yobx.xoptim.patterns.onnx_transpose — yet-another-onnx-builder 0.1.0 documentation

class yobx.xoptim.patterns.onnx_transpose.SwapUnsqueezeTransposePattern(verbose: int = 0, priority: int = 0)[source]#

Swaps Unsqueeze and Transpose.

Model with nodes to be fused:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_X(["X FLOAT(a, b, c)"])
    I_axes(["axes INT64(2)"])

    Constant_0[["Constant() -#gt; axes"]]
    Unsqueeze_1[["Unsqueeze(., .)"]]
    Transpose_2[["Transpose(., perm=[0, 2, 1, 4, 3])"]]

    I_X -->|"FLOAT(a, b, c)"| Unsqueeze_1
    Constant_0 -->|"INT64(2)"| Unsqueeze_1
    Unsqueeze_1 -->|"FLOAT(a, 1, 1, b, c)"| Transpose_2

    O_Y(["Y FLOAT(e, f, g, h, i)"])
    Transpose_2 --> O_Y

    class I_X,I_axes,O_Y ioNode
    class Constant_0 constNode
    class Unsqueeze_1,Transpose_2 opNode

Outcome of the fusion:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_X(["X FLOAT(a, b, c)"])
    I_axes(["axes INT64(2)"])

    Transpose_0[["Transpose(., perm=[0, 2, 1])"]]
    Unsqueeze_1[["Unsqueeze(., .)"]]

    I_X -->|"FLOAT(a, b, c)"| Transpose_0
    Transpose_0 -->|"FLOAT(a, c, b)"| Unsqueeze_1
    I_axes -->|"INT64(2)"| Unsqueeze_1

    O_Y(["Y FLOAT(e, f, g, h, i)"])
    Unsqueeze_1 --> O_Y

    class I_X,I_axes,O_Y ioNode
    class Transpose_0,Unsqueeze_1 opNode

apply(g: GraphBuilder, unsqueeze_node: NodeProto, transpose_node: NodeProto) → List[NodeProto][source]#

The method does the rewriting. It assumes it can happen. It takes a list of nodes impacted by the rewriting assumes no other pattern optimizer will be modify them. It receives the list of nodes returned by method apply. Since it is a list of argument, method match can include None values. The method returns the new nodes. The optimizer considers that any node given to this function is removed from the graph, and any node returned by it are added. If a received node must be kept, it must be added to the list of returned node.

Parameters:: nodes – nodes returned by method match, there are then removed
Returns:: nodes to add to graph.

match(g: GraphBuilderPatternOptimization, node: NodeProto, matched: List[MatchResult]) → MatchResult | None[source]#

Determines nodes around node which can be rewritten.

Parameters:

g – is a GraphBuilderPatternOptimization, it holds all the existing nodes, is able to return any information about type, shape, the node before, the node after another one.
node – the matching must determine if some nodes around this one are part of set of nodes this pattern optimizer can rewrite. From there, the function explores wherever it needs, checking any condition it needs.
matched – usually unused, it returns of nodes already matching a pattern

The method must not modify the graph. The method returns None if no match is found or an instance of class MatchResult. It must contain:

a list of nodes involved in the rewriting. It does not mean all of them will be removed but all of them are needed to do the rewriting and must not be impacted by other pattern optimizer.
A function doing the rewriting (usually method apply of the pattern class).
An existing node where the rewritten nodes can be inserted. Knowing it makes it faster to rewriter. If not specified, the optimizer will automatically determine the position of the new nodes.

class yobx.xoptim.patterns.onnx_transpose.TransposeEqualReshapePattern(verbose: int = 0, priority: int = 1, min_opset: int = 1)[source]#

Replaces a Transpose by a Reshape when switched dimensions are all equal to 1 but one.

Model with nodes to be fused:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_X(["X FLOAT(3, 2, 1, 5)"])

    Transpose_0[["Transpose(., perm=[0, 2, 1, 3])"]]

    I_X -->|"FLOAT(3, 2, 1, 5)"| Transpose_0

    O_Y(["Y FLOAT(a, b, c, d)"])
    Transpose_0 --> O_Y

    class I_X,O_Y ioNode
    class Transpose_0 opNode

Outcome of the fusion:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_X(["X FLOAT(3, 2, 1, 5)"])

    Reshape_0[["Reshape(., [0, 1, -1, 0])"]]

    I_X -->|"FLOAT(3, 2, 1, 5)"| Reshape_0

    O_Y(["Y FLOAT(a, b, c, d)"])
    Reshape_0 --> O_Y

    class I_X,O_Y ioNode
    class Reshape_0 opNode

apply(g: GraphBuilder, transpose_node: NodeProto) → List[NodeProto][source]#

The method does the rewriting. It assumes it can happen. It takes a list of nodes impacted by the rewriting assumes no other pattern optimizer will be modify them. It receives the list of nodes returned by method apply. Since it is a list of argument, method match can include None values. The method returns the new nodes. The optimizer considers that any node given to this function is removed from the graph, and any node returned by it are added. If a received node must be kept, it must be added to the list of returned node.

Parameters:: nodes – nodes returned by method match, there are then removed
Returns:: nodes to add to graph.

match(g: GraphBuilderPatