Introduce MLIR Op Properties

[mehdi_amini]

Summary

This new features enabled to dedicate custom storage inline within operations.
This storage can be used as an alternative to attributes to store data that is
specific to an operation. Attribute can also be stored inside the properties
storage if desired, but any kind of data can be present as well. This offers
a way to store and mutate data without uniquing in the Context like Attribute.
See the OpPropertiesTest.cpp for an example where a struct with a
std::vector<> is attached to an operation and mutated in-place:

struct TestProperties {

int a = -1;
float b = -1.;
std::vector<int64_t> array = {-33};

};

More complex scheme (including reference-counting) are also possible.

The only constraint to enable storing a C++ object as "properties" on an
operation is to implement three functions:

• convert from the candidate object to an Attribute
• convert from the Attribute to the candidate object
• hash the object

Optional the parsing and printing can also be customized with 2 extra
functions.

A new options is introduced to ODS to allow dialects to specify:

let usePropertiesForAttributes = 1;

When set to true, the inherent attributes for all the ops in this dialect
will be using properties instead of being stored alongside discardable
attributes.
The TestDialect showcases this feature.

[Mogball]

Nice. This is an exciting evolution of the operation model in MLIR. Attributes are really slow for many trivial things like storing an integer.

mlir/include/mlir/IR/OpBase.td

182

that's a lot of newlines

229

should this field be near the top?

mlir/include/mlir/IR/OpDefinition.h

1643

can you document these a little?

1758

docs please

mlir/include/mlir/IR/Operation.h

238

nice cider autoformatting

838

Can you use a fixed-width type for the properties object size field? I imagine int32_t would be sufficient

mlir/include/mlir/IR/OperationSupport.h

503

539–638

1257

Maybe we should add an ODSSupport.h file? DialectImplementation.h has the FieldParser stuff used by ODS too

1259

Please drop all the leading ::mlir::'s

mlir/lib/AsmParser/Parser.cpp

1992

does std::move even do anything here?

mlir/test/IR/properties.mlir

21

newline pls

mlir/tools/mlir-tblgen/FormatGen.cpp

180

keyword and name of the token don't line up?

mlir/tools/mlir-tblgen/OpDefinitionsGen.cpp

1162

these should get all the ::llvm::/::mlir:: prefixes

[mehdi_amini]

minor cleanups

[mehdi_amini]

Address Jeff's comments (easy ones)

mlir/include/mlir/IR/Operation.h

838

Reminds me that I'm concerned about the alignment for the Properties pointer here, need to think more about this...
Leaving this comment open for now.

[mehdi_amini]

Move propertiesStorageSize to Operation member, stealing 8 bits from numRegions. This limits the maximum size of Properties to 2048B (anything that large does not seem reasonnable anyway!)

[mehdi_amini]

Rebase

[mehdi_amini]

Add a new ODS dialect flag "usePropertiesForAttributes" which when set to true makes all inherent attributes stored as properties for this dialect. The test dialect is updated to showcase this.

The idea is to make this the default in the future and deprecate storing inherent attributes alongside discardable ones.

[mehdi_amini]

rebase and some small fixes

[mehdi_amini]

Rebase and minor updates to make initialization more robust

[jpienaar]

mlir/lib/Dialect/Shape/IR/Shape.cpp

1143

Mmmm, if we always used OpAdaptor, then these could all have been hidden below that here right?

Oh (and I haven't look at rest) is OpAdaptor also expanded?

[rriddle]

Started going through, made it about half way, will need another pass or two.

mlir/include/mlir/IR/OpBase.td

197

Tablegen constructs have generally used $_blah for these. Why not do that for properties?

199–200

Can you add a Format reference like the other fields?

1159

N -> n

1164

Extra newline here

2475–2479

When will this be used? What benefit is there to making this open?

mlir/include/mlir/IR/OpDefinition.h

1730

// -> ///

Same below

1733

1741

1818–1820

Can you document this? Also, why not wrap this in a constexpr function?

1919–1931

Not putting this in the else branch?

1979

Did you mean customPrintProperties here?

mlir/include/mlir/IR/Operation.h

97

Can you define a strongly typed wrapper class for this void *? The use of void * makes me a tad uncomfortable here.

435

Why std::optional? That doesn't match any other API usage we have for this type of method, I'd rather not diverge.

437–439

Can you document these methods?

495–496

503–504

Same throughout

838

Did this get resolved?

841–846

Let's please use strongly typed API here instead of void *.

996

// -> ///

mlir/include/mlir/IR/OperationSupport.h

124

552

Why cast just to resolve back to op? Also it'd be nice to add context as to why it goes to getDiscardableAttr in this case, which realistically it could just be getAttr until those methods get remove.

[rriddle]

mlir/include/mlir/IR/OperationSupport.h

583

585

955–957

Please document this method.

1256–1257

Let's resolve this. I'd be fine with a support file as Jeff suggested.

mlir/include/mlir/TableGen/Operator.h

355

Please document this API.

[mehdi_amini]

mlir/include/mlir/IR/OpBase.td

2475–2479

What I had in mind when I added this was the ability to use the same class for multiple ops in the same dialect. It does not seem to carry its weight right now though, removing.

mlir/lib/Dialect/Shape/IR/Shape.cpp

1143

Yes the adaptor constructor has been updated, see here for example: https://reviews.llvm.org/D141742#change-ewmZYaiO8SOS

[mehdi_amini]

address a very small part of the comments :)

[jpienaar]

mlir/lib/Dialect/Shape/IR/Shape.cpp

1143

What is stopping us from switching to those? :) (e.g., if we have folks going through all the interfaces, perhaps we can fold it in). I know that couples 2 changes, I could try and "quickly" do that if you think that makes sense. Then your change here becomes smaller. WDYT? (I could also delay till later/after next 2 conferences :))

[mehdi_amini]

Address comments

mlir/include/mlir/IR/Operation.h

97

Ah! Indeed I hated the void* and wrongly punted on this...

435

We need to differentiate whether the attribute is set from whether there is no inherent attribute with this name.
That is: returning std::nullopt is signaling a different case than returning an Attribute{} ; and both cases can happen here.

838

Yes: there is now a field (propertiesStorageSize) on the operation that stores the size, and this is solving my alignment issue as well.

mlir/include/mlir/IR/OperationSupport.h

552

Why cast just to resolve back to op?

It's a trick: it delays resolving the call to later when the Operation class is available. Without it I get:error: member access into incomplete type 'mlir::Operation'

Also it'd be nice to add context as to why it goes to getDiscardableAttr in this case, which realistically it could just be getAttr until those methods get remove.

Right it could be getAttr(), but seemed more logical to me to resolve the call to where it'll end up: if someone traces the code I'm saving them some mental load in carrying invariants through the call stack.

mlir/include/mlir/TableGen/Operator.h

355

Removed API

mlir/lib/Dialect/Shape/IR/Shape.cpp

1143

We should look into a default implementation building the adaptor and forwarding to a method to be implemented by the class author.
After our offline discussion, I'll punt this to future work.

[mehdi_amini]

Add Langref doc

[mehdi_amini]

Rebase

[mehdi_amini]

Rebase

[mehdi_amini]

Fix builders

[mehdi_amini]

Fix UB when binding a nullptr as a reference

[mehdi_amini]

Rebase after improving diagnostic on invalid IR

[mehdi_amini]

rebase and fix build

[rriddle]

Alright, I've now looked at everything but OpDefinitionsGen. Will take a look at that one tomorrow.

mlir/docs/LangRef.md

744

mlir/include/mlir/IR/ODSSupport.h

19–21

Should we wrap this in a nested namespace? Is this intended to be called from user code? Feels cleaner to put this in a namespace to add an extra layer of visibility from user code.

mlir/include/mlir/IR/Operation.h

523–524

Can you inline the optional into the if for each of these?

mlir/lib/AsmParser/Parser.cpp

1293–1294

mlir/lib/IR/MLIRContext.cpp

811–812

814

I thought that the optional return was supposed to differentiate nullptr from the name being an inherent attribute, but I don't see any difference here (nullopt is always used for failure).

mlir/test/IR/dynamic.mlir

4 ↗

(On Diff #515982)

Why are all of these being deleted?

mlir/test/Rewrite/pdl-bytecode.mlir

3 ↗

(On Diff #515982)

Why are all of these tests being deleted? This is extremely concerning to me.

mlir/test/lib/Dialect/Test/TestDialect.cpp

47–48

Why is this necessary?

1833–1834

Same here, why is this necessary?

mlir/test/lib/Dialect/Test/TestDialect.h

63

// -> /// for each of these.

mlir/test/lib/Dialect/Test/TestOps.td

3301–3309

Please use normal strings if you don't need multi-line, it makes the examples much more compact and cleaner.

3345

Can you wrap this at 80?

[mehdi_amini]

Fix one more verifier during custom parsing that involves inherent attributes

[mehdi_amini]

mlir/include/mlir/IR/ODSSupport.h

19–21

I just tried, but this is supposed to work with arbitrary types using ADL.
There are matching the "builtin" Attributes (IntegerAttr and DenseI64ArrayAttr only right now), but one could have their own types.

If I nest these here, I would need ODS to emit a using namespace ::mlir::ods_support; in the generated header and the generated implementation for the op definition, which would be more annoying I think?

mlir/lib/IR/MLIRContext.cpp

814

This is the implementation for unregistered operation: so for these we can't know if there is a missing inherent attr from the DictionaryAttr, so can't return nullptr. We have to assume that if a name isn't in the properties dictionary then there is no inherent attr for this name.

Any other ideas?

[mehdi_amini]

Address comments (and rebase)

[zero9178]

mlir/include/mlir/IR/OpBase.td

212

All TableGen code currently uses $_ctxt, this would otherwise be the first use of $_ctx

[mehdi_amini]

Use $_ctxt

[mehdi_amini]

Rebase

[martin-luecke]

mlir/lib/AsmParser/Parser.cpp

1416

1421

[mehdi_amini]

Fix typos reported by Martin

[Mogball]

mlir/include/mlir/IR/OpBase.td

198

nit: other tablegen stuff typically uses $_ctxt

mlir/tools/mlir-tblgen/OpDefinitionsGen.cpp

1053

drop trivial braces

1087

why do some of these have the leading :: and some don't?

[mehdi_amini]

mlir/include/mlir/IR/OpBase.td

198

The code was updated, I forgot to fix the doc here.

mlir/tools/mlir-tblgen/OpDefinitionsGen.cpp

1087

Which ones don't?

All of the mlir:: or llvm:: utilities / symbols should be fully qualified.
Note: Properties is defined by the current class, not by MLIR.

[mehdi_amini]

Address Jeff's comments

mlir/docs/LangRef.md

 ## Operations
 
 Syntax:
 
 ```
 operation            ::= op-result-list? (generic-operation | custom-operation)
                          trailing-location?
 generic-operation    ::= string-literal `(` value-use-list? `)`  successor-list?
-                         region-list? dictionary-attribute? `:` function-type
+                         dictionary-properties? region-list? dictionary-attribute?
+                         `:` function-type
 custom-operation     ::= bare-id custom-operation-format
 op-result-list       ::= op-result (`,` op-result)* `=`
 op-result            ::= value-id (`:` integer-literal)
 successor-list       ::= `[` successor (`,` successor)* `]`
 successor            ::= caret-id (`:` block-arg-list)?
+dictionary-propertes ::= `<` dictionary-attribute `>`
 region-list          ::= `(` region (`,` region)* `)`
 dictionary-attribute ::= `{` (attribute-entry (`,` attribute-entry)*)? `}`
 trailing-location    ::= (`loc` `(` location `)`)?
 ```
 
 MLIR introduces a uniform concept called *operations* to enable describing many
 different levels of abstractions and computations. Operations in MLIR are fully
 extensible (there is no fixed list of operations) and have application-specific
 semantics. For example, MLIR supports
 [target-independent operations](Dialects/MemRef.md),
 [affine operations](Dialects/Affine.md), and
 [target-specific machine operations](#target-specific-operations).
 
 The internal representation of an operation is simple: an operation is
 identified by a unique string (e.g. `dim`, `tf.Conv2d`, `x86.repmovsb`,
 `ppc.eieio`, etc), can return zero or more results, take zero or more operands,
-has a dictionary of [attributes](#attributes), has zero or more successors, and
-zero or more enclosed [regions](#regions). The generic printing form includes
-all these elements literally, with a function type to indicate the types of the
+has storage for [properties](#properties), has a dictionary of
+[attributes](#attributes), has zero or more successors, and zero or more
+enclosed [regions](#regions). The generic printing form includes all these
+elements literally, with a function type to indicate the types of the
 results and operands.
 
 Example:
 
 ```mlir
 // An operation that produces two results.
 // The results of %result can be accessed via the <name> `#` <opNo> syntax.
 %result:2 = "foo_div"() : () -> (f32, i32)
 
 // Pretty form that defines a unique name for each result.
 %foo, %bar = "foo_div"() : () -> (f32, i32)
 
 // Invoke a TensorFlow function called tf.scramble with two inputs
-// and an attribute "fruit".
-%2 = "tf.scramble"(%result#0, %bar) {fruit = "banana"} : (f32, i32) -> f32
+// and an attribute "fruit" stored in properties.
+%2 = "tf.scramble"(%result#0, %bar) <{fruit = "banana"}> : (f32, i32) -> f32
+
+// Invoke an operation with some discardable attributes
+%foo, %bar = "foo_div"() {some_attr = "value", other_attr = 42 : i64} : () -> (f32, i32)
 ```
 
 In addition to the basic syntax above, dialects may register known operations.
 This allows those dialects to support *custom assembly form* for parsing and
 printing operations. In the operation sets listed below, we show both forms.
 
 ### Builtin Operations
 

 See [here](DefiningDialects/AttributesAndTypes.md) to learn how to define dialect types.
 
 ### Builtin Types
 
 The [builtin dialect](Dialects/Builtin.md) defines a set of types that are
 directly usable by any other dialect in MLIR. These types cover a range from
 primitive integer and floating-point types, function types, and more.
 
+## Properties
+
+Properties are extra data members stored directly on an Operation class. They
+provide a way to store [inherent attributes](#attributes) and other arbitrary
+data. The semantics of the data is specific to a given operation, and may be
+exposed through [Interfaces](Interfaces.md) accessors and other methods.
+Properties can always be serialized to Attribute in order to be printed
+generically.
+
 ## Attributes
 
 Syntax:
 
 ```
 attribute-entry ::= (bare-id | string-literal) `=` attribute-value
 attribute-value ::= attribute-alias | dialect-attribute | builtin-attribute
 ```
 
 Attributes are the mechanism for specifying constant data on operations in
 places where a variable is never allowed - e.g. the comparison predicate of a
 [`cmpi` operation](Dialects/ArithOps.md#arithcmpi-mlirarithcmpiop). Each operation has an
 attribute dictionary, which associates a set of attribute names to attribute
 values. MLIR's builtin dialect provides a rich set of
 [builtin attribute values](#builtin-attribute-values) out of the box (such as
 arrays, dictionaries, strings, etc.). Additionally, dialects can define their
 own [dialect attribute values](#dialect-attribute-values).
 
-The top-level attribute dictionary attached to an operation has special
-semantics. The attribute entries are considered to be of two different kinds
-based on whether their dictionary key has a dialect prefix:
+For dialects which haven't adopted properties yet, the top-level attribute
+dictionary attached to an operation has special semantics. The attribute
+entries are considered to be of two different kinds based on whether their
+dictionary key has a dialect prefix:
 
 -   *inherent attributes* are inherent to the definition of an operation's
     semantics. The operation itself is expected to verify the consistency of
     these attributes. An example is the `predicate` attribute of the
     `arith.cmpi` op. These attributes must have names that do not start with a
     dialect prefix.
 
 -   *discardable attributes* have semantics defined externally to the operation
     itself, but must be compatible with the operations's semantics. These
     attributes must have names that start with a dialect prefix. The dialect
     indicated by the dialect prefix is expected to verify these attributes. An
     example is the `gpu.container_module` attribute.
 
 Note that attribute values are allowed to themselves be dictionary attributes,
 but only the top-level dictionary attribute attached to the operation is subject
 to the classification above.
 
+When properties are adopted, only discardable attributes are stored in the
+top-level dictionary, while inherent attributes are stored in the properties
+storage.
+
 ### Attribute Value Aliases
 
 ```
 attribute-alias-def ::= '#' alias-name '=' attribute-value
 attribute-alias ::= '#' alias-name
 ```
 
 MLIR supports defining named aliases for attribute values. An attribute alias is

mlir/include/mlir/Conversion/LLVMCommon/Pattern.h

explicit ConvertOpToLLVMPattern(LLVMTypeConverter &typeConverter,
                                   PatternBenefit benefit = 1)
       : ConvertToLLVMPattern(SourceOp::getOperationName(),
                              &typeConverter.getContext(), typeConverter,
                              benefit) {}
 
   /// Wrappers around the RewritePattern methods that pass the derived op type.
   void rewrite(Operation *op, ArrayRef<Value> operands,
                ConversionPatternRewriter &rewriter) const final {
+    if constexpr (SourceOp::hasProperties())
+      rewrite(cast<SourceOp>(op),
+              OpAdaptor(operands, op->getAttrDictionary(),
+                        cast<SourceOp>(op).getProperties()),
+              rewriter);
     rewrite(cast<SourceOp>(op), OpAdaptor(operands, op->getAttrDictionary()),
             rewriter);
   }
   LogicalResult match(Operation *op) const final {
     return match(cast<SourceOp>(op));
   }
   LogicalResult
   matchAndRewrite(Operation *op, ArrayRef<Value> operands,
                   ConversionPatternRewriter &rewriter) const final {
+    if constexpr (SourceOp::hasProperties())
+      return matchAndRewrite(cast<SourceOp>(op),
+                             OpAdaptor(operands, op->getAttrDictionary(),
+                                       cast<SourceOp>(op).getProperties()),
+                             rewriter);
     return matchAndRewrite(cast<SourceOp>(op),
                            OpAdaptor(operands, op->getAttrDictionary()),
                            rewriter);
   }
 
   /// Rewrite and Match methods that operate on the SourceOp type. These must be
   /// overridden by the derived pattern class.
   virtual LogicalResult match(SourceOp op) const {

mlir/include/mlir/IR/DialectBase.td

class Dialect {
   // individual type directly.
   bit useDefaultTypePrinterParser = 0;
 
   // If this dialect overrides the hook for canonicalization patterns.
   bit hasCanonicalizer = 0;
 
   // If this dialect can be extended at runtime with new operations or types.
   bit isExtensible = 0;
+
+  // Whether inherent Attributes defined in ODS will be stored as Properties.
+  bit usePropertiesForAttributes = 0;
 }
 
 #endif // DIALECTBASE_TD

mlir/include/mlir/IR/ExtensibleDialect.h

 #define MLIR_IR_EXTENSIBLEDIALECT_H
 
 #include "mlir/IR/Dialect.h"
 #include "mlir/IR/DialectInterface.h"
 #include "mlir/IR/MLIRContext.h"
 #include "mlir/IR/OpDefinition.h"
 #include "mlir/Support/TypeID.h"
 #include "llvm/ADT/StringMap.h"
+#include "llvm/Support/ErrorHandling.h"
+#include <optional>
 
 namespace mlir {
 class AsmParser;
 class AsmPrinter;
 class DynamicAttr;
 class DynamicType;
 class ExtensibleDialect;
 class MLIRContext;
void printAssembly(Operation *op, OpAsmPrinter &printer,
                      StringRef name) final {
     printFn(op, printer, name);
   }
   LogicalResult verifyInvariants(Operation *op) final { return verifyFn(op); }
   LogicalResult verifyRegionInvariants(Operation *op) final {
     return verifyRegionFn(op);
   }
 
+  /// Implementation for properties (unsupported right now here).
+  std::optional<Attribute> getInherentAttr(Operation *op,
+                                           StringRef name) final {
+    llvm::report_fatal_error("Unsupported getInherentAttr on Dynamic dialects");
+  }
+  void setInherentAttr(Operation *op, StringAttr name, Attribute value) final {
+    llvm::report_fatal_error("Unsupported setInherentAttr on Dynamic dialects");
+  }
+  void populateInherentAttrs(Operation *op, NamedAttrList &attrs) final {}
+  LogicalResult
+  verifyInherentAttrs(OperationName opName, NamedAttrList &attributes,
+                      function_ref<InFlightDiagnostic()> getDiag) final {
+    return success();
+  }
+  int getOpPropertyByteSize() final { return 0; }
+  void initProperties(OperationName opName, OpaqueProperties storage,
+                      OpaqueProperties init) final {}
+  void deleteProperties(OpaqueProperties prop) final {}
+  void populateDefaultProperties(OperationName opName,
+                                 OpaqueProperties properties) final {}
+
+  LogicalResult setPropertiesFromAttr(Operation *op, Attribute attr,
+                                      InFlightDiagnostic *diag) final {
+    return failure();
+  }
+  Attribute getPropertiesAsAttr(Operation *op) final { return {}; }
+  void copyProperties(OpaqueProperties lhs, OpaqueProperties rhs) final {}
+  llvm::hash_code hashProperties(OpaqueProperties prop) final { return {}; }
+
 private:
   DynamicOpDefinition(
       StringRef name, ExtensibleDialect *dialect,
       OperationName::VerifyInvariantsFn &&verifyFn,
       OperationName::VerifyRegionInvariantsFn &&verifyRegionFn,
       OperationName::ParseAssemblyFn &&parseFn,
       OperationName::PrintAssemblyFn &&printFn,
       OperationName::FoldHookFn &&foldHookFn,

mlir/include/mlir/IR/ODSSupport.h

+//===- ODSSupport.h ---------------------------------------------*- 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 defines a number of support method for ODS generated code.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_IR_ODSSUPPORT_H
+#define MLIR_IR_ODSSUPPORT_H
+
+#include "mlir/IR/Attributes.h"
+
+namespace mlir {
+
+//===----------------------------------------------------------------------===//
+// Support for properties
+//===----------------------------------------------------------------------===//
+
+/// Convert an IntegerAttr attribute to an int64_t, or return an error if the
+/// attribute isn't an IntegerAttr. If the optional diagnostic is provided an
+/// error message is also emitted.
+LogicalResult convertFromAttribute(int64_t &storage, Attribute attr,
+                                   InFlightDiagnostic *diag);
+
+/// Convert the provided int64_t to an IntegerAttr attribute.
+Attribute convertToAttribute(MLIRContext *ctx, int64_t storage);
+
+/// Convert a DenseI64ArrayAttr to the provided storage. It is expected that the
+/// storage has the same size as the array. An error is returned if the
+/// attribute isn't a DenseI64ArrayAttr or it does not have the same size. If
+/// the optional diagnostic is provided an error message is also emitted.
+LogicalResult convertFromAttribute(MutableArrayRef<int64_t> storage,
+                                   Attribute attr, InFlightDiagnostic *diag);
+
+/// Convert the provided ArrayRef<int64_t> to a DenseI64ArrayAttr attribute.
+Attribute convertToAttribute(MLIRContext *ctx, ArrayRef<int64_t> storage);
+
+} // namespace mlir
+
+#endif // MLIR_IR_ODSSUPPORT_H
 No newline at end of file

mlir/include/mlir/IR/OpBase.td

 // Subclass for constraints on a type.
 class TypeConstraint<Pred predicate, string summary = "",
                      string cppClassNameParam = "::mlir::Type"> :
     Constraint<predicate, summary> {
   // The name of the C++ Type class if known, or Type if not.
   string cppClassName = cppClassNameParam;
 }
 
+// Base class for defining properties.
+class Property<string storageTypeParam = "", string desc = ""> {
+  // User-readable one line summary used in error reporting messages. If empty,
+  // a generic message will be used.
+  string summary = desc;
+  // The full description of this property.
+  string description = "";
+  code storageType = storageTypeParam;
+  code interfaceType = storageTypeParam;
+
+  // The expression to convert from the storage type to the Interface
+  // type. For example, an enum can be stored as an int but returned as an
+  // enum class.
+  //
+  // Format:
+  // - `$_storage` will contain the property in the storage type.
+  // - `$_ctxt` will contain an `MLIRContext *`.
+  code convertFromStorage = "$_storage";
+
+  // The call expression to build a property storage from the interface type.
+  //
+  // Format:
+  // - `$_storage` will contain the property in the storage type.
+  // - `$_value` will contain the property in the user interface type.
+  code assignToStorage = "$_storage = $_value";
+
+  // The call expression to convert from the storage type to an attribute.
+  //
+  // Format:
+  // - `$_storage` is the storage type value.
+  // - `$_ctxt` is a `MLIRContext *`.
+  //
+  // The expression must result in an Attribute.
+  code convertToAttribute = [{
+    convertToAttribute($_ctxt, $_storage)
+  }];
+
+  // The call expression to convert from an Attribute to the storage type.
+  //
+  // Format:
+  // - `$_storage` is the storage type value.
+  // - `$_attr` is the attribute.
+  // - `$_diag` is an optional Diagnostic pointer to emit error.
+  //
+  // The expression must return a LogicalResult
+  code convertFromAttribute = [{
+    return convertFromAttribute($_storage, $_attr, $_diag);
+  }];
+
+  // The call expression to hash the property.
+  //
+  // Format:
+  // - `$_storage` is the variable to hash.
+  //
+  // The expression should define a llvm::hash_code.
+  code hashProperty = [{
+    llvm::hash_value($_storage);
+  }];
+
+  // Default value for the property.
+  string defaultValue = ?;
+}
+
 // Subclass for constraints on an attribute.
 class AttrConstraint<Pred predicate, string summary = ""> :
     Constraint<predicate, summary>;
 
 // Subclass for constraints on a region.
 class RegionConstraint<Pred predicate, string summary = ""> :
     Constraint<predicate, summary>;
 

 // Default-valued string-based attribute. Wraps the default value in escaped
 // quotes.
 class DefaultValuedStrAttr<Attr attr, string val>
     : DefaultValuedAttr<attr, "\"" # val # "\"">;
 class DefaultValuedOptionalStrAttr<Attr attr, string val>
     : DefaultValuedOptionalAttr<attr, "\"" # val # "\"">;
 
 //===----------------------------------------------------------------------===//
+// Primitive property kinds
+
+class ArrayProperty<string storageTypeParam = "", int n, string desc = ""> :
+  Property<storageTypeParam # "[" # n # "]", desc> {
+  let interfaceType = "::llvm::ArrayRef<" # storageTypeParam # ">";
+  let convertFromStorage = "$_storage";
+  let assignToStorage = "::llvm::copy($_value, $_storage)";
+}
+
+//===----------------------------------------------------------------------===//
 // Primitive attribute kinds
 
 // A generic attribute that must be constructed around a specific buildable type
 // `attrValType`. Backed by MLIR attribute kind `attrKind`.
 class TypedAttrBase<Type attrValType, string attrKind, Pred condition,
                     string descr> :
     Attr<condition, descr> {
   let constBuilderCall = "$_builder.get" # attrKind # "(" #
class Op<Dialect dialect, string mnemonic, list<Trait> props = []> {
 
   // Whether this op has a folder.
   bit hasFolder = 0;
 
   // Op traits.
   // Note: The list of traits will be uniqued by ODS.
   list<Trait> traits = props;
 
   // Additional code that will be added to the public part of the generated
   // C++ code of the op declaration.
   code extraClassDeclaration = ?;
 
   // Additional code that will be added to the generated source file. The
   // generated code is placed inside the op's C++ namespace. `$cppClass` is
   // replaced by the op's C++ class name.
   code extraClassDefinition = ?;
 }
 
 // The arguments of an op.
 class Arguments<dag args> {
   dag arguments = args;

mlir/include/mlir/IR/OpDefinition.h

void ensureRegionTerminator(
     Region &region, OpBuilder &builder, Location loc,
     function_ref<Operation *(OpBuilder &, Location)> buildTerminatorOp);
 void ensureRegionTerminator(
     Region &region, Builder &builder, Location loc,
     function_ref<Operation *(OpBuilder &, Location)> buildTerminatorOp);
 
 } // namespace impl
 
+/// Structure used by default as a "marker" when no "Properties" are set on an
+/// Operation.
+struct EmptyProperties {};
+
+/// Traits to detect whether an Operation defined a `Properties` type, otherwise
+/// it'll default to `EmptyProperties`.
+template <class Op, class = void>
+struct PropertiesSelector {
+  using type = EmptyProperties;
+};
+template <class Op>
+struct PropertiesSelector<Op, std::void_t<typename Op::Properties>> {
+  using type = typename Op::Properties;
+};
+
 /// This is the concrete base class that holds the operation pointer and has
 /// non-generic methods that only depend on State (to avoid having them
 /// instantiated on template types that don't affect them.
 ///
 /// This also has the fallback implementations of customization hooks for when
 /// they aren't customized.
 class OpState {
 public:
protected:
   /// result with the fields to use.
   static ParseResult parse(OpAsmParser &parser, OperationState &result);
 
   /// Print the operation. Unless overridden, this method will first try to get
   /// an operation printer from the dialect. Otherwise, it prints the operation
   /// in generic form.
   static void print(Operation *op, OpAsmPrinter &p, StringRef defaultDialect);
 
+  /// Parse properties as a Attribute.
+  static ParseResult genericParseProperties(OpAsmParser &parser,
+                                            Attribute &result);
+
+  /// Print the properties as a Attribute.
+  static void genericPrintProperties(OpAsmPrinter &p, Attribute properties);
+
   /// Print an operation name, eliding the dialect prefix if necessary.
   static void printOpName(Operation *op, OpAsmPrinter &p,
                           StringRef defaultDialect);
 
   /// Mutability management is handled by the OpWrapper/OpConstWrapper classes,
   /// so we can cast it away here.
   explicit OpState(Operation *state) : state(state) {}
 
+  /// For all op which don't have properties, we keep a single instance of
+  /// `EmptyProperties` to be used where a reference to a properties is needed:
+  /// this allow to bind a pointer to the reference without triggering UB.
+  static EmptyProperties &getEmptyProperties() {
+    static EmptyProperties emptyProperties;
+    return emptyProperties;
+  }
+
 private:
   Operation *state;
 
   /// Allow access to internal hook implementation methods.
   friend RegisteredOperationName;
 };
 
 // Allow comparing operators.

 
 namespace op_definition_impl {
 //===----------------------------------------------------------------------===//
 // Trait Existence
 
 /// Returns true if this given Trait ID matches the IDs of any of the provided
 /// trait types `Traits`.
 template <template <typename T> class... Traits>
-static bool hasTrait(TypeID traitID) {
+inline bool hasTrait(TypeID traitID) {
   TypeID traitIDs[] = {TypeID::get<Traits>()...};
   for (unsigned i = 0, e = sizeof...(Traits); i != e; ++i)
     if (traitIDs[i] == traitID)
       return true;
   return false;
 }
+template <>
+inline bool hasTrait<>(TypeID traitID) {
+  return false;
+}
 
 //===----------------------------------------------------------------------===//
 // Trait Folding
 
 /// Trait to check if T provides a 'foldTrait' method for single result
 /// operations.
 template <typename T, typename... Args>
 using has_single_result_fold_trait = decltype(T::foldTrait(

 
 /// Given a set of traits, return the result of verifying the regions of the
 /// given operation.
 template <typename... Ts>
 LogicalResult verifyRegionTraits(Operation *op) {
   return success((succeeded(verifyRegionTrait<Ts>(op)) && ...));
 }
 } // namespace op_definition_impl
 
 //===----------------------------------------------------------------------===//
 // Operation Definition classes
 //===----------------------------------------------------------------------===//
 
 /// This provides public APIs that all operations should have.  The template
 /// argument 'ConcreteType' should be the concrete type by CRTP and the others
 /// are base classes by the policy pattern.
 template <typename ConcreteType, template <typename T> class... Traits>
std::optional<RegisteredOperationName> info =
                                         &context);
     if (!info)
       llvm::report_fatal_error(
           "Attempting to attach an interface to an unregistered operation " +
           ConcreteType::getOperationName() + ".");
     (checkInterfaceTarget<Models>(), ...);
     info->attachInterface<Models...>();
   }
+  /// Convert the provided attribute to a property and assigned it to the
+  /// provided properties. This default implementation forwards to a free
+  /// function `setPropertiesFromAttribute` that can be looked up with ADL in
+  /// the namespace where the properties are defined. It can also be overridden
+  /// in the derived ConcreteOp.
+  template <typename PropertiesTy>
+  static LogicalResult setPropertiesFromAttr(PropertiesTy &prop, Attribute attr,
+                                             InFlightDiagnostic *diag) {
+    return setPropertiesFromAttribute(prop, attr, diag);
+  }
+  /// Convert the provided properties to an attribute. This default
+  /// implementation forwards to a free function `getPropertiesAsAttribute` that
+  /// can be looked up with ADL in the namespace where the properties are
+  /// defined. It can also be overridden in the derived ConcreteOp.
+  template <typename PropertiesTy>
+  static Attribute getPropertiesAsAttr(MLIRContext *ctx,
+                                       const PropertiesTy &prop) {
+    return getPropertiesAsAttribute(ctx, prop);
+  }
+  /// Hash the provided properties. This default implementation forwards to a
+  /// free function `computeHash` that can be looked up with ADL in the
+  /// namespace where the properties are defined. It can also be overridden in
+  /// the derived ConcreteOp.
+  template <typename PropertiesTy>
+  static llvm::hash_code computePropertiesHash(const PropertiesTy &prop) {
+    return computeHash(prop);
+  }
 
 private:
   /// Trait to check if T provides a 'fold' method for a single result op.
   template <typename T, typename... Args>
   using has_single_result_fold_t =
       decltype(std::declval<T>().fold(std::declval<ArrayRef<Attribute>>()));
   template <typename T>
   constexpr static bool has_single_result_fold_v =
       llvm::is_detected<has_single_result_fold_t, T>::value;
   /// Trait to check if T provides a general 'fold' method.
private:
 
   /// Trait to check if T provides a 'print' method.
   template <typename T, typename... Args>
   using has_print =
       decltype(std::declval<T>().print(std::declval<OpAsmPrinter &>()));
   template <typename T>
   using detect_has_print = llvm::is_detected<has_print, T>;
 
+  /// Trait to check if printProperties(OpAsmPrinter, T) exist
+  template <typename T, typename... Args>
+  using has_print_properties = decltype(printProperties(
+      std::declval<OpAsmPrinter &>(), std::declval<T>()));
+  template <typename T>
+  using detect_has_print_properties =
+      llvm::is_detected<has_print_properties, T>;
+
+  /// Trait to check if parseProperties(OpAsmParser, T) exist
+  template <typename T, typename... Args>
+  using has_parse_properties = decltype(parseProperties(
+      std::declval<OpAsmParser &>(), std::declval<T &>()));
+  template <typename T>
+  using detect_has_parse_properties =
+      llvm::is_detected<has_parse_properties, T>;
+
   /// Trait to check if T provides a 'ConcreteEntity' type alias.
   template <typename T>
   using has_concrete_entity_t = typename T::ConcreteEntity;
 
+public:
+  /// Returns true if this operation defines a `Properties` inner type.
+  static constexpr bool hasProperties() {
+    return !std::is_same_v<
+        typename ConcreteType::template InferredProperties<ConcreteType>,
+        EmptyProperties>;
+  }
+
+private:
   /// A struct-wrapped type alias to T::ConcreteEntity if provided and to
   /// ConcreteType otherwise. This is akin to std::conditional but doesn't fail
   /// on the missing typedef. Useful for checking if the interface is targeting
   /// the right class.
   template <typename T,
             bool = llvm::is_detected<has_concrete_entity_t, T>::value>
   struct InterfaceTargetOrOpT {
     using type = typename T::ConcreteEntity;
private:
   }
   /// Return the result of folding a single result operation that defines a
   /// `fold` method.
   template <typename ConcreteOpT>
   static LogicalResult
   foldSingleResultHook(Operation *op, ArrayRef<Attribute> operands,
                        SmallVectorImpl<OpFoldResult> &results) {
     OpFoldResult result;
-    if constexpr (has_fold_adaptor_single_result_v<ConcreteOpT>)
+    if constexpr (has_fold_adaptor_single_result_v<ConcreteOpT>) {
+      if constexpr (hasProperties()) {
         result = cast<ConcreteOpT>(op).fold(typename ConcreteOpT::FoldAdaptor(
-          operands, op->getAttrDictionary(), op->getRegions()));
-    else
+            operands, op->getAttrDictionary(),
+            cast<ConcreteOpT>(op).getProperties(), op->getRegions()));
+      } else {
+        result = cast<ConcreteOpT>(op).fold(typename ConcreteOpT::FoldAdaptor(
+            operands, op->getAttrDictionary(), {}, op->getRegions()));
+      }
+    } else {
       result = cast<ConcreteOpT>(op).fold(operands);
+    }
 
     // If the fold failed or was in-place, try to fold the traits of the
     // operation.
     if (!result || result.template dyn_cast<Value>() == op->getResult(0)) {
       if (succeeded(op_definition_impl::foldTraits<Traits<ConcreteType>...>(
               op, operands, results)))
         return success();
       return success(static_cast<bool>(result));
     }
     results.push_back(result);
     return success();
   }
   /// Return the result of folding an operation that defines a `fold` method.
   template <typename ConcreteOpT>
   static LogicalResult foldHook(Operation *op, ArrayRef<Attribute> operands,
                                 SmallVectorImpl<OpFoldResult> &results) {
     auto result = LogicalResult::failure();
     if constexpr (has_fold_adaptor_v<ConcreteOpT>) {
+      if constexpr (hasProperties()) {
+        result = cast<ConcreteOpT>(op).fold(
+            typename ConcreteOpT::FoldAdaptor(
+                operands, op->getAttrDictionary(),
+                cast<ConcreteOpT>(op).getProperties(), op->getRegions()),
+            results);
+      } else {
         result = cast<ConcreteOpT>(op).fold(
             typename ConcreteOpT::FoldAdaptor(operands, op->getAttrDictionary(),
-                                            op->getRegions()),
+                                              {}, op->getRegions()),
             results);
+      }
     } else {
       result = cast<ConcreteOpT>(op).fold(operands, results);
     }
 
     // If the fold failed or was in-place, try to fold the traits of the
     // operation.
     if (failed(result) || results.empty()) {
       if (succeeded(op_definition_impl::foldTraits<Traits<ConcreteType>...>(
if constexpr (detect_has_print<ConcreteType>::value)
         OpState::printOpName(op, p, defaultDialect);
         return cast<ConcreteType>(op).print(p);
       };
     return [](Operation *op, OpAsmPrinter &printer, StringRef defaultDialect) {
       return OpState::print(op, printer, defaultDialect);
     };
   }
 
+public:
+  template <typename T>
+  using InferredProperties = typename PropertiesSelector<T>::type;
+  template <typename T = ConcreteType>
+  InferredProperties<T> &getProperties() {
+    if constexpr (!hasProperties())
+      return getEmptyProperties();
+    return *getOperation()
+                ->getPropertiesStorage()
+                .template as<InferredProperties<T> *>();
+  }
+
+  /// This hook populates any unset default attrs when mapped to properties.
+  template <typename T = ConcreteType>
+  static void populateDefaultProperties(OperationName opName,
+                                        InferredProperties<T> &properties) {}
+
+  /// Print the operation properties. Unless overridden, this method will try to
+  /// dispatch to a `printProperties` free-function if it exists, and otherwise
+  /// by converting the properties to an Attribute.
+  template <typename T>
+  static void printProperties(MLIRContext *ctx, OpAsmPrinter &p,
+                              const T &properties) {
+    if constexpr (detect_has_print_properties<T>::value)
+      return printProperties(p, properties);
+    genericPrintProperties(p,
+                           ConcreteType::getPropertiesAsAttr(ctx, properties));
+  }
+
+  /// Parser the properties. Unless overridden, this method will print by
+  /// converting the properties to an Attribute.
+  template <typename T = ConcreteType>
+  static ParseResult parseProperties(OpAsmParser &parser,
+                                     OperationState &result) {
+    if constexpr (detect_has_parse_properties<InferredProperties<T>>::value) {
+      return parseProperties(
+          parser, result.getOrAddProperties<InferredProperties<T>>());
+    }
+    return genericParseProperties(parser, result.propertiesAttr);
+  }
+
+private:
   /// Implementation of `PopulateDefaultAttrsFn` OperationName hook.
   static OperationName::PopulateDefaultAttrsFn getPopulateDefaultAttrsFn() {
     return ConcreteType::populateDefaultAttrs;
   }
   /// Implementation of `VerifyInvariantsFn` OperationName hook.
   static LogicalResult verifyInvariants(Operation *op) {
     static_assert(hasNoDataMembers(),
                   "Op class shouldn't define new data members");

mlir/include/mlir/IR/OpImplementation.h

parseCustomAttributeWithFallback(AttrType &result, Type type,
     return parseAttribute(result, type, attrName, attrs);
   }
 
   /// Parse a custom attribute of a given type unless the next token is `#`, in
   /// which case the generic parser is invoked. The parsed attribute is
   /// populated in `result`.
   template <typename AttrType>
   std::enable_if_t<detect_has_parse_method<AttrType>::value, ParseResult>
-  parseCustomAttributeWithFallback(AttrType &result) {
+  parseCustomAttributeWithFallback(AttrType &result, Type type = {}) {
     SMLoc loc = getCurrentLocation();
 
     // Parse any kind of attribute.
     Attribute attr;
     if (parseCustomAttributeWithFallback(
-            attr, {}, [&](Attribute &result, Type type) -> ParseResult {
+            attr, type, [&](Attribute &result, Type type) -> ParseResult {
               result = AttrType::parse(*this, type);
               return success(!!result);
             }))
       return failure();
 
     // Check for the right kind of attribute.
     result = attr.dyn_cast<AttrType>();
     if (!result)
       return emitError(loc, "invalid kind of attribute specified");
     return success();
   }
 
   /// SFINAE parsing method for Attribute that don't implement a parse method.
   template <typename AttrType>
   std::enable_if_t<!detect_has_parse_method<AttrType>::value, ParseResult>
-  parseCustomAttributeWithFallback(AttrType &result) {
-    return parseAttribute(result);
+  parseCustomAttributeWithFallback(AttrType &result, Type type = {}) {
+    return parseAttribute(result, type);
   }
 
   /// Parse an arbitrary optional attribute of a given type and return it in
   /// result.
   virtual OptionalParseResult parseOptionalAttribute(Attribute &result,
                                                      Type type = {}) = 0;
 
   /// Parse an optional array attribute and return it in result.
public:
   virtual ParseResult parseGenericOperationAfterOpName(
       OperationState &result,
       std::optional<ArrayRef<UnresolvedOperand>> parsedOperandType =
           std::nullopt,
       std::optional<ArrayRef<Block *>> parsedSuccessors = std::nullopt,
       std::optional<MutableArrayRef<std::unique_ptr<Region>>> parsedRegions =
           std::nullopt,
       std::optional<ArrayRef<NamedAttribute>> parsedAttributes = std::nullopt,
+      std::optional<Attribute> parsedPropertiesAttribute = std::nullopt,
       std::optional<FunctionType> parsedFnType = std::nullopt) = 0;
 
   /// Parse a single SSA value operand name along with a result number if
   /// `allowResultNumber` is true.
   virtual ParseResult parseOperand(UnresolvedOperand &result,
                                    bool allowResultNumber = true) = 0;
 
   /// Parse a single operand if present.

mlir/include/mlir/IR/Operation.h

 #include "mlir/IR/BuiltinAttributes.h"
 #include "mlir/IR/Diagnostics.h"
 #include "mlir/IR/OperationSupport.h"
 #include "mlir/IR/Region.h"
 #include "llvm/ADT/Twine.h"
 #include <optional>
 
 namespace mlir {
+namespace detail {
+/// This is a "tag" used for mapping the properties storage in
+/// llvm::TrailingObjects.
+enum class OpProperties : char {};
+} // namespace detail
+
 /// Operation is the basic unit of execution within MLIR.
 ///
 /// The following documentation are recommended to understand this class:
 /// - https://mlir.llvm.org/docs/LangRef/#operations
 /// - https://mlir.llvm.org/docs/Tutorials/UnderstandingTheIRStructure/
 ///
 /// An Operation is defined first by its name, which is a unique string. The
 /// name is interpreted so that if it contains a '.' character, the part before

 ///
 /// An Operation may contain optionally one or multiple Regions, stored in a
 /// tail allocated array. Each `Region` is a list of Blocks. Each `Block` is
 /// itself a list of Operations. This structure is effectively forming a tree.
 ///
 /// Some operations like branches also refer to other Block, in which case they
 /// would have an array of `BlockOperand`.
 ///
+/// An Operation may contain optionally a "Properties" object: this is a
+/// pre-defined C++ object with a fixed size. This object is owned by the
+/// operation and deleted with the operation. It can be converted to an
+/// Attribute on demand, or loaded from an Attribute.
+///
+///
 /// Finally an Operation also contain an optional `DictionaryAttr`, a Location,
 /// and a pointer to its parent Block (if any).
 class alignas(8) Operation final
     : public llvm::ilist_node_with_parent<Operation, Block>,
       private llvm::TrailingObjects<Operation, detail::OperandStorage,
-                                    BlockOperand, Region, OpOperand> {
+                                    detail::OpProperties, BlockOperand, Region,
+                                    OpOperand> {
 public:
   /// Create a new Operation with the specific fields. This constructor
   /// populates the provided attribute list with default attributes if
   /// necessary.
   static Operation *create(Location location, OperationName name,
                            TypeRange resultTypes, ValueRange operands,
-                           NamedAttrList &&attributes, BlockRange successors,
+                           NamedAttrList &&attributes,
+                           OpaqueProperties properties, BlockRange successors,
                            unsigned numRegions);
 
   /// Create a new Operation with the specific fields. This constructor uses an
   /// existing attribute dictionary to avoid uniquing a list of attributes.
   static Operation *create(Location location, OperationName name,
                            TypeRange resultTypes, ValueRange operands,
-                           DictionaryAttr attributes, BlockRange successors,
+                           DictionaryAttr attributes,
+                           OpaqueProperties properties, BlockRange successors,
                            unsigned numRegions);
 
   /// Create a new Operation from the fields stored in `state`.
   static Operation *create(const OperationState &state);
 
   /// Create a new Operation with the specific fields.
   static Operation *create(Location location, OperationName name,
                            TypeRange resultTypes, ValueRange operands,
                            NamedAttrList &&attributes,
+                           OpaqueProperties properties,
                            BlockRange successors = {},
                            RegionRange regions = {});
 
   /// The name of an operation is the key identifier for it.
   OperationName getName() { return name; }
 
   /// If this operation has a registered operation description, return it.
   /// Otherwise return std::nullopt.
public:
   Region *getParentRegion() { return block ? block->getParent() : nullptr; }
 
   /// Returns the closest surrounding operation that contains this operation
   /// or nullptr if this is a top-level operation.
   Operation *getParentOp() { return block ? block->getParentOp() : nullptr; }
 
   /// Return the closest surrounding parent operation that is of type 'OpTy'.
   template <typename OpTy>
   OpTy getParentOfType() {
     auto *op = this;
     while ((op = op->getParentOp()))
       if (auto parentOp = dyn_cast<OpTy>(op))
         return parentOp;
     return OpTy();
   }
 
   /// Returns the closest surrounding parent operation with trait `Trait`.
public:
   //===--------------------------------------------------------------------===//
   // Attributes
   //===--------------------------------------------------------------------===//
 
   // Operations may optionally carry a list of attributes that associate
   // constants to names.  Attributes may be dynamically added and removed over
   // the lifetime of an operation.
 
