+]]: index, %[[ARG5:[a-zA-Z0-9]+]]: index,
 //  CHECK-SAME:     %[[ARG6:[a-zA-Z0-9]+]]: index, %[[ARG7:[a-zA-Z0-9]+]]: index,
-//  CHECK-SAME:     %[[ARG8:[a-zA-Z0-9]+]]: i32, %[[ARG9:[a-zA-Z0-9]+]]: i32)
-//   CHECK-DAG:       %[[T3:.+]] = affine.apply #[[MAP5]](%[[ARG2]], %[[ARG3]])
-//   CHECK-DAG:       %[[T4:.+]] = affine.apply #[[MAP6]](%[[ARG4]], %[[ARG5]], %[[ARG6]])
-//   CHECK-DAG:       %[[T5:.+]] = addi %[[ARG8]], %[[ARG9]]
-//       CHECK:       %[[T6:.+]] = index_cast %[[T3]]
-//       CHECK:       %[[T7:.+]] = addi %[[T5]], %[[T6]]
-//       CHECK:       %[[T8:.+]] = index_cast %[[T4]]
-//       CHECK:       %[[T9:.+]] = addi %[[T7]], %[[T8]]
-//       CHECK:       %[[T10:.+]] = index_cast %[[ARG7]]
-//       CHECK:       %[[T11:.+]] = addi %[[T9]], %[[T10]]
+//  CHECK-SAME:     %[[ARG8:[a-zA-Z0-9]+]]: i32, %[[ARG9:[a-zA-Z0-9]+]]: i32,
+//  CHECK-SAME:     %[[ARG10:[a-zA-Z0-9]+]]: i32)
+//   CHECK-DAG:       %[[T4:.+]] = affine.apply #[[MAP5]](%[[ARG2]], %[[ARG3]])
+//   CHECK-DAG:       %[[T5:.+]] = affine.apply #[[MAP6]](%[[ARG4]], %[[ARG5]], %[[ARG6]])
+//   CHECK-DAG:       %[[T6:.+]] = addi %[[ARG8]], %[[ARG9]]
+//       CHECK:       %[[T7:.+]] = index_cast %[[T4]]
+//       CHECK:       %[[T8:.+]] = addi %[[T6]], %[[T7]]
+//       CHECK:       %[[T9:.+]] = index_cast %[[T5]]
+//       CHECK:       %[[T10:.+]] = addi %[[T8]], %[[T9]]
+//       CHECK:       %[[T11:.+]] = index_cast %[[ARG7]]
+//       CHECK:       %[[T12:.+]] = addi %[[T10]], %[[T11]]
 
 // -----
 
-func @reshape_as_producer_projected_permutation
-  (%arg0 : tensor<33x8x?xi32>) -> tensor<264x?x4xi32> {
+func @reshape_as_producer_projected_permutation(
+    %arg0 : tensor<33x8x?xi32>, %shape : tensor<264x?x4xi32>) -> tensor<264x?x4xi32>
+{
   %0 = linalg.tensor_reshape %arg0 [affine_map<(d0, d1, d2) -> (d0, d1)>,
                                     affine_map<(d0, d1, d2) -> (d2)>]
     : tensor<33x8x?xi32> into tensor<264x?xi32>
   %1 = linalg.indexed_generic
     {indexing_maps = [affine_map<(d0, d1, d2) -> (d0, d1)>,
                       affine_map<(d0, d1, d2) -> (d0, d1, d2)>],
-     iterator_types = ["parallel", "parallel", "parallel"]} ins(%0 : tensor<264x?xi32>) {
-  ^bb0(%arg1: index, %arg2: index, %arg3: index, %arg4: i32):  // no predecessors
+     iterator_types = ["parallel", "parallel", "parallel"]}
+     ins(%0 : tensor<264x?xi32>)
+    outs(%shape : tensor<264x?x4xi32>) {
+  ^bb0(%arg1: index, %arg2: index, %arg3: index, %arg4: i32, %s: i32):  // no predecessors
     %2 = index_cast %arg1 : index to i32
     %3 = addi %arg4, %2 : i32
     %4 = index_cast %arg2 : index to i32
     %5 = addi %3, %4 : i32
     %6 = index_cast %arg3 : index to i32
     %7 = addi %5, %6 : i32
     linalg.yield %7 : i32
   } -> tensor<264x?x4xi32>

 //       CHECK:   %[[RES:.+]] = linalg.indexed_generic
 //  CHECK-SAME:     indexing_maps = [#[[MAP0]], #[[MAP1]]]
 //  CHECK-SAME:     ins(%[[ARG0]] : tensor<33x8x?xi32>)
 //       CHECK:   ^{{.+}}(
 //  CHECK-SAME:     %[[ARG1:[a-zA-Z0-9]+]]: index,
 //  CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]+]]: index,
 //  CHECK-SAME:     %[[ARG3:[a-zA-Z0-9]+]]: index,
 //  CHECK-SAME:     %[[ARG4:[a-zA-Z0-9]+]]: index,
-//  CHECK-SAME:     %[[ARG5:[a-zA-Z0-9]+]]: i32)
+//  CHECK-SAME:     %[[ARG5:[a-zA-Z0-9]+]]: i32,
+//  CHECK-SAME:     %[[ARG7:[a-zA-Z0-9]+]]: i32)
 //       CHECK:       %[[T0:.+]] = affine.apply #[[MAP2]](%[[ARG1]], %[[ARG2]])
 //       CHECK:       %[[T1:.+]] = index_cast %[[T0]] : index to i32
 //       CHECK:       %[[T2:.+]] = addi %[[ARG5]], %[[T1]] : i32
 //       CHECK:       %[[T3:.+]] = index_cast %[[ARG3]] : index to i32
 //       CHECK:       %[[T4:.+]] = addi %[[T2]], %[[T3]] : i32
 //       CHECK:       %[[T5:.+]] = index_cast %[[ARG4]] : index to i32
 //       CHECK:       %[[T6:.+]] = addi %[[T4]], %[[T5]] : i32
 //       CHECK:       linalg.yield %[[T6]] : i32
 //       CHECK:    %[[RES2:.+]] = linalg.tensor_reshape %[[RES]]
 //  CHECK-SAME:      [#[[MAP3]], #[[MAP4]], #[[MAP5]]]
 //  CHECK-SAME:    : tensor<33x8x?x4xi32> into tensor<264x?x4xi32>
 //       CHECK:  return %[[RES2]] : tensor<264x?x4xi32>
 
 // -----
 
 #map0 = affine_map<(d0, d1) -> (d0, d1)>
 #map1 = affine_map<(d0, d1) -> (d1, d0)>
 func @generic_op_reshape_consumer_fusion_projected(%arg0 : tensor<?x?xf32>,
                                                    %arg1 : tensor<?x?xf32>) ->
                                                    tensor<?x?x4x5xf32>
 {
   %0 = linalg.generic {
      indexing_maps = [#map0, #map0, #map1],
      iterator_types = ["parallel", "parallel"]}
-      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
+       ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%arg0 : tensor<?x?xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %s: f32):       // no predecessors
       %1 = mulf %arg3, %arg4 : f32
       linalg.yield %1 : f32
   } -> tensor<?x?xf32>
   %1 = linalg.tensor_reshape %0 [affine_map<(i, j, k, l) -> (i)>,
                                  affine_map<(i, j, k, l) -> (j, k, l)>] :
     tensor<?x?xf32> into tensor<?x?x4x5xf32>
   return %1 : tensor<?x?x4x5xf32>
 }

mlir/test/Dialect/Linalg/reshape_linearization_fusion.mlir

 // RUN: mlir-opt -split-input-file -linalg-fold-reshape-ops-by-linearization %s | FileCheck %s
 
-
-// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1 * 4 + d2, d3)>
-// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
-
 #map0 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 func @generic_op_reshape_producer_fusion(%arg0 : tensor<?x?x?xf32>,
                                          %arg1 : tensor<?x?x4x?xf32>) ->
                                          tensor<?x?x4x?xf32>
 {
   %0 = linalg.tensor_reshape %arg0 [affine_map<(i, j, k, l) -> (i)>,
                                     affine_map<(i, j, k, l) -> (j, k)>,
                                     affine_map<(i, j, k, l) -> (l)>] :
     tensor<?x?x?xf32> into tensor<?x?x4x?xf32>
   %1 = linalg.generic {
-     indexing_maps = [#map0, #map0, #map0],
-     iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
-      ins(%0, %arg1 : tensor<?x?x4x?xf32>, tensor<?x?x4x?xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
+    ins(%0, %arg1 : tensor<?x?x4x?xf32>, tensor<?x?x4x?xf32>)
+    outs(%0 : tensor<?x?x4x?xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
       %1 = mulf %arg3, %arg4 : f32
       linalg.yield %1 : f32
   } -> tensor<?x?x4x?xf32>
   return %1 : tensor<?x?x4x?xf32>
 }
 
-// CHECK-LABEL: func @generic_op_reshape_producer_fusion
+//   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1 * 4 + d2, d3)>
+//   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
+//       CHECK: func @generic_op_reshape_producer_fusion
+//  CHECK-SAME:   %[[ARG0:.+]]: tensor<?x?x?xf32>
 //       CHECK: linalg.generic
 //  CHECK-SAME:   indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP1]]]
-//   CHECK-NOT: linalg.generic
-
+//  CHECK-SAME:   ins(%[[ARG0]], %{{.+}} : tensor<?x?x?xf32>, tensor<?x?x4x?xf32>)
+//  CHECK-SAME:   outs(%{{.+}} : tensor<?x?x4x?xf32>)
 
 // -----
 
-// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
-// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1 * 20 + d2 * 5 + d3)>
-
 #map0 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 func @generic_op_reshape_consumer_fusion(%arg0 : tensor<?x?x4x5xf32>,
                                          %arg1 : tensor<?x?x4x5xf32>) ->
                                          tensor<?x?xf32>
 {
   %0 = linalg.generic {
-     indexing_maps = [#map0, #map0, #map0],
-     iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
-      ins(%arg0, %arg1 : tensor<?x?x4x5xf32>, tensor<?x?x4x5xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
+    ins(%arg0, %arg1 : tensor<?x?x4x5xf32>, tensor<?x?x4x5xf32>)
+    outs(%arg0 : tensor<?x?x4x5xf32>){
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
       %1 = mulf %arg3, %arg4 : f32
       linalg.yield %1 : f32
   } -> tensor<?x?x4x5xf32>
   %1 = linalg.tensor_reshape %0 [affine_map<(i, j, k, l) -> (i)>,
                                  affine_map<(i, j, k, l) -> (j, k, l)>] :
     tensor<?x?x4x5xf32> into tensor<?x?xf32>
   return %1 : tensor<?x?xf32>
 }
 
-// CHECK-LABEL: func @generic_op_reshape_consumer_fusion
-//       CHECK: linalg.generic
-//  CHECK-SAME:   indexing_maps = [#[[$MAP0]], #[[$MAP0]], #[[$MAP1]]]
-//   CHECK-NOT: linalg.generic
+
+//   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
+//   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1 * 20 + d2 * 5 + d3)>
+//       CHECK: func @generic_op_reshape_consumer_fusion
+//  CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: tensor<?x?x4x5xf32>
+//   CHECK-DAG:   %[[C0:.+]] = constant 0 : index
+//   CHECK-DAG:   %[[C1:.+]] = constant 1 : index
+//   CHECK-DAG:   %[[C20:.+]] = constant 20 : index
+//       CHECK:   %[[T0:.+]] = dim %[[ARG0]], %[[C0]]
+//       CHECK:   %[[T1:.+]] = dim %[[ARG0]], %[[C1]]
+//       CHECK:   %[[T2:.+]] = muli %[[T1]], %[[C20]]
+//       CHECK:   %[[T3:.+]] = linalg.init_tensor [%[[T0]], %[[T2]]]
+//       CHECK:   linalg.generic
+//  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP0]], #[[$MAP1]]]
+//  CHECK-SAME:     outs(%[[T3]] : tensor<?x?xf32>)
 
 // -----
 
 #map0 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 func @generic_op_reshape_consumer_nofusion(%arg0 : tensor<?x?x?x5xf32>,
                                            %arg1 : tensor<?x?x?x5xf32>) ->
                                            tensor<?x?xf32>
 {
   %0 = linalg.generic {
      indexing_maps = [#map0, #map0, #map0],
      iterator_types = ["parallel", "parallel", "parallel", "parallel"]}
-      ins(%arg0, %arg1 : tensor<?x?x?x5xf32>, tensor<?x?x?x5xf32>) {
-    ^bb0(%arg3: f32, %arg4: f32):       // no predecessors
+      ins(%arg0, %arg1 : tensor<?x?x?x5xf32>, tensor<?x?x?x5xf32>)
+      outs(%arg0 : tensor<?x?x?x5xf32>) {
+    ^bb0(%arg3: f32, %arg4: f32, %arg5: f32):       // no predecessors
       %1 = mulf %arg3, %arg4 : f32
       linalg.yield %1 : f32
   } -> tensor<?x?x?x5xf32>
   %1 = linalg.tensor_reshape %0 [affine_map<(i, j, k, l) -> (i)>,
                                  affine_map<(i, j, k, l) -> (j, k, l)>] :
     tensor<?x?x?x5xf32> into tensor<?x?xf32>
   return %1 : tensor<?x?xf32>
 }
 
 // CHECK-LABEL: func @generic_op_reshape_consumer_nofusion
-//       CHECK: linalg.tensor_reshape
+//       CHECK:   %[[T0:.+]] = linalg.generic
+//       CHECK:   linalg.tensor_reshape %[[T0]]
 
