[mlir] Add an interface for Cast-Like operations

[rriddle]

Summary

A cast-like operation is one that converts from a set of input types to a set of output types. The arity of the inputs may be from 0-N, whereas the arity of the outputs may be anything from 1-N. Cast-like operations are removable in cases where they produce a "no-op", i.e when the input types and output types match 1-1.

[silvas]

mlir/lib/IR/Operation.cpp

1229

This fold doesn't seem right, as many things that we could consider to be "casts" can be lossy. for example, consider if sitofp and fptosi were a single op (like we do for index_cast an others). Then this could fold away an intermediate truncation to integer.

Thoughts?

[rriddle]

mlir/lib/IR/Operation.cpp

1229

Thanks for the catch, I wrote the builtin op first and then just copied it over. This shouldn't be here.

[silvas]

mlir/include/mlir/Interfaces/CastInterfaces.td

23

what is the 0 case? That just seems more like a constant op.

[rriddle]

mlir/include/mlir/Interfaces/CastInterfaces.td

23

I was thinking of the cases that happen during dialect conversion. During conversion you can end up with a 1->0 conversion (most often during function signature conversion), i.e. where a type gets dropped. If uses of the original value still linger around, we have to insert a cast operation that temporarily materializes that value.

[silvas]

Thanks for cleaning this up!

mlir/test/IR/invalid-ops.mlir

1077 ↗

(On Diff #317082)

nit: kind of confusing to see this plural but only used for one type. Would suggest saying type(s).

Also, I don't see a test for the multiple-result case, multiple operand case, or 0-operand case. I don't think the anycast patch will include it (it allows all type combinations, right?), so it might be worth adding a test op.

[rriddle]

Fix accidental incorrect behavior for casts

[lattner]

nice!

mlir/docs/Tutorials/Toy/Ch-4.md

224

Can't this code be simpler, something like:

return input && output && input == output && input.hasRank())

since the tensor types are uniqued? You only need the more complicated code if you're using verifyCompatibleShape instead of shape equality IIRC.

mlir/examples/toy/Ch4/mlir/Dialect.cpp

249

same code

mlir/examples/toy/Ch5/mlir/Dialect.cpp

249

same code

mlir/examples/toy/Ch6/mlir/Dialect.cpp

250

same code

mlir/examples/toy/Ch7/mlir/Dialect.cpp

301

same code

mlir/include/mlir/Dialect/StandardOps/IR/Ops.td

48–49

btw Paul taught me that !listconcat(x, y) can be spelled as x # y which is much nicer for this sort of use case

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

2740

Random question, but should this be an assert that input/output.size() == 1 instead of a check?

Shouldn't this be called only when the arity is already checked by the verifier?

[rriddle]

Address comments

[rriddle]

mlir/include/mlir/Dialect/StandardOps/IR/Ops.td

48–49

Nice!

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

2740

These methods are sometimes called from outside of the verifier. That coupled with the fact that these are now interface methods, I figured it's better to gracefully fail instead of assert.

[lattner]

ok

[herhut]

Nice cleanup, thanks!

mlir/docs/Tutorials/Toy/Ch-4.md

216

nit: They - The

mlir/examples/toy/Ch4/mlir/Dialect.cpp

241

nit: They -> The

mlir/examples/toy/Ch5/mlir/Dialect.cpp

241

Same nit here and in every copy :)

[rriddle]

Address feedback

mlir/docs/Tutorials/Toy/Ch-4.md

 piece: there is a hidden type conversion on the edge of the call. If we look at
 the above, the operands to the generic_call are of type `tensor<2x3xf64>`, while
 the inputs to the function expect `tensor<*xf64>`. To resolve this difference,
 the inliner expects an explicit cast operation to be inserted. For this, we need
 to add a new operation to the Toy dialect, `ToyCastOp`(toy.cast), to represent
 casts between two different shapes.
 
 ```tablegen
-def CastOp : Toy_Op<"cast", [NoSideEffect, SameOperandsAndResultShape]> {
+def CastOp : Toy_Op<"cast", [
+    DeclareOpInterfaceMethods<CastOpInterface>,
+    NoSideEffect,
+    SameOperandsAndResultShape]
+  > {
   let summary = "shape cast operation";
   let description = [{
     The "cast" operation converts a tensor from one type to an equivalent type
     without changing any data elements. The source and destination types
-    must both be tensor types with the same element type. If both are ranked
-    then the rank should be the same and static dimensions should match. The
-    operation is invalid if converting to a mismatching constant dimension.
+    must both be tensor types with the same element type. If both are ranked,
+    then shape is required to match. The operation is invalid if converting
+    to a mismatching constant dimension.
   }];
 
   let arguments = (ins F64Tensor:$input);
   let results = (outs F64Tensor:$output);
+}
+```
+
+Note that the definition of this cast operation adds a `CastOpInterface` to the
+traits list. This interface provides several utilities for cast-like operation,
+such as folding identity casts and verification. We hook into this interface by
+providing a definition for the `areCastCompatible` method:
 
-  // Set the folder bit so that we can fold redundant cast operations.
-  let hasFolder = 1;
+```c++
+/// Returns true if the given set of input and result types are compatible with
+/// this cast operation. This is required by the `CastOpInterface` to verify
+/// this operation and provide other additional utilities.
+bool CastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  // The inputs must be Tensors with the same element type.
+  TensorType input = inputs.front().dyn_cast<TensorType>();
+  TensorType output = outputs.front().dyn_cast<TensorType>();
+  if (!input || !output || input.getElementType() != output.getElementType())
+    return false;
+  // The shape is required to match if both types are ranked.
+  return !input.hasRank() || !output.hasRank() || input == output;
 }
+
 ```
 
