mlir._mlir_libs._mlir.rewrite¶

MLIR Rewrite Bindings

Classes¶

`GreedyRewriteStrictness`	Generic enumeration.
`GreedySimplifyRegionLevel`	Generic enumeration.
`DialectConversionFoldingMode`	Generic enumeration.
`PatternRewriter`
`RewritePatternSet`
`ConversionPatternRewriter`
`ConversionTarget`
`TypeConverter`
`PDLResultList`
`PDLModule`
`GreedyRewriteConfig`
`ConversionConfig`
`FrozenRewritePatternSet`

Functions¶

`apply_patterns_and_fold_greedily`(…)	Applys the given patterns to the given op greedily while folding results.
`walk_and_apply_patterns`(→ None)	Applies the given patterns to the given op by a fast walk-based driver.
`apply_partial_conversion`(→ None)	Applies a partial conversion on the given operation.
`apply_full_conversion`(→ None)	Applies a full conversion on the given operation.

Module Contents¶

class mlir._mlir_libs._mlir.rewrite.GreedyRewriteStrictness¶

Bases: enum.Enum

Generic enumeration.

Derive from this class to define new enumerations.

ANY_OP = 0¶

EXISTING_AND_NEW_OPS = 1¶

EXISTING_OPS = 2¶

class mlir._mlir_libs._mlir.rewrite.GreedySimplifyRegionLevel¶

Bases: enum.Enum

Generic enumeration.

Derive from this class to define new enumerations.

DISABLED = 0¶

NORMAL = 1¶

AGGRESSIVE = 2¶

class mlir._mlir_libs._mlir.rewrite.DialectConversionFoldingMode¶

Bases: enum.Enum

Generic enumeration.

Derive from this class to define new enumerations.

NEVER = 0¶

BEFORE_PATTERNS = 1¶

AFTER_PATTERNS = 2¶

class mlir._mlir_libs._mlir.rewrite.PatternRewriter¶

property ip: _mlir.ir.InsertionPoint¶: The current insertion point of the PatternRewriter.

replace_op(op: _mlir.ir._OperationBase, new_op: _mlir.ir._OperationBase) → None¶
replace_op(op: _mlir.ir._OperationBase, values: collections.abc.Sequence[_mlir.ir.Value]) → None: Replace an operation with a list of values.

erase_op(op: _mlir.ir.Operation) → None¶: Erase an operation.

class mlir._mlir_libs._mlir.rewrite.RewritePatternSet(context: Context | None = None)¶

add(root: type | str, fn: collections.abc.Callable[[mlir.ir.Operation, PatternRewriter], Any], benefit: int = 1) → None¶

Add a new rewrite pattern on the specified root operation, using the provided callable for matching and rewriting, and assign it the given benefit.

Parameters:

root – The root operation to which this pattern applies. This may be either an OpView subclass (e.g., arith.AddIOp) or an operation name string (e.g., "arith.addi").
fn – The callable to use for matching and rewriting, which takes an operation and a pattern rewriter as arguments. The match is considered successful iff the callable returns a value where bool(value) is False (e.g. None). If possible, the operation is cast to its corresponding OpView subclass before being passed to the callable.
benefit – The benefit of the pattern, defaulting to 1.

add_conversion(root: object, fn: collections.abc.Callable, type_converter: TypeConverter, benefit: int = 1) → None¶

Add a new conversion pattern on the specified root operation, using the provided callable for matching and rewriting, and assign it the given benefit.