 // -----
 
-
-// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1 * 4 + d2, d3)>
-// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
-
 #map0 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 func @indexed_generic_op_reshape_producer_fusion(%arg0 : tensor<?x?x?xi32>)
   -> tensor<?x?x4x?xi32> {
   %0 = linalg.tensor_reshape %arg0 [affine_map<(i, j, k, l) -> (i)>,
                                     affine_map<(i, j, k, l) -> (j, k)>,
                                     affine_map<(i, j, k, l) -> (l)>] :
     tensor<?x?x?xi32> into tensor<?x?x4x?xi32>
   %1 = linalg.indexed_generic {
     indexing_maps = [#map0, #map0],
     iterator_types = ["parallel", "parallel", "parallel", "parallel"] }
-    ins(%0 : tensor<?x?x4x?xi32>) {
-  ^bb0(%arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: i32):       // no predecessors
+    ins(%0 : tensor<?x?x4x?xi32>)
+    outs(%0 : tensor<?x?x4x?xi32>) {
+  ^bb0(%arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: i32, %arg7 : i32):       // no predecessors
     %2 = index_cast %arg2 : index to i32
     %3 = addi %arg6, %2 : i32
     linalg.yield %3 : i32
   } -> tensor<?x?x4x?xi32>
   return %1 : tensor<?x?x4x?xi32>
 }
 
-// CHECK-LABEL: func @indexed_generic_op_reshape_producer_fusion
-//   CHECK-NOT: linalg.tensor_reshape
-//       CHECK: linalg.indexed_generic
-//  CHECK-SAME:   indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
-//   CHECK-NOT: linalg.tensor_reshape
+//   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1 * 4 + d2, d3)>
+//   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
+//       CHECK: func @indexed_generic_op_reshape_producer_fusion
+//  CHECK-SAME:   %[[ARG0:.+]]: tensor<?x?x?xi32>
+//       CHECK:   linalg.indexed_generic
+//  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
+//  CHECK-SAME:     ins(%[[ARG0]] : tensor<?x?x?xi32>)
 
 // -----
 
-// CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
-// CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1 * 20 + d2 * 5 + d3)>
-
 #map0 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
 func @indexed_generic_op_reshape_consumer_fusion(%arg0 : tensor<?x?x4x5xi32>)
   -> tensor<?x?xi32> {
   %0 = linalg.indexed_generic {
     indexing_maps = [#map0, #map0],
     iterator_types = ["parallel", "parallel", "parallel", "parallel"] }
-    ins(%arg0 : tensor<?x?x4x5xi32>) {
-  ^bb0(%arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: i32):       // no predecessors
+    ins(%arg0 : tensor<?x?x4x5xi32>) outs(%arg0 : tensor<?x?x4x5xi32>) {
+  ^bb0(%arg2: index, %arg3: index, %arg4: index, %arg5: index, %arg6: i32, %arg7: i32):       // no predecessors
     %2 = index_cast %arg2 : index to i32
     %3 = addi %arg6, %2 : i32
     linalg.yield %3 : i32
   } -> tensor<?x?x4x5xi32>
   %1 = linalg.tensor_reshape %0 [affine_map<(i, j, k, l) -> (i)>,
                                  affine_map<(i, j, k, l) -> (j, k, l)>] :
     tensor<?x?x4x5xi32> into tensor<?x?xi32>
   return %1 : tensor<?x?xi32>
 }
 
+//   CHECK-DAG: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
+//   CHECK-DAG: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1 * 20 + d2 * 5 + d3)>
 // CHECK-LABEL: func @indexed_generic_op_reshape_consumer_fusion
-//   CHECK-NOT: linalg.tensor_reshape
-//       CHECK: linalg.indexed_generic
-//  CHECK-SAME:   indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
-//   CHECK-NOT: linalg.tensor_reshape
+//  CHECK-SAME:   %[[ARG0:[a-zA-Z0-9_]+]]: tensor<?x?x4x5xi32>
+//   CHECK-DAG:   %[[C0:.+]] = constant 0 : index
+//   CHECK-DAG:   %[[C1:.+]] = constant 1 : index
+//   CHECK-DAG:   %[[C20:.+]] = constant 20 : index
+//       CHECK:   %[[T0:.+]] = dim %[[ARG0]], %[[C0]]
+//       CHECK:   %[[T1:.+]] = dim %[[ARG0]], %[[C1]]
+//       CHECK:   %[[T2:.+]] = muli %[[T1]], %[[C20]]
+//       CHECK:   %[[T3:.+]] = linalg.init_tensor [%[[T0]], %[[T2]]]
+//       CHECK:   linalg.indexed_generic
+//  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
+//  CHECK-SAME:     outs(%[[T3]] : tensor<?x?xi32>)
+//   CHECK-NOT:   linalg.tensor_reshape
 
 // -----
 
-// CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d0, d1 + d2 * 7)>
-// CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
-
 #map0 = affine_map<(d0, d1, d2) -> (d0)>
 #map1 = affine_map<(d0, d1, d2) -> (d1, d2)>
 #map2 = affine_map<(d0, d1, d2) -> (d0, d2, d1)>
 #map3 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 func @generic_op_021_permultation_reshape_producer_fusion(%arg0 : tensor<3x35xf32>) -> tensor<3x7x5xf32> {
   %0 = linalg.tensor_reshape %arg0 [#map0, #map1] : tensor<3x35xf32> into tensor<3x5x7xf32>
-  %1 = linalg.generic {indexing_maps = [#map2, #map3], iterator_types = ["parallel", "parallel", "parallel"]} ins(%0 : tensor<3x5x7xf32>) {
-    ^bb0(%arg2: f32):  // no predecessors
+  %1 = linalg.init_tensor [3, 7, 5] : tensor<3x7x5xf32>
+  %2 = linalg.generic
+    {indexing_maps = [#map2, #map3],
+     iterator_types = ["parallel", "parallel", "parallel"]}
+    ins(%0 : tensor<3x5x7xf32>) outs(%1 : tensor<3x7x5xf32>) {
+    ^bb0(%arg2: f32, %arg3 : f32):  // no predecessors
       linalg.yield %arg2 : f32
     } -> tensor<3x7x5xf32>
-    return %1 : tensor<3x7x5xf32>
+    return %2 : tensor<3x7x5xf32>
 }
 
+// CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d0, d1 + d2 * 7)>
+// CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 // CHECK-LABEL: func @generic_op_021_permultation_reshape_producer_fusion
-//   CHECK-NOT: linalg.tensor_reshape
-//       CHECK: linalg.generic
-//  CHECK-SAME:   indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
-//   CHECK-NOT: linalg.tensor_reshape
+//   CHECK-NOT:   linalg.tensor_reshape
+//       CHECK:   linalg.generic
+//  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
 
 // -----
 
-// CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d2, d0 * 7 + d1)>
-// CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
-
 #map0 = affine_map<(d0, d1, d2) -> (d0)>
 #map1 = affine_map<(d0, d1, d2) -> (d1, d2)>
 #map2 = affine_map<(d0, d1, d2) -> (d1, d2, d0)>
 #map3 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 func @generic_op_120_permultation_reshape_producer_fusion(%arg0 : tensor<3x35xf32>) -> tensor<5x7x3xf32> {
   %0 = linalg.tensor_reshape %arg0 [#map0, #map1] : tensor<3x35xf32> into tensor<3x5x7xf32>
-  %1 = linalg.generic {indexing_maps = [#map2, #map3], iterator_types = ["parallel", "parallel", "parallel"]} ins(%0 : tensor<3x5x7xf32>) {
-    ^bb0(%arg2: f32):  // no predecessors
+  %1 = linalg.init_tensor [5, 7, 3] : tensor<5x7x3xf32>
+  %2 = linalg.generic
+    {indexing_maps = [#map2, #map3],
+     iterator_types = ["parallel", "parallel", "parallel"]}
+    ins(%0 : tensor<3x5x7xf32>) outs(%1 : tensor<5x7x3xf32>) {
+    ^bb0(%arg2: f32, %arg3: f32):  // no predecessors
       linalg.yield %arg2 : f32
     } -> tensor<5x7x3xf32>
-    return %1 : tensor<5x7x3xf32>
+    return %2 : tensor<5x7x3xf32>
 }
 
-// CHECK-LABEL: func @generic_op_120_permultation_reshape_producer_fusion
-//   CHECK-NOT: linalg.tensor_reshape
-//       CHECK: linalg.generic
-//  CHECK-SAME:   indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
-//   CHECK-NOT: linalg.tensor_reshape
+//   CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d2, d0 * 7 + d1)>
+//   CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
+//       CHECK: func @generic_op_120_permultation_reshape_producer_fusion
+//   CHECK-NOT:   linalg.tensor_reshape
+//       CHECK:   linalg.generic
+//  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
 
 // -----
 
-// CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d1, d0 * 7 + d2)>
-// CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
-
 #map0 = affine_map<(d0, d1, d2) -> (d0)>
 #map1 = affine_map<(d0, d1, d2) -> (d1, d2)>
 #map2 = affine_map<(d0, d1, d2) -> (d1, d0, d2)>
 #map3 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 func @generic_op_102_permultation_reshape_producer_fusion(%arg0 : tensor<3x35xf32>) -> tensor<5x3x7xf32> {
   %0 = linalg.tensor_reshape %arg0 [#map0, #map1] : tensor<3x35xf32> into tensor<3x5x7xf32>
-  %1 = linalg.generic {indexing_maps = [#map2, #map3], iterator_types = ["parallel", "parallel", "parallel"]} ins(%0 : tensor<3x5x7xf32>) {
-    ^bb0(%arg2: f32):  // no predecessors
+  %1 = linalg.init_tensor [5, 3, 7] : tensor<5x3x7xf32>
+  %2 = linalg.generic
+    {indexing_maps = [#map2, #map3],
+     iterator_types = ["parallel", "parallel", "parallel"]}
+    ins(%0 : tensor<3x5x7xf32>) outs(%1 : tensor<5x3x7xf32>) {
+    ^bb0(%arg2: f32, %arg3: f32):  // no predecessors
       linalg.yield %arg2 : f32
     } -> tensor<5x3x7xf32>
-    return %1 : tensor<5x3x7xf32>
+    return %2 : tensor<5x3x7xf32>
 }
 
-// CHECK-LABEL: func @generic_op_102_permultation_reshape_producer_fusion
-//   CHECK-NOT: linalg.tensor_reshape
-//       CHECK: linalg.generic
-//  CHECK-SAME:   indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
-//   CHECK-NOT: linalg.tensor_reshape
+
+//   CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d1, d0 * 7 + d2)>
+//   CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
+//       CHECK: func @generic_op_102_permultation_reshape_producer_fusion
+//   CHECK-NOT:   linalg.tensor_reshape
+//       CHECK:   linalg.generic
+//  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
 
 // -----
 
-// CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
-// CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d1, d0 * 7 + d2)>
-
-
 #map0 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
 #map1 = affine_map<(d0, d1, d2) -> (d1, d0, d2)>
 #map2 = affine_map<(d0, d1, d2) -> (d0)>
 #map3 = affine_map<(d0, d1, d2) -> (d1, d2)>
 func @generic_op_102_permultation_reshape_consumer_fusion(%arg0 : tensor<3x5x7xf32>) -> tensor<5x21xf32> {
-  %0 = linalg.generic {indexing_maps = [#map0, #map1], iterator_types = ["parallel", "parallel", "parallel"]} ins(%arg0 : tensor<3x5x7xf32>) {
-    ^bb0(%arg2: f32):  // no predecessors
+  %0 = linalg.init_tensor [5, 3, 7] : tensor<5x3x7xf32>
+  %1 = linalg.generic
+    {indexing_maps = [#map0, #map1],
+     iterator_types = ["parallel", "parallel", "parallel"]}
+    ins(%arg0 : tensor<3x5x7xf32>) outs(%0 : tensor<5x3x7xf32>) {
+    ^bb0(%arg2: f32, %arg3 : f32):  // no predecessors
       linalg.yield %arg2 : f32
   } -> tensor<5x3x7xf32>
-  %1 = linalg.tensor_reshape %0 [#map2, #map3] : tensor<5x3x7xf32> into tensor<5x21xf32>
-  return %1 : tensor<5x21xf32>
+  %2 = linalg.tensor_reshape %1 [#map2, #map3] : tensor<5x3x7xf32> into tensor<5x21xf32>
+  return %2 : tensor<5x21xf32>
 }
 
-// CHECK-LABEL: func @generic_op_102_permultation_reshape_consumer_fusion
-//   CHECK-NOT: linalg.tensor_reshape
-//       CHECK: linalg.generic
-//  CHECK-SAME:   indexing_maps = [#[[$MAP0]], #[[$MAP1]]]
-//   CHECK-NOT: linalg.tensor_reshape
+
+//   CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
+//   CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0, d1, d2) -> (d1, d0 * 7 + d2)>
+//       CHECK: func @generic_op_102_permultation_reshape_consumer_fusion
+//   CHECK-NOT:   linalg.tensor_reshape
+//       CHECK:   linalg.generic
+//  CHECK-SAME:     indexing_maps = [#[[$MAP0]], #[[$MAP1]]]

mlir/test/Dialect/Linalg/roundtrip.mlir

#trait = {
   indexing_maps = #accesses,
   iterator_types = ["parallel", "parallel", "parallel"],
   library_call = "some_external_function_name_1"
 }
 
 func @generic(%arg0: memref<?x?xvector<3x4xi4>, offset: ?, strides: [?, 1]>,
               %arg1: memref<?x?x?xf32, offset: ?, strides: [?, ?, 1]>) {
   linalg.generic #trait
        ins(%arg0 : memref<?x?xvector<3x4xi4>, offset: ?, strides: [?, 1]>)
       outs(%arg1 : memref<?x?x?xf32, offset: ?, strides: [?, ?, 1]>)
       attrs = {foo = 1} {
     ^bb(%0: vector<3x4xi4>, %1: f32) :
       %f0 = constant 0.0 : f32
       linalg.yield %f0 : f32
   }
   return
 }
 // CHECK-LABEL: func @generic
 //       CHECK:   linalg.generic {
 //  CHECK-SAME:     indexing_maps = [#{{[0-9a-z]*}}, #{{[0-9a-z]*}}],
 //  CHECK-SAME:     iterator_types = ["parallel", "parallel", "parallel"],
 //  CHECK-SAME:     library_call = "some_external_function_name_1"}
-//  CHECK-SAME:     ins({{.*}} : memref<?x?xvector<3x4xi4>, #[[$strided2D]]>)
+//  CHECK-SAME:      ins({{.*}} : memref<?x?xvector<3x4xi4>, #[[$strided2D]]>)
 //  CHECK-SAME:     outs({{.*}} : memref<?x?x?xf32, #[[$strided3D]]>)
 //  CHECK-SAME:     {foo = 1 : i64}
 
 func @generic_with_tensor_input(%arg0: tensor<?x?xvector<3x4xi4>>,
                                 %arg1: memref<?x?x?xf32, offset: ?, strides: [?, ?, 1]>) {
   linalg.generic #trait
        ins(%arg0 : tensor<?x?xvector<3x4xi4>>)
       outs(%arg1 : memref<?x?x?xf32, offset: ?, strides: [?, ?, 1]>)
       attrs = {foo = 1} {
     ^bb(%0: vector<3x4xi4>, %1: f32) :
       %f0 = constant 0.0 : f32
       linalg.yield %f0 : f32
   }
   return
 }

 }
 
 // CHECK-LABEL: func @generic_without_inputs
 //       CHECK:   linalg.generic
 //   CHECK-NOT:     ins
 
 // -----
 
-#accesses = [
+#accesses2 = [
   affine_map<(i, j, k) -> (j, i)>,
   affine_map<(i, j, k) -> (i, k, i + j)>,
   affine_map<(i, j, k) -> (i, k, i + j)>
 ]
 
 #trait2 = {
-  indexing_maps = #accesses,
+  indexing_maps = #accesses2,
   iterator_types = ["parallel", "parallel", "parallel"],
   library_call = "some_external_function_name_1"
 }
 
 func @generic_with_tensor_input_and_output(
     %arg0: tensor<?x?xvector<3x4xi4>>, %arg1: tensor<?x?x?xf32>)
     -> (tensor<?x?x?xf32>) {
   %0 = linalg.generic #trait2
        ins(%arg0, %arg1 : tensor<?x?xvector<3x4xi4>>, tensor<?x?x?xf32>)
+      outs(%arg1 : tensor<?x?x?xf32>)
       attrs = {foo = 1} {
-    ^bb(%0: vector<3x4xi4>, %1: f32) :
+    ^bb(%0: vector<3x4xi4>, %1: f32, %2: f32) :
       %f0 = constant 0.0 : f32
       linalg.yield %f0 : f32
   } -> tensor<?x?x?xf32>
   return %0 : tensor<?x?x?xf32>
 }
 // CHECK-LABEL: func @generic_with_tensor_input_and_output
 //       CHECK:   linalg.generic {
 //  CHECK-SAME:     indexing_maps = [#{{.*}}, #{{.*}}], iterator_types = ["parallel", "parallel", "parallel"],
 //  CHECK-SAME:     library_call = "some_external_function_name_1"}
-//  CHECK-SAME:     ins({{.*}} : tensor<?x?xvector<3x4xi4>>, tensor<?x?x?xf32>)
+//  CHECK-SAME:      ins({{.*}} : tensor<?x?xvector<3x4xi4>>, tensor<?x?x?xf32>)
+//  CHECK-SAME:     outs({{.*}} : tensor<?x?x?xf32>)
 //  CHECK-SAME:     {foo = 1 : i64}
 //       CHECK:     -> tensor<?x?x?xf32>
 //       CHECK:   return {{.*}} : tensor<?x?x?xf32>
 
 // -----
 
-#accesses = [
+#accesses3 = [
   affine_map<(i, j, k) -> (j, i)>,
   affine_map<(i, j, k) -> (i, k, i + j)>,
   affine_map<(i, j, k) -> (i, k, i + j)>
 ]
 
-#trait2 = {
-  indexing_maps = #accesses,
+#trait3 = {
+  indexing_maps = #accesses3,
   iterator_types = ["parallel", "parallel", "parallel"],
   library_call = "some_external_function_name_1"
 }
 
 func @indexed_generic_with_tensor_input_and_output(
     %arg0: tensor<?x?xvector<3x4xi4>>, %arg1: tensor<?x?x?xf32>)
     -> (tensor<?x?x?xf32>) {
-  %0 = linalg.indexed_generic #trait2
+  %0 = linalg.indexed_generic #trait3
        ins(%arg0, %arg1 : tensor<?x?xvector<3x4xi4>>, tensor<?x?x?xf32>)
+      outs(%arg1 : tensor<?x?x?xf32>)
       attrs = {foo = 1} {
-    ^bb(%i: index, %j: index, %k: index, %0: vector<3x4xi4>, %1: f32) :
+    ^bb(%i: index, %j: index, %k: index, %0: vector<3x4xi4>, %1: f32, %2: f32) :
       %f0 = constant 0.0 : f32
       linalg.yield %f0 : f32
   } -> tensor<?x?x?xf32>
   return %0 : tensor<?x?x?xf32>
 }
 // CHECK-LABEL: func @indexed_generic_with_tensor_input_and_output
 //       CHECK:   linalg.indexed_generic {
 //  CHECK-SAME:     indexing_maps = [#{{.*}}, #{{.*}}], iterator_types = ["parallel", "parallel", "parallel"],
 //  CHECK-SAME:     library_call = "some_external_function_name_1"}
-//  CHECK-SAME:     ins({{.*}} : tensor<?x?xvector<3x4xi4>>, tensor<?x?x?xf32>)
+//  CHECK-SAME:      ins({{.*}} : tensor<?x?xvector<3x4xi4>>, tensor<?x?x?xf32>)
+//  CHECK-SAME:     outs({{.*}} : tensor<?x?x?xf32>)
 //  CHECK-SAME:     {foo = 1 : i64}
 //       CHECK:     -> tensor<?x?x?xf32>
 //       CHECK:   return {{.*}} : tensor<?x?x?xf32>
 
 // -----
 
 #broadcast_access = [
   affine_map<(i, j) -> ()>,
   affine_map<(i, j) -> (i, j)>
 ]
 
 #trait_broadcast = {
   indexing_maps = #broadcast_access,
   iterator_types = ["parallel", "parallel"],
   library_call = "some_broadcast_external_fn"
 }
 
-func @generic_op_zero_rank(%arg0: tensor<f32>) -> (tensor<3x4xf32>)
+func @generic_op_zero_rank(%arg0: tensor<f32>, %arg1 : tensor<3x4xf32>) -> (tensor<3x4xf32>)
 {
   %0 = linalg.generic #trait_broadcast
-      ins(%arg0 : tensor<f32>) {
-    ^bb(%a: f32) :
+       ins(%arg0 : tensor<f32>)
+      outs(%arg1 : tensor<3x4xf32>) {
+    ^bb(%a: f32, %b: f32) :
       linalg.yield %a : f32
   } -> tensor<3x4xf32>
   return %0 : tensor<3x4xf32>
 }
 
-func @indexed_generic_op_zero_rank(%arg0: tensor<f32>) -> (tensor<3x4xf32>)
+func @indexed_generic_op_zero_rank(%arg0: tensor<f32>, %arg1 : tensor<3x4xf32>) -> (tensor<3x4xf32>)
 {
   %0 = linalg.indexed_generic #trait_broadcast
-      ins(%arg0 : tensor<f32>) {
-    ^bb(%i: index, %j: index, %a: f32) :
+       ins(%arg0 : tensor<f32>)
+      outs(%arg1 : tensor<3x4xf32>) {
+    ^bb(%i: index, %j: index, %a: f32, %b: f32) :
       linalg.yield %a : f32
   } -> tensor<3x4xf32>
   return %0 : tensor<3x4xf32>
 }
 
 // -----
 
 // CHECK-DAG: #[[$strided2D:.*]] = affine_map<(d0, d1)[s0, s1] -> (d0 * s1 + s0 + d1)>
 // CHECK-DAG: #[[$strided3D:.*]] = affine_map<(d0, d1, d2)[s0, s1, s2] -> (d0 * s1 + s0 + d1 * s2 + d2)>
 
 #accesses = [
   affine_map<(i, j, k) -> (j, i)>,
   affine_map<(i, j, k) -> (i, k, i + j)>
 ]
 
 #trait3 = {
   indexing_maps = #accesses,
   iterator_types = ["parallel", "parallel", "parallel"],
   library_call = "some_external_function_name_2"
 }
 
 func @generic_region(%arg0: memref<?x?xvector<3x4xi4>, offset: ?, strides: [?, 1]>,
                      %arg1: memref<?x?x?xf32, offset: ?, strides: [?, ?, 1]>) {
   linalg.generic #trait3
        ins(%arg0 : memref<?x?xvector<3x4xi4>, offset: ?, strides: [?, 1]>)
       outs(%arg1 : memref<?x?x?xf32, offset: ?, strides: [?, ?, 1]>)
       attrs = {foo = 1} {
     ^bb(%a: vector<3x4xi4>, %b: f32) :
       linalg.yield %b : f32
   }
   return
 }
 // CHECK-LABEL: func @generic_region
 //       CHECK:   linalg.generic {
 //  CHECK-SAME:     indexing_maps = [#{{[0-9a-z]*}}, #{{[0-9a-z]*}}],
 //  CHECK-SAME:     iterator_types = ["parallel", "parallel", "parallel"],
 //  CHECK-SAME:     library_call = "some_external_function_name_2"
-//  CHECK-SAME:     ins({{.*}} : memref<?x?xvector<3x4xi4>, #[[$strided2D]]>)
+//  CHECK-SAME:      ins({{.*}} : memref<?x?xvector<3x4xi4>, #[[$strided2D]]>)
 //  CHECK-SAME:     outs({{.*}} : memref<?x?x?xf32, #[[$strided3D]]>)
 //  CHECK-SAME:     attrs = {foo = 1 : i64} {
 //       CHECK:  ^{{.*}}(%{{.*}}: vector<3x4xi4>, %{{.*}}: f32):
 //       CHECK:    linalg.yield %{{.*}} : f32
 
 func @indexed_generic(%arg0: memref<?x?xvector<3x4xi4>, offset: ?, strides: [?, 1]>,
                       %arg1: memref<?x?x?xf32, offset: ?, strides: [?, ?, 1]>) {
   linalg.indexed_generic #trait3
        ins(%arg0 : memref<?x?xvector<3x4xi4>, offset: ?, strides: [?, 1]>)
       outs(%arg1 : memref<?x?x?xf32, offset: ?, strides: [?, ?, 1]>)
       attrs = {foo = 1} {
     ^bb(%i: index, %j: index, %k: index, %a: vector<3x4xi4>, %b: f32) :
       linalg.yield %b : f32
   }
   return
 }
 // CHECK-LABEL: func @indexed_generic
%r3 = linalg.reshape %3 [affine_map<(i, j, k, l, m) -> (i, j)>,
                            affine_map<(i, j, k, l, m) -> (l, m)>] :
     memref<1x3x4x1x5xf32> into memref<3x4x5xf32>
   // Reshapes on tensors.
   %t0 = linalg.tensor_reshape %arg1 [affine_map<(i, j, k, l, m) -> (i, j)>,
                                      affine_map<(i, j, k, l, m) -> (k)>,
                                      affine_map<(i, j, k, l, m) -> (l, m)>] :
     tensor<3x4x5xf32> into tensor<1x3x4x1x5xf32>
   %rt0 = linalg.tensor_reshape %t0 [affine_map<(i, j, k, l, m) -> (i, j)>,
                                     affine_map<(i, j, k, l, m) -> (k)>,
                                     affine_map<(i, j, k, l, m) -> (l, m)>] :
     tensor<1x3x4x1x5xf32> into tensor<3x4x5xf32>
   %t1 = linalg.tensor_reshape %arg2 [affine_map<(i, j, k, l, m) -> (i, j)>,
                                      affine_map<(i, j, k, l, m) -> (k)>,
                                      affine_map<(i, j, k, l, m) -> (l, m)>] :
     tensor<3x?x5xf32> into tensor<1x3x?x1x5xf32>
   %rt1 = linalg.tensor_reshape %t1 [affine_map<(i, j, k, l, m) -> (i)>,
                                     affine_map<(i, j, k, l, m) -> (j, k)>,
                                     affine_map<(i, j, k, l, m) -> (l, m)>] :

 func @named_ops(%a3: memref<?x?x?xf32>, %b3: memref<?x?x?xf32>, %c3: memref<?x?x?xf32>,
                 %ta3: tensor<?x?x?xf32>, %tb3: tensor<?x?x?xf32>, %tc3: tensor<?x?x?xf32>)
   -> (tensor<?x?x?xf32>, tensor<?x?x?xf32>)
 {
   linalg.batch_matmul ins(%a3, %b3: memref<?x?x?xf32>, memref<?x?x?xf32>)
                      outs(%c3: memref<?x?x?xf32>)
   linalg.batch_matmul ins(%ta3, %tb3: tensor<?x?x?xf32>, tensor<?x?x?xf32>)
                      outs(%c3: memref<?x?x?xf32>)
-  %res1 = linalg.batch_matmul ins(%ta3, %tb3: tensor<?x?x?xf32>, tensor<?x?x?xf32>)
-                     init(%tc3: tensor<?x?x?xf32>)
+  %res1 = linalg.batch_matmul
+                      ins(%ta3, %tb3: tensor<?x?x?xf32>, tensor<?x?x?xf32>)
+                     outs(%tc3: tensor<?x?x?xf32>)
                   -> tensor<?x?x?xf32>
-  %res2 = linalg.batch_matmul ins(%ta3, %b3: tensor<?x?x?xf32>, memref<?x?x?xf32>)
-                     init(%tc3: tensor<?x?x?xf32>)
+  %res2 = linalg.batch_matmul
+                      ins(%ta3, %b3: tensor<?x?x?xf32>, memref<?x?x?xf32>)
+                     outs(%tc3: tensor<?x?x?xf32>)
                   -> tensor<?x?x?xf32>
   return %res1, %res2 : tensor<?x?x?xf32>, tensor<?x?x?xf32>
 }
 // CHECK-LABEL: func @named_ops
 //       CHECK:   linalg.batch_matmul
 //       CHECK:   linalg.batch_matmul
 //       CHECK:   linalg.batch_matmul
 //       CHECK:   linalg.batch_matmul

mlir/test/Dialect/Linalg/sparse_1d.mlir

 // CHECK:             %[[VAL_9:.*]] = addf %[[VAL_8]], %[[VAL_1]] : f32
 // CHECK:             store %[[VAL_9]], %[[VAL_6]]{{\[}}%[[VAL_7]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_10:.*]] = tensor_load %[[VAL_6]] : memref<32xf32>
 // CHECK:           return %[[VAL_10]] : tensor<32xf32>
 // CHECK:         }
 func @add_d(%arga: tensor<32xf32>, %argb: f32) -> tensor<32xf32> {
   %0 = linalg.generic #trait_d
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32xf32>)
+    outs(%arga: tensor<32xf32>) {
+      ^bb(%a: f32, %s : f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // CHECK-LABEL:   func @mul_d(
 // CHECK-SAME:                %[[VAL_0:.*]]: tensor<32xf32>,
 // CHECK-SAME:                %[[VAL_1:.*]]: f32) -> tensor<32xf32> {
 // CHECK:           %[[VAL_2:.*]] = constant 32 : index
 // CHECK:           %[[VAL_3:.*]] = constant 0 : index
 // CHECK:           %[[VAL_4:.*]] = constant 1 : index
 // CHECK:           %[[VAL_5:.*]] = alloca() : memref<32xf32>
 // CHECK:           %[[VAL_6:.*]] = alloca() : memref<32xf32>
 // CHECK:           scf.for %[[VAL_7:.*]] = %[[VAL_3]] to %[[VAL_2]] step %[[VAL_4]] {
 // CHECK:             %[[VAL_8:.*]] = load %[[VAL_5]]{{\[}}%[[VAL_7]]] : memref<32xf32>
 // CHECK:             %[[VAL_9:.*]] = mulf %[[VAL_8]], %[[VAL_1]] : f32
 // CHECK:             store %[[VAL_9]], %[[VAL_6]]{{\[}}%[[VAL_7]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_10:.*]] = tensor_load %[[VAL_6]] : memref<32xf32>
 // CHECK:           return %[[VAL_10]] : tensor<32xf32>
 // CHECK:         }
 func @mul_d(%arga: tensor<32xf32>, %argb: f32) -> tensor<32xf32> {
   %0 = linalg.generic #trait_d
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32xf32>)
+    outs(%arga: tensor<32xf32>) {
+      ^bb(%a: f32, %s : f32):
         %0 = mulf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 #trait_s = {
   indexing_maps = [

 // CHECK:           scf.for %[[VAL_27:.*]] = %[[VAL_28:.*]]#1 to %[[VAL_3]] step %[[VAL_6]] {
 // CHECK:             store %[[VAL_1]], %[[VAL_10]]{{\[}}%[[VAL_27]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_29:.*]] = tensor_load %[[VAL_10]] : memref<32xf32>
 // CHECK:           return %[[VAL_29]] : tensor<32xf32>
 // CHECK:         }
 func @add_s(%arga: tensor<32xf32>, %argb: f32) -> tensor<32xf32> {
   %0 = linalg.generic #trait_s
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32xf32>)
+    outs(%arga: tensor<32xf32>) {
+      ^bb(%a: f32, %s : f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // CHECK-LABEL:   func @repeated_add_s(
 // CHECK-SAME:                         %[[VAL_0:.*]]: tensor<32xf32>) -> tensor<32xf32> {

 // CHECK:             %[[VAL_18:.*]] = addf %[[VAL_14]], %[[VAL_17]] : f32
 // CHECK:             store %[[VAL_18]], %[[VAL_7]]{{\[}}%[[VAL_11]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_19:.*]] = tensor_load %[[VAL_7]] : memref<32xf32>
 // CHECK:           return %[[VAL_19]] : tensor<32xf32>
 // CHECK:         }
 func @repeated_add_s(%arga: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_s
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32xf32>)
+    outs(%arga: tensor<32xf32>) {
+      ^bb(%a: f32, %s : f32):
         %0 = addf %a, %a  : f32  // same tensor
         %1 = addf %a, %a  : f32  // should yield
         %2 = addf %0, %1  : f32  // one guard
         linalg.yield %2 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 

 // CHECK:             %[[VAL_14:.*]] = mulf %[[VAL_13]], %[[VAL_1]] : f32
 // CHECK:             store %[[VAL_14]], %[[VAL_8]]{{\[}}%[[VAL_12]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_15:.*]] = tensor_load %[[VAL_8]] : memref<32xf32>
 // CHECK:           return %[[VAL_15]] : tensor<32xf32>
 // CHECK:         }
 func @mul_s(%arga: tensor<32xf32>, %argb: f32) -> tensor<32xf32> {
   %0 = linalg.generic #trait_s
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32xf32>)
+    outs(%arga: tensor<32xf32>) {
+      ^bb(%a: f32, %s : f32):
         %0 = mulf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 #trait_dd = {
   indexing_maps = [

 // CHECK:             %[[VAL_11:.*]] = addf %[[VAL_9]], %[[VAL_10]] : f32
 // CHECK:             store %[[VAL_11]], %[[VAL_7]]{{\[}}%[[VAL_8]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_12:.*]] = tensor_load %[[VAL_7]] : memref<32xf32>
 // CHECK:           return %[[VAL_12]] : tensor<32xf32>
 // CHECK:         }
 func @add_dd(%arga: tensor<32xf32>, %argb: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_dd
-    ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>)
+    outs(%arga : tensor<32xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // CHECK-LABEL:   func @mul_dd(
 // CHECK-SAME:                 %[[VAL_0:.*]]: tensor<32xf32>,

 // CHECK:             %[[VAL_11:.*]] = mulf %[[VAL_9]], %[[VAL_10]] : f32
 // CHECK:             store %[[VAL_11]], %[[VAL_7]]{{\[}}%[[VAL_8]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_12:.*]] = tensor_load %[[VAL_7]] : memref<32xf32>
 // CHECK:           return %[[VAL_12]] : tensor<32xf32>
 // CHECK:         }
 func @mul_dd(%arga: tensor<32xf32>, %argb: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_dd
-    ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>)
+    outs(%arga : tensor<32xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 #trait_ds = {
   indexing_maps = [

 // CHECK:             %[[VAL_32:.*]] = load %[[VAL_7]]{{\[}}%[[VAL_30]]] : memref<32xf32>
 // CHECK:             store %[[VAL_32]], %[[VAL_11]]{{\[}}%[[VAL_30]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_33:.*]] = tensor_load %[[VAL_11]] : memref<32xf32>
 // CHECK:           return %[[VAL_33]] : tensor<32xf32>
 // CHECK:         }
 func @add_ds(%arga: tensor<32xf32>, %argb: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_ds
-    ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>)
+    outs(%arga : tensor<32xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // CHECK-LABEL:   func @mul_ds(
 // CHECK-SAME:                 %[[VAL_0:.*]]: tensor<32xf32>,

 // CHECK:             %[[VAL_16:.*]] = mulf %[[VAL_14]], %[[VAL_15]] : f32
 // CHECK:             store %[[VAL_16]], %[[VAL_9]]{{\[}}%[[VAL_13]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_17:.*]] = tensor_load %[[VAL_9]] : memref<32xf32>
 // CHECK:           return %[[VAL_17]] : tensor<32xf32>
 // CHECK:         }
 func @mul_ds(%arga: tensor<32xf32>, %argb: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_ds
-    ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>)
+    outs(%arga : tensor<32xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 #trait_sd = {
   indexing_maps = [

 // CHECK:             %[[VAL_32:.*]] = load %[[VAL_10]]{{\[}}%[[VAL_30]]] : memref<32xf32>
 // CHECK:             store %[[VAL_32]], %[[VAL_11]]{{\[}}%[[VAL_30]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_33:.*]] = tensor_load %[[VAL_11]] : memref<32xf32>
 // CHECK:           return %[[VAL_33]] : tensor<32xf32>
 // CHECK:         }
 func @add_sd(%arga: tensor<32xf32>, %argb: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_sd
-    ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>)
+    outs(%arga : tensor<32xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // CHECK-LABEL:   func @mul_sd(
 // CHECK-SAME:                 %[[VAL_0:.*]]: tensor<32xf32>,

 // CHECK:             %[[VAL_16:.*]] = mulf %[[VAL_14]], %[[VAL_15]] : f32
 // CHECK:             store %[[VAL_16]], %[[VAL_9]]{{\[}}%[[VAL_13]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_17:.*]] = tensor_load %[[VAL_9]] : memref<32xf32>
 // CHECK:           return %[[VAL_17]] : tensor<32xf32>
 // CHECK:         }
 func @mul_sd(%arga: tensor<32xf32>, %argb: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_sd
-    ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>)
+    outs(%arga : tensor<32xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 #trait_ss = {
   indexing_maps = [

 // CHECK:             %[[VAL_51:.*]] = load %[[VAL_10]]{{\[}}%[[VAL_48]]] : memref<?xf32>
 // CHECK:             store %[[VAL_51]], %[[VAL_11]]{{\[}}%[[VAL_50]]] : memref<32xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_52:.*]] = tensor_load %[[VAL_11]] : memref<32xf32>
 // CHECK:           return %[[VAL_52]] : tensor<32xf32>
 // CHECK:         }
 func @add_ss(%arga: tensor<32xf32>, %argb: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_ss
-    ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>)
+    outs(%arga : tensor<32xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // CHECK-LABEL:   func @mul_ss(
 // CHECK-SAME:                 %[[VAL_0:.*]]: tensor<32xf32>,

 // CHECK:             %[[VAL_39:.*]] = select %[[VAL_37]], %[[VAL_38]], %[[VAL_23]] : index
 // CHECK:             scf.yield %[[VAL_36]], %[[VAL_39]] : index, index
 // CHECK:           }
 // CHECK:           %[[VAL_40:.*]] = tensor_load %[[VAL_11]] : memref<32xf32>
 // CHECK:           return %[[VAL_40]] : tensor<32xf32>
 // CHECK:         }
 func @mul_ss(%arga: tensor<32xf32>, %argb: tensor<32xf32>) -> tensor<32xf32> {
   %0 = linalg.generic #trait_ss
-    ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32xf32>, tensor<32xf32>)
+    outs(%arga : tensor<32xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 #trait_two_way_inv = {
   indexing_maps = [

 // CHECK:             store %[[VAL_58]], %[[VAL_12]]{{\[}}%[[VAL_56]]] : memref<16xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_59:.*]] = tensor_load %[[VAL_12]] : memref<16xf32>
 // CHECK:           return %[[VAL_59]] : tensor<16xf32>
 // CHECK:         }
 func @two_way_inv(%arga: tensor<16xf32>,
                   %argb: tensor<16xf32>, %argc: f32) -> tensor<16xf32> {
   %0 = linalg.generic #trait_two_way_inv
-    ins(%arga, %argb : tensor<16xf32>, tensor<16xf32>) {
-      ^bb(%a : f32, %b : f32):
+    ins(%arga, %argb : tensor<16xf32>, tensor<16xf32>)
+    outs(%argb : tensor<16xf32>) {
+      ^bb(%a : f32, %b : f32, %c : f32):
         %0 = mulf %a, %argc : f32
         %1 = mulf %b, %argc : f32
         %2 = addf %0, %1 : f32
         linalg.yield %2: f32
   } -> tensor<16xf32>
   return %0 : tensor<16xf32>
 }
 

 // CHECK:           }
 // CHECK:           %[[VAL_58:.*]] = tensor_load %[[VAL_12]] : memref<16xf32>
 // CHECK:           return %[[VAL_58]] : tensor<16xf32>
 // CHECK:         }
 func @two_way_inv_alt(%arga: tensor<16xf32>,
                       %argb: tensor<16xf32>, %argc: f32) -> tensor<16xf32> {
   // Same kernel, but now expressed as "x(i) = (a(i) + b(i)) * c".
   %0 = linalg.generic #trait_two_way_inv
-    ins(%arga, %argb : tensor<16xf32>, tensor<16xf32>) {
-      ^bb(%a : f32, %b : f32):
+    ins(%arga, %argb : tensor<16xf32>, tensor<16xf32>)
+    outs(%argb : tensor<16xf32>) {
+      ^bb(%a : f32, %b : f32, %c : f32):
         %0 = addf %a, %b : f32
         %1 = mulf %0, %argc : f32
         linalg.yield %1: f32
   } -> tensor<16xf32>
   return %0 : tensor<16xf32>
 }
 
 #trait_sum_reduction = {

 // CHECK:           }
 // CHECK:           store %[[VAL_17:.*]], %[[VAL_8]][] : memref<f32>
 // CHECK:           %[[VAL_18:.*]] = tensor_load %[[VAL_8]] : memref<f32>
 // CHECK:           return %[[VAL_18]] : tensor<f32>
 // CHECK:         }
 func @sum_reduction(%arga: tensor<?xf32>, %argx: tensor<f32>) -> tensor<f32> {
   %0 = linalg.generic #trait_sum_reduction
     ins(%arga : tensor<?xf32>)
-    init(%argx : tensor<f32>) {
+    outs(%argx : tensor<f32>) {
       ^bb(%a : f32, %x : f32):
         %0 = addf %x, %a  : f32
         linalg.yield %0: f32
   } -> tensor<f32>
   return %0 : tensor<f32>
 }
 