-We can then override the necessary hook on the ToyInlinerInterface to insert
-this for us when necessary:
+With a proper cast operation, we can now override the necessary hook on the
+ToyInlinerInterface to insert it for us when necessary:
 
 ```c++
 struct ToyInlinerInterface : public DialectInlinerInterface {
   ...
 
   /// Attempts to materialize a conversion for a type mismatch between a call
   /// from this dialect, and a callable region. This method should generate an
   /// operation that takes 'input' as the only operand, and produces a single

mlir/examples/toy/Ch4/CMakeLists.txt

add_toy_chapter(toyc-ch4
   ToyCh4CombineIncGen
   )
 
 include_directories(${CMAKE_CURRENT_BINARY_DIR})
 include_directories(${CMAKE_CURRENT_BINARY_DIR}/include/)
 target_link_libraries(toyc-ch4
   PRIVATE
     MLIRAnalysis
+    MLIRCastInterfaces
     MLIRCallInterfaces
     MLIRIR
     MLIRParser
     MLIRPass
     MLIRSideEffectInterfaces
     MLIRTransforms)
 

mlir/examples/toy/Ch4/include/toy/Dialect.h

 //===----------------------------------------------------------------------===//
 
 #ifndef MLIR_TUTORIAL_TOY_DIALECT_H_
 #define MLIR_TUTORIAL_TOY_DIALECT_H_
 
 #include "mlir/IR/BuiltinOps.h"
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Dialect.h"
+#include "mlir/Interfaces/CastInterfaces.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 #include "toy/ShapeInferenceInterface.h"
 
 namespace mlir {
 namespace toy {
 
 /// This is the definition of the Toy dialect. A dialect inherits from
 /// mlir::Dialect and registers custom attributes, operations, and types (in its

mlir/examples/toy/Ch4/include/toy/Ops.td

 //===- Ops.td - Toy dialect operation definitions ----------*- tablegen -*-===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // Defines the operations of the Toy dialect.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef TOY_OPS
 #define TOY_OPS
 
 include "mlir/Interfaces/CallInterfaces.td"
+include "mlir/Interfaces/CastInterfaces.td"
 include "mlir/Interfaces/SideEffectInterfaces.td"
 include "toy/ShapeInferenceInterface.td"
 
 // Provide a definition of the 'toy' dialect in the ODS framework so that we
 // can define our operations.
 def Toy_Dialect : Dialect {
   let name = "toy";
   let cppNamespace = "::mlir::toy";
def AddOp : Toy_Op<"add",
   let printer = [{ return ::printBinaryOp(p, *this); }];
 
   // Allow building an AddOp with from the two input operands.
   let builders = [
     OpBuilderDAG<(ins "Value":$lhs, "Value":$rhs)>
   ];
 }
 
-def CastOp : Toy_Op<"cast",
-    [DeclareOpInterfaceMethods<ShapeInferenceOpInterface>, NoSideEffect,
-     SameOperandsAndResultShape]> {
+def CastOp : Toy_Op<"cast", [
+     DeclareOpInterfaceMethods<CastOpInterface>,
+     DeclareOpInterfaceMethods<ShapeInferenceOpInterface>,
+     NoSideEffect,
+     SameOperandsAndResultShape
+  ]> {
   let summary = "shape cast operation";
   let description = [{
     The "cast" operation converts a tensor from one type to an equivalent type
-    without changing any data elements. The source and destination types
-    must both be tensor types with the same element type. If both are ranked
-    then the rank should be the same and static dimensions should match. The
-    operation is invalid if converting to a mismatching constant dimension.
+    without changing any data elements. The source and destination types must
+    both be tensor types with the same element type. If both are ranked, then
+    shape is required to match. The operation is invalid if converting to a
+    mismatching constant dimension.
   }];
 
   let arguments = (ins F64Tensor:$input);
   let results = (outs F64Tensor:$output);
 
   let assemblyFormat = "$input attr-dict `:` type($input) `to` type($output)";
-
-  // Set the folder bit so that we can fold redundant cast operations.
-  let hasFolder = 1;
 }
 
 def GenericCallOp : Toy_Op<"generic_call",
     [DeclareOpInterfaceMethods<CallOpInterface>]> {
   let summary = "generic call operation";
   let description = [{
     Generic calls represent calls to a user defined function that needs to
     be specialized for the shape of its arguments. The callee name is attached

mlir/examples/toy/Ch4/mlir/Dialect.cpp

 
 //===----------------------------------------------------------------------===//
 // CastOp
 
 /// Infer the output shape of the CastOp, this is required by the shape
 /// inference interface.
 void CastOp::inferShapes() { getResult().setType(getOperand().getType()); }
 
+/// Returns true if the given set of input and result types are compatible with
+/// this cast operation. This is required by the `CastOpInterface` to verify
+/// this operation and provide other additional utilities.
+bool CastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  // The inputs must be Tensors with the same element type.
+  TensorType input = inputs.front().dyn_cast<TensorType>();
+  TensorType output = outputs.front().dyn_cast<TensorType>();
+  if (!input || !output || input.getElementType() != output.getElementType())
+    return false;
+  // The shape is required to match if both types are ranked.
+  return !input.hasRank() || !output.hasRank() || input == output;
+}
+
 //===----------------------------------------------------------------------===//
 // GenericCallOp
 
 void GenericCallOp::build(mlir::OpBuilder &builder, mlir::OperationState &state,
                           StringRef callee, ArrayRef<mlir::Value> arguments) {
   // Generic call always returns an unranked Tensor initially.
   state.addTypes(UnrankedTensorType::get(builder.getF64Type()));
   state.addOperands(arguments);

mlir/examples/toy/Ch4/mlir/ToyCombine.cpp

 using namespace mlir;
 using namespace toy;
 
 namespace {
 /// Include the patterns defined in the Declarative Rewrite framework.
 #include "ToyCombine.inc"
 } // end anonymous namespace
 
-/// Fold simple cast operations that return the same type as the input.
-OpFoldResult CastOp::fold(ArrayRef<Attribute> operands) {
-  return mlir::impl::foldCastOp(*this);
-}
-
 /// This is an example of a c++ rewrite pattern for the TransposeOp. It
 /// optimizes the following scenario: transpose(transpose(x)) -> x
 struct SimplifyRedundantTranspose : public mlir::OpRewritePattern<TransposeOp> {
   /// We register this pattern to match every toy.transpose in the IR.
   /// The "benefit" is used by the framework to order the patterns and process
   /// them in order of profitability.
   SimplifyRedundantTranspose(mlir::MLIRContext *context)
       : OpRewritePattern<TransposeOp>(context, /*benefit=*/1) {}

mlir/examples/toy/Ch5/CMakeLists.txt

 include_directories(${CMAKE_CURRENT_BINARY_DIR})
 include_directories(${CMAKE_CURRENT_BINARY_DIR}/include/)
 get_property(dialect_libs GLOBAL PROPERTY MLIR_DIALECT_LIBS)
 target_link_libraries(toyc-ch5
   PRIVATE
     ${dialect_libs}
     MLIRAnalysis
     MLIRCallInterfaces
+    MLIRCastInterfaces
     MLIRIR
     MLIRParser
     MLIRPass
     MLIRSideEffectInterfaces
     MLIRSupport
     MLIRTransforms)

mlir/examples/toy/Ch5/include/toy/Dialect.h

 //===----------------------------------------------------------------------===//
 
 #ifndef MLIR_TUTORIAL_TOY_DIALECT_H_
 #define MLIR_TUTORIAL_TOY_DIALECT_H_
 
 #include "mlir/IR/BuiltinOps.h"
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Dialect.h"
+#include "mlir/Interfaces/CastInterfaces.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 #include "toy/ShapeInferenceInterface.h"
 
 namespace mlir {
 namespace toy {
 
 /// This is the definition of the Toy dialect. A dialect inherits from
 /// mlir::Dialect and registers custom attributes, operations, and types (in its

mlir/examples/toy/Ch5/include/toy/Ops.td

 //===- Ops.td - Toy dialect operation definitions ----------*- tablegen -*-===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // Defines the operations of the Toy dialect.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef TOY_OPS
 #define TOY_OPS
 
 include "mlir/Interfaces/CallInterfaces.td"
+include "mlir/Interfaces/CastInterfaces.td"
 include "mlir/Interfaces/SideEffectInterfaces.td"
 include "toy/ShapeInferenceInterface.td"
 
 // Provide a definition of the 'toy' dialect in the ODS framework so that we
 // can define our operations.
 def Toy_Dialect : Dialect {
   let name = "toy";
   let cppNamespace = "::mlir::toy";
def AddOp : Toy_Op<"add",
   let printer = [{ return ::printBinaryOp(p, *this); }];
 
   // Allow building an AddOp with from the two input operands.
   let builders = [
     OpBuilderDAG<(ins "Value":$lhs, "Value":$rhs)>
   ];
 }
 
-def CastOp : Toy_Op<"cast",
-    [DeclareOpInterfaceMethods<ShapeInferenceOpInterface>, NoSideEffect,
-     SameOperandsAndResultShape]> {
+def CastOp : Toy_Op<"cast", [
+     DeclareOpInterfaceMethods<CastOpInterface>,
+     DeclareOpInterfaceMethods<ShapeInferenceOpInterface>,
+     NoSideEffect,
+     SameOperandsAndResultShape
+  ]> {
   let summary = "shape cast operation";
   let description = [{
     The "cast" operation converts a tensor from one type to an equivalent type
-    without changing any data elements. The source and destination types
-    must both be tensor types with the same element type. If both are ranked
-    then the rank should be the same and static dimensions should match. The
-    operation is invalid if converting to a mismatching constant dimension.
+    without changing any data elements. The source and destination types must
+    both be tensor types with the same element type. If both are ranked, then
+    shape is required to match. The operation is invalid if converting to a
+    mismatching constant dimension.
   }];
 
   let arguments = (ins F64Tensor:$input);
   let results = (outs F64Tensor:$output);
 
   let assemblyFormat = "$input attr-dict `:` type($input) `to` type($output)";
-
-  // Set the folder bit so that we can fold redundant cast operations.
-  let hasFolder = 1;
 }
 
 def GenericCallOp : Toy_Op<"generic_call",
     [DeclareOpInterfaceMethods<CallOpInterface>]> {
   let summary = "generic call operation";
   let description = [{
     Generic calls represent calls to a user defined function that needs to
     be specialized for the shape of its arguments. The callee name is attached

mlir/examples/toy/Ch5/mlir/Dialect.cpp

 
 //===----------------------------------------------------------------------===//
 // CastOp
 
 /// Infer the output shape of the CastOp, this is required by the shape
 /// inference interface.
 void CastOp::inferShapes() { getResult().setType(getOperand().getType()); }
 
+/// Returns true if the given set of input and result types are compatible with
+/// this cast operation. This is required by the `CastOpInterface` to verify
+/// this operation and provide other additional utilities.
+bool CastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  // The inputs must be Tensors with the same element type.
+  TensorType input = inputs.front().dyn_cast<TensorType>();
+  TensorType output = outputs.front().dyn_cast<TensorType>();
+  if (!input || !output || input.getElementType() != output.getElementType())
+    return false;
+  // The shape is required to match if both types are ranked.
+  return !input.hasRank() || !output.hasRank() || input == output;
+}
+
 //===----------------------------------------------------------------------===//
 // GenericCallOp
 
 void GenericCallOp::build(mlir::OpBuilder &builder, mlir::OperationState &state,
                           StringRef callee, ArrayRef<mlir::Value> arguments) {
   // Generic call always returns an unranked Tensor initially.
   state.addTypes(UnrankedTensorType::get(builder.getF64Type()));
   state.addOperands(arguments);

mlir/examples/toy/Ch5/mlir/ToyCombine.cpp

 using namespace mlir;
 using namespace toy;
 
 namespace {
 /// Include the patterns defined in the Declarative Rewrite framework.
 #include "ToyCombine.inc"
 } // end anonymous namespace
 
-/// Fold simple cast operations that return the same type as the input.
-OpFoldResult CastOp::fold(ArrayRef<Attribute> operands) {
-  return mlir::impl::foldCastOp(*this);
-}
-
 /// This is an example of a c++ rewrite pattern for the TransposeOp. It
 /// optimizes the following scenario: transpose(transpose(x)) -> x
 struct SimplifyRedundantTranspose : public mlir::OpRewritePattern<TransposeOp> {
   /// We register this pattern to match every toy.transpose in the IR.
   /// The "benefit" is used by the framework to order the patterns and process
   /// them in order of profitability.
   SimplifyRedundantTranspose(mlir::MLIRContext *context)
       : OpRewritePattern<TransposeOp>(context, /*benefit=*/1) {}

mlir/examples/toy/Ch6/CMakeLists.txt

 get_property(dialect_libs GLOBAL PROPERTY MLIR_DIALECT_LIBS)
 get_property(conversion_libs GLOBAL PROPERTY MLIR_CONVERSION_LIBS)
 target_link_libraries(toyc-ch6
   PRIVATE
     ${dialect_libs}
     ${conversion_libs}
     MLIRAnalysis
     MLIRCallInterfaces
+    MLIRCastInterfaces
     MLIRExecutionEngine
     MLIRIR
     MLIRLLVMIR
     MLIRParser
     MLIRPass
     MLIRSideEffectInterfaces
     MLIRSupport
     MLIRTargetLLVMIR
     MLIRTransforms
     )

mlir/examples/toy/Ch6/include/toy/Dialect.h

 //===----------------------------------------------------------------------===//
 
 #ifndef MLIR_TUTORIAL_TOY_DIALECT_H_
 #define MLIR_TUTORIAL_TOY_DIALECT_H_
 
 #include "mlir/IR/BuiltinOps.h"
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Dialect.h"
+#include "mlir/Interfaces/CastInterfaces.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 #include "toy/ShapeInferenceInterface.h"
 
 namespace mlir {
 namespace toy {
 
 /// This is the definition of the Toy dialect. A dialect inherits from
 /// mlir::Dialect and registers custom attributes, operations, and types (in its

mlir/examples/toy/Ch6/include/toy/Ops.td

 //===- Ops.td - Toy dialect operation definitions ----------*- tablegen -*-===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // Defines the operations of the Toy dialect.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef TOY_OPS
 #define TOY_OPS
 
 include "mlir/Interfaces/CallInterfaces.td"
+include "mlir/Interfaces/CastInterfaces.td"
 include "mlir/Interfaces/SideEffectInterfaces.td"
 include "toy/ShapeInferenceInterface.td"
 
 // Provide a definition of the 'toy' dialect in the ODS framework so that we
 // can define our operations.
 def Toy_Dialect : Dialect {
   let name = "toy";
   let cppNamespace = "::mlir::toy";
def AddOp : Toy_Op<"add",
   let printer = [{ return ::printBinaryOp(p, *this); }];
 
   // Allow building an AddOp with from the two input operands.
   let builders = [
     OpBuilderDAG<(ins "Value":$lhs, "Value":$rhs)>
   ];
 }
 
-def CastOp : Toy_Op<"cast",
-    [DeclareOpInterfaceMethods<ShapeInferenceOpInterface>, NoSideEffect,
-     SameOperandsAndResultShape]> {
+def CastOp : Toy_Op<"cast", [
+     DeclareOpInterfaceMethods<CastOpInterface>,
+     DeclareOpInterfaceMethods<ShapeInferenceOpInterface>,
+     NoSideEffect,
+     SameOperandsAndResultShape
+  ]> {
   let summary = "shape cast operation";
   let description = [{
     The "cast" operation converts a tensor from one type to an equivalent type
-    without changing any data elements. The source and destination types
-    must both be tensor types with the same element type. If both are ranked
-    then the rank should be the same and static dimensions should match. The
-    operation is invalid if converting to a mismatching constant dimension.
+    without changing any data elements. The source and destination types must
+    both be tensor types with the same element type. If both are ranked, then
+    shape is required to match. The operation is invalid if converting to a
+    mismatching constant dimension.
   }];
 
   let arguments = (ins F64Tensor:$input);
   let results = (outs F64Tensor:$output);
 
   let assemblyFormat = "$input attr-dict `:` type($input) `to` type($output)";
-
-  // Set the folder bit so that we can fold redundant cast operations.
-  let hasFolder = 1;
 }
 
 def GenericCallOp : Toy_Op<"generic_call",
     [DeclareOpInterfaceMethods<CallOpInterface>]> {
   let summary = "generic call operation";
   let description = [{
     Generic calls represent calls to a user defined function that needs to
     be specialized for the shape of its arguments. The callee name is attached

mlir/examples/toy/Ch6/mlir/Dialect.cpp

 
 //===----------------------------------------------------------------------===//
 // CastOp
 
 /// Infer the output shape of the CastOp, this is required by the shape
 /// inference interface.
 void CastOp::inferShapes() { getResult().setType(getOperand().getType()); }
 
+/// Returns true if the given set of input and result types are compatible with
+/// this cast operation. This is required by the `CastOpInterface` to verify
+/// this operation and provide other additional utilities.
+bool CastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  // The inputs must be Tensors with the same element type.
+  TensorType input = inputs.front().dyn_cast<TensorType>();