Parameters:

root – The root operation to which this pattern applies. This may be either an OpView subclass (e.g., arith.AddIOp) or an operation name string (e.g., "arith.addi").
fn – The callable to use for matching and rewriting, which takes an operation, its adaptor, the type converter and a pattern rewriter as arguments. The match is considered successful iff the callable returns a value where bool(value) is False (e.g. None). If possible, the operation is cast to its corresponding OpView subclass before being passed to the callable.
type_converter – The type converter to convert types in the IR.
benefit – The benefit of the pattern, defaulting to 1.

freeze() → FrozenRewritePatternSet¶: Freeze the pattern set into a frozen one.

class mlir._mlir_libs._mlir.rewrite.ConversionPatternRewriter¶: Bases: PatternRewriter

class mlir._mlir_libs._mlir.rewrite.ConversionTarget(context: Context | None = None)¶

add_legal_op(*ops) → None¶: Mark the given operations as legal.

add_illegal_op(*ops) → None¶: Mark the given operations as illegal.

add_legal_dialect(*dialects) → None¶: Mark the given dialects as legal.

add_illegal_dialect(*dialects) → None¶: Mark the given dialect as illegal.

class mlir._mlir_libs._mlir.rewrite.TypeConverter¶

add_conversion(convert: collections.abc.Callable) → None¶: Register a type conversion function.

class mlir._mlir_libs._mlir.rewrite.PDLResultList¶

append(arg: _mlir.ir.Value, /) → None¶
append(arg: _mlir.ir.Operation, /) → None
append(arg: _mlir.ir.Type, /) → None
append(arg: _mlir.ir.Attribute, /) → None

class mlir._mlir_libs._mlir.rewrite.PDLModule(module: _mlir.ir.Module)¶

class mlir._mlir_libs._mlir.rewrite.PDLModule(module: _mlir.ir.Module)

freeze() → FrozenRewritePatternSet¶

register_rewrite_function(arg0: str, arg1: collections.abc.Callable, /) → None¶

register_constraint_function(arg0: str, arg1: collections.abc.Callable, /) → None¶

class mlir._mlir_libs._mlir.rewrite.GreedyRewriteConfig¶

property max_iterations: int¶: Maximum number of iterations

property max_num_rewrites: int¶: Maximum number of rewrites per iteration

property use_top_down_traversal: bool¶: Whether to use top-down traversal

property enable_folding: bool¶: Enable or disable folding

property strictness: GreedyRewriteStrictness¶: Rewrite strictness level

property region_simplification_level: GreedySimplifyRegionLevel¶: Region simplification level

property enable_constant_cse: bool¶: Enable or disable constant CSE

class mlir._mlir_libs._mlir.rewrite.ConversionConfig¶

property folding_mode: DialectConversionFoldingMode¶: folding behavior during dialect conversion

property build_materializations: bool¶: Whether the dialect conversion attempts to build source/target materializations

class mlir._mlir_libs._mlir.rewrite.FrozenRewritePatternSet¶

property _CAPIPtr: object¶

_CAPICreate() → object¶

mlir._mlir_libs._mlir.rewrite.apply_patterns_and_fold_greedily(module: _mlir.ir.Module, set: FrozenRewritePatternSet, config: GreedyRewriteConfig | None = None) → None¶
mlir._mlir_libs._mlir.rewrite.apply_patterns_and_fold_greedily(op: _mlir.ir._OperationBase, set: FrozenRewritePatternSet, config: GreedyRewriteConfig | None = None) → None: Applys the given patterns to the given op greedily while folding results.

mlir._mlir_libs._mlir.rewrite.walk_and_apply_patterns(op: _mlir.ir._OperationBase, set: FrozenRewritePatternSet) → None¶: Applies the given patterns to the given op by a fast walk-based driver.

mlir._mlir_libs._mlir.rewrite.apply_partial_conversion(op: _mlir.ir._OperationBase, target: ConversionTarget, set: FrozenRewritePatternSet, config: ConversionConfig | None = None) → None¶: Applies a partial conversion on the given operation.

mlir._mlir_libs._mlir.rewrite.apply_full_conversion(op: _mlir.ir._OperationBase, target: ConversionTarget, set: FrozenRewritePatternSet, config: ConversionConfig | None = None) → None¶: Applies a full conversion on the given operation.