Multi-Level IR Compiler Framework

Defining Dialect Attributes and Types

This document describes how to define dialect attributes and types.

LangRef Refresher 

Before diving into how to define these constructs, below is a quick refresher from the MLIR LangRef.


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 arith.cmpi operation, or the underlying value of a arith.constant operation. Each operation has an attribute dictionary, which associates a set of attribute names to attribute values.


Every SSA value, such as operation results or block arguments, in MLIR has a type defined by the type system. MLIR has an open type system with no fixed list of types, and there are no restrictions on the abstractions they represent. For example, take the following Arithmetic AddI operation:

  %result = arith.addi %lhs, %rhs : i64

It takes two input SSA values (%lhs and %rhs), and returns a single SSA value (%result). The inputs and outputs of this operation are of type i64, which is an instance of the Builtin IntegerType.

Attributes and Types 

The C++ Attribute and Type classes in MLIR (like Ops, and many other things) are value-typed. This means that instances of Attribute or Type are passed around by-value, as opposed to by-pointer or by-reference. The Attribute and Type classes act as wrappers around internal storage objects that are uniqued within an instance of an MLIRContext.

The structure for defining Attributes and Types is nearly identical, with only a few differences depending on the context. As such, a majority of this document describes the process for defining both Attributes and Types side-by-side with examples for both. If necessary, a section will explicitly call out any distinct differences.

Adding a new Attribute or Type definition 

As described above, C++ Attribute and Type objects in MLIR are value-typed and essentially function as helpful wrappers around an internal storage object that holds the actual data for the type. Similarly to Operations, Attributes and Types are defined declaratively via TableGen; a generic language with tooling to maintain records of domain-specific information. It is highly recommended that users review the TableGen Programmer’s Reference for an introduction to its syntax and constructs.

Starting the definition of a new attribute or type simply requires adding a specialization for either the AttrDef or TypeDef class respectively. Instances of the classes correspond to unqiue Attribute or Type classes.

Below show cases an example Attribute and Type definition. We generally recommend defining Attribute and Type classes in different .td files to better encapsulate the different constructs, and define a proper layering between them. This recommendation extends to all of the MLIR constructs, including Interfaces, Operations, etc.

// Include the definition of the necessary tablegen constructs for defining
// our types.
include "mlir/IR/"

// It's common to define a base classes for types in the same dialect. This
// removes the need to pass in the dialect for each type, and can also be used
// to define a few fields ahead of time.
class MyDialect_Type<string name, string typeMnemonic, list<Trait> traits = []>
    : TypeDef<My_Dialect, name, traits> {
  let mnemonic = typeMnemonic;

// Here is a simple definition of an "integer" type, with a width parameter.
def My_IntegerType : MyDialect_Type<"Integer", "int"> {
  let summary = "Integer type with arbitrary precision up to a fixed limit";
  let description = [{
    Integer types have a designated bit width.
  /// Here we defined a single parameter for the type, which is the bitwidth.
  let parameters = (ins "unsigned":$width);

  /// Here we define the textual format of the type declaratively, which will
  /// automatically generate parser and printer logic. This will allow for
  /// instances of the type to be output as, for example:
  ///    !<10> // a 10-bit integer.
  let assemblyFormat = "`<` $width `>`";

  /// Indicate that our type will add additional verification to the parameters.
  let genVerifyDecl = 1;

Below is an example of an Attribute:

// Include the definition of the necessary tablegen constructs for defining
// our attributes.
include "mlir/IR/"

// It's common to define a base classes for attributes in the same dialect. This
// removes the need to pass in the dialect for each attribute, and can also be used
// to define a few fields ahead of time.
class MyDialect_Attr<string name, string attrMnemonic, list<Trait> traits = []>
    : AttrDef<My_Dialect, name, traits> {
  let mnemonic = attrMnemonic;

// Here is a simple definition of an "integer" attribute, with a type and value parameter.
def My_IntegerAttr : MyDialect_Attr<"Integer", "int"> {
  let summary = "An Attribute containing a integer value";
  let description = [{
    An integer attribute is a literal attribute that represents an integral
    value of the specified integer type.
  /// Here we've defined two parameters, one is a "self" type parameter, and the
  /// other is the integer value of the attribute. The self type parameter is
  /// specially handled by the assembly format.
  let parameters = (ins AttributeSelfTypeParameter<"">:$type, "APInt":$value);