 #trait_sum_reduction_ss = {

 // CHECK:         }
 func @sum_reduction_ss(%arga: tensor<16xf32>,
                        %argb: tensor<16xf32>,
                        %argx: tensor<f32>) -> tensor<f32> {
   // Just for testing. This case would be better expressed
   // as two separate reductions kernels.
   %0 = linalg.generic #trait_sum_reduction_ss
     ins(%arga, %argb: tensor<16xf32>, tensor<16xf32>)
-    init(%argx : tensor<f32>) {
+    outs(%argx : tensor<f32>) {
       ^bb(%a : f32, %b : f32, %x : f32):
         %0 = addf %a, %b  : f32
         %1 = addf %x, %0  : f32
         linalg.yield %1: f32
   } -> tensor<f32>
   return %0 : tensor<f32>
 }
 

 func @sum_reduction_inv(%arga: tensor<16xf32>,
                         %argb: tensor<f32>,
                         %argc: tensor<16xf32>,
                         %argx: tensor<f32>) -> tensor<f32> {
   // Just for testing. This case would be better expressed
   // as two separate reductions kernels.
   %0 = linalg.generic #trait_sum_reduction_inv_ss
     ins(%arga, %argb, %argc : tensor<16xf32>, tensor<f32>, tensor<16xf32>)
-    init(%argx : tensor<f32>) {
+    outs(%argx : tensor<f32>) {
       ^bb(%a : f32, %b : f32, %c : f32, %x : f32):
         %0 = mulf %a, %b  : f32
         %1 = addf %0, %c  : f32
         %2 = addf %x, %1  : f32
         linalg.yield %2: f32
   } -> tensor<f32>
   return %0 : tensor<f32>
 }

mlir/test/Dialect/Linalg/sparse_2d.mlir

 // CHECK:               store %[[VAL_13]], %[[VAL_8]]{{\[}}%[[VAL_9]], %[[VAL_10]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_14:.*]] = tensor_load %[[VAL_8]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_14]] : tensor<32x16xf32>
 // CHECK:         }
 func @add_dd(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_dd
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga: tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 // CHECK-LABEL:   func @mul_dd(
 // CHECK-SAME:                 %[[VAL_0:.*]]: tensor<32x16xf32>,

 // CHECK:               store %[[VAL_13]], %[[VAL_8]]{{\[}}%[[VAL_9]], %[[VAL_10]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_14:.*]] = tensor_load %[[VAL_8]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_14]] : tensor<32x16xf32>
 // CHECK:         }
 func @mul_dd(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_dd
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 #trait_ds = {
   indexing_maps = [

 // CHECK:               store %[[VAL_35]], %[[VAL_12]]{{\[}}%[[VAL_13]], %[[VAL_33]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_36:.*]] = tensor_load %[[VAL_12]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_36]] : tensor<32x16xf32>
 // CHECK:         }
 func @add_ds(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_ds
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 // CHECK-LABEL:   func @mul_ds(
 // CHECK-SAME:                 %[[VAL_0:.*]]: tensor<32x16xf32>,

 // CHECK:               store %[[VAL_19]], %[[VAL_10]]{{\[}}%[[VAL_11]], %[[VAL_16]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_20:.*]] = tensor_load %[[VAL_10]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_20]] : tensor<32x16xf32>
 // CHECK:         }
 func @mul_ds(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_ds
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 #trait_sd = {
   indexing_maps = [

 // CHECK:               store %[[VAL_38]], %[[VAL_12]]{{\[}}%[[VAL_35]], %[[VAL_37]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_39:.*]] = tensor_load %[[VAL_12]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_39]] : tensor<32x16xf32>
 // CHECK:         }
 func @add_sd(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_sd
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 // CHECK-LABEL:   func @mul_sd(
 // CHECK-SAME:                 %[[VAL_0:.*]]: tensor<32x16xf32>,

 // CHECK:               store %[[VAL_20]], %[[VAL_10]]{{\[}}%[[VAL_14]], %[[VAL_15]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_21:.*]] = tensor_load %[[VAL_10]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_21]] : tensor<32x16xf32>
 // CHECK:         }
 func @mul_sd(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_sd
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 #trait_ss = {
   indexing_maps = [

 // CHECK:               store %[[VAL_56]], %[[VAL_14]]{{\[}}%[[VAL_53]], %[[VAL_55]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_57:.*]] = tensor_load %[[VAL_14]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_57]] : tensor<32x16xf32>
 // CHECK:         }
 func @add_ss(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_ss
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 // CHECK-LABEL:   func @mul_ss(
 // CHECK-SAME:                 %[[VAL_0:.*]]: tensor<32x16xf32>,

 // CHECK:               store %[[VAL_23]], %[[VAL_11]]{{\[}}%[[VAL_15]], %[[VAL_20]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_24:.*]] = tensor_load %[[VAL_11]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_24]] : tensor<32x16xf32>
 // CHECK:         }
 func @mul_ss(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_ss
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 #trait_ss_ss = {
   indexing_maps = [

 // CHECK:               store %[[VAL_114]], %[[VAL_15]]{{\[}}%[[VAL_108]], %[[VAL_113]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_115:.*]] = tensor_load %[[VAL_15]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_115]] : tensor<32x16xf32>
 // CHECK:         }
 func @add_ss_ss(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_ss_ss
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 // CHECK-LABEL:   func @mul_ss_ss(
 // CHECK-SAME:                    %[[VAL_0:.*]]: tensor<32x16xf32>,

 // CHECK:             %[[VAL_70:.*]] = select %[[VAL_68]], %[[VAL_69]], %[[VAL_27]] : index
 // CHECK:             scf.yield %[[VAL_67]], %[[VAL_70]] : index, index
 // CHECK:           }
 // CHECK:           %[[VAL_71:.*]] = tensor_load %[[VAL_15]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_71]] : tensor<32x16xf32>
 // CHECK:         }
 func @mul_ss_ss(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_ss_ss
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 #trait_sd_ds = {
   indexing_maps = [

 // CHECK:               store %[[VAL_114]], %[[VAL_15]]{{\[}}%[[VAL_108]], %[[VAL_113]]] : memref<32x16xf32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_115:.*]] = tensor_load %[[VAL_15]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_115]] : tensor<32x16xf32>
 // CHECK:         }
 func @add_sd_ds(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_ss_ss
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 // CHECK-LABEL:   func @mul_sd_ds(
 // CHECK-SAME:                    %[[VAL_0:.*]]: tensor<32x16xf32>,

 // CHECK:             %[[VAL_70:.*]] = select %[[VAL_68]], %[[VAL_69]], %[[VAL_27]] : index
 // CHECK:             scf.yield %[[VAL_67]], %[[VAL_70]] : index, index
 // CHECK:           }
 // CHECK:           %[[VAL_71:.*]] = tensor_load %[[VAL_15]] : memref<32x16xf32>
 // CHECK:           return %[[VAL_71]] : tensor<32x16xf32>
 // CHECK:         }
 func @mul_sd_ds(%arga: tensor<32x16xf32>, %argb: tensor<32x16xf32>) -> tensor<32x16xf32> {
   %0 = linalg.generic #trait_ss_ss
-    ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16xf32>, tensor<32x16xf32>)
+    outs(%arga : tensor<32x16xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 #trait_matvec = {
   indexing_maps = [

 // CHECK:             }
 // CHECK:             store %[[VAL_25:.*]], %[[VAL_11]]{{\[}}%[[VAL_12]]] : memref<16xf32>
 // CHECK:           }
 // CHECK:           %[[VAL_26:.*]] = tensor_load %[[VAL_11]] : memref<16xf32>
 // CHECK:           return %[[VAL_26]] : tensor<16xf32>
 // CHECK:         }
 func @matvec(%argA: tensor<16x32xf32>, %argb: tensor<32xf32>, %argx: tensor<16xf32>) -> tensor<16xf32> {
   %0 = linalg.generic #trait_matvec
        ins(%argA, %argb : tensor<16x32xf32>, tensor<32xf32>)
-      init(%argx : tensor<16xf32>) {
+      outs(%argx : tensor<16xf32>) {
     ^bb(%A: f32, %b: f32, %x: f32):
       %0 = mulf %A, %b : f32
       %1 = addf %0, %x : f32
       linalg.yield %1 : f32
   } -> tensor<16xf32>
   return %0 : tensor<16xf32>
 }
 

 // CHECK:             }
 // CHECK:             store %[[VAL_20:.*]], %[[VAL_9]][] : memref<f32>
 // CHECK:           }
 // CHECK:           %[[VAL_21:.*]] = tensor_load %[[VAL_9]] : memref<f32>
 // CHECK:           return %[[VAL_21]] : tensor<f32>
 // CHECK:         }
 func @sum_reduction(%arga: tensor<10x20xf32>, %argx: tensor<f32>) -> tensor<f32> {
   %0 = linalg.generic #trait_sum_reduction
      ins(%arga : tensor<10x20xf32>)
-    init(%argx : tensor<f32>) {
+    outs(%argx : tensor<f32>) {
       ^bb(%a : f32, %x : f32):
         %0 = addf %x, %a  : f32
         linalg.yield %0: f32
   } -> tensor<f32>
   return %0 : tensor<f32>
 }
 
 #trait_scale = {

 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_19:.*]] = tensor_load %[[VAL_10]] : memref<?x?xf64>
 // CHECK:           return %[[VAL_19]] : tensor<?x?xf64>
 // CHECK:         }
 func @scale(%arga: tensor<?x?xf64>) -> tensor<?x?xf64> {
   %0 = constant 2.0 : f64
   %1 = linalg.generic #trait_scale
-    ins(%arga: tensor<?x?xf64>) {
-      ^bb(%a: f64):
+     ins(%arga: tensor<?x?xf64>)
+    outs(%arga: tensor<?x?xf64>) {
+      ^bb(%a: f64, %s: f64):
         %2 = mulf %a, %0  : f64
         linalg.yield %2 : f64
   } -> tensor<?x?xf64>
   return %1 : tensor<?x?xf64>
 }
 
 #trait_sampled_dense_dense = {
   indexing_maps = [

 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_38:.*]] = tensor_load %[[VAL_20]] : memref<?x?xf32>
 // CHECK:           return %[[VAL_38]] : tensor<?x?xf32>
 // CHECK:         }
 func @sampled_dense_dense(%args: tensor<?x?xf32>,
                           %arga: tensor<?x?xf32>,
                           %argb: tensor<?x?xf32>,
-        %argx: tensor<?x?xf32>) -> tensor<?x?xf32> {
+                          %argx: tensor<?x?xf32>) -> tensor<?x?xf32> {
   %0 = linalg.generic #trait_sampled_dense_dense
      ins(%args, %arga, %argb : tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>)
-    init(%argx : tensor<?x?xf32>) {
+    outs(%argx : tensor<?x?xf32>) {
       ^bb(%s : f32, %a : f32, %b : f32, %x : f32):
         %0 = mulf %a, %b  : f32
         %1 = mulf %s, %0  : f32
         %2 = addf %x, %1  : f32
         linalg.yield %2: f32
   } -> tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }
func @sum_kernel_with_inv(%arga: tensor<?x?xf32>,
                           %arge: tensor<f32>,
                           %argx: tensor<?xf32>) -> tensor<?xf32> {
   %0 = linalg.generic #trait_sum_kernel_with_inv
     ins(%arga, %argb, %argc, %argd, %arge : tensor<?x?xf32>,
                                             tensor<?x?xf32>,
                                             tensor<?x?xf32>,
                                             tensor<?xf32>,
                                             tensor<f32>)
-    init(%argx : tensor<?xf32>) {
+    outs(%argx : tensor<?xf32>) {
       ^bb(%a : f32, %b : f32, %c : f32, %d : f32, %e : f32, %x : f32):
         %0 = mulf %a, %b  : f32
         %1 = mulf %0, %d  : f32
         %2 = mulf %1, %e  : f32
         %3 = addf %2, %c  : f32
         %4 = addf %x, %3  : f32
         linalg.yield %4: f32
   } -> tensor<?xf32>
   return %0 : tensor<?xf32>
 }

mlir/test/Dialect/Linalg/sparse_3d.mlir

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_16:.*]] = tensor_load %[[VAL_9]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_16]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @add_ddd(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_ddd
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s: f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 // CHECK-LABEL:   func @mul_ddd(
 // CHECK-SAME:                  %[[VAL_0:.*]]: tensor<32x16x8xf32>,

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_16:.*]] = tensor_load %[[VAL_9]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_16]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @mul_ddd(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_ddd
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 #trait_dds = {
   indexing_maps = [

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_40:.*]] = tensor_load %[[VAL_13]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_40]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @add_dds(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_dds
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 // CHECK-LABEL:   func @mul_dds(
 // CHECK-SAME:                  %[[VAL_0:.*]]: tensor<32x16x8xf32>,

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_24:.*]] = tensor_load %[[VAL_11]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_24]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @mul_dds(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_dds
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 #trait_dsd = {
   indexing_maps = [

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_42:.*]] = tensor_load %[[VAL_13]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_42]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @add_dsd(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_dsd
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 // CHECK-LABEL:   func @mul_dsd(
 // CHECK-SAME:                  %[[VAL_0:.*]]: tensor<32x16x8xf32>,

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_24:.*]] = tensor_load %[[VAL_11]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_24]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @mul_dsd(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_dsd
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 #trait_dss = {
   indexing_maps = [

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_60:.*]] = tensor_load %[[VAL_15]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_60]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @add_dss(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_dss
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 // CHECK-LABEL:   func @mul_dss(
 // CHECK-SAME:                  %[[VAL_0:.*]]: tensor<32x16x8xf32>,

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_27:.*]] = tensor_load %[[VAL_12]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_27]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @mul_dss(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_dss
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 #trait_sdd = {
   indexing_maps = [

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_45:.*]] = tensor_load %[[VAL_13]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_45]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @add_sdd(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_sdd
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 // CHECK-LABEL:   func @mul_sdd(
 // CHECK-SAME:                  %[[VAL_0:.*]]: tensor<32x16x8xf32>,

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_25:.*]] = tensor_load %[[VAL_11]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_25]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @mul_sdd(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_sdd
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 #trait_sds = {
   indexing_maps = [