+  TensorType output = outputs.front().dyn_cast<TensorType>();
+  if (!input || !output || input.getElementType() != output.getElementType())
+    return false;
+  // The shape is required to match if both types are ranked.
+  return !input.hasRank() || !output.hasRank() || input == output;
+}
+
 //===----------------------------------------------------------------------===//
 // GenericCallOp
 
 void GenericCallOp::build(mlir::OpBuilder &builder, mlir::OperationState &state,
                           StringRef callee, ArrayRef<mlir::Value> arguments) {
   // Generic call always returns an unranked Tensor initially.
   state.addTypes(UnrankedTensorType::get(builder.getF64Type()));
   state.addOperands(arguments);
   state.addAttribute("callee", builder.getSymbolRefAttr(callee));

mlir/examples/toy/Ch6/mlir/ToyCombine.cpp

 using namespace mlir;
 using namespace toy;
 
 namespace {
 /// Include the patterns defined in the Declarative Rewrite framework.
 #include "ToyCombine.inc"
 } // end anonymous namespace
 
-/// Fold simple cast operations that return the same type as the input.
-OpFoldResult CastOp::fold(ArrayRef<Attribute> operands) {
-  return mlir::impl::foldCastOp(*this);
-}
-
 /// This is an example of a c++ rewrite pattern for the TransposeOp. It
 /// optimizes the following scenario: transpose(transpose(x)) -> x
 struct SimplifyRedundantTranspose : public mlir::OpRewritePattern<TransposeOp> {
   /// We register this pattern to match every toy.transpose in the IR.
   /// The "benefit" is used by the framework to order the patterns and process
   /// them in order of profitability.
   SimplifyRedundantTranspose(mlir::MLIRContext *context)
       : OpRewritePattern<TransposeOp>(context, /*benefit=*/1) {}

mlir/examples/toy/Ch7/CMakeLists.txt

 get_property(dialect_libs GLOBAL PROPERTY MLIR_DIALECT_LIBS)
 get_property(conversion_libs GLOBAL PROPERTY MLIR_CONVERSION_LIBS)
 target_link_libraries(toyc-ch7
   PRIVATE
     ${dialect_libs}
     ${conversion_libs}
     MLIRAnalysis
     MLIRCallInterfaces
+    MLIRCastInterfaces
     MLIRExecutionEngine
     MLIRIR
     MLIRParser
     MLIRPass
     MLIRSideEffectInterfaces
     MLIRTargetLLVMIR
     MLIRTransforms
     )

mlir/examples/toy/Ch7/include/toy/Dialect.h

 //===----------------------------------------------------------------------===//
 
 #ifndef MLIR_TUTORIAL_TOY_DIALECT_H_
 #define MLIR_TUTORIAL_TOY_DIALECT_H_
 
 #include "mlir/IR/BuiltinOps.h"
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Dialect.h"
+#include "mlir/Interfaces/CastInterfaces.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 #include "toy/ShapeInferenceInterface.h"
 
 namespace mlir {
 namespace toy {
 namespace detail {
 struct StructTypeStorage;
 } // end namespace detail

mlir/examples/toy/Ch7/include/toy/Ops.td

 //===- Ops.td - Toy dialect operation definitions ----------*- tablegen -*-===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // Defines the operations of the Toy dialect.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef TOY_OPS
 #define TOY_OPS
 
 include "mlir/Interfaces/CallInterfaces.td"
+include "mlir/Interfaces/CastInterfaces.td"
 include "mlir/Interfaces/SideEffectInterfaces.td"
 include "toy/ShapeInferenceInterface.td"
 
 // Provide a definition of the 'toy' dialect in the ODS framework so that we
 // can define our operations.
 def Toy_Dialect : Dialect {
   let name = "toy";
   let cppNamespace = "::mlir::toy";
def AddOp : Toy_Op<"add",
   let printer = [{ return ::printBinaryOp(p, *this); }];
 
   // Allow building an AddOp with from the two input operands.
   let builders = [
     OpBuilderDAG<(ins "Value":$lhs, "Value":$rhs)>
   ];
 }
 
-def CastOp : Toy_Op<"cast",
-    [DeclareOpInterfaceMethods<ShapeInferenceOpInterface>, NoSideEffect,
-     SameOperandsAndResultShape]> {
+def CastOp : Toy_Op<"cast", [
+     DeclareOpInterfaceMethods<CastOpInterface>,
+     DeclareOpInterfaceMethods<ShapeInferenceOpInterface>,
+     NoSideEffect,
+     SameOperandsAndResultShape
+  ]> {
   let summary = "shape cast operation";
   let description = [{
     The "cast" operation converts a tensor from one type to an equivalent type
-    without changing any data elements. The source and destination types
-    must both be tensor types with the same element type. If both are ranked
-    then the rank should be the same and static dimensions should match. The
-    operation is invalid if converting to a mismatching constant dimension.
+    without changing any data elements. The source and destination types must
+    both be tensor types with the same element type. If both are ranked, then
+    shape is required to match. The operation is invalid if converting to a
+    mismatching constant dimension.
   }];
 
   let arguments = (ins F64Tensor:$input);
   let results = (outs F64Tensor:$output);
 
   let assemblyFormat = "$input attr-dict `:` type($input) `to` type($output)";
-
-  // Set the folder bit so that we can fold redundant cast operations.
-  let hasFolder = 1;
 }
 
 def GenericCallOp : Toy_Op<"generic_call",
     [DeclareOpInterfaceMethods<CallOpInterface>]> {
   let summary = "generic call operation";
   let description = [{
     Generic calls represent calls to a user defined function that needs to
     be specialized for the shape of its arguments. The callee name is attached

mlir/examples/toy/Ch7/mlir/Dialect.cpp

 
 //===----------------------------------------------------------------------===//
 // CastOp
 
 /// Infer the output shape of the CastOp, this is required by the shape
 /// inference interface.
 void CastOp::inferShapes() { getResult().setType(getOperand().getType()); }
 
+/// Returns true if the given set of input and result types are compatible with
+/// this cast operation. This is required by the `CastOpInterface` to verify
+/// this operation and provide other additional utilities.
+bool CastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  // The inputs must be Tensors with the same element type.
+  TensorType input = inputs.front().dyn_cast<TensorType>();
+  TensorType output = outputs.front().dyn_cast<TensorType>();
+  if (!input || !output || input.getElementType() != output.getElementType())
+    return false;
+  // The shape is required to match if both types are ranked.
+  return !input.hasRank() || !output.hasRank() || input == output;
+}
+
 //===----------------------------------------------------------------------===//
 // GenericCallOp
 
 void GenericCallOp::build(mlir::OpBuilder &builder, mlir::OperationState &state,
                           StringRef callee, ArrayRef<mlir::Value> arguments) {
   // Generic call always returns an unranked Tensor initially.
   state.addTypes(UnrankedTensorType::get(builder.getF64Type()));
   state.addOperands(arguments);

mlir/examples/toy/Ch7/mlir/ToyCombine.cpp

 using namespace mlir;
 using namespace toy;
 
 namespace {
 /// Include the patterns defined in the Declarative Rewrite framework.
 #include "ToyCombine.inc"
 } // end anonymous namespace
 
-/// Fold simple cast operations that return the same type as the input.
-OpFoldResult CastOp::fold(ArrayRef<Attribute> operands) {
-  return mlir::impl::foldCastOp(*this);
-}
-
 /// Fold constants.
 OpFoldResult ConstantOp::fold(ArrayRef<Attribute> operands) { return value(); }
 
 /// Fold struct constants.
 OpFoldResult StructConstantOp::fold(ArrayRef<Attribute> operands) {
   return value();
 }
 

mlir/include/mlir/Dialect/StandardOps/IR/Ops.h

 #ifndef MLIR_DIALECT_STANDARDOPS_IR_OPS_H
 #define MLIR_DIALECT_STANDARDOPS_IR_OPS_H
 
 #include "mlir/IR/Builders.h"
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Dialect.h"
 #include "mlir/IR/OpImplementation.h"
 #include "mlir/Interfaces/CallInterfaces.h"
+#include "mlir/Interfaces/CastInterfaces.h"
 #include "mlir/Interfaces/ControlFlowInterfaces.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 #include "mlir/Interfaces/VectorInterfaces.h"
 #include "mlir/Interfaces/ViewLikeInterface.h"
 
 // Pull in all enum type definitions and utility function declarations.
 #include "mlir/Dialect/StandardOps/IR/OpsEnums.h.inc"
 

mlir/include/mlir/Dialect/StandardOps/IR/Ops.td

 
 #ifndef STANDARD_OPS
 #define STANDARD_OPS
 
 include "mlir/Dialect/StandardOps/IR/StandardOpsBase.td"
 include "mlir/IR/OpAsmInterface.td"
 include "mlir/IR/SymbolInterfaces.td"
 include "mlir/Interfaces/CallInterfaces.td"
+include "mlir/Interfaces/CastInterfaces.td"
 include "mlir/Interfaces/ControlFlowInterfaces.td"