+  /// Access an inherent attribute by name: returns an empty optional if there
+  /// is no inherent attribute with this name.
+  ///
+  /// This method is available as a transient facility in the migration process
+  /// to use Properties instead.
+  std::optional<Attribute> getInherentAttr(StringRef name);
+
+  /// Set an inherent attribute by name.
+  ///
+  /// This method is available as a transient facility in the migration process
+  /// to use Properties instead.
+  void setInherentAttr(StringAttr name, Attribute value);
+
+  /// Access a discardable attribute by name, returns an null Attribute if the
+  /// discardable attribute does not exist.
+  Attribute getDiscardableAttr(StringRef name) { return attrs.get(name); }
+
+  /// Access a discardable attribute by name, returns an null Attribute if the
+  /// discardable attribute does not exist.
+  Attribute getDiscardableAttr(StringAttr name) { return attrs.get(name); }
+
+  /// Set a discardable attribute by name.
+  void setDiscardableAttr(StringAttr name, Attribute value) {
+    NamedAttrList attributes(attrs);
+    if (attributes.set(name, value) != value)
+      attrs = attributes.getDictionary(getContext());
+  }
+
+  /// Return all of the discardable attributes on this operation.
+  ArrayRef<NamedAttribute> getDiscardableAttrs() { return attrs.getValue(); }
+
+  /// Return all of the discardable attributes on this operation as a
+  /// DictionaryAttr.
+  DictionaryAttr getDiscardableAttrDictionary() { return attrs; }
+
   /// Return all of the attributes on this operation.
-  ArrayRef<NamedAttribute> getAttrs() { return attrs.getValue(); }
+  ArrayRef<NamedAttribute> getAttrs() {
+    if (!getPropertiesStorage())
+      return getDiscardableAttrs();
+    return getAttrDictionary().getValue();
+  }
 
   /// Return all of the attributes on this operation as a DictionaryAttr.
-  DictionaryAttr getAttrDictionary() { return attrs; }
+  DictionaryAttr getAttrDictionary();
 
-  /// Set the attribute dictionary on this operation.
-  void setAttrs(DictionaryAttr newAttrs) {
+  /// Set the attributes from a dictionary on this operation.
+  /// These methods are expensive: if the dictionnary only contains discardable
+  /// attributes, `setDiscardableAttrs` is more efficient.
+  void setAttrs(DictionaryAttr newAttrs);
+  void setAttrs(ArrayRef<NamedAttribute> newAttrs);
+  /// Set the discardable attribute dictionary on this operation.
+  void setDiscardableAttrs(DictionaryAttr newAttrs) {
     assert(newAttrs && "expected valid attribute dictionary");
     attrs = newAttrs;
   }
-  void setAttrs(ArrayRef<NamedAttribute> newAttrs) {
-    setAttrs(DictionaryAttr::get(getContext(), newAttrs));
+  void setDiscardableAttrs(ArrayRef<NamedAttribute> newAttrs) {
+    setDiscardableAttrs(DictionaryAttr::get(getContext(), newAttrs));
   }
 
   /// Return the specified attribute if present, null otherwise.
-  Attribute getAttr(StringAttr name) { return attrs.get(name); }
-  Attribute getAttr(StringRef name) { return attrs.get(name); }
+  /// These methods are expensive: if the dictionnary only contains discardable
+  /// attributes, `getDiscardableAttr` is more efficient.
+  Attribute getAttr(StringAttr name) {
+    if (getPropertiesStorageSize()) {
+      if (std::optional<Attribute> inherentAttr = getInherentAttr(name))
+        return *inherentAttr;
+    }
+    return attrs.get(name);
+  }
+  Attribute getAttr(StringRef name) {
+    if (getPropertiesStorageSize()) {
+      if (std::optional<Attribute> inherentAttr = getInherentAttr(name))
+        return *inherentAttr;
+    }
+    return attrs.get(name);
+  }
 
   template <typename AttrClass>
   AttrClass getAttrOfType(StringAttr name) {
     return getAttr(name).dyn_cast_or_null<AttrClass>();
   }
   template <typename AttrClass>
   AttrClass getAttrOfType(StringRef name) {
     return getAttr(name).dyn_cast_or_null<AttrClass>();
   }
 
   /// Return true if the operation has an attribute with the provided name,
   /// false otherwise.
-  bool hasAttr(StringAttr name) { return attrs.contains(name); }
-  bool hasAttr(StringRef name) { return attrs.contains(name); }
+  bool hasAttr(StringAttr name) {
+    if (getPropertiesStorageSize()) {
+      if (std::optional<Attribute> inherentAttr = getInherentAttr(name))
+        return (bool)*inherentAttr;
+    }
+    return attrs.contains(name);
+  }
+  bool hasAttr(StringRef name) {
+    if (getPropertiesStorageSize()) {
+      if (std::optional<Attribute> inherentAttr = getInherentAttr(name))
+        return (bool)*inherentAttr;
+    }
+    return attrs.contains(name);
+  }
   template <typename AttrClass, typename NameT>
   bool hasAttrOfType(NameT &&name) {
     return static_cast<bool>(
         getAttrOfType<AttrClass>(std::forward<NameT>(name)));
   }
 
   /// If the an attribute exists with the specified name, change it to the new
   /// value. Otherwise, add a new attribute with the specified name/value.
   void setAttr(StringAttr name, Attribute value) {
+    if (getPropertiesStorageSize()) {
+      if (std::optional<Attribute> inherentAttr = getInherentAttr(name)) {
+        setInherentAttr(name, value);
+        return;
+      }
+    }
     NamedAttrList attributes(attrs);
     if (attributes.set(name, value) != value)
       attrs = attributes.getDictionary(getContext());
   }
   void setAttr(StringRef name, Attribute value) {
     setAttr(StringAttr::get(getContext(), name), value);
   }
 
   /// Remove the attribute with the specified name if it exists. Return the
   /// attribute that was erased, or nullptr if there was no attribute with such
   /// name.
   Attribute removeAttr(StringAttr name) {
+    if (getPropertiesStorageSize()) {
+      if (std::optional<Attribute> inherentAttr = getInherentAttr(name)) {
+        setInherentAttr(name, {});
+        return *inherentAttr;
+      }
+    }
     NamedAttrList attributes(attrs);
     Attribute removedAttr = attributes.erase(name);
     if (removedAttr)
       attrs = attributes.getDictionary(getContext());
     return removedAttr;
   }
   Attribute removeAttr(StringRef name) {
     return removeAttr(StringAttr::get(getContext(), name));
dialect_attr_iterator dialect_attr_begin() {
     auto attrs = getAttrs();
     return dialect_attr_iterator(attrs.begin(), attrs.end());
   }
   dialect_attr_iterator dialect_attr_end() {
     auto attrs = getAttrs();
     return dialect_attr_iterator(attrs.end(), attrs.end());
   }
 
-  /// Set the dialect attributes for this operation, and preserve all dependent.
+  /// Set the dialect attributes for this operation, and preserve all inherent.
   template <typename DialectAttrT>
   void setDialectAttrs(DialectAttrT &&dialectAttrs) {
     NamedAttrList attrs;
     attrs.append(std::begin(dialectAttrs), std::end(dialectAttrs));
     for (auto attr : getAttrs())
       if (!attr.getName().strref().contains('.'))
         attrs.push_back(attr);
     setAttrs(attrs.getDictionary(getContext()));
public:
   /// Emit a warning about this operation, reporting up to any diagnostic
   /// handlers that may be listening.
   InFlightDiagnostic emitWarning(const Twine &message = {});
 
   /// Emit a remark about this operation, reporting up to any diagnostic
   /// handlers that may be listening.
   InFlightDiagnostic emitRemark(const Twine &message = {});
 
+  /// Returns the properties storage size.
+  int getPropertiesStorageSize() const {
+    return ((int)propertiesStorageSize) * 8;
+  }
+  /// Returns the properties storage.
+  OpaqueProperties getPropertiesStorage() {
+    if (propertiesStorageSize)
+      return {
+          reinterpret_cast<void *>(getTrailingObjects<detail::OpProperties>())};
+    return {nullptr};
+  }
+  OpaqueProperties getPropertiesStorage() const {
+    if (propertiesStorageSize)
+      return {reinterpret_cast<void *>(const_cast<detail::OpProperties *>(
+          getTrailingObjects<detail::OpProperties>()))};
+    return {nullptr};
+  }
+
+  /// Return the properties converted to an attribute.
+  /// This is expensive, and mostly useful when dealing with unregistered
+  /// operation. Returns an empty attribute if no properties are present.
+  Attribute getPropertiesAsAttribute();
+
+  /// Set the properties from the provided  attribute.
+  /// This is an expensive operation that can fail if the attribute is not
+  /// matching the expectations of the properties for this operation. This is
+  /// mostly useful for unregistered operations or used when parsing the
+  /// generic format. An optional diagnostic can be passed in for richer errors.
+  LogicalResult setPropertiesFromAttribute(Attribute attr,
+                                           InFlightDiagnostic *diagnostic);
+
+  /// Copy properties from an existing other properties object. The two objects
+  /// must be the same type.
+  void copyProperties(OpaqueProperties rhs);
+
+  /// Compute a hash for the op properties (if any).
+  llvm::hash_code hashProperties();
+
 private:
   //===--------------------------------------------------------------------===//
   // Ordering
   //===--------------------------------------------------------------------===//
 
   /// This value represents an invalid index ordering for an operation within a
   /// block.
   static constexpr unsigned kInvalidOrderIdx = -1;
 
   /// This value represents the stride to use when computing a new order for an
   /// operation.
   static constexpr unsigned kOrderStride = 5;
 
   /// Update the order index of this operation of this operation if necessary,
   /// potentially recomputing the order of the parent block.
   void updateOrderIfNecessary();
 
   /// Returns true if this operation has a valid order.
   bool hasValidOrder() { return orderIndex != kInvalidOrderIdx; }
 
 private:
   Operation(Location location, OperationName name, unsigned numResults,
             unsigned numSuccessors, unsigned numRegions,
-            DictionaryAttr attributes, bool hasOperandStorage);
+            int propertiesStorageSize, DictionaryAttr attributes,
+            OpaqueProperties properties, bool hasOperandStorage);
 
   // Operations are deleted through the destroy() member because they are
   // allocated with malloc.
   ~Operation();
 
   /// Returns the additional size necessary for allocating the given objects
   /// before an Operation in-memory.
   static size_t prefixAllocSize(unsigned numOutOfLineResults,
#endif
   Location location;
 
   /// Relative order of this operation in its parent block. Used for
   /// O(1) local dominance checks between operations.
   mutable unsigned orderIndex = 0;
 
   const unsigned numResults;
   const unsigned numSuccs;
-  const unsigned numRegions : 31;
+  const unsigned numRegions : 23;
 
   /// This bit signals whether this operation has an operand storage or not. The
   /// operand storage may be elided for operations that are known to never have
   /// operands.
   bool hasOperandStorage : 1;
 
+  /// The size of the storage for properties (if any), divided by 8: since the
+  /// Properties storage will always be rounded up to the next multiple of 8 we
+  /// save some bits here.
+  unsigned char propertiesStorageSize : 8;
+  /// This is the maximum size we support to allocate properties inline with an
+  /// operation: this must match the bitwidth above.
+  static constexpr int64_t propertiesCapacity = 8 * 256;
+
   /// This holds the name of the operation.
   OperationName name;
 
   /// This holds general named attributes for the operation.
   DictionaryAttr attrs;
 
   // allow ilist_traits access to 'block' field.
   friend struct llvm::ilist_traits<Operation>;
 
   // allow block to access the 'orderIndex' field.
   friend class Block;
 
   // allow value to access the 'ResultStorage' methods.
   friend class Value;
 
   // allow ilist_node_with_parent to access the 'getParent' method.
   friend class llvm::ilist_node_with_parent<Operation, Block>;
 
   // This stuff is used by the TrailingObjects template.
-  friend llvm::TrailingObjects<Operation, detail::OperandStorage, BlockOperand,
-                               Region, OpOperand>;
+  friend llvm::TrailingObjects<Operation, detail::OperandStorage,
+                               detail::OpProperties, BlockOperand, Region,
+                               OpOperand>;
   size_t numTrailingObjects(OverloadToken<detail::OperandStorage>) const {
     return hasOperandStorage ? 1 : 0;
   }
   size_t numTrailingObjects(OverloadToken<BlockOperand>) const {
     return numSuccs;
   }
   size_t numTrailingObjects(OverloadToken<Region>) const { return numRegions; }
+  size_t numTrailingObjects(OverloadToken<detail::OpProperties>) const {
+    return getPropertiesStorageSize();
+  }
 };
 
 inline raw_ostream &operator<<(raw_ostream &os, const Operation &op) {
   const_cast<Operation &>(op).print(os, OpPrintingFlags().useLocalScope());
   return os;
 }
 
 } // namespace mlir

mlir/include/mlir/IR/OperationSupport.h

 //===- OperationSupport.h ---------------------------------------*- 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 defines a number of support types that Operation and related
 // classes build on top of.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef MLIR_IR_OPERATIONSUPPORT_H
 #define MLIR_IR_OPERATIONSUPPORT_H
 
+#include "mlir/IR/Attributes.h"
 #include "mlir/IR/BlockSupport.h"
 #include "mlir/IR/BuiltinAttributes.h"
+#include "mlir/IR/Diagnostics.h"
 #include "mlir/IR/Location.h"
 #include "mlir/IR/TypeRange.h"
 #include "mlir/IR/Types.h"
 #include "mlir/IR/Value.h"
 #include "mlir/Support/InterfaceSupport.h"
 #include "llvm/ADT/BitmaskEnum.h"
 #include "llvm/ADT/PointerUnion.h"
 #include "llvm/ADT/STLFunctionalExtras.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/PointerLikeTypeTraits.h"
 #include "llvm/Support/TrailingObjects.h"
 #include <memory>
 #include <optional>
 
 namespace llvm {
 class BitVector;
 } // namespace llvm
 
 namespace mlir {
 class Dialect;
 class DictionaryAttr;
 class ElementsAttr;
+struct EmptyProperties;
 class MutableOperandRangeRange;
 class NamedAttrList;
 class Operation;
 struct OperationState;
 class OpAsmParser;
 class OpAsmParserResult;
 class OpAsmPrinter;
 class OperandRange;
 class OperandRangeRange;
 class OpFoldResult;
 class ParseResult;
 class Pattern;
 class Region;
 class ResultRange;
 class RewritePattern;
 class RewritePatternSet;
 class Type;
 class Value;
 class ValueRange;
 template <typename ValueRangeT>
 class ValueTypeRange;
 
 //===----------------------------------------------------------------------===//
+// OpaqueProperties
+//===----------------------------------------------------------------------===//
+
+/// Simple wrapper around a void* in order to express generically how to pass
+/// in op properties through APIs.
+class OpaqueProperties {
+public:
+  OpaqueProperties(void *prop) : properties(prop) {}
+  operator bool() const { return properties != nullptr; }
+  template <typename Dest>
+  Dest as() const {
+    return static_cast<Dest>(const_cast<void *>(properties));
+  }
+
+private:
+  void *properties;
+};
+
+//===----------------------------------------------------------------------===//
 // OperationName
 //===----------------------------------------------------------------------===//
 
 class OperationName {
 public:
   using FoldHookFn = llvm::unique_function<LogicalResult(
       Operation *, ArrayRef<Attribute>, SmallVectorImpl<OpFoldResult> &) const>;
   using HasTraitFn = llvm::unique_function<bool(TypeID) const>;
virtual void getCanonicalizationPatterns(RewritePatternSet &,
                                              MLIRContext *) = 0;
     virtual bool hasTrait(TypeID) = 0;
     virtual OperationName::ParseAssemblyFn getParseAssemblyFn() = 0;
     virtual void populateDefaultAttrs(const OperationName &,
                                       NamedAttrList &) = 0;
     virtual void printAssembly(Operation *, OpAsmPrinter &, StringRef) = 0;
     virtual LogicalResult verifyInvariants(Operation *) = 0;
     virtual LogicalResult verifyRegionInvariants(Operation *) = 0;
+    /// Implementation for properties
+    virtual std::optional<Attribute> getInherentAttr(Operation *,
+                                                     StringRef name) = 0;
+    virtual void setInherentAttr(Operation *op, StringAttr name,
+                                 Attribute value) = 0;
+    virtual void populateInherentAttrs(Operation *op, NamedAttrList &attrs) = 0;
+    virtual LogicalResult
+    verifyInherentAttrs(OperationName opName, NamedAttrList &attributes,
+                        function_ref<InFlightDiagnostic()> getDiag) = 0;
+    virtual int getOpPropertyByteSize() = 0;
+    virtual void initProperties(OperationName opName, OpaqueProperties storage,
+                                OpaqueProperties init) = 0;
+    virtual void deleteProperties(OpaqueProperties) = 0;
+    virtual void populateDefaultProperties(OperationName opName,
+                                           OpaqueProperties properties) = 0;
+    virtual LogicalResult setPropertiesFromAttr(Operation *, Attribute,
+                                                InFlightDiagnostic *) = 0;
+    virtual Attribute getPropertiesAsAttr(Operation *) = 0;
+    virtual void copyProperties(OpaqueProperties, OpaqueProperties) = 0;
+    virtual llvm::hash_code hashProperties(OpaqueProperties) = 0;
   };
 
 public:
   class Impl : public InterfaceConcept {
   public:
     Impl(StringRef, Dialect *dialect, TypeID typeID,
          detail::InterfaceMap interfaceMap);
     Impl(StringAttr name, Dialect *dialect, TypeID typeID,
LogicalResult foldHook(Operation *, ArrayRef<Attribute>,
                            SmallVectorImpl<OpFoldResult> &) final;
     void getCanonicalizationPatterns(RewritePatternSet &, MLIRContext *) final;
     bool hasTrait(TypeID) final;
     OperationName::ParseAssemblyFn getParseAssemblyFn() final;
     void populateDefaultAttrs(const OperationName &, NamedAttrList &) final;
     void printAssembly(Operation *, OpAsmPrinter &, StringRef) final;
     LogicalResult verifyInvariants(Operation *) final;
     LogicalResult verifyRegionInvariants(Operation *) final;
+    /// Implementation for properties
+    std::optional<Attribute> getInherentAttr(Operation *op,
+                                             StringRef name) final;
+    void setInherentAttr(Operation *op, StringAttr name, Attribute value) final;
+    void populateInherentAttrs(Operation *op, NamedAttrList &attrs) final;
+    LogicalResult
+    verifyInherentAttrs(OperationName opName, NamedAttrList &attributes,
+                        function_ref<InFlightDiagnostic()> getDiag) final;
+    int getOpPropertyByteSize() final;
+    void initProperties(OperationName opName, OpaqueProperties storage,
+                        OpaqueProperties init) final;
+    void deleteProperties(OpaqueProperties) final;
+    void populateDefaultProperties(OperationName opName,
+                                   OpaqueProperties properties) final;
+    LogicalResult setPropertiesFromAttr(Operation *, Attribute,
+                                        InFlightDiagnostic *) final;
+    Attribute getPropertiesAsAttr(Operation *) final;
+    void copyProperties(OpaqueProperties, OpaqueProperties) final;
+    llvm::hash_code hashProperties(OpaqueProperties) final;
   };
 
 public:
   OperationName(StringRef name, MLIRContext *context);
 
   /// Return if this operation is registered.
   bool isRegistered() const { return getImpl()->isRegistered(); }
 
public:
   template <typename T>
   bool mightHaveInterface() const {
     return mightHaveInterface(TypeID::get<T>());
   }
   bool mightHaveInterface(TypeID interfaceID) const {
     return !isRegistered() || hasInterface(interfaceID);
   }
 
+  /// Lookup an inherent attribute by name, this method isn't recommended
+  /// and may be removed in the future.
+  std::optional<Attribute> getInherentAttr(Operation *op,
+                                           StringRef name) const {
+    return getImpl()->getInherentAttr(op, name);
+  }
+
+  void setInherentAttr(Operation *op, StringAttr name, Attribute value) const {
+    return getImpl()->setInherentAttr(op, name, value);
+  }
+
+  void populateInherentAttrs(Operation *op, NamedAttrList &attrs) const {
+    return getImpl()->populateInherentAttrs(op, attrs);
+  }
+  /// This method exists for backward compatibility purpose when using
+  /// properties to store inherent attributes, it enables validating the
+  /// attributes when parsed from the older generic syntax pre-Properties.
+  LogicalResult
+  verifyInherentAttrs(NamedAttrList &attributes,
+                      function_ref<InFlightDiagnostic()> getDiag) const {
+    return getImpl()->verifyInherentAttrs(*this, attributes, getDiag);
+  }
+  /// This hooks return the number of bytes to allocate for the op properties.
+  int getOpPropertyByteSize() const {
+    return getImpl()->getOpPropertyByteSize();
+  }
+
+  /// This hooks destroy the op properties.
+  void destroyOpProperties(OpaqueProperties properties) const {
+    getImpl()->deleteProperties(properties);
+  }
+
+  /// Initialize the op properties.
+  void initOpProperties(OpaqueProperties storage, OpaqueProperties init) const {
+    getImpl()->initProperties(*this, storage, init);
+  }
+
+  /// Set the default values on the ODS attribute in the properties.
+  void populateDefaultProperties(OpaqueProperties properties) const {
+    getImpl()->populateDefaultProperties(*this, properties);
+  }
+
+  /// Return the op properties converted to an Attribute.
+  Attribute getOpPropertiesAsAttribute(Operation *op) const {
+    return getImpl()->getPropertiesAsAttr(op);
+  }
+
+  /// Define the op properties from the provided Attribute.
+  LogicalResult
+  setOpPropertiesFromAttribute(Operation *op, Attribute properties,
+                               InFlightDiagnostic *diagnostic) const {
+    return getImpl()->setPropertiesFromAttr(op, properties, diagnostic);
+  }
+
+  void copyOpProperties(OpaqueProperties lhs, OpaqueProperties rhs) const {
+    return getImpl()->copyProperties(lhs, rhs);
+  }
+
+  llvm::hash_code hashOpProperties(OpaqueProperties properties) const {
+    return getImpl()->hashProperties(properties);
+  }
+
   /// Return the dialect this operation is registered to if the dialect is
   /// loaded in the context, or nullptr if the dialect isn't loaded.
   Dialect *getDialect() const {
     return isRegistered() ? getImpl()->getDialect()
                           : getImpl()->getName().getReferencedDialect();
   }
 
   /// Return the name of the dialect this operation is registered to.

 }
 
 //===----------------------------------------------------------------------===//
 // RegisteredOperationName
 //===----------------------------------------------------------------------===//
 
 /// This is a "type erased" representation of a registered operation. This
 /// should only be used by things like the AsmPrinter and other things that need
 /// to be parameterized by generic operation hooks. Most user code should use
 /// the concrete operation types.
 class RegisteredOperationName : public OperationName {
 public:
   /// Implementation of the InterfaceConcept for operation APIs that forwarded
   /// to a concrete op implementation.
   template <typename ConcreteOp>
   struct Model : public Impl {
     Model(Dialect *dialect)
void printAssembly(Operation *op, OpAsmPrinter &printer,
                        StringRef name) final {
       ConcreteOp::getPrintAssemblyFn()(op, printer, name);
     }
     LogicalResult verifyInvariants(Operation *op) final {
       return ConcreteOp::getVerifyInvariantsFn()(op);
     }
     LogicalResult verifyRegionInvariants(Operation *op) final {
       return ConcreteOp::getVerifyRegionInvariantsFn()(op);
     }
+
+    /// Implementation for "Properties"
+
+    using Properties = std::remove_reference_t<
+        decltype(std::declval<ConcreteOp>().getProperties())>;
+
+    std::optional<Attribute> getInherentAttr(Operation *op,
+                                             StringRef name) final {
+      if constexpr (hasProperties) {
+        auto concreteOp = cast<ConcreteOp>(op);
+        return ConcreteOp::getInherentAttr(concreteOp.getProperties(), name);
+      }
+      // If the op does not have support for properties, we dispatch back to the
+      // dictionnary of discardable attributes for now.
+      return cast<ConcreteOp>(op)->getDiscardableAttr(name);
+    }
+    void setInherentAttr(Operation *op, StringAttr name,
+                         Attribute value) final {
+      if constexpr (hasProperties) {
+        auto concreteOp = cast<ConcreteOp>(op);
+        return ConcreteOp::setInherentAttr(concreteOp.getProperties(), name,
+                                           value);
+      }
+      // If the op does not have support for properties, we dispatch back to the
+      // dictionnary of discardable attributes for now.
+      return cast<ConcreteOp>(op)->setDiscardableAttr(name, value);
+    }
+    void populateInherentAttrs(Operation *op, NamedAttrList &attrs) final {
+      if constexpr (hasProperties) {
+        auto concreteOp = cast<ConcreteOp>(op);
+        ConcreteOp::populateInherentAttrs(concreteOp.getProperties(), attrs);
+      }
+    }
+    LogicalResult
+    verifyInherentAttrs(OperationName opName, NamedAttrList &attributes,
+                        function_ref<InFlightDiagnostic()> getDiag) final {
+      if constexpr (hasProperties)
+        return ConcreteOp::verifyInherentAttrs(opName, attributes, getDiag);
+      return success();
+    }
+    // Detect if the concrete operation defined properties.
+    static constexpr bool hasProperties = !std::is_same_v<
+        typename ConcreteOp::template InferredProperties<ConcreteOp>,
+        EmptyProperties>;
+
+    int getOpPropertyByteSize() final {
+      if constexpr (hasProperties)
+        return sizeof(Properties);
+      return 0;
+    }
+    void initProperties(OperationName opName, OpaqueProperties storage,
+                        OpaqueProperties init) final {
+      using Properties =
+          typename ConcreteOp::template InferredProperties<ConcreteOp>;
+      if (init)
+        new (storage.as<Properties *>()) Properties(*init.as<Properties *>());
+      else
+        new (storage.as<Properties *>()) Properties();
+      if constexpr (hasProperties)
+        ConcreteOp::populateDefaultProperties(opName,
+                                              *storage.as<Properties *>());
+    }
+    void deleteProperties(OpaqueProperties prop) final {
+      prop.as<Properties *>()->~Properties();
+    }
+    void populateDefaultProperties(OperationName opName,
+                                   OpaqueProperties properties) final {
+      if constexpr (hasProperties)
+        ConcreteOp::populateDefaultProperties(opName,
+                                              *properties.as<Properties *>());
+    }
+
+    LogicalResult setPropertiesFromAttr(Operation *op, Attribute attr,
+                                        InFlightDiagnostic *diag) final {
+      if constexpr (hasProperties)
+        return ConcreteOp::setPropertiesFromAttr(
+            cast<ConcreteOp>(op).getProperties(), attr, diag);
+      if (diag)
+        *diag << "This operation does not support properties";
+      return failure();
+    }
+    Attribute getPropertiesAsAttr(Operation *op) final {
+      if constexpr (hasProperties) {
+        auto concreteOp = cast<ConcreteOp>(op);
+        return ConcreteOp::getPropertiesAsAttr(concreteOp->getContext(),
+                                               concreteOp.getProperties());
+      }
+      return {};
+    }
+    void copyProperties(OpaqueProperties lhs, OpaqueProperties rhs) final {
+      *lhs.as<Properties *>() = *rhs.as<Properties *>();
+    }
+    llvm::hash_code hashProperties(OpaqueProperties prop) final {
+      if constexpr (hasProperties)
+        return ConcreteOp::computePropertiesHash(*prop.as<Properties *>());
+
+      return {};
+    }
   };
 
   /// Lookup the registered operation information for the given operation.
   /// Returns std::nullopt if the operation isn't registered.
   static std::optional<RegisteredOperationName> lookup(StringRef name,
                                                        MLIRContext *ctx);
 
   /// Register a new operation in a Dialect object.
   /// This constructor is used by Dialect objects when they register the list
public:
   /// Replaces the attributes with new list of attributes.
   void assign(const_iterator inStart, const_iterator inEnd);
 
   /// Replaces the attributes with new list of attributes.
   void assign(ArrayRef<NamedAttribute> range) {
     assign(range.begin(), range.end());
   }
 
+  void clear() {
+    attrs.clear();
+    dictionarySorted.setPointerAndInt(nullptr, false);
+  }
+
   bool empty() const { return attrs.empty(); }
 
   void reserve(size_type N) { attrs.reserve(N); }
 
   /// Add an attribute with the specified name.
   void push_back(NamedAttribute newAttribute);
 
   /// Pop last element from list.
struct OperationState {
   /// Types of the results of this operation.
   SmallVector<Type, 4> types;
   NamedAttrList attributes;
   /// Successors of this operation and their respective operands.
   SmallVector<Block *, 1> successors;
   /// Regions that the op will hold.
   SmallVector<std::unique_ptr<Region>, 1> regions;
 
+  // If we're creating an unregistered operation, this Attribute is used to
+  // build the properties. Otherwise it is ignored. For registered operations
+  // see the `getOrAddProperties` method.
+  Attribute propertiesAttr;
+
+private:
+  OpaqueProperties properties = nullptr;
+  TypeID propertiesId;
+  llvm::function_ref<void(OpaqueProperties)> propertiesDeleter;
+  llvm::function_ref<void(OpaqueProperties, const OpaqueProperties)>
+      propertiesSetter;
+  friend class Operation;
+
 public:
   OperationState(Location location, StringRef name);
   OperationState(Location location, OperationName name);
 
   OperationState(Location location, OperationName name, ValueRange operands,
                  TypeRange types, ArrayRef<NamedAttribute> attributes = {},
                  BlockRange successors = {},
                  MutableArrayRef<std::unique_ptr<Region>> regions = {});
   OperationState(Location location, StringRef name, ValueRange operands,
                  TypeRange types, ArrayRef<NamedAttribute> attributes = {},
                  BlockRange successors = {},
                  MutableArrayRef<std::unique_ptr<Region>> regions = {});
+  OperationState(OperationState &&other) = default;
+  OperationState(const OperationState &other) = default;
+  OperationState &operator=(OperationState &&other) = default;
+  OperationState &operator=(const OperationState &other) = default;
+  ~OperationState();
+
+  /// Get (or create) a properties of the provided type to be set on the
+  /// operation on creation.
+  template <typename T>
+  T &getOrAddProperties() {
+    if (!properties) {
+      T *p = new T{};
+      properties = p;
+      propertiesDeleter = [](OpaqueProperties prop) {
+        delete prop.as<const T *>();
+      };
+      propertiesSetter = [](OpaqueProperties new_prop,
+                            const OpaqueProperties prop) {
+        *new_prop.as<T *>() = *prop.as<const T *>();
+      };
+      propertiesId = TypeID::get<T>();
+    }
+    assert(propertiesId == TypeID::get<T>() && "Inconsistent properties");
+    return *properties.as<T *>();
+  }
+  OpaqueProperties getRawProperties() { return properties; }
+
+  // Set the properties defined on this OpState on the given operation,
+  // optionally emit diagnostics on error through the provided diagnostic.
+  LogicalResult setProperties(Operation *op,
+                              InFlightDiagnostic *diagnostic) const;
 
   void addOperands(ValueRange newOperands);
 
   void addTypes(ArrayRef<Type> newTypes) {
     types.append(newTypes.begin(), newTypes.end());
   }
   template <typename RangeT>
   std::enable_if_t<!std::is_convertible<RangeT, ArrayRef<Type>>::value>
struct OperationEquivalence {
   }
   /// Helper that can be used with `isEquivalentTo` above to consider ops
   /// equivalent only if their operands are the exact same SSA values.
   static LogicalResult exactValueMatch(Value lhs, Value rhs) {
     return success(lhs == rhs);
   }
 };
 
 /// Enable Bitmask enums for OperationEquivalence::Flags.
 LLVM_ENABLE_BITMASK_ENUMS_IN_NAMESPACE();
 
 //===----------------------------------------------------------------------===//
 // OperationFingerPrint
 //===----------------------------------------------------------------------===//
 
 /// A unique fingerprint for a specific operation, and all of it's internal
 /// operations.
 class OperationFingerPrint {
 public:
   OperationFingerPrint(Operation *topOp);

mlir/include/mlir/Interfaces/InferTypeOpInterface.h

 // shape inference function.
 //
 // TODO: Consider generating typedefs for trait member functions if this usage
 // becomes more common.
 LogicalResult inferReturnTensorTypes(
     function_ref<
         LogicalResult(MLIRContext *, std::optional<Location> location,
                       ValueShapeRange operands, DictionaryAttr attributes,
-                      RegionRange regions,
+                      OpaqueProperties properties, RegionRange regions,
                       SmallVectorImpl<ShapedTypeComponents> &retComponents)>
         componentTypeFn,
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes);
 
 /// Verifies that the inferred result types match the actual result types for
 /// the op. Precondition: op implements InferTypeOpInterface.
 LogicalResult verifyInferredResultTypes(Operation *op);
 } // namespace detail
 
 namespace OpTrait {

 ///   trait is currently only used where the interfaces are, so keep it
 ///   restricted for now).
 template <typename ConcreteType>
 class InferTensorType : public TraitBase<ConcreteType, InferTensorType> {
 public:
   static LogicalResult
   inferReturnTypes(MLIRContext *context, std::optional<Location> location,
                    ValueRange operands, DictionaryAttr attributes,
-                   RegionRange regions,
+                   OpaqueProperties properties, RegionRange regions,
                    SmallVectorImpl<Type> &inferredReturnTypes) {
     static_assert(
         ConcreteType::template hasTrait<InferShapedTypeOpInterface::Trait>(),
         "requires InferShapedTypeOpInterface to ensure succesful invocation");
     static_assert(
         ConcreteType::template hasTrait<InferTypeOpInterface::Trait>(),
         "requires InferTypeOpInterface to ensure succesful invocation");
     return ::mlir::detail::inferReturnTensorTypes(
         ConcreteType::inferReturnTypeComponents, context, location, operands,
-        attributes, regions, inferredReturnTypes);
+        attributes, properties, regions, inferredReturnTypes);
   }
 };
 
 } // namespace OpTrait
 } // namespace mlir
 
 #endif // MLIR_INTERFACES_INFERTYPEOPINTERFACE_H_

mlir/include/mlir/Interfaces/InferTypeOpInterface.td

StaticInterfaceMethod<
       in an undefined behavior.
       }],
       /*retTy=*/"::mlir::LogicalResult",
       /*methodName=*/"inferReturnTypes",
       /*args=*/(ins "::mlir::MLIRContext *":$context,
                     "::std::optional<::mlir::Location>":$location,
                     "::mlir::ValueRange":$operands,
                     "::mlir::DictionaryAttr":$attributes,
+                    "::mlir::OpaqueProperties":$properties,
                     "::mlir::RegionRange":$regions,
                     "::llvm::SmallVectorImpl<::mlir::Type>&":$inferredReturnTypes)
     >,
     StaticInterfaceMethod<
       /*desc=*/[{Refine the return types that an op would generate.
 
       This method computes the return types as `inferReturnTypes` does but
       additionally takes the existing result types as input. The existing
StaticInterfaceMethod<
       e.g., operands are verified, or it may result in an undefined behavior.
       }],
       /*retTy=*/"::mlir::LogicalResult",
       /*methodName=*/"refineReturnTypes",
       /*args=*/(ins "::mlir::MLIRContext *":$context,
                     "::std::optional<::mlir::Location>":$location,
                     "::mlir::ValueRange":$operands,
                     "::mlir::DictionaryAttr":$attributes,
+                    "::mlir::OpaqueProperties":$properties,
                     "::mlir::RegionRange":$regions,
                     "::llvm::SmallVectorImpl<::mlir::Type>&":$returnTypes),
       /*methodBody=*/[{}],
       /*defaultImplementation=*/[{
           llvm::SmallVector<Type, 4> inferredReturnTypes;
           if (failed(ConcreteOp::inferReturnTypes(context, location, operands,
-                                                  attributes, regions,
+                                                  attributes, properties, regions,
                                                   inferredReturnTypes)))
             return failure();
           if (!ConcreteOp::isCompatibleReturnTypes(inferredReturnTypes,
                                                    returnTypes)) {
             return emitOptionalError(
                 location, "'", ConcreteOp::getOperationName(),
                 "' op inferred type(s) ", inferredReturnTypes,
                 " are incompatible with return type(s) of operation ",
StaticInterfaceMethod<
       population of components is not error condition.
       }],
       /*retTy=*/"::mlir::LogicalResult",
       /*methodName=*/"inferReturnTypeComponents",
       /*args=*/(ins "::mlir::MLIRContext*":$context,
                     "::std::optional<::mlir::Location>":$location,
                     "::mlir::ValueShapeRange":$operands,
                     "::mlir::DictionaryAttr":$attributes,
+                    "::mlir::OpaqueProperties":$properties,
                     "::mlir::RegionRange":$regions,
                     "::llvm::SmallVectorImpl<::mlir::ShapedTypeComponents>&":
                       $inferredReturnShapes),
       /*methodBody=*/[{}],
       /*defaultImplementation=*/[{ return ::mlir::failure(); }]
     >,
     InterfaceMethod<
       /*desc=*/[{Reify the shape computation for the operation.

mlir/include/mlir/TableGen/Argument.h

 // TableGen because we need similar pattern matching mechanisms for them.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef MLIR_TABLEGEN_ARGUMENT_H_
 #define MLIR_TABLEGEN_ARGUMENT_H_
 
 #include "mlir/TableGen/Attribute.h"
+#include "mlir/TableGen/Property.h"
 #include "mlir/TableGen/Type.h"
 #include "llvm/ADT/PointerUnion.h"
 #include <string>
 
 namespace llvm {
 class StringRef;
 } // namespace llvm
 
struct NamedTypeConstraint {
   // Returns true if this is a variable length type constraint. This is either
   // variadic or optional.
   bool isVariableLength() const { return isOptional() || isVariadic(); }
 
   llvm::StringRef name;
   TypeConstraint constraint;
 };
 
-// Operation argument: either attribute or operand
-using Argument = llvm::PointerUnion<NamedAttribute *, NamedTypeConstraint *>;
+// Operation argument: either attribute, property, or operand
+using Argument = llvm::PointerUnion<NamedAttribute *, NamedProperty *,
+                                    NamedTypeConstraint *>;
 
 } // namespace tblgen
 } // namespace mlir
 
 #endif // MLIR_TABLEGEN_ARGUMENT_H_

mlir/include/mlir/TableGen/Class.h

 /// Unstructured extra class declarations and definitions, from TableGen
 /// definitions. The default visibility of extra class declarations is up to the
 /// owning class.
 class ExtraClassDeclaration
     : public ClassDeclarationBase<ClassDeclaration::ExtraClassDeclaration> {
 public:
   /// Create an extra class declaration.
   ExtraClassDeclaration(StringRef extraClassDeclaration,
-                        StringRef extraClassDefinition = "")
+                        std::string extraClassDefinition = "")
+      : ExtraClassDeclaration(extraClassDeclaration.str(),
+                              std::move(extraClassDefinition)) {}
+
+  ExtraClassDeclaration(std::string extraClassDeclaration,
+                        std::string extraClassDefinition = "")
       : extraClassDeclaration(extraClassDeclaration),
         extraClassDefinition(extraClassDefinition) {}
 
   /// Write the extra class declarations.
   void writeDeclTo(raw_indented_ostream &os) const override;
 
   /// Write the extra class definitions.
   void writeDefTo(raw_indented_ostream &os,
                   StringRef namePrefix) const override;
 
 private:
   /// The string of the extra class declarations. It is re-indented before
   /// printed.
-  StringRef extraClassDeclaration;
+  std::string extraClassDeclaration;
   /// The string of the extra class definitions. It is re-indented before
   /// printed.
   std::string extraClassDefinition;
 };
 
 /// A class used to emit C++ classes from Tablegen.  Contains a list of public
 /// methods and a list of private fields to be emitted.
 class Class {

mlir/include/mlir/TableGen/Dialect.h

public:
   /// Returns true if this dialect should generate the default dispatch for
   /// type printing/parsing.
   bool useDefaultTypePrinterParser() const;
 
   /// Returns true if this dialect can be extended at runtime with new
   /// operations or types.
   bool isExtensible() const;
 
+  /// Default to use properties for storing Attributes for operations in this
+  /// dialect.
+  bool usePropertiesForAttributes() const;
+
   // Returns whether two dialects are equal by checking the equality of the
   // underlying record.
   bool operator==(const Dialect &other) const;
 
   bool operator!=(const Dialect &other) const { return !(*this == other); }
 
   // Compares two dialects by comparing the names of the dialects.
   bool operator<(const Dialect &other) const;

mlir/include/mlir/TableGen/Operator.h

 #ifndef MLIR_TABLEGEN_OPERATOR_H_
 #define MLIR_TABLEGEN_OPERATOR_H_
 
 #include "mlir/Support/LLVM.h"
 #include "mlir/TableGen/Argument.h"
 #include "mlir/TableGen/Attribute.h"
 #include "mlir/TableGen/Builder.h"
 #include "mlir/TableGen/Dialect.h"
+#include "mlir/TableGen/Property.h"
 #include "mlir/TableGen/Region.h"
 #include "mlir/TableGen/Successor.h"
 #include "mlir/TableGen/Trait.h"
 #include "mlir/TableGen/Type.h"
 #include "llvm/ADT/PointerUnion.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringMap.h"
 #include "llvm/ADT/StringRef.h"
public:
   StringRef getResultName(int index) const;
   /// Returns the `index`-th result's decorators.
   var_decorator_range getResultDecorators(int index) const;
 
   /// Returns the number of variable length results in this operation.
   unsigned getNumVariableLengthResults() const;
 
   /// Op attribute iterators.
-  using attribute_iterator = const NamedAttribute *;
-  attribute_iterator attribute_begin() const;
-  attribute_iterator attribute_end() const;
-  llvm::iterator_range<attribute_iterator> getAttributes() const;
+  using const_attribute_iterator = const NamedAttribute *;
+  const_attribute_iterator attribute_begin() const;
+  const_attribute_iterator attribute_end() const;
+  llvm::iterator_range<const_attribute_iterator> getAttributes() const;
+  using attribute_iterator = NamedAttribute *;
+  attribute_iterator attribute_begin();
+  attribute_iterator attribute_end();
+  llvm::iterator_range<attribute_iterator> getAttributes();
 
   int getNumAttributes() const { return attributes.size(); }
   int getNumNativeAttributes() const { return numNativeAttributes; }
 
   /// Op attribute accessors.
   NamedAttribute &getAttribute(int index) { return attributes[index]; }
   const NamedAttribute &getAttribute(int index) const {
     return attributes[index];
   }
 
   /// Op operand iterators.
   const_value_iterator operand_begin() const;
   const_value_iterator operand_end() const;
   const_value_range getOperands() const;
 
+  // Op properties iterators.
+  using const_property_iterator = const NamedProperty *;
+  const_property_iterator properties_begin() const {
+    return properties.begin();
+  }
+  const_property_iterator properties_end() const { return properties.end(); }
+  llvm::iterator_range<const_property_iterator> getProperties() const {
+    return properties;
+  }
+  using property_iterator = NamedProperty *;
+  property_iterator properties_begin() { return properties.begin(); }
+  property_iterator properties_end() { return properties.end(); }
+  llvm::iterator_range<property_iterator> getProperties() { return properties; }
+  int getNumCoreAttributes() const { return properties.size(); }
+
+  // Op properties accessors.
+  NamedProperty &getProperty(int index) { return properties[index]; }
+  const NamedProperty &getProperty(int index) const {
+    return properties[index];
+  }
+
   int getNumOperands() const { return operands.size(); }
   NamedTypeConstraint &getOperand(int index) { return operands[index]; }
   const NamedTypeConstraint &getOperand(int index) const {
     return operands[index];
   }
 
   /// Returns the number of variadic operands in this operation.
   unsigned getNumVariableLengthOperands() const;
public:
   /// Returns the setter name for the accessor of `name`.
   std::string getSetterName(StringRef name) const;
 
   /// Returns the remove name for the accessor of `name`.
   std::string getRemoverName(StringRef name) const;
 
   bool hasFolder() const;
 
 private:
   /// Populates the vectors containing operands, attributes, results and traits.
   void populateOpStructure();
 
   /// Populates type inference info (mostly equality) with input a mapping from
   /// names to indices for arguments and results.
   void populateTypeInferenceInfo(
       const llvm::StringMap<int> &argumentsAndResultsIndex);
 
private:
   SmallVector<NamedTypeConstraint, 4> operands;
 
   /// The attributes of the op.  Contains native attributes (corresponding to
   /// the actual stored attributed of the operation) followed by derived
   /// attributes (corresponding to dynamic properties of the operation that are
   /// computed upon request).
   SmallVector<NamedAttribute, 4> attributes;
 
+  /// The properties of the op.
+  SmallVector<NamedProperty> properties;
+
   /// The arguments of the op (operands and native attributes).
   SmallVector<Argument, 4> arguments;
 
   /// The results of the op.
   SmallVector<NamedTypeConstraint, 4> results;
 
   /// The successors of this op.
   SmallVector<NamedSuccessor, 0> successors;

mlir/include/mlir/TableGen/Property.h

+//===- Property.h - Property wrapper class --------------------*- 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Property wrapper to simplify using TableGen Record defining a MLIR
+// Property.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_TABLEGEN_PROPERTY_H_
+#define MLIR_TABLEGEN_PROPERTY_H_
+
+#include "mlir/Support/LLVM.h"
+#include "mlir/TableGen/Constraint.h"
+#include "llvm/ADT/StringRef.h"
+
+namespace llvm {
+class DefInit;
+class Record;
+} // namespace llvm
+
+namespace mlir {
+namespace tblgen {
+class Dialect;
+class Type;
+
+// Wrapper class providing helper methods for accessing MLIR Property defined
+// in TableGen. This class should closely reflect what is defined as class
+// `Property` in TableGen.
+class Property {
+public:
+  explicit Property(const llvm::Record *record);
+  explicit Property(const llvm::DefInit *init);
+
+  // Returns the storage type.
+  StringRef getStorageType() const;
+
+  // Returns the interface type for this property.
+  StringRef getInterfaceType() const;
+
+  // Returns the template getter method call which reads this property's
+  // storage and returns the value as of the desired return type.
+  StringRef getConvertFromStorageCall() const;
+
+  // Returns the template setter method call which reads this property's
+  // in the provided interface type and assign it to the storage.
+  StringRef getAssignToStorageCall() const;
+
+  // Returns the conversion method call which reads this property's
+  // in the storage type and builds an attribute.
+  StringRef getConvertToAttributeCall() const;
+
+  // Returns the setter method call which reads this property's
+  // in the provided interface type and assign it to the storage.
+  StringRef getConvertFromAttributeCall() const;
+
+  // Returns the code to compute the hash for this property.
+  StringRef getHashPropertyCall() const;
+
+  // Returns whether this Property has a default value.
+  bool hasDefaultValue() const;
+  // Returns the default value for this Property.
+  StringRef getDefaultValue() const;
+
+  // Returns the TableGen definition this Property was constructed from.
+  const llvm::Record &getDef() const;
+
+private:
+  // The TableGen definition of this constraint.
+  const llvm::Record *def;
+};
+
+// A struct wrapping an op property and its name together
+struct NamedProperty {
+  llvm::StringRef name;
+  Property prop;
+};
+
+} // namespace tblgen
+} // namespace mlir
+
+#endif // MLIR_TABLEGEN_PROPERTY_H_

mlir/include/mlir/Transforms/DialectConversion.h

public:
   void rewrite(Operation *op, ArrayRef<Value> operands,
                ConversionPatternRewriter &rewriter) const final {
     rewrite(cast<SourceOp>(op), OpAdaptor(operands, op->getAttrDictionary()),
             rewriter);
   }
   LogicalResult
   matchAndRewrite(Operation *op, ArrayRef<Value> operands,
                   ConversionPatternRewriter &rewriter) const final {
-    return matchAndRewrite(cast<SourceOp>(op),
-                           OpAdaptor(operands, op->getAttrDictionary()),
+    auto sourceOp = cast<SourceOp>(op);
+    if constexpr (SourceOp::hasProperties())
+      return matchAndRewrite(sourceOp,
+                             OpAdaptor(operands, op->getAttrDictionary(),
+                                       sourceOp.getProperties()),
                              rewriter);
+    return matchAndRewrite(
+        sourceOp, OpAdaptor(operands, op->getAttrDictionary()), rewriter);
   }
 
   /// Rewrite and Match methods that operate on the SourceOp type. These must be
   /// overridden by the derived pattern class.
   virtual LogicalResult match(SourceOp op) const {
     llvm_unreachable("must override match or matchAndRewrite");
   }
   virtual void rewrite(SourceOp op, OpAdaptor adaptor,

mlir/include/mlir/Transforms/OneToNTypeConversion.h

public:
   /// namely the N values each original operand gets converted to. Concretely,
   /// this makes the result type of the accessor functions of the adaptor class
   /// be a `ValueRange`.
   class OpAdaptor
       : public SourceOp::template GenericAdaptor<ArrayRef<ValueRange>> {
   public:
     using RangeT = ArrayRef<ValueRange>;
     using BaseT = typename SourceOp::template GenericAdaptor<RangeT>;
+    using Properties = typename SourceOp::template InferredProperties<SourceOp>;
 
     OpAdaptor(const OneToNTypeMapping *operandMapping,
               const OneToNTypeMapping *resultMapping,
               const ValueRange *convertedOperands, RangeT values,
-              DictionaryAttr attrs = nullptr, RegionRange regions = {})
-        : BaseT(values, attrs, regions), operandMapping(operandMapping),
-          resultMapping(resultMapping), convertedOperands(convertedOperands) {}
+              DictionaryAttr attrs = nullptr, Properties &properties = {},
+              RegionRange regions = {})
+        : BaseT(values, attrs, properties, regions),
+          operandMapping(operandMapping), resultMapping(resultMapping),
+          convertedOperands(convertedOperands) {}
 
     /// Get the type mapping of the original operands to the converted operands.
     const OneToNTypeMapping &getOperandMapping() const {
       return *operandMapping;
     }
 
     /// Get the type mapping of the original results to the converted results.
     const OneToNTypeMapping &getResultMapping() const { return *resultMapping; }
LogicalResult matchAndRewrite(Operation *op, OneToNPatternRewriter &rewriter,
                                 ValueRange convertedOperands) const final {
     // Wrap converted operands and type mappings into an adaptor.
     SmallVector<ValueRange> valueRanges;
     for (int64_t i = 0; i < op->getNumOperands(); i++) {
       auto values = operandMapping.getConvertedValues(convertedOperands, i);
       valueRanges.push_back(values);
     }
     OpAdaptor adaptor(&operandMapping, &resultMapping, &convertedOperands,
-                      valueRanges, op->getAttrDictionary(), op->getRegions());
+                      valueRanges, op->getAttrDictionary(),
+                      cast<SourceOp>(op).getProperties(), op->getRegions());
 
     // Call overload implemented by the derived class.
     return matchAndRewrite(cast<SourceOp>(op), adaptor, rewriter);
   }
 };
 
 /// Applies the given set of patterns recursively on the given op and adds user
 /// materializations where necessary. The patterns are expected to be

mlir/lib/AsmParser/Parser.cpp

public:
   ParseResult parseGenericOperationAfterOpName(
       OperationState &result,
       std::optional<ArrayRef<UnresolvedOperand>> parsedOperandUseInfo =
           std::nullopt,
       std::optional<ArrayRef<Block *>> parsedSuccessors = std::nullopt,
       std::optional<MutableArrayRef<std::unique_ptr<Region>>> parsedRegions =
           std::nullopt,
       std::optional<ArrayRef<NamedAttribute>> parsedAttributes = std::nullopt,
+      std::optional<Attribute> propertiesAttribute = std::nullopt,
       std::optional<FunctionType> parsedFnType = std::nullopt);
 
   /// Parse an operation instance that is in the generic form and insert it at
   /// the provided insertion point.
   Operation *parseGenericOperation(Block *insertBlock,
                                    Block::iterator insertPt);
 
