MLIR  19.0.0git
Arith.h
Go to the documentation of this file.
1 //===- Arith.h - Arith dialect ------------------------------------*- C++-*-==//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #ifndef MLIR_DIALECT_ARITH_IR_ARITH_H_
10 #define MLIR_DIALECT_ARITH_IR_ARITH_H_
11 
13 #include "mlir/IR/Dialect.h"
14 #include "mlir/IR/OpDefinition.h"
21 #include "llvm/ADT/StringExtras.h"
22 
23 //===----------------------------------------------------------------------===//
24 // ArithDialect
25 //===----------------------------------------------------------------------===//
26 
27 #include "mlir/Dialect/Arith/IR/ArithOpsDialect.h.inc"
28 
29 //===----------------------------------------------------------------------===//
30 // Arith Dialect Enum Attributes
31 //===----------------------------------------------------------------------===//
32 
33 #include "mlir/Dialect/Arith/IR/ArithOpsEnums.h.inc"
34 #define GET_ATTRDEF_CLASSES
35 #include "mlir/Dialect/Arith/IR/ArithOpsAttributes.h.inc"
36 
37 //===----------------------------------------------------------------------===//
38 // Arith Interfaces
39 //===----------------------------------------------------------------------===//
40 #include "mlir/Dialect/Arith/IR/ArithOpsInterfaces.h.inc"
41 
42 //===----------------------------------------------------------------------===//
43 // Arith Dialect Operations
44 //===----------------------------------------------------------------------===//
45 
46 #define GET_OP_CLASSES
47 #include "mlir/Dialect/Arith/IR/ArithOps.h.inc"
48 
49 namespace mlir {
50 namespace arith {
51 
52 /// Specialization of `arith.constant` op that returns an integer value.
53 class ConstantIntOp : public arith::ConstantOp {
54 public:
55  using arith::ConstantOp::ConstantOp;
56  static ::mlir::TypeID resolveTypeID() { return TypeID::get<ConstantOp>(); }
57 
58  /// Build a constant int op that produces an integer of the specified width.
59  static void build(OpBuilder &builder, OperationState &result, int64_t value,
60  unsigned width);
61 
62  /// Build a constant int op that produces an integer of the specified type,
63  /// which must be an integer type.
64  static void build(OpBuilder &builder, OperationState &result, int64_t value,
65  Type type);
66 
67  inline int64_t value() {
68  return cast<IntegerAttr>(arith::ConstantOp::getValue()).getInt();
69  }
70 
71  static bool classof(Operation *op);
72 };
73 
74 /// Specialization of `arith.constant` op that returns a floating point value.
75 class ConstantFloatOp : public arith::ConstantOp {
76 public:
77  using arith::ConstantOp::ConstantOp;
78  static ::mlir::TypeID resolveTypeID() { return TypeID::get<ConstantOp>(); }
79 
80  /// Build a constant float op that produces a float of the specified type.
81  static void build(OpBuilder &builder, OperationState &result,
82  const APFloat &value, FloatType type);
83 
84  inline APFloat value() {
85  return cast<FloatAttr>(arith::ConstantOp::getValue()).getValue();
86  }
87 
88  static bool classof(Operation *op);
89 };
90 
91 /// Specialization of `arith.constant` op that returns an integer of index type.
92 class ConstantIndexOp : public arith::ConstantOp {
93 public:
94  using arith::ConstantOp::ConstantOp;
95  static ::mlir::TypeID resolveTypeID() { return TypeID::get<ConstantOp>(); }
96  /// Build a constant int op that produces an index.
97  static void build(OpBuilder &builder, OperationState &result, int64_t value);
98 
99  inline int64_t value() {
100  return cast<IntegerAttr>(arith::ConstantOp::getValue()).getInt();
101  }
102 
103  static bool classof(Operation *op);
104 };
105 
106 } // namespace arith
107 } // namespace mlir
108 
109 //===----------------------------------------------------------------------===//
110 // Utility Functions
111 //===----------------------------------------------------------------------===//
112 
113 namespace mlir {
114 namespace arith {
115 
116 /// Compute `lhs` `pred` `rhs`, where `pred` is one of the known integer
117 /// comparison predicates.
118 bool applyCmpPredicate(arith::CmpIPredicate predicate, const APInt &lhs,
119  const APInt &rhs);
120 
121 /// Compute `lhs` `pred` `rhs`, where `pred` is one of the known floating point
122 /// comparison predicates.
123 bool applyCmpPredicate(arith::CmpFPredicate predicate, const APFloat &lhs,
124  const APFloat &rhs);
125 
126 /// Returns the identity value attribute associated with an AtomicRMWKind op.
127 /// `useOnlyFiniteValue` defines whether the identity value should steer away
128 /// from infinity representations or anything that is not a proper finite
129 /// number.
130 /// E.g., The identity value for maxf is in theory `-Inf`, but if we want to
131 /// stay in the finite range, it would be `BiggestRepresentableNegativeFloat`.
132 /// The purpose of this boolean is to offer constants that will play nice
133 /// with fast math related optimizations.
134 TypedAttr getIdentityValueAttr(AtomicRMWKind kind, Type resultType,
135  OpBuilder &builder, Location loc,
136  bool useOnlyFiniteValue = false);
137 
138 /// Return the identity numeric value associated to the give op. Return
139 /// std::nullopt if there is no known neutral element.
140 /// If `op` has `FastMathFlags::ninf`, only finite values will be used
141 /// as neutral element.
142 std::optional<TypedAttr> getNeutralElement(Operation *op);
143 
144 /// Returns the identity value associated with an AtomicRMWKind op.
145 /// \see getIdentityValueAttr for a description of what `useOnlyFiniteValue`
146 /// does.
147 Value getIdentityValue(AtomicRMWKind op, Type resultType, OpBuilder &builder,
148  Location loc, bool useOnlyFiniteValue = false);
149 
150 /// Returns the value obtained by applying the reduction operation kind
151 /// associated with a binary AtomicRMWKind op to `lhs` and `rhs`.
152 Value getReductionOp(AtomicRMWKind op, OpBuilder &builder, Location loc,
153  Value lhs, Value rhs);
154 
155 arith::CmpIPredicate invertPredicate(arith::CmpIPredicate pred);
156 } // namespace arith
157 } // namespace mlir
158 
159 #endif // MLIR_DIALECT_ARITH_IR_ARITH_H_
This class helps build Operations.
Definition: Builders.h:209
Operation is the basic unit of execution within MLIR.
Definition: Operation.h:88
Instances of the Type class are uniqued, have an immutable identifier and an optional mutable compone...
Definition: Types.h:74
Specialization of arith.constant op that returns a floating point value.
Definition: Arith.h:75
static void build(OpBuilder &builder, OperationState &result, const APFloat &value, FloatType type)
Build a constant float op that produces a float of the specified type.
Definition: ArithOps.cpp:271
static bool classof(Operation *op)
Definition: ArithOps.cpp:277
::mlir::TypeID resolveTypeID()
Definition: Arith.h:78
Specialization of arith.constant op that returns an integer of index type.
Definition: Arith.h:92
static void build(OpBuilder &builder, OperationState &result, int64_t value)
Build a constant int op that produces an index.
Definition: ArithOps.cpp:283
::mlir::TypeID resolveTypeID()
Definition: Arith.h:95
static bool classof(Operation *op)
Definition: ArithOps.cpp:289
Specialization of arith.constant op that returns an integer value.
Definition: Arith.h:53
::mlir::TypeID resolveTypeID()
Definition: Arith.h:56
static void build(OpBuilder &builder, OperationState &result, int64_t value, unsigned width)
Build a constant int op that produces an integer of the specified width.
Definition: ArithOps.cpp:250
static bool classof(Operation *op)
Definition: ArithOps.cpp:265
std::optional< TypedAttr > getNeutralElement(Operation *op)
Return the identity numeric value associated to the give op.
Definition: ArithOps.cpp:2489
bool applyCmpPredicate(arith::CmpIPredicate predicate, const APInt &lhs, const APInt &rhs)
Compute lhs pred rhs, where pred is one of the known integer comparison predicates.
Definition: ArithOps.cpp:1718
TypedAttr getIdentityValueAttr(AtomicRMWKind kind, Type resultType, OpBuilder &builder, Location loc, bool useOnlyFiniteValue=false)
Returns the identity value attribute associated with an AtomicRMWKind op.
Definition: ArithOps.cpp:2434
Value getReductionOp(AtomicRMWKind op, OpBuilder &builder, Location loc, Value lhs, Value rhs)
Returns the value obtained by applying the reduction operation kind associated with a binary AtomicRM...
Definition: ArithOps.cpp:2540
Value getIdentityValue(AtomicRMWKind op, Type resultType, OpBuilder &builder, Location loc, bool useOnlyFiniteValue=false)
Returns the identity value associated with an AtomicRMWKind op.
Definition: ArithOps.cpp:2530
arith::CmpIPredicate invertPredicate(arith::CmpIPredicate pred)
Invert an integer comparison predicate.
Definition: ArithOps.cpp:68
Include the generated interface declarations.
This represents an operation in an abstracted form, suitable for use with the builder APIs.