  /// Here we've defined a custom builder for the type, that removes the need to pass
  /// in an MLIRContext instance; as it can be infered from the `type`.
  let builders = [
    AttrBuilderWithInferredContext<(ins "Type":$type,
                                        "const APInt &":$value), [{
      return $_get(type.getContext(), type, value);

  /// Here we define the textual format of the attribute declaratively, which will
  /// automatically generate parser and printer logic. This will allow for
  /// instances of the attribute to be output as, for example:
  ///<50> : !<32> // a 32-bit integer of value 50.
  /// Note that the self type parameter is not included in the assembly format.
  /// Its value is derived from the optional trailing type on all attributes.
  let assemblyFormat = "`<` $value `>`";

  /// Indicate that our attribute will add additional verification to the parameters.
  let genVerifyDecl = 1;

  /// Indicate to the ODS generator that we do not want the default builders,
  /// as we have defined our own simpler ones.
  let skipDefaultBuilders = 1;

Class Name 

The name of the C++ class which gets generated defaults to <classParamName>Attr or <classParamName>Type for attributes and types respectively. In the examples above, this was the name template parameter that was provided to MyDialect_Attr and MyDialect_Type. For the definitions we added above, we would get C++ classes named IntegerType and IntegerAttr respectively. This can be explicitly overridden via the cppClassName field.


The summary and description fields allow for providing user documentation for the attribute or type. The summary field expects a simple single-line string, with the description field used for long and extensive documentation. This documentation can be used to generate markdown documentation for the dialect and is used by upstream MLIR dialects.


The mnemonic field, i.e. the template parameters attrMnemonic and typeMnemonic we specified above, are used to specify a name for use during parsing. This allows for more easily dispatching to the current attribute or type class when parsing IR. This field is generally optional, and custom parsing/printing logic can be added without defining it, though most classes will want to take advantage of the convenience it provides. This is why we added it as a template parameter in the examples above.


The parameters field is a variable length list containing the attribute or type’s parameters. If no parameters are specified (the default), this type is considered a singleton type (meaning there is only one possible instance). Parameters in this list take the form: "c++Type":$paramName. Parameter types with a C++ type that requires allocation when constructing the storage instance in the context require one of the following:

  • Utilize the AttrParameter or TypeParameter classes instead of the raw “c++Type” string. This allows for providing custom allocation code when using that parameter. StringRefParameter and ArrayRefParameter are examples of common parameter types that require allocation.
  • Set the genAccessors field to 1 (the default) to generate accessor methods for each parameter (e.g. int getWidth() const in the Type example above).
  • Set the hasCustomStorageConstructor field to 1 to generate a storage class that only declares the constructor, allowing for you to specialize it with whatever allocation code necessary.

AttrParameter, TypeParameter, and AttrOrTypeParameter 

As hinted at above, these classes allow for specifying parameter types with additional functionality. This is generally useful for complex parameters, or those with additional invariants that prevent using the raw C++ class. Examples include documentation (e.g. the summary and syntax field), the C++ type, a custom allocator to use in the storage constructor method, a custom comparator to decide if two instances of the parameter type are equal, etc. As the names may suggest, AttrParameter is intended for parameters on Attributes, TypeParameter for Type parameters, and AttrOrTypeParameters for either.

Below is an easy parameter pitfall, and highlights when to use these parameter classes.

let parameters = (ins "ArrayRef<int>":$dims);

The above seems innocuous, but it is often a bug! The default storage constructor blindly copies parameters by value. It does not know anything about the types, meaning that the data of this ArrayRef will be copied as-is and is likely to lead to use-after-free errors when using the created Attribute or Type if the underlying does not have a lifetime exceeding that of the MLIRContext. If the lifetime of the data can’t be guaranteed, the ArrayRef<int> requires allocation to ensure that its elements reside within the MLIRContext, e.g. with dims = allocator.copyInto(dims).

Here is a simple example for the exact situation above:

def ArrayRefIntParam : TypeParameter<"::llvm::ArrayRef<int>", "Array of int"> {
  let allocator = "$_dst = $_allocator.copyInto($_self);";

The parameter can then be used as so:

let parameters = (ins ArrayRefIntParam:$dims);

Below contains descriptions for other various available fields:

The allocator code block has the following substitutions:

  • $_allocator is the TypeStorageAllocator in which to allocate objects.
  • $_dst is the variable in which to place the allocated data.

The comparator code block has the following substitutions:

  • $_lhs is an instance of the parameter type.
  • $_rhs is an instance of the parameter type.

MLIR includes several specialized classes for common situations:

  • APFloatParameter for APFloats.

  • StringRefParameter<descriptionOfParam> for StringRefs.

  • ArrayRefParameter<arrayOf, descriptionOfParam> for ArrayRefs of value types.

  • SelfAllocationParameter<descriptionOfParam> for C++ classes which contain a method called allocateInto(StorageAllocator &allocator) to allocate itself into allocator.

  • ArrayRefOfSelfAllocationParameter<arrayOf, descriptionOfParam> for arrays of objects which self-allocate as per the last specialization.

  • AttributeSelfTypeParameter is a special AttrParameter that represents parameters derived from the optional trailing type on attributes.


Similarly to operations, Attribute and Type classes may attach Traits that provide additional mixin methods and other data. Traits may be attached via the trailing template argument, i.e. the traits list parameter in the example above. See the main Trait documentation for more information on defining and using traits.


Attribute and Type classes may attach Interfaces to provide an virtual interface into the Attribute or Type. Interfaces are added in the same way as Traits, by using the traits list template parameter of the AttrDef or TypeDef. See the main Interface documentation for more information on defining and using interfaces.


For each attribute or type, there are a few builders(get/getChecked) automatically generated based on the parameters of the type. These are used to construct instances of the corresponding attribute or type. For example, given the following definition:

def MyAttrOrType : ... {
  let parameters = (ins "int":$intParam);

The following builders are generated:

// Builders are named `get`, and return a new instance for a given set of parameters.
static MyAttrOrType get(MLIRContext *context, int intParam);

// If `genVerifyDecl` is set to 1, the following method is also generated. This method
// is similar to `get`, but is failable and on error will return nullptr.
static MyAttrOrType getChecked(function_ref<InFlightDiagnostic()> emitError,
                               MLIRContext *context, int intParam);

If these autogenerated methods are not desired, such as when they conflict with a custom builder method, the skipDefaultBuilders field may be set to 1 to signal that the default builders should not be generated.

Custom builder methods 

The default builder methods may cover a majority of the simple cases related to construction, but when they cannot satisfy all of an attribute or type’s needs, additional builders may be defined via the builders field. The builders field is a list of custom builders, either using TypeBuilder for types or AttrBuilder for attributes, that are added to the attribute or type class. The following will showcase several examples for defining builders for a custom type MyType, the process is the same for attributes except that attributes use AttrBuilder instead of TypeBuilder.

def MyType : ... {
  let parameters = (ins "int":$intParam);

  let builders = [
    TypeBuilder<(ins "int":$intParam)>,
    TypeBuilder<(ins CArg<"int", "0">:$intParam)>,
    TypeBuilder<(ins CArg<"int", "0">:$intParam), [{
      // Write the body of the `get` builder inline here.
      return Base::get($_ctxt, intParam);
    TypeBuilderWithInferredContext<(ins "Type":$typeParam), [{
      // This builder states that it can infer an MLIRContext instance from
      // its arguments.
      return Base::get(typeParam.getContext(), ...);
    TypeBuilder<(ins "int":$intParam), [{}], "IntegerType">,

In this example, we provide several different convenience builders that are useful in different scenarios. The ins prefix is common to many function declarations in ODS, which use a TableGen dag. What follows is a comma-separated list of types (quoted string or CArg) and names prefixed with the $ sign. The use of CArg allows for providing a default value to that argument. Let’s take a look at each of these builders individually

The first builder will generate the declaration of a builder method that looks like:

  let builders = [
    TypeBuilder<(ins "int":$intParam)>,
class MyType : /*...*/ {
  static MyType get(::mlir::MLIRContext *context, int intParam);

This builder is identical to the one that will be automatically generated for MyType. The context parameter is implicitly added by the generator, and is used when building the Type instance (with Base::get). The distinction here is that we can provide the implementation of this get method. With this style of builder definition only the declaration is generated, the implementor of MyType will need to provide a definition of MyType::get.

The second builder will generate the declaration of a builder method that looks like:

  let builders = [
    TypeBuilder<(ins CArg<"int", "0">:$intParam)>,
class MyType : /*...*/ {
  static MyType get(::mlir::MLIRContext *context, int intParam = 0);

The constraints here are identical to the first builder example except for the fact that intParam now has a default value attached.

The third builder will generate the declaration of a builder method that looks like:

  let builders = [
    TypeBuilder<(ins CArg<"int", "0">:$intParam), [{
      // Write the body of the `get` builder inline here.
      return Base::get($_ctxt, intParam);
class MyType : /*...*/ {
  static MyType get(::mlir::MLIRContext *context, int intParam = 0);

MyType MyType::get(::mlir::MLIRContext *context, int intParam) {
  // Write the body of the `get` builder inline here.
  return Base::get(context, intParam);

This is identical to the second builder example. The difference is that now, a definition for the builder method will be generated automatically using the provided code block as the body. When specifying the body inline, $_ctxt may be used to access the MLIRContext * parameter.