   /// This type is used to keep track of things that are either an Operation or
Value OperationParser::createForwardRefPlaceholder(SMLoc loc, Type type) {
   // something with a def/use chain.
   //
   // We create these placeholders as having an empty name, which we know
   // cannot be created through normal user input, allowing us to distinguish
   // them.
   auto name = OperationName("builtin.unrealized_conversion_cast", getContext());
   auto *op = Operation::create(
       getEncodedSourceLocation(loc), name, type, /*operands=*/{},
-      /*attributes=*/std::nullopt, /*successors=*/{}, /*numRegions=*/0);
+      /*attributes=*/std::nullopt, /*properties=*/nullptr, /*successors=*/{},
+      /*numRegions=*/0);
   forwardRefPlaceholders[op->getResult(0)] = loc;
   return op->getResult(0);
 }
 
 //===----------------------------------------------------------------------===//
 // Operation Parsing
 //===----------------------------------------------------------------------===//
 

 } // namespace
 
 ParseResult OperationParser::parseGenericOperationAfterOpName(
     OperationState &result,
     std::optional<ArrayRef<UnresolvedOperand>> parsedOperandUseInfo,
     std::optional<ArrayRef<Block *>> parsedSuccessors,
     std::optional<MutableArrayRef<std::unique_ptr<Region>>> parsedRegions,
     std::optional<ArrayRef<NamedAttribute>> parsedAttributes,
+    std::optional<Attribute> propertiesAttribute,
     std::optional<FunctionType> parsedFnType) {
 
   // Parse the operand list, if not explicitly provided.
   SmallVector<UnresolvedOperand, 8> opInfo;
   if (!parsedOperandUseInfo) {
     if (parseToken(Token::l_paren, "expected '(' to start operand list") ||
         parseOptionalSSAUseList(opInfo) ||
         parseToken(Token::r_paren, "expected ')' to end operand list")) {
if (getToken().is(Token::l_square)) {
       if (parseSuccessors(successors))
         return failure();
       result.addSuccessors(successors);
     }
   } else {
     result.addSuccessors(*parsedSuccessors);
   }
 
+  // Parse the properties, if not explicitly provided.
+  if (propertiesAttribute) {
+    result.propertiesAttr = *propertiesAttribute;
+  } else if (consumeIf(Token::less)) {
+    result.propertiesAttr = parseAttribute();
+    if (!result.propertiesAttr)
+      return failure();
+    if (parseToken(Token::greater, "expected '>' to close properties"))
+      return failure();
+  }
   // Parse the region list, if not explicitly provided.
   if (!parsedRegions) {
     if (consumeIf(Token::l_paren)) {
       do {
         // Create temporary regions with the top level region as parent.
         result.regions.emplace_back(new Region(topLevelOp));
         if (parseRegion(*result.regions.back(), /*entryArguments=*/{}))
           return failure();
Operation *OperationParser::parseGenericOperation() {
 
   // If we are populating the parser state, start a new operation definition.
   if (state.asmState)
     state.asmState->startOperationDefinition(result.name);
 
   if (parseGenericOperationAfterOpName(result))
     return nullptr;
 
+  // Operation::create() is not allowed to fail, however setting the properties
+  // from an attribute is a failable operation. So we save the attribute here
+  // and set it on the operation post-parsing.
+  Attribute properties;
+  std::swap(properties, result.propertiesAttr);
+
+  // If we don't have properties in the textual IR, but the operation now has
+  // support for properties, we support some backward-compatible generic syntax
+  // for the operation and as such we accept inherent attributes mixed in the
+  // dictionary of discardable attributes. We pre-validate these here because
+  // invalid attributes can't be casted to the properties storage and will be
+  // silently dropped. For example an attribute { foo = 0 : i32 } that is
+  // declared as F32Attr in ODS would have a C++ type of FloatAttr in the
+  // properties array. When setting it we would do something like:
+  //
+  //   properties.foo = dyn_cast<FloatAttr>(fooAttr);
+  //
+  // which would end up with a null Attribute. The diagnostic from the verifier
+  // would be "missing foo attribute" instead of something like "expects a 32
+  // bits float attribute but got a 32 bits integer attribute".
+  if (!properties && !result.getRawProperties()) {
+    Optional<RegisteredOperationName> info = result.name.getRegisteredInfo();
+    if (info) {
+      if (failed(info->verifyInherentAttrs(result.attributes, [&]() {
+            return mlir::emitError(srcLocation) << "'" << name << "' op ";
+          })))
+        return nullptr;
+    }
+  }
+
   // Create the operation and try to parse a location for it.
   Operation *op = opBuilder.create(result);
   if (parseTrailingLocationSpecifier(op))
     return nullptr;
+
+  // Try setting the properties for the operation, using a diagnostic to print
+  // errors.
+  if (properties) {
+    InFlightDiagnostic diagnostic =
+        mlir::emitError(srcLocation, "invalid properties ")
+        << properties << " for op " << name << ": ";
+    if (failed(op->setPropertiesFromAttribute(properties, &diagnostic)))
+      return nullptr;
+    diagnostic.abandon();
+  }
+
   return op;
 }
 
 Operation *OperationParser::parseGenericOperation(Block *insertBlock,
                                                   Block::iterator insertPt) {
   Token nameToken = getToken();
 
   OpBuilder::InsertionGuard restoreInsertionPoint(opBuilder);
public:
   }
 
   ParseResult parseGenericOperationAfterOpName(
       OperationState &result,
       std::optional<ArrayRef<UnresolvedOperand>> parsedUnresolvedOperands,
       std::optional<ArrayRef<Block *>> parsedSuccessors,
       std::optional<MutableArrayRef<std::unique_ptr<Region>>> parsedRegions,
       std::optional<ArrayRef<NamedAttribute>> parsedAttributes,
+      std::optional<Attribute> parsedPropertiesAttribute,
       std::optional<FunctionType> parsedFnType) final {
     return parser.parseGenericOperationAfterOpName(
         result, parsedUnresolvedOperands, parsedSuccessors, parsedRegions,
-        parsedAttributes, parsedFnType);
+        parsedAttributes, parsedPropertiesAttribute, parsedFnType);
   }
   //===--------------------------------------------------------------------===//
   // Utilities
   //===--------------------------------------------------------------------===//
 
   /// Return the name of the specified result in the specified syntax, as well
   /// as the subelement in the name.  For example, in this operation:
   ///
CustomOpAsmParser opAsmParser(opLoc, resultIDs, parseAssemblyFn,
                                 isIsolatedFromAbove, opName, *this);
   if (opAsmParser.parseOperation(opState))
     return nullptr;
 
   // If it emitted an error, we failed.
   if (opAsmParser.didEmitError())
     return nullptr;
 
+  Attribute properties = opState.propertiesAttr;
+  opState.propertiesAttr = Attribute{};
+
   // Otherwise, create the operation and try to parse a location for it.
   Operation *op = opBuilder.create(opState);
   if (parseTrailingLocationSpecifier(op))
     return nullptr;
+
+  // Try setting the properties for the operation.
+  if (properties) {
+    InFlightDiagnostic diagnostic =
+        mlir::emitError(srcLocation, "invalid properties ")
+        << properties << " for op " << op->getName().getStringRef() << ": ";
+    if (failed(op->setPropertiesFromAttribute(properties, &diagnostic)))
+      return nullptr;
+    diagnostic.abandon();
+  }
   return op;
 }
 
 ParseResult OperationParser::parseLocationAlias(LocationAttr &loc) {
   Token tok = getToken();
   consumeToken(Token::hash_identifier);
   StringRef identifier = tok.getSpelling().drop_front();
   if (identifier.contains('.')) {

mlir/lib/CAPI/IR/IR.cpp

emitError(state.location)
         << "type inference was requested for the operation " << state.name
         << ", but the operation does not support type inference. Result "
            "types must be specified explicitly.";
     return failure();
   }
 
   if (succeeded(inferInterface->inferReturnTypes(
           context, state.location, state.operands,
-          state.attributes.getDictionary(context), state.regions, state.types)))
+          state.attributes.getDictionary(context), state.getRawProperties(),
+          state.regions, state.types)))
     return success();
 
   // Diagnostic emitted by interface.
   return failure();
 }
 
 MlirOperation mlirOperationCreate(MlirOperationState *state) {
   assert(state);

mlir/lib/Dialect/GPU/Transforms/AsyncRegionRewriter.cpp

LogicalResult rewriteAsyncOp(gpu::AsyncOpInterface asyncOp) {
 
     // Clone the op to return a token in addition to the other results.
     SmallVector<Type, 1> resultTypes;
     resultTypes.reserve(1 + op->getNumResults());
     copy(op->getResultTypes(), std::back_inserter(resultTypes));
     resultTypes.push_back(tokenType);
     auto *newOp = Operation::create(op->getLoc(), op->getName(), resultTypes,
                                     op->getOperands(), op->getAttrDictionary(),
+                                    op->getPropertiesStorage(),
                                     op->getSuccessors(), op->getNumRegions());
 
     // Clone regions into new op.
     IRMapping mapping;
     for (auto pair : llvm::zip_first(op->getRegions(), newOp->getRegions()))
       std::get<0>(pair).cloneInto(&std::get<1>(pair), mapping);
 
     // Replace the op with the async clone.

mlir/lib/Dialect/MemRef/IR/MemRefOps.cpp

 #include "mlir/Dialect/Arith/Utils/Utils.h"
 #include "mlir/Dialect/MemRef/IR/MemRef.h"
 #include "mlir/Dialect/MemRef/Utils/MemRefUtils.h"
 #include "mlir/Dialect/Utils/StaticValueUtils.h"
 #include "mlir/IR/AffineMap.h"
 #include "mlir/IR/Builders.h"
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Matchers.h"
+#include "mlir/IR/OpDefinition.h"
 #include "mlir/IR/PatternMatch.h"
 #include "mlir/IR/TypeUtilities.h"
 #include "mlir/Interfaces/InferTypeOpInterface.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 #include "mlir/Interfaces/ViewLikeInterface.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallBitVector.h"
 

 //===----------------------------------------------------------------------===//
 // ExtractStridedMetadataOp
 //===----------------------------------------------------------------------===//
 
 /// The number and type of the results are inferred from the
 /// shape of the source.
 LogicalResult ExtractStridedMetadataOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
-  ExtractStridedMetadataOpAdaptor extractAdaptor(operands, attributes, regions);
+  ExtractStridedMetadataOpAdaptor extractAdaptor(
+      operands, attributes, *properties.as<EmptyProperties *>(), regions);
   auto sourceType = extractAdaptor.getSource().getType().dyn_cast<MemRefType>();
   if (!sourceType)
     return failure();
 
   unsigned sourceRank = sourceType.getRank();
   IndexType indexType = IndexType::get(context);
   auto memrefType =
       MemRefType::get({}, sourceType.getElementType(),

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

matchAndRewrite(memref::ExtractStridedMetadataOp extractStridedMetadataOp,
     if (!reinterpretCastOp)
       return failure();
 
     Location loc = extractStridedMetadataOp.getLoc();
     // Check if the source is suitable for extract_strided_metadata.
     SmallVector<Type> inferredReturnTypes;
     if (failed(extractStridedMetadataOp.inferReturnTypes(
             rewriter.getContext(), loc, {reinterpretCastOp.getSource()},
-            /*attributes=*/{}, /*regions=*/{}, inferredReturnTypes)))
+            /*attributes=*/{}, /*properties=*/nullptr, /*regions=*/{},
+            inferredReturnTypes)))
       return rewriter.notifyMatchFailure(
           reinterpretCastOp, "reinterpret_cast source's type is incompatible");
 
     auto memrefType =
         reinterpretCastOp.getResult().getType().cast<MemRefType>();
     unsigned rank = memrefType.getRank();
     SmallVector<OpFoldResult> results;
     results.resize_for_overwrite(rank * 2 + 2);

mlir/lib/Dialect/SCF/IR/SCF.cpp

bool mlir::scf::insideMutuallyExclusiveBranches(Operation *a, Operation *b) {
 
   // Could not find a common IfOp among a's and b's ancestors.
   return false;
 }
 
 LogicalResult
 IfOp::inferReturnTypes(MLIRContext *ctx, std::optional<Location> loc,
                        ValueRange operands, DictionaryAttr attrs,
-                       RegionRange regions,
+                       OpaqueProperties properties, RegionRange regions,
                        SmallVectorImpl<Type> &inferredReturnTypes) {
   if (regions.empty())
     return failure();
   Region *r = regions.front();
   if (r->empty())
     return failure();
   Block &b = r->front();
   if (b.empty())
if (elseBuilder) {
     elseBuilder(builder, result.location);
   }
 
   // Infer result types.
   SmallVector<Type> inferredReturnTypes;
   MLIRContext *ctx = builder.getContext();
   auto attrDict = DictionaryAttr::get(ctx, result.attributes);
   if (succeeded(inferReturnTypes(ctx, std::nullopt, result.operands, attrDict,
-                                 result.regions, inferredReturnTypes))) {
+                                 /*properties=*/nullptr, result.regions,
+                                 inferredReturnTypes))) {
     result.addTypes(inferredReturnTypes);
   }
 }
 
 LogicalResult IfOp::verify() {
   if (getNumResults() != 0 && getElseRegion().empty())
     return emitOpError("must have an else block if defining values");
   return success();

mlir/lib/Dialect/Shape/IR/Shape.cpp

 }
 
 //===----------------------------------------------------------------------===//
 // AddOp
 //===----------------------------------------------------------------------===//
 
 LogicalResult mlir::shape::AddOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType().isa<SizeType>() ||
       operands[1].getType().isa<SizeType>())
     inferredReturnTypes.assign({SizeType::get(context)});
   else
     inferredReturnTypes.assign({IndexType::get(context)});
   return success();
 }

 
 void ConstShapeOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
                                                MLIRContext *context) {
   patterns.add<TensorCastConstShape>(context);
 }
 
 LogicalResult mlir::shape::ConstShapeOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   Builder b(context);
   auto shape = attributes.getAs<DenseIntElementsAttr>("shape");
   if (!shape)
     return emitOptionalError(location, "missing shape attribute");
   inferredReturnTypes.assign({RankedTensorType::get(
       {static_cast<int64_t>(shape.size())}, b.getIndexType())});
   return success();
OpFoldResult DimOp::fold(FoldAdaptor adaptor) {
   auto extent = valShapedType.getDimSize(*index);
   if (ShapedType::isDynamic(extent))
     return nullptr;
   return IntegerAttr::get(IndexType::get(getContext()), extent);
 }
 
 LogicalResult mlir::shape::DimOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   DimOpAdaptor dimOp(operands);
   inferredReturnTypes.assign({dimOp.getIndex().getType()});
   return success();
 }
 
 bool mlir::shape::DimOp::isCompatibleReturnTypes(TypeRange l, TypeRange r) {
   return eachHasOnlyOneOfTypes<SizeType, IndexType>(l, r);
OpFoldResult DivOp::fold(FoldAdaptor adaptor) {
   }
 
   Type indexTy = IndexType::get(getContext());
   return IntegerAttr::get(indexTy, quotient);
 }
 
 LogicalResult mlir::shape::DivOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType().isa<SizeType>() ||
       operands[1].getType().isa<SizeType>())
     inferredReturnTypes.assign({SizeType::get(context)});
   else
     inferredReturnTypes.assign({IndexType::get(context)});
   return success();
 }
if (shape.getType().isa<ShapeType>()) {
     Value dim =
         builder.create<arith::ConstantOp>(loc, builder.getIndexType(), dimAttr);
     build(builder, result, builder.getIndexType(), shape, dim);
   }
 }
 
 LogicalResult mlir::shape::GetExtentOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   inferredReturnTypes.assign({IndexType::get(context)});
   return success();
 }
 
 bool mlir::shape::GetExtentOp::isCompatibleReturnTypes(TypeRange l,
                                                        TypeRange r) {
   // SizeType is compatible with IndexType.

 }
 
 //===----------------------------------------------------------------------===//
 // MeetOp
 //===----------------------------------------------------------------------===//
 
 LogicalResult mlir::shape::MeetOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands.empty())
     return failure();
 
   auto isShapeType = [](Type arg) {
     if (arg.isa<ShapeType>())
       return true;
     return isExtentTensorType(arg);

 
 void shape::RankOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
                                                 MLIRContext *context) {
   patterns.add<RankShapeOfCanonicalizationPattern>(context);
 }
 
 LogicalResult mlir::shape::RankOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType().isa<ShapeType>())
     inferredReturnTypes.assign({SizeType::get(context)});
   else
     inferredReturnTypes.assign({IndexType::get(context)});
   return success();
 }
 
OpFoldResult NumElementsOp::fold(FoldAdaptor adaptor) {
   for (auto value : shape.cast<DenseIntElementsAttr>())
     product *= value;
   Builder builder(getContext());
   return builder.getIndexAttr(product.getLimitedValue());
 }
 
 LogicalResult mlir::shape::NumElementsOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType().isa<ShapeType>())
     inferredReturnTypes.assign({SizeType::get(context)});
   else
     inferredReturnTypes.assign({IndexType::get(context)});
   return success();
 }
 
OpFoldResult MaxOp::fold(FoldAdaptor adaptor) {
   // If operands are equal, just propagate one.
   if (getLhs() == getRhs())
     return getLhs();
   return nullptr;
 }
 
 LogicalResult mlir::shape::MaxOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType() == operands[1].getType())
     inferredReturnTypes.assign({operands[0].getType()});
   else
     inferredReturnTypes.assign({SizeType::get(context)});
   return success();
 }
 
OpFoldResult MinOp::fold(FoldAdaptor adaptor) {
   // If operands are equal, just propagate one.
   if (getLhs() == getRhs())
     return getLhs();
   return nullptr;
 }
 
 LogicalResult mlir::shape::MinOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType() == operands[1].getType())
     inferredReturnTypes.assign({operands[0].getType()});
   else
     inferredReturnTypes.assign({SizeType::get(context)});
   return success();
 }
 
if (!rhs)
     return nullptr;
   APInt folded = lhs.getValue() * rhs.getValue();
   Type indexTy = IndexType::get(getContext());
   return IntegerAttr::get(indexTy, folded);
 }
 
 LogicalResult mlir::shape::MulOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType().isa<SizeType>() ||
       operands[1].getType().isa<SizeType>())
     inferredReturnTypes.assign({SizeType::get(context)});
   else
     inferredReturnTypes.assign({IndexType::get(context)});
   return success();
 }

 void ShapeOfOp::getCanonicalizationPatterns(RewritePatternSet &patterns,
                                             MLIRContext *context) {
   patterns.add<ShapeOfCastExtentTensor, ShapeOfWithTensor,
                ExtractFromShapeOfExtentTensor>(context);
 }
 
 LogicalResult mlir::shape::ShapeOfOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType().isa<ValueShapeType>())
     inferredReturnTypes.assign({ShapeType::get(context)});
   else {
     auto shapedTy = operands[0].getType().cast<ShapedType>();
     int64_t rank =
         shapedTy.hasRank() ? shapedTy.getRank() : ShapedType::kDynamic;
     Type indexTy = IndexType::get(context);

mlir/lib/Dialect/Tosa/IR/TosaOps.cpp

for (int i = 0, e = operands.size(); i != e; ++i) {
     }
   }
 
   return success();
 }
 
 LogicalResult tosa::ArgMaxOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ShapeAdaptor inputShape = operands.getShape(0);
   IntegerAttr axis = attributes.get("axis").cast<IntegerAttr>();
   int32_t axisVal = axis.getValue().getSExtValue();
 
   if (!inputShape.hasRank()) {
     inferredReturnShapes.push_back(ShapedTypeComponents());
     return success();
   }
 
   SmallVector<int64_t> outShape;
   outShape.reserve(inputShape.getRank() - 1);
   for (int i = 0, s = inputShape.getRank(); i < s; i++) {
     if (i == axisVal)
       continue;
     outShape.push_back(inputShape.getDimSize(i));
   }
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outShape));
   return success();
 }
 
 LogicalResult tosa::RFFT2dOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ShapeAdaptor inputShape = operands.getShape(0);
 
   if (!inputShape.hasRank())
     return failure();
 
   llvm::SmallVector<int64_t> outputShape;
   outputShape.resize(3, ShapedType::kDynamic);
LogicalResult tosa::RFFT2dOp::inferReturnTypeComponents(
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
 
   return success();
 }
 
 LogicalResult tosa::FFT2dOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   inferredReturnShapes.push_back(ShapedTypeComponents(operands.getShape(0)));
   inferredReturnShapes.push_back(ShapedTypeComponents(operands.getShape(1)));
   return success();
 }
 
 LogicalResult tosa::ConcatOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   // Infer all dimension sizes by reducing based on inputs.
   int32_t axis =
       attributes.get("axis").cast<IntegerAttr>().getValue().getSExtValue();
   llvm::SmallVector<int64_t> outputShape;
   bool hasRankedInput = false;
   for (auto operand : operands) {
     ShapeAdaptor operandShape = operands.getShape(operand);
LogicalResult tosa::ConcatOp::inferReturnTypeComponents(
   outputShape[axis] = concatDimSize;
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape, inputType));
   return success();
 }
 
 LogicalResult tosa::EqualOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   llvm::SmallVector<int64_t> outShape;
   if (resolveBroadcastShape(operands, outShape).failed()) {
     inferredReturnShapes.push_back(ShapedTypeComponents());
     return success();
   }
 
   inferredReturnShapes.push_back(
       ShapedTypeComponents(outShape, IntegerType::get(context, /*width=*/1)));
   return success();
 }
 
 bool tosa::EqualOp::isCompatibleReturnTypes(TypeRange l, TypeRange r) {
   if (l.size() != r.size() || l.size() != 1)
     return false;
   return succeeded(verifyCompatibleShape(l[0], r[0]));
 }
 
 LogicalResult tosa::FullyConnectedOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ShapeAdaptor inputShape = operands.getShape(0);
   ShapeAdaptor weightShape = operands.getShape(1);
   ShapeAdaptor biasShape = operands.getShape(2);
 
   // All shapes are dynamic.
   SmallVector<int64_t> outShape;
   outShape.resize(2, ShapedType::kDynamic);
LogicalResult tosa::FullyConnectedOp::inferReturnTypeComponents(
   inferredReturnShapes.push_back(ShapedTypeComponents(outShape));
   return success();
 }
 
 LogicalResult FullyConnectedOp::verify() { return verifyConvOp(*this); }
 
 LogicalResult tosa::MatMulOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ShapeAdaptor lhsShape = operands.getShape(0);
   ShapeAdaptor rhsShape = operands.getShape(1);
 
   // All shapes are dynamic.
   SmallVector<int64_t> outShape;
   outShape.resize(3, ShapedType::kDynamic);
 
LogicalResult tosa::MatMulOp::inferReturnTypeComponents(
   }
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outShape));
   return success();
 }
 
 LogicalResult tosa::PadOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ShapeAdaptor inputShape = operands.getShape(0);
   ShapeAdaptor paddingShape = operands.getShape(1);
   SmallVector<int64_t> outputShape;
 
   // If both inputs have unknown shape, we cannot determine the shape of the
   // output.
   if (!inputShape.hasRank() && !paddingShape.hasRank()) {

 static SmallVector<int64_t> convertToMlirShape(ArrayRef<int64_t> shape) {
   return to_vector(llvm::map_range(shape, [](int64_t dim) {
     return dim == -1 ? ShapedType::kDynamic : dim;
   }));
 }
 
 LogicalResult tosa::SliceOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   inferredReturnShapes.push_back(ShapedTypeComponents(
       convertToMlirShape(SliceOpAdaptor(operands, attributes).getSize())));
   return success();
 }
 
 LogicalResult tosa::TableOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ShapeAdaptor inputShape = operands.getShape(0);
 
   if (!inputShape.hasRank()) {
     inferredReturnShapes.push_back(ShapedTypeComponents());
     return success();
   }
 
   inferredReturnShapes.resize(1);
   inputShape.getDims(inferredReturnShapes[0]);
   return success();
 }
 
 LogicalResult tosa::TileOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   TileOpAdaptor adaptor(operands, attributes);
   ArrayRef<int64_t> multiples = adaptor.getMultiples();
   ShapeAdaptor inputShape = operands.getShape(0);
   SmallVector<int64_t> outputShape;
   if (!inputShape.hasRank()) {
     outputShape.resize(multiples.size(), ShapedType::kDynamic);
     inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));

 bool tosa::ReshapeOp::isCompatibleReturnTypes(TypeRange l, TypeRange r) {
   if (l.size() != r.size() || l.size() != 1)
     return false;
   return getElementTypeOrSelf(l[0]) == getElementTypeOrSelf(r[0]);
 }
 
 LogicalResult tosa::ReshapeOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ReshapeOpAdaptor adaptor(operands, attributes);
   ShapeAdaptor inputShape = operands.getShape(0);
   Type inputType = getElementTypeOrSelf(operands.getType()[0]);
   llvm::SmallVector<int64_t> newShapeValue =
       convertToMlirShape(adaptor.getNewShape());
 
   // We cannot infer from the total number of elements so we must take the
perms = llvm::to_vector(
       llvm::map_range(permsAttr.getValues<APInt>(),
                       [](const APInt &val) { return val.getSExtValue(); }));
 
   return success();
 }
 
 LogicalResult tosa::TransposeOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ShapeAdaptor inputShape = operands.getShape(0);
   ShapeAdaptor permsShape = operands.getShape(1);
 
   // If input rank and permutation length is unknown, the output rank is
   // unknown.
   if (!inputShape.hasRank() || !permsShape.hasRank() ||
       permsShape.isDynamicDim(0)) {
LogicalResult tosa::TransposeOp::inferReturnTypeComponents(
   }
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
 LogicalResult tosa::GatherOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   llvm::SmallVector<int64_t> outputShape;
   outputShape.resize(3, ShapedType::kDynamic);
 
   ShapeAdaptor valuesShape = operands.getShape(0);
   if (valuesShape.hasRank()) {
     outputShape[0] = valuesShape.getDimSize(0);
     outputShape[2] = valuesShape.getDimSize(2);
   }
 
   ShapeAdaptor indicesShape = operands.getShape(1);
   if (indicesShape.hasRank()) {
     if (outputShape[0] == ShapedType::kDynamic)
       outputShape[0] = indicesShape.getDimSize(0);
     if (outputShape[1] == ShapedType::kDynamic)
       outputShape[1] = indicesShape.getDimSize(1);
   }
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
 LogicalResult tosa::ResizeOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   ResizeOpAdaptor adaptor(operands, attributes);
   llvm::SmallVector<int64_t, 4> outputShape;
   outputShape.resize(4, ShapedType::kDynamic);
 
   ShapeAdaptor inputShape = operands.getShape(adaptor.getInput());
   if (!inputShape.hasRank())
     return failure();
outputShape[2] =
       1;
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
 LogicalResult tosa::ScatterOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   llvm::SmallVector<int64_t> outputShape;
   outputShape.resize(3, ShapedType::kDynamic);
 
   ShapeAdaptor valuesInShape = operands.getShape(0);
   if (valuesInShape.hasRank()) {
     outputShape[0] = valuesInShape.getDimSize(0);
     outputShape[1] = valuesInShape.getDimSize(1);
if (getElementTypeOrSelf(l[0]) != getElementTypeOrSelf(r[0]))              \
       return false;                                                            \
     return succeeded(verifyCompatibleShape(l[0], r[0]));                       \
   }
 
 #define REDUCE_SHAPE_INFER(OP)                                                 \
   LogicalResult OP::inferReturnTypeComponents(                                 \
       MLIRContext *context, ::std::optional<Location> location,                \
       ValueShapeRange operands, DictionaryAttr attributes,                     \
-      RegionRange regions,                                                     \
+      OpaqueProperties properties, RegionRange regions,                        \
       SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {           \
     Type inputType =                                                           \
         operands.getType()[0].cast<TensorType>().getElementType();             \
     return ReduceInferReturnTypes(operands.getShape(0), inputType,             \
                                   attributes.get("axis").cast<IntegerAttr>(),  \
                                   inferredReturnShapes);                       \
   }                                                                            \
   COMPATIBLE_RETURN_TYPES(OP)
static LogicalResult NAryInferReturnTypes(
   }
   return success();
 }
 
 #define NARY_SHAPE_INFER(OP)                                                   \
   LogicalResult OP::inferReturnTypeComponents(                                 \
       MLIRContext *context, ::std::optional<Location> location,                \
       ValueShapeRange operands, DictionaryAttr attributes,                     \
-      RegionRange regions,                                                     \
+      OpaqueProperties properties, RegionRange regions,                        \
       SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {           \
     return NAryInferReturnTypes(operands, inferredReturnShapes);               \
   }
 
 NARY_SHAPE_INFER(tosa::AbsOp)
 NARY_SHAPE_INFER(tosa::AddOp)
 NARY_SHAPE_INFER(tosa::ArithmeticRightShiftOp)
 NARY_SHAPE_INFER(tosa::BitwiseAndOp)
static LogicalResult poolingInferReturnTypes(
   }
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
 LogicalResult Conv2DOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   llvm::SmallVector<int64_t> outputShape(4, ShapedType::kDynamic);
   Conv2DOp::Adaptor adaptor(operands.getValues(), attributes);
 
   int64_t inputWidth = ShapedType::kDynamic;
   int64_t inputHeight = ShapedType::kDynamic;
   int64_t weightWidth = ShapedType::kDynamic;
   int64_t weightHeight = ShapedType::kDynamic;
LogicalResult Conv2DOp::inferReturnTypeComponents(
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
 LogicalResult Conv2DOp::verify() { return verifyConvOp(*this); }
 
 LogicalResult Conv3DOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   llvm::SmallVector<int64_t> outputShape(5, ShapedType::kDynamic);
   Conv3DOp::Adaptor adaptor(operands.getValues(), attributes);
 
   int64_t inputWidth = ShapedType::kDynamic;
   int64_t inputHeight = ShapedType::kDynamic;
   int64_t inputDepth = ShapedType::kDynamic;
 
LogicalResult Conv3DOp::inferReturnTypeComponents(
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
 LogicalResult Conv3DOp::verify() { return verifyConvOp(*this); }
 
 LogicalResult AvgPool2dOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   return poolingInferReturnTypes(operands, attributes, inferredReturnShapes);
 }
 
 LogicalResult MaxPool2dOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   return poolingInferReturnTypes(operands, attributes, inferredReturnShapes);
 }
 
 LogicalResult DepthwiseConv2DOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   llvm::SmallVector<int64_t> outputShape(4, ShapedType::kDynamic);
   DepthwiseConv2DOp::Adaptor adaptor(operands.getValues(), attributes);
 
   int64_t inputWidth = ShapedType::kDynamic;
   int64_t inputHeight = ShapedType::kDynamic;
   int64_t inputChannels = ShapedType::kDynamic;
 
LogicalResult DepthwiseConv2DOp::inferReturnTypeComponents(
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
 LogicalResult DepthwiseConv2DOp::verify() { return verifyConvOp(*this); }
 
 LogicalResult TransposeConv2DOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   TransposeConv2DOp::Adaptor adaptor(operands.getValues(), attributes);
   // outputShape is mutable.
   llvm::SmallVector<int64_t> outputShape =
       convertToMlirShape(adaptor.getOutShape());
 
   int64_t inputWidth = ShapedType::kDynamic;
   int64_t inputHeight = ShapedType::kDynamic;
LogicalResult TransposeConv2DOp::inferReturnTypeComponents(
   }
 
   inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
 LogicalResult IfOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   llvm::SmallVector<tosa::YieldOp> yieldOps;
   for (Region *region : regions) {
     for (auto &block : *region)
       if (auto returnOp = dyn_cast<tosa::YieldOp>(block.getTerminator()))
         yieldOps.push_back(returnOp);
   }
 
for (const ValueKnowledge &result : resultKnowledge) {
     inferredReturnShapes.push_back(result.getShapedTypeComponents());
   }
 
   return success();
 }
 
 LogicalResult WhileOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   llvm::SmallVector<tosa::YieldOp> yieldOps;
   for (auto &block : *regions[1])
     if (auto returnOp = dyn_cast<tosa::YieldOp>(block.getTerminator()))
       yieldOps.push_back(returnOp);
 
   // TOSA's while must have a tosa.yield as its terminator. If not found this
   // tosa.while is invalid.

mlir/lib/Dialect/Tosa/Transforms/TosaDecomposeTransposeConv.cpp

InferShapedTypeOpInterface shapeInterface =
       dyn_cast<InferShapedTypeOpInterface>(op.getOperation());
   if (!shapeInterface)
     return op;
 
   SmallVector<ShapedTypeComponents> returnedShapes;
   if (shapeInterface
           .inferReturnTypeComponents(op.getContext(), op.getLoc(),
                                      op->getOperands(), op->getAttrDictionary(),
+                                     op->getPropertiesStorage(),
                                      op->getRegions(), returnedShapes)
           .failed())
     return op;
 
   // We need to use the element type of the existing result type to generate
   // the new result shaped type. This is because rescale can include a cast to
   // different bit-width types and does not have a TypeAttr to define the
   // target type.

mlir/lib/Dialect/Tosa/Transforms/TosaInferShapes.cpp

for (Operation &op : block) {
           dyn_cast<InferShapedTypeOpInterface>(op);
       if (!shapeInterface)
         continue;
 
       SmallVector<ShapedTypeComponents> returnedShapes;
 
       ValueShapeRange range(op.getOperands(), operandShape);
       if (shapeInterface
-              .inferReturnTypeComponents(op.getContext(), op.getLoc(), range,
-                                         op.getAttrDictionary(),
-                                         op.getRegions(), returnedShapes)
+              .inferReturnTypeComponents(
+                  op.getContext(), op.getLoc(), range, op.getAttrDictionary(),
+                  op.getPropertiesStorage(), op.getRegions(), returnedShapes)
               .succeeded()) {
         for (auto it : llvm::zip(op.getResults(), returnedShapes)) {
           Value result = std::get<0>(it);
           ShapedTypeComponents predictedShape = std::get<1>(it);
 
           // Check whether this use case is replaceable. We define an op as
           // being replaceable if it is used by a ReturnOp or a TosaOp.
           bool replaceable = true;

mlir/lib/Dialect/Vector/IR/VectorOps.cpp

SmallVector<int64_t, 4> positionConstants =
         return pos.getDefiningOp<arith::ConstantIndexOp>().value();
       }));
   build(builder, result, source, positionConstants);
 }
 
 LogicalResult
 ExtractOp::inferReturnTypes(MLIRContext *, std::optional<Location>,
                             ValueRange operands, DictionaryAttr attributes,
-                            RegionRange,
+                            OpaqueProperties properties, RegionRange,
                             SmallVectorImpl<Type> &inferredReturnTypes) {
   ExtractOp::Adaptor op(operands, attributes);
   auto vectorType = op.getVector().getType().cast<VectorType>();
   if (static_cast<int64_t>(op.getPosition().size()) == vectorType.getRank()) {
     inferredReturnTypes.push_back(vectorType.getElementType());
   } else {
     auto n =
         std::min<size_t>(op.getPosition().size(), vectorType.getRank() - 1);
if (!attr || attr.getInt() < 0 || attr.getInt() >= indexSize)
       return emitOpError("mask index #") << (en.index() + 1) << " out of range";
   }
   return success();
 }
 
 LogicalResult
 ShuffleOp::inferReturnTypes(MLIRContext *, std::optional<Location>,
                             ValueRange operands, DictionaryAttr attributes,
-                            RegionRange,
+                            OpaqueProperties properties, RegionRange,
                             SmallVectorImpl<Type> &inferredReturnTypes) {
   ShuffleOp::Adaptor op(operands, attributes);
   auto v1Type = op.getV1().getType().cast<VectorType>();
   auto v1Rank = v1Type.getRank();
   // Construct resulting type: leading dimension matches mask
   // length, all trailing dimensions match the operands.
   SmallVector<int64_t, 4> shape;
   shape.reserve(v1Rank);

mlir/lib/IR/AsmPrinter.cpp

void AsmPrinter::Impl::printOptionalAttrDict(ArrayRef<NamedAttribute> attrs,
   llvm::SmallDenseSet<StringRef> elidedAttrsSet(elidedAttrs.begin(),
                                                 elidedAttrs.end());
   auto filteredAttrs = llvm::make_filter_range(attrs, [&](NamedAttribute attr) {
     return !elidedAttrsSet.contains(attr.getName().strref());
   });
   if (!filteredAttrs.empty())
     printFilteredAttributesFn(filteredAttrs);
 }
-
 void AsmPrinter::Impl::printNamedAttribute(NamedAttribute attr) {
   // Print the name without quotes if possible.
   ::printKeywordOrString(attr.getName().strref(), os);
 
   // Pretty printing elides the attribute value for unit attributes.
   if (attr.getValue().isa<UnitAttr>())
     return;
 
void OperationPrinter::printGenericOp(Operation *op, bool printOpName) {
   // For terminators, print the list of successors and their operands.
   if (op->getNumSuccessors() != 0) {
     os << '[';
     interleaveComma(op->getSuccessors(),
                     [&](Block *successor) { printBlockName(successor); });
     os << ']';
   }
 
+  // Print the properties.
+  if (Attribute prop = op->getPropertiesAsAttribute())
+    os << " <" << prop << '>';
+
   // Print regions.
   if (op->getNumRegions() != 0) {
     os << " (";
     interleaveComma(op->getRegions(), [&](Region &region) {
       printRegion(region, /*printEntryBlockArgs=*/true,
                   /*printBlockTerminators=*/true, /*printEmptyBlock=*/true);
     });
     os << ')';
   }
 
-  auto attrs = op->getAttrs();
+  auto attrs = op->getDiscardableAttrs();
   printOptionalAttrDict(attrs);
 
   // Print the type signature of the operation.
   os << " : ";
   printFunctionalType(op);
 }
 
 void OperationPrinter::printBlockName(Block *block) {
if (!region.empty()) {
     for (auto &b : llvm::drop_begin(region.getBlocks(), 1))
       print(&b);
   }
   os.indent(currentIndent) << "}";
 }
 
 void OperationPrinter::printAffineMapOfSSAIds(AffineMapAttr mapAttr,
                                               ValueRange operands) {
+  if (!mapAttr) {
+    os << "<<NULL AFFINE MAP>>";
+    return;
+  }
   AffineMap map = mapAttr.getValue();
   unsigned numDims = map.getNumDims();
   auto printValueName = [&](unsigned pos, bool isSymbol) {
     unsigned index = isSymbol ? numDims + pos : pos;
     assert(index < operands.size());
     if (isSymbol)
       os << "symbol(";
     printValueID(operands[index]);

mlir/lib/IR/CMakeLists.txt

add_mlir_library(MLIRIR
   DialectResourceBlobManager.cpp
   Dominance.cpp
   ExtensibleDialect.cpp
   FunctionImplementation.cpp
   FunctionInterfaces.cpp
   IntegerSet.cpp
   Location.cpp
   MLIRContext.cpp
+  ODSSupport.cpp
   Operation.cpp
   OperationSupport.cpp
   PatternMatch.cpp
   Region.cpp
   RegionKindInterface.cpp
   SymbolTable.cpp
   TensorEncoding.cpp
   Types.cpp

mlir/lib/IR/MLIRContext.cpp

 #include "AffineMapDetail.h"
 #include "AttributeDetail.h"
 #include "IntegerSetDetail.h"
 #include "TypeDetail.h"
 #include "mlir/IR/Action.h"
 #include "mlir/IR/AffineExpr.h"
 #include "mlir/IR/AffineMap.h"
 #include "mlir/IR/Attributes.h"
+#include "mlir/IR/BuiltinAttributes.h"
 #include "mlir/IR/BuiltinDialect.h"
 #include "mlir/IR/Diagnostics.h"
 #include "mlir/IR/Dialect.h"
 #include "mlir/IR/ExtensibleDialect.h"
 #include "mlir/IR/IntegerSet.h"
 #include "mlir/IR/Location.h"
 #include "mlir/IR/OpImplementation.h"
 #include "mlir/IR/Types.h"

 OperationName::UnregisteredOpModel::verifyInvariants(Operation *) {
   return success();
 }
 LogicalResult
 OperationName::UnregisteredOpModel::verifyRegionInvariants(Operation *) {
   return success();
 }
 
+Optional<Attribute>
+OperationName::UnregisteredOpModel::getInherentAttr(Operation *op,
+                                                    StringRef name) {
+  auto dict = dyn_cast_or_null<DictionaryAttr>(getPropertiesAsAttr(op));
+  if (!dict)
+    return std::nullopt;
+  if (Attribute attr = dict.get(name))
+    return attr;
+  return std::nullopt;
+}
+void OperationName::UnregisteredOpModel::setInherentAttr(Operation *op,
+                                                         StringAttr name,
+                                                         Attribute value) {
+  auto dict = dyn_cast_or_null<DictionaryAttr>(getPropertiesAsAttr(op));
+  assert(dict);
+  NamedAttrList attrs(dict);
+  attrs.set(name, value);
+  *op->getPropertiesStorage().as<Attribute *>() =
+      attrs.getDictionary(op->getContext());
+}
+void OperationName::UnregisteredOpModel::populateInherentAttrs(
+    Operation *op, NamedAttrList &attrs) {}
+LogicalResult OperationName::UnregisteredOpModel::verifyInherentAttrs(
+    OperationName opName, NamedAttrList &attributes,
+    function_ref<InFlightDiagnostic()> getDiag) {
+  return success();
+}
+int OperationName::UnregisteredOpModel::getOpPropertyByteSize() {
+  return sizeof(Attribute);
+}
+void OperationName::UnregisteredOpModel::initProperties(
+    OperationName opName, OpaqueProperties storage, OpaqueProperties init) {
+  new (storage.as<Attribute *>()) Attribute();
+}
+void OperationName::UnregisteredOpModel::deleteProperties(
+    OpaqueProperties prop) {
+  prop.as<Attribute *>()->~Attribute();
+}
+void OperationName::UnregisteredOpModel::populateDefaultProperties(
+    OperationName opName, OpaqueProperties properties) {}
+LogicalResult OperationName::UnregisteredOpModel::setPropertiesFromAttr(
+    Operation *op, Attribute attr, InFlightDiagnostic *diag) {
+  *op->getPropertiesStorage().as<Attribute *>() = attr;
+  return success();
+}
+Attribute
+OperationName::UnregisteredOpModel::getPropertiesAsAttr(Operation *op) {
+  return *op->getPropertiesStorage().as<Attribute *>();
+}
+void OperationName::UnregisteredOpModel::copyProperties(OpaqueProperties lhs,
+                                                        OpaqueProperties rhs) {
+  *lhs.as<Attribute *>() = *rhs.as<Attribute *>();
+}
+llvm::hash_code
+OperationName::UnregisteredOpModel::hashProperties(OpaqueProperties prop) {
+  return llvm::hash_combine(*prop.as<Attribute *>());
+}
+
 //===----------------------------------------------------------------------===//
 // RegisteredOperationName
 //===----------------------------------------------------------------------===//
 
 std::optional<RegisteredOperationName>
 RegisteredOperationName::lookup(StringRef name, MLIRContext *ctx) {
   auto &impl = ctx->getImpl();
   auto it = impl.registeredOperations.find(name);

mlir/lib/IR/ODSSupport.cpp

+//===- ODSSupport.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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains out-of-line implementations of the support types that
+// Operation and related classes build on top of.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/IR/ODSSupport.h"
+#include "mlir/IR/BuiltinAttributes.h"
+#include "mlir/IR/BuiltinTypes.h"
+#include "mlir/IR/Diagnostics.h"
+
+using namespace mlir;
+
+LogicalResult mlir::convertFromAttribute(int64_t &storage,
+                                         ::mlir::Attribute attr,
+                                         ::mlir::InFlightDiagnostic *diag) {
+  auto valueAttr = dyn_cast<IntegerAttr>(attr);
+  if (!valueAttr) {
+    if (diag)
+      *diag << "expected IntegerAttr for key `value`";
+    return failure();
+  }
+  storage = valueAttr.getValue().getSExtValue();
+  return success();
+}
+Attribute mlir::convertToAttribute(MLIRContext *ctx, int64_t storage) {
+  return IntegerAttr::get(IntegerType::get(ctx, 64), storage);
+}
+LogicalResult mlir::convertFromAttribute(MutableArrayRef<int64_t> storage,
+                                         ::mlir::Attribute attr,
+                                         ::mlir::InFlightDiagnostic *diag) {
+  auto valueAttr = dyn_cast<DenseI64ArrayAttr>(attr);
+  if (!valueAttr) {
+    if (diag)
+      *diag << "expected DenseI64ArrayAttr for key `value`";
+    return failure();
+  }
+  if (valueAttr.size() != static_cast<int64_t>(storage.size())) {
+    if (diag)
+      *diag << "Size mismatch in attribute conversion: " << valueAttr.size()
+            << " vs " << storage.size();
+    return failure();
+  }
+  llvm::copy(valueAttr.asArrayRef(), storage.begin());
+  return success();
+}
+Attribute mlir::convertToAttribute(MLIRContext *ctx,
+                                   ArrayRef<int64_t> storage) {
+  return DenseI64ArrayAttr::get(ctx, storage);
+}

mlir/lib/IR/Operation.cpp

 //===- Operation.cpp - Operation support code -----------------------------===//
 //
 // 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/IR/Operation.h"
+#include "mlir/IR/Attributes.h"
+#include "mlir/IR/BuiltinAttributes.h"
 #include "mlir/IR/BuiltinTypes.h"
 #include "mlir/IR/Dialect.h"
 #include "mlir/IR/IRMapping.h"
 #include "mlir/IR/OpImplementation.h"
+#include "mlir/IR/OperationSupport.h"
 #include "mlir/IR/PatternMatch.h"
 #include "mlir/IR/TypeUtilities.h"
 #include "mlir/Interfaces/FoldInterfaces.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringExtras.h"
 #include <numeric>
 
 using namespace mlir;
 
 //===----------------------------------------------------------------------===//
 // Operation
 //===----------------------------------------------------------------------===//
 
 /// Create a new Operation from operation state.
 Operation *Operation::create(const OperationState &state) {
-  return create(state.location, state.name, state.types, state.operands,
-                state.attributes.getDictionary(state.getContext()),
-                state.successors, state.regions);
+  Operation *op =
+      create(state.location, state.name, state.types, state.operands,
+             state.properties, state.successors, state.regions);
+  if (LLVM_UNLIKELY(state.propertiesAttr)) {
+    assert(!state.properties);
+    LogicalResult result =
+        op->setPropertiesFromAttribute(state.propertiesAttr,
+                                       /*diagnostic=*/nullptr);
+    assert(result.succeeded() && "invalid properties in op creation");
+    (void)result;
+  }
+  return op;
 }
 
 /// Create a new Operation with the specific fields.
 Operation *Operation::create(Location location, OperationName name,
                              TypeRange resultTypes, ValueRange operands,
-                             NamedAttrList &&attributes, BlockRange successors,
+                             NamedAttrList &&attributes,
+                             OpaqueProperties properties, BlockRange successors,
                              RegionRange regions) {
   unsigned numRegions = regions.size();
-  Operation *op = create(location, name, resultTypes, operands,
-                         std::move(attributes), successors, numRegions);
+  Operation *op =
+      create(location, name, resultTypes, operands, std::move(attributes),
+             properties, successors, numRegions);
   for (unsigned i = 0; i < numRegions; ++i)
     if (regions[i])
       op->getRegion(i).takeBody(*regions[i]);
   return op;
 }
 
 /// Create a new Operation with the specific fields.
 Operation *Operation::create(Location location, OperationName name,
                              TypeRange resultTypes, ValueRange operands,
-                             NamedAttrList &&attributes, BlockRange successors,
+                             NamedAttrList &&attributes,
+                             OpaqueProperties properties, BlockRange successors,
                              unsigned numRegions) {
   // Populate default attributes.
   name.populateDefaultAttrs(attributes);
 
   return create(location, name, resultTypes, operands,
-                attributes.getDictionary(location.getContext()), successors,
-                numRegions);
+                attributes.getDictionary(location.getContext()), properties,
+                successors, numRegions);
 }
 
 /// Overload of create that takes an existing DictionaryAttr to avoid
 /// unnecessarily uniquing a list of attributes.
 Operation *Operation::create(Location location, OperationName name,
                              TypeRange resultTypes, ValueRange operands,
-                             DictionaryAttr attributes, BlockRange successors,
+                             DictionaryAttr attributes,
+                             OpaqueProperties properties, BlockRange successors,
                              unsigned numRegions) {
   assert(llvm::all_of(resultTypes, [](Type t) { return t; }) &&
          "unexpected null result type");
 
   // We only need to allocate additional memory for a subset of results.
   unsigned numTrailingResults = OpResult::getNumTrailing(resultTypes.size());
   unsigned numInlineResults = OpResult::getNumInline(resultTypes.size());
   unsigned numSuccessors = successors.size();
   unsigned numOperands = operands.size();
   unsigned numResults = resultTypes.size();
+  int opPropertiesAllocSize = name.getOpPropertyByteSize();
 
   // If the operation is known to have no operands, don't allocate an operand
   // storage.
   bool needsOperandStorage =
       operands.empty() ? !name.hasTrait<OpTrait::ZeroOperands>() : true;
 