 include "mlir/Interfaces/SideEffectInterfaces.td"
 include "mlir/Interfaces/VectorInterfaces.td"
 include "mlir/Interfaces/ViewLikeInterface.td"
 
 def StandardOps_Dialect : Dialect {
   let name = "std";
   let cppNamespace = "";
class Std_Op<string mnemonic, list<OpTrait> traits = []> :
   // functions.
   let printer = [{ return ::print(p, *this); }];
   let verifier = [{ return ::verify(*this); }];
   let parser = [{ return ::parse$cppClass(parser, result); }];
 }
 
 // Base class for standard cast operations. Requires single operand and result,
 // but does not constrain them to specific types.
 class CastOp<string mnemonic, list<OpTrait> traits = []> :
-    Std_Op<mnemonic,
-           !listconcat(traits, [NoSideEffect, SameOperandsAndResultShape])> {
-
+    Std_Op<mnemonic, traits # [
+      NoSideEffect, SameOperandsAndResultShape,
+      DeclareOpInterfaceMethods<CastOpInterface>
+    ]> {
   let results = (outs AnyType);
 
   let builders = [
     OpBuilderDAG<(ins "Value":$source, "Type":$destType), [{
        impl::buildCastOp($_builder, $_state, source, destType);
     }]>
   ];
 
   let parser = [{
     return impl::parseCastOp(parser, result);
   }];
   let printer = [{
     return printStandardCastOp(this->getOperation(), p);
   }];
-  let verifier = [{ return impl::verifyCastOp(*this, areCastCompatible); }];
 
-  let hasFolder = 1;
+  // Cast operations are fully verified by its traits.
+  let verifier = ?;
 }
 
 // Base class for arithmetic cast operations.
 class ArithmeticCastOp<string mnemonic, list<OpTrait> traits = []> :
     CastOp<mnemonic,
     !listconcat(traits, [ElementwiseMappable,
                          DeclareOpInterfaceMethods<VectorUnrollOpInterface>])> {
 }

 
 def FPExtOp : ArithmeticCastOp<"fpext">, Arguments<(ins AnyType:$in)> {
   let summary = "cast from floating-point to wider floating-point";
   let description = [{
     Cast a floating-point value to a larger floating-point-typed value.
     The destination type must to be strictly wider than the source type.
     Only scalars are currently supported.
   }];
-
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-  }];
-
-  let hasFolder = 0;
 }
 
 //===----------------------------------------------------------------------===//
 // FPToSIOp
 //===----------------------------------------------------------------------===//
 
 def FPToSIOp : ArithmeticCastOp<"fptosi">, Arguments<(ins AnyType:$in)> {
   let summary = "cast from floating-point type to integer type";
   let description = [{
     Cast from a value interpreted as floating-point to the nearest (rounding
     towards zero) signed integer value.
   }];
-
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-  }];
-
-  let hasFolder = 0;
 }
 
 //===----------------------------------------------------------------------===//
 // FPToUIOp
 //===----------------------------------------------------------------------===//
 
 def FPToUIOp : ArithmeticCastOp<"fptoui">, Arguments<(ins AnyType:$in)> {
   let summary = "cast from floating-point type to integer type";
   let description = [{
     Cast from a value interpreted as floating-point to the nearest (rounding
     towards zero) unsigned integer value.
   }];
-
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-  }];
-
-  let hasFolder = 0;
 }
 
 //===----------------------------------------------------------------------===//
 // FPTruncOp
 //===----------------------------------------------------------------------===//
 
 def FPTruncOp : ArithmeticCastOp<"fptrunc">, Arguments<(ins AnyType:$in)> {
   let summary = "cast from floating-point to narrower floating-point";
   let description = [{
     Truncate a floating-point value to a smaller floating-point-typed value.
     The destination type must be strictly narrower than the source type.
     If the value cannot be exactly represented, it is rounded using the default
     rounding mode. Only scalars are currently supported.
   }];
-
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-  }];
-
-  let hasFolder = 0;
 }
 
 //===----------------------------------------------------------------------===//
 // GlobalMemrefOp
 //===----------------------------------------------------------------------===//
 
 def GlobalMemrefOp : Std_Op<"global_memref", [Symbol]> {
   let summary = "declare or define a global memref variable";

 def IndexCastOp : ArithmeticCastOp<"index_cast">, Arguments<(ins AnyType:$in)> {
   let summary = "cast between index and integer types";
   let description = [{
     Casts between integer scalars and 'index' scalars. Index is an integer of
     platform-specific bit width. If casting to a wider integer, the value is
     sign-extended. If casting to a narrower integer, the value is truncated.
   }];
 
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-  }];
-
   let hasFolder = 1;
 }
 
 //===----------------------------------------------------------------------===//
 // LoadOp
 //===----------------------------------------------------------------------===//
 
 def LoadOp : Std_Op<"load",
let description = [{
     Erase rank information.
         %5 = memref_cast %1 : memref<4x?xf32> to memref<*xf32>
     ```
   }];
 
   let arguments = (ins AnyRankedOrUnrankedMemRef:$source);
   let results = (outs AnyRankedOrUnrankedMemRef);
 
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-
-    /// The result of a memref_cast is always a memref.
-    Type getType() { return getResult().getType(); }
-  }];
+  let hasFolder = 1;
 }
 
 
 //===----------------------------------------------------------------------===//
 // MemRefReinterpretCastOp
 //===----------------------------------------------------------------------===//
 
 def MemRefReinterpretCastOp:

 def SIToFPOp : ArithmeticCastOp<"sitofp">, Arguments<(ins AnyType:$in)> {
   let summary = "cast from integer type to floating-point";
   let description = [{
     Cast from a value interpreted as signed or vector of signed integers to the
     corresponding floating-point scalar or vector value. If the value cannot be
     exactly represented, it is rounded using the default rounding mode. Scalars
     and vector types are currently supported.
   }];
-
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-  }];
-
-  let hasFolder = 0;
 }
 
 //===----------------------------------------------------------------------===//
 // SplatOp
 //===----------------------------------------------------------------------===//
 
 def SplatOp : Std_Op<"splat", [NoSideEffect,
      TypesMatchWith<"operand type matches element type of result",

 def UIToFPOp : ArithmeticCastOp<"uitofp">, Arguments<(ins AnyType:$in)> {
   let summary = "cast from unsigned integer type to floating-point";
   let description = [{
     Cast from a value interpreted as unsigned integer or vector of unsigned
     integers to the corresponding scalar or vector floating-point value. If the
     value cannot be exactly represented, it is rounded using the default
     rounding mode. Scalars and vector types are currently supported.
   }];
-
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-  }];
-
-  let hasFolder = 0;
 }
 
 //===----------------------------------------------------------------------===//
 // UnsignedDivIOp
 //===----------------------------------------------------------------------===//
 
 def UnsignedDivIOp : IntArithmeticOp<"divi_unsigned"> {
   let summary = "unsigned integer division operation";

mlir/include/mlir/Dialect/Tensor/IR/Tensor.h

 //===- Tensor.h - Tensor dialect --------------------------------*- 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_TENSOR_IR_TENSOR_H_
 #define MLIR_DIALECT_TENSOR_IR_TENSOR_H_
 
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Dialect.h"
 #include "mlir/IR/OpDefinition.h"
 #include "mlir/IR/OpImplementation.h"
+#include "mlir/Interfaces/CastInterfaces.h"
 #include "mlir/Interfaces/ControlFlowInterfaces.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 
 //===----------------------------------------------------------------------===//
 // Tensor Dialect
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Dialect/Tensor/IR/TensorOpsDialect.h.inc"

mlir/include/mlir/Dialect/Tensor/IR/TensorOps.td

 //===- TensorOps.td - Tensor op definitions ----------------*- tablegen -*-===//
 //
 // 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 TENSOR_OPS
 #define TENSOR_OPS
 
 include "mlir/Dialect/Tensor/IR/TensorBase.td"
+include "mlir/Interfaces/CastInterfaces.td"
 include "mlir/Interfaces/ControlFlowInterfaces.td"
 include "mlir/Interfaces/SideEffectInterfaces.td"
 
 class Tensor_Op<string mnemonic, list<OpTrait> traits = []>
     : Op<Tensor_Dialect, mnemonic, traits> {
   let printer = [{ return ::print(p, *this); }];
   let verifier = [{ return ::verify(*this); }];
   let parser = [{ return ::parse$cppClass(parser, result); }];
 }
 
 //===----------------------------------------------------------------------===//
 // CastOp
 //===----------------------------------------------------------------------===//
 
-def Tensor_CastOp : Tensor_Op<"cast", [NoSideEffect]> {
+def Tensor_CastOp : Tensor_Op<"cast", [
+    DeclareOpInterfaceMethods<CastOpInterface>, NoSideEffect
+  ]> {
   let summary = "tensor cast operation";
   let description = [{
     Convert a tensor from one type to an equivalent type without changing any
     data elements. The source and destination types must both be tensor types
     with the same element type. If both are ranked, then the rank should be the
     same and static dimensions should match. The operation is invalid if
     converting to a mismatching constant dimension.
 
let description = [{
     %4 = tensor.cast %3 : tensor<4x?xf32> to tensor<?x?xf32>
     %5 = tensor.cast %4 : tensor<?x?xf32> to tensor<*xf32>
     ```
   }];
 
   let arguments = (ins AnyTensor:$source);
   let results = (outs AnyTensor:$dest);
   let assemblyFormat = "$source attr-dict `:` type($source) `to` type($dest)";
-  let verifier = "return impl::verifyCastOp(*this, areCastCompatible);";
-
-  let extraClassDeclaration = [{
-    /// Return true if `a` and `b` are valid operand and result pairs for
-    /// the operation.
-    static bool areCastCompatible(Type a, Type b);
-
-    /// The result of a tensor.cast is always a tensor.
-    TensorType getType() { return getResult().getType().cast<TensorType>(); }
-  }];
 
-  let hasFolder = 1;
   let hasCanonicalizer = 1;
+  let verifier = ?;
 }
 
 //===----------------------------------------------------------------------===//
 // ExtractOp
 //===----------------------------------------------------------------------===//
 
 def Tensor_ExtractOp : Tensor_Op<"extract",
     [NoSideEffect,

mlir/include/mlir/IR/Diagnostics.h

 
 //===----------------------------------------------------------------------===//
 // DiagnosticArgument
 //===----------------------------------------------------------------------===//
 
 /// A variant type that holds a single argument for a diagnostic.
 class DiagnosticArgument {
 public:
+  /// Note: The constructors below are only exposed due to problems accessing
+  /// constructors from type traits, they should not be used directly by users.
+  // Construct from an Attribute.
+  explicit DiagnosticArgument(Attribute attr);
+  // Construct from a floating point number.
+  explicit DiagnosticArgument(double val)
+      : kind(DiagnosticArgumentKind::Double), doubleVal(val) {}
+  explicit DiagnosticArgument(float val) : DiagnosticArgument(double(val)) {}
+  // Construct from a signed integer.
+  template <typename T>
+  explicit DiagnosticArgument(
+      T val, typename std::enable_if<std::is_signed<T>::value &&
+                                     std::numeric_limits<T>::is_integer &&
+                                     sizeof(T) <= sizeof(int64_t)>::type * = 0)
+      : kind(DiagnosticArgumentKind::Integer), opaqueVal(int64_t(val)) {}
+  // Construct from an unsigned integer.
+  template <typename T>
+  explicit DiagnosticArgument(
+      T val, typename std::enable_if<std::is_unsigned<T>::value &&
+                                     std::numeric_limits<T>::is_integer &&
+                                     sizeof(T) <= sizeof(uint64_t)>::type * = 0)
+      : kind(DiagnosticArgumentKind::Unsigned), opaqueVal(uint64_t(val)) {}
+  // Construct from a string reference.
+  explicit DiagnosticArgument(StringRef val)
+      : kind(DiagnosticArgumentKind::String), stringVal(val) {}
+  // Construct from a Type.
+  explicit DiagnosticArgument(Type val);
+
   /// Enum that represents the different kinds of diagnostic arguments
   /// supported.
   enum class DiagnosticArgumentKind {
     Attribute,
     Double,
     Integer,
     String,
     Type,
public:
   uint64_t getAsUnsigned() const {
     assert(getKind() == DiagnosticArgumentKind::Unsigned);
     return static_cast<uint64_t>(opaqueVal);
   }
 
 private:
   friend class Diagnostic;
 
-  // Construct from an Attribute.
-  explicit DiagnosticArgument(Attribute attr);
-
-  // Construct from a floating point number.
-  explicit DiagnosticArgument(double val)
-      : kind(DiagnosticArgumentKind::Double), doubleVal(val) {}
-  explicit DiagnosticArgument(float val) : DiagnosticArgument(double(val)) {}
-
-  // Construct from a signed integer.
-  template <typename T>
-  explicit DiagnosticArgument(
-      T val, typename std::enable_if<std::is_signed<T>::value &&
-                                     std::numeric_limits<T>::is_integer &&
-                                     sizeof(T) <= sizeof(int64_t)>::type * = 0)
-      : kind(DiagnosticArgumentKind::Integer), opaqueVal(int64_t(val)) {}
-
-  // Construct from an unsigned integer.
-  template <typename T>
-  explicit DiagnosticArgument(
-      T val, typename std::enable_if<std::is_unsigned<T>::value &&
-                                     std::numeric_limits<T>::is_integer &&
-                                     sizeof(T) <= sizeof(uint64_t)>::type * = 0)
-      : kind(DiagnosticArgumentKind::Unsigned), opaqueVal(uint64_t(val)) {}
-
-  // Construct from a string reference.
-  explicit DiagnosticArgument(StringRef val)
-      : kind(DiagnosticArgumentKind::String), stringVal(val) {}
-
-  // Construct from a Type.
-  explicit DiagnosticArgument(Type val);
-
   /// The kind of this argument.
   DiagnosticArgumentKind kind;
 
   /// The value of this argument.
   union {
     double doubleVal;
     intptr_t opaqueVal;
     StringRef stringVal;
public:
   Location getLocation() const { return loc; }
 
   /// Returns the current list of diagnostic arguments.
   MutableArrayRef<DiagnosticArgument> getArguments() { return arguments; }
   ArrayRef<DiagnosticArgument> getArguments() const { return arguments; }
 
   /// Stream operator for inserting new diagnostic arguments.
   template <typename Arg>
-  typename std::enable_if<!std::is_convertible<Arg, StringRef>::value,
-                          Diagnostic &>::type
+  typename std::enable_if<
+      !std::is_convertible<Arg, StringRef>::value &&
+          std::is_constructible<DiagnosticArgument, Arg>::value,
   operator<<(Arg &&val) {
     arguments.push_back(DiagnosticArgument(std::forward<Arg>(val)));
     return *this;
   }
 
   /// Stream in a string literal.
   Diagnostic &operator<<(const char *val) {
     arguments.push_back(DiagnosticArgument(val));
public:
 
   /// Stream in an Operation.
   Diagnostic &operator<<(Operation &val);
   Diagnostic &operator<<(Operation *val) {
     return *this << *val;
   }
 
   /// Stream in a range.
-  template <typename T> Diagnostic &operator<<(iterator_range<T> range) {
-    return appendRange(range);
-  }
-  template <typename T> Diagnostic &operator<<(ArrayRef<T> range) {
+  template <typename T, typename ValueT = llvm::detail::ValueOfRange<T>>
+  std::enable_if_t<!std::is_constructible<DiagnosticArgument, T>::value,
+                   Diagnostic &>
+  operator<<(T &&range) {
     return appendRange(range);
   }
 
   /// Append a range to the diagnostic. The default delimiter between elements
   /// is ','.
-  template <typename T, template <typename> class Container>
-  Diagnostic &appendRange(const Container<T> &c, const char *delim = ", ") {
+  template <typename T>
+  Diagnostic &appendRange(const T &c, const char *delim = ", ") {
     llvm::interleave(
         c, [this](const auto &a) { *this << a; }, [&]() { *this << delim; });
     return *this;
   }
 
   /// Append arguments to the diagnostic.
   template <typename Arg1, typename Arg2, typename... Args>
   Diagnostic &append(Arg1 &&arg1, Arg2 &&arg2, Args &&... args) {

mlir/include/mlir/IR/OpDefinition.h

ParseResult parseOneResultSameOperandTypeOp(OpAsmParser &parser,
                                             OperationState &result);
 
 // Prints the given binary `op` in custom assembly form if both the two operands
 // and the result have the same time. Otherwise, prints the generic assembly
 // form.
 void printOneResultOp(Operation *op, OpAsmPrinter &p);
 } // namespace impl
 
-// These functions are out-of-line implementations of the methods in CastOp,