The fourth builder will generate the declaration of a builder method that looks like:

  let builders = [
    TypeBuilderWithInferredContext<(ins "Type":$typeParam), [{
      // This builder states that it can infer an MLIRContext instance from
      // its arguments.
      return Base::get(typeParam.getContext(), ...);
class MyType : /*...*/ {
  static MyType get(Type typeParam);

MyType MyType::get(Type typeParam) {
  // This builder states that it can infer an MLIRContext instance from its
  // arguments.
  return Base::get(typeParam.getContext(), ...);

In this builder example, the main difference from the third builder example there is that the MLIRContext parameter is no longer added. This is because the builder used TypeBuilderWithInferredContext implies that the context parameter is not necessary as it can be inferred from the arguments to the builder.

The fifth builder will generate the declaration of a builder method with a custom return type, like:

  let builders = [
    TypeBuilder<(ins "int":$intParam), [{}], "IntegerType">,
class MyType : /*...*/ {
  static IntegerType get(::mlir::MLIRContext *context, int intParam);


This generates a builder declaration the same as the first three examples, but the return type of the builder is user-specified instead of the attribute or type class. This is useful for defining builders of attributes and types that may fold or canonicalize on construction.

Parsing and Printing 

If a mnemonic was specified, the hasCustomAssemblyFormat and assemblyFormat fields may be used to specify the assembly format of an attribute or type. Attributes and Types with no parameters need not use either of these fields, in which case the syntax for the Attribute or Type is simply the mnemonic.

For each dialect, two “dispatch” functions will be created: one for parsing and one for printing. These static functions placed alongside the class definitions and have the following function signatures:

static ParseResult generatedAttributeParser(DialectAsmParser& parser, StringRef *mnemonic, Type attrType, Attribute &result);
static LogicalResult generatedAttributePrinter(Attribute attr, DialectAsmPrinter& printer);

static ParseResult generatedTypeParser(DialectAsmParser& parser, StringRef *mnemonic, Type &result);
static LogicalResult generatedTypePrinter(Type type, DialectAsmPrinter& printer);

The above functions should be added to the respective in your Dialect::printType and Dialect::parseType methods, or consider using the useDefaultAttributePrinterParser and useDefaultTypePrinterParser ODS Dialect options if all attributes or types define a mnemonic.

The mnemonic, hasCustomAssemblyFormat, and assemblyFormat fields are optional. If none are defined, the generated code will not include any parsing or printing code and omit the attribute or type from the dispatch functions above. In this case, the dialect author is responsible for parsing/printing in the respective Dialect::parseAttribute/Dialect::printAttribute and Dialect::parseType/Dialect::printType methods.

Using hasCustomAssemblyFormat 

Attributes and types defined in ODS with a mnemonic can define an hasCustomAssemblyFormat to specify custom parsers and printers defined in C++. When set to 1 a corresponding parse and print method will be declared on the Attribute or Type class to be defined by the user.

For Types, these methods will have the form:

  • static Type MyType::parse(AsmParser &parser)

  • Type MyType::print(AsmPrinter &p) const

For Attributes, these methods will have the form:

  • static Attribute MyAttr::parse(AsmParser &parser, Type attrType)

  • Attribute MyAttr::print(AsmPrinter &p) const

Using assemblyFormat 

Attributes and types defined in ODS with a mnemonic can define an assemblyFormat to declaratively describe custom parsers and printers. The assembly format consists of literals, variables, and directives.

  • A literal is a keyword or valid punctuation enclosed in backticks, e.g. `keyword` or `<`.
  • A variable is a parameter name preceded by a dollar sign, e.g. $param0, which captures one attribute or type parameter.
  • A directive is a keyword followed by an optional argument list that defines special parser and printer behaviour.
// An example type with an assembly format.
def MyType : TypeDef<My_Dialect, "MyType"> {
  // Define a mnemonic to allow the dialect's parser hook to call into the
  // generated parser.
  let mnemonic = "my_type";

  // Define two parameters whose C++ types are indicated in string literals.
  let parameters = (ins "int":$count, "AffineMap":$map);

  // Define the assembly format. Surround the format with less `<` and greater
  // `>` so that MLIR's printer uses the pretty format.
  let assemblyFormat = "`<` $count `,` `map` `=` $map `>`";

The declarative assembly format for MyType results in the following format in the IR:

!my_dialect.my_type<42, map = affine_map<(i, j) -> (j, i)>>
Parameter Parsing and Printing 

For many basic parameter types, no additional work is needed to define how these parameters are parsed or printed.