   // Compute the byte size for the operation and the operand storage. This takes
   // into account the size of the operation, its trailing objects, and its
   // prefixed objects.
   size_t byteSize =
-      totalSizeToAlloc<detail::OperandStorage, BlockOperand, Region, OpOperand>(
-          needsOperandStorage ? 1 : 0, numSuccessors, numRegions, numOperands);
+      totalSizeToAlloc<detail::OperandStorage, detail::OpProperties,
+                       BlockOperand, Region, OpOperand>(
+          needsOperandStorage ? 1 : 0, opPropertiesAllocSize, numSuccessors,
+          numRegions, numOperands);
   size_t prefixByteSize = llvm::alignTo(
       Operation::prefixAllocSize(numTrailingResults, numInlineResults),
       alignof(Operation));
   char *mallocMem = reinterpret_cast<char *>(malloc(byteSize + prefixByteSize));
   void *rawMem = mallocMem + prefixByteSize;
 
   // Create the new Operation.
-  Operation *op =
-      ::new (rawMem) Operation(location, name, numResults, numSuccessors,
-                               numRegions, attributes, needsOperandStorage);
+  Operation *op = ::new (rawMem) Operation(
+      location, name, numResults, numSuccessors, numRegions,
+      opPropertiesAllocSize, attributes, properties, needsOperandStorage);
 
   assert((numSuccessors == 0 || op->mightHaveTrait<OpTrait::IsTerminator>()) &&
          "unexpected successors in a non-terminator operation");
 
   // Initialize the results.
   auto resultTypeIt = resultTypes.begin();
   for (unsigned i = 0; i < numInlineResults; ++i, ++resultTypeIt)
     new (op->getInlineOpResult(i)) detail::InlineOpResult(*resultTypeIt, i);
new (&op->getOperandStorage()) detail::OperandStorage(
         op, op->getTrailingObjects<OpOperand>(), operands);
   }
 
   // Initialize the successors.
   auto blockOperands = op->getBlockOperands();
   for (unsigned i = 0; i != numSuccessors; ++i)
     new (&blockOperands[i]) BlockOperand(op, successors[i]);
 
+  // This must be done after properties are initalized.
+  op->setAttrs(attributes);
+
   return op;
 }
 
 Operation::Operation(Location location, OperationName name, unsigned numResults,
                      unsigned numSuccessors, unsigned numRegions,
-                     DictionaryAttr attributes, bool hasOperandStorage)
+                     int fullPropertiesStorageSize, DictionaryAttr attributes,
+                     OpaqueProperties properties, bool hasOperandStorage)
     : location(location), numResults(numResults), numSuccs(numSuccessors),
-      numRegions(numRegions), hasOperandStorage(hasOperandStorage), name(name),
-      attrs(attributes) {
+      numRegions(numRegions), hasOperandStorage(hasOperandStorage),
+      propertiesStorageSize((fullPropertiesStorageSize + 7) / 8), name(name) {
   assert(attributes && "unexpected null attribute dictionary");
+  assert(fullPropertiesStorageSize <= propertiesCapacity &&
+         "Properties size overflow");
 #ifndef NDEBUG
   if (!getDialect() && !getContext()->allowsUnregisteredDialects())
     llvm::report_fatal_error(
         name.getStringRef() +
         " created with unregistered dialect. If this is intended, please call "
         "allowUnregisteredDialects() on the MLIRContext, or use "
         "-allow-unregistered-dialect with the MLIR tool used.");
 #endif
+  if (fullPropertiesStorageSize)
+    name.initOpProperties(getPropertiesStorage(), properties);
 }
 
 // Operations are deleted through the destroy() member because they are
 // allocated via malloc.
 Operation::~Operation() {
   assert(block == nullptr && "operation destroyed but still in a block");
 #ifndef NDEBUG
   if (!use_empty()) {
#endif
 
   // Explicitly run the destructors for the successors.
   for (auto &successor : getBlockOperands())
     successor.~BlockOperand();
 
   // Explicitly destroy the regions.
   for (auto &region : getRegions())
     region.~Region();
+  if (propertiesStorageSize)
+    name.destroyOpProperties(getPropertiesStorage());
 }
 
 /// Destroy this operation or one of its subclasses.
 void Operation::destroy() {
   // Operations may have additional prefixed allocation, which needs to be
   // accounted for here when computing the address to free.
   char *rawMem = reinterpret_cast<char *>(this) -
                  llvm::alignTo(prefixAllocSize(), alignof(Operation));

 /// handlers that may be listening.
 InFlightDiagnostic Operation::emitRemark(const Twine &message) {
   InFlightDiagnostic diag = mlir::emitRemark(getLoc(), message);
   if (getContext()->shouldPrintOpOnDiagnostic())
     diag.attachNote(getLoc()) << "see current operation: " << *this;
   return diag;
 }
 
+DictionaryAttr Operation::getAttrDictionary() {
+  if (getPropertiesStorageSize()) {
+    NamedAttrList attrsList = attrs;
+    getName().populateInherentAttrs(this, attrsList);
+    return attrsList.getDictionary(getContext());
+  }
+  return attrs;
+}
+
+void Operation::setAttrs(DictionaryAttr newAttrs) {
+  assert(newAttrs && "expected valid attribute dictionary");
+  if (getPropertiesStorageSize()) {
+    attrs = DictionaryAttr::get(getContext(), {});
+    for (const NamedAttribute &attr : newAttrs)
+      setAttr(attr.getName(), attr.getValue());
+    return;
+  }
+  attrs = newAttrs;
+}
+void Operation::setAttrs(ArrayRef<NamedAttribute> newAttrs) {
+  if (getPropertiesStorageSize()) {
+    setAttrs(DictionaryAttr::get(getContext(), {}));
+    for (const NamedAttribute &attr : newAttrs)
+      setAttr(attr.getName(), attr.getValue());
+    return;
+  }
+  attrs = DictionaryAttr::get(getContext(), newAttrs);
+}
+
+std::optional<Attribute> Operation::getInherentAttr(StringRef name) {
+  return getName().getInherentAttr(this, name);
+}
+
+void Operation::setInherentAttr(StringAttr name, Attribute value) {
+  getName().setInherentAttr(this, name, value);
+}
+
+Attribute Operation::getPropertiesAsAttribute() {
+  Optional<RegisteredOperationName> info = getRegisteredInfo();
+  if (LLVM_UNLIKELY(!info))
+    return *getPropertiesStorage().as<Attribute *>();
+  return info->getOpPropertiesAsAttribute(this);
+}
+LogicalResult
+Operation::setPropertiesFromAttribute(Attribute attr,
+                                      InFlightDiagnostic *diagnostic) {
+  Optional<RegisteredOperationName> info = getRegisteredInfo();
+  if (LLVM_UNLIKELY(!info)) {
+    *getPropertiesStorage().as<Attribute *>() = attr;
+    return success();
+  }
+  return info->setOpPropertiesFromAttribute(this, attr, diagnostic);
+}
+
+void Operation::copyProperties(OpaqueProperties rhs) {
+  name.copyOpProperties(getPropertiesStorage(), rhs);
+}
+
+llvm::hash_code Operation::hashProperties() {
+  return name.hashOpProperties(getPropertiesStorage());
+}
+
 //===----------------------------------------------------------------------===//
 // Operation Ordering
 //===----------------------------------------------------------------------===//
 
 constexpr unsigned Operation::kInvalidOrderIdx;
 constexpr unsigned Operation::kOrderStride;
 
 /// Given an operation 'other' that is within the same parent block, return
Operation *Operation::clone(IRMapping &mapper, CloneOptions options) {
 
   // Remap the successors.
   successors.reserve(getNumSuccessors());
   for (Block *successor : getSuccessors())
     successors.push_back(mapper.lookupOrDefault(successor));
 
   // Create the new operation.
   auto *newOp = create(getLoc(), getName(), getResultTypes(), operands, attrs,
-                       successors, getNumRegions());
+                       getPropertiesStorage(), successors, getNumRegions());
   mapper.map(this, newOp);
 
   // Clone the regions.
   if (options.shouldCloneRegions()) {
     for (unsigned i = 0; i != numRegions; ++i)
       getRegion(i).cloneInto(&newOp->getRegion(i), mapper);
   }
 

 void OpState::printOpName(Operation *op, OpAsmPrinter &p,
                           StringRef defaultDialect) {
   StringRef name = op->getName().getStringRef();
   if (name.startswith((defaultDialect + ".").str()) && name.count('.') == 1)
     name = name.drop_front(defaultDialect.size() + 1);
   p.getStream() << name;
 }
 
+/// Parse properties as a Attribute.
+ParseResult OpState::genericParseProperties(OpAsmParser &parser,
+                                            Attribute &result) {
+  if (parser.parseLess() || parser.parseAttribute(result) ||
+      parser.parseGreater())
+    return failure();
+  return success();
+}
+
+/// Print the properties as a Attribute.
+void OpState::genericPrintProperties(OpAsmPrinter &p, Attribute properties) {
+  p << "<" << properties << ">";
+}
+
 /// Emit an error about fatal conditions with this operation, reporting up to
 /// any diagnostic handlers that may be listening.
 InFlightDiagnostic OpState::emitError(const Twine &message) {
   return getOperation()->emitError(message);
 }
 
 /// Emit an error with the op name prefixed, like "'dim' op " which is
 /// convenient for verifiers.

mlir/lib/IR/OperationSupport.cpp

 OperationState::OperationState(Location location, StringRef name,
                                ValueRange operands, TypeRange types,
                                ArrayRef<NamedAttribute> attributes,
                                BlockRange successors,
                                MutableArrayRef<std::unique_ptr<Region>> regions)
     : OperationState(location, OperationName(name, location.getContext()),
                      operands, types, attributes, successors, regions) {}
 
+OperationState::~OperationState() {
+  if (properties)
+    propertiesDeleter(properties);
+}
+
+LogicalResult
+OperationState::setProperties(Operation *op,
+                              InFlightDiagnostic *diagnostic) const {
+  if (LLVM_UNLIKELY(propertiesAttr)) {
+    assert(!properties);
+    return op->setPropertiesFromAttribute(propertiesAttr, diagnostic);
+  }
+  if (properties)
+    propertiesSetter(op->getPropertiesStorage(), properties);
+  return success();
+}
+
 void OperationState::addOperands(ValueRange newOperands) {
   operands.append(newOperands.begin(), newOperands.end());
 }
 
 void OperationState::addSuccessors(BlockRange newSuccessors) {
   successors.append(newSuccessors.begin(), newSuccessors.end());
 }
 

 
 llvm::hash_code OperationEquivalence::computeHash(
     Operation *op, function_ref<llvm::hash_code(Value)> hashOperands,
     function_ref<llvm::hash_code(Value)> hashResults, Flags flags) {
   // Hash operations based upon their:
   //   - Operation Name
   //   - Attributes
   //   - Result Types
-  llvm::hash_code hash = llvm::hash_combine(
-      op->getName(), op->getAttrDictionary(), op->getResultTypes());
+  llvm::hash_code hash =
+      llvm::hash_combine(op->getName(), op->getAttrDictionary(),
+                         op->getResultTypes(), op->hashProperties());
 
   //   - Operands
   ValueRange operands = op->getOperands();
   SmallVector<Value> operandStorage;
   if (op->hasTrait<mlir::OpTrait::IsCommutative>()) {
     operandStorage.append(operands.begin(), operands.end());
     llvm::sort(operandStorage, [](Value a, Value b) -> bool {
       return a.getAsOpaquePointer() < b.getAsOpaquePointer();

mlir/lib/Interfaces/InferTypeOpInterface.cpp

if (index < 0 || static_cast<size_t>(index) >= size())
     return nullptr;
   return getShape(operator[](index));
 }
 
 LogicalResult mlir::detail::inferReturnTensorTypes(
     function_ref<
         LogicalResult(MLIRContext *, std::optional<Location> location,
                       ValueShapeRange operands, DictionaryAttr attributes,
-                      RegionRange regions,
+                      OpaqueProperties properties, RegionRange regions,
                       SmallVectorImpl<ShapedTypeComponents> &retComponents)>
         componentTypeFn,
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   SmallVector<ShapedTypeComponents, 2> retComponents;
-  if (failed(componentTypeFn(context, location, operands, attributes, regions,
-                             retComponents)))
+  if (failed(componentTypeFn(context, location, operands, attributes,
+                             properties, regions, retComponents)))
     return failure();
   for (const auto &shapeAndType : retComponents) {
     Type elementTy = shapeAndType.getElementType();
     assert(elementTy && "element type required to construct tensor");
 
     Attribute attr = shapeAndType.getAttribute();
     if (shapeAndType.hasRank()) {
       inferredReturnTypes.push_back(
           RankedTensorType::get(shapeAndType.getDims(), elementTy, attr));
     } else {
       assert(attr == nullptr && "attribute not supported");
       inferredReturnTypes.push_back(UnrankedTensorType::get(elementTy));
     }
   }
   return success();
 }
 
 LogicalResult mlir::detail::verifyInferredResultTypes(Operation *op) {
   SmallVector<Type, 4> inferredReturnTypes(op->getResultTypes());
   auto retTypeFn = cast<InferTypeOpInterface>(op);
-  return retTypeFn.refineReturnTypes(op->getContext(), op->getLoc(),
-                                     op->getOperands(), op->getAttrDictionary(),
-                                     op->getRegions(), inferredReturnTypes);
+  auto result = retTypeFn.refineReturnTypes(
+      op->getContext(), op->getLoc(), op->getOperands(),
+      op->getAttrDictionary(), op->getPropertiesStorage(), op->getRegions(),
+      inferredReturnTypes);
+  if (failed(result))
+    op->emitOpError() << "failed to infer returned types";
+
+  return result;
 }

mlir/lib/Rewrite/ByteCode.cpp

if (numResults == kInferTypesMarker) {
     InferTypeOpInterface::Concept *inferInterface =
         state.name.getInterface<InferTypeOpInterface>();
     assert(inferInterface &&
            "expected operation to provide InferTypeOpInterface");
 
     // TODO: Handle failure.
     if (failed(inferInterface->inferReturnTypes(
             state.getContext(), state.location, state.operands,
-            state.attributes.getDictionary(state.getContext()), state.regions,
-            state.types)))
+            state.attributes.getDictionary(state.getContext()),
+            state.getRawProperties(), state.regions, state.types)))
       return;
   } else {
     // Otherwise, this is a fixed number of results.
     for (unsigned i = 0; i != numResults; ++i) {
       if (read<PDLValue::Kind>() == PDLValue::Kind::Type) {
         state.types.push_back(read<Type>());
       } else {
         TypeRange *resultTypes = read<TypeRange *>();

mlir/lib/TableGen/CMakeLists.txt

llvm_add_library(MLIRTableGen STATIC
   Dialect.cpp
   Format.cpp
   GenInfo.cpp
   Interfaces.cpp
   Operator.cpp
   Pass.cpp
   Pattern.cpp
   Predicate.cpp
+  Property.cpp
   Region.cpp
   SideEffects.cpp
   Successor.cpp
   Trait.cpp
   Type.cpp
 
   DISABLE_LLVM_LINK_LLVM_DYLIB
 
   ADDITIONAL_HEADER_DIRS
   ${MLIR_MAIN_INCLUDE_DIR}/mlir/TableGen
 )
 
 mlir_check_all_link_libraries(MLIRTableGen)
 
 add_mlir_library_install(MLIRTableGen)

mlir/lib/TableGen/CodeGenHelpers.cpp

 /// Code for an attribute constraint. These may be called from ops only.
 /// Attribute constraints cannot reference anything other than `$_self` and
 /// `$_op`.
 ///
 /// TODO: Unique constraints for adaptors. However, most Adaptor::verify
 /// functions are stripped anyways.
 static const char *const attrConstraintCode = R"(
 static ::mlir::LogicalResult {0}(
-    ::mlir::Operation *op, ::mlir::Attribute attr, ::llvm::StringRef attrName) {
-  if (attr && !({1})) {
-    return op->emitOpError("attribute '") << attrName
+    ::mlir::Attribute attr, ::llvm::StringRef attrName, llvm::function_ref<::mlir::InFlightDiagnostic()> getDiag) {{
+  if (attr && !({1}))
+    return getDiag() << "attribute '" << attrName
         << "' failed to satisfy constraint: {2}";
-  }
   return ::mlir::success();
 }
+static ::mlir::LogicalResult {0}(
+    ::mlir::Operation *op, ::mlir::Attribute attr, ::llvm::StringRef attrName) {{
+  return {0}(attr, attrName, [op]() {{
+    return op->emitOpError();
+  });
+}
 )";
 
 /// Code for a successor constraint.
 static const char *const successorConstraintCode = R"(
 static ::mlir::LogicalResult {0}(
     ::mlir::Operation *op, ::mlir::Block *successor,
     ::llvm::StringRef successorName, unsigned successorIndex) {
   if (!({1})) {

mlir/lib/TableGen/Dialect.cpp

 bool Dialect::useDefaultTypePrinterParser() const {
   return def->getValueAsBit("useDefaultTypePrinterParser");
 }
 
 bool Dialect::isExtensible() const {
   return def->getValueAsBit("isExtensible");
 }
 
+bool Dialect::usePropertiesForAttributes() const {
+  return def->getValueAsBit("usePropertiesForAttributes");
+}
+
 bool Dialect::operator==(const Dialect &other) const {
   return def == other.def;
 }
 
 bool Dialect::operator<(const Dialect &other) const {
   return getName() < other.getName();
 }

mlir/lib/TableGen/Operator.cpp

 //===- Operator.cpp - Operator class --------------------------------------===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // Operator wrapper to simplify using TableGen Record defining a MLIR Op.
 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/TableGen/Operator.h"
+#include "mlir/TableGen/Argument.h"
 #include "mlir/TableGen/Predicate.h"
 #include "mlir/TableGen/Trait.h"
 #include "mlir/TableGen/Type.h"
 #include "llvm/ADT/EquivalenceClasses.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/Sequence.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/StringExtras.h"

 }
 auto Operator::trait_end() const -> const_trait_iterator {
   return traits.end();
 }
 auto Operator::getTraits() const -> llvm::iterator_range<const_trait_iterator> {
   return {trait_begin(), trait_end()};
 }
 
-auto Operator::attribute_begin() const -> attribute_iterator {
+auto Operator::attribute_begin() const -> const_attribute_iterator {
   return attributes.begin();
 }
-auto Operator::attribute_end() const -> attribute_iterator {
+auto Operator::attribute_end() const -> const_attribute_iterator {
   return attributes.end();
 }
 auto Operator::getAttributes() const
-    -> llvm::iterator_range<attribute_iterator> {
+    -> llvm::iterator_range<const_attribute_iterator> {
+  return {attribute_begin(), attribute_end()};
+}
+auto Operator::attribute_begin() -> attribute_iterator {
+  return attributes.begin();
+}
+auto Operator::attribute_end() -> attribute_iterator {
+  return attributes.end();
+}
+auto Operator::getAttributes() -> llvm::iterator_range<attribute_iterator> {
   return {attribute_begin(), attribute_end()};
 }
 
 auto Operator::operand_begin() const -> const_value_iterator {
   return operands.begin();
 }
 auto Operator::operand_end() const -> const_value_iterator {
   return operands.end();
for (const ResultTypeInference &infer : inference)
       resultTypeMapping.push_back(infer.sources.front());
   }
 }
 
 void Operator::populateOpStructure() {
   auto &recordKeeper = def.getRecords();
   auto *typeConstraintClass = recordKeeper.getClass("TypeConstraint");
   auto *attrClass = recordKeeper.getClass("Attr");
+  auto *propertyClass = recordKeeper.getClass("Property");
   auto *derivedAttrClass = recordKeeper.getClass("DerivedAttr");
   auto *opVarClass = recordKeeper.getClass("OpVariable");
   numNativeAttributes = 0;
 
   DagInit *argumentValues = def.getValueAsDag("arguments");
   unsigned numArgs = argumentValues->getNumArgs();
 
   // Mapping from name of to argument or result index. Arguments are indexed
for (unsigned i = 0; i != numArgs; ++i) {
     } else if (argDef->isSubClassOf(attrClass)) {
       if (givenName.empty())
         PrintFatalError(argDef->getLoc(), "attributes must be named");
       if (argDef->isSubClassOf(derivedAttrClass))
         PrintFatalError(argDef->getLoc(),
                         "derived attributes not allowed in argument list");
       attributes.push_back({givenName, Attribute(argDef)});
       ++numNativeAttributes;
+    } else if (argDef->isSubClassOf(propertyClass)) {
+      if (givenName.empty())
+        PrintFatalError(argDef->getLoc(), "properties must be named");
+      properties.push_back({givenName, Property(argDef)});
     } else {
-      PrintFatalError(def.getLoc(), "unexpected def type; only defs deriving "
-                                    "from TypeConstraint or Attr are allowed");
+      PrintFatalError(def.getLoc(),
+                      "unexpected def type; only defs deriving "
+                      "from TypeConstraint or Attr or Property are allowed");
     }
     if (!givenName.empty())
       argumentsAndResultsIndex[givenName] = i;
   }
 
   // Handle derived attributes.
   for (const auto &val : def.getValues()) {
     if (auto *record = dyn_cast<llvm::RecordRecTy>(val.getType())) {
if (auto *record = dyn_cast<llvm::RecordRecTy>(val.getType())) {
            Attribute(cast<DefInit>(val.getValue()))});
     }
   }
 
   // Populate `arguments`. This must happen after we've finalized `operands` and
   // `attributes` because we will put their elements' pointers in `arguments`.
   // SmallVector may perform re-allocation under the hood when adding new
   // elements.
-  int operandIndex = 0, attrIndex = 0;
+  int operandIndex = 0, attrIndex = 0, propIndex = 0;
   for (unsigned i = 0; i != numArgs; ++i) {
     Record *argDef = dyn_cast<DefInit>(argumentValues->getArg(i))->getDef();
     if (argDef->isSubClassOf(opVarClass))
       argDef = argDef->getValueAsDef("constraint");
 
     if (argDef->isSubClassOf(typeConstraintClass)) {
       attrOrOperandMapping.push_back(
           {OperandOrAttribute::Kind::Operand, operandIndex});
       arguments.emplace_back(&operands[operandIndex++]);
-    } else {
-      assert(argDef->isSubClassOf(attrClass));
+    } else if (argDef->isSubClassOf(attrClass)) {
       attrOrOperandMapping.push_back(
           {OperandOrAttribute::Kind::Attribute, attrIndex});
       arguments.emplace_back(&attributes[attrIndex++]);
+    } else {
+      assert(argDef->isSubClassOf(propertyClass));
+      arguments.emplace_back(&properties[propIndex++]);
     }
   }
 
   auto *resultsDag = def.getValueAsDag("results");
   auto *outsOp = dyn_cast<DefInit>(resultsDag->getOperator());
   if (!outsOp || outsOp->getDef()->getName() != "outs") {
     PrintFatalError(def.getLoc(), "'results' must have 'outs' directive");
   }

mlir/lib/TableGen/Property.cpp

+//===- Property.cpp - Property wrapper class ----------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Property wrapper to simplify using TableGen Record defining a MLIR
+// Property.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/TableGen/Property.h"
+#include "mlir/TableGen/Format.h"
+#include "mlir/TableGen/Operator.h"
+#include "llvm/TableGen/Record.h"
+
+using namespace mlir;
+using namespace mlir::tblgen;
+
+using llvm::DefInit;
+using llvm::Init;
+using llvm::Record;
+using llvm::StringInit;
+
+// Returns the initializer's value as string if the given TableGen initializer
+// is a code or string initializer. Returns the empty StringRef otherwise.
+static StringRef getValueAsString(const Init *init) {
+  if (const auto *str = dyn_cast<StringInit>(init))
+    return str->getValue().trim();
+  return {};
+}
+
+Property::Property(const Record *record) : def(record) {
+  assert((record->isSubClassOf("Property") || record->isSubClassOf("Attr")) &&
+         "must be subclass of TableGen 'Property' class");
+}
+
+Property::Property(const DefInit *init) : Property(init->getDef()) {}
+
+StringRef Property::getStorageType() const {
+  const auto *init = def->getValueInit("storageType");
+  auto type = getValueAsString(init);
+  if (type.empty())
+    return "Property";
+  return type;
+}
+
+StringRef Property::getInterfaceType() const {
+  const auto *init = def->getValueInit("interfaceType");
+  return getValueAsString(init);
+}
+
+StringRef Property::getConvertFromStorageCall() const {
+  const auto *init = def->getValueInit("convertFromStorage");
+  return getValueAsString(init);
+}
+
+StringRef Property::getAssignToStorageCall() const {
+  const auto *init = def->getValueInit("assignToStorage");
+  return getValueAsString(init);
+}
+
+StringRef Property::getConvertToAttributeCall() const {
+  const auto *init = def->getValueInit("convertToAttribute");
+  return getValueAsString(init);
+}
+
+StringRef Property::getConvertFromAttributeCall() const {
+  const auto *init = def->getValueInit("convertFromAttribute");
+  return getValueAsString(init);
+}
+
+StringRef Property::getHashPropertyCall() const {
+  return getValueAsString(def->getValueInit("hashProperty"));
+}
+
+bool Property::hasDefaultValue() const { return !getDefaultValue().empty(); }
+
+StringRef Property::getDefaultValue() const {
+  const auto *init = def->getValueInit("defaultValue");
+  return getValueAsString(init);
+}
+
+const llvm::Record &Property::getDef() const { return *def; }

mlir/test/Bytecode/versioning/versioned_attr.mlir

 // This file contains a test case representative of a dialect parsing an
 // attribute with versioned custom encoding.
 
 // Bytecode currently does not support big-endian platforms
 // UNSUPPORTED: target=s390x-{{.*}}
 
 //===--------------------------------------------------------------------===//
 // Test attribute upgrade
 //===--------------------------------------------------------------------===//
 
 // COM: bytecode contains
 // COM: module {
 // COM:   version: 1.12
-// COM:   "test.versionedB"() {attribute = #test.attr_params<24, 42>} : () -> ()
+// COM:   "test.versionedB"() <{attribute = #test.attr_params<24, 42>}> : () -> ()
 // COM: }
 // RUN: mlir-opt %S/versioned-attr-1.12.mlirbc 2>&1 | FileCheck %s --check-prefix=CHECK1
-// CHECK1: "test.versionedB"() {attribute = #test.attr_params<42, 24>} : () -> ()
+// CHECK1: "test.versionedB"() <{attribute = #test.attr_params<42, 24>}> : () -> ()
 
 //===--------------------------------------------------------------------===//
 // Test attribute upgrade
 //===--------------------------------------------------------------------===//
 
 // COM: bytecode contains
 // COM: module {
 // COM:   version: 2.0
-// COM:   "test.versionedB"() {attribute = #test.attr_params<42, 24>} : () -> ()
+// COM:   "test.versionedB"() <{attribute = #test.attr_params<42, 24>}> : () -> ()
 // COM: }
 // RUN: mlir-opt %S/versioned-attr-2.0.mlirbc 2>&1 | FileCheck %s --check-prefix=CHECK2
-// CHECK2: "test.versionedB"() {attribute = #test.attr_params<42, 24>} : () -> ()
+// CHECK2: "test.versionedB"() <{attribute = #test.attr_params<42, 24>}> : () -> ()

mlir/test/Bytecode/versioning/versioned_op.mlir

 // This file contains test cases related to the dialect post-parsing upgrade
 // mechanism.
 
 // Bytecode currently does not support big-endian platforms
 // UNSUPPORTED: target=s390x-{{.*}}
 
 //===--------------------------------------------------------------------===//
 // Test generic
 //===--------------------------------------------------------------------===//
 
 // COM: bytecode contains
 // COM: module {
 // COM:   version: 2.0
-// COM:   "test.versionedA"() {dims = 123 : i64, modifier = false} : () -> ()
+// COM:   "test.versionedA"() <{dims = 123 : i64, modifier = false}> : () -> ()
 // COM: }
 // RUN: mlir-opt %S/versioned-op-2.0.mlirbc 2>&1 | FileCheck %s --check-prefix=CHECK1
-// CHECK1: "test.versionedA"() {dims = 123 : i64, modifier = false} : () -> ()
+// CHECK1: "test.versionedA"() <{dims = 123 : i64, modifier = false}> : () -> ()
 
 //===--------------------------------------------------------------------===//
 // Test upgrade
 //===--------------------------------------------------------------------===//
 
 // COM: bytecode contains
 // COM: module {
 // COM:   version: 1.12
-// COM:   "test.versionedA"() {dimensions = 123 : i64} : () -> ()
+// COM:   "test.versionedA"() <{dimensions = 123 : i64}> : () -> ()
 // COM: }
 // RUN: mlir-opt %S/versioned-op-1.12.mlirbc 2>&1 | FileCheck %s --check-prefix=CHECK2
-// CHECK2: "test.versionedA"() {dims = 123 : i64, modifier = false} : () -> ()
+// CHECK2: "test.versionedA"() <{dims = 123 : i64, modifier = false}> : () -> ()
 
 //===--------------------------------------------------------------------===//
 // Test forbidden downgrade
 //===--------------------------------------------------------------------===//
 
 // COM: bytecode contains
 // COM: module {
 // COM:   version: 2.2
-// COM:   "test.versionedA"() {dims = 123 : i64, modifier = false} : () -> ()
+// COM:   "test.versionedA"() <{dims = 123 : i64, modifier = false}> : () -> ()
 // COM: }
 // RUN: not mlir-opt %S/versioned-op-2.2.mlirbc 2>&1 | FileCheck %s --check-prefix=ERR_NEW_VERSION
 // ERR_NEW_VERSION: current test dialect version is 2.0, can't parse version: 2.2

mlir/test/Conversion/OneToNTypeConversion/one-to-n-type-conversion.mlir

 
 // If we only convert the tuple ops, the original `get_tuple_element` ops will
 // disappear but one target materialization will be inserted from the
 // unconverted function arguments to each of the return values (which have
 // redundancy among themselves).
 //
 // CHECK-TUP-LABEL: func.func @materializations_tuple_args(
 // CHECK-TUP-SAME:                                         %[[ARG0:.*]]: tuple<tuple<>, i1, tuple<tuple<i2>>>) -> (i1, i2) {
-// CHECK-TUP-DAG:     %[[V0:.*]] = "test.get_tuple_element"(%[[ARG0]]) {index = 0 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<>
-// CHECK-TUP-DAG:     %[[V1:.*]] = "test.get_tuple_element"(%[[ARG0]]) {index = 1 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> i1
-// CHECK-TUP-DAG:     %[[V2:.*]] = "test.get_tuple_element"(%[[ARG0]]) {index = 2 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
-// CHECK-TUP-DAG:     %[[V3:.*]] = "test.get_tuple_element"(%[[V2]]) {index = 0 : i32} : (tuple<tuple<i2>>) -> tuple<i2>
-// CHECK-TUP-DAG:     %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 0 : i32} : (tuple<i2>) -> i2
-// CHECK-TUP-DAG:     %[[V5:.*]] = "test.get_tuple_element"(%[[ARG0]]) {index = 0 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<>
-// CHECK-TUP-DAG:     %[[V6:.*]] = "test.get_tuple_element"(%[[ARG0]]) {index = 1 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> i1
-// CHECK-TUP-DAG:     %[[V7:.*]] = "test.get_tuple_element"(%[[ARG0]]) {index = 2 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
-// CHECK-TUP-DAG:     %[[V8:.*]] = "test.get_tuple_element"(%[[V7]]) {index = 0 : i32} : (tuple<tuple<i2>>) -> tuple<i2>
-// CHECK-TUP-DAG:     %[[V9:.*]] = "test.get_tuple_element"(%[[V8]]) {index = 0 : i32} : (tuple<i2>) -> i2
+// CHECK-TUP-DAG:     %[[V0:.*]] = "test.get_tuple_element"(%[[ARG0]]) <{index = 0 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<>
+// CHECK-TUP-DAG:     %[[V1:.*]] = "test.get_tuple_element"(%[[ARG0]]) <{index = 1 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> i1
+// CHECK-TUP-DAG:     %[[V2:.*]] = "test.get_tuple_element"(%[[ARG0]]) <{index = 2 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
+// CHECK-TUP-DAG:     %[[V3:.*]] = "test.get_tuple_element"(%[[V2]]) <{index = 0 : i32}> : (tuple<tuple<i2>>) -> tuple<i2>
+// CHECK-TUP-DAG:     %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 0 : i32}> : (tuple<i2>) -> i2
+// CHECK-TUP-DAG:     %[[V5:.*]] = "test.get_tuple_element"(%[[ARG0]]) <{index = 0 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<>
+// CHECK-TUP-DAG:     %[[V6:.*]] = "test.get_tuple_element"(%[[ARG0]]) <{index = 1 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> i1
+// CHECK-TUP-DAG:     %[[V7:.*]] = "test.get_tuple_element"(%[[ARG0]]) <{index = 2 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
+// CHECK-TUP-DAG:     %[[V8:.*]] = "test.get_tuple_element"(%[[V7]]) <{index = 0 : i32}> : (tuple<tuple<i2>>) -> tuple<i2>
+// CHECK-TUP-DAG:     %[[V9:.*]] = "test.get_tuple_element"(%[[V8]]) <{index = 0 : i32}> : (tuple<i2>) -> i2
 // CHECK-TUP-DAG:     return %[[V1]], %[[V9]] : i1, i2
 
 // If we only convert the func ops, argument materializations are created from
 // the converted tuple elements back to the tuples that the `get_tuple_element`
 // ops expect.
 //
 // CHECK-FUNC-LABEL: func.func @materializations_tuple_args(
 // CHECK-FUNC-SAME:                                         %[[ARG0:.*]]: i1,
 // CHECK-FUNC-SAME:                                         %[[ARG1:.*]]: i2) -> (i1, i2) {
 // CHECK-FUNC-DAG:     %[[V0:.*]] = "test.make_tuple"() : () -> tuple<>
 // CHECK-FUNC-DAG:     %[[V1:.*]] = "test.make_tuple"(%[[ARG1]]) : (i2) -> tuple<i2>
 // CHECK-FUNC-DAG:     %[[V2:.*]] = "test.make_tuple"(%[[V1]]) : (tuple<i2>) -> tuple<tuple<i2>>
 // CHECK-FUNC-DAG:     %[[V3:.*]] = "test.make_tuple"(%[[V0]], %[[ARG0]], %[[V2]]) : (tuple<>, i1, tuple<tuple<i2>>) -> tuple<tuple<>, i1, tuple<tuple<i2>>>
-// CHECK-FUNC-DAG:     %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 0 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<>
-// CHECK-FUNC-DAG:     %[[V5:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 1 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> i1
-// CHECK-FUNC-DAG:     %[[V6:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 2 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
-// CHECK-FUNC-DAG:     %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) {index = 0 : i32} : (tuple<tuple<i2>>) -> tuple<i2>
-// CHECK-FUNC-DAG:     %[[V8:.*]] = "test.get_tuple_element"(%[[V7]]) {index = 0 : i32} : (tuple<i2>) -> i2
+// CHECK-FUNC-DAG:     %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 0 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<>
+// CHECK-FUNC-DAG:     %[[V5:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 1 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> i1
+// CHECK-FUNC-DAG:     %[[V6:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 2 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
+// CHECK-FUNC-DAG:     %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) <{index = 0 : i32}> : (tuple<tuple<i2>>) -> tuple<i2>
+// CHECK-FUNC-DAG:     %[[V8:.*]] = "test.get_tuple_element"(%[[V7]]) <{index = 0 : i32}> : (tuple<i2>) -> i2
 // CHECK-FUNC-DAG:     return %[[V5]], %[[V8]] : i1, i2
 
 // If we convert both tuple and func ops, basically everything disappears.
 //
 // CHECK-BOTH-LABEL: func.func @materializations_tuple_args(
 // CHECK-BOTH-SAME:                                         %[[ARG0:.*]]: i1,
 // CHECK-BOTH-SAME:                                         %[[ARG1:.*]]: i2) -> (i1, i2) {
 // CHECK-BOTH-DAG:     return %[[ARG0]], %[[ARG1]] : i1, i2

 //
 // CHECK-FUNC-LABEL: func.func @materializations_tuple_return(
 // CHECK-FUNC-SAME:                                           %[[ARG0:.*]]: i1,
 // CHECK-FUNC-SAME:                                           %[[ARG1:.*]]: i2) -> (i1, i2) {
 // CHECK-FUNC-DAG:     %[[V0:.*]] = "test.make_tuple"() : () -> tuple<>
 // CHECK-FUNC-DAG:     %[[V1:.*]] = "test.make_tuple"(%[[ARG1]]) : (i2) -> tuple<i2>
 // CHECK-FUNC-DAG:     %[[V2:.*]] = "test.make_tuple"(%[[V1]]) : (tuple<i2>) -> tuple<tuple<i2>>
 // CHECK-FUNC-DAG:     %[[V3:.*]] = "test.make_tuple"(%[[V0]], %[[ARG0]], %[[V2]]) : (tuple<>, i1, tuple<tuple<i2>>) -> tuple<tuple<>, i1, tuple<tuple<i2>>>
-// CHECK-FUNC-DAG:     %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 0 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<>
-// CHECK-FUNC-DAG:     %[[V5:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 1 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> i1
-// CHECK-FUNC-DAG:     %[[V6:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 2 : i32} : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
-// CHECK-FUNC-DAG:     %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) {index = 0 : i32} : (tuple<tuple<i2>>) -> tuple<i2>
-// CHECK-FUNC-DAG:     %[[V8:.*]] = "test.get_tuple_element"(%[[V7]]) {index = 0 : i32} : (tuple<i2>) -> i2
+// CHECK-FUNC-DAG:     %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 0 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<>
+// CHECK-FUNC-DAG:     %[[V5:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 1 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> i1
+// CHECK-FUNC-DAG:     %[[V6:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 2 : i32}> : (tuple<tuple<>, i1, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
+// CHECK-FUNC-DAG:     %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) <{index = 0 : i32}> : (tuple<tuple<i2>>) -> tuple<i2>
+// CHECK-FUNC-DAG:     %[[V8:.*]] = "test.get_tuple_element"(%[[V7]]) <{index = 0 : i32}> : (tuple<i2>) -> i2
 // CHECK-FUNC-DAG:     return %[[V5]], %[[V8]] : i1, i2
 
 // If we convert both tuple and func ops, basically everything disappears.
 //
 // CHECK-BOTH-LABEL: func.func @materializations_tuple_return(
 // CHECK-BOTH-SAME:                                           %[[ARG0:.*]]: i1,
 // CHECK-BOTH-SAME:                                           %[[ARG1:.*]]: i2) -> (i1, i2) {
 // CHECK-BOTH-DAG:     return %[[ARG0]], %[[ARG1]] : i1, i2
 
 func.func @materializations_tuple_return(%arg0: i1, %arg1: i2) -> tuple<tuple<>, i1, tuple<tuple<i2>>> {
   %0 = "test.make_tuple"() : () -> tuple<>
   %1 = "test.make_tuple"(%arg1) : (i2) -> tuple<i2>
   %2 = "test.make_tuple"(%1) : (tuple<i2>) -> tuple<tuple<i2>>
   %3 = "test.make_tuple"(%0, %arg0, %2) : (tuple<>, i1, tuple<tuple<i2>>) -> tuple<tuple<>, i1, tuple<tuple<i2>>>
   return %3 : tuple<tuple<>, i1, tuple<tuple<i2>>>
 }

mlir/test/Conversion/OneToNTypeConversion/scf-structural-one-to-n-type-conversion.mlir

 
 // CHECK-LABEL: func.func @if_tuple_ops(
 // CHECK-SAME:                          %[[ARG0:.*]]: i1,
 // CHECK-SAME:                          %[[ARG1:.*]]: i1) -> i1 {
 // CHECK-NEXT:    %[[V0:.*]] = "test.make_tuple"() : () -> tuple<>
 // CHECK-NEXT:    %[[V1:.*]] = "test.make_tuple"(%[[V0]], %[[ARG0]]) : (tuple<>, i1) -> tuple<tuple<>, i1>
 // CHECK-NEXT:    %[[V2:.*]] = scf.if %[[ARG1]] -> (i1) {
 // CHECK-NEXT:      %[[V3:.*]] = "test.op"(%[[V1]]) : (tuple<tuple<>, i1>) -> tuple<tuple<>, i1>
-// CHECK-NEXT:      %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 0 : i32} : (tuple<tuple<>, i1>) -> tuple<>
-// CHECK-NEXT:      %[[V5:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 1 : i32} : (tuple<tuple<>, i1>) -> i1
+// CHECK-NEXT:      %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 0 : i32}> : (tuple<tuple<>, i1>) -> tuple<>
+// CHECK-NEXT:      %[[V5:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 1 : i32}> : (tuple<tuple<>, i1>) -> i1
 // CHECK-NEXT:      scf.yield %[[V5]] : i1
 // CHECK-NEXT:    } else {
 // CHECK-NEXT:      %[[V6:.*]] = "test.source"() : () -> tuple<tuple<>, i1>
-// CHECK-NEXT:      %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) {index = 0 : i32} : (tuple<tuple<>, i1>) -> tuple<>
-// CHECK-NEXT:      %[[V8:.*]] = "test.get_tuple_element"(%[[V6]]) {index = 1 : i32} : (tuple<tuple<>, i1>) -> i1
+// CHECK-NEXT:      %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) <{index = 0 : i32}> : (tuple<tuple<>, i1>) -> tuple<>
+// CHECK-NEXT:      %[[V8:.*]] = "test.get_tuple_element"(%[[V6]]) <{index = 1 : i32}> : (tuple<tuple<>, i1>) -> i1
 // CHECK-NEXT:      scf.yield %[[V8]] : i1
 // CHECK-NEXT:    }
 // CHECK-NEXT:    return %[[V2]] : i1
 func.func @if_tuple_ops(%arg0: tuple<tuple<>, i1>, %arg1: i1) -> tuple<tuple<>, i1> {
   %0 = scf.if %arg1 -> (tuple<tuple<>, i1>) {
     %1 = "test.op"(%arg0) : (tuple<tuple<>, i1>) -> tuple<tuple<>, i1>
     scf.yield %1 : tuple<tuple<>, i1>
   } else {

 
 // CHECK-LABEL: func.func @while_tuple_ops(
 // CHECK-SAME:                             %[[ARG0:.*]]: i1,
 // CHECK-SAME:                             %[[ARG1:.*]]: i1) -> i1 {
 // CHECK-NEXT:    %[[V0:.*]] = scf.while (%[[ARG2:.*]] = %[[ARG0]]) : (i1) -> i1 {
 // CHECK-NEXT:      %[[V1:.*]] = "test.make_tuple"() : () -> tuple<>
 // CHECK-NEXT:      %[[V2:.*]] = "test.make_tuple"(%[[V1]], %[[ARG2]]) : (tuple<>, i1) -> tuple<tuple<>, i1>
 // CHECK-NEXT:      %[[V3:.*]] = "test.op"(%[[V2]]) : (tuple<tuple<>, i1>) -> tuple<tuple<>, i1>
-// CHECK-NEXT:      %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 0 : i32} : (tuple<tuple<>, i1>) -> tuple<>
-// CHECK-NEXT:      %[[V5:.*]] = "test.get_tuple_element"(%[[V3]]) {index = 1 : i32} : (tuple<tuple<>, i1>) -> i1
+// CHECK-NEXT:      %[[V4:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 0 : i32}> : (tuple<tuple<>, i1>) -> tuple<>
+// CHECK-NEXT:      %[[V5:.*]] = "test.get_tuple_element"(%[[V3]]) <{index = 1 : i32}> : (tuple<tuple<>, i1>) -> i1
 // CHECK-NEXT:      scf.condition(%[[ARG1]]) %[[V5]] : i1
 // CHECK-NEXT:    } do {
 // CHECK-NEXT:    ^bb0(%[[ARG3:.*]]: i1):
 // CHECK-NEXT:      %[[V6:.*]] = "test.source"() : () -> tuple<tuple<>, i1>
-// CHECK-NEXT:      %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) {index = 0 : i32} : (tuple<tuple<>, i1>) -> tuple<>
-// CHECK-NEXT:      %[[V8:.*]] = "test.get_tuple_element"(%[[V6]]) {index = 1 : i32} : (tuple<tuple<>, i1>) -> i1
+// CHECK-NEXT:      %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) <{index = 0 : i32}> : (tuple<tuple<>, i1>) -> tuple<>
+// CHECK-NEXT:      %[[V8:.*]] = "test.get_tuple_element"(%[[V6]]) <{index = 1 : i32}> : (tuple<tuple<>, i1>) -> i1
 // CHECK-NEXT:      scf.yield %[[V8]] : i1
 // CHECK-NEXT:    }
 // CHECK-NEXT:    return %[[V0]] : i1
 func.func @while_tuple_ops(%arg0: tuple<tuple<>, i1>, %arg1: i1) -> tuple<tuple<>, i1> {
   %0 = scf.while (%arg2 = %arg0) : (tuple<tuple<>, i1>) -> tuple<tuple<>, i1> {
     %1 = "test.op"(%arg2) : (tuple<tuple<>, i1>) -> tuple<tuple<>, i1>
     scf.condition(%arg1) %1 : tuple<tuple<>, i1>
   } do {
   ^bb0(%arg2: tuple<tuple<>, i1>):
     %1 = "test.source"() : () -> tuple<tuple<>, i1>
     scf.yield %1 : tuple<tuple<>, i1>
   }
   return %0 : tuple<tuple<>, i1>
 }

mlir/test/Dialect/Shape/invalid.mlir

func.func @shape_of(%value_arg : !shape.value_shape,
   // expected-error@+1 {{if at least one of the operands can hold error values then the result must be of type `shape` to propagate them}}
   %0 = shape.shape_of %value_arg : !shape.value_shape -> tensor<?xindex>
   return
 }
 
 // -----
 
 func.func @shape_of_incompatible_return_types(%value_arg : tensor<1x2xindex>) {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{'shape.shape_of' op inferred type(s) 'tensor<2xindex>' are incompatible with return type(s) of operation 'tensor<3xindex>'}}
   %0 = shape.shape_of %value_arg : tensor<1x2xindex> -> tensor<3xindex>
   return
 }
 
 // -----
 
 func.func @rank(%arg : !shape.shape) {
func.func @fn(%arg: !shape.shape) -> !shape.witness {
   return %0 : !shape.witness
 }
 
 // -----
 
 // Test that type inference flags the wrong return type.
 
 func.func @const_shape() {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{'tensor<3xindex>' are incompatible with return type(s) of operation 'tensor<2xindex>'}}
   %0 = shape.const_shape [4, 5, 6] : tensor<2xindex>
   return
 }
 
 // -----
 
 func.func @invalid_meet(%arg0 : !shape.shape, %arg1 : index) -> index {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{requires all sizes or shapes}}
   %result = shape.meet %arg0, %arg1 : !shape.shape, index -> index
   return %result : index
 }
 
 // -----
 
 func.func @invalid_meet(%arg0 : tensor<2xindex>, %arg1 : tensor<3xindex>) -> tensor<?xindex> {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{unequal shape cardinality}}
   %result = shape.meet %arg0, %arg1 : tensor<2xindex>, tensor<3xindex> -> tensor<?xindex>
   return %result : tensor<?xindex>
 }
 

mlir/test/Dialect/Tosa/invalid.mlir

func.func @test_conv2d(%arg0: tensor<1x29x29x4xi8>, %arg1: tensor<16x3x3x4xi8>, %arg2: tensor<16xi8>) -> tensor<1x27x27x16xi8> {
   %0 = "tosa.conv2d"(%arg0, %arg1, %arg2) {dilation = array<i64: 1, 1>, pad = array<i64: 0, 0, 0, 0>, stride = array<i64: 1, 1>}
            : (tensor<1x29x29x4xi8>, tensor<16x3x3x4xi8>, tensor<16xi8>) -> tensor<1x27x27x16xi8>
   return %0 : tensor<1x27x27x16xi8>
 }
 
 // -----
 
 func.func @test_concat(%arg0 : tensor<2x1xf32>, %arg1 : tensor<2x2xf32>) -> tensor<?x?xf32> {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{Cannot concat tensors with different sizes on the non-axis dimension 1}}
   %0 = "tosa.concat"(%arg0, %arg1) {axis = 0 : i64} : (tensor<2x1xf32>, tensor<2x2xf32>) -> tensor<?x?xf32>
   return %0 : tensor<?x?xf32>
 }
 
 // -----
 
 func.func @test_concat_element_type_mismatch(%arg0 : tensor<1x2xf32>, %arg1 : tensor<2x2xf32>) -> tensor<?x?xi8> {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{'tosa.concat' op inferred type(s) 'tensor<3x2xf32>' are incompatible with return type(s) of operation 'tensor<?x?xi8>}}
   %0 = "tosa.concat"(%arg0, %arg1) {axis = 0 : i64} : (tensor<1x2xf32>, tensor<2x2xf32>) -> tensor<?x?xi8>
   return %0 : tensor<?x?xi8>
 }
 
 // -----
 
 func.func @test_pad_non_const(%arg0: tensor<13x21x3xf32>, %arg1: tensor<3x2xi32>) -> tensor<13x21x3xf32> {
func.func @test_fully_connected_non_const(%arg0: tensor<13x21x3xf32>, %arg1: tensor<2xf32>) -> tensor<273x2xf32> {
   // expected-error@+1 {{'tosa.fully_connected' op bias of fully_connected is not constant}}
   %2 = "tosa.fully_connected"(%1, %0, %arg1) : (tensor<273x3xf32>, tensor<2x3xf32>, tensor<2xf32>) -> tensor<273x2xf32>
   return %2 : tensor<273x2xf32>
 }
 
 // -----
 
 func.func @test_reduce_sum_type_mismatch(%arg0 : tensor<2x3x4x5xf32>) -> () {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{'tosa.reduce_sum' op inferred type(s) 'tensor<1x3x4x5xf32>' are incompatible with return type(s) of operation 'tensor<1x3x4x5xi32>'}}
   %0 = "tosa.reduce_sum"(%arg0) {axis = 0 : i64} : (tensor<2x3x4x5xf32>) -> tensor<1x3x4x5xi32>
   return
 }
 
 // -----
 
 func.func @test_reduce_max_type_mismatch(%arg0 : tensor<2x3x4x5xf32>) -> () {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{'tosa.reduce_max' op inferred type(s) 'tensor<2x3x4x1xf32>' are incompatible with return type(s) of operation 'tensor<2x3x4x1xi32>'}}
   %0 = "tosa.reduce_max"(%arg0) {axis = 3 : i64} : (tensor<2x3x4x5xf32>) -> tensor<2x3x4x1xi32>
   return
 }
 
 // -----
 
 func.func @test_reduce_min_type_mismatch(%arg0 : tensor<2x3x4x5xf32>) -> () {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{'tosa.reduce_min' op inferred type(s) 'tensor<2x1x4x5xf32>' are incompatible with return type(s) of operation 'tensor<2x1x4x5xi32>'}}
   %0 = "tosa.reduce_min"(%arg0) {axis = 1 : i64} : (tensor<2x3x4x5xf32>) -> tensor<2x1x4x5xi32>
   return
 }
 
 // -----
 
 func.func @test_reduce_prod_type_mismatch(%arg0 : tensor<2x3x4x5xf32>) -> () {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{'tosa.reduce_prod' op inferred type(s) 'tensor<2x1x4x5xf32>' are incompatible with return type(s) of operation 'tensor<2x3x4x5xf32>'}}
   %0 = "tosa.reduce_prod"(%arg0) {axis = 1 : i64} : (tensor<2x3x4x5xf32>) -> tensor<2x3x4x5xf32>
   return
 }
 
 // -----
 
 func.func @test_reshape_type_mismatch(%arg0 : tensor<13x21x3xf32>) -> () {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{'tosa.reshape' op inferred type(s) 'tensor<13x21x3x1xf32>' are incompatible with return type(s) of operation 'tensor<13x21x3x1xi32>'}}
   %0 = "tosa.reshape"(%arg0) {new_shape = array<i64: 13, 21, 3, 1>} : (tensor<13x21x3xf32>) -> tensor<13x21x3x1xi32>
   return
 }

mlir/test/IR/greedy-pattern-rewriter-driver.mlir

 // RUN: mlir-opt %s -test-patterns="max-iterations=1" \
 // RUN:     -allow-unregistered-dialect --split-input-file | FileCheck %s
 
 // CHECK-LABEL: func @add_to_worklist_after_inplace_update()
 func.func @add_to_worklist_after_inplace_update() {
   // The following op is updated in-place and should be added back to the
   // worklist of the GreedyPatternRewriteDriver (regardless of the value of
   // config.max_iterations).
 