-// which avoids them being template instantiated/duplicated.
+// These functions are out-of-line implementations of the methods in
+// CastOpInterface, which avoids them being template instantiated/duplicated.
 namespace impl {
+/// Attempt to fold the given cast operation.
+LogicalResult foldCastInterfaceOp(Operation *op,
+                                  ArrayRef<Attribute> attrOperands,
+                                  SmallVectorImpl<OpFoldResult> &foldResults);
+/// Attempt to verify the given cast operation.
+LogicalResult verifyCastInterfaceOp(
+    Operation *op, function_ref<bool(TypeRange, TypeRange)> areCastCompatible);
+
 // TODO: Remove the parse/print/build here (new ODS functionality obsoletes the
 // need for them, but some older ODS code in `std` still depends on them).
 void buildCastOp(OpBuilder &builder, OperationState &result, Value source,
                  Type destType);
 ParseResult parseCastOp(OpAsmParser &parser, OperationState &result);
 void printCastOp(Operation *op, OpAsmPrinter &p);
-// TODO: Create a CastOpInterface with a method areCastCompatible.
-// Also, consider adding functionality to CastOpInterface to be able to perform
-// the ChainedTensorCast canonicalization generically.
+// TODO: These methods are deprecated in favor of CastOpInterface. Remove them
+// when all uses have been updated. Also, consider adding functionality to
+// CastOpInterface to be able to perform the ChainedTensorCast canonicalization
+// generically.
 Value foldCastOp(Operation *op);
 LogicalResult verifyCastOp(Operation *op,
                            function_ref<bool(Type, Type)> areCastCompatible);
 } // namespace impl
 } // end namespace mlir
 
 #endif

mlir/include/mlir/Interfaces/CMakeLists.txt

 add_mlir_interface(CallInterfaces)
+add_mlir_interface(CastInterfaces)
 add_mlir_interface(ControlFlowInterfaces)
 add_mlir_interface(CopyOpInterface)
 add_mlir_interface(DerivedAttributeOpInterface)
 add_mlir_interface(InferTypeOpInterface)
 add_mlir_interface(LoopLikeInterface)
 add_mlir_interface(SideEffectInterfaces)
 add_mlir_interface(VectorInterfaces)
 add_mlir_interface(ViewLikeInterface)
 

mlir/include/mlir/Interfaces/CastInterfaces.h

+//===- CastInterfaces.h - Cast Interfaces for MLIR --------------*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the definitions of the cast interfaces defined in
+// `CastInterfaces.td`.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_INTERFACES_CASTINTERFACES_H
+#define MLIR_INTERFACES_CASTINTERFACES_H
+
+#include "mlir/IR/OpDefinition.h"
+
+/// Include the generated interface declarations.
+#include "mlir/Interfaces/CastInterfaces.h.inc"
+
+#endif // MLIR_INTERFACES_CASTINTERFACES_H

mlir/include/mlir/Interfaces/CastInterfaces.td

+//===- CastInterfaces.td - Cast Interfaces for ops ---------*- tablegen -*-===//
+//
+// 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 contains a set of interfaces that can be used to define information
+// related to cast-like operations.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_INTERFACES_CASTINTERFACES
+#define MLIR_INTERFACES_CASTINTERFACES
+
+include "mlir/IR/OpBase.td"
+
+def CastOpInterface : OpInterface<"CastOpInterface"> {
+  let description = [{
+    A cast-like operation is one that converts from a set of input types to a
+    set of output types. The arity of the inputs may be from 0-N, whereas the
+    arity of the outputs may be anything from 1-N. Cast-like operations are
+    trivially removable in cases where they produce an No-op, i.e when the
+    input types and output types match 1-1.
+  }];
+  let cppNamespace = "::mlir";
+
+  let methods = [
+    StaticInterfaceMethod<[{
+        Returns true if the given set of input and result types are compatible
+        to cast using this cast operation.
+      }],
+      "bool", "areCastCompatible",
+      (ins "mlir::TypeRange":$inputs, "mlir::TypeRange":$outputs)
+    >,
+  ];
+
+  let extraTraitClassDeclaration = [{
+    /// Attempt to fold the given cast operation.
+    static LogicalResult foldTrait(Operation *op, ArrayRef<Attribute> operands,
+                                   SmallVectorImpl<OpFoldResult> &results) {
+      return impl::foldCastInterfaceOp(op, operands, results);
+    }
+  }];
+  let verify = [{
+    return impl::verifyCastInterfaceOp($_op, ConcreteOp::areCastCompatible);
+  }];
+}
+
+#endif // MLIR_INTERFACES_CASTINTERFACES

mlir/lib/Dialect/Shape/IR/CMakeLists.txt

 set(LLVM_TARGET_DEFINITIONS ShapeCanonicalization.td)
 mlir_tablegen(ShapeCanonicalization.inc -gen-rewriters)
 add_public_tablegen_target(MLIRShapeCanonicalizationIncGen)
 
 add_mlir_dialect_library(MLIRShape
   Shape.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${MLIR_MAIN_INCLUDE_DIR}/mlir/Dialect/Shape
 
   DEPENDS
   MLIRShapeOpsIncGen
 
   LINK_LIBS PUBLIC
+  MLIRCastInterfaces
   MLIRControlFlowInterfaces
   MLIRDialect
   MLIRInferTypeOpInterface
   MLIRIR
   MLIRSideEffectInterfaces
   MLIRStandard
   MLIRTensor
   )

mlir/lib/Dialect/StandardOps/CMakeLists.txt

 add_mlir_dialect_library(MLIRStandard
   IR/Ops.cpp
   EDSC/Builders.cpp
   EDSC/Intrinsics.cpp
   Utils/Utils.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${MLIR_MAIN_INCLUDE_DIR}/mlir/Dialect/StandardOps
 
   DEPENDS
   MLIRStandardOpsIncGen
 
   LINK_LIBS PUBLIC
   MLIRCallInterfaces
+  MLIRCastInterfaces
   MLIRControlFlowInterfaces
   MLIREDSC
   MLIRIR
   MLIRSideEffectInterfaces
   MLIRTensor
   MLIRVectorInterfaces
   MLIRViewLikeInterface
   )
 
 add_subdirectory(Transforms)

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

 
 /// This method checks for cast compatibility of vector types.
 /// If 'a' and 'b' are vector types, and they are cast compatible,
 /// it calls the 'areElementsCastCompatible' function to check for
 /// element cast compatibility.
 /// Returns 'true' if the vector types are cast compatible,  and 'false'
 /// otherwise.
 static bool areVectorCastSimpleCompatible(
-    Type a, Type b, function_ref<bool(Type, Type)> areElementsCastCompatible) {
+    Type a, Type b,
+    function_ref<bool(TypeRange, TypeRange)> areElementsCastCompatible) {
   if (auto va = a.dyn_cast<VectorType>())
     if (auto vb = b.dyn_cast<VectorType>())
       return va.getShape().equals(vb.getShape()) &&
              areElementsCastCompatible(va.getElementType(),
                                        vb.getElementType());
   return false;
 }
 
LogicalResult DmaWaitOp::verify() {
 
   return success();
 }
 
 //===----------------------------------------------------------------------===//
 // FPExtOp
 //===----------------------------------------------------------------------===//
 
-bool FPExtOp::areCastCompatible(Type a, Type b) {
+bool FPExtOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   if (auto fa = a.dyn_cast<FloatType>())
     if (auto fb = b.dyn_cast<FloatType>())
       return fa.getWidth() < fb.getWidth();
   return areVectorCastSimpleCompatible(a, b, areCastCompatible);
 }
 
 //===----------------------------------------------------------------------===//
 // FPToSIOp
 //===----------------------------------------------------------------------===//
 
-bool FPToSIOp::areCastCompatible(Type a, Type b) {
+bool FPToSIOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   if (a.isa<FloatType>() && b.isSignlessInteger())
     return true;
   return areVectorCastSimpleCompatible(a, b, areCastCompatible);
 }
 
 //===----------------------------------------------------------------------===//
 // FPToUIOp
 //===----------------------------------------------------------------------===//
 
-bool FPToUIOp::areCastCompatible(Type a, Type b) {
+bool FPToUIOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   if (a.isa<FloatType>() && b.isSignlessInteger())
     return true;
   return areVectorCastSimpleCompatible(a, b, areCastCompatible);
 }
 
 //===----------------------------------------------------------------------===//
 // FPTruncOp
 //===----------------------------------------------------------------------===//
 
-bool FPTruncOp::areCastCompatible(Type a, Type b) {
+bool FPTruncOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   if (auto fa = a.dyn_cast<FloatType>())
     if (auto fb = b.dyn_cast<FloatType>())
       return fa.getWidth() > fb.getWidth();
   return areVectorCastSimpleCompatible(a, b, areCastCompatible);
 }
 
 //===----------------------------------------------------------------------===//
 // GlobalMemrefOp
GetGlobalMemrefOp::verifySymbolUses(SymbolTableCollection &symbolTable) {