  • The default printer for any parameter is $_printer << $_self, where $_self is the C++ value of the parameter and $_printer is an AsmPrinter.
  • The default parser for a parameter is FieldParser<$cppClass>::parse($_parser), where $cppClass is the C++ type of the parameter and $_parser is an AsmParser.

Printing and parsing behaviour can be added to additional C++ types by overloading these functions or by defining a parser and printer in an ODS parameter class.

Example of overloading:

using MyParameter = std::pair<int, int>;

AsmPrinter &operator<<(AsmPrinter &printer, MyParameter param) {
  printer << param.first << " * " << param.second;

template <> struct FieldParser<MyParameter> {
  static FailureOr<MyParameter> parse(AsmParser &parser) {
    int a, b;
    if (parser.parseInteger(a) || parser.parseStar() ||
      return failure();
    return MyParameter(a, b);

Example of using ODS parameter classes:

def MyParameter : TypeParameter<"std::pair<int, int>", "pair of ints"> {
  let printer = [{ $_printer << $_self.first << " * " << $_self.second }];
  let parser = [{ [&] -> FailureOr<std::pair<int, int>> {
    int a, b;
    if ($_parser.parseInteger(a) || $_parser.parseStar() ||
      return failure();
    return std::make_pair(a, b);
  }() }];

A type using this parameter with the assembly format `<` $myParam `>` will look as follows in the IR:

!my_dialect.my_type<42 * 24>
Non-POD Parameters 

Parameters that aren’t plain-old-data (e.g. references) may need to define a cppStorageType to contain the data until it is copied into the allocator. For example, StringRefParameter uses std::string as its storage type, whereas ArrayRefParameter uses SmallVector as its storage type. The parsers for these parameters are expected to return FailureOr<$cppStorageType>.

To add a custom conversion between the cppStorageType and the C++ type of the parameter, parameters can override convertFromStorage, which by default is "$_self" (i.e., it attempts an implicit conversion from cppStorageType).

Optional Parameters 

Optional parameters in the assembly format can be indicated by setting isOptional. The C++ type of an optional parameter is required to satisfy the following requirements:

  • is default-constructible
  • is contextually convertible to bool
  • only the default-constructed value is false

The parameter parser should return the default-constructed value to indicate “no value present”. The printer will guard on the presence of a value to print the parameter.

If a value was not parsed for an optional parameter, then the parameter will be set to its default-constructed C++ value. For example, Optional<int> will be set to llvm::None and Attribute will be set to nullptr.

Only optional parameters or directives that only capture optional parameters can be used in optional groups. An optional group is a set of elements optionally printed based on the presence of an anchor. Suppose parameter a is an IntegerAttr.

( `(` $a^ `)` ) : (`x`)?

In the above assembly format, if a is present (non-null), then it will be printed as (5 : i32). If it is not present, it will be x. Directives that are used inside optional groups are allowed only if all captured parameters are also optional.

Default-Valued Parameters 

Optional parameters can be given default values by setting defaultValue, a string of the C++ default value, or by using DefaultValuedParameter. If a value for the parameter was not encountered during parsing, it is set to this default value. If a parameter is equal to its default value, it is not printed. The comparator field of the parameter is used, but if one is not specified, the equality operator is used.

For example:

let parameters = (ins DefaultValuedParameter<"Optional<int>", "5">:$a)
let mnemonic = "default_valued";
let assemblyFormat = "(`<` $a^ `>`)?";

Which will look like:

!test.default_valued     // a = 5
!test.default_valued<10> // a = 10

For optional Attribute or Type parameters, the current MLIR context is available through $_ctxt. E.g.

DefaultValuedParameter<"IntegerType", "IntegerType::get($_ctxt, 32)">

The value of parameters that appear before the default-valued parameter in the parameter declaration list are available as substitutions. E.g.

let parameters = (ins
  DefaultValuedParameter<"Type", "$value.getType()">:$type
Attribute Self Type Parameter 

An attribute optionally has a trailing type after the assembly format of the attribute value itself. MLIR parses over the attribute value and optionally parses a colon-type before passing the Type into the dialect parser hook.

dialect-attribute  ::= `#` dialect-namespace `<` attr-data `>`
                       (`:` type)?
                     | `#` alias-name pretty-dialect-sym-body? (`:` type)?

AttributeSelfTypeParameter is an attribute parameter specially handled by the assembly format generator. Only one such parameter can be specified, and its value is derived from the trailing type. This parameter’s default value is NoneType::get($_ctxt).

In order for the type to be printed by MLIR, however, the attribute must implement TypedAttrInterface. For example,

// This attribute has only a self type parameter.
def MyExternAttr : AttrDef<MyDialect, "MyExtern", [TypedAttrInterface]> {
  let parameters = (AttributeSelfTypeParameter<"">:$type);
  let mnemonic = "extern";
  let assemblyFormat = "";

This attribute can look like:

#my_dialect.extern // none
#my_dialect.extern : i32
#my_dialect.extern : tensor<4xi32>
#my_dialect.extern : !my_dialect.my_type
Assembly Format Directives 

Attribute and type assembly formats have the following directives:

  • params: capture all parameters of an attribute or type.
  • qualified: mark a parameter to be printed with its leading dialect and mnemonic.
  • struct: generate a “struct-like” parser and printer for a list of key-value pairs.
  • custom: dispatch a call to user-define parser and printer functions
  • ref: in a custom directive, references a previously bound variable
params Directive 

This directive is used to refer to all parameters of an attribute or type. When used as a top-level directive, params generates a parser and printer for a comma-separated list of the parameters. For example:

def MyPairType : TypeDef<My_Dialect, "MyPairType"> {
  let parameters = (ins "int":$a, "int":$b);
  let mnemonic = "pair";
  let assemblyFormat = "`<` params `>`";

In the IR, this type will appear as:

!my_dialect.pair<42, 24>

The params directive can also be passed to other directives, such as struct, as an argument that refers to all parameters in place of explicitly listing all parameters as variables.

qualified Directive 

This directive can be used to wrap attribute or type parameters such that they are printed in a fully qualified form, i.e., they include the dialect name and mnemonic prefix.

For example:

def OuterType : TypeDef<My_Dialect, "MyOuterType"> {
  let parameters = (ins MyPairType:$inner);
  let mnemonic = "outer";
  let assemblyFormat = "`<` pair `:` $inner `>`";
def OuterQualifiedType : TypeDef<My_Dialect, "MyOuterQualifiedType"> {
  let parameters = (ins MyPairType:$inner);
  let mnemonic = "outer_qual";
  let assemblyFormat = "`<` pair `:` qualified($inner) `>`";

In the IR, the types will appear as:

!my_dialect.outer<pair : <42, 24>>
!my_dialect.outer_qual<pair : !mydialect.pair<42, 24>>

If optional parameters are present, they are not printed in the parameter list if they are not present.

struct Directive 

The struct directive accepts a list of variables to capture and will generate a parser and printer for a comma-separated list of key-value pairs. If an optional parameter is included in the struct, it can be elided. The variables are printed in the order they are specified in the argument list but can be parsed in any order. For example:

def MyStructType : TypeDef<My_Dialect, "MyStructType"> {
  let parameters = (ins StringRefParameter<>:$sym_name,
                        "int":$a, "int":$b, "int":$c);
  let mnemonic = "struct";
  let assemblyFormat = "`<` $sym_name `->` struct($a, $b, $c) `>`";

In the IR, this type can appear with any permutation of the order of the parameters captured in the directive.

!my_dialect.struct<"foo" -> a = 1, b = 2, c = 3>
!my_dialect.struct<"foo" -> b = 2, c = 3, a = 1>

Passing params as the only argument to struct makes the directive capture all the parameters of the attribute or type. For the same type above, an assembly format of `<` struct(params) `>` will result in:

!my_dialect.struct<b = 2, sym_name = "foo", c = 3, a = 1>

The order in which the parameters are printed is the order in which they are declared in the attribute’s or type’s parameter list.

custom and ref directive 

The custom directive is used to dispatch calls to user-defined printer and parser functions. For example, suppose we had the following type:

let parameters = (ins "int":$foo, "int":$bar);
let assemblyFormat = "custom<Foo>($foo) custom<Bar>($bar, ref($foo))";

The custom directive custom<Foo>($foo) will in the parser and printer respectively generate calls to:

LogicalResult parseFoo(AsmParser &parser, FailureOr<int> &foo);
void printFoo(AsmPrinter &printer, int foo);

A previously bound variable can be passed as a parameter to a custom directive by wrapping it in a ref directive. In the previous example, $foo is bound by the first directive. The second directive references it and expects the following printer and parser signatures:

LogicalResult parseBar(AsmParser &parser, FailureOr<int> &bar, int foo);
void printBar(AsmPrinter &printer, int bar, int foo);

More complex C++ types can be used with the custom directive. The only caveat is that the parameter for the parser must use the storage type of the parameter. For example, StringRefParameter expects the parser and printer signatures as:

LogicalResult parseStringParam(AsmParser &parser,
                               FailureOr<std::string> &value);
void printStringParam(AsmPrinter &printer, StringRef value);

The custom parser is considered to have failed if it returns failure or if any bound parameters have failure values afterwards.