-  // CHECK: "test.any_attr_of_i32_str"() {attr = 3 : i32} : () -> ()
+  // CHECK: "test.any_attr_of_i32_str"() <{attr = 3 : i32}> : () -> ()
   "test.any_attr_of_i32_str"() {attr = 0 : i32} : () -> ()
   return
 }
 
 // -----
 
 // CHECK-LABEL: func @add_ancestors_to_worklist()
 func.func @add_ancestors_to_worklist() {
        // CHECK: "foo.maybe_eligible_op"() {eligible} : () -> index
   // CHECK-NEXT: "test.one_region_op"()
   "test.one_region_op"() ({
     %0 = "foo.maybe_eligible_op" () : () -> (index)
     "foo.yield"(%0) : (index) -> ()
   }) {hoist_eligible_ops}: () -> ()
   return
 }

mlir/test/IR/invalid.mlir

 
 // -----
 
 // expected-error @+1 {{unbalanced '<' character in pretty dialect name}}
 func.func @bad_arrow(%arg : !unreg.ptr<(i32)->)
 
 // -----
 
-// expected-error @+1 {{attribute 'attr' occurs more than once in the attribute list}}
-test.format_symbol_name_attr_op @name { attr = "xx" }
-
-// -----
-
 func.func @forward_reference_type_check() -> (i8) {
   cf.br ^bb2
 
 ^bb1:
   // expected-note @+1 {{previously used here with type 'i8'}}
   return %1 : i8
 
 ^bb2:

mlir/test/IR/parser.mlir

 // This is a registered operation with no custom parser and printer, and should
 // be handled by the dialect.
 // CHECK: test.dialect_custom_format_fallback custom_format_fallback
 test.dialect_custom_format_fallback custom_format_fallback
 
 // Check that an op with an optional result parses f80 as type.
 // CHECK: test.format_optional_result_d_op : f80
 test.format_optional_result_d_op : f80
+

mlir/test/IR/properties.mlir

+// # RUN: mlir-opt %s -split-input-file | mlir-opt  |FileCheck %s
+// # RUN: mlir-opt %s -mlir-print-op-generic -split-input-file  | mlir-opt -mlir-print-op-generic | FileCheck %s --check-prefix=GENERIC
+
+// CHECK:   test.with_properties
+// CHECK-SAME: <{a = 32 : i64, array = array<i64: 1, 2, 3, 4>, b = "foo"}>
+// GENERIC:   "test.with_properties"()
+// GENERIC-SAME: <{a = 32 : i64, array = array<i64: 1, 2, 3, 4>, b = "foo"}> : () -> ()
+test.with_properties <{a = 32 : i64, array = array<i64: 1, 2, 3, 4>, b = "foo"}>
+
+// CHECK:   test.with_nice_properties
+// CHECK-SAME:    "foo bar" is -3
+// GENERIC: "test.with_nice_properties"()
+// GENERIC-SAME:  <{prop = {label = "foo bar", value = -3 : i32}}> : () -> ()
+test.with_nice_properties "foo bar" is -3
+
+// CHECK:   test.with_wrapped_properties
+// CHECK-SAME:    "content for properties"
+// GENERIC: "test.with_wrapped_properties"()
+// GENERIC-SAME:  <{prop = "content for properties"}> : () -> ()
+test.with_wrapped_properties <{prop = "content for properties"}>

mlir/test/IR/test-fold-adaptor.mlir

 // RUN: mlir-opt %s -canonicalize -split-input-file | FileCheck %s
 
 func.func @test() -> i32 {
   %c5 = "test.constant"() {value = 5 : i32} : () -> i32
   %c1 = "test.constant"() {value = 1 : i32} : () -> i32
   %c2 = "test.constant"() {value = 2 : i32} : () -> i32
   %c3 = "test.constant"() {value = 3 : i32} : () -> i32
   %res = test.fold_with_fold_adaptor %c5, [ %c1, %c2], { (%c3), (%c3) } {
     %c0 = "test.constant"() {value = 0 : i32} : () -> i32
   }
   return %res : i32
 }
 
 // CHECK-LABEL: func.func @test
-// CHECK-NEXT: %[[C:.*]] = "test.constant"() {value = 33 : i32}
+// CHECK-NEXT: %[[C:.*]] = "test.constant"() <{value = 33 : i32}>
 // CHECK-NEXT: return %[[C]]

mlir/test/IR/test-manual-cpp-fold.mlir

 // RUN: mlir-opt %s -canonicalize -split-input-file | FileCheck %s
 
 func.func @test() -> i32 {
   %c5 = "test.constant"() {value = 5 : i32} : () -> i32
   %res = "test.manual_cpp_op_with_fold"(%c5) : (i32) -> i32
   return %res : i32
 }
 
 // CHECK-LABEL: func.func @test
-// CHECK-NEXT: %[[C:.*]] = "test.constant"() {value = 5 : i32}
+// CHECK-NEXT: %[[C:.*]] = "test.constant"() <{value = 5 : i32}>
 // CHECK-NEXT: return %[[C]]

mlir/test/IR/traits.mlir

 
 // Test the invariants of operations with the Symbol Trait.
 
 // expected-error@+1 {{op requires attribute 'sym_name'}}
 "test.symbol"() {} : () -> ()
 
 // -----
 
-// expected-error@+1 {{op attribute 'sym_visibility' failed to satisfy constraint: string attribute}}
-"test.symbol"() {sym_name = "foo_2", sym_visibility} : () -> ()
+// expected-error@+1 {{invalid properties {sym_name = "foo_2", sym_visibility} for op test.symbol: Invalid attribute `sym_visibility` in property conversion: unit}}
+"test.symbol"() <{sym_name = "foo_2", sym_visibility}> : () -> ()
 
 // -----
 
 // expected-error@+1 {{visibility expected to be one of ["public", "private", "nested"]}}
 "test.symbol"() {sym_name = "foo_2", sym_visibility = "foo"} : () -> ()
 
 // -----
 

 func.func @failedMissingOperandSizeAttr(%arg: i32) {
   // expected-error @+1 {{requires dense i32 array attribute 'operand_segment_sizes'}}
   "test.attr_sized_operands"(%arg, %arg, %arg, %arg) : (i32, i32, i32, i32) -> ()
 }
 
 // -----
 
 func.func @failedOperandSizeAttrWrongType(%arg: i32) {
-  // expected-error @+1 {{requires dense i32 array attribute 'operand_segment_sizes'}}
+  // expected-error @+1 {{attribute 'operand_segment_sizes' failed to satisfy constraint: i32 dense array attribute}}
   "test.attr_sized_operands"(%arg, %arg, %arg, %arg) {operand_segment_sizes = 10} : (i32, i32, i32, i32) -> ()
 }
 
 // -----
 
 func.func @failedOperandSizeAttrWrongElementType(%arg: i32) {
-  // expected-error @+1 {{requires dense i32 array attribute 'operand_segment_sizes'}}
+  // expected-error @+1 {{attribute 'operand_segment_sizes' failed to satisfy constraint: i32 dense array attribute}}
   "test.attr_sized_operands"(%arg, %arg, %arg, %arg) {operand_segment_sizes = array<i64: 1, 1, 1, 1>} : (i32, i32, i32, i32) -> ()
 }
 
 // -----
 
 func.func @failedOperandSizeAttrNegativeValue(%arg: i32) {
   // expected-error @+1 {{'operand_segment_sizes' attribute cannot have negative elements}}
   "test.attr_sized_operands"(%arg, %arg, %arg, %arg) {operand_segment_sizes = array<i32: 1, 1, -1, 1>} : (i32, i32, i32, i32) -> ()
func.func @failed_type_traits() {
   "test.result_type_with_trait"() : () -> i32
   return
 }
 
 // -----
 
 // Check that we can query traits in attributes
 func.func @succeeded_attr_traits() {
-  // CHECK: "test.attr_with_trait"() {attr = #test.attr_with_trait} : () -> ()
+  // CHECK: "test.attr_with_trait"() <{attr = #test.attr_with_trait}> : () -> ()
   "test.attr_with_trait"() {attr = #test.attr_with_trait} : () -> ()
   return
 }
 
 // -----
 
 // Check that we can query traits in attributes
 func.func @failed_attr_traits() {
   // expected-error@+1 {{'attr' attribute should have trait 'TestAttrTrait'}}
   "test.attr_with_trait"() {attr = 42 : i32} : () -> ()
   return
 }

mlir/test/Interfaces/InferIntRangeInterface/infer-int-range-test-ops.mlir

 // RUN: mlir-opt -test-int-range-inference %s | FileCheck %s
 
 // CHECK-LABEL: func @constant
-// CHECK: %[[cst:.*]] = "test.constant"() {value = 3 : index}
+// CHECK: %[[cst:.*]] = "test.constant"() <{value = 3 : index}
 // CHECK: return %[[cst]]
 func.func @constant() -> index {
   %0 = test.with_bounds { umin = 3 : index, umax = 3 : index,
                                smin = 3 : index, smax = 3 : index}
   func.return %0 : index
 }
 
 // CHECK-LABEL: func @increment
-// CHECK: %[[cst:.*]] = "test.constant"() {value = 4 : index}
+// CHECK: %[[cst:.*]] = "test.constant"() <{value = 4 : index}
 // CHECK: return %[[cst]]
 func.func @increment() -> index {
   %0 = test.with_bounds { umin = 3 : index, umax = 3 : index, smin = 0 : index, smax = 0x7fffffffffffffff : index }
   %1 = test.increment %0
   func.return %1 : index
 }
 
 // CHECK-LABEL: func @maybe_increment

 // CHECK: test.reflect_bounds {smax = 9223372036854775807 : index, smin = -9223372036854775808 : index, umax = -1 : index, umin = 0 : index}
 func.func @func_args_unbound(%arg0 : index) -> index {
   %0 = test.reflect_bounds %arg0
   func.return %0 : index
 }
 
 // CHECK-LABEL: func @propagate_across_while_loop_false()
 func.func @propagate_across_while_loop_false() -> index {
-  // CHECK-DAG: %[[C0:.*]] = "test.constant"() {value = 0
-  // CHECK-DAG: %[[C1:.*]] = "test.constant"() {value = 1
+  // CHECK-DAG: %[[C0:.*]] = "test.constant"() <{value = 0
+  // CHECK-DAG: %[[C1:.*]] = "test.constant"() <{value = 1
   %0 = test.with_bounds { umin = 0 : index, umax = 0 : index,
                           smin = 0 : index, smax = 0 : index }
   %1 = scf.while : () -> index {
     %false = arith.constant false
     // CHECK: scf.condition(%{{.*}}) %[[C0]]
     scf.condition(%false) %0 : index
   } do {
   ^bb0(%i1: index):
     scf.yield
   }
   // CHECK: return %[[C1]]
   %2 = test.increment %1
   return %2 : index
 }
 
 // CHECK-LABEL: func @propagate_across_while_loop
 func.func @propagate_across_while_loop(%arg0 : i1) -> index {
-  // CHECK-DAG: %[[C0:.*]] = "test.constant"() {value = 0
-  // CHECK-DAG: %[[C1:.*]] = "test.constant"() {value = 1
+  // CHECK-DAG: %[[C0:.*]] = "test.constant"() <{value = 0
+  // CHECK-DAG: %[[C1:.*]] = "test.constant"() <{value = 1
   %0 = test.with_bounds { umin = 0 : index, umax = 0 : index,
                           smin = 0 : index, smax = 0 : index }
   %1 = scf.while : () -> index {
     // CHECK: scf.condition(%{{.*}}) %[[C0]]
     scf.condition(%arg0) %0 : index
   } do {
   ^bb0(%i1: index):
     scf.yield
   }
   // CHECK: return %[[C1]]
   %2 = test.increment %1
   return %2 : index
 }
 
 // CHECK-LABEL: func @dont_propagate_across_infinite_loop()
 func.func @dont_propagate_across_infinite_loop() -> index {
-  // CHECK: %[[C0:.*]] = "test.constant"() {value = 0
+  // CHECK: %[[C0:.*]] = "test.constant"() <{value = 0
   %0 = test.with_bounds { umin = 0 : index, umax = 0 : index,
                           smin = 0 : index, smax = 0 : index }
   // CHECK: %[[loopRes:.*]] = scf.while
   %1 = scf.while : () -> index {
     %true = arith.constant true
     // CHECK: scf.condition(%{{.*}}) %[[C0]]
     scf.condition(%true) %0 : index
   } do {

mlir/test/Transforms/decompose-call-graph-types.mlir

 // RUN: mlir-opt %s -split-input-file -test-decompose-call-graph-types | FileCheck %s
 
 // RUN: mlir-opt %s -split-input-file \
 // RUN:   -test-one-to-n-type-conversion="convert-func-ops" \
 // RUN: | FileCheck %s --check-prefix=CHECK-12N
 
 // Test case: Most basic case of a 1:N decomposition, an identity function.
 
 // CHECK-LABEL:   func @identity(
 // CHECK-SAME:                   %[[ARG0:.*]]: i1,
 // CHECK-SAME:                   %[[ARG1:.*]]: i32) -> (i1, i32) {
 // CHECK:           %[[ARG_MATERIALIZED:.*]] = "test.make_tuple"(%[[ARG0]], %[[ARG1]]) : (i1, i32) -> tuple<i1, i32>
-// CHECK:           %[[RET0:.*]] = "test.get_tuple_element"(%[[ARG_MATERIALIZED]]) {index = 0 : i32} : (tuple<i1, i32>) -> i1
-// CHECK:           %[[RET1:.*]] = "test.get_tuple_element"(%[[ARG_MATERIALIZED]]) {index = 1 : i32} : (tuple<i1, i32>) -> i32
+// CHECK:           %[[RET0:.*]] = "test.get_tuple_element"(%[[ARG_MATERIALIZED]]) <{index = 0 : i32}> : (tuple<i1, i32>) -> i1
+// CHECK:           %[[RET1:.*]] = "test.get_tuple_element"(%[[ARG_MATERIALIZED]]) <{index = 1 : i32}> : (tuple<i1, i32>) -> i32
 // CHECK:           return %[[RET0]], %[[RET1]] : i1, i32
 // CHECK-12N-LABEL:   func @identity(
 // CHECK-12N-SAME:                   %[[ARG0:.*]]: i1,
 // CHECK-12N-SAME:                   %[[ARG1:.*]]: i32) -> (i1, i32) {
 // CHECK-12N:           return %[[ARG0]], %[[ARG1]] : i1, i32
 func.func @identity(%arg0: tuple<i1, i32>) -> tuple<i1, i32> {
   return %arg0 : tuple<i1, i32>
 }

 // CHECK-LABEL:   func @mixed_recursive_decomposition(
 // CHECK-SAME:                 %[[ARG0:.*]]: i1,
 // CHECK-SAME:                 %[[ARG1:.*]]: i2) -> (i1, i2) {
 // CHECK:           %[[V0:.*]] = "test.make_tuple"() : () -> tuple<>
 // CHECK:           %[[V1:.*]] = "test.make_tuple"(%[[ARG0]]) : (i1) -> tuple<i1>
 // CHECK:           %[[V2:.*]] = "test.make_tuple"(%[[ARG1]]) : (i2) -> tuple<i2>
 // CHECK:           %[[V3:.*]] = "test.make_tuple"(%[[V2]]) : (tuple<i2>) -> tuple<tuple<i2>>
 // CHECK:           %[[V4:.*]] = "test.make_tuple"(%[[V0]], %[[V1]], %[[V3]]) : (tuple<>, tuple<i1>, tuple<tuple<i2>>) -> tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>
-// CHECK:           %[[V5:.*]] = "test.get_tuple_element"(%[[V4]]) {index = 0 : i32} : (tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>) -> tuple<>
-// CHECK:           %[[V6:.*]] = "test.get_tuple_element"(%[[V4]]) {index = 1 : i32} : (tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>) -> tuple<i1>
-// CHECK:           %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) {index = 0 : i32} : (tuple<i1>) -> i1
-// CHECK:           %[[V8:.*]] = "test.get_tuple_element"(%[[V4]]) {index = 2 : i32} : (tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
-// CHECK:           %[[V9:.*]] = "test.get_tuple_element"(%[[V8]]) {index = 0 : i32} : (tuple<tuple<i2>>) -> tuple<i2>
-// CHECK:           %[[V10:.*]] = "test.get_tuple_element"(%[[V9]]) {index = 0 : i32} : (tuple<i2>) -> i2
+// CHECK:           %[[V5:.*]] = "test.get_tuple_element"(%[[V4]]) <{index = 0 : i32}> : (tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>) -> tuple<>
+// CHECK:           %[[V6:.*]] = "test.get_tuple_element"(%[[V4]]) <{index = 1 : i32}> : (tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>) -> tuple<i1>
+// CHECK:           %[[V7:.*]] = "test.get_tuple_element"(%[[V6]]) <{index = 0 : i32}> : (tuple<i1>) -> i1
+// CHECK:           %[[V8:.*]] = "test.get_tuple_element"(%[[V4]]) <{index = 2 : i32}> : (tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>) -> tuple<tuple<i2>>
+// CHECK:           %[[V9:.*]] = "test.get_tuple_element"(%[[V8]]) <{index = 0 : i32}> : (tuple<tuple<i2>>) -> tuple<i2>
+// CHECK:           %[[V10:.*]] = "test.get_tuple_element"(%[[V9]]) <{index = 0 : i32}> : (tuple<i2>) -> i2
 // CHECK:           return %[[V7]], %[[V10]] : i1, i2
 // CHECK-12N-LABEL:   func @mixed_recursive_decomposition(
 // CHECK-12N-SAME:                 %[[ARG0:.*]]: i1,
 // CHECK-12N-SAME:                 %[[ARG1:.*]]: i2) -> (i1, i2) {
 // CHECK-12N:           return %[[ARG0]], %[[ARG1]] : i1, i2
 func.func @mixed_recursive_decomposition(%arg0: tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>) -> tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>> {
   return %arg0 : tuple<tuple<>, tuple<i1>, tuple<tuple<i2>>>
 }
 
 // -----
 
 // Test case: Check decomposition of calls.
 
 // CHECK-LABEL:   func private @callee(i1, i32) -> (i1, i32)
 // CHECK-12N-LABEL:   func private @callee(i1, i32) -> (i1, i32)
 func.func private @callee(tuple<i1, i32>) -> tuple<i1, i32>
 
 // CHECK-LABEL:   func @caller(
 // CHECK-SAME:                 %[[ARG0:.*]]: i1,
 // CHECK-SAME:                 %[[ARG1:.*]]: i32) -> (i1, i32) {
 // CHECK:           %[[ARG_MATERIALIZED:.*]] = "test.make_tuple"(%[[ARG0]], %[[ARG1]]) : (i1, i32) -> tuple<i1, i32>
-// CHECK:           %[[CALL_ARG0:.*]] = "test.get_tuple_element"(%[[ARG_MATERIALIZED]]) {index = 0 : i32} : (tuple<i1, i32>) -> i1
-// CHECK:           %[[CALL_ARG1:.*]] = "test.get_tuple_element"(%[[ARG_MATERIALIZED]]) {index = 1 : i32} : (tuple<i1, i32>) -> i32
+// CHECK:           %[[CALL_ARG0:.*]] = "test.get_tuple_element"(%[[ARG_MATERIALIZED]]) <{index = 0 : i32}> : (tuple<i1, i32>) -> i1
+// CHECK:           %[[CALL_ARG1:.*]] = "test.get_tuple_element"(%[[ARG_MATERIALIZED]]) <{index = 1 : i32}> : (tuple<i1, i32>) -> i32
 // CHECK:           %[[DECOMPOSED:.*]]:2 = call @callee(%[[CALL_ARG0]], %[[CALL_ARG1]]) : (i1, i32) -> (i1, i32)
 // CHECK:           %[[CALL_RESULT_RECOMPOSED:.*]] = "test.make_tuple"(%[[DECOMPOSED]]#0, %[[DECOMPOSED]]#1) : (i1, i32) -> tuple<i1, i32>
-// CHECK:           %[[RET0:.*]] = "test.get_tuple_element"(%[[CALL_RESULT_RECOMPOSED]]) {index = 0 : i32} : (tuple<i1, i32>) -> i1
-// CHECK:           %[[RET1:.*]] = "test.get_tuple_element"(%[[CALL_RESULT_RECOMPOSED]]) {index = 1 : i32} : (tuple<i1, i32>) -> i32
+// CHECK:           %[[RET0:.*]] = "test.get_tuple_element"(%[[CALL_RESULT_RECOMPOSED]]) <{index = 0 : i32}> : (tuple<i1, i32>) -> i1
+// CHECK:           %[[RET1:.*]] = "test.get_tuple_element"(%[[CALL_RESULT_RECOMPOSED]]) <{index = 1 : i32}> : (tuple<i1, i32>) -> i32
 // CHECK:           return %[[RET0]], %[[RET1]] : i1, i32
 // CHECK-12N-LABEL:   func @caller(
 // CHECK-12N-SAME:                 %[[ARG0:.*]]: i1,
 // CHECK-12N-SAME:                 %[[ARG1:.*]]: i32) -> (i1, i32) {
 // CHECK-12N:           %[[V0:.*]]:2 = call @callee(%[[ARG0]], %[[ARG1]]) : (i1, i32) -> (i1, i32)
 // CHECK-12N:           return %[[V0]]#0, %[[V0]]#1 : i1, i32
 func.func @caller(%arg0: tuple<i1, i32>) -> tuple<i1, i32> {
   %0 = call @callee(%arg0) : (tuple<i1, i32>) -> tuple<i1, i32>

 
 // -----
 
 // Test case: Ensure decompositions are inserted properly around results of
 // unconverted ops.
 
 // CHECK-LABEL:   func @unconverted_op_result() -> (i1, i32) {
 // CHECK:           %[[UNCONVERTED_VALUE:.*]] = "test.source"() : () -> tuple<i1, i32>
-// CHECK:           %[[RET0:.*]] = "test.get_tuple_element"(%[[UNCONVERTED_VALUE]]) {index = 0 : i32} : (tuple<i1, i32>) -> i1
-// CHECK:           %[[RET1:.*]] = "test.get_tuple_element"(%[[UNCONVERTED_VALUE]]) {index = 1 : i32} : (tuple<i1, i32>) -> i32
+// CHECK:           %[[RET0:.*]] = "test.get_tuple_element"(%[[UNCONVERTED_VALUE]]) <{index = 0 : i32}> : (tuple<i1, i32>) -> i1
+// CHECK:           %[[RET1:.*]] = "test.get_tuple_element"(%[[UNCONVERTED_VALUE]]) <{index = 1 : i32}> : (tuple<i1, i32>) -> i32
 // CHECK:           return %[[RET0]], %[[RET1]] : i1, i32
 // CHECK-12N-LABEL:   func @unconverted_op_result() -> (i1, i32) {
 // CHECK-12N:           %[[UNCONVERTED_VALUE:.*]] = "test.source"() : () -> tuple<i1, i32>
-// CHECK-12N:           %[[RET0:.*]] = "test.get_tuple_element"(%[[UNCONVERTED_VALUE]]) {index = 0 : i32} : (tuple<i1, i32>) -> i1
-// CHECK-12N:           %[[RET1:.*]] = "test.get_tuple_element"(%[[UNCONVERTED_VALUE]]) {index = 1 : i32} : (tuple<i1, i32>) -> i32
+// CHECK-12N:           %[[RET0:.*]] = "test.get_tuple_element"(%[[UNCONVERTED_VALUE]]) <{index = 0 : i32}> : (tuple<i1, i32>) -> i1
+// CHECK-12N:           %[[RET1:.*]] = "test.get_tuple_element"(%[[UNCONVERTED_VALUE]]) <{index = 1 : i32}> : (tuple<i1, i32>) -> i32
 // CHECK-12N:           return %[[RET0]], %[[RET1]] : i1, i32
 func.func @unconverted_op_result() -> tuple<i1, i32> {
   %0 = "test.source"() : () -> (tuple<i1, i32>)
   return %0 : tuple<i1, i32>
 }
 
 // -----
 
 // Test case: Ensure decompositions are inserted properly around results of
 // unconverted ops in the case of different nesting levels.
 
 // CHECK-LABEL:   func @nested_unconverted_op_result(
 // CHECK-SAME:                 %[[ARG0:.*]]: i1,
 // CHECK-SAME:                 %[[ARG1:.*]]: i32) -> (i1, i32) {
 // CHECK:           %[[V0:.*]] = "test.make_tuple"(%[[ARG1]]) : (i32) -> tuple<i32>
 // CHECK:           %[[V1:.*]] = "test.make_tuple"(%[[ARG0]], %[[V0]]) : (i1, tuple<i32>) -> tuple<i1, tuple<i32>>
 // CHECK:           %[[V2:.*]] = "test.op"(%[[V1]]) : (tuple<i1, tuple<i32>>) -> tuple<i1, tuple<i32>>
-// CHECK:           %[[V3:.*]] = "test.get_tuple_element"(%[[V2]]) {index = 0 : i32} : (tuple<i1, tuple<i32>>) -> i1
-// CHECK:           %[[V4:.*]] = "test.get_tuple_element"(%[[V2]]) {index = 1 : i32} : (tuple<i1, tuple<i32>>) -> tuple<i32>
-// CHECK:           %[[V5:.*]] = "test.get_tuple_element"(%[[V4]]) {index = 0 : i32} : (tuple<i32>) -> i32
+// CHECK:           %[[V3:.*]] = "test.get_tuple_element"(%[[V2]]) <{index = 0 : i32}> : (tuple<i1, tuple<i32>>) -> i1
+// CHECK:           %[[V4:.*]] = "test.get_tuple_element"(%[[V2]]) <{index = 1 : i32}> : (tuple<i1, tuple<i32>>) -> tuple<i32>
+// CHECK:           %[[V5:.*]] = "test.get_tuple_element"(%[[V4]]) <{index = 0 : i32}> : (tuple<i32>) -> i32
 // CHECK:           return %[[V3]], %[[V5]] : i1, i32
 // CHECK-12N-LABEL:   func @nested_unconverted_op_result(
 // CHECK-12N-SAME:                 %[[ARG0:.*]]: i1,
 // CHECK-12N-SAME:                 %[[ARG1:.*]]: i32) -> (i1, i32) {
 // CHECK-12N:           %[[V0:.*]] = "test.make_tuple"(%[[ARG1]]) : (i32) -> tuple<i32>
 // CHECK-12N:           %[[V1:.*]] = "test.make_tuple"(%[[ARG0]], %[[V0]]) : (i1, tuple<i32>) -> tuple<i1, tuple<i32>>
 // CHECK-12N:           %[[V2:.*]] = "test.op"(%[[V1]]) : (tuple<i1, tuple<i32>>) -> tuple<i1, tuple<i32>>
-// CHECK-12N:           %[[V3:.*]] = "test.get_tuple_element"(%[[V2]]) {index = 0 : i32} : (tuple<i1, tuple<i32>>) -> i1
-// CHECK-12N:           %[[V4:.*]] = "test.get_tuple_element"(%[[V2]]) {index = 1 : i32} : (tuple<i1, tuple<i32>>) -> tuple<i32>
-// CHECK-12N:           %[[V5:.*]] = "test.get_tuple_element"(%[[V4]]) {index = 0 : i32} : (tuple<i32>) -> i32
+// CHECK-12N:           %[[V3:.*]] = "test.get_tuple_element"(%[[V2]]) <{index = 0 : i32}> : (tuple<i1, tuple<i32>>) -> i1
+// CHECK-12N:           %[[V4:.*]] = "test.get_tuple_element"(%[[V2]]) <{index = 1 : i32}> : (tuple<i1, tuple<i32>>) -> tuple<i32>
+// CHECK-12N:           %[[V5:.*]] = "test.get_tuple_element"(%[[V4]]) <{index = 0 : i32}> : (tuple<i32>) -> i32
 // CHECK-12N:           return %[[V3]], %[[V5]] : i1, i32
 func.func @nested_unconverted_op_result(%arg: tuple<i1, tuple<i32>>) -> tuple<i1, tuple<i32>> {
   %0 = "test.op"(%arg) : (tuple<i1, tuple<i32>>) -> (tuple<i1, tuple<i32>>)
   return %0 : tuple<i1, tuple<i32>>
 }
 
 // -----
 
 // Test case: Check mixed decomposed and non-decomposed args.
 // This makes sure to test the cases if 1:0, 1:1, and 1:N decompositions.
 
 // CHECK-LABEL:   func private @callee(i1, i2, i3, i4, i5, i6) -> (i1, i2, i3, i4, i5, i6)
 // CHECK-12N-LABEL:   func private @callee(i1, i2, i3, i4, i5, i6) -> (i1, i2, i3, i4, i5, i6)
 func.func private @callee(tuple<>, i1, tuple<i2>, i3, tuple<i4, i5>, i6) -> (tuple<>, i1, tuple<i2>, i3, tuple<i4, i5>, i6)
 
 // CHECK-LABEL:   func @caller(
 // CHECK-SAME:                 %[[I1:.*]]: i1,
 // CHECK-SAME:                 %[[I2:.*]]: i2,
 // CHECK-SAME:                 %[[I3:.*]]: i3,
 // CHECK-SAME:                 %[[I4:.*]]: i4,
 // CHECK-SAME:                 %[[I5:.*]]: i5,
 // CHECK-SAME:                 %[[I6:.*]]: i6) -> (i1, i2, i3, i4, i5, i6) {
 // CHECK:           %[[ARG_TUPLE:.*]] = "test.make_tuple"(%[[I4]], %[[I5]]) : (i4, i5) -> tuple<i4, i5>
-// CHECK:           %[[ARG_TUPLE_0:.*]] = "test.get_tuple_element"(%[[ARG_TUPLE]]) {index = 0 : i32} : (tuple<i4, i5>) -> i4
-// CHECK:           %[[ARG_TUPLE_1:.*]] = "test.get_tuple_element"(%[[ARG_TUPLE]]) {index = 1 : i32} : (tuple<i4, i5>) -> i5
+// CHECK:           %[[ARG_TUPLE_0:.*]] = "test.get_tuple_element"(%[[ARG_TUPLE]]) <{index = 0 : i32}> : (tuple<i4, i5>) -> i4
+// CHECK:           %[[ARG_TUPLE_1:.*]] = "test.get_tuple_element"(%[[ARG_TUPLE]]) <{index = 1 : i32}> : (tuple<i4, i5>) -> i5
 // CHECK:           %[[CALL:.*]]:6 = call @callee(%[[I1]], %[[I2]], %[[I3]], %[[ARG_TUPLE_0]], %[[ARG_TUPLE_1]], %[[I6]]) : (i1, i2, i3, i4, i5, i6) -> (i1, i2, i3, i4, i5, i6)
 // CHECK:           %[[RET_TUPLE:.*]] = "test.make_tuple"(%[[CALL]]#3, %[[CALL]]#4) : (i4, i5) -> tuple<i4, i5>
-// CHECK:           %[[RET_TUPLE_0:.*]] = "test.get_tuple_element"(%[[RET_TUPLE]]) {index = 0 : i32} : (tuple<i4, i5>) -> i4
-// CHECK:           %[[RET_TUPLE_1:.*]] = "test.get_tuple_element"(%[[RET_TUPLE]]) {index = 1 : i32} : (tuple<i4, i5>) -> i5
+// CHECK:           %[[RET_TUPLE_0:.*]] = "test.get_tuple_element"(%[[RET_TUPLE]]) <{index = 0 : i32}> : (tuple<i4, i5>) -> i4
+// CHECK:           %[[RET_TUPLE_1:.*]] = "test.get_tuple_element"(%[[RET_TUPLE]]) <{index = 1 : i32}> : (tuple<i4, i5>) -> i5
 // CHECK:           return %[[CALL]]#0, %[[CALL]]#1, %[[CALL]]#2, %[[RET_TUPLE_0]], %[[RET_TUPLE_1]], %[[CALL]]#5 : i1, i2, i3, i4, i5, i6
 // CHECK-12N-LABEL:   func @caller(
 // CHECK-12N-SAME:                 %[[I1:.*]]: i1,
 // CHECK-12N-SAME:                 %[[I2:.*]]: i2,
 // CHECK-12N-SAME:                 %[[I3:.*]]: i3,
 // CHECK-12N-SAME:                 %[[I4:.*]]: i4,
 // CHECK-12N-SAME:                 %[[I5:.*]]: i5,
 // CHECK-12N-SAME:                 %[[I6:.*]]: i6) -> (i1, i2, i3, i4, i5, i6) {
 // CHECK-12N:           %[[CALL:.*]]:6 = call @callee(%[[I1]], %[[I2]], %[[I3]], %[[I4]], %[[I5]], %[[I6]]) : (i1, i2, i3, i4, i5, i6) -> (i1, i2, i3, i4, i5, i6)
 // CHECK-12N:           return %[[CALL]]#0, %[[CALL]]#1, %[[CALL]]#2, %[[CALL]]#3, %[[CALL]]#4, %[[CALL]]#5 : i1, i2, i3, i4, i5, i6
 func.func @caller(%arg0: tuple<>, %arg1: i1, %arg2: tuple<i2>, %arg3: i3, %arg4: tuple<i4, i5>, %arg5: i6) -> (tuple<>, i1, tuple<i2>, i3, tuple<i4, i5>, i6) {
   %0, %1, %2, %3, %4, %5 = call @callee(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5) : (tuple<>, i1, tuple<i2>, i3, tuple<i4, i5>, i6) -> (tuple<>, i1, tuple<i2>, i3, tuple<i4, i5>, i6)
   return %0, %1, %2, %3, %4, %5 : tuple<>, i1, tuple<i2>, i3, tuple<i4, i5>, i6
 }

mlir/test/Transforms/test-legalizer.mlir

 // RUN: mlir-opt -allow-unregistered-dialect -split-input-file -test-legalize-patterns -verify-diagnostics %s | FileCheck %s
 
 // CHECK-LABEL: verifyDirectPattern
 func.func @verifyDirectPattern() -> i32 {
-  // CHECK-NEXT:  "test.legal_op_a"() {status = "Success"}
+  // CHECK-NEXT:  "test.legal_op_a"() <{status = "Success"}
   %result = "test.illegal_op_a"() : () -> (i32)
   // expected-remark@+1 {{op 'func.return' is not legalizable}}
   return %result : i32
 }
 
 // CHECK-LABEL: verifyLargerBenefit
 func.func @verifyLargerBenefit() -> i32 {
-  // CHECK-NEXT:  "test.legal_op_a"() {status = "Success"}
+  // CHECK-NEXT:  "test.legal_op_a"() <{status = "Success"}
   %result = "test.illegal_op_c"() : () -> (i32)
   // expected-remark@+1 {{op 'func.return' is not legalizable}}
   return %result : i32
 }
 
 // CHECK-LABEL: func private @remap_input_1_to_0()
 func.func private @remap_input_1_to_0(i16)
 

mlir/test/Transforms/test-operation-folder.mlir

 // RUN: mlir-opt -test-patterns='top-down=false' %s | FileCheck %s
 // RUN: mlir-opt -test-patterns='top-down=true' %s | FileCheck %s
 
 func.func @foo() -> i32 {
   %c42 = arith.constant 42 : i32
 
   // The new operation should be present in the output and contain an attribute
   // with value "42" that results from folding.
 
-  // CHECK: "test.op_in_place_fold"(%{{.*}}) {attr = 42 : i32}
+  // CHECK: "test.op_in_place_fold"(%{{.*}}) <{attr = 42 : i32}
   %0 = "test.op_in_place_fold_anchor"(%c42) : (i32) -> (i32)
   return %0 : i32
 }
 
 func.func @test_fold_before_previously_folded_op() -> (i32, i32) {
   // When folding two constants will be generated and uniqued. Check that the
   // uniqued constant properly dominates both uses.
   // CHECK: %[[CST:.+]] = arith.constant true

mlir/test/lib/Dialect/Test/TestDialect.h

 #include "mlir/Interfaces/CopyOpInterface.h"
 #include "mlir/Interfaces/DerivedAttributeOpInterface.h"
 #include "mlir/Interfaces/InferIntRangeInterface.h"
 #include "mlir/Interfaces/InferTypeOpInterface.h"
 #include "mlir/Interfaces/LoopLikeInterface.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 #include "mlir/Interfaces/ViewLikeInterface.h"
 
+#include <memory>
+
 namespace mlir {
 class DLTIDialect;
 class RewritePatternSet;
 } // namespace mlir
 
 //===----------------------------------------------------------------------===//
 // TestDialect
 //===----------------------------------------------------------------------===//
 
 #include "TestOpInterfaces.h.inc"
 #include "TestOpsDialect.h.inc"
 
+namespace test {
+// Define some classes to exercises the Properties feature.
+
+struct PropertiesWithCustomPrint {
+  /// A shared_ptr to a const object is safe: it is equivalent to a value-based
+  /// member. Here the label will be deallocated when the last operation
+  /// refering to it is destroyed. However there is no pool-allocation: this is
+  /// offloaded to the client.
+  std::shared_ptr<const std::string> label;
+  int value;
+};
+class MyPropStruct {
+public:
+  std::string content;
+  // These three methods are invoked through the  `MyStructProperty` wrapper
+  // defined in TestOps.td
+  mlir::Attribute asAttribute(mlir::MLIRContext *ctx) const;
+  static mlir::LogicalResult setFromAttr(MyPropStruct &prop,
+                                         mlir::Attribute attr,
+                                         mlir::InFlightDiagnostic *diag);
+  llvm::hash_code hash() const;
+};
+} // namespace test
+
 #define GET_OP_CLASSES
 #include "TestOps.h.inc"
 
 namespace test {
 
 // Op deliberately defined in C++ code rather than ODS to test that C++
 // Ops can still use the old `fold` method.
 class ManualCppOpWithFold

mlir/test/lib/Dialect/Test/TestDialect.cpp

 #include "mlir/Dialect/Tensor/IR/Tensor.h"
 #include "mlir/IR/AsmState.h"
 #include "mlir/IR/BuiltinAttributes.h"
 #include "mlir/IR/BuiltinOps.h"
 #include "mlir/IR/Diagnostics.h"
 #include "mlir/IR/ExtensibleDialect.h"
 #include "mlir/IR/FunctionImplementation.h"
 #include "mlir/IR/MLIRContext.h"
+#include "mlir/IR/ODSSupport.h"
 #include "mlir/IR/OperationSupport.h"
 #include "mlir/IR/PatternMatch.h"
 #include "mlir/IR/TypeUtilities.h"
 #include "mlir/IR/Verifier.h"
 #include "mlir/Interfaces/InferIntRangeInterface.h"
 #include "mlir/Reducer/ReductionPatternInterface.h"
 #include "mlir/Transforms/FoldUtils.h"
 #include "mlir/Transforms/InliningUtils.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringSwitch.h"
 
 #include <numeric>
 #include <optional>
 
 // Include this before the using namespace lines below to
 // test that we don't have namespace dependencies.
 #include "TestOpsDialect.cpp.inc"
 
 using namespace mlir;
 using namespace test;
 
+Attribute MyPropStruct::asAttribute(MLIRContext *ctx) const {
+  return StringAttr::get(ctx, content);
+}
+LogicalResult MyPropStruct::setFromAttr(MyPropStruct &prop, Attribute attr,
+                                        InFlightDiagnostic *diag) {
+  StringAttr strAttr = attr.dyn_cast<StringAttr>();
+  if (!strAttr) {
+    if (diag)
+      *diag << "Expect StringAttr but got " << attr;
+    return failure();
+  }
+  prop.content = strAttr.getValue();
+  return success();
+}
+llvm::hash_code MyPropStruct::hash() const {
+  return hash_value(StringRef(content));
+}
+
+static LogicalResult setPropertiesFromAttribute(PropertiesWithCustomPrint &prop,
+                                                Attribute attr,
+                                                InFlightDiagnostic *diagnostic);
+static DictionaryAttr
+getPropertiesAsAttribute(MLIRContext *ctx,
+                         const PropertiesWithCustomPrint &prop);
+static llvm::hash_code computeHash(const PropertiesWithCustomPrint &prop);
+static void customPrintProperties(OpAsmPrinter &p,
+                                  const PropertiesWithCustomPrint &prop);
+static ParseResult customParseProperties(OpAsmParser &parser,
+                                         PropertiesWithCustomPrint &prop);
+
 void test::registerTestDialect(DialectRegistry &registry) {
   registry.insert<TestDialect>();
 }
 
 //===----------------------------------------------------------------------===//
 // TestDialect version utilities
 //===----------------------------------------------------------------------===//
 

 Operation *TestDialect::materializeConstant(OpBuilder &builder, Attribute value,
                                             Type type, Location loc) {
   return builder.create<TestOpConstant>(loc, type, value);
 }
 
 ::mlir::LogicalResult FormatInferType2Op::inferReturnTypes(
     ::mlir::MLIRContext *context, ::std::optional<::mlir::Location> location,
     ::mlir::ValueRange operands, ::mlir::DictionaryAttr attributes,
-    ::mlir::RegionRange regions,
+    OpaqueProperties properties, ::mlir::RegionRange regions,
     ::llvm::SmallVectorImpl<::mlir::Type> &inferredReturnTypes) {
   inferredReturnTypes.assign({::mlir::IntegerType::get(context, 16)});
   return ::mlir::success();
 }
 
 void *TestDialect::getRegisteredInterfaceForOp(TypeID typeID,
                                                OperationName opName) {
   if (opName.getIdentifier() == "test.unregistered_side_effect_op" &&
LogicalResult TestOpWithVariadicResultsAndFolder::fold(
     FoldAdaptor adaptor, SmallVectorImpl<OpFoldResult> &results) {
   for (Value input : this->getOperands()) {
     results.push_back(input);
   }
   return success();
 }
 
 OpFoldResult TestOpInPlaceFold::fold(FoldAdaptor adaptor) {
-  if (adaptor.getOp() && !(*this)->hasAttr("attr")) {
+  if (adaptor.getOp() && !(*this)->getAttr("attr")) {
     // The folder adds "attr" if not present.
     (*this)->setAttr("attr", adaptor.getOp());
     return getResult();
   }
   return {};
 }
 
 OpFoldResult TestPassthroughFold::fold(FoldAdaptor adaptor) {
OpFoldResult TestOpFoldWithFoldAdaptor::fold(FoldAdaptor adaptor) {
 
   sum += 4 * std::distance(adaptor.getBody().begin(), adaptor.getBody().end());
 
   return IntegerAttr::get(getType(), sum);
 }
 
 LogicalResult OpWithInferTypeInterfaceOp::inferReturnTypes(
     MLIRContext *, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &inferredReturnTypes) {
   if (operands[0].getType() != operands[1].getType()) {
     return emitOptionalError(location, "operand type mismatch ",
                              operands[0].getType(), " vs ",
                              operands[1].getType());
   }
   inferredReturnTypes.assign({operands[0].getType()});
   return success();
 }
 
 // TODO: We should be able to only define either inferReturnType or
 // refineReturnType, currently only refineReturnType can be omitted.
 LogicalResult OpWithRefineTypeInterfaceOp::inferReturnTypes(
     MLIRContext *context, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &returnTypes) {
   returnTypes.clear();
   return OpWithRefineTypeInterfaceOp::refineReturnTypes(
-      context, location, operands, attributes, regions, returnTypes);
+      context, location, operands, attributes, properties, regions,
+      returnTypes);
 }
 
 LogicalResult OpWithRefineTypeInterfaceOp::refineReturnTypes(
     MLIRContext *, std::optional<Location> location, ValueRange operands,
-    DictionaryAttr attributes, RegionRange regions,
+    DictionaryAttr attributes, OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<Type> &returnTypes) {
   if (operands[0].getType() != operands[1].getType()) {
     return emitOptionalError(location, "operand type mismatch ",
                              operands[0].getType(), " vs ",
                              operands[1].getType());
   }
   // TODO: Add helper to make this more concise to write.
   if (returnTypes.empty())
     returnTypes.resize(1, nullptr);
   if (returnTypes[0] && returnTypes[0] != operands[0].getType())
     return emitOptionalError(location,
                              "required first operand and result to match");
   returnTypes[0] = operands[0].getType();
   return success();
 }
 
 LogicalResult OpWithShapedTypeInferTypeInterfaceOp::inferReturnTypeComponents(
     MLIRContext *context, std::optional<Location> location,
-    ValueShapeRange operands, DictionaryAttr attributes, RegionRange regions,
+    ValueShapeRange operands, DictionaryAttr attributes,
+    OpaqueProperties properties, RegionRange regions,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
   // Create return type consisting of the last element of the first operand.
   auto operandType = operands.front().getType();
   auto sval = operandType.dyn_cast<ShapedType>();
   if (!sval) {
     return emitOptionalError(location, "only shaped type operands allowed");
   }
   int64_t dim = sval.hasRank() ? sval.getShape().front() : ShapedType::kDynamic;

 OpFoldResult ManualCppOpWithFold::fold(ArrayRef<Attribute> attributes) {
   // Just a simple fold for testing purposes that reads an operands constant
   // value and returns it.
   if (!attributes.empty())
     return attributes.front();
   return nullptr;
 }
 
+static LogicalResult
+setPropertiesFromAttribute(PropertiesWithCustomPrint &prop, Attribute attr,
+                           InFlightDiagnostic *diagnostic) {
+  DictionaryAttr dict = dyn_cast<DictionaryAttr>(attr);
+  if (!dict) {
+    if (diagnostic)
+      *diagnostic << "expected DictionaryAttr to set TestProperties";
+    return failure();
+  }
+  auto label = dict.getAs<mlir::StringAttr>("label");
+  if (!label) {
+    if (diagnostic)
+      *diagnostic << "expected StringAttr for key `label`";
+    return failure();
+  }
+  auto valueAttr = dict.getAs<IntegerAttr>("value");
+  if (!valueAttr) {
+    if (diagnostic)
+      *diagnostic << "expected IntegerAttr for key `value`";
+    return failure();
+  }
+
+  prop.label = std::make_shared<std::string>(label.getValue());
+  prop.value = valueAttr.getValue().getSExtValue();
+  return success();
+}
+static DictionaryAttr
+getPropertiesAsAttribute(MLIRContext *ctx,
+                         const PropertiesWithCustomPrint &prop) {
+  SmallVector<NamedAttribute> attrs;
+  Builder b{ctx};
+  attrs.push_back(b.getNamedAttr("label", b.getStringAttr(*prop.label)));
+  attrs.push_back(b.getNamedAttr("value", b.getI32IntegerAttr(prop.value)));
+  return b.getDictionaryAttr(attrs);
+}
+static llvm::hash_code computeHash(const PropertiesWithCustomPrint &prop) {
+  return llvm::hash_combine(prop.value, StringRef(*prop.label));
+}
+static void customPrintProperties(OpAsmPrinter &p,
+                                  const PropertiesWithCustomPrint &prop) {
+  p.printKeywordOrString(*prop.label);
+  p << " is " << prop.value;
+}
+static ParseResult customParseProperties(OpAsmParser &parser,
+                                         PropertiesWithCustomPrint &prop) {
+  std::string label;
+  if (parser.parseKeywordOrString(&label) || parser.parseKeyword("is") ||
+      parser.parseInteger(prop.value))
+    return failure();
+  prop.label = std::make_shared<std::string>(std::move(label));
+  return success();
+}
+
 #include "TestOpEnums.cpp.inc"
 #include "TestOpInterfaces.cpp.inc"
 #include "TestTypeInterfaces.cpp.inc"
 