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_63:.*]] = tensor_load %[[VAL_15]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_63]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @add_sds(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_sds
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 // CHECK-LABEL:   func @mul_sds(
 // CHECK-SAME:                  %[[VAL_0:.*]]: tensor<32x16x8xf32>,

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_28:.*]] = tensor_load %[[VAL_12]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_28]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @mul_sds(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_sds
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 #trait_ssd = {
   indexing_maps = [

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_65:.*]] = tensor_load %[[VAL_15]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_65]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @add_ssd(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_ssd
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 // CHECK-LABEL:   func @mul_ssd(
 // CHECK-SAME:                  %[[VAL_0:.*]]: tensor<32x16x8xf32>,

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_28:.*]] = tensor_load %[[VAL_12]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_28]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @mul_ssd(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_ssd
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 #trait_sss = {
   indexing_maps = [

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_83:.*]] = tensor_load %[[VAL_17]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_83]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @add_sss(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_sss
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = addf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 // CHECK-LABEL:   func @mul_sss(
 // CHECK-SAME:                  %[[VAL_0:.*]]: tensor<32x16x8xf32>,

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_31:.*]] = tensor_load %[[VAL_13]] : memref<32x16x8xf32>
 // CHECK:           return %[[VAL_31]] : tensor<32x16x8xf32>
 // CHECK:         }
 func @mul_sss(%arga: tensor<32x16x8xf32>, %argb: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> {
   %0 = linalg.generic #trait_sss
-    ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>) {
-      ^bb(%a: f32, %b: f32):
+     ins(%arga, %argb: tensor<32x16x8xf32>, tensor<32x16x8xf32>)
+    outs(%arga : tensor<32x16x8xf32>) {
+      ^bb(%a: f32, %b: f32, %s : f32):
         %0 = mulf %a, %b  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16x8xf32>
   return %0 : tensor<32x16x8xf32>
 }
 
 #trait_kernel_3d = {
   indexing_maps = [

 // CHECK:           %[[VAL_36:.*]] = tensor_load %[[VAL_18]] : memref<?x?xf32>
 // CHECK:           return %[[VAL_36]] : tensor<?x?xf32>
 // CHECK:         }
 func @kernel_3d(%arga: tensor<?x?xf32>,
                 %argb: tensor<?x?x?xf32>,
                 %argc: tensor<?x?xf32>,
           %argd: tensor<?x?xf32>) -> tensor<?x?xf32> {
   %0 = linalg.generic #trait_kernel_3d
        ins(%argb, %argc, %argd : tensor<?x?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>)
-      init(%arga : tensor<?x?xf32>) {
+      outs(%arga : tensor<?x?xf32>) {
     ^bb(%b: f32, %c: f32, %d : f32, %a : f32):
       %0 = mulf %b, %c : f32
       %1 = mulf %0, %d : f32
       %2 = addf %1, %a : f32
       linalg.yield %2 : f32
   } -> tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }

 // CHECK:               store %[[VAL_29:.*]], %[[VAL_12]][] : memref<f32>
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_30:.*]] = tensor_load %[[VAL_12]] : memref<f32>
 // CHECK:           return %[[VAL_30]] : tensor<f32>
 // CHECK:         }
 func @sum_reduction(%arga: tensor<10x20x30xf32>, %argx: tensor<f32>) -> tensor<f32> {
   %0 = linalg.generic #trait_sum_reduction
      ins(%arga : tensor<10x20x30xf32>)
-    init(%argx : tensor<f32>) {
+    outs(%argx : tensor<f32>) {
       ^bb(%a : f32, %x : f32):
         %0 = addf %x, %a  : f32
         linalg.yield %0: f32
   } -> tensor<f32>
   return %0 : tensor<f32>
 }
 
 #trait_sum_reduction_inv = {

 // CHECK:           %[[VAL_24:.*]] = tensor_load %[[VAL_12]] : memref<f32>
 // CHECK:           return %[[VAL_24]] : tensor<f32>
 // CHECK:         }
 func @sum_reduction_inv(%arga: tensor<?x?x?xf32>,
                         %argb: tensor<?xf32>,
             %argx: tensor<f32>) -> tensor<f32> {
   %0 = linalg.generic #trait_sum_reduction_inv
     ins(%arga, %argb : tensor<?x?x?xf32>, tensor<?xf32>)
-    init(%argx : tensor<f32>) {
+    outs(%argx : tensor<f32>) {
       ^bb(%a : f32, %b : f32, %x : f32):
         %0 = mulf %a, %b  : f32
         %1 = addf %x, %0  : f32
         linalg.yield %1: f32
   } -> tensor<f32>
   return %0 : tensor<f32>
 }
 
#trait_invariants = {
   ],
   iterator_types = ["parallel", "parallel", "parallel"],
   doc = "X(i,j,k) = a(i) * b(j) * c(k)"
 }
 
 // CHECK-LABEL:   func @invariants(
 // CHECK-SAME:                     %[[VAL_0:.*]]: tensor<10xf32>,
 // CHECK-SAME:                     %[[VAL_1:.*]]: tensor<20xf32>,
-// CHECK-SAME:                     %[[VAL_2:.*]]: tensor<30xf32>) -> tensor<10x20x30xf32> {
+// CHECK-SAME:                     %[[VAL_2:.*]]: tensor<30xf32>,
+// CHECK-SAME:                     %[[SHAPE:.*]]: tensor<10x20x30xf32>) -> tensor<10x20x30xf32> {
 // CHECK:           %[[VAL_3:.*]] = constant 10 : index
 // CHECK:           %[[VAL_4:.*]] = constant 20 : index
 // CHECK:           %[[VAL_5:.*]] = constant 30 : index
 // CHECK:           %[[VAL_6:.*]] = constant 0 : index
 // CHECK:           %[[VAL_7:.*]] = constant 1 : index
 // CHECK:           %[[VAL_8:.*]] = alloca() : memref<10xf32>
 // CHECK:           %[[VAL_9:.*]] = alloca() : memref<20xf32>
 // CHECK:           %[[VAL_10:.*]] = alloca() : memref<30xf32>

 // CHECK:               }
 // CHECK:             }
 // CHECK:           }
 // CHECK:           %[[VAL_20:.*]] = tensor_load %[[VAL_11]] : memref<10x20x30xf32>
 // CHECK:           return %[[VAL_20]] : tensor<10x20x30xf32>
 // CHECK:         }
 func @invariants(%arga: tensor<10xf32>,
                  %argb: tensor<20xf32>,
-                 %argc: tensor<30xf32>) -> tensor<10x20x30xf32> {
+                 %argc: tensor<30xf32>,
+                 %shape : tensor<10x20x30xf32>) -> tensor<10x20x30xf32> {
   %0 = linalg.generic #trait_invariants
-    ins(%arga, %argb, %argc : tensor<10xf32>, tensor<20xf32>, tensor<30xf32>) {
-      ^bb(%a : f32, %b : f32, %c : f32):
+     ins(%arga, %argb, %argc : tensor<10xf32>, tensor<20xf32>, tensor<30xf32>)
+    outs(%shape : tensor<10x20x30xf32>) {
+      ^bb(%a : f32, %b : f32, %c : f32, %s : f32):
         %0 = mulf %a, %b  : f32
         %1 = mulf %0, %c  : f32
         linalg.yield %1: f32
   } -> tensor<10x20x30xf32>
   return %0 : tensor<10x20x30xf32>
 }

mlir/test/Dialect/Linalg/sparse_invalid.mlir

 // RUN: mlir-opt %s -split-input-file -verify-diagnostics
 
 #trait_memref = {
   indexing_maps = [
     affine_map<(i) -> (i)>,  // a
     affine_map<(i) -> (i)>   // x (out)
   ],
   sparse = [
     [ "S" ],  // a
     [ "D" ]   // x
   ],
   iterator_types = ["parallel"]
 }
 
-func @invalid_memref(%arga: memref<32xf32>, %argb: f32) -> tensor<32xf32> {
+func @invalid_memref(%arga: memref<32xf32>, %argb: f32, %shape: tensor<32xf32>)
+  -> tensor<32xf32>
+{
   // expected-error@+1 {{'linalg.generic' op expected sparse annotations on tensors only}}
   %0 = linalg.generic #trait_memref
-    ins(%arga: memref<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: memref<32xf32>)
+    outs(%shape: tensor<32xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // -----
 
-#trait_two_out = {
-  indexing_maps = [
-    affine_map<(i) -> (i)>,  // a
-    affine_map<(i) -> (i)>,  // x (out)
-    affine_map<(i) -> (i)>   // y (out)
-  ],
-  sparse = [
-    [ "S" ],  // a
-    [ "D" ],  // x
-    [ "D" ]   // y
-  ],
-  iterator_types = ["parallel"]
-}
-
-func @invalid_two_out(%arga: tensor<32xf32>) -> tensor<32xf32> {
-  // expected-error@+1 {{'linalg.generic' op expected single output tensor}}
-  %0, %1 = linalg.generic #trait_two_out
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
-        %0 = addf %a, %a : f32
-        linalg.yield %a, %0 : f32, f32
-  } -> tensor<32xf32>, tensor<32xf32>
-  return %1 : tensor<32xf32>
-}
-
-// -----
-
-#trait_two_blocks = {
-  indexing_maps = [
-    affine_map<(i) -> (i)>,  // a
-    affine_map<(i) -> (i)>   // x (out)
-  ],
-  sparse = [
-    [ "S" ],  // a
-    [ "D" ]   // x
-  ],
-  iterator_types = ["parallel"]
-}
-
-func @invalid_two_blocks(%arga: tensor<32xf32>) -> tensor<32xf32> {
-  // expected-error@+1 {{'linalg.generic' op expects region #0 to have 0 or 1 blocks}}
-  %0 = linalg.generic #trait_two_blocks
-    ins(%arga: tensor<32xf32>) {
-      ^bb1(%a: f32):
-        %0 = addf %a, %a : f32
-      ^bb2:
-        linalg.yield %0 : f32
-  } -> tensor<32xf32>
-  return %0 : tensor<32xf32>
-}
-
-// -----
-
-#trait_no_block = {
-  indexing_maps = [
-    affine_map<(i) -> (i)>  // a
-  ],
-  sparse = [
-    [ "S" ]  // a
-  ],
-  iterator_types = ["parallel"]
-}
-
-func @invalid_no_block(%arga: tensor<32xf32>) {
-  // expected-error@+1 {{'linalg.generic' op expected region with 1 block}}
-  linalg.generic #trait_no_block
-    ins(%arga: tensor<32xf32>) {
-    }
-  return
-}
-
-// -----
-
 #trait_too_many = {
   indexing_maps = [
     affine_map<(i) -> (i)>,  // a
     affine_map<(i) -> (i)>   // x (out)
   ],
   sparse = [
     [ "S" ],  // a
     [ "S" ],  // b
     [ "D" ]   // x
   ],
   iterator_types = ["parallel"]
 }
 
 func @invalid_too_many(%arga: tensor<32xf32>, %argb: f32) -> tensor<32xf32> {
   // expected-error@+1 {{'linalg.generic' op expected one sparse annotation for each tensor}}
   %0 = linalg.generic #trait_too_many
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32xf32>)
+    outs(%arga: tensor<32xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // -----
 
 #trait_no_array = {
   indexing_maps = [
     affine_map<(i) -> (i)>,  // a
     affine_map<(i) -> (i)>   // x (out)
   ],
   sparse = [ 1, 2 ],
   iterator_types = ["parallel"]
 }
 
 func @invalid_no_array(%arga: tensor<32xf32>, %argb: f32) -> tensor<32xf32> {
   // expected-error@+1 {{'linalg.generic' op expected sparse annotation array for tensor 0}}
   %0 = linalg.generic #trait_no_array
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32xf32>)
+    outs(%arga: tensor<32xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // -----
 
 #trait_wrong_rank = {
   indexing_maps = [
     affine_map<(i) -> (i)>,  // a
     affine_map<(i) -> (i)>   // x (out)
   ],
   sparse = [
     [ "S" ],
     [ "D", "D" ]
   ],
   iterator_types = ["parallel"]
 }
 
 func @invalid_wrong_rank(%arga: tensor<32xf32>, %argb: f32) -> tensor<32xf32> {
   // expected-error@+1 {{'linalg.generic' op expected sparse annotation with rank 1 for tensor 1}}
   %0 = linalg.generic #trait_wrong_rank
-    ins(%arga: tensor<32xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32xf32>)
+    outs(%arga: tensor<32xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32xf32>
   return %0 : tensor<32xf32>
 }
 
 // -----
 
 #trait_no_string = {
   indexing_maps = [
     affine_map<(i,j) -> (i,j)>,  // a
     affine_map<(i,j) -> (i,j)>   // x (out)
   ],
   sparse = [
     [ "S", 1 ],
     [ "D", "D" ]
   ],
   iterator_types = ["parallel","parallel"]
 }
 
 func @invalid_no_string(%arga: tensor<32x16xf32>, %argb: f32) -> tensor<32x16xf32> {
   // expected-error@+1 {{'linalg.generic' op expected sparse annotation at position 1 for tensor 0}}
   %0 = linalg.generic #trait_no_string
-    ins(%arga: tensor<32x16xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32x16xf32>)
+    outs(%arga: tensor<32x16xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 // -----
 
 #trait_wrong_symbol = {
   indexing_maps = [
     affine_map<(i,j) -> (i,j)>,  // a
     affine_map<(i,j) -> (i,j)>   // x (out)
   ],
   sparse = [
     [ "S", "S" ],
     [ "D", "X" ]
   ],
   iterator_types = ["parallel","parallel"]
 }
 
 func @invalid_wrong_symbol(%arga: tensor<32x16xf32>, %argb: f32) -> tensor<32x16xf32> {
   // expected-error@+1 {{'linalg.generic' op expected sparse annotation at position 1 for tensor 1}}
   %0 = linalg.generic #trait_wrong_symbol
-    ins(%arga: tensor<32x16xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32x16xf32>)
+    outs(%arga: tensor<32x16xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }
 
