Source code for experimental_experiment.xoptim.patterns

import inspect
from typing import List, Optional
from onnx import NodeProto
from ..patterns_api import PatternOptimization, MatchResult

# onnx patterns
from .onnx_any import IdentityPattern, SameChildrenPattern
from .onnx_cast import (
    CastPattern,
    CastCastBinaryPattern,
    CastOpCastPattern,
    ComputationCastOpCastPattern,
)
from .onnx_clip import ClipClipPattern
from .onnx_conv import ConvBiasNullPattern
from .onnx_dropout import DropoutPattern
from .onnx_equal import UnsqueezeEqualPattern
from .onnx_expand import ExpandPattern, ExpandBroadcastPattern, ExpandSwapPattern
from .onnx_functions import GeluPattern, LeakyReluPattern, SoftmaxCrossEntropyLossCastPattern
from .onnx_layer_normalization import (
    BatchNormalizationPattern,
    BatchNormalizationTrainingPattern,
    CastLayerNormalizationCastPattern,
    LayerNormalizationPattern,
    LayerNormalizationScalePattern,
)
from .onnx_mul import (
    MulMulMulScalarPattern,
    SwitchOrderBinaryPattern,
)
from .onnx_matmul import (
    GemmTransposePattern,
    MatMulAddPattern,
    MatMulReshape2Of3Pattern,
    MulMulMatMulPattern,
    ReshapeMatMulReshapePattern,
    TransposeMatMulPattern,
    TransposeReshapeMatMulPattern,
)
from .onnx_reduce import ReduceSumNormalizePattern
from .onnx_reshape import (
    ReshapePattern,
    ReduceReshapePattern,
    Reshape2Of3Pattern,
    ReshapeReshapeBinaryPattern,
    ReshapeReshapePattern,
)
from .onnx_rotary import RotaryConcatPartPattern
from .onnx_sequence import SequenceConstructAtPattern
from .onnx_slice import SliceSlicePattern
from .onnx_split import SlicesSplitPattern
from .onnx_sub import Sub1MulPattern
from .onnx_transpose import TransposeTransposePattern, TransposeReshapeTransposePattern
from .onnx_unsqueeze import UnsqueezeUnsqueezePattern




class AlmostDoNothingPattern(PatternOptimization):
    """
    Checks that a Expand is really needed.
    """

    n_count = 0



    def match(
        self,
        g: "GraphBuilderPatternOptimization",  # noqa: F821
        node: NodeProto,
        matched: List[MatchResult],
    ) -> Optional[MatchResult]:
        if self.n_count >= 0:
            return self.none()
        if node.op_type != "Pow" or node.domain != "":
            return self.none()
        if node.name is not None and "AlmostDoNothing" in node.name:
            return self.none(node, inspect.currentframe().f_lineno)
        self.n_count += 1
        return MatchResult(self, [node], self.apply, insert_at=node)




    def apply(
        self,
        g: "GraphBuilder",  # noqa: F821
        node: NodeProto,
    ) -> List[NodeProto]:
        return [
            g.make_node(
                node.op_type,
                node.input,
                node.output,
                name=f"AlmostDoNothing--{node.name}",
            )
        ]






def get_default_patterns(verbose: int = 0) -> List[PatternOptimization]:
    """
    Returns a default list of optimization patterns.
    It is equal to the following list.

    .. runpython::
        :showcode:

        import pprint
        from experimental_experiment.xoptim.patterns import get_default_patterns
        pprint.pprint(get_default_patterns())
    """
    return [
        # AlmostDoNothingPattern(verbose=verbose),
        BatchNormalizationPattern(verbose=verbose),
        BatchNormalizationTrainingPattern(verbose=verbose),
        CastLayerNormalizationCastPattern(verbose=verbose),
        CastPattern(verbose=verbose),
        CastCastBinaryPattern(verbose=verbose),
        CastOpCastPattern(verbose=verbose),
        ClipClipPattern(verbose=verbose),
        ComputationCastOpCastPattern(verbose=verbose),
        ConvBiasNullPattern(verbose=verbose),
        DropoutPattern(verbose=verbose),
        ExpandPattern(verbose=verbose),
        ExpandBroadcastPattern(verbose=verbose),
        ExpandSwapPattern(verbose=verbose),
        GeluPattern(verbose=verbose),
        IdentityPattern(verbose=verbose),
        LayerNormalizationPattern(verbose=verbose),
        LayerNormalizationScalePattern(verbose=verbose),
        LeakyReluPattern(verbose=verbose),
        MulMulMulScalarPattern(verbose=verbose),
        ReduceReshapePattern(verbose=verbose),
        ReduceSumNormalizePattern(verbose=verbose),
        ReshapePattern(verbose=verbose),
        ReshapeMatMulReshapePattern(verbose=verbose),
        Reshape2Of3Pattern(verbose=verbose),
        ReshapeReshapeBinaryPattern(verbose=verbose),
        MatMulAddPattern(verbose=verbose),
        GemmTransposePattern(verbose=verbose),
        MatMulReshape2Of3Pattern(verbose=verbose),
        MulMulMatMulPattern(verbose=verbose),
        ReshapeReshapePattern(verbose=verbose),
        RotaryConcatPartPattern(verbose=verbose),
        SameChildrenPattern(verbose=verbose),
        SequenceConstructAtPattern(verbose=verbose),
        SliceSlicePattern(verbose=verbose),
        SlicesSplitPattern(verbose=verbose),
        SoftmaxCrossEntropyLossCastPattern(verbose=verbose),
        Sub1MulPattern(verbose=verbose),
        SwitchOrderBinaryPattern(verbose=verbose),
        TransposeMatMulPattern(verbose=verbose),
        TransposeReshapeMatMulPattern(verbose=verbose),
        TransposeReshapeTransposePattern(verbose=verbose),
        TransposeTransposePattern(verbose=verbose),
        UnsqueezeEqualPattern(verbose=verbose),
        UnsqueezeUnsqueezePattern(verbose=verbose),
    ]