   return success();
 }
 
 //===----------------------------------------------------------------------===//
 // IndexCastOp
 //===----------------------------------------------------------------------===//
 
 // Index cast is applicable from index to integer and backwards.
-bool IndexCastOp::areCastCompatible(Type a, Type b) {
+bool IndexCastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   if (a.isa<ShapedType>() && b.isa<ShapedType>()) {
     auto aShaped = a.cast<ShapedType>();
     auto bShaped = b.cast<ShapedType>();
 
     return (aShaped.getShape() == bShaped.getShape()) &&
            areCastCompatible(aShaped.getElementType(),
                              bShaped.getElementType());
   }

 }
 
 //===----------------------------------------------------------------------===//
 // MemRefCastOp
 //===----------------------------------------------------------------------===//
 
 Value MemRefCastOp::getViewSource() { return source(); }
 
-bool MemRefCastOp::areCastCompatible(Type a, Type b) {
+bool MemRefCastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   auto aT = a.dyn_cast<MemRefType>();
   auto bT = b.dyn_cast<MemRefType>();
 
   auto uaT = a.dyn_cast<UnrankedMemRefType>();
   auto ubT = b.dyn_cast<UnrankedMemRefType>();
 
   if (aT && bT) {
     if (aT.getElementType() != bT.getElementType())
if (aT && bT) {
 
     return true;
   }
 
   return false;
 }
 
 OpFoldResult MemRefCastOp::fold(ArrayRef<Attribute> operands) {
-  if (Value folded = impl::foldCastOp(*this))
-    return folded;
   return succeeded(foldMemRefCast(*this)) ? getResult() : Value();
 }
 
 //===----------------------------------------------------------------------===//
 // MemRefReinterpretCastOp
 //===----------------------------------------------------------------------===//
 
 void mlir::MemRefReinterpretCastOp::build(
OpFoldResult SignedRemIOp::fold(ArrayRef<Attribute> operands) {
   return IntegerAttr::get(lhs.getType(), lhs.getValue().srem(rhsValue));
 }
 
 //===----------------------------------------------------------------------===//
 // SIToFPOp
 //===----------------------------------------------------------------------===//
 
 // sitofp is applicable from integer types to float types.
-bool SIToFPOp::areCastCompatible(Type a, Type b) {
+bool SIToFPOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   if (a.isSignlessInteger() && b.isa<FloatType>())
     return true;
   return areVectorCastSimpleCompatible(a, b, areCastCompatible);
 }
 
 //===----------------------------------------------------------------------===//
 // SplatOp
 //===----------------------------------------------------------------------===//
return constFoldBinaryOp<IntegerAttr>(operands,
                                         [](APInt a, APInt b) { return a - b; });
 }
 
 //===----------------------------------------------------------------------===//
 // UIToFPOp
 //===----------------------------------------------------------------------===//
 
 // uitofp is applicable from integer types to float types.
-bool UIToFPOp::areCastCompatible(Type a, Type b) {
+bool UIToFPOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   if (a.isSignlessInteger() && b.isa<FloatType>())
     return true;
   return areVectorCastSimpleCompatible(a, b, areCastCompatible);
 }
 
 //===----------------------------------------------------------------------===//
 // SubViewOp
 //===----------------------------------------------------------------------===//

mlir/lib/Dialect/Tensor/IR/CMakeLists.txt

 add_mlir_dialect_library(MLIRTensor
   TensorDialect.cpp
   TensorOps.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${PROJECT_SOURCE_DIR}/include/mlir/Dialect/Tensor
 
   DEPENDS
   MLIRTensorOpsIncGen
 
   LINK_COMPONENTS
   Core
 
   LINK_LIBS PUBLIC
+  MLIRCastInterfaces
   MLIRIR
   MLIRSideEffectInterfaces
   MLIRSupport
   )

mlir/lib/Dialect/Tensor/IR/TensorOps.cpp

for (auto t : llvm::zip(sourceType.getShape(), resultType.getShape())) {
     if (ShapedType::isDynamic(std::get<0>(t)) &&
         !ShapedType::isDynamic(std::get<1>(t)))
       return false;
   }
 
   return true;
 }
 
-bool CastOp::areCastCompatible(Type a, Type b) {
+bool CastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
+  if (inputs.size() != 1 || outputs.size() != 1)
+    return false;
+  Type a = inputs.front(), b = outputs.front();
   auto aT = a.dyn_cast<TensorType>();
   auto bT = b.dyn_cast<TensorType>();
   if (!aT || !bT)
     return false;
 
   if (aT.getElementType() != bT.getElementType())
     return false;
 
   return succeeded(verifyCompatibleShape(aT, bT));
 }
 
-OpFoldResult CastOp::fold(ArrayRef<Attribute> operands) {
-  return impl::foldCastOp(*this);
-}
-
 /// Compute a TensorType that has the joined shape knowledge of the two
 /// given TensorTypes. The element types need to match.
 static TensorType joinShapes(TensorType one, TensorType two) {
   assert(one.getElementType() == two.getElementType());
 
   if (!one.hasRank())
     return two;
   if (!two.hasRank())

mlir/lib/IR/Operation.cpp

void impl::printOneResultOp(Operation *op, OpAsmPrinter &p) {
   // Now we can output only one type for all operands and the result.
   p << " : " << resultType;
 }
 
 //===----------------------------------------------------------------------===//
 // CastOp implementation
 //===----------------------------------------------------------------------===//
 
+/// Attempt to fold the given cast operation.
+LogicalResult
+impl::foldCastInterfaceOp(Operation *op, ArrayRef<Attribute> attrOperands,
+                          SmallVectorImpl<OpFoldResult> &foldResults) {
+  OperandRange operands = op->getOperands();
+  if (operands.empty())
+    return failure();
+  ResultRange results = op->getResults();
+
+  // Check for the case where the input and output types match 1-1.
+  if (operands.getTypes() == results.getTypes()) {
+    foldResults.append(operands.begin(), operands.end());
+    return success();
+  }
+
+  return failure();
+}
+
+/// Attempt to verify the given cast operation.
+LogicalResult impl::verifyCastInterfaceOp(
+    Operation *op, function_ref<bool(TypeRange, TypeRange)> areCastCompatible) {
+  auto resultTypes = op->getResultTypes();
+  if (llvm::empty(resultTypes))
+    return op->emitOpError()
+           << "expected at least one result for cast operation";
+
+  auto operandTypes = op->getOperandTypes();
+  if (!areCastCompatible(operandTypes, resultTypes)) {
+    InFlightDiagnostic diag = op->emitOpError("operand type");
+    if (llvm::empty(operandTypes))
+      diag << "s []";
+    else if (llvm::size(operandTypes) == 1)
+      diag << " " << *operandTypes.begin();
+    else
+      diag << "s " << operandTypes;
+    return diag << " and result type" << (resultTypes.size() == 1 ? " " : "s ")
+                << resultTypes << " are cast incompatible";
+  }
+
+  return success();
+}
+
 void impl::buildCastOp(OpBuilder &builder, OperationState &result, Value source,
                        Type destType) {
   result.addOperands(source);
   result.addTypes(destType);
 }
 
 ParseResult impl::parseCastOp(OpAsmParser &parser, OperationState &result) {
   OpAsmParser::OperandType srcInfo;

mlir/lib/Interfaces/CMakeLists.txt

 set(LLVM_OPTIONAL_SOURCES
   CallInterfaces.cpp
+  CastInterfaces.cpp
   ControlFlowInterfaces.cpp
   CopyOpInterface.cpp
   DerivedAttributeOpInterface.cpp
   InferTypeOpInterface.cpp
   LoopLikeInterface.cpp
   SideEffectInterfaces.cpp
   VectorInterfaces.cpp
   ViewLikeInterface.cpp
add_mlir_library(MLIR${name}
 
     LINK_LIBS PUBLIC
     MLIRIR
     )
 endfunction(add_mlir_interface_library)
 
 
 add_mlir_interface_library(CallInterfaces)
+add_mlir_interface_library(CastInterfaces)
 add_mlir_interface_library(ControlFlowInterfaces)
 add_mlir_interface_library(CopyOpInterface)
 add_mlir_interface_library(DerivedAttributeOpInterface)
 add_mlir_interface_library(InferTypeOpInterface)
 add_mlir_interface_library(LoopLikeInterface)
 add_mlir_interface_library(SideEffectInterfaces)
 add_mlir_interface_library(VectorInterfaces)
 add_mlir_interface_library(ViewLikeInterface)
 

mlir/lib/Interfaces/CastInterfaces.cpp

+//===- CastInterfaces.cpp -------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Interfaces/CastInterfaces.h"
+
+using namespace mlir;
+
+//===----------------------------------------------------------------------===//
+// Table-generated class definitions
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Interfaces/CastInterfaces.cpp.inc"