 // -----
 
 #trait_no_sparse_output = {
   indexing_maps = [
     affine_map<(i,j) -> (i,j)>,  // a
     affine_map<(i,j) -> (i,j)>   // x (out)
   ],
   sparse = [
     [ "S", "S" ],
     [ "D", "S" ]
   ],
   iterator_types = ["parallel","parallel"]
 }
 
 func @invalid_no_sparse_output(%arga: tensor<32x16xf32>, %argb: f32) -> tensor<32x16xf32> {
   // expected-error@+1 {{'linalg.generic' op sparse output tensors not supported (yet)}}
   %0 = linalg.generic #trait_no_sparse_output
-    ins(%arga: tensor<32x16xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<32x16xf32>)
+    outs(%arga: tensor<32x16xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = addf %a, %argb  : f32
         linalg.yield %0 : f32
   } -> tensor<32x16xf32>
   return %0 : tensor<32x16xf32>
 }

mlir/test/Dialect/Linalg/sparse_parallel.mlir

 //
 // CHECK-PAR4-LABEL: func @scale_dd
 // CHECK-PAR4:         scf.parallel
 // CHECK-PAR4:           scf.parallel
 // CHECK-PAR4:         return
 //
 func @scale_dd(%scale: f32, %arga: tensor<?x?xf32>) -> tensor<?x?xf32> {
   %0 = linalg.generic #trait_dd
-    ins(%arga: tensor<?x?xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<?x?xf32>)
+    outs(%arga: tensor<?x?xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = mulf %a, %scale  : f32
         linalg.yield %0 : f32
   } -> tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }
 
 #trait_ss = {
   indexing_maps = [

 //
 // CHECK-PAR4-LABEL: func @scale_ss
 // CHECK-PAR4:         scf.parallel
 // CHECK-PAR4:           scf.parallel
 // CHECK-PAR4:         return
 //
 func @scale_ss(%scale: f32, %arga: tensor<?x?xf32>) -> tensor<?x?xf32> {
   %0 = linalg.generic #trait_ss
-    ins(%arga: tensor<?x?xf32>) {
-      ^bb(%a: f32):
+     ins(%arga: tensor<?x?xf32>)
+    outs(%arga: tensor<?x?xf32>) {
+      ^bb(%a: f32, %s: f32):
         %0 = mulf %a, %scale  : f32
         linalg.yield %0 : f32
   } -> tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }
 
 #trait_matvec = {
   indexing_maps = [

 // CHECK-PAR4-LABEL: func @matvec
 // CHECK-PAR4:         scf.parallel
 // CHECK-PAR4:           scf.for
 // CHECK-PAR4:         return
 //
 func @matvec(%argA: tensor<16x32xf32>, %argb: tensor<32xf32>, %argx: tensor<16xf32>) -> tensor<16xf32> {
   %0 = linalg.generic #trait_matvec
       ins(%argA, %argb : tensor<16x32xf32>, tensor<32xf32>)
-      init(%argx : tensor<16xf32>) {
+     outs(%argx : tensor<16xf32>) {
     ^bb(%A: f32, %b: f32, %x: f32):
       %0 = mulf %A, %b : f32
       %1 = addf %0, %x : f32
       linalg.yield %1 : f32
   } -> tensor<16xf32>
   return %0 : tensor<16xf32>
 }

mlir/test/Dialect/Linalg/sparse_storage.mlir

 // CHECK-TYPE3:   %[[VAL0:.*]] = load %{{.*}}[%[[I]]] : memref<?xf64>
 // CHECK-TYPE3:   %[[VAL1:.*]] = load %{{.*}}[%[[INDC]]] : memref<32xf64>
 // CHECK-TYPE3:   %[[MUL:.*]] = mulf %[[VAL0]], %[[VAL1]] : f64
 // CHECK-TYPE3:   store %[[MUL]], %{{.*}}[%[[INDC]]] : memref<32xf64>
 // CHECK-TYPE3: }
 
 func @mul_dd(%arga: tensor<32xf64>, %argb: tensor<32xf64>) -> tensor<32xf64> {
   %0 = linalg.generic #trait_mul_1d
-    ins(%arga, %argb: tensor<32xf64>, tensor<32xf64>) {
-      ^bb(%a: f64, %b: f64):
+     ins(%arga, %argb: tensor<32xf64>, tensor<32xf64>)
+    outs(%arga : tensor<32xf64>) {
+      ^bb(%a: f64, %b: f64, %s: f64):
         %0 = mulf %a, %b  : f64
         linalg.yield %0 : f64
   } -> tensor<32xf64>
   return %0 : tensor<32xf64>
 }
 

mlir/test/Dialect/Linalg/tile-and-distribute.mlir

 //      CHECK: %[[LBX:.*]] = muli %[[BIDX]], %[[C8]] : index
 //      CHECK: %[[STEPX:.*]] = muli %[[NBLOCKSX]], %[[C8]] : index
 //      CHECK:   %[[TD1:.*]] = scf.for {{.*}} to {{.*}} step {{.*}} iter_args(%[[TC1:.*]] = %[[TC0]]) -> (tensor<?x?xf32>) {
 //      CHECK:     %[[TD2:.*]] = scf.for {{.*}} to {{.*}} step {{.*}} iter_args(%[[TC2:.*]] = %[[TC1]]) -> (tensor<?x?xf32>) {
 //      CHECK:       %[[sTA:.*]] = subtensor %[[TA]][{{.*}}] : tensor<?x?xf32> to tensor<?x?xf32>
 //      CHECK:       %[[sTB:.*]] = subtensor %[[TB]][{{.*}}] : tensor<?x?xf32> to tensor<?x?xf32>
 //      CHECK:       %[[sTC:.*]] = subtensor %[[TC2]][{{.*}}] : tensor<?x?xf32> to tensor<?x?xf32>
 //      CHECK:       %[[sTD:.*]] = linalg.matmul ins(%[[sTA]], %[[sTB]] : tensor<?x?xf32>, tensor<?x?xf32>)
-// CHECK-SAME:                                  init(%[[sTC]] : tensor<?x?xf32>)  -> tensor<?x?xf32>
+// CHECK-SAME:                                  outs(%[[sTC]] : tensor<?x?xf32>)  -> tensor<?x?xf32>
 //      CHECK:       %[[TD:.*]] = subtensor_insert %[[sTD]] into %[[TC2]][{{.*}}]  : tensor<?x?xf32> into tensor<?x?xf32>
 //      CHECK:       scf.yield %[[TD]] : tensor<?x?xf32>
 //      CHECK:     scf.yield %[[TD2]] : tensor<?x?xf32>
 //      CHECK:   scf.yield %[[TD1]] : tensor<?x?xf32>
   %0 = linalg.matmul {__internal_linalg_transform__ = "tensors_distribute1"}
        ins(%arg0, %arg1: tensor<?x?xf32>, tensor<?x?xf32>)
-      init(%arg2: tensor<?x?xf32>)
+      outs(%arg2: tensor<?x?xf32>)
     -> tensor<?x?xf32>
 
 //      CHECK: return %[[TD0]] : tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }
 

mlir/test/Dialect/Linalg/tile-and-fuse-tensors.mlir

 // RUN: mlir-opt %s -test-linalg-greedy-fusion -split-input-file | FileCheck %s
 
 #map0 = affine_map<(d0)[s0] -> (2, -d0 + s0)>
 #map1 = affine_map<(d0)[s0] -> (4, -d0 + s0)>
 #map2 = affine_map<(d0)[s0] -> (3, -d0 + s0)>
 #map3 = affine_map<(d0, d1) -> (2, d0 - d1)>
 #map4 = affine_map<(d0, d1) -> (3, d0 - d1)>
 
 func @matmul_tensors(%arg0: tensor<?x?xf32>, %arg1: tensor<?x?xf32>, %arg2: tensor<?x?xf32>) -> tensor<?x?xf32> {
   %t0 = linalg.matmul ins(%arg0, %arg1: tensor<?x?xf32>, tensor<?x?xf32>)
-                     init(%arg2: tensor<?x?xf32>)
+                     outs(%arg2: tensor<?x?xf32>)
     -> tensor<?x?xf32>
 
   %c4 = constant 4 : index
   %c2 = constant 2 : index
   %c0 = constant 0 : index
   %c3 = constant 3 : index
   %c1 = constant 1 : index
   %0 = dim %t0, %c0 : tensor<?x?xf32>
   %1 = dim %t0, %c1 : tensor<?x?xf32>
   %2 = dim %arg1, %c1 : tensor<?x?xf32>
   %3 = scf.for %arg3 = %c0 to %0 step %c2 iter_args(%arg4 = %arg2) -> (tensor<?x?xf32>) {
     %4 = scf.for %arg5 = %c0 to %2 step %c3 iter_args(%arg6 = %arg4) -> (tensor<?x?xf32>) {
       %5 = scf.for %arg7 = %c0 to %1 step %c4 iter_args(%arg8 = %arg6) -> (tensor<?x?xf32>) {
         %6 = subtensor %t0[%arg3, %arg7][%c2, 4][1, 1] : tensor<?x?xf32> to tensor<?x4xf32>
         %7 = subtensor %arg1[%arg7, %arg5][4, %c3][1, 1] : tensor<?x?xf32> to tensor<4x?xf32>
         %8 = subtensor %arg8[%arg3, %arg5][%c2, %c3][1, 1] : tensor<?x?xf32> to tensor<?x?xf32>
-        %9 = linalg.matmul ins(%6, %7 : tensor<?x4xf32>, tensor<4x?xf32>) init(%8 : tensor<?x?xf32>) -> tensor<?x?xf32>
+        %9 = linalg.matmul ins(%6, %7 : tensor<?x4xf32>, tensor<4x?xf32>) outs(%8 : tensor<?x?xf32>) -> tensor<?x?xf32>
         %10 = subtensor_insert %9 into %arg8[%arg3, %arg5] [%c2, %c3] [1, 1]  : tensor<?x?xf32> into tensor<?x?xf32>
         scf.yield %10 : tensor<?x?xf32>
       }
       scf.yield %5 : tensor<?x?xf32>
     }
     scf.yield %4 : tensor<?x?xf32>
   }
   return %3 : tensor<?x?xf32>

 //  CHECK-NEXT:   scf.for %[[J:[0-9a-z]*]]
 //  CHECK-NEXT:     scf.for %[[K:[0-9a-z]*]] {{.*}} iter_args(%[[RES:[0-9a-z]*]]
 //   CHECK-DAG:       %[[stB1:.*]] = subtensor %[[B]][%[[K]], %[[J]]] [4, 3] [1, 1]  : tensor<?x?xf32> to tensor<4x3xf32>
 //   CHECK-DAG:       %[[stF:.*]] = subtensor %[[RES]][%[[I]], %[[J]]] [2, 3] [1, 1]  : tensor<?x?xf32> to tensor<2x3xf32>
 //
 // subtensors of the producing matmul.
 //   CHECK-DAG:       %[[stB2:.*]] = subtensor %[[B]][0, %[[K]]] [%[[dA1]], 4] [1, 1]  : tensor<?x?xf32> to tensor<?x4xf32>
 //   CHECK-DAG:       %[[stC:.*]] = subtensor %[[C]][%[[I]], %[[K]]] [2, 4] [1, 1]  : tensor<?x?xf32> to tensor<2x4xf32>
-//       CHECK:       %[[stD:.*]] = linalg.matmul ins(%[[stA]], %[[stB2]] : tensor<2x?xf32>, tensor<?x4xf32>) init(%[[stC]] : tensor<2x4xf32>)  -> tensor<2x4xf32>
-//  CHECK-NEXT:       %[[stG:.*]] = linalg.matmul ins(%[[stD]], %[[stB1]] : tensor<2x4xf32>, tensor<4x3xf32>) init(%[[stF]] : tensor<2x3xf32>)  -> tensor<2x3xf32>
+//       CHECK:       %[[stD:.*]] = linalg.matmul ins(%[[stA]], %[[stB2]] : tensor<2x?xf32>, tensor<?x4xf32>) outs(%[[stC]] : tensor<2x4xf32>)  -> tensor<2x4xf32>
+//  CHECK-NEXT:       %[[stG:.*]] = linalg.matmul ins(%[[stD]], %[[stB1]] : tensor<2x4xf32>, tensor<4x3xf32>) outs(%[[stF]] : tensor<2x3xf32>)  -> tensor<2x3xf32>
 //  CHECK-NEXT:       subtensor_insert %[[stG]] into %[[RES]][%[[I]], %[[J]]]

mlir/test/Dialect/Linalg/tile-tensors.mlir

-// RUN: mlir-opt %s -linalg-tile="linalg-tile-sizes=2,3,4" -mlir-disable-threading=true | FileCheck %s
+// RUN: mlir-opt %s -linalg-tile="linalg-tile-sizes=2,3,4" | FileCheck %s
 