 #define GET_OP_CLASSES
 #include "TestOps.cpp.inc"

mlir/test/lib/Dialect/Test/TestDialect.td

def Test_Dialect : Dialect {
   let hasOperationAttrVerify = 1;
   let hasRegionArgAttrVerify = 1;
   let hasRegionResultAttrVerify = 1;
   let hasOperationInterfaceFallback = 1;
   let hasNonDefaultDestructor = 1;
   let useDefaultTypePrinterParser = 0;
   let useDefaultAttributePrinterParser = 1;
   let isExtensible = 1;
+  let usePropertiesForAttributes = 1;
   let dependentDialects = ["::mlir::DLTIDialect"];
 
   let extraClassDeclaration = [{
     void registerAttributes();
     void registerTypes();
 
     // Provides a custom printing/parsing for some operations.
     ::std::optional<ParseOpHook>

mlir/test/lib/Dialect/Test/TestOps.td

 def VariadicRegionInferredTypesOp : TEST_Op<"variadic_region_inferred",
                                             [InferTypeOpInterface]> {
   let regions = (region VariadicRegion<AnyRegion>:$bodies);
   let results = (outs Variadic<AnyType>);
 
   let extraClassDeclaration = [{
     static mlir::LogicalResult inferReturnTypes(mlir::MLIRContext *context,
           std::optional<::mlir::Location> location, mlir::ValueRange operands,
-          mlir::DictionaryAttr attributes, mlir::RegionRange regions,
+          mlir::DictionaryAttr attributes, mlir::OpaqueProperties properties, mlir::RegionRange regions,
           llvm::SmallVectorImpl<mlir::Type> &inferredReturnTypes) {
       inferredReturnTypes.assign({mlir::IntegerType::get(context, 16)});
       return mlir::success();
     }
   }];
 }
 
 def OneRegionWithOperandsOp : TEST_Op<"one_region_with_operands_op", []> {

 
 def FormatInferTypeOp : TEST_Op<"format_infer_type", [InferTypeOpInterface]> {
   let results = (outs AnyType);
   let assemblyFormat = "attr-dict";
 
   let extraClassDeclaration = [{
     static ::mlir::LogicalResult inferReturnTypes(::mlir::MLIRContext *context,
           ::std::optional<::mlir::Location> location, ::mlir::ValueRange operands,
-          ::mlir::DictionaryAttr attributes, ::mlir::RegionRange regions,
+          ::mlir::DictionaryAttr attributes, mlir::OpaqueProperties properties, ::mlir::RegionRange regions,
           ::llvm::SmallVectorImpl<::mlir::Type> &inferredReturnTypes) {
       inferredReturnTypes.assign({::mlir::IntegerType::get(context, 16)});
       return ::mlir::success();
     }
    }];
 }
 
 // Check that formatget supports DeclareOpInterfaceMethods.
 def FormatInferType2Op : TEST_Op<"format_infer_type2", [DeclareOpInterfaceMethods<InferTypeOpInterface>]> {
   let results = (outs AnyType);
   let assemblyFormat = "attr-dict";
 }
 
 // Base class for testing mixing allOperandTypes, allOperands, and
 // inferResultTypes.
 class FormatInferAllTypesBaseOp<string mnemonic, list<Trait> traits = []>
     : TEST_Op<mnemonic, [InferTypeOpInterface] # traits> {
   let arguments = (ins Variadic<AnyType>:$args);
   let results = (outs Variadic<AnyType>:$outs);
   let extraClassDeclaration = [{
     static ::mlir::LogicalResult inferReturnTypes(::mlir::MLIRContext *context,
           ::std::optional<::mlir::Location> location, ::mlir::ValueRange operands,
-          ::mlir::DictionaryAttr attributes, ::mlir::RegionRange regions,
+          ::mlir::DictionaryAttr attributes, mlir::OpaqueProperties properties, ::mlir::RegionRange regions,
           ::llvm::SmallVectorImpl<::mlir::Type> &inferredReturnTypes) {
       ::mlir::TypeRange operandTypes = operands.getTypes();
       inferredReturnTypes.assign(operandTypes.begin(), operandTypes.end());
       return ::mlir::success();
     }
    }];
 }
 

 def FormatInferTypeRegionsOp
     : TEST_Op<"format_infer_type_regions", [InferTypeOpInterface]> {
   let results = (outs Variadic<AnyType>:$outs);
   let regions = (region AnyRegion:$region);
   let assemblyFormat = "$region attr-dict";
   let extraClassDeclaration = [{
     static ::mlir::LogicalResult inferReturnTypes(::mlir::MLIRContext *context,
           ::std::optional<::mlir::Location> location, ::mlir::ValueRange operands,
-          ::mlir::DictionaryAttr attributes, ::mlir::RegionRange regions,
+          ::mlir::DictionaryAttr attributes, mlir::OpaqueProperties properties, ::mlir::RegionRange regions,
           ::llvm::SmallVectorImpl<::mlir::Type> &inferredReturnTypes) {
       if (regions.empty())
         return ::mlir::failure();
       auto types = regions.front()->getArgumentTypes();
       inferredReturnTypes.assign(types.begin(), types.end());
       return ::mlir::success();
     }
   }];
 }
 
 // Test inferReturnTypes coupled with variadic operands (operand_segment_sizes).
 def FormatInferTypeVariadicOperandsOp
     : TEST_Op<"format_infer_type_variadic_operands",
               [InferTypeOpInterface, AttrSizedOperandSegments]> {
   let arguments = (ins Variadic<I32>:$a, Variadic<I64>:$b);
   let results = (outs Variadic<AnyType>:$outs);
   let assemblyFormat = "`(` $a `:` type($a) `)` `(` $b `:` type($b) `)` attr-dict";
   let extraClassDeclaration = [{
     static ::mlir::LogicalResult inferReturnTypes(::mlir::MLIRContext *context,
           ::std::optional<::mlir::Location> location, ::mlir::ValueRange operands,
-          ::mlir::DictionaryAttr attributes, ::mlir::RegionRange regions,
+          ::mlir::DictionaryAttr attributes, mlir::OpaqueProperties properties, ::mlir::RegionRange regions,
           ::llvm::SmallVectorImpl<::mlir::Type> &inferredReturnTypes) {
-      FormatInferTypeVariadicOperandsOpAdaptor adaptor(operands, attributes);
+      FormatInferTypeVariadicOperandsOpAdaptor adaptor(
+          operands, attributes, *properties.as<Properties *>(), {});
       auto aTypes = adaptor.getA().getTypes();
       auto bTypes = adaptor.getB().getTypes();
       inferredReturnTypes.append(aTypes.begin(), aTypes.end());
       inferredReturnTypes.append(bTypes.begin(), bTypes.end());
       return ::mlir::success();
     }
   }];
 }
class TableGenBuildInferReturnTypeBaseOp<string mnemonic,
                                          list<Trait> traits = []>
     : TEST_Op<mnemonic, [InferTypeOpInterface] # traits> {
   let arguments = (ins Variadic<AnyType>:$inputs);
   let results = (outs AnyType:$result);
 
   let extraClassDeclaration = [{
     static ::mlir::LogicalResult inferReturnTypes(::mlir::MLIRContext *,
           ::std::optional<::mlir::Location> location, ::mlir::ValueRange operands,
-          ::mlir::DictionaryAttr attributes, ::mlir::RegionRange regions,
+          ::mlir::DictionaryAttr attributes, mlir::OpaqueProperties properties, ::mlir::RegionRange regions,
           ::llvm::SmallVectorImpl<::mlir::Type> &inferredReturnTypes) {
       inferredReturnTypes.assign({operands[0].getType()});
       return ::mlir::success();
     }
    }];
 }
 
 // Op with InferTypeOpInterface and regions.
def TestVersionedOpB : TEST_Op<"versionedB"> {
   //
   // We support loading old IR through a custom readAttribute method, see
   // `readAttribute()` in `TestBytecodeDialectInterface`
   let arguments = (ins
     TestAttrParams:$attribute
   );
 }
 
+//===----------------------------------------------------------------------===//
+// Test Properties
+//===----------------------------------------------------------------------===//
+
+
+// Op with a properties struct defined inline.
+def TestOpWithProperties : TEST_Op<"with_properties"> {
+  let assemblyFormat = "prop-dict attr-dict";
+  let arguments = (ins
+    Property<"int64_t">:$a,
+    StrAttr:$b, // Attributes can directly be used here.
+    ArrayProperty<"int64_t", 4>:$array // example of an array
+  );
+}
+
+// Demonstrate how to wrap an existing C++ class named MyPropStruct.
+def MyStructProperty : Property<"MyPropStruct"> {
+  let convertToAttribute = "$_storage.asAttribute($_ctxt)";
+  let convertFromAttribute = "return MyPropStruct::setFromAttr($_storage, $_attr, $_diag);";
+  let hashProperty = "$_storage.hash();";
+}
+
+def TestOpWithWrappedProperties : TEST_Op<"with_wrapped_properties"> {
+  let assemblyFormat = "prop-dict attr-dict";
+  let arguments = (ins
+    MyStructProperty:$prop
+  );
+}
+
+// Op with a properties struct defined out-of-line. The struct has custom
+// printer/parser.
+
+def PropertiesWithCustomPrint : Property<"PropertiesWithCustomPrint"> {
+  let convertToAttribute = [{
+    getPropertiesAsAttribute($_ctxt, $_storage)
+  }];
+  let convertFromAttribute = [{
+    return setPropertiesFromAttribute($_storage, $_attr, $_diag);
+  }];
+  let hashProperty = [{
+    computeHash($_storage);
+  }];
+}
+
+def TestOpWithNiceProperties : TEST_Op<"with_nice_properties"> {
+  let assemblyFormat = "prop-dict attr-dict";
+  let arguments = (ins
+     PropertiesWithCustomPrint:$prop
+  );
+  let extraClassDeclaration = [{
+    void printProperties(::mlir::MLIRContext *ctx, ::mlir::OpAsmPrinter &p,
+                         const Properties &prop);
+    static ::mlir::ParseResult  parseProperties(::mlir::OpAsmParser &parser,
+                                     ::mlir::OperationState &result);
+  }];
+  let extraClassDefinition = [{
+    void TestOpWithNiceProperties::printProperties(::mlir::MLIRContext *ctx,
+            ::mlir::OpAsmPrinter &p, const Properties &prop) {
+      customPrintProperties(p, prop.prop);
+    }
+    ::mlir::ParseResult TestOpWithNiceProperties::parseProperties(
+        ::mlir::OpAsmParser &parser,
+        ::mlir::OperationState &result) {
+      Properties &prop = result.getOrAddProperties<Properties>();
+      if (customParseProperties(parser, prop.prop))
+        return failure();
+      return success();
+    }
+  }];
+}
+
+
+
 #endif // TEST_OPS

mlir/test/lib/Dialect/Test/TestPatterns.cpp

static void invokeCreateWithInferredReturnType(Operation *op) {
   // Use permutations of 2 args as operands.
   assert(fop.getNumArguments() >= 2);
   for (int i = 0, e = fop.getNumArguments(); i < e; ++i) {
     for (int j = 0; j < e; ++j) {
       std::array<Value, 2> values = {{fop.getArgument(i), fop.getArgument(j)}};
       SmallVector<Type, 2> inferredReturnTypes;
       if (succeeded(OpTy::inferReturnTypes(
               context, std::nullopt, values, op->getAttrDictionary(),
-              op->getRegions(), inferredReturnTypes))) {
+              op->getPropertiesStorage(), op->getRegions(),
+              inferredReturnTypes))) {
         OperationState state(location, OpTy::getOperationName());
         // TODO: Expand to regions.
         OpTy::build(b, state, values, op->getAttrs());
         (void)b.create(state);
       }
     }
   }
 }

mlir/test/mlir-tblgen/constraint-unique.td

 /// Test that a type constraint with a different description was generated.
 // CHECK:    static ::mlir::LogicalResult [[$O_TYPE_CONSTRAINT:__mlir_ods_local_type_constraint.*]](
 // CHECK:      if (!((typePred(type, *op)))) {
 // CHECK-NEXT:   return op->emitOpError(valueKind) << " #" << valueIndex
 // CHECK-NEXT:       << " must be another type, but got " << type;
 
 /// Test that an attribute contraint was generated.
 // CHECK:    static ::mlir::LogicalResult [[$A_ATTR_CONSTRAINT:__mlir_ods_local_attr_constraint.*]](
-// CHECK:      if (attr && !((attrPred(attr, *op)))) {
-// CHECK-NEXT:   return op->emitOpError("attribute '") << attrName
+// CHECK:      if (attr && !((attrPred(attr, *op))))
+// CHECK-NEXT:   return getDiag() << "attribute '" << attrName
 // CHECK-NEXT:       << "' failed to satisfy constraint: an attribute";
 
 /// Test that duplicate attribute constraint was not generated.
 // CHECK-NOT:        << "' failed to satisfy constraint: an attribute";
 
 /// Test that a attribute constraint with a different description was generated.
 // CHECK:    static ::mlir::LogicalResult [[$O_ATTR_CONSTRAINT:__mlir_ods_local_attr_constraint.*]](
-// CHECK:      if (attr && !((attrPred(attr, *op)))) {
-// CHECK-NEXT:   return op->emitOpError("attribute '") << attrName
+// CHECK:    static ::mlir::LogicalResult [[$O_ATTR_CONSTRAINT:__mlir_ods_local_attr_constraint.*]](
+// CHECK:      if (attr && !((attrPred(attr, *op))))
+// CHECK-NEXT:   return getDiag() << "attribute '" << attrName
 // CHECK-NEXT:       << "' failed to satisfy constraint: another attribute";
 
 /// Test that a successor contraint was generated.
 // CHECK:    static ::mlir::LogicalResult [[$A_SUCCESSOR_CONSTRAINT:__mlir_ods_local_successor_constraint.*]](
 // CHECK:      if (!((successorPred(successor, *op)))) {
 // CHECK-NEXT:   return op->emitOpError("successor #") << successorIndex << " ('"
 // CHECK-NEXT:       << successorName << ")' failed to verify constraint: a successor";
 

mlir/test/mlir-tblgen/interfaces-as-constraints.td

 // CHECK-NEXT:        << " must be TopLevelTypeInterface instance, but got " << type;
 
 // CHECK: static ::mlir::LogicalResult {{__mlir_ods_local_type_constraint.*}}(
 // CHECK:   if (!((type.isa<test::TypeInterfaceInNamespace>()))) {
 // CHECK-NEXT:    return op->emitOpError(valueKind) << " #" << valueIndex
 // CHECK-NEXT:        << " must be TypeInterfaceInNamespace instance, but got " << type;
 
 // CHECK: static ::mlir::LogicalResult {{__mlir_ods_local_attr_constraint.*}}(
-// CHECK:   if (attr && !((attr.isa<TopLevelAttrInterface>()))) {
-// CHECK-NEXT:    return op->emitOpError("attribute '") << attrName
+// CHECK:   if (attr && !((attr.isa<TopLevelAttrInterface>())))
+// CHECK-NEXT:     return getDiag() << "attribute '" << attrName
 // CHECK-NEXT:        << "' failed to satisfy constraint: TopLevelAttrInterface instance";
 
 // CHECK: static ::mlir::LogicalResult {{__mlir_ods_local_attr_constraint.*}}(
-// CHECK:   if (attr && !((attr.isa<test::AttrInterfaceInNamespace>()))) {
-// CHECK-NEXT:    return op->emitOpError("attribute '") << attrName
+// CHECK:   if (attr && !((attr.isa<test::AttrInterfaceInNamespace>())))
+// CHECK-NEXT:    return getDiag() << "attribute '" << attrName
 // CHECK-NEXT:        << "' failed to satisfy constraint: AttrInterfaceInNamespace instance";
 
 // CHECK: TopLevelAttrInterface OpUsingAllOfThose::getAttr1()
 // CHECK: test::AttrInterfaceInNamespace OpUsingAllOfThose::getAttr2()

mlir/test/mlir-tblgen/op-attribute.td

 
 // DEF-LABEL: AOp definitions
 
 // Test verify method
 // ---
 
 // DEF:      ::mlir::LogicalResult AOpAdaptor::verify
 // DEF:      ::mlir::Attribute tblgen_aAttr;
-// DEF-NEXT: while (true) {
+// DEF: while (true) {
 // DEF-NEXT:   if (namedAttrIt == namedAttrRange.end())
 // DEF-NEXT:     return emitError(loc, "'test.a_op' op ""requires attribute 'aAttr'");
 // DEF-NEXT:   if (namedAttrIt->getName() == AOp::getAAttrAttrName(*odsOpName)) {
 // DEF-NEXT:     tblgen_aAttr = namedAttrIt->getValue();
 // DEF-NEXT:     break;
 // DEF:      ::mlir::Attribute tblgen_bAttr;
 // DEF-NEXT: ::mlir::Attribute tblgen_cAttr;
 // DEF-NEXT: ::mlir::Attribute tblgen_dAttr;

 
 // DEF-LABEL: AgetOp definitions
 
 // Test verify method
 // ---
 
 // DEF:      ::mlir::LogicalResult AgetOpAdaptor::verify
 // DEF:      ::mlir::Attribute tblgen_aAttr;
-// DEF-NEXT: while (true)
+// DEF: while (true)
 // DEF:      ::mlir::Attribute tblgen_bAttr;
 // DEF-NEXT: ::mlir::Attribute tblgen_cAttr;
-// DEF-NEXT: while (true)
+// DEF: while (true)
 // DEF:      if (tblgen_aAttr && !((some-condition)))
 // DEF-NEXT:   return emitError(loc, "'test2.a_get_op' op ""attribute 'aAttr' failed to satisfy constraint: some attribute kind");
 // DEF:      if (tblgen_bAttr && !((some-condition)))
 // DEF-NEXT:   return emitError(loc, "'test2.a_get_op' op ""attribute 'bAttr' failed to satisfy constraint: some attribute kind");
 // DEF:      if (tblgen_cAttr && !((some-condition)))
 // DEF-NEXT:   return emitError(loc, "'test2.a_get_op' op ""attribute 'cAttr' failed to satisfy constraint: some attribute kind");
 
 // Test getter methods

mlir/test/mlir-tblgen/op-decl-and-defs.td

 // CHECK:   static void getCanonicalizationPatterns(::mlir::RewritePatternSet &results, ::mlir::MLIRContext *context);
 // CHECK:   ::mlir::LogicalResult fold(FoldAdaptor adaptor, ::llvm::SmallVectorImpl<::mlir::OpFoldResult> &results);
 // CHECK:   // Display a graph for debugging purposes.
 // CHECK:   void displayGraph();
 // CHECK: };
 
 // DEFS-LABEL: NS::AOp definitions
 
-// DEFS: AOpGenericAdaptorBase::AOpGenericAdaptorBase(::mlir::DictionaryAttr attrs, ::mlir::RegionRange regions) : odsAttrs(attrs), odsRegions(regions)
+// DEFS: AOpGenericAdaptorBase::AOpGenericAdaptorBase(::mlir::DictionaryAttr attrs, ::mlir::EmptyProperties properties, ::mlir::RegionRange regions) : odsAttrs(attrs), odsRegions(regions)
 // DEFS: ::mlir::RegionRange AOpGenericAdaptorBase::getSomeRegions()
 // DEFS-NEXT: return odsRegions.drop_front(1);
 // DEFS: ::mlir::RegionRange AOpGenericAdaptorBase::getRegions()
 
 // Check AttrSizedOperandSegments
 // ---
 
 def NS_AttrSizedOperandOp : NS_Op<"attr_sized_operands",

mlir/test/mlir-tblgen/op-format.td

 }]>, Arguments<(ins Optional<I1>:$a)>;
 
 // CHECK-LABEL: OptionalGroupB::parse
 // CHECK: if (::mlir::succeeded(parser.parseOptionalKeyword("foo")))
 // CHECK-NEXT: else
 // CHECK-NEXT: result.addAttribute("a", parser.getBuilder().getUnitAttr())
 // CHECK: parser.parseKeyword("bar")
 // CHECK-LABEL: OptionalGroupB::print
-// CHECK: if (!(*this)->getAttr("a"))
+// CHECK: if (!getAAttr())
 // CHECK-NEXT: odsPrinter << ' ' << "foo"
 // CHECK-NEXT: else
 // CHECK-NEXT: odsPrinter << ' ' << "bar"
 def OptionalGroupB : TestFormat_Op<[{
   (`foo`) : (`bar` $a^)? attr-dict
 }]>, Arguments<(ins UnitAttr:$a)>;
 
 // Optional group anchored on a default-valued attribute:
 // CHECK-LABEL: OptionalGroupC::parse
-//       CHECK: if ((*this)->getAttr("a") != ::mlir::OpBuilder((*this)->getContext()).getStringAttr("default")) {
+
+//       CHECK: if (getAAttr() && getAAttr() != ::mlir::OpBuilder((*this)->getContext()).getStringAttr("default")) {
 //  CHECK-NEXT:   odsPrinter << ' ';
 //  CHECK-NEXT:   odsPrinter.printAttributeWithoutType(getAAttr());
 //  CHECK-NEXT: }
 def OptionalGroupC : TestFormat_Op<[{
   ($a^)? attr-dict
 }]>, Arguments<(ins DefaultValuedStrAttr<StrAttr, "default">:$a)>;

mlir/test/mlir-tblgen/op-result.td

 def OpL3 : NS_Op<"op_with_all_types_constraint",
     [AllTypesMatch<["a", "b"]>]> {
   let arguments = (ins I32Attr:$a);
   let results = (outs AnyType:$b);
 }
 
 // CHECK-LABEL: LogicalResult OpL3::inferReturnTypes
 // CHECK-NOT: }
-// CHECK: ::mlir::Type odsInferredType0 = attributes.get("a").cast<::mlir::TypedAttr>().getType();
+// CHECK: ::mlir::Type odsInferredType0 = odsInferredTypeAttr0.getType();
 // CHECK: inferredReturnTypes[0] = odsInferredType0;
 
 def OpL4 : NS_Op<"two_inference_edges", [
     TypesMatchWith<"", "a", "b", "infer0($_self)">,
     TypesMatchWith<"", "b", "c", "infer1($_self)">,
     TypesMatchWith<"", "input", "a", "fromInput($_self)">]> {
   let arguments = (ins I32:$input);
   let results = (outs AnyType:$a, AnyType:$b, AnyType:$c);

mlir/test/mlir-tblgen/pattern.mlir

 // RUN: mlir-opt -test-patterns -mlir-print-debuginfo -mlir-print-local-scope %s | FileCheck %s
 
 // CHECK-LABEL: verifyFusedLocs
 func.func @verifyFusedLocs(%arg0 : i32) -> i32 {
   %0 = "test.op_a"(%arg0) {attr = 10 : i32} : (i32) -> i32 loc("a")
   %result = "test.op_a"(%0) {attr = 20 : i32} : (i32) -> i32 loc("b")
 
-  // CHECK: "test.op_b"(%arg0) {attr = 10 : i32} : (i32) -> i32 loc("a")
-  // CHECK: "test.op_b"(%arg0) {attr = 20 : i32} : (i32) -> i32 loc(fused["b", "a"])
+  // CHECK: "test.op_b"(%arg0) <{attr = 10 : i32}> : (i32) -> i32 loc("a")
+  // CHECK: "test.op_b"(%arg0) <{attr = 20 : i32}> : (i32) -> i32 loc(fused["b", "a"])
   return %result : i32
 }
 
 // CHECK-LABEL: verifyDesignatedLoc
 func.func @verifyDesignatedLoc(%arg0 : i32) -> i32 {
   %0 = "test.loc_src"(%arg0) : (i32) -> i32 loc("loc3")
   %1 = "test.loc_src"(%0) : (i32) -> i32 loc("loc2")
   %2 = "test.loc_src"(%1) : (i32) -> i32 loc("loc1")
func.func @verifyZeroArg() -> i32 {
   // CHECK: "test.op_k"() : () -> i32
   %0 = "test.op_j"() : () -> i32
   return %0 : i32
 }
 
 // CHECK-LABEL: testIgnoreArgMatch
 // CHECK-SAME: (%{{[a-z0-9]*}}: i32 loc({{[^)]*}}), %[[ARG1:[a-z0-9]*]]: i32 loc({{[^)]*}}),
 func.func @testIgnoreArgMatch(%arg0: i32, %arg1: i32, %arg2: i32, %arg3: f32) {
-  // CHECK: "test.ignore_arg_match_dst"(%[[ARG1]]) {f = 15 : i64}
+  // CHECK: "test.ignore_arg_match_dst"(%[[ARG1]]) <{f = 15 : i64}>
   "test.ignore_arg_match_src"(%arg0, %arg1, %arg2) {d = 42, e = 24, f = 15} : (i32, i32, i32) -> ()
 
   // CHECK: test.ignore_arg_match_src
   // Not match because wrong type for $c.
   "test.ignore_arg_match_src"(%arg0, %arg1, %arg3) {d = 42, e = 24, f = 15} : (i32, i32, f32) -> ()
 
   // CHECK: test.ignore_arg_match_src
   // Not match because wrong type for $f.
   "test.ignore_arg_match_src"(%arg0, %arg1, %arg2) {d = 42 : i32, e = 24, f = 15} : (i32, i32, i32) -> ()
   return
 }
 
 // CHECK-LABEL: verifyInterleavedOperandAttribute
 // CHECK-SAME:    %[[ARG0:.*]]: i32 loc({{[^)]*}}), %[[ARG1:.*]]: i32 loc({{[^)]*}})
 func.func @verifyInterleavedOperandAttribute(%arg0: i32, %arg1: i32) {
-  // CHECK: "test.interleaved_operand_attr2"(%[[ARG0]], %[[ARG1]]) {attr1 = 15 : i64, attr2 = 42 : i64}
+  // CHECK: "test.interleaved_operand_attr2"(%[[ARG0]], %[[ARG1]]) <{attr1 = 15 : i64, attr2 = 42 : i64}>
   "test.interleaved_operand_attr1"(%arg0, %arg1) {attr1 = 15, attr2 = 42} : (i32, i32) -> ()
   return
 }
 
 // CHECK-LABEL: verifyBenefit
 func.func @verifyBenefit(%arg0 : i32) -> i32 {
   %0 = "test.op_d"(%arg0) : (i32) -> i32
   %1 = "test.op_g"(%arg0) : (i32) -> i32
   %2 = "test.op_g"(%1) : (i32) -> i32
 
   // CHECK: "test.op_f"(%arg0)
-  // CHECK: "test.op_b"(%arg0) {attr = 34 : i32}
+  // CHECK: "test.op_b"(%arg0) <{attr = 34 : i32}>
   return %0 : i32
 }
 
 // CHECK-LABEL: verifyNativeCodeCall
 func.func @verifyNativeCodeCall(%arg0: i32, %arg1: i32) -> (i32, i32) {
-  // CHECK: %0 = "test.native_code_call2"(%arg0) {attr = [42, 24]} : (i32) -> i32
+  // CHECK: %0 = "test.native_code_call2"(%arg0) <{attr = [42, 24]}> : (i32) -> i32
   // CHECK:  return %0, %arg1
   %0 = "test.native_code_call1"(%arg0, %arg1) {choice = true, attr1 = 42, attr2 = 24} : (i32, i32) -> (i32)
   %1 = "test.native_code_call1"(%arg0, %arg1) {choice = false, attr1 = 42, attr2 = 24} : (i32, i32) -> (i32)
   return %0, %1: i32, i32
 }
 
 // CHECK-LABEL: verifyAuxiliaryNativeCodeCall
 func.func @verifyAuxiliaryNativeCodeCall(%arg0: i32) -> (i32) {

 // Test Symbol Binding
 //===----------------------------------------------------------------------===//
 
 // CHECK-LABEL: symbolBinding
 func.func @symbolBinding(%arg0: i32) -> i32 {
   // An op with one use is matched.
   // CHECK: %0 = "test.symbol_binding_b"(%arg0)
   // CHECK: %1 = "test.symbol_binding_c"(%0)
-  // CHECK: %2 = "test.symbol_binding_d"(%0, %1) {attr = 42 : i64}
+  // CHECK: %2 = "test.symbol_binding_d"(%0, %1) <{attr = 42 : i64}>
   %0 = "test.symbol_binding_a"(%arg0) {attr = 42} : (i32) -> (i32)
 
   // An op without any use is not matched.
   // CHECK: "test.symbol_binding_a"(%arg0)
   %1 = "test.symbol_binding_a"(%arg0) {attr = 42} : (i32) -> (i32)
 
   // CHECK: return %2
   return %0: i32
 }
 
 // CHECK-LABEL: symbolBindingNoResult
 func.func @symbolBindingNoResult(%arg0: i32) {
   // CHECK: test.symbol_binding_b
   "test.symbol_binding_no_result"(%arg0) : (i32) -> ()
   return
 }
 
 //===----------------------------------------------------------------------===//
 // Test Attributes
 //===----------------------------------------------------------------------===//
 
 // CHECK-LABEL: succeedMatchOpAttr
 func.func @succeedMatchOpAttr() -> i32 {
-  // CHECK: "test.match_op_attribute2"() {default_valued_attr = 3 : i32, more_attr = 4 : i32, optional_attr = 2 : i32, required_attr = 1 : i32}
+  // CHECK: "test.match_op_attribute2"() <{default_valued_attr = 3 : i32, more_attr = 4 : i32, optional_attr = 2 : i32, required_attr = 1 : i32}>
   %0 = "test.match_op_attribute1"() {required_attr = 1: i32, optional_attr = 2: i32, default_valued_attr = 3: i32, more_attr = 4: i32} : () -> (i32)
   return %0: i32
 }
 
 // CHECK-LABEL: succeedMatchMissingOptionalAttr
 func.func @succeedMatchMissingOptionalAttr() -> i32 {
-  // CHECK: "test.match_op_attribute2"() {default_valued_attr = 3 : i32, more_attr = 4 : i32, required_attr = 1 : i32}
+  // CHECK: "test.match_op_attribute2"() <{default_valued_attr = 3 : i32, more_attr = 4 : i32, required_attr = 1 : i32}>
   %0 = "test.match_op_attribute1"() {required_attr = 1: i32, default_valued_attr = 3: i32, more_attr = 4: i32} : () -> (i32)
   return %0: i32
 }
 
 // CHECK-LABEL: succeedMatchMissingDefaultValuedAttr
 func.func @succeedMatchMissingDefaultValuedAttr() -> i32 {
-  // CHECK: "test.match_op_attribute2"() {default_valued_attr = 42 : i32, more_attr = 4 : i32, optional_attr = 2 : i32, required_attr = 1 : i32}
+  // CHECK: "test.match_op_attribute2"() <{default_valued_attr = 42 : i32, more_attr = 4 : i32, optional_attr = 2 : i32, required_attr = 1 : i32}>
   %0 = "test.match_op_attribute1"() {required_attr = 1: i32, optional_attr = 2: i32, more_attr = 4: i32} : () -> (i32)
   return %0: i32
 }
 
 // CHECK-LABEL: failedMatchAdditionalConstraintNotSatisfied
 func.func @failedMatchAdditionalConstraintNotSatisfied() -> i32 {
   // CHECK: "test.match_op_attribute1"()
   %0 = "test.match_op_attribute1"() {required_attr = 1: i32, optional_attr = 2: i32, more_attr = 5: i32} : () -> (i32)
   return %0: i32
 }
 
 // CHECK-LABEL: verifyConstantAttr
 func.func @verifyConstantAttr(%arg0 : i32) -> i32 {
-  // CHECK: "test.op_b"(%arg0) {attr = 17 : i32} : (i32) -> i32 loc("a")
+  // CHECK: "test.op_b"(%arg0) <{attr = 17 : i32}> : (i32) -> i32 loc("a")
   %0 = "test.op_c"(%arg0) : (i32) -> i32 loc("a")
   return %0 : i32
 }
 
 // CHECK-LABEL: verifyUnitAttr
 func.func @verifyUnitAttr() -> (i32, i32) {
   // Unit attribute present in the matched op is propagated as attr2.
-  // CHECK: "test.match_op_attribute4"() {attr1, attr2} : () -> i32
+  // CHECK: "test.match_op_attribute4"() <{attr1, attr2}> : () -> i32
   %0 = "test.match_op_attribute3"() {attr} : () -> i32
 
   // Since the original op doesn't have the unit attribute, the new op
   // only has the constant-constructed unit attribute attr1.
-  // CHECK: "test.match_op_attribute4"() {attr1} : () -> i32
+  // CHECK: "test.match_op_attribute4"() <{attr1}> : () -> i32
   %1 = "test.match_op_attribute3"() : () -> i32
   return %0, %1 : i32, i32
 }
 
 //===----------------------------------------------------------------------===//
 // Test Constant Matching
 //===----------------------------------------------------------------------===//
 
 // CHECK-LABEL: testConstOp
 func.func @testConstOp() -> (i32) {
-  // CHECK-NEXT: [[C0:%.+]] = "test.constant"() {value = 1
+  // CHECK-NEXT: [[C0:%.+]] = "test.constant"() <{value = 1
   %0 = "test.constant"() {value = 1 : i32} : () -> i32
 
   // CHECK-NEXT: return [[C0]]
   return %0 : i32
 }
 
 // CHECK-LABEL: testConstOpUsed
 func.func @testConstOpUsed() -> (i32) {
-  // CHECK-NEXT: [[C0:%.+]] = "test.constant"() {value = 1
+  // CHECK-NEXT: [[C0:%.+]] = "test.constant"() <{value = 1
   %0 = "test.constant"() {value = 1 : i32} : () -> i32
 
   // CHECK-NEXT: [[V0:%.+]] = "test.op_s"([[C0]])
   %1 = "test.op_s"(%0) {value = 1 : i32} : (i32) -> i32
 
   // CHECK-NEXT: return [[V0]]
   return %1 : i32
 }
 
 // CHECK-LABEL: testConstOpReplaced
 func.func @testConstOpReplaced() -> (i32) {
-  // CHECK-NEXT: [[C0:%.+]] = "test.constant"() {value = 1
+  // CHECK-NEXT: [[C0:%.+]] = "test.constant"() <{value = 1
   %0 = "test.constant"() {value = 1 : i32} : () -> i32
   %1 = "test.constant"() {value = 2 : i32} : () -> i32
 
-  // CHECK: [[V0:%.+]] = "test.op_s"([[C0]]) {value = 2 : i32}
+  // CHECK: [[V0:%.+]] = "test.op_s"([[C0]]) <{value = 2 : i32}
   %2 = "test.op_r"(%0, %1) : (i32, i32) -> i32
 
   // CHECK: [[V0]]
   return %2 : i32
 }
 
 // CHECK-LABEL: testConstOpMatchFailure
 func.func @testConstOpMatchFailure() -> (i64) {
-  // CHECK-DAG: [[C0:%.+]] = "test.constant"() {value = 1
+  // CHECK-DAG: [[C0:%.+]] = "test.constant"() <{value = 1
   %0 = "test.constant"() {value = 1 : i64} : () -> i64
 
-  // CHECK-DAG: [[C1:%.+]] = "test.constant"() {value = 2
+  // CHECK-DAG: [[C1:%.+]] = "test.constant"() <{value = 2
   %1 = "test.constant"() {value = 2 : i64} : () -> i64
 
   // CHECK: [[V0:%.+]] = "test.op_r"([[C0]], [[C1]])
   %2 = "test.op_r"(%0, %1) : (i64, i64) -> i64
 
   // CHECK: [[V0]]
   return %2 : i64
 }
 
 // CHECK-LABEL: testConstOpMatchNonConst
 func.func @testConstOpMatchNonConst(%arg0 : i32) -> (i32) {
-  // CHECK-DAG: [[C0:%.+]] = "test.constant"() {value = 1
+  // CHECK-DAG: [[C0:%.+]] = "test.constant"() <{value = 1
   %0 = "test.constant"() {value = 1 : i32} : () -> i32
 
   // CHECK: [[V0:%.+]] = "test.op_r"([[C0]], %arg0)
   %1 = "test.op_r"(%0, %arg0) : (i32, i32) -> i32
 
   // CHECK: [[V0]]
   return %1 : i32
 }
 
 
 
 //===----------------------------------------------------------------------===//
 // Test Enum Attributes
 //===----------------------------------------------------------------------===//
 
 // CHECK-LABEL: verifyI32EnumAttr
 func.func @verifyI32EnumAttr() -> i32 {
-  // CHECK: "test.i32_enum_attr"() {attr = 10 : i32}
+  // CHECK: "test.i32_enum_attr"() <{attr = 10 : i32}
   %0 = "test.i32_enum_attr"() {attr = 5: i32} : () -> i32
   return %0 : i32
 }
 
 // CHECK-LABEL: verifyI64EnumAttr
 func.func @verifyI64EnumAttr() -> i32 {
-  // CHECK: "test.i64_enum_attr"() {attr = 10 : i64}
+  // CHECK: "test.i64_enum_attr"() <{attr = 10 : i64}
   %0 = "test.i64_enum_attr"() {attr = 5: i64} : () -> i32
   return %0 : i32
 }
 
 //===----------------------------------------------------------------------===//
 // Test ElementsAttr
 //===----------------------------------------------------------------------===//
 

 
 //===----------------------------------------------------------------------===//
 // Test that natives calls are only called once during rewrites.
 //===----------------------------------------------------------------------===//
 
 // CHECK-LABEL: redundantTest
 func.func @redundantTest(%arg0: i32) -> i32 {
   %0 = "test.op_m"(%arg0) : (i32) -> i32
-  // CHECK: "test.op_m"(%arg0) {optional_attr = 314159265 : i32} : (i32) -> i32
+  // CHECK: "test.op_m"(%arg0) <{optional_attr = 314159265 : i32}> : (i32) -> i32
   return %0 : i32
 }
 
 //===----------------------------------------------------------------------===//
 // Test either directive
 //===----------------------------------------------------------------------===//
 
 // CHECK: @either_dag_leaf_only

mlir/test/mlir-tblgen/return-types.mlir

 // CHECK: "test.op_with_infer_type_if"
 // CHECK-SAME: (tensor<20xi32>, tensor<20xi32>) -> tensor<20xi32>
   return
 }
 
 // -----
 
 func.func @testReturnTypeOpInterface(%arg0 : tensor<10xf32>) {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{incompatible with return type}}
   %bad = "test.op_with_infer_type_if"(%arg0, %arg0) : (tensor<10xf32>, tensor<10xf32>) -> tensor<*xf32>
   return
 }
 
 // -----
 
 func.func @testReturnTypeOpInterfaceMismatch(%arg0 : tensor<10xf32>, %arg1 : tensor<20xf32>) {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{operand type mismatch}}
   %bad = "test.op_with_infer_type_if"(%arg0, %arg1) : (tensor<10xf32>, tensor<20xf32>) -> tensor<*xf32>
   return
 }
 
 // -----
 
 func.func @testReturnTypeOpInterface(%arg0 : tensor<10xf32>) {
+  // expected-error@+2 {{failed to infer returned types}}
   // expected-error@+1 {{required first operand and result to match}}
   %bad = "test.op_with_refine_type_if"(%arg0, %arg0) : (tensor<10xf32>, tensor<10xf32>) -> tensor<*xf32>
   return
 }
 
 // -----
 // CHECK-LABEL: testReifyFunctions
 func.func @testReifyFunctions(%arg0 : tensor<10xf32>, %arg1 : tensor<20xf32>) {
   // expected-remark@+1 {{arith.constant 10}}
   %0 = "test.op_with_shaped_type_infer_type_if"(%arg0, %arg1) : (tensor<10xf32>, tensor<20xf32>) -> tensor<10xi17>
   // expected-remark@+1 {{arith.constant 20}}
   %1 = "test.op_with_shaped_type_infer_type_if"(%arg1, %arg0) : (tensor<20xf32>, tensor<10xf32>) -> tensor<20xi17>
   return
 }

mlir/tools/mlir-tblgen/FormatGen.h

enum Kind {
     question,
     star,
     pipe,
 
     // Keywords.
     keyword_start,
     kw_attr_dict,
     kw_attr_dict_w_keyword,
+    kw_prop_dict,
     kw_custom,
     kw_functional_type,
     kw_oilist,
     kw_operands,
     kw_params,
     kw_qualified,
     kw_ref,
     kw_regions,
private:
   StringRef value;
 };
 
 class DirectiveElement : public FormatElementBase<FormatElement::Directive> {
 public:
   /// These are the kinds of directives.
   enum Kind {
     AttrDict,
+    PropDict,
     Custom,
     FunctionalType,
     OIList,
     Operands,
     Ref,
     Regions,
     Results,
     Successors,

mlir/tools/mlir-tblgen/FormatGen.cpp

while (isalnum(*curPtr) || *curPtr == '_' || *curPtr == '-')
     ++curPtr;
 
   // Check to see if this identifier is a keyword.
   StringRef str(tokStart, curPtr - tokStart);
   auto kind =
       StringSwitch<FormatToken::Kind>(str)
           .Case("attr-dict", FormatToken::kw_attr_dict)
           .Case("attr-dict-with-keyword", FormatToken::kw_attr_dict_w_keyword)
+          .Case("prop-dict", FormatToken::kw_prop_dict)
           .Case("custom", FormatToken::kw_custom)
           .Case("functional-type", FormatToken::kw_functional_type)
           .Case("oilist", FormatToken::kw_oilist)
           .Case("operands", FormatToken::kw_operands)
           .Case("params", FormatToken::kw_params)
           .Case("ref", FormatToken::kw_ref)
           .Case("regions", FormatToken::kw_regions)
           .Case("results", FormatToken::kw_results)

mlir/tools/mlir-tblgen/OpClass.cpp

declare<UsingDeclaration>("GenericAdaptor",
       ->addTemplateParam("RangeT");
   declare<UsingDeclaration>(
       "FoldAdaptor", "GenericAdaptor<::llvm::ArrayRef<::mlir::Attribute>>");
 }
 
 void OpClass::finalize() {
   Class::finalize();
   declare<VisibilityDeclaration>(Visibility::Public);
-  declare<ExtraClassDeclaration>(extraClassDeclaration, extraClassDefinition);
+  declare<ExtraClassDeclaration>(extraClassDeclaration.str(),
+                                 extraClassDefinition);
 }

mlir/tools/mlir-tblgen/OpDefinitionsGen.cpp

 //===- OpDefinitionsGen.cpp - MLIR op definitions generator ---------------===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // OpDefinitionsGen uses the description of operations to generate C++
 // definitions for ops.
 //
 //===----------------------------------------------------------------------===//
 
 #include "OpClass.h"
 #include "OpFormatGen.h"
 #include "OpGenHelpers.h"
+#include "mlir/TableGen/Argument.h"
+#include "mlir/TableGen/Attribute.h"
 #include "mlir/TableGen/Class.h"
 #include "mlir/TableGen/CodeGenHelpers.h"
 #include "mlir/TableGen/Format.h"
 #include "mlir/TableGen/GenInfo.h"
 #include "mlir/TableGen/Interfaces.h"
 #include "mlir/TableGen/Operator.h"
+#include "mlir/TableGen/Property.h"
 #include "mlir/TableGen/SideEffects.h"
 #include "mlir/TableGen/Trait.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/MapVector.h"
 #include "llvm/ADT/Sequence.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringSet.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Signals.h"
 #include "llvm/TableGen/Error.h"
 #include "llvm/TableGen/Record.h"
 #include "llvm/TableGen/TableGenBackend.h"
 
 #define DEBUG_TYPE "mlir-tblgen-opdefgen"
 
 using namespace llvm;
 using namespace mlir;
 using namespace mlir::tblgen;
 
 static const char *const tblgenNamePrefix = "tblgen_";
 static const char *const generatedArgName = "odsArg";
 static const char *const odsBuilder = "odsBuilder";
 static const char *const builderOpState = "odsState";
+static const char *const propertyStorage = "propStorage";
+static const char *const propertyValue = "propValue";
+static const char *const propertyAttr = "propAttr";
+static const char *const propertyDiag = "propDiag";
 
 /// The names of the implicit attributes that contain variadic operand and
 /// result segment sizes.
 static const char *const operandSegmentAttrName = "operand_segment_sizes";
 static const char *const resultSegmentAttrName = "result_segment_sizes";
 
 /// Code for an Op to lookup an attribute. Uses cached identifiers and subrange
 /// lookup.
static const char *const attrSizedSegmentValueRangeCalcCode = R"(
   for (unsigned i = 0; i < index; ++i)
     start += sizeAttr[i];
   return {start, sizeAttr[index]};
 )";
 /// The code snippet to initialize the sizes for the value range calculation.
 ///
 /// {0}: The code to get the attribute.
 static const char *const adapterSegmentSizeAttrInitCode = R"(
-  assert(odsAttrs && "missing segment size attribute for op");
+  assert({0} && "missing segment size attribute for op");
   auto sizeAttr = {0}.cast<::mlir::DenseI32ArrayAttr>();
 )";
 /// The code snippet to initialize the sizes for the value range calculation.
 ///
 /// {0}: The code to get the attribute.
 static const char *const opSegmentSizeAttrInitCode = R"(
   auto sizeAttr = {0}.cast<::mlir::DenseI32ArrayAttr>();
 )";
private:
     }
   };
 
   // Generate code for getting an attribute.
   Formatter getAttr(StringRef attrName, bool isNamed = false) const {
     assert(attrMetadata.count(attrName) && "expected attribute metadata");
     return [this, attrName, isNamed](raw_ostream &os) -> raw_ostream & {
       const AttributeMetadata &attr = attrMetadata.find(attrName)->second;
+      if (hasProperties()) {
+        assert(!isNamed);
+        return os << "getProperties()." << attrName;
+      }
       return os << formatv(subrangeGetAttr, getAttrName(attrName),
                            attr.lowerBound, attr.upperBound, getAttrRange(),
                            isNamed ? "Named" : "");
     };
   }
 
   // Generate code for getting the name of an attribute.
   Formatter getAttrName(StringRef attrName) const {
public:
   // Return the ODS operation wrapper.
   const Operator &getOp() const { return op; }
 
   // Get the attribute metadata sorted by name.
   const llvm::MapVector<StringRef, AttributeMetadata> &getAttrMetadata() const {
     return attrMetadata;
   }
 
+  /// Returns whether to emit a `Properties` struct for this operation or not.
+  bool hasProperties() const {
+    if (!op.getProperties().empty())
+      return true;
+    if (!op.getDialect().usePropertiesForAttributes())
+      return false;
+    return llvm::any_of(getAttrMetadata(),
+                        [](const std::pair<StringRef, AttributeMetadata> &it) {
+                          return !it.second.constraint ||
+                                 !it.second.constraint->isDerivedAttr();
+                        });
+  }
+
 private:
   // Compute the attribute metadata.
   void computeAttrMetadata();
 
   // The operation ODS wrapper.
   const Operator &op;
   // True if code is being generate for an op. False for an adaptor.
   const bool emitForOp;
private:
   void genAttrNameGetters();
 
   // Generates the OpAsmOpInterface for this operation if possible.
   void genOpAsmInterface();
 
   // Generates the `getOperationName` method for this op.
   void genOpNameGetter();
 
+  // Generates code to manage the properties, if any!
+  void genPropertiesSupport();
+
   // Generates getters for the attributes.
   void genAttrGetters();
 
   // Generates setter for the attributes.
   void genAttrSetters();
 
   // Generates removers for optional attributes.
   void genOptionalAttrRemovers();
body << formatv(checkAttrSizedValueSegmentsCode, operandSegmentAttrName,
                     emitHelper.emitErrorPrefix());
   }
   if (op.getTrait("::mlir::OpTrait::AttrSizedResultSegments")) {
     body << formatv(checkAttrSizedValueSegmentsCode, resultSegmentAttrName,
                     op.getNumResults(), "result", emitHelper.emitErrorPrefix());
   }
 }
 
+// Return true if a verifier can be emitted for the attribute: it is not a
+// derived attribute, it has a predicate, its condition is not empty, and, for
+// adaptors, the condition does not reference the op.
+static bool canEmitAttrVerifier(Attribute attr, bool isEmittingForOp) {
+  if (attr.isDerivedAttr())
+    return false;
+  Pred pred = attr.getPredicate();
+  if (pred.isNull())
+    return false;
+  std::string condition = pred.getCondition();
+  return !condition.empty() &&
+         (!StringRef(condition).contains("$_op") || isEmittingForOp);
+}
+
 // Generate attribute verification. If an op instance is not available, then
 // attribute checks that require one will not be emitted.
 //
 // Attribute verification is performed as follows:
 //
 // 1. Verify that all required attributes are present in sorted order. This
 // ensures that we can use subrange lookup even with potentially missing
 // attributes.
 // 2. Verify native trait attributes so that other attributes may call methods
 // that depend on the validity of these attributes, e.g. segment size attributes
 // and operand or result getters.
 // 3. Verify the constraints on all present attributes.
-static void genAttributeVerifier(
-    const OpOrAdaptorHelper &emitHelper, FmtContext &ctx, MethodBody &body,
-    const StaticVerifierFunctionEmitter &staticVerifierEmitter) {
+static void
+genAttributeVerifier(const OpOrAdaptorHelper &emitHelper, FmtContext &ctx,
+                     MethodBody &body,
+                     const StaticVerifierFunctionEmitter &staticVerifierEmitter,
+                     bool useProperties) {
   if (emitHelper.getAttrMetadata().empty())
     return;
 
   // Verify the attribute if it is present. This assumes that default values
   // are valid. This code snippet pastes the condition inline.
   //
   // TODO: verify the default value is valid (perhaps in debug mode only).
   //

 )";
 
   // Traverse the array until the required attribute is found. Return an error
   // if the traversal reached the end.
   //
   // {0}: Code to get the name of the attribute.
   // {1}: The emit error prefix.
   // {2}: The name of the attribute.
-  const char *const findRequiredAttr = R"(while (true) {{
+  const char *const findRequiredAttr = R"(
+while (true) {{
   if (namedAttrIt == namedAttrRange.end())
     return {1}"requires attribute '{2}'");
   if (namedAttrIt->getName() == {0}) {{
     tblgen_{2} = namedAttrIt->getValue();
     break;
   })";
 
   // Emit a check to see if the iteration has encountered an optional attribute.
   //
   // {0}: Code to get the name of the attribute.
   // {1}: The name of the attribute.
   const char *const checkOptionalAttr = R"(
   else if (namedAttrIt->getName() == {0}) {{
     tblgen_{1} = namedAttrIt->getValue();
   })";
 
   // Emit the start of the loop for checking trailing attributes.
   const char *const checkTrailingAttrs = R"(while (true) {
   if (namedAttrIt == namedAttrRange.end()) {
     break;
   })";
 
-  // Return true if a verifier can be emitted for the attribute: it is not a
-  // derived attribute, it has a predicate, its condition is not empty, and, for
-  // adaptors, the condition does not reference the op.
-  const auto canEmitVerifier = [&](Attribute attr) {
-    if (attr.isDerivedAttr())
-      return false;
-    Pred pred = attr.getPredicate();
-    if (pred.isNull())
-      return false;
-    std::string condition = pred.getCondition();
-    return !condition.empty() && (!StringRef(condition).contains("$_op") ||
-                                  emitHelper.isEmittingForOp());
-  };
-
   // Emit the verifier for the attribute.
   const auto emitVerifier = [&](Attribute attr, StringRef attrName,
                                 StringRef varName) {
     std::string condition = attr.getPredicate().getCondition();
 
     std::optional<StringRef> constraintFn;
     if (emitHelper.isEmittingForOp() &&
         (constraintFn = staticVerifierEmitter.getAttrConstraintFn(attr))) {
       body << formatv(verifyAttrUnique, *constraintFn, varName, attrName);
     } else {
       body << formatv(verifyAttrInline, varName,
                       tgfmt(condition, &ctx.withSelf(varName)),
                       emitHelper.emitErrorPrefix(), attrName,
                       escapeString(attr.getSummary()));
     }
   };
 
   // Prefix variables with `tblgen_` to avoid hiding the attribute accessor.
   const auto getVarName = [&](StringRef attrName) {
     return (tblgenNamePrefix + attrName).str();
   };
 
-  body.indent() << formatv("auto namedAttrRange = {0};\n",
-                           emitHelper.getAttrRange());
+  body.indent();
+  if (useProperties) {
+    for (const std::pair<StringRef, AttributeMetadata> &it :
+         emitHelper.getAttrMetadata()) {
+      body << formatv(
+          "auto tblgen_{0} = getProperties().{0}; (void)tblgen_{0};\n",
+          it.first);
+      const AttributeMetadata &metadata = it.second;
+      if (metadata.isRequired)
+        body << formatv(
+            "if (!tblgen_{0}) return {1}\"requires attribute '{0}'\");\n",
+            it.first, emitHelper.emitErrorPrefix());
+    }
+  } else {
+    body << formatv("auto namedAttrRange = {0};\n", emitHelper.getAttrRange());
     body << "auto namedAttrIt = namedAttrRange.begin();\n";
 
     // Iterate over the attributes in sorted order. Keep track of the optional
     // attributes that may be encountered along the way.
     SmallVector<const AttributeMetadata *> optionalAttrs;
+
     for (const std::pair<StringRef, AttributeMetadata> &it :
          emitHelper.getAttrMetadata()) {
       const AttributeMetadata &metadata = it.second;
       if (!metadata.isRequired) {
         optionalAttrs.push_back(&metadata);
         continue;
       }
 
       body << formatv("::mlir::Attribute {0};\n", getVarName(it.first));
       for (const AttributeMetadata *optional : optionalAttrs) {
         body << formatv("::mlir::Attribute {0};\n",
                         getVarName(optional->attrName));
       }
       body << formatv(findRequiredAttr, emitHelper.getAttrName(it.first),
                       emitHelper.emitErrorPrefix(), it.first);
       for (const AttributeMetadata *optional : optionalAttrs) {
         body << formatv(checkOptionalAttr,
                         emitHelper.getAttrName(optional->attrName),
                         optional->attrName);
       }
       body << "\n  ++namedAttrIt;\n}\n";
       optionalAttrs.clear();
     }
     // Get trailing optional attributes.
     if (!optionalAttrs.empty()) {
       for (const AttributeMetadata *optional : optionalAttrs) {
         body << formatv("::mlir::Attribute {0};\n",
                         getVarName(optional->attrName));
       }
       body << checkTrailingAttrs;
       for (const AttributeMetadata *optional : optionalAttrs) {
         body << formatv(checkOptionalAttr,
                         emitHelper.getAttrName(optional->attrName),
                         optional->attrName);
       }
       body << "\n  ++namedAttrIt;\n}\n";
     }
+  }
   body.unindent();
 
-  // Emit the checks for segment attributes first so that the other constraints
-  // can call operand and result getters.
+  // Emit the checks for segment attributes first so that the other
+  // constraints can call operand and result getters.
   genNativeTraitAttrVerifier(body, emitHelper);
 
+  bool isEmittingForOp = emitHelper.isEmittingForOp();
   for (const auto &namedAttr : emitHelper.getOp().getAttributes())
-    if (canEmitVerifier(namedAttr.attr))
+    if (canEmitAttrVerifier(namedAttr.attr, isEmittingForOp))
       emitVerifier(namedAttr.attr, namedAttr.name, getVarName(namedAttr.name));
 }
 
 /// Op extra class definitions have a `$cppClass` substitution that is to be
 /// replaced by the C++ class name.
 static std::string formatExtraDefinitions(const Operator &op) {
   FmtContext ctx = FmtContext().addSubst("cppClass", op.getCppClassName());
   return tgfmt(op.getExtraClassDefinition(), &ctx).str();
OpEmitter::OpEmitter(const Operator &op,
   genAttrNameGetters();
   genOpAsmInterface();
   genOpNameGetter();
   genNamedOperandGetters();
   genNamedOperandSetters();
   genNamedResultGetters();
   genNamedRegionGetters();
   genNamedSuccessorGetters();
+  genPropertiesSupport();
   genAttrGetters();
   genAttrSetters();
   genOptionalAttrRemovers();
   genBuilder();
   genPopulateDefaultAttributes();
   genParser();
   genPrinter();
   genVerifier();
body << "    if (!attr)\n      return "
                   &fctx.withSelf(defaultValue))
          << ";\n";
   }
   body << "  return "
        << tgfmt(attr.getConvertFromStorageCall(), &fctx.withSelf("attr"))
        << ";\n";
 }
 
+void OpEmitter::genPropertiesSupport() {
+  if (!emitHelper.hasProperties())
+    return;
+  using ConstArgument =
+      llvm::PointerUnion<const AttributeMetadata *, const NamedProperty *>;
+
+  SmallVector<ConstArgument> attrOrProperties;
+  for (const std::pair<StringRef, AttributeMetadata> &it :
+       emitHelper.getAttrMetadata()) {
+    if (!it.second.constraint || !it.second.constraint->isDerivedAttr())
+      attrOrProperties.push_back(&it.second);
+  }
+  for (const NamedProperty &prop : op.getProperties())
+    attrOrProperties.push_back(&prop);
+  if (attrOrProperties.empty())
+    return;
+  auto &setPropMethod =
+      opClass
+          .addStaticMethod(
+              "::mlir::LogicalResult", "setPropertiesFromAttr",
+              MethodParameter("Properties &", "prop"),
+              MethodParameter("::mlir::Attribute", "attr"),
+              MethodParameter("::mlir::InFlightDiagnostic *", "diag"))
+          ->body();
+  auto &getPropMethod =
+      opClass
+          .addStaticMethod("::mlir::Attribute", "getPropertiesAsAttr",
+                           MethodParameter("::mlir::MLIRContext *", "ctx"),
+                           MethodParameter("const Properties &", "prop"))
+          ->body();
+  auto &hashMethod =
+      opClass
+          .addStaticMethod("llvm::hash_code", "computePropertiesHash",
+                           MethodParameter("const Properties &", "prop"))
+          ->body();
+  auto &getInherentAttrMethod =
+      opClass
+          .addStaticMethod("std::optional<mlir::Attribute>", "getInherentAttr",
+                           MethodParameter("const Properties &", "prop"),
+                           MethodParameter("llvm::StringRef", "name"))
+          ->body();
+  auto &setInherentAttrMethod =
+      opClass
+          .addStaticMethod("void", "setInherentAttr",
+                           MethodParameter("Properties &", "prop"),
+                           MethodParameter("llvm::StringRef", "name"),
+                           MethodParameter("mlir::Attribute", "value"))
+          ->body();
+  auto &populateInherentAttrsMethod =
+      opClass
+          .addStaticMethod("void", "populateInherentAttrs",
+                           MethodParameter("const Properties &", "prop"),
+                           MethodParameter("::mlir::NamedAttrList &", "attrs"))
+          ->body();
+  auto &verifyInherentAttrsMethod =
+      opClass
+          .addStaticMethod(
+              "::mlir::LogicalResult", "verifyInherentAttrs",
+              MethodParameter("::mlir::OperationName", "opName"),
+              MethodParameter("::mlir::NamedAttrList &", "attrs"),
+              MethodParameter(
+                  "llvm::function_ref<::mlir::InFlightDiagnostic()>",
+                  "getDiag"))
+          ->body();
+
+  opClass.declare<UsingDeclaration>("Properties", "FoldAdaptor::Properties");
+
+  // Convert the property to the attribute form.
+
+  setPropMethod << R"decl(
+  ::mlir::DictionaryAttr dict = dyn_cast<::mlir::DictionaryAttr>(attr);
+  if (!dict) {
+    if (diag)
+      *diag << "expected DictionaryAttr to set properties";
+    return failure();
+  }
+    )decl";
+  // TODO: properties might be optional as well.
+  const char *propFromAttrFmt = R"decl(;
+    {{
+      auto setFromAttr = [] (auto &propStorage, ::mlir::Attribute propAttr, 
+                             ::mlir::InFlightDiagnostic *propDiag) {{
+        {0};
+      };
+      auto attr = dict.get("{1}");
+      if (!attr) {{
+        if (diag)
+          *diag << "expected key entry for {1} in DictionaryAttr to set "
+                   "Properties.";
+        return failure();
+      }
+      if (failed(setFromAttr(prop.{1}, attr, diag))) return ::mlir::failure();
+    }
+)decl";
+  for (const auto &attrOrProp : attrOrProperties) {
+    if (const auto *namedProperty =
+            attrOrProp.dyn_cast<const NamedProperty *>()) {
+      StringRef name = namedProperty->name;
+      auto &prop = namedProperty->prop;
+      FmtContext fctx;
+      setPropMethod << formatv(propFromAttrFmt,
+                               tgfmt(prop.getConvertFromAttributeCall(),
+                                     &fctx.addSubst("_attr", propertyAttr)
+                                          .addSubst("_storage", propertyStorage)
+                                          .addSubst("_diag", propertyDiag)),
+                               name);
+    } else {
+      const auto *namedAttr = attrOrProp.dyn_cast<const AttributeMetadata *>();
+      StringRef name = namedAttr->attrName;
+      setPropMethod << formatv(R"decl(
+  {{
+    auto &propStorage = prop.{0};
+    auto attr = dict.get("{0}");
+    if (attr || /*isRequired=*/{1}) {{
+      if (!attr) {{
+        if (diag)
+          *diag << "expected key entry for {0} in DictionaryAttr to set "
+                   "Properties.";
+        return failure();
+      }
+      auto convertedAttr = dyn_cast<std::remove_reference_t<decltype(propStorage)>>(attr);
+      if (convertedAttr) {{
+        propStorage = convertedAttr;
+      } else {{
+        if (diag)
+          *diag << "Invalid attribute `{0}` in property conversion: " << attr;
+        return failure();
+      }
+    }
+  }
+)decl",
+                               name, namedAttr->isRequired);
+    }
+  }
+  setPropMethod << "  return ::mlir::success();\n";
+
+  // Convert the attribute form to the property.
+
+  getPropMethod << "    ::mlir::SmallVector<::mlir::NamedAttribute> attrs;\n"
+                << "    ::mlir::Builder odsBuilder{ctx};\n";
+  const char *propToAttrFmt = R"decl(
+    {
+      const auto &propStorage = prop.{0};
+      attrs.push_back(odsBuilder.getNamedAttr("{0}",
+                                              {1}));
+    }
+)decl";
+  for (const auto &attrOrProp : attrOrProperties) {
+    if (const auto *namedProperty =
+            attrOrProp.dyn_cast<const NamedProperty *>()) {
+      StringRef name = namedProperty->name;
+      auto &prop = namedProperty->prop;
+      FmtContext fctx;
+      getPropMethod << formatv(
+          propToAttrFmt, name,
+          tgfmt(prop.getConvertToAttributeCall(),
+                &fctx.addSubst("_ctxt", "ctx")
+                     .addSubst("_storage", propertyStorage)));
+      continue;
+    }
+    const auto *namedAttr = attrOrProp.dyn_cast<const AttributeMetadata *>();
+    StringRef name = namedAttr->attrName;
+    getPropMethod << formatv(R"decl(
+    {{
+      const auto &propStorage = prop.{0};
+      if (propStorage)
+        attrs.push_back(odsBuilder.getNamedAttr("{0}",
+                                       propStorage));
+    }
+)decl",
+                             name);
+  }
+  getPropMethod << R"decl(
+  if (!attrs.empty())
+    return odsBuilder.getDictionaryAttr(attrs);
+  return {};    
+)decl";
+
+  // Hashing for the property
+
+  const char *propHashFmt = R"decl(
+  auto hash_{0} = [] (const auto &propStorage) -> llvm::hash_code {
+    return {1};
+  };
+)decl";
+  for (const auto &attrOrProp : attrOrProperties) {
+    if (const auto *namedProperty =
+            attrOrProp.dyn_cast<const NamedProperty *>()) {
+      StringRef name = namedProperty->name;
+      auto &prop = namedProperty->prop;
+      FmtContext fctx;
+      hashMethod << formatv(propHashFmt, name,
+                            tgfmt(prop.getHashPropertyCall(),
+                                  &fctx.addSubst("_storage", propertyStorage)));
+    }
+  }
+  hashMethod << "  return llvm::hash_combine(";
+  llvm::interleaveComma(
+      attrOrProperties, hashMethod, [&](const ConstArgument &attrOrProp) {
+        if (const auto *namedProperty =
+                attrOrProp.dyn_cast<const NamedProperty *>()) {
+          hashMethod << "\n    hash_" << namedProperty->name << "(prop."
+                     << namedProperty->name << ")";
+          return;
+        }
+        const auto *namedAttr =
+            attrOrProp.dyn_cast<const AttributeMetadata *>();
+        StringRef name = namedAttr->attrName;
+        hashMethod << "\n    llvm::hash_value(prop." << name
+                   << ".getAsOpaquePointer())";
+      });
+  hashMethod << ");\n";
+
+  const char *getInherentAttrMethodFmt = R"decl(
+    if (name == "{0}")
+      return prop.{0};
+)decl";
+  const char *setInherentAttrMethodFmt = R"decl(
+    if (name == "{0}") {{
+       prop.{0} = dyn_cast_or_null<std::remove_reference_t<decltype(prop.{0})>>(value);
+       return;
+    }
+)decl";
+  const char *populateInherentAttrsMethodFmt = R"decl(
+    if (prop.{0}) attrs.append("{0}", prop.{0});
+)decl";
+  for (const auto &attrOrProp : attrOrProperties) {
+    if (const auto *namedAttr =
+            attrOrProp.dyn_cast<const AttributeMetadata *>()) {
+      StringRef name = namedAttr->attrName;
+      getInherentAttrMethod << formatv(getInherentAttrMethodFmt, name);
+      setInherentAttrMethod << formatv(setInherentAttrMethodFmt, name);
+      populateInherentAttrsMethod
+          << formatv(populateInherentAttrsMethodFmt, name);
+      continue;
+    }
+  }
+  getInherentAttrMethod << "  return std::nullopt;\n";
+
+  // Emit the verifiers method for backward compatibility with the generic
+  // syntax. This method verifies the constraint on the properties attributes
+  // before they are set, since dyn_cast<> will silently omit failures.
+  for (const auto &attrOrProp : attrOrProperties) {
+    const auto *namedAttr = attrOrProp.dyn_cast<const AttributeMetadata *>();
+    if (!namedAttr || !namedAttr->constraint)
+      continue;
+    Attribute attr = *namedAttr->constraint;
+    std::optional<StringRef> constraintFn =
+        staticVerifierEmitter.getAttrConstraintFn(attr);
+    if (!constraintFn)
+      continue;
+    if (canEmitAttrVerifier(attr,
+                            /*isEmittingForOp=*/false)) {
+      std::string name = op.getGetterName(namedAttr->attrName);
+      verifyInherentAttrsMethod
+          << formatv(R"(
+    {{
+      ::mlir::Attribute attr = attrs.get({0}AttrName(opName));
+      if (attr && ::mlir::failed({1}(attr, "{2}", getDiag)))
+        return ::mlir::failure();
+    }
+)",
+                     name, constraintFn, namedAttr->attrName);
+    }
+  }
+  verifyInherentAttrsMethod << "    return ::mlir::success();";
+}
+
 void OpEmitter::genAttrGetters() {
   FmtContext fctx;
   fctx.withBuilder("::mlir::Builder((*this)->getContext())");
 
   // Emit the derived attribute body.
   auto emitDerivedAttr = [&](StringRef name, Attribute attr) {
     if (auto *method = opClass.addMethod(attr.getReturnType(), name))
       method->body() << "  " << attr.getDerivedCodeBody() << "\n";
   };
 
-  // Generate named accessor with Attribute return type. This is a wrapper class
-  // that allows referring to the attributes via accessors instead of having to
-  // use the string interface for better compile time verification.
+  // Generate named accessor with Attribute return type. This is a wrapper
+  // class that allows referring to the attributes via accessors instead of
+  // having to use the string interface for better compile time verification.
   auto emitAttrWithStorageType = [&](StringRef name, StringRef attrName,
                                      Attribute attr) {
     auto *method = opClass.addMethod(attr.getStorageType(), name + "Attr");
     if (!method)
       return;
     method->body() << formatv(
         "  return {0}.{1}<{2}>();", emitHelper.getAttr(attrName),
         attr.isOptional() || attr.hasDefaultValue() ? "dyn_cast_or_null"
// Generate method to materialize derived attributes as a DictionaryAttr.
     interleave(
         derivedAttrs, body,
         [&](const NamedAttribute &namedAttr) {
           auto tmpl = namedAttr.attr.getConvertFromStorageCall();
           std::string name = op.getGetterName(namedAttr.name);
           body << "    {" << name << "AttrName(),\n"
                << tgfmt(tmpl, &fctx.withSelf(name + "()")
                                    .withBuilder("odsBuilder")
-                                   .addSubst("_ctxt", "ctx"))
+                                   .addSubst("_ctxt", "ctx")
+                                   .addSubst("_storage", "ctx"))
                << "}";
         },
         ",\n");
     body << "});";
   }
 }
 
 void OpEmitter::genAttrSetters() {
for (const NamedAttribute &namedAttr : op.getAttributes()) {
     emitAttrWithStorageType(setterName, getterName, namedAttr.attr);
     emitAttrWithReturnType(setterName, getterName, namedAttr.attr);
   }
 }
 
 void OpEmitter::genOptionalAttrRemovers() {
   // Generate methods for removing optional attributes, instead of having to
   // use the string interface. Enables better compile time verification.
-  auto emitRemoveAttr = [&](StringRef name) {
+  auto emitRemoveAttr = [&](StringRef name, bool useProperties) {
     auto upperInitial = name.take_front().upper();
     auto *method = opClass.addMethod("::mlir::Attribute",
                                      op.getRemoverName(name) + "Attr");
     if (!method)
       return;
-    method->body() << formatv("  return (*this)->removeAttr({0}AttrName());",
+    if (useProperties) {
+      method->body() << formatv(R"(
+    auto &attr = getProperties().{0};
+    attr = {{};
+    return attr;
+)",
+                                name);
+      return;
+    }
+    method->body() << formatv("return (*this)->removeAttr({0}AttrName());",
                               op.getGetterName(name));
   };
 
   for (const NamedAttribute &namedAttr : op.getAttributes())
     if (namedAttr.attr.isOptional())
-      emitRemoveAttr(namedAttr.name);
+      emitRemoveAttr(namedAttr.name,
+                     op.getDialect().usePropertiesForAttributes());
 }
 
 // Generates the code to compute the start and end index of an operand or result
 // range.
 template <typename RangeT>
 static void generateValueRangeStartAndEnd(
     Class &opClass, bool isGenericAdaptorBase, StringRef methodName,
     int numVariadic, int numNonVariadic, StringRef rangeSizeCall,
auto *m = opClass.addMethod(operand.isVariadicOfVariadic()
                                 name + "Mutable");
     ERROR_IF_PRUNED(m, name, op);
     auto &body = m->body();
     body << "  auto range = getODSOperandIndexAndLength(" << i << ");\n"
          << "  auto mutableRange = "
             "::mlir::MutableOperandRange(getOperation(), "
             "range.first, range.second";
     if (attrSizedOperands) {
+      if (emitHelper.hasProperties())
+        body << formatv(
+            ", ::mlir::MutableOperandRange::OperandSegment({0}u, "
+            "{getOperandSegmentSizesAttrName(), getProperties().{1}})",
+            i, operandSegmentAttrName);
+      else
         body << formatv(
             ", ::mlir::MutableOperandRange::OperandSegment({0}u, *{1})", i,
             emitHelper.getAttr(operandSegmentAttrName, /*isNamed=*/true));
     }
     body << ");\n";
 
     // If this operand is a nested variadic, we split the range into a
     // MutableOperandRangeRange that provides a range over all of the
     // sub-ranges.
     if (operand.isVariadicOfVariadic()) {
       body << "  return "
if (inferType) {
       // Generate builder that infers type too.
       // TODO: Subsume this with general checking if type can be
       // inferred automatically.
       body << formatv(R"(
         ::llvm::SmallVector<::mlir::Type, 2> inferredReturnTypes;
         if (::mlir::succeeded({0}::inferReturnTypes(odsBuilder.getContext(),
                       {1}.location, {1}.operands,
                       {1}.attributes.getDictionary({1}.getContext()),
+                      {1}.getRawProperties(),
                       {1}.regions, inferredReturnTypes)))
           {1}.addTypes(inferredReturnTypes);
         else
           ::llvm::report_fatal_error("Failed to infer result type(s).");)",
                       opClass.getClassName(), builderOpState);
       return;
     }
 
     switch (paramKind) {
     case TypeParamKind::None:
       return;
     case TypeParamKind::Separate:
       for (int i = 0, e = op.getNumResults(); i < e; ++i) {
         if (op.getResult(i).isOptional())
           body << "  if (" << resultNames[i] << ")\n  ";
         body << "  " << builderOpState << ".addTypes(" << resultNames[i]
              << ");\n";
       }
 
       // Automatically create the 'result_segment_sizes' attribute using
       // the length of the type ranges.
       if (op.getTrait("::mlir::OpTrait::AttrSizedResultSegments")) {
+        if (op.getDialect().usePropertiesForAttributes()) {
+          body << "  (" << builderOpState
+               << ".getOrAddProperties<Properties>()." << resultSegmentAttrName
+               << " = \n"
+                  "        odsBuilder.getDenseI32ArrayAttr({";
+        } else {
           std::string getterName = op.getGetterName(resultSegmentAttrName);
           body << " " << builderOpState << ".addAttribute(" << getterName
                << "AttrName(" << builderOpState << ".name), "
                << "odsBuilder.getDenseI32ArrayAttr({";
-
+        }
         interleaveComma(
             llvm::seq<int>(0, op.getNumResults()), body, [&](int i) {
               const NamedTypeConstraint &result = op.getResult(i);
               if (!result.isVariableLength()) {
                 body << "1";
               } else if (result.isOptional()) {
                 body << "(" << resultNames[i] << " ? 1 : 0)";
               } else {

 
 void OpEmitter::genPopulateDefaultAttributes() {
   // All done if no attributes, except optional ones, have default values.
   if (llvm::all_of(op.getAttributes(), [](const NamedAttribute &named) {
         return !named.attr.hasDefaultValue() || named.attr.isOptional();
       }))
     return;
 
+  if (op.getDialect().usePropertiesForAttributes()) {
+    SmallVector<MethodParameter> paramList;
+    paramList.emplace_back("::mlir::OperationName", "opName");
+    paramList.emplace_back("Properties &", "properties");
+    auto *m =
+        opClass.addStaticMethod("void", "populateDefaultProperties", paramList);
+    ERROR_IF_PRUNED(m, "populateDefaultProperties", op);
+    auto &body = m->body();
+    body.indent();
+    body << "::mlir::Builder " << odsBuilder << "(opName.getContext());\n";
+    for (const NamedAttribute &namedAttr : op.getAttributes()) {
+      auto &attr = namedAttr.attr;
+      if (!attr.hasDefaultValue() || attr.isOptional())
+        continue;
+      StringRef name = namedAttr.name;
+      FmtContext fctx;
+      fctx.withBuilder(odsBuilder);
+      body << "if (!properties." << name << ")\n"
+           << "  properties." << name << " = "
+           << std::string(tgfmt(attr.getConstBuilderTemplate(), &fctx,
+                                tgfmt(attr.getDefaultValue(), &fctx)))
+           << ";\n";
+    }
+    return;
+  }
+
   SmallVector<MethodParameter> paramList;
   paramList.emplace_back("const ::mlir::OperationName &", "opName");
   paramList.emplace_back("::mlir::NamedAttrList &", "attributes");
   auto *m = opClass.addStaticMethod("void", "populateDefaultAttrs", paramList);
   ERROR_IF_PRUNED(m, "populateDefaultAttrs", op);
   auto &body = m->body();
   body.indent();
 
void OpEmitter::genInferredTypeCollectiveParamBuilder() {
   }
 
   // Result types
   body << formatv(R"(
   ::llvm::SmallVector<::mlir::Type, 2> inferredReturnTypes;
   if (::mlir::succeeded({0}::inferReturnTypes(odsBuilder.getContext(),
           {1}.location, operands,
           {1}.attributes.getDictionary({1}.getContext()),
+          {1}.getRawProperties(),
           {1}.regions, inferredReturnTypes))) {{)",
                   opClass.getClassName(), builderOpState);
   if (numVariadicResults == 0 || numNonVariadicResults != 0)
     body << "\n    assert(inferredReturnTypes.size()"
          << (numVariadicResults != 0 ? " >= " : " == ") << numNonVariadicResults
          << "u && \"mismatched number of return types\");";
   body << "\n    " << builderOpState << ".addTypes(inferredReturnTypes);";
 
if (const auto *operand = arg.dyn_cast<NamedTypeConstraint *>()) {
         type = "::mlir::ValueRange";
       else
         type = "::mlir::Value";
 
       paramList.emplace_back(type, getArgumentName(op, numOperands++),
                              operand->isOptional());
       continue;
     }
+    if (const auto *operand = arg.dyn_cast<NamedProperty *>()) {
+      // TODO
+      continue;
+    }
     const NamedAttribute &namedAttr = *arg.get<NamedAttribute *>();
     const Attribute &attr = namedAttr.attr;
 
     // Inferred attributes don't need to be added to the param list.
     if (inferredAttributes.contains(namedAttr.name))
       continue;
 
     StringRef type;
if (operand.constraint.isVariadicOfVariadic()) {
       body << "  for (::mlir::ValueRange range : " << argName << ")\n   "
            << builderOpState << ".addOperands(range);\n";
 
       // Add the segment attribute.
       body << "  {\n"
            << "    ::llvm::SmallVector<int32_t> rangeSegments;\n"
            << "    for (::mlir::ValueRange range : " << argName << ")\n"
            << "      rangeSegments.push_back(range.size());\n"
-           << "    " << builderOpState << ".addAttribute("
+           << "    auto rangeAttr = " << odsBuilder
+           << ".getDenseI32ArrayAttr(rangeSegments);\n";
+      if (op.getDialect().usePropertiesForAttributes()) {
+        body << "    " << builderOpState << ".getOrAddProperties<Properties>()."
+             << operand.constraint.getVariadicOfVariadicSegmentSizeAttr()
+             << " = rangeAttr;";
+      } else {
+        body << "    " << builderOpState << ".addAttribute("
              << op.getGetterName(
                     operand.constraint.getVariadicOfVariadicSegmentSizeAttr())
-           << "AttrName(" << builderOpState << ".name), " << odsBuilder
-           << ".getDenseI32ArrayAttr(rangeSegments));"
-           << "  }\n";
+             << "AttrName(" << builderOpState << ".name), rangeAttr);";
+      }
+      body << "  }\n";
       continue;
     }
 
     if (operand.isOptional())
       body << "  if (" << argName << ")\n  ";
     body << "  " << builderOpState << ".addOperands(" << argName << ");\n";
   }
 
   // If the operation has the operand segment size attribute, add it here.
   if (op.getTrait("::mlir::OpTrait::AttrSizedOperandSegments")) {
     std::string sizes = op.getGetterName(operandSegmentAttrName);
+    if (op.getDialect().usePropertiesForAttributes()) {
+      body << "   (" << builderOpState << ".getOrAddProperties<Properties>()."
+           << operandSegmentAttrName << "= "
+           << "odsBuilder.getDenseI32ArrayAttr({";
+    } else {
       body << "  " << builderOpState << ".addAttribute(" << sizes << "AttrName("
            << builderOpState << ".name), "
            << "odsBuilder.getDenseI32ArrayAttr({";
+    }
     interleaveComma(llvm::seq<int>(0, op.getNumOperands()), body, [&](int i) {
       const NamedTypeConstraint &operand = op.getOperand(i);
       if (!operand.isVariableLength()) {
         body << "1";
         return;
       }
 
       std::string operandName = getArgumentName(op, i);
for (const auto &namedAttr : op.getAttributes()) {
     if (emitNotNullCheck)
       body.indent() << formatv("if ({0}) ", namedAttr.name) << "{\n";
 
     if (isRawValueAttr && canUseUnwrappedRawValue(attr)) {
       // If this is a raw value, then we need to wrap it in an Attribute
       // instance.
       FmtContext fctx;
       fctx.withBuilder("odsBuilder");
+      if (op.getDialect().usePropertiesForAttributes()) {
+        body << formatv("  {0}.getOrAddProperties<Properties>().{1} = {2};\n",
+                        builderOpState, namedAttr.name,
+                        constBuildAttrFromParam(attr, fctx, namedAttr.name));
+      } else {
         body << formatv("  {0}.addAttribute({1}AttrName({0}.name), {2});\n",
                         builderOpState, op.getGetterName(namedAttr.name),
                         constBuildAttrFromParam(attr, fctx, namedAttr.name));
+      }
+    } else {
+      if (op.getDialect().usePropertiesForAttributes()) {
+        body << formatv("  {0}.getOrAddProperties<Properties>().{1} = {1};\n",
+                        builderOpState, namedAttr.name);
       } else {
         body << formatv("  {0}.addAttribute({1}AttrName({0}.name), {2});\n",
                         builderOpState, op.getGetterName(namedAttr.name),
                         namedAttr.name);
       }
+    }
     if (emitNotNullCheck)
       body.unindent() << "  }\n";
   }
 
   // Create the correct number of regions.
   for (const NamedRegion &region : op.getRegions()) {
     if (region.isVariadic())
       body << formatv("  for (unsigned i = 0; i < {0}Count; ++i)\n  ",
void OpEmitter::genSideEffectInterfaceMethods() {
   /// Attributes and Operands.
   for (unsigned i = 0, operandIt = 0, e = op.getNumArgs(); i != e; ++i) {
     Argument arg = op.getArg(i);
     if (arg.is<NamedTypeConstraint *>()) {
       resolveDecorators(op.getArgDecorators(i), operandIt, EffectKind::Operand);
       ++operandIt;
       continue;
     }
+    if (arg.is<NamedProperty *>())
+      continue;
     const NamedAttribute *attr = arg.get<NamedAttribute *>();
     if (attr->attr.getBaseAttr().isSymbolRefAttr())
       resolveDecorators(op.getArgDecorators(i), i, EffectKind::Symbol);
   }
   /// Results.
   for (unsigned i = 0, e = op.getNumResults(); i != e; ++i)
     resolveDecorators(op.getResultDecorators(i), i, EffectKind::Result);
 
void OpEmitter::genTypeInterfaceMethods() {
   SmallVector<int> constructedIndices(op.getNumResults(), -1);
   int inferredTypeIdx = 0;
   for (int numResults = op.getNumResults(); inferredTypeIdx != numResults;) {
     for (int i = 0, e = op.getNumResults(); i != e; ++i) {
       if (constructedIndices[i] >= 0)
         continue;
       const InferredResultType &infer = op.getInferredResultType(i);
       std::string typeStr;
-      body << "  ::mlir::Type odsInferredType" << inferredTypeIdx++ << " = ";
       if (infer.isArg()) {
         // If this is an operand, just index into operand list to access the
         // type.
         auto arg = op.getArgToOperandOrAttribute(infer.getIndex());
         if (arg.kind() == Operator::OperandOrAttribute::Kind::Operand) {
           typeStr = ("operands[" + Twine(arg.operandOrAttributeIndex()) +
                      "].getType()")
                         .str();
 
           // If this is an attribute, index into the attribute dictionary.
         } else {
           auto *attr =
               op.getArg(arg.operandOrAttributeIndex()).get<NamedAttribute *>();
-          typeStr = ("attributes.get(\"" + attr->name +
-                     "\").cast<::mlir::TypedAttr>().getType()")
-                        .str();
+          body << "  ::mlir::TypedAttr odsInferredTypeAttr" << inferredTypeIdx
+               << " = ";
+          if (op.getDialect().usePropertiesForAttributes()) {
+            body << "(properties ? properties.as<Properties *>()->"
+                 << attr->name
+                 << " : attributes.get(\"" + attr->name +
+                        "\").dyn_cast_or_null<::mlir::TypedAttr>());\n";
+          } else {
+            body << "attributes.get(\"" + attr->name +
+                        "\").dyn_cast_or_null<::mlir::TypedAttr>();\n";
+          }
+          body << "  if (!odsInferredTypeAttr" << inferredTypeIdx
+               << ") return ::mlir::failure();\n";
+          typeStr =
+              ("odsInferredTypeAttr" + Twine(inferredTypeIdx) + ".getType()")
         }
       } else if (std::optional<StringRef> builder =
                      op.getResult(infer.getResultIndex())
                          .constraint.getBuilderCall()) {
         typeStr = tgfmt(*builder, &fctx).str();
       } else if (int index = constructedIndices[infer.getResultIndex()];
                  index >= 0) {
         typeStr = ("odsInferredType" + Twine(index)).str();
       } else {
         continue;
       }
-      body << tgfmt(infer.getTransformer(), &fctx.withSelf(typeStr)) << ";\n";
+      body << "  ::mlir::Type odsInferredType" << inferredTypeIdx++ << " = "
+           << tgfmt(infer.getTransformer(), &fctx.withSelf(typeStr)) << ";\n";
       constructedIndices[i] = inferredTypeIdx - 1;
     }
   }
   for (auto [i, index] : llvm::enumerate(constructedIndices))
     body << "  inferredReturnTypes[" << i << "] = odsInferredType" << index
          << ";\n";
   body << "  return ::mlir::success();";
 }
void OpEmitter::genPrinter() {
   ERROR_IF_PRUNED(method, "print", op);
 }
 
 void OpEmitter::genVerifier() {
   auto *implMethod =
       opClass.addMethod("::mlir::LogicalResult", "verifyInvariantsImpl");
   ERROR_IF_PRUNED(implMethod, "verifyInvariantsImpl", op);
   auto &implBody = implMethod->body();
+  bool useProperties = emitHelper.hasProperties();
 
   populateSubstitutions(emitHelper, verifyCtx);
-  genAttributeVerifier(emitHelper, verifyCtx, implBody, staticVerifierEmitter);
+  genAttributeVerifier(emitHelper, verifyCtx, implBody, staticVerifierEmitter,
+                       useProperties);
   genOperandResultVerifier(implBody, op.getOperands(), "operand");
   genOperandResultVerifier(implBody, op.getResults(), "result");
 
   for (auto &trait : op.getTraits()) {
     if (auto *t = dyn_cast<tblgen::PredTrait>(&trait)) {
       implBody << tgfmt("  if (!($0))\n    "
                         "return emitOpError(\"failed to verify that $1\");\n",
                         &verifyCtx, tgfmt(t->getPredTemplate(), &verifyCtx),

 OpOperandAdaptorEmitter::OpOperandAdaptorEmitter(
     const Operator &op,
     const StaticVerifierFunctionEmitter &staticVerifierEmitter)
     : op(op), genericAdaptorBase(op.getGenericAdaptorName() + "Base"),
       genericAdaptor(op.getGenericAdaptorName()), adaptor(op.getAdaptorName()),
       staticVerifierEmitter(staticVerifierEmitter),
       emitHelper(op, /*emitForOp=*/false) {
 
+  genericAdaptorBase.declare<VisibilityDeclaration>(Visibility::Public);
+  bool useProperties = emitHelper.hasProperties();
+  if (useProperties) {
+    // Define the properties struct with multiple members.
+    using ConstArgument =
+        llvm::PointerUnion<const AttributeMetadata *, const NamedProperty *>;
+    SmallVector<ConstArgument> attrOrProperties;
+    for (const std::pair<StringRef, AttributeMetadata> &it :
+         emitHelper.getAttrMetadata()) {
+      if (!it.second.constraint || !it.second.constraint->isDerivedAttr())
+        attrOrProperties.push_back(&it.second);
+    }
+    for (const NamedProperty &prop : op.getProperties())
+      attrOrProperties.push_back(&prop);
+    assert(!attrOrProperties.empty());
+    std::string declarations = "  struct Properties {\n";
+    llvm::raw_string_ostream os(declarations);
+    for (const auto &attrOrProp : attrOrProperties) {
+      if (const auto *namedProperty =
+              attrOrProp.dyn_cast<const NamedProperty *>()) {
+        StringRef name = namedProperty->name;
+        if (name.empty())
+          report_fatal_error("missing name for property");
+        std::string camelName =
+            convertToCamelFromSnakeCase(name, /*capitalizeFirst=*/true);
+        auto &prop = namedProperty->prop;
+        // Generate the data member using the storage type.
+        os << "    using " << name << "Ty = " << prop.getStorageType() << ";\n"
+           << "    " << name << "Ty " << name;
+        if (prop.hasDefaultValue())
+          os << " = " << prop.getDefaultValue();
+
+        // Emit accessors using the interface type.
+        const char *accessorFmt = R"decl(;
+    {0} get{1}() {
+      auto &propStorage = this->{2};
+      return {3};
+    }
+    void set{1}(const {0} &propValue) {
+      auto &propStorage = this->{2};
+      {4};
+    }
+)decl";
+        FmtContext fctx;
+        os << formatv(accessorFmt, prop.getInterfaceType(), camelName, name,
+                      tgfmt(prop.getConvertFromStorageCall(),
+                            &fctx.addSubst("_storage", propertyStorage)),
+                      tgfmt(prop.getAssignToStorageCall(),
+                            &fctx.addSubst("_value", propertyValue)
+                                 .addSubst("_storage", propertyStorage)));
+        continue;
+      }
+      const auto *namedAttr = attrOrProp.dyn_cast<const AttributeMetadata *>();
+      const Attribute *attr = nullptr;
+      if (namedAttr->constraint)
+        attr = &*namedAttr->constraint;
+      StringRef name = namedAttr->attrName;
+      if (name.empty())
+        report_fatal_error("missing name for property attr");
+      std::string camelName =
+          convertToCamelFromSnakeCase(name, /*capitalizeFirst=*/true);
+      // Generate the data member using the storage type.
+      StringRef storageType;
+      if (attr) {
+        storageType = attr->getStorageType();
+      } else {
+        if (name != operandSegmentAttrName && name != resultSegmentAttrName) {
+          report_fatal_error("unexpected AttributeMetadata");
+        }
+        // TODO: update to use native integers.
+        storageType = "::mlir::DenseI32ArrayAttr";
+      }
+      os << "    using " << name << "Ty = " << storageType << ";\n"
+         << "    " << name << "Ty " << name << ";\n";
+
+      // Emit accessors using the interface type.
+      if (attr) {
+        const char *accessorFmt = R"decl(
+    auto get{0}() {
+      auto &propStorage = this->{1};
+      return propStorage.{2}<{3}>();
+    }
+    void set{0}(const {3} &propValue) {
+      this->{1} = propValue;
+    }
+)decl";
+        os << formatv(accessorFmt, camelName, name,
+                      attr->isOptional() || attr->hasDefaultValue()
+                          ? "dyn_cast_or_null"
+                          : "cast",
+                      storageType);
+      }
+    }
+    os << "  };\n";
+    os.flush();
+    genericAdaptorBase.declare<ExtraClassDeclaration>(std::move(declarations));
+  }
   genericAdaptorBase.declare<VisibilityDeclaration>(Visibility::Protected);
   genericAdaptorBase.declare<Field>("::mlir::DictionaryAttr", "odsAttrs");
-  genericAdaptorBase.declare<Field>("::mlir::RegionRange", "odsRegions");
   genericAdaptorBase.declare<Field>("::std::optional<::mlir::OperationName>",
                                     "odsOpName");
+  if (useProperties)
+    genericAdaptorBase.declare<Field>("Properties", "properties");
+  genericAdaptorBase.declare<Field>("::mlir::RegionRange", "odsRegions");
 
   genericAdaptor.addTemplateParam("RangeT");
   genericAdaptor.addField("RangeT", "odsOperands");
   genericAdaptor.addParent(
       ParentClass("detail::" + genericAdaptorBase.getClassName()));
   genericAdaptor.declare<UsingDeclaration>(
       "ValueT", "::llvm::detail::ValueOfRange<RangeT>");
   genericAdaptor.declare<UsingDeclaration>(
       "Base", "detail::" + genericAdaptorBase.getClassName());
 
   const auto *attrSizedOperands =
       op.getTrait("::mlir::OpTrait::AttrSizedOperandSegments");
   {
     SmallVector<MethodParameter> paramList;
     paramList.emplace_back("::mlir::DictionaryAttr", "attrs",
                            attrSizedOperands ? "" : "nullptr");
+    if (useProperties)
+      paramList.emplace_back("const Properties &", "properties", "{}");
+    else
+      paramList.emplace_back("::mlir::EmptyProperties", "properties", "{}");
     paramList.emplace_back("::mlir::RegionRange", "regions", "{}");
     auto *baseConstructor = genericAdaptorBase.addConstructor(paramList);
     baseConstructor->addMemberInitializer("odsAttrs", "attrs");
+    if (useProperties)
+      baseConstructor->addMemberInitializer("properties", "properties");
     baseConstructor->addMemberInitializer("odsRegions", "regions");
 
     MethodBody &body = baseConstructor->body();
     body.indent() << "if (odsAttrs)\n";
     body.indent() << formatv(
         "odsOpName.emplace(\"{0}\", odsAttrs.getContext());\n",
         op.getOperationName());
 
     paramList.insert(paramList.begin(), MethodParameter("RangeT", "values"));
     auto *constructor = genericAdaptor.addConstructor(std::move(paramList));
-    constructor->addMemberInitializer("Base", "attrs, regions");
+    constructor->addMemberInitializer("Base", "attrs, properties, regions");
     constructor->addMemberInitializer("odsOperands", "values");
   }
 
   std::string sizeAttrInit;
   if (op.getTrait("::mlir::OpTrait::AttrSizedOperandSegments")) {
     sizeAttrInit = formatv(adapterSegmentSizeAttrInitCode,
                            emitHelper.getAttr(operandSegmentAttrName));
   }
   generateNamedOperandGetters(op, genericAdaptor,
                               /*genericAdaptorBase=*/&genericAdaptorBase,
                               /*sizeAttrInit=*/sizeAttrInit,
                               /*rangeType=*/"RangeT",
                               /*rangeElementType=*/"ValueT",
                               /*rangeBeginCall=*/"odsOperands.begin()",
                               /*rangeSizeCall=*/"odsOperands.size()",
                               /*getOperandCallPattern=*/"odsOperands[{0}]");
 
-  // Any invalid overlap for `getOperands` will have been diagnosed before here
-  // already.
+  // Any invalid overlap for `getOperands` will have been diagnosed before
+  // here already.
   if (auto *m = genericAdaptor.addMethod("RangeT", "getOperands"))
     m->body() << "  return odsOperands;";
 
   FmtContext fctx;
   fctx.withBuilder("::mlir::Builder(odsAttrs.getContext())");
 
   // Generate named accessor with Attribute return type.
   auto emitAttrWithStorageType = [&](StringRef name, StringRef emitName,
                                      Attribute attr) {
     auto *method =
         genericAdaptorBase.addMethod(attr.getStorageType(), emitName + "Attr");
     ERROR_IF_PRUNED(method, "Adaptor::" + emitName + "Attr", op);
     auto &body = method->body().indent();
-    body << "assert(odsAttrs && \"no attributes when constructing adapter\");\n"
-         << formatv("auto attr = {0}.{1}<{2}>();\n", emitHelper.getAttr(name),
+    if (!useProperties)
+      body << "assert(odsAttrs && \"no attributes when constructing "
+              "adapter\");\n";
+    body << formatv("auto attr = {0}.{1}<{2}>();\n", emitHelper.getAttr(name),
                     attr.hasDefaultValue() || attr.isOptional()
                         ? "dyn_cast_or_null"
                         : "cast",
                     attr.getStorageType());
 
     if (attr.hasDefaultValue() && attr.isOptional()) {
       // Use the default value if attribute is not set.
       // TODO: this is inefficient, we are recreating the attribute for every
       // call. This should be set instead.
       std::string defaultValue = std::string(
           tgfmt(attr.getConstBuilderTemplate(), &fctx, attr.getDefaultValue()));
       body << "if (!attr)\n  attr = " << defaultValue << ";\n";
     }
     body << "return attr;\n";
   };
 
+  if (useProperties) {
+    auto *m = genericAdaptorBase.addInlineMethod("const Properties &",
+                                                 "getProperties");
+    ERROR_IF_PRUNED(m, "Adaptor::getProperties", op);
+    m->body() << "  return properties;";
+  }
   {
     auto *m =
         genericAdaptorBase.addMethod("::mlir::DictionaryAttr", "getAttributes");
     ERROR_IF_PRUNED(m, "Adaptor::getAttributes", op);
     m->body() << "  return odsAttrs;";
   }
   for (auto &namedAttr : op.getAttributes()) {
     const auto &name = namedAttr.name;
if (region.isVariadic()) {
       continue;
     }
 
     auto *m = genericAdaptorBase.addMethod("::mlir::Region &", name);
     ERROR_IF_PRUNED(m, "Adaptor::" + name, op);
     m->body() << formatv("  return *odsRegions[{0}];", i);
   }
   if (numRegions > 0) {
-    // Any invalid overlap for `getRegions` will have been diagnosed before here
-    // already.
+    // Any invalid overlap for `getRegions` will have been diagnosed before
+    // here already.
     if (auto *m =
             genericAdaptorBase.addMethod("::mlir::RegionRange", "getRegions"))
       m->body() << "  return odsRegions;";
   }
 
   StringRef genericAdaptorClassName = genericAdaptor.getClassName();
   adaptor.addParent(ParentClass(genericAdaptorClassName))
       .addTemplateParam("::mlir::ValueRange");
   adaptor.declare<VisibilityDeclaration>(Visibility::Public);
   adaptor.declare<UsingDeclaration>(genericAdaptorClassName +
                                     "::" + genericAdaptorClassName);
   {
     // Constructor taking the Op as single parameter.
     auto *constructor =
         adaptor.addConstructor(MethodParameter(op.getCppClassName(), "op"));
     constructor->addMemberInitializer(
-        adaptor.getClassName(),
-        "op->getOperands(), op->getAttrDictionary(), op->getRegions()");
+        adaptor.getClassName(), "op->getOperands(), op->getAttrDictionary(), "
+                                "op.getProperties(), op->getRegions()");
   }
 
   // Add verification function.
   addVerification();
 
   genericAdaptorBase.finalize();
   genericAdaptor.finalize();
   adaptor.finalize();
 }
 
 void OpOperandAdaptorEmitter::addVerification() {
   auto *method = adaptor.addMethod("::mlir::LogicalResult", "verify",
                                    MethodParameter("::mlir::Location", "loc"));
   ERROR_IF_PRUNED(method, "verify", op);
   auto &body = method->body();
+  bool useProperties = emitHelper.hasProperties();
 
   FmtContext verifyCtx;
   populateSubstitutions(emitHelper, verifyCtx);
-  genAttributeVerifier(emitHelper, verifyCtx, body, staticVerifierEmitter);
+  genAttributeVerifier(emitHelper, verifyCtx, body, staticVerifierEmitter,
+                       useProperties);
 
   body << "  return ::mlir::success();";
 }
 
 void OpOperandAdaptorEmitter::emitDecl(
     const Operator &op,
     const StaticVerifierFunctionEmitter &staticVerifierEmitter,
     raw_ostream &os) {

mlir/tools/mlir-tblgen/OpFormatGen.cpp

public:
   /// keyword.
   bool isWithKeyword() const { return withKeyword; }
 
 private:
   /// If the dictionary should be printed with the 'attributes' keyword.
   bool withKeyword;
 };
 