If the genVerifyDecl field is set, additional verification methods are generated on the class.

  • static LogicalResult verify(function_ref<InFlightDiagnostic()> emitError, parameters...)

These methods are used to verify the parameters provided to the attribute or type class on construction, and emit any necessary diagnostics. This method is automatically invoked from the builders of the attribute or type class.

  • AttrOrType getChecked(function_ref<InFlightDiagnostic()> emitError, parameters...)

As noted in the Builders section, these methods are companions to get builders that are failable. If the verify invocation fails when these methods are called, they return nullptr instead of asserting.

Storage Classes 

Somewhat alluded to in the sections above is the concept of a “storage class” (often abbreviated to “storage”). Storage classes contain all of the data necessary to construct and unique a attribute or type instance. These classes are the “immortal” objects that get uniqued within an MLIRContext and get wrapped by the Attribute and Type classes. Every Attribute or Type class has a corresponding storage class, that can be accessed via the protected getImpl() method.

In most cases the storage class is auto generated, but if necessary it can be manually defined by setting the genStorageClass field to 0. The name and namespace (defaults to detail) can additionally be controlled via the The storageClass and storageNamespace fields.

Defining a storage class 

User defined storage classes must adhere to the following:

  • Inherit from the base type storage class of AttributeStorage or TypeStorage respectively.
  • Define a type alias, KeyTy, that maps to a type that uniquely identifies an instance of the derived type. For example, this could be a std::tuple of all of the storage parameters.
  • Provide a construction method that is used to allocate a new instance of the storage class.
    • static Storage *construct(StorageAllocator &allocator, const KeyTy &key)
  • Provide a comparison method between an instance of the storage and the KeyTy.
    • bool operator==(const KeyTy &) const
  • Provide a method to generate the KeyTy from a list of arguments passed to the uniquer when building an Attribute or Type. (Note: This is only necessary if the KeyTy cannot be default constructed from these arguments).
    • static KeyTy getKey(Args...&& args)
  • Provide a method to hash an instance of the KeyTy. (Note: This is not necessary if an llvm::DenseMapInfo<KeyTy> specialization exists)
    • static llvm::hash_code hashKey(const KeyTy &)

Let’s look at an example:

/// Here we define a storage class for a ComplexType, that holds a non-zero
/// integer and an integer type.
struct ComplexTypeStorage : public TypeStorage {
  ComplexTypeStorage(unsigned nonZeroParam, Type integerType)
      : nonZeroParam(nonZeroParam), integerType(integerType) {}

  /// The hash key for this storage is a pair of the integer and type params.
  using KeyTy = std::pair<unsigned, Type>;

  /// Define the comparison function for the key type.
  bool operator==(const KeyTy &key) const {
    return key == KeyTy(nonZeroParam, integerType);

  /// Define a hash function for the key type.
  /// Note: This isn't necessary because std::pair, unsigned, and Type all have
  /// hash functions already available.
  static llvm::hash_code hashKey(const KeyTy &key) {
    return llvm::hash_combine(key.first, key.second);

  /// Define a construction function for the key type.
  /// Note: This isn't necessary because KeyTy can be directly constructed with
  /// the given parameters.
  static KeyTy getKey(unsigned nonZeroParam, Type integerType) {
    return KeyTy(nonZeroParam, integerType);

  /// Define a construction method for creating a new instance of this storage.
  static ComplexTypeStorage *construct(StorageAllocator &allocator, const KeyTy &key) {
    return new (allocator.allocate<ComplexTypeStorage>())
        ComplexTypeStorage(key.first, key.second);

  /// The parametric data held by the storage class.
  unsigned nonZeroParam;
  Type integerType;

Mutable attributes and types 

Attributes and Types are immutable objects uniqued within an MLIRContext. That being said, some parameters may be treated as “mutable” and modified after construction. Mutable parameters should be reserved for parameters that can not be reasonably initialized during construction time. Given the mutable component, these parameters do not take part in the uniquing of the Attribute or Type.

TODO: Mutable parameters are currently not supported in the declarative specification of attributes and types, and thus requires defining the Attribute or Type class in C++.

Defining a mutable storage 

In addition to the base requirements for a storage class, instances with a mutable component must additionally adhere to the following:

  • The mutable component must not participate in the storage KeyTy.
  • Provide a mutation method that is used to modify an existing instance of the storage. This method modifies the mutable component based on arguments, using allocator for any newly dynamically-allocated storage, and indicates whether the modification was successful.
    • LogicalResult mutate(StorageAllocator &allocator, Args ...&& args)

Let’s define a simple storage for recursive types, where a type is identified by its name and may contain another type including itself.

/// Here we define a storage class for a RecursiveType that is identified by its
/// name and contains another type.
struct RecursiveTypeStorage : public TypeStorage {
  /// The type is uniquely identified by its name. Note that the contained type
  /// is _not_ a part of the key.
  using KeyTy = StringRef;

  /// Construct the storage from the type name. Explicitly initialize the
  /// containedType to nullptr, which is used as marker for the mutable
  /// component being not yet initialized.
  RecursiveTypeStorage(StringRef name) : name(name), containedType(nullptr) {}

  /// Define the comparison function.
  bool operator==(const KeyTy &key) const { return key == name; }

  /// Define a construction method for creating a new instance of the storage.
  static RecursiveTypeStorage *construct(StorageAllocator &allocator,
                                         const KeyTy &key) {
    // Note that the key string is copied into the allocator to ensure it
    // remains live as long as the storage itself.
    return new (allocator.allocate<RecursiveTypeStorage>())

  /// Define a mutation method for changing the type after it is created. In
  /// many cases, we only want to set the mutable component once and reject
  /// any further modification, which can be achieved by returning failure from
  /// this function.
  LogicalResult mutate(StorageAllocator &, Type body) {
    // If the contained type has been initialized already, and the call tries
    // to change it, reject the change.
    if (containedType && containedType != body)
      return failure();

    // Change the body successfully.
    containedType = body;
    return success();

  StringRef name;
  Type containedType;

Type class definition 

Having defined the storage class, we can define the type class itself. Type::TypeBase provides a mutate method that forwards its arguments to the mutate method of the storage and ensures the mutation happens safely.

class RecursiveType : public Type::TypeBase<RecursiveType, Type,
                                            RecursiveTypeStorage> {
  /// Inherit parent constructors.
  using Base::Base;

  /// Creates an instance of the Recursive type. This only takes the type name
  /// and returns the type with uninitialized body.
  static RecursiveType get(MLIRContext *ctx, StringRef name) {
    // Call into the base to get a uniqued instance of this type. The parameter
    // (name) is passed after the context.
    return Base::get(ctx, name);

  /// Now we can change the mutable component of the type. This is an instance
  /// method callable on an already existing RecursiveType.
  void setBody(Type body) {
    // Call into the base to mutate the type.
    LogicalResult result = Base::mutate(body);

    // Most types expect the mutation to always succeed, but types can implement
    // custom logic for handling mutation failures.
    assert(succeeded(result) &&
           "attempting to change the body of an already-initialized type");

    // Avoid unused-variable warning when building without assertions.
    (void) result;

  /// Returns the contained type, which may be null if it has not been
  /// initialized yet.
  Type getBody() { return getImpl()->containedType; }

  /// Returns the name.
  StringRef getName() { return getImpl()->name; }

Extra declarations 

The declarative Attribute and Type definitions try to auto-generate as much logic and methods as possible. With that said, there will always be long-tail cases that won’t be covered. For such cases, extraClassDeclaration and extraClassDefinition can be used. Code within the extraClassDeclaration field will be copied literally to the generated C++ Attribute or Type class. Code within extraClassDefinition will be added to the generated source file inside the class’s C++ namespace. The substitution $cppClass will be replaced by the Attribute or Type’s C++ class name.

Note that these are mechanisms intended for long-tail cases by power users; for not-yet-implemented widely-applicable cases, improving the infrastructure is preferable.

Registering with the Dialect 

Once the attributes and types have been defined, they must then be registered with the parent Dialect. This is done via the addAttributes and addTypes methods. Note that when registering, the full definition of the storage classes must be visible.

void MyDialect::initialize() {
    /// Add the defined attributes to the dialect.
#include "MyDialect/"

    /// Add the defined types to the dialect.
#include "MyDialect/"