 // CHECK-LABEL: func @matmul_tensors(
 // CHECK-SAME:    %[[TA:[0-9a-z]+]]: tensor<?x?xf32>
 // CHECK-SAME:    %[[TB:[0-9a-z]+]]: tensor<?x?xf32>
 // CHECK-SAME:    %[[TC:[0-9a-z]+]]: tensor<?x?xf32>) -> tensor<?x?xf32> {
 func @matmul_tensors(
   %arg0: tensor<?x?xf32>, %arg1: tensor<?x?xf32>, %arg2: tensor<?x?xf32>)
     -> tensor<?x?xf32> {
 //      CHECK: %[[TD0:.*]] = scf.for {{.*}} to {{.*}} step {{.*}} iter_args(%[[TC0:.*]] = %[[TC]]) -> (tensor<?x?xf32>) {
 //      CHECK:   %[[TD1:.*]] = scf.for {{.*}} to {{.*}} step {{.*}} iter_args(%[[TC1:.*]] = %[[TC0]]) -> (tensor<?x?xf32>) {
 //      CHECK:     %[[TD2:.*]] = scf.for {{.*}} to {{.*}} step {{.*}} iter_args(%[[TC2:.*]] = %[[TC1]]) -> (tensor<?x?xf32>) {
 //      CHECK:       %[[sTA:.*]] = subtensor %[[TA]][{{.*}}] : tensor<?x?xf32> to tensor<?x?xf32>
 //      CHECK:       %[[sTB:.*]] = subtensor %[[TB]][{{.*}}] : tensor<?x?xf32> to tensor<?x?xf32>
 //      CHECK:       %[[sTC:.*]] = subtensor %[[TC2]][{{.*}}] : tensor<?x?xf32> to tensor<?x?xf32>
 //      CHECK:       %[[sTD:.*]] = linalg.matmul ins(%[[sTA]], %[[sTB]] : tensor<?x?xf32>, tensor<?x?xf32>)
-// CHECK-SAME:                                  init(%[[sTC]] : tensor<?x?xf32>)  -> tensor<?x?xf32>
+// CHECK-SAME:                                  outs(%[[sTC]] : tensor<?x?xf32>)  -> tensor<?x?xf32>
 //      CHECK:       %[[TD:.*]] = subtensor_insert %[[sTD]] into %[[TC2]][{{.*}}]  : tensor<?x?xf32> into tensor<?x?xf32>
 //      CHECK:       scf.yield %[[TD]] : tensor<?x?xf32>
 //      CHECK:     scf.yield %[[TD2]] : tensor<?x?xf32>
 //      CHECK:   scf.yield %[[TD1]] : tensor<?x?xf32>
   %0 = linalg.matmul  ins(%arg0, %arg1: tensor<?x?xf32>, tensor<?x?xf32>)
-                     init(%arg2: tensor<?x?xf32>)
+                     outs(%arg2: tensor<?x?xf32>)
     -> tensor<?x?xf32>
 
 //      CHECK: return %[[TD0]] : tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }

mlir/test/EDSC/builder-api-test.cpp

 //       CHECK:     mulf
 //       CHECK:   } -> tensor<?x?xf32>
 //       CHECK:   linalg.generic {
 //  CHECK-SAME:     indexing_maps = [affine_map<(d0, d1, d2) -> (d0, d2)>,
 //  CHECK-SAME:                      affine_map<(d0, d1, d2) -> (d2, d1)>,
 //  CHECK-SAME:                      affine_map<(d0, d1, d2) -> (d0, d1)>],
 //  CHECK-SAME:     iterator_types = ["parallel", "parallel", "reduction"]
 // CHECK-SAME: ins(%{{[a-z0-9]*}}, %{{[a-z0-9]*}} : tensor<?x?xf32>, memref<?x?xf32>)
-// CHECK-SAME: init(%{{[a-z0-9]*}} : tensor<?x?xf32>)
+// CHECK-SAME: outs(%{{[a-z0-9]*}} : tensor<?x?xf32>)
 //       CHECK:     mulf
 //       CHECK:     addf
 //       CHECK:   } -> tensor<?x?xf32>
 // clang-format on
 TEST_FUNC(linalg_tensors_test) {
   using namespace edsc;
   using namespace edsc::ops;
 
   auto f32Type = FloatType::getF32(&globalContext());
   auto memrefType = MemRefType::get(
       {ShapedType::kDynamicSize, ShapedType::kDynamicSize}, f32Type, {}, 0);
   auto tensorType = RankedTensorType::get(
       {ShapedType::kDynamicSize, ShapedType::kDynamicSize}, f32Type);
-  auto f = makeFunction("linalg_tensors", {}, {tensorType, memrefType});
+  auto f =
+      makeFunction("linalg_tensors", {}, {tensorType, memrefType, tensorType});
 
   OpBuilder builder(f.getBody());
   ScopedContext scope(builder, f.getLoc());
-  Value A(f.getArgument(0)), B(f.getArgument(1));
+  Value A(f.getArgument(0)), B(f.getArgument(1)), C(f.getArgument(2));
   AffineExpr i, j;
   bindDims(&globalContext(), i, j);
-  StructuredIndexed SA(A), SB(B), SC(tensorType);
+  StructuredIndexed SA(A), SB(B), SC(C);
   Value added = linalg_generic_pointwise_add(SA({i, j}), SB({i, j}), SC({i, j}))
                     ->getResult(0);
   Value maxed = linalg_generic_pointwise_max(
                     SA({i, j}), StructuredIndexed(added)({i, j}), SC({i, j}))
                     ->getResult(0);
   Value tanhed = linalg_generic_pointwise_tanh(StructuredIndexed(maxed)({i, j}),
                                                SC({i, j}))
                      ->getResult(0);
TEST_FUNC(builder_loop_for_yield) {
   Value init1 = std_constant_float(llvm::APFloat(2.0f), f32Type);
   Value a(f.getArgument(0)), b(f.getArgument(1)), c(f.getArgument(2)),
       d(f.getArgument(3));
   using namespace edsc::op;
   auto results = loopNestBuilder(a - b, c + d, a, {init0, init1},
                                  [&](Value iv, ValueRange args) {
                                    Value sum = args[0] + args[1];
                                    return scf::ValueVector{args[1], sum};
-                                 }).getResults();
+                                 })
+                     .getResults();
   results[0] + results[1];
 
   // clang-format off
   // CHECK-LABEL: func @builder_loop_for_yield(%{{.*}}: index, %{{.*}}: index, %{{.*}}: index, %{{.*}}: index) {
   // CHECK:     [[init0:%.*]] = constant
   // CHECK:     [[init1:%.*]] = constant
   // CHECK-DAG:    [[r0:%[0-9]+]] = affine.apply affine_map<()[s0, s1] -> (s0 - s1)>()[%{{.*}}, %{{.*}}]
   // CHECK-DAG:    [[r1:%[0-9]+]] = affine.apply affine_map<()[s0, s1] -> (s0 + s1)>()[%{{.*}}, %{{.*}}]

mlir/test/mlir-linalg-ods-gen/test-linalg-ods-gen.tc

 // RUN: mlir-linalg-ods-gen %s -gen-ods-decl=1 | FileCheck %s --check-prefix=ODS
 // RUN: mlir-linalg-ods-gen %s -gen-impl=1 | FileCheck %s --check-prefix=IMPL
 
 // ODS-LABEL: def Test1Op : LinalgStructuredBase_Op<"test1", [
 //  ODS-NEXT:   AttrSizedOperandSegments
 //  ODS-NEXT:   DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
-//  ODS-NEXT:   NamedStructuredOpTrait
 //  ODS-NEXT:   SingleBlockImplicitTerminator<"YieldOp">
 //
 // IMPL-LABEL:  ArrayAttr Test1Op::iterator_types() {
 //       IMPL:  { {{.*}}Parallel{{.*}}, {{.*}}Reduction{{.*}} }
 //
 //       IMPL:  ArrayAttr Test1Op::indexing_maps() {
 //       IMPL:  AffineMap::get(2, 0, {d0, d1}, context),
 //  IMPL-NEXT:  AffineMap::get(2, 0, {d1}, context),
 //  IMPL-NEXT:  AffineMap::get(2, 0, {d0}, context) });
 //
 //       IMPL:  void Test1Op::regionBuilder(Block &block) {
 //       IMPL:  Value [[a:.*]](args[0]), [[b:.*]](args[1]), [[c:.*]](args[2]);
 //       IMPL:  Value [[d:.*]] = std_mulf([[a]], [[b]]);
 //       IMPL:  Value [[e:.*]] = std_addf([[c]], [[d]]);
 //       IMPL:  (linalg_yield(ValueRange{ [[e]] }));
 //
 ods_def<Test1Op> :
 def test1(A: f32(M, K), B: f32(K)) -> (C: f32(M)) {
   C(m) = std_addf<k>(std_mulf(A(m, k), B(k)));
 }
 
 // ODS-LABEL: def Test2Op : LinalgStructuredBase_Op<"test2", [
 //  ODS-NEXT:   AttrSizedOperandSegments
 //  ODS-NEXT:   DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
-//  ODS-NEXT:   NamedStructuredOpTrait
 //  ODS-NEXT:   SingleBlockImplicitTerminator<"YieldOp">
 //
 // IMPL-LABEL:  ArrayAttr Test2Op::iterator_types() {
 //       IMPL:  { {{.*}}Parallel{{.*}}, {{.*}}Parallel{{.*}}, {{.*}}Reduction{{.*}} }
 //
 //       IMPL:  ArrayAttr Test2Op::indexing_maps() {
 //       IMPL:  AffineMap::get(3, 0, {d0, d2}, context),
 //  IMPL-NEXT:  AffineMap::get(3, 0, {d2, d1}, context),
 //  IMPL-NEXT:  AffineMap::get(3, 0, {d0, d1}, context) });
 //
 //       IMPL:  Test2Op::regionBuilder(Block &block) {
 //       IMPL:  Value [[a:.*]](args[0]), [[b:.*]](args[1]), [[c:.*]](args[2]);
 //       IMPL:  Value [[d:.*]] = std_mulf([[a]], [[b]]);
 //       IMPL:  Value [[e:.*]] = std_addf([[c]], [[d]]);
 //       IMPL:  (linalg_yield(ValueRange{ [[e]] }));
 //
 ods_def<Test2Op> :
 def test2(A: f32(M, K), B: f32(K, N)) -> (C: f32(M, N)) {
   C(m, n) = std_addf<k>(std_mulf(A(m, k), B(k, n)));
 }
 
 // ODS-LABEL: def Test3Op : LinalgStructuredBase_Op<"test3", [
 //  ODS-NEXT:   AttrSizedOperandSegments
 //  ODS-NEXT:   DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
-//  ODS-NEXT:   NamedStructuredOpTrait
 //  ODS-NEXT:   SingleBlockImplicitTerminator<"YieldOp">
 //
 // IMPL-LABEL:  ArrayAttr Test3Op::iterator_types() {
 //       IMPL:  { {{.*}}Parallel{{.*}}, {{.*}}Parallel{{.*}}, {{.*}}Reduction{{.*}} }
 //
 //       IMPL:  ArrayAttr Test3Op::indexing_maps() {
 //       IMPL:  AffineMap::get(4, 0, {d0, d1, d3}, context),
 //  IMPL-NEXT:  AffineMap::get(4, 0, {d3, d2}, context),

mlir/tools/mlir-linalg-ods-gen/mlir-linalg-ods-gen.cpp

 
 /// Print the ODS class that defines a new `cppOpName` for a `linalgOpName`.
 void TCParser::printODS(llvm::raw_ostream &os, StringRef cppOpName,
                         StringRef linalgOpName,
                         ComprehensionParsingState &state) {
   const char *header = R"FMT(  def {0} : LinalgStructuredBase_Op<"{1}", [
     AttrSizedOperandSegments,
     DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
-    NamedStructuredOpTrait,
     SingleBlockImplicitTerminator<"YieldOp">]> {
       let arguments = (ins Variadic<AnyShaped>:$inputs,
-                           Variadic<AnyMemRef>:$output_buffers,
-                           Variadic<AnyRankedTensor>:$init_tensors);
+                           Variadic<AnyShaped>:$outputs);
       let results = (outs Variadic<AnyRankedTensor>:$result_tensors);
       let regions = (region AnyRegion:$region);
 
       let skipDefaultBuilders = 1;
       let builders = [ OpBuilderDAG<
-        (ins "ValueRange":$inputs, "ValueRange":$outputBuffers),
+        (ins "ValueRange":$inputs, "ValueRange":$outputs),
         [{{
           $_state.addOperands(inputs);
-          $_state.addOperands(outputBuffers);
+          $_state.addOperands(outputs);
           $_state.addAttribute(
             "operand_segment_sizes",
             $_builder.getI32VectorAttr({{
               static_cast<int32_t>(inputs.size()),
-              static_cast<int32_t>(outputBuffers.size()),
-              static_cast<int32_t>(0)}));
+              static_cast<int32_t>(outputs.size())}));
           buildNamedStructuredOpRegionAndAttributes<{0}>(
             $_builder,
             $_state,
             TypeRange(inputs),
-            TypeRange(outputBuffers),
-            TypeRange(),
-            TypeRange());
+            TypeRange(outputs));
         }]>, OpBuilderDAG<
         (ins "TypeRange":$resultTensorTypes, "ValueRange":$inputs,
-             "ValueRange":$outputBuffers, "ValueRange":$initTensors),
+             "ValueRange":$outputs),
         [{{
           $_state.addOperands(inputs);
-          $_state.addOperands(outputBuffers);
-          $_state.addOperands(initTensors);
+          $_state.addOperands(outputs);
           $_state.addTypes(resultTensorTypes);
           $_state.addAttribute(
             "operand_segment_sizes",
             $_builder.getI32VectorAttr({{
               static_cast<int32_t>(inputs.size()),
-              static_cast<int32_t>(outputBuffers.size()),
-              static_cast<int32_t>(initTensors.size())}));
+              static_cast<int32_t>(outputs.size())}));
           buildNamedStructuredOpRegionAndAttributes<{0}>(
             $_builder,
             $_state,
             TypeRange(inputs),
-            TypeRange(outputBuffers),
-            TypeRange(initTensors),
-            resultTensorTypes);
+            TypeRange(outputs));
         }]>, OpBuilderDAG<
         (ins "TypeRange":$resultTensorTypes, "ValueRange":$operands,
              CArg<"ArrayRef<NamedAttribute>", "{{}">:$attributes),
         [{{
           $_state.addOperands(operands);
           $_state.addAttributes(attributes);
           $_state.addTypes(resultTensorTypes);
           (void)$_state.addRegion();
         }]>
       ];
       let printer = [{{ return ::printNamedStructuredOp(p, *this); }];
       let parser = [{{ return ::parseNamedStructuredOp<{0}>(parser, result); }];
-      let verifier = [{{ return ::verifyNamedStructuredOp(*this); }];
       let hasFolder = 1;
       let hasCanonicalizer = 1;
 
       let extraClassDeclaration = [{{
         // Auto-generated.
         ArrayAttr iterator_types();
         ArrayAttr indexing_maps();
         static void regionBuilder(Block &block);