+/// This class represents the `prop-dict` directive. This directive represents
+/// the properties of the operation, expressed as a directionary.
+class PropDictDirective
+    : public DirectiveElementBase<DirectiveElement::PropDict> {
+public:
+  explicit PropDictDirective() = default;
+};
+
 /// This class represents the `functional-type` directive. This directive takes
 /// two arguments and formats them, respectively, as the inputs and results of a
 /// FunctionType.
 class FunctionalTypeDirective
     : public DirectiveElementBase<DirectiveElement::FunctionalType> {
 public:
   FunctionalTypeDirective(FormatElement *inputs, FormatElement *results)
       : inputs(inputs), results(results) {}
struct OperationFormat {
   enum class GenContext {
     /// The element is generated at the top-level or with the same behaviour.
     Normal,
     /// The element is generated inside an optional group.
     Optional
   };
 
   OperationFormat(const Operator &op)
-
-  {
+      : useProperties(op.getDialect().usePropertiesForAttributes() &&
+                      !op.getAttributes().empty()),
+        opCppClassName(op.getCppClassName()) {
     operandTypes.resize(op.getNumOperands(), TypeResolution());
     resultTypes.resize(op.getNumResults(), TypeResolution());
 
     hasImplicitTermTrait = llvm::any_of(op.getTraits(), [](const Trait &trait) {
       return trait.getDef().isSubClassOf("SingleBlockImplicitTerminator");
     });
 
     hasSingleBlockTrait =
struct OperationFormat {
 
   /// A flag indicating if this operation has the SingleBlockImplicitTerminator
   /// trait.
   bool hasImplicitTermTrait;
 
   /// A flag indicating if this operation has the SingleBlock trait.
   bool hasSingleBlockTrait;
 
+  /// Indicate whether attribute are stored in properties.
+  bool useProperties;
+
+  /// The Operation class name
+  StringRef opCppClassName;
+
   /// A map of buildable types to indices.
   llvm::MapVector<StringRef, int, llvm::StringMap<int>> buildableTypes;
 
   /// The index of the buildable type, if valid, for every operand and result.
   std::vector<TypeResolution> operandTypes, resultTypes;
 
   /// The set of attributes explicitly used within the format.
   SmallVector<const NamedAttribute *, 8> usedAttributes;
static bool shouldFormatSymbolNameAttr(const NamedAttribute *attr) {
   return attr->attr.getBaseAttr().getAttrDefName() == "SymbolNameAttr";
 }
 
 /// The code snippet used to generate a parser call for an attribute.
 ///
 /// {0}: The name of the attribute.
 /// {1}: The type for the attribute.
 const char *const attrParserCode = R"(
-  if (parser.parseCustomAttributeWithFallback({0}Attr, {1}, "{0}",
-          result.attributes)) {{
+  if (parser.parseCustomAttributeWithFallback({0}Attr, {1})) {{
     return ::mlir::failure();
   }
 )";
 
 /// The code snippet used to generate a parser call for an attribute.
 ///
 /// {0}: The name of the attribute.
 /// {1}: The type for the attribute.
 const char *const genericAttrParserCode = R"(
-  if (parser.parseAttribute({0}Attr, {1}, "{0}", result.attributes))
+  if (parser.parseAttribute({0}Attr, {1}))
     return ::mlir::failure();
 )";
 
 const char *const optionalAttrParserCode = R"(
-  {
-    ::mlir::OptionalParseResult parseResult =
-      parser.parseOptionalAttribute({0}Attr, {1}, "{0}", result.attributes);
-    if (parseResult.has_value() && failed(*parseResult))
-      return ::mlir::failure();
-  }
+  ::mlir::OptionalParseResult parseResult{0}Attr =
+    parser.parseOptionalAttribute({0}Attr, {1});
+  if (parseResult{0}Attr.has_value() && failed(*parseResult{0}Attr))
+  if (parseResult{0}Attr.has_value() && succeeded(*parseResult{0}Attr))
 )";
 
 /// The code snippet used to generate a parser call for a symbol name attribute.
 ///
 /// {0}: The name of the attribute.
 const char *const symbolNameAttrParserCode = R"(
-  if (parser.parseSymbolName({0}Attr, "{0}", result.attributes))
+  if (parser.parseSymbolName({0}Attr))
     return ::mlir::failure();
 )";
 const char *const optionalSymbolNameAttrParserCode = R"(
   // Parsing an optional symbol name doesn't fail, so no need to check the
   // result.
-  (void)parser.parseOptionalSymbolName({0}Attr, "{0}", result.attributes);
+  (void)parser.parseOptionalSymbolName({0}Attr);
 )";
 
 /// The code snippet used to generate a parser call for an enum attribute.
 ///
 /// {0}: The name of the attribute.
 /// {1}: The c++ namespace for the enum symbolize functions.
 /// {2}: The function to symbolize a string of the enum.
 /// {3}: The constant builder call to create an attribute of the enum type.
 /// {4}: The set of allowed enum keywords.
 /// {5}: The error message on failure when the enum isn't present.
+/// {6}: The attribute assignment expression
 const char *const enumAttrParserCode = R"(
   {
     ::llvm::StringRef attrStr;
     ::mlir::NamedAttrList attrStorage;
     auto loc = parser.getCurrentLocation();
     if (parser.parseOptionalKeyword(&attrStr, {4})) {
       ::mlir::StringAttr attrVal;
       ::mlir::OptionalParseResult parseResult =
const char *const enumAttrParserCode = R"(
     }
     if (!attrStr.empty()) {
       auto attrOptional = {1}::{2}(attrStr);
       if (!attrOptional)
         return parser.emitError(loc, "invalid ")
                << "{0} attribute specification: \"" << attrStr << '"';;
 
       {0}Attr = {3};
-      result.addAttribute("{0}", {0}Attr);
+      {6}
     }
   }
 )";
 
 /// The code snippet used to generate a parser call for an operand.
 ///
 /// {0}: The name of the operand.
 const char *const variadicOperandParserCode = R"(

 /// The code snippet used to generate a parser call to infer return types.
 ///
 /// {0}: The operation class name
 const char *const inferReturnTypesParserCode = R"(
   ::llvm::SmallVector<::mlir::Type> inferredReturnTypes;
   if (::mlir::failed({0}::inferReturnTypes(parser.getContext(),
       result.location, result.operands,
       result.attributes.getDictionary(parser.getContext()),
+      result.getRawProperties(),
       result.regions, inferredReturnTypes)))
     return ::mlir::failure();
   result.addTypes(inferredReturnTypes);
 )";
 
 /// The code snippet used to generate a parser call for a region list.
 ///
 /// {0}: The name for the region list.
if (auto *attr = dyn_cast<AttributeVariable>(param)) {
     body << tgfmt(string->getValue(), &ctx);
 
   } else {
     llvm_unreachable("unknown custom directive parameter");
   }
 }
 
 /// Generate the parser for a custom directive.
-static void genCustomDirectiveParser(CustomDirective *dir, MethodBody &body) {
+static void genCustomDirectiveParser(CustomDirective *dir, MethodBody &body,
+                                     bool useProperties,
+                                     StringRef opCppClassName) {
   body << "  {\n";
 
   // Preprocess the directive variables.
   // * Add a local variable for optional operands and types. This provides a
   //   better API to the user defined parser methods.
   // * Set the location of operand variables.
   for (FormatElement *param : dir->getArguments()) {
     if (auto *operand = dyn_cast<OperandVariable>(param)) {
body << "))\n"
        << "      return ::mlir::failure();\n";
 
   // After parsing, add handling for any of the optional constructs.
   for (FormatElement *param : dir->getArguments()) {
     if (auto *attr = dyn_cast<AttributeVariable>(param)) {
       const NamedAttribute *var = attr->getVar();
       if (var->attr.isOptional() || var->attr.hasDefaultValue())
         body << llvm::formatv("    if ({0}Attr)\n  ", var->name);
-
+      if (useProperties) {
+        body << formatv(
+            "    result.getOrAddProperties<{1}::Properties>().{0} = {0}Attr;\n",
+            var->name, opCppClassName);
+      } else {
         body << llvm::formatv("    result.addAttribute(\"{0}\", {0}Attr);\n",
                               var->name);
+      }
+
     } else if (auto *operand = dyn_cast<OperandVariable>(param)) {
       const NamedTypeConstraint *var = operand->getVar();
       if (var->isOptional()) {
         body << llvm::formatv("    if ({0}Operand.has_value())\n"
                               "      {0}Operands.push_back(*{0}Operand);\n",
                               var->name);
       } else if (var->isVariadicOfVariadic()) {
         body << llvm::formatv(
for (FormatElement *param : dir->getArguments()) {
     }
   }
 
   body << "  }\n";
 }
 
 /// Generate the parser for a enum attribute.
 static void genEnumAttrParser(const NamedAttribute *var, MethodBody &body,
-                              FmtContext &attrTypeCtx, bool parseAsOptional) {
+                              FmtContext &attrTypeCtx, bool parseAsOptional,
+                              bool useProperties, StringRef opCppClassName) {
   Attribute baseAttr = var->attr.getBaseAttr();
   const EnumAttr &enumAttr = cast<EnumAttr>(baseAttr);
   std::vector<EnumAttrCase> cases = enumAttr.getAllCases();
 
   // Generate the code for building an attribute for this enum.
   std::string attrBuilderStr;
   {
     llvm::raw_string_ostream os(attrBuilderStr);
errorMessageOS
         << "return parser.emitError(loc, \"expected string or "
            "keyword containing one of the following enum values for attribute '"
         << var->name << "' [";
     llvm::interleaveComma(cases, errorMessageOS, [&](const auto &attrCase) {
       errorMessageOS << attrCase.getStr();
     });
     errorMessageOS << "]\");";
   }
+  std::string attrAssignment;
+  if (useProperties) {
+    attrAssignment =
+        formatv("  "
+                "result.getOrAddProperties<{1}::Properties>().{0} = {0}Attr;",
+                var->name, opCppClassName);
+  } else {
+    attrAssignment =
+        formatv("result.addAttribute(\"{0}\", {0}Attr);", var->name);
+  }
 
   body << formatv(enumAttrParserCode, var->name, enumAttr.getCppNamespace(),
                   enumAttr.getStringToSymbolFnName(), attrBuilderStr,
-                  validCaseKeywordsStr, errorMessage);
+                  validCaseKeywordsStr, errorMessage, attrAssignment);
 }
 
 // Generate the parser for an attribute.
 static void genAttrParser(AttributeVariable *attr, MethodBody &body,
-                          FmtContext &attrTypeCtx, bool parseAsOptional) {
+                          FmtContext &attrTypeCtx, bool parseAsOptional,
+                          bool useProperties, StringRef opCppClassName) {
   const NamedAttribute *var = attr->getVar();
 
   // Check to see if we can parse this as an enum attribute.
   if (canFormatEnumAttr(var))
-    return genEnumAttrParser(var, body, attrTypeCtx, parseAsOptional);
+    return genEnumAttrParser(var, body, attrTypeCtx, parseAsOptional,
+                             useProperties, opCppClassName);
 
   // Check to see if we should parse this as a symbol name attribute.
   if (shouldFormatSymbolNameAttr(var)) {
     body << formatv(parseAsOptional ? optionalSymbolNameAttrParserCode
                                     : symbolNameAttrParserCode,
                     var->name);
-    return;
-  }
+  } else {
 
     // If this attribute has a buildable type, use that when parsing the
     // attribute.
     std::string attrTypeStr;
     if (std::optional<StringRef> typeBuilder = attr->getTypeBuilder()) {
       llvm::raw_string_ostream os(attrTypeStr);
       os << tgfmt(*typeBuilder, &attrTypeCtx);
     } else {
       attrTypeStr = "::mlir::Type{}";
     }
     if (parseAsOptional) {
       body << formatv(optionalAttrParserCode, var->name, attrTypeStr);
     } else {
       if (attr->shouldBeQualified() ||
           var->attr.getStorageType() == "::mlir::Attribute")
         body << formatv(genericAttrParserCode, var->name, attrTypeStr);
       else
         body << formatv(attrParserCode, var->name, attrTypeStr);
     }
   }
+  if (useProperties) {
+    body << formatv(
+        "  if ({0}Attr) result.getOrAddProperties<{1}::Properties>().{0} = "
+        "{0}Attr;\n",
+        var->name, opCppClassName);
+  } else {
+    body << formatv(
+        "  if ({0}Attr) result.attributes.append(\"{0}\", {0}Attr);\n",
+        var->name);
+  }
+}
 
 void OperationFormat::genParser(Operator &op, OpClass &opClass) {
   SmallVector<MethodParameter> paramList;
   paramList.emplace_back("::mlir::OpAsmParser &", "parser");
   paramList.emplace_back("::mlir::OperationState &", "result");
 
   auto *method = opClass.addStaticMethod("::mlir::ParseResult", "parse",
                                          std::move(paramList));
auto genElementParsers = [&](FormatElement *firstElement,
       FormatElement *elidedAnchorElement = nullptr;
       auto *anchorAttr = dyn_cast<AttributeVariable>(optional->getAnchor());
       if (anchorAttr && anchorAttr != firstElement &&
           anchorAttr->isUnitAttr()) {
         elidedAnchorElement = anchorAttr;
 
         if (!thenGroup == optional->isInverted()) {
           // Add the anchor unit attribute to the operation state.
+          if (useProperties) {
+            body << formatv(
+                "    result.getOrAddProperties<{1}::Properties>().{0} = "
+                "parser.getBuilder().getUnitAttr();",
+                anchorAttr->getVar()->name, opCppClassName);
+          } else {
             body << "    result.addAttribute(\"" << anchorAttr->getVar()->name
                  << "\", parser.getBuilder().getUnitAttr());\n";
           }
         }
+      }
 
       // Generate the rest of the elements inside an optional group. Elements in
       // an optional group after the guard are parsed as required.
       for (FormatElement *childElement : elements)
         if (childElement != elidedAnchorElement)
           genElementParser(childElement, body, attrTypeCtx,
                            GenContext::Optional);
     };
 
     ArrayRef<FormatElement *> thenElements =
         optional->getThenElements(/*parseable=*/true);
 
     // Generate a special optional parser for the first element to gate the
     // parsing of the rest of the elements.
     FormatElement *firstElement = thenElements.front();
     if (auto *attrVar = dyn_cast<AttributeVariable>(firstElement)) {
-      genAttrParser(attrVar, body, attrTypeCtx, /*parseAsOptional=*/true);
+      genAttrParser(attrVar, body, attrTypeCtx, /*parseAsOptional=*/true,
+                    useProperties, opCppClassName);
       body << "  if (" << attrVar->getVar()->name << "Attr) {\n";
     } else if (auto *literal = dyn_cast<LiteralElement>(firstElement)) {
       body << "  if (::mlir::succeeded(parser.parseOptional";
       genLiteralParser(literal->getSpelling(), body);
       body << ")) {\n";
     } else if (auto *opVar = dyn_cast<OperandVariable>(firstElement)) {
       genElementParser(opVar, body, attrTypeCtx);
       body << "  if (!" << opVar->getVar()->name << "Operands.empty()) {\n";
for (auto clause : oilist->getClauses()) {
       ArrayRef<FormatElement *> pelement = std::get<1>(clause);
       body << "if (succeeded(parser.parseOptional";
       genLiteralParser(lelement->getSpelling(), body);
       body << ")) {\n";
       StringRef lelementName = lelement->getSpelling();
       body << formatv(oilistParserCode, lelementName);
       if (AttributeVariable *unitAttrElem =
               oilist->getUnitAttrParsingElement(pelement)) {
+        if (useProperties) {
+          body << formatv(
+              "    result.getOrAddProperties<{1}::Properties>().{0} = "
+              "parser.getBuilder().getUnitAttr();",
+              unitAttrElem->getVar()->name, opCppClassName);
+        } else {
           body << "  result.addAttribute(\"" << unitAttrElem->getVar()->name
                << "\", UnitAttr::get(parser.getContext()));\n";
+        }
       } else {
         for (FormatElement *el : pelement)
           genElementParser(el, body, attrTypeCtx);
       }
       body << "    } else ";
     }
     body << " {\n";
     body << "    break;\n";
if (auto *optional = dyn_cast<OptionalElement>(element)) {
     /// Whitespaces.
   } else if (isa<WhitespaceElement>(element)) {
     // Nothing to parse.
 
     /// Arguments.
   } else if (auto *attr = dyn_cast<AttributeVariable>(element)) {
     bool parseAsOptional =
         (genCtx == GenContext::Normal && attr->getVar()->attr.isOptional());
-    genAttrParser(attr, body, attrTypeCtx, parseAsOptional);
+    genAttrParser(attr, body, attrTypeCtx, parseAsOptional, useProperties,
+                  opCppClassName);
 
   } else if (auto *operand = dyn_cast<OperandVariable>(element)) {
     ArgumentLengthKind lengthKind = getArgumentLengthKind(operand->getVar());
     StringRef name = operand->getVar()->name;
     if (lengthKind == ArgumentLengthKind::VariadicOfVariadic)
       body << llvm::formatv(
           variadicOfVariadicOperandParserCode, name,
           operand->getVar()->constraint.getVariadicOfVariadicSegmentSizeAttr());
void OperationFormat::genElementParser(FormatElement *element, MethodBody &body,
 
   } else if (auto *successor = dyn_cast<SuccessorVariable>(element)) {
     bool isVariadic = successor->getVar()->isVariadic();
     body << formatv(isVariadic ? successorListParserCode : successorParserCode,
                     successor->getVar()->name);
 
     /// Directives.
   } else if (auto *attrDict = dyn_cast<AttrDictDirective>(element)) {
-    body << "  if (parser.parseOptionalAttrDict"
+    body.indent() << "{\n";
+    body.indent() << "auto loc = parser.getCurrentLocation();(void)loc;\n"
+                  << "if (parser.parseOptionalAttrDict"
                   << (attrDict->isWithKeyword() ? "WithKeyword" : "")
                   << "(result.attributes))\n"
-         << "    return ::mlir::failure();\n";
+                  << "  return ::mlir::failure();\n";
+    if (useProperties) {
+      body << "if (failed(verifyInherentAttrs(result.name, result.attributes, "
+              "[&]() {\n"
+           << "    return parser.emitError(loc) << \"'\" << "
+              "result.name.getStringRef() << \"' op \";\n"
+           << "  })))\n"
+           << "  return ::mlir::failure();\n";
+    }
+    body.unindent() << "}\n";
+    body.unindent();
+  } else if (auto *attrDict = dyn_cast<PropDictDirective>(element)) {
+    body << "  if (parseProperties(parser, result))\n"
+         << "    return ::mlir::failure();\n";
   } else if (auto *customDir = dyn_cast<CustomDirective>(element)) {
-    genCustomDirectiveParser(customDir, body);
-
+    genCustomDirectiveParser(customDir, body, useProperties, opCppClassName);
   } else if (isa<OperandsDirective>(element)) {
     body << "  ::llvm::SMLoc allOperandLoc = parser.getCurrentLocation();\n"
          << "  if (parser.parseOperandList(allOperands))\n"
          << "    return ::mlir::failure();\n";
 
   } else if (isa<RegionsDirective>(element)) {
     body << llvm::formatv(regionListParserCode, "full");
     if (hasImplicitTermTrait)
else
       body << "  result.addSuccessors(" << successor.name << "Successor);\n";
   }
 }
 
 void OperationFormat::genParserVariadicSegmentResolution(Operator &op,
                                                          MethodBody &body) {
   if (!allOperands) {
     if (op.getTrait("::mlir::OpTrait::AttrSizedOperandSegments")) {
+      if (op.getDialect().usePropertiesForAttributes()) {
+        body << formatv("  "
+                        "result.getOrAddProperties<{0}::Properties>().operand_"
+                        "segment_sizes = "
+                        "(parser.getBuilder().getDenseI32ArrayAttr({{",
+                        op.getCppClassName());
+      } else {
         body << "  result.addAttribute(\"operand_segment_sizes\", "
              << "parser.getBuilder().getDenseI32ArrayAttr({";
+      }
       auto interleaveFn = [&](const NamedTypeConstraint &operand) {
         // If the operand is variadic emit the parsed size.
         if (operand.isVariableLength())
           body << "static_cast<int32_t>(" << operand.name << "Operands.size())";
         else
           body << "1";
       };
       llvm::interleaveComma(op.getOperands(), body, interleaveFn);
       body << "}));\n";
     }
     for (const NamedTypeConstraint &operand : op.getOperands()) {
       if (!operand.isVariadicOfVariadic())
         continue;
+      if (op.getDialect().usePropertiesForAttributes()) {
+        body << llvm::formatv(
+            "  result.getOrAddProperties<{0}::Properties>().{1} = "
+            "parser.getBuilder().getDenseI32ArrayAttr({2}OperandGroupSizes);\n",
+            op.getCppClassName(),
+            operand.constraint.getVariadicOfVariadicSegmentSizeAttr(),
+            operand.name);
+      } else {
         body << llvm::formatv(
             "  result.addAttribute(\"{0}\", "
-          "parser.getBuilder().getDenseI32ArrayAttr({1}OperandGroupSizes));\n",
+            "parser.getBuilder().getDenseI32ArrayAttr({1}OperandGroupSizes));"
+            "\n",
             operand.constraint.getVariadicOfVariadicSegmentSizeAttr(),
             operand.name);
       }
     }
+  }
 
   if (!allResultTypes &&
       op.getTrait("::mlir::OpTrait::AttrSizedResultSegments")) {
+    if (op.getDialect().usePropertiesForAttributes()) {
+      body << formatv(
+          "  "
+          "result.getOrAddProperties<{0}::Properties>().result_segment_sizes = "
+          "(parser.getBuilder().getDenseI32ArrayAttr({{",
+          op.getCppClassName());
+    } else {
       body << "  result.addAttribute(\"result_segment_sizes\", "
            << "parser.getBuilder().getDenseI32ArrayAttr({";
+    }
     auto interleaveFn = [&](const NamedTypeConstraint &result) {
       // If the result is variadic emit the parsed size.
       if (result.isVariableLength())
         body << "static_cast<int32_t>(" << result.name << "Types.size())";
       else
         body << "1";
     };
     llvm::interleaveComma(op.getResults(), body, interleaveFn);

 /// {0}: The name of the enum attribute.
 /// {1}: The name of the enum attributes symbolToString function.
 const char *enumAttrBeginPrinterCode = R"(
   {
     auto caseValue = {0}();
     auto caseValueStr = {1}(caseValue);
 )";
 
+/// Generate the printer for the 'prop-dict' directive.
+static void genPropDictPrinter(OperationFormat &fmt, Operator &op,
+                               MethodBody &body) {
+  body << "  _odsPrinter << \" \";\n"
+       << "  printProperties(this->getContext(), _odsPrinter, "
+          "getProperties());\n";
+}
+
 /// Generate the printer for the 'attr-dict' directive.
 static void genAttrDictPrinter(OperationFormat &fmt, Operator &op,
                                MethodBody &body, bool withKeyword) {
   body << "  ::llvm::SmallVector<::llvm::StringRef, 2> elidedAttrs;\n";
   // Elide the variadic segment size attributes if necessary.
   if (!fmt.allOperands &&
       op.getTrait("::mlir::OpTrait::AttrSizedOperandSegments"))
     body << "  elidedAttrs.push_back(\"operand_segment_sizes\");\n";
TypeSwitch<FormatElement *>(anchor)
       .Case<TypeDirective>([&](TypeDirective *element) {
         genOptionalGroupPrinterAnchor(element->getArg(), op, body);
       })
       .Case<FunctionalTypeDirective>([&](FunctionalTypeDirective *element) {
         genOptionalGroupPrinterAnchor(element->getInputs(), op, body);
       })
       .Case<AttributeVariable>([&](AttributeVariable *element) {
         Attribute attr = element->getVar()->attr;
-        body << "(*this)->getAttr(\"" << element->getVar()->name << "\")";
+        body << op.getGetterName(element->getVar()->name) << "Attr()";
         if (attr.isOptional())
           return; // done
         if (attr.hasDefaultValue()) {
           // Consider a default-valued attribute as present if it's not the
           // default value.
           FmtContext fctx;
           fctx.withBuilder("::mlir::OpBuilder((*this)->getContext())");
-          body << " != "
+          body << " && " << op.getGetterName(element->getVar()->name)
+               << "Attr() != "
                << tgfmt(attr.getConstBuilderTemplate(), &fctx,
                         attr.getDefaultValue());
           return;
         }
         llvm_unreachable("attribute must be optional or default-valued");
       });
 }
 
void OperationFormat::genElementPrinter(FormatElement *element,
 
   // Emit the attribute dictionary.
   if (auto *attrDict = dyn_cast<AttrDictDirective>(element)) {
     genAttrDictPrinter(*this, op, body, attrDict->isWithKeyword());
     lastWasPunctuation = false;
     return;
   }
 
+  // Emit the attribute dictionary.
+  if (auto *propDict = dyn_cast<PropDictDirective>(element)) {
+    genPropDictPrinter(*this, op, body);
+    lastWasPunctuation = false;
+    return;
+  }
+
   // Optionally insert a space before the next element. The AttrDict printer
   // already adds a space as necessary.
   if (shouldEmitSpace || !lastWasPunctuation)
     body << "  _odsPrinter << ' ';\n";
   lastWasPunctuation = false;
   shouldEmitSpace = true;
 
   if (auto *attr = dyn_cast<AttributeVariable>(element)) {
void handleTypesMatchConstraint(
       llvm::StringMap<TypeResolutionInstance> &variableTyResolver,
       const llvm::Record &def);
 
   /// Returns an argument or attribute with the given name that has been seen
   /// within the format.
   ConstArgument findSeenArg(StringRef name);
 
   /// Parse the various different directives.
+  FailureOr<FormatElement *> parsePropDictDirective(SMLoc loc, Context context);
   FailureOr<FormatElement *> parseAttrDictDirective(SMLoc loc, Context context,
                                                     bool withKeyword);
   FailureOr<FormatElement *> parseFunctionalTypeDirective(SMLoc loc,
                                                           Context context);
   FailureOr<FormatElement *> parseOIListDirective(SMLoc loc, Context context);
   LogicalResult verifyOIListParsingElement(FormatElement *element, SMLoc loc);
   FailureOr<FormatElement *> parseOperandsDirective(SMLoc loc, Context context);
   FailureOr<FormatElement *> parseQualifiedDirective(SMLoc loc,
private:
   //===--------------------------------------------------------------------===//
 
   OperationFormat &fmt;
   Operator &op;
 
   // The following are various bits of format state used for verification
   // during parsing.
   bool hasAttrDict = false;
+  bool hasPropDict = false;
   bool hasAllRegions = false, hasAllSuccessors = false;
   bool canInferResultTypes = false;
   llvm::SmallBitVector seenOperandTypes, seenResultTypes;
   llvm::SmallSetVector<const NamedAttribute *, 8> seenAttrs;
   llvm::DenseSet<const NamedTypeConstraint *> seenOperands;
   llvm::DenseSet<const NamedRegion *> seenRegions;
   llvm::DenseSet<const NamedSuccessor *> seenSuccessors;
 };
OpFormatParser::parseVariableImpl(SMLoc loc, StringRef name, Context ctx) {
   return emitError(loc, "expected variable to refer to an argument, region, "
                         "result, or successor");
 }
 
 FailureOr<FormatElement *>
 OpFormatParser::parseDirectiveImpl(SMLoc loc, FormatToken::Kind kind,
                                    Context ctx) {
   switch (kind) {
+  case FormatToken::kw_prop_dict:
+    return parsePropDictDirective(loc, ctx);
   case FormatToken::kw_attr_dict:
     return parseAttrDictDirective(loc, ctx,
                                   /*withKeyword=*/false);
   case FormatToken::kw_attr_dict_w_keyword:
     return parseAttrDictDirective(loc, ctx,
                                   /*withKeyword=*/true);
   case FormatToken::kw_functional_type:
     return parseFunctionalTypeDirective(loc, ctx);
if (context == RefDirectiveContext) {
     if (hasAttrDict)
       return emitError(loc, "'attr-dict' directive has already been seen");
     hasAttrDict = true;
   }
 
   return create<AttrDictDirective>(withKeyword);
 }
 
+FailureOr<FormatElement *>
+OpFormatParser::parsePropDictDirective(SMLoc loc, Context context) {
+  if (context == TypeDirectiveContext)
+    return emitError(loc, "'prop-dict' directive can only be used as a "
+                          "top-level directive");
+
+  if (context == RefDirectiveContext)
+    llvm::report_fatal_error("'ref' of 'prop-dict' unsupported");
+  // Otherwise, this is a top-level context.
+
+  if (hasPropDict)
+    return emitError(loc, "'prop-dict' directive has already been seen");
+  hasPropDict = true;
+
+  return create<PropDictDirective>();
+}
+
 LogicalResult OpFormatParser::verifyCustomDirectiveArguments(
     SMLoc loc, ArrayRef<FormatElement *> arguments) {
   for (FormatElement *argument : arguments) {
     if (!isa<StringElement, RefDirective, TypeDirective, AttrDictDirective,
              AttributeVariable, OperandVariable, RegionVariable,
              SuccessorVariable>(argument)) {
       // TODO: FormatElement should have location info attached.
       return emitError(loc, "only variables and types may be used as "

mlir/unittests/Debug/FileLineColLocBreakpointManagerTest.cpp

 using namespace mlir::tracing;
 
 static Operation *createOp(MLIRContext *context, Location loc,
                            StringRef operationName,
                            unsigned int numRegions = 0) {
   context->allowUnregisteredDialects();
   return Operation::create(loc, OperationName(operationName, context),
                            std::nullopt, std::nullopt, std::nullopt,
-                           std::nullopt, numRegions);
+                           OpaqueProperties(nullptr), std::nullopt, numRegions);
 }
 
 namespace {
 struct FileLineColLocTestingAction
     : public ActionImpl<FileLineColLocTestingAction> {
   MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(FileLineColLocTestingAction)
   static constexpr StringLiteral tag = "file-line-col-loc-testing-action";
   FileLineColLocTestingAction(ArrayRef<IRUnit> irUnits)

mlir/unittests/IR/AdaptorTest.cpp

TEST(Adaptor, GenericAdaptorsOperandAccess) {
 
   // Has optional arguments.
   OIListSimple::FoldAdaptor c({}, nullptr);
   {
     // Optional arguments return the default constructed value if not present.
     // Using optional instead of plain int here to differentiate absence of
     // value from the value 0.
     SmallVector<std::optional<int>> v = {0, 4};
-    OIListSimple::GenericAdaptor<ArrayRef<std::optional<int>>> d(
-        v, builder.getDictionaryAttr({builder.getNamedAttr(
-               "operand_segment_sizes",
-               builder.getDenseI32ArrayAttr({1, 0, 1}))}));
+    OIListSimple::Properties prop;
+    prop.operand_segment_sizes = builder.getDenseI32ArrayAttr({1, 0, 1});
+    OIListSimple::GenericAdaptor<ArrayRef<std::optional<int>>> d(v, {}, prop,
+                                                                 {});
     EXPECT_EQ(d.getArg0(), 0);
     EXPECT_EQ(d.getArg1(), std::nullopt);
     EXPECT_EQ(d.getArg2(), 4);
   }
 
   // Has VariadicOfVariadic arguments.
   FormatVariadicOfVariadicOperand::FoldAdaptor e({});
   {
     SmallVector<int> v = {0, 1, 2, 3, 4};
-    FormatVariadicOfVariadicOperand::GenericAdaptor<ArrayRef<int>> f(
-        v, builder.getDictionaryAttr({builder.getNamedAttr(
-               "operand_segments", builder.getDenseI32ArrayAttr({3, 2, 0}))}));
+    FormatVariadicOfVariadicOperand::Properties prop;
+    prop.operand_segments = builder.getDenseI32ArrayAttr({3, 2, 0});
+    FormatVariadicOfVariadicOperand::GenericAdaptor<ArrayRef<int>> f(v, {},
+                                                                     prop, {});
     SmallVector<ArrayRef<int>> operand = f.getOperand();
     ASSERT_EQ(operand.size(), (std::size_t)3);
     EXPECT_THAT(operand[0], ElementsAre(0, 1, 2));
     EXPECT_THAT(operand[1], ElementsAre(3, 4));
     EXPECT_THAT(operand[2], ElementsAre());
   }
 }

mlir/unittests/IR/CMakeLists.txt

 add_mlir_unittest(MLIRIRTests
   AdaptorTest.cpp
   AttributeTest.cpp
   DialectTest.cpp
   InterfaceTest.cpp
   IRMapping.cpp
   InterfaceAttachmentTest.cpp
   OperationSupportTest.cpp
   PatternMatchTest.cpp
   ShapedTypeTest.cpp
   TypeTest.cpp
+  OpPropertiesTest.cpp
 
   DEPENDS
   MLIRTestInterfaceIncGen
 )
 target_include_directories(MLIRIRTests PRIVATE "${MLIR_BINARY_DIR}/test/lib/Dialect/Test")
 target_link_libraries(MLIRIRTests
   PRIVATE
   MLIRIR
   MLIRTestDialect)

mlir/unittests/IR/OpPropertiesTest.cpp

+//===- TestOpProperties.cpp - Test all properties-related APIs ------------===//
+//
+// 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/IR/OpDefinition.h"
+#include "mlir/Parser/Parser.h"
+#include "gtest/gtest.h"
+#include <optional>
+
+using namespace mlir;
+
+namespace {
+/// Simple structure definining a struct to define "properties" for a given
+/// operation. Default values are honored when creating an operation.
+struct TestProperties {
+  int a = -1;
+  float b = -1.;
+  std::vector<int64_t> array = {-33};
+  /// A shared_ptr to a const object is safe: it is equivalent to a value-based
+  /// member. Here the label will be deallocated when the last operation
+  /// referring to it is destroyed. However there is no pool-allocation: this is
+  /// offloaded to the client.
+  std::shared_ptr<const std::string> label;
+  MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(TestProperties)
+};
+
+/// Convert a DictionaryAttr to a TestProperties struct, optionally emit errors
+/// through the provided diagnostic if any. This is used for example during
+/// parsing with the generic format.
+static LogicalResult
+setPropertiesFromAttribute(TestProperties &prop, Attribute attr,
+                           InFlightDiagnostic *diagnostic) {
+  DictionaryAttr dict = dyn_cast<DictionaryAttr>(attr);
+  if (!dict) {
+    if (diagnostic)
+      *diagnostic << "expected DictionaryAttr to set TestProperties";
+    return failure();
+  }
+  auto aAttr = dict.getAs<IntegerAttr>("a");
+  if (!aAttr) {
+    if (diagnostic)
+      *diagnostic << "expected IntegerAttr for key `a`";
+    return failure();
+  }
+  auto bAttr = dict.getAs<FloatAttr>("b");
+  if (!bAttr ||
+      &bAttr.getValue().getSemantics() != &llvm::APFloatBase::IEEEsingle()) {
+    if (diagnostic)
+      *diagnostic << "expected FloatAttr for key `b`";
+    return failure();
+  }
+
+  auto arrayAttr = dict.getAs<DenseI64ArrayAttr>("array");
+  if (!arrayAttr) {
+    if (diagnostic)
+      *diagnostic << "expected DenseI64ArrayAttr for key `array`";
+    return failure();
+  }
+
+  auto label = dict.getAs<mlir::StringAttr>("label");
+  if (!label) {
+    if (diagnostic)
+      *diagnostic << "expected StringAttr for key `label`";
+    return failure();
+  }
+
+  prop.a = aAttr.getValue().getSExtValue();
+  prop.b = bAttr.getValue().convertToFloat();
+  prop.array.assign(arrayAttr.asArrayRef().begin(),
+                    arrayAttr.asArrayRef().end());
+  prop.label = std::make_shared<std::string>(label.getValue());
+  return success();
+}
+
+/// Convert a TestProperties struct to a DictionaryAttr, this is used for
+/// example during printing with the generic format.
+static Attribute getPropertiesAsAttribute(MLIRContext *ctx,
+                                          const TestProperties &prop) {
+  SmallVector<NamedAttribute> attrs;
+  Builder b{ctx};
+  attrs.push_back(b.getNamedAttr("a", b.getI32IntegerAttr(prop.a)));
+  attrs.push_back(b.getNamedAttr("b", b.getF32FloatAttr(prop.b)));
+  attrs.push_back(b.getNamedAttr("array", b.getDenseI64ArrayAttr(prop.array)));
+  attrs.push_back(b.getNamedAttr(
+      "label", b.getStringAttr(prop.label ? *prop.label : "<nullptr>")));
+  return b.getDictionaryAttr(attrs);
+}
+
+inline llvm::hash_code computeHash(const TestProperties &prop) {
+  // We hash `b` which is a float using its underlying array of char:
+  unsigned char const *p = reinterpret_cast<unsigned char const *>(&prop.b);
+  ArrayRef<unsigned char> bBytes{p, sizeof(prop.b)};
+  return llvm::hash_combine(
+      prop.a, llvm::hash_combine_range(bBytes.begin(), bBytes.end()),
+      llvm::hash_combine_range(prop.array.begin(), prop.array.end()),
+      StringRef(*prop.label));
+}
+
+/// A custom operation for the purpose of showcasing how to use "properties".
+class OpWithProperties : public Op<OpWithProperties> {
+public:
+  // Begin boilerplate
+  MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(OpWithProperties)
+  using Op::Op;
+  static ArrayRef<StringRef> getAttributeNames() { return {}; }
+  static StringRef getOperationName() {
+    return "test_op_properties.op_with_properties";
+  }
+  // End boilerplate
+
+  // This alias is the only definition needed for enabling "properties" for this
+  // operation.
+  using Properties = TestProperties;
+  static std::optional<mlir::Attribute> getInherentAttr(const Properties &prop,
+                                                        StringRef name) {
+    return std::nullopt;
+  }
+  static void setInherentAttr(Properties &prop, StringRef name,
+                              mlir::Attribute value) {}
+  static void populateInherentAttrs(const Properties &prop,
+                                    NamedAttrList &attrs) {}
+  static LogicalResult
+  verifyInherentAttrs(OperationName opName, NamedAttrList &attrs,
+                      function_ref<InFlightDiagnostic()> getDiag) {
+    return success();
+  }
+};
+
+// A trivial supporting dialect to register the above operation.
+class TestOpPropertiesDialect : public Dialect {
+public:
+  MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(TestOpPropertiesDialect)
+  static constexpr StringLiteral getDialectNamespace() {
+    return StringLiteral("test_op_properties");
+  }
+  explicit TestOpPropertiesDialect(MLIRContext *context)
+      : Dialect(getDialectNamespace(), context,
+                TypeID::get<TestOpPropertiesDialect>()) {
+    addOperations<OpWithProperties>();
+  }
+};
+
+constexpr StringLiteral mlirSrc = R"mlir(
+    "test_op_properties.op_with_properties"()
+      <{a = -42 : i32,
+        b = -4.200000e+01 : f32,
+        array = array<i64: 40, 41>,
+        label = "bar foo"}> : () -> ()
+)mlir";
+
+TEST(OpPropertiesTest, Properties) {
+  MLIRContext context;
+  context.getOrLoadDialect<TestOpPropertiesDialect>();
+  ParserConfig config(&context);
+  // Parse the operation with some properties.
+  OwningOpRef<Operation *> op = parseSourceString(mlirSrc, config);
+  ASSERT_TRUE(op.get() != nullptr);
+  auto opWithProp = dyn_cast<OpWithProperties>(op.get());
+  ASSERT_TRUE(opWithProp);
+  {
+    std::string output;
+    llvm::raw_string_ostream os(output);
+    opWithProp.print(os);
+    ASSERT_STREQ("\"test_op_properties.op_with_properties\"() "
+                 "<{a = -42 : i32, "
+                 "array = array<i64: 40, 41>, "
+                 "b = -4.200000e+01 : f32, "
+                 "label = \"bar foo\"}> : () -> ()\n",
+                 os.str().c_str());
+  }
+  // Get a mutable reference to the properties for this operation and modify it
+  // in place one member at a time.
+  TestProperties &prop = opWithProp.getProperties();
+  prop.a = 42;
+  {
+    std::string output;
+    llvm::raw_string_ostream os(output);
+    opWithProp.print(os);
+    EXPECT_TRUE(StringRef(os.str()).contains("a = 42"));
+    EXPECT_TRUE(StringRef(os.str()).contains("b = -4.200000e+01"));
+    EXPECT_TRUE(StringRef(os.str()).contains("array = array<i64: 40, 41>"));
+    EXPECT_TRUE(StringRef(os.str()).contains("label = \"bar foo\""));
+  }
+  prop.b = 42.;
+  {
+    std::string output;
+    llvm::raw_string_ostream os(output);
+    opWithProp.print(os);
+    EXPECT_TRUE(StringRef(os.str()).contains("a = 42"));
+    EXPECT_TRUE(StringRef(os.str()).contains("b = 4.200000e+01"));
+    EXPECT_TRUE(StringRef(os.str()).contains("array = array<i64: 40, 41>"));
+    EXPECT_TRUE(StringRef(os.str()).contains("label = \"bar foo\""));
+  }
+  prop.array.push_back(42);
+  {
+    std::string output;
+    llvm::raw_string_ostream os(output);
+    opWithProp.print(os);
+    EXPECT_TRUE(StringRef(os.str()).contains("a = 42"));
+    EXPECT_TRUE(StringRef(os.str()).contains("b = 4.200000e+01"));
+    EXPECT_TRUE(StringRef(os.str()).contains("array = array<i64: 40, 41, 42>"));
+    EXPECT_TRUE(StringRef(os.str()).contains("label = \"bar foo\""));
+  }
+  prop.label = std::make_shared<std::string>("foo bar");
+  {
+    std::string output;
+    llvm::raw_string_ostream os(output);
+    opWithProp.print(os);
+    EXPECT_TRUE(StringRef(os.str()).contains("a = 42"));
+    EXPECT_TRUE(StringRef(os.str()).contains("b = 4.200000e+01"));
+    EXPECT_TRUE(StringRef(os.str()).contains("array = array<i64: 40, 41, 42>"));
+    EXPECT_TRUE(StringRef(os.str()).contains("label = \"foo bar\""));
+  }
+}
+
+// Test diagnostic emission when using invalid dictionary.
+TEST(OpPropertiesTest, FailedProperties) {
+  MLIRContext context;
+  context.getOrLoadDialect<TestOpPropertiesDialect>();
+  std::string diagnosticStr;
+  context.getDiagEngine().registerHandler([&](Diagnostic &diag) {
+    diagnosticStr += diag.str();
+    return success();
+  });
+
+  // Parse the operation with some properties.
+  ParserConfig config(&context);
+
+  // Parse an operation with invalid (incomplete) properties.
+  OwningOpRef<Operation *> owningOp =
+      parseSourceString("\"test_op_properties.op_with_properties\"() "
+                        "<{a = -42 : i32}> : () -> ()\n",
+                        config);
+  ASSERT_EQ(owningOp.get(), nullptr);
+  EXPECT_STREQ(
+      "invalid properties {a = -42 : i32} for op "
+      "test_op_properties.op_with_properties: expected FloatAttr for key `b`",
+      diagnosticStr.c_str());
+  diagnosticStr.clear();
+
+  owningOp = parseSourceString(mlirSrc, config);
+  Operation *op = owningOp.get();
+  ASSERT_TRUE(op != nullptr);
+  Location loc = op->getLoc();
+  auto opWithProp = dyn_cast<OpWithProperties>(op);
+  ASSERT_TRUE(opWithProp);
+
+  OperationState state(loc, op->getName());
+  Builder b{&context};
+  NamedAttrList attrs;
+  attrs.push_back(b.getNamedAttr("a", b.getStringAttr("foo")));
+  state.propertiesAttr = attrs.getDictionary(&context);
+  {
+    auto diag = op->emitError("setting properties failed: ");
+    auto result = state.setProperties(op, &diag);
+    EXPECT_TRUE(result.failed());
+  }
+  EXPECT_STREQ("setting properties failed: expected IntegerAttr for key `a`",
+               diagnosticStr.c_str());
+}
+
+TEST(OpPropertiesTest, DefaultValues) {
+  MLIRContext context;
+  context.getOrLoadDialect<TestOpPropertiesDialect>();
+  OperationState state(UnknownLoc::get(&context),
+                       "test_op_properties.op_with_properties");
+  Operation *op = Operation::create(state);
+  ASSERT_TRUE(op != nullptr);
+  {
+    std::string output;
+    llvm::raw_string_ostream os(output);
+    op->print(os);
+    EXPECT_TRUE(StringRef(os.str()).contains("a = -1"));
+    EXPECT_TRUE(StringRef(os.str()).contains("b = -1"));
+    EXPECT_TRUE(StringRef(os.str()).contains("array = array<i64: -33>"));
+  }
+  op->erase();
+}
+
+TEST(OpPropertiesTest, Cloning) {
+  MLIRContext context;
+  context.getOrLoadDialect<TestOpPropertiesDialect>();
+  ParserConfig config(&context);
+  // Parse the operation with some properties.
+  OwningOpRef<Operation *> op = parseSourceString(mlirSrc, config);
+  ASSERT_TRUE(op.get() != nullptr);
+  auto opWithProp = dyn_cast<OpWithProperties>(op.get());
+  ASSERT_TRUE(opWithProp);
+  Operation *clone = opWithProp->clone();
+
+  // Check that op and its clone prints equally
+  std::string opStr;
+  std::string cloneStr;
+  {
+    llvm::raw_string_ostream os(opStr);
+    op.get()->print(os);
+  }
+  {
+    llvm::raw_string_ostream os(cloneStr);
+    clone->print(os);
+  }
+  clone->erase();
+  EXPECT_STREQ(opStr.c_str(), cloneStr.c_str());
+}
+
+TEST(OpPropertiesTest, Equivalence) {
+  MLIRContext context;
+  context.getOrLoadDialect<TestOpPropertiesDialect>();
+  ParserConfig config(&context);
+  // Parse the operation with some properties.
+  OwningOpRef<Operation *> op = parseSourceString(mlirSrc, config);
+  ASSERT_TRUE(op.get() != nullptr);
+  auto opWithProp = dyn_cast<OpWithProperties>(op.get());
+  ASSERT_TRUE(opWithProp);
+  llvm::hash_code reference = OperationEquivalence::computeHash(opWithProp);
+  TestProperties &prop = opWithProp.getProperties();
+  prop.a = 42;
+  EXPECT_NE(reference, OperationEquivalence::computeHash(opWithProp));
+  prop.a = -42;
+  EXPECT_EQ(reference, OperationEquivalence::computeHash(opWithProp));
+  prop.b = 42.;
+  EXPECT_NE(reference, OperationEquivalence::computeHash(opWithProp));
+  prop.b = -42.;
+  EXPECT_EQ(reference, OperationEquivalence::computeHash(opWithProp));
+  prop.array.push_back(42);
+  EXPECT_NE(reference, OperationEquivalence::computeHash(opWithProp));
+  prop.array.pop_back();
+  EXPECT_EQ(reference, OperationEquivalence::computeHash(opWithProp));
+}
+
+TEST(OpPropertiesTest, getOrAddProperties) {
+  MLIRContext context;
+  context.getOrLoadDialect<TestOpPropertiesDialect>();
+  OperationState state(UnknownLoc::get(&context),
+                       "test_op_properties.op_with_properties");
+  // Test `getOrAddProperties` API on OperationState.
+  TestProperties &prop = state.getOrAddProperties<TestProperties>();
+  prop.a = 1;
+  prop.b = 2;
+  prop.array = {3, 4, 5};
+  Operation *op = Operation::create(state);
+  ASSERT_TRUE(op != nullptr);
+  {
+    std::string output;
+    llvm::raw_string_ostream os(output);
+    op->print(os);
+    EXPECT_TRUE(StringRef(os.str()).contains("a = 1"));
+    EXPECT_TRUE(StringRef(os.str()).contains("b = 2"));
+    EXPECT_TRUE(StringRef(os.str()).contains("array = array<i64: 3, 4, 5>"));
+  }
+  op->erase();
+}
+
+} // namespace

mlir/unittests/IR/OperationSupportTest.cpp

 using namespace mlir;
 using namespace mlir::detail;
 
 static Operation *createOp(MLIRContext *context,
                            ArrayRef<Value> operands = std::nullopt,
                            ArrayRef<Type> resultTypes = std::nullopt,
                            unsigned int numRegions = 0) {
   context->allowUnregisteredDialects();
-  return Operation::create(UnknownLoc::get(context),
-                           OperationName("foo.bar", context), resultTypes,
-                           operands, std::nullopt, std::nullopt, numRegions);
+  return Operation::create(
+      UnknownLoc::get(context), OperationName("foo.bar", context), resultTypes,
+      operands, std::nullopt, nullptr, std::nullopt, numRegions);
 }
 
 namespace {
 TEST(OperandStorageTest, NonResizable) {
   MLIRContext context;
   Builder builder(&context);
 
   Operation *useOp =

mlir/unittests/Transforms/DialectConversion.cpp

 //===- DialectConversion.cpp - Dialect conversion unit tests --------------===//
 //
 // 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/Transforms/DialectConversion.h"
 #include "gtest/gtest.h"
 
 using namespace mlir;
 
 static Operation *createOp(MLIRContext *context) {
   context->allowUnregisteredDialects();
-  return Operation::create(UnknownLoc::get(context),
-                           OperationName("foo.bar", context), std::nullopt,
-                           std::nullopt, std::nullopt, std::nullopt, 0);
+  return Operation::create(
+      UnknownLoc::get(context), OperationName("foo.bar", context), std::nullopt,
+      std::nullopt, std::nullopt, /*properties=*/nullptr, std::nullopt, 0);
 }
 
 namespace {
 struct DummyOp {
   MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(DummyOp)
 
   static StringRef getOperationName() { return "foo.bar"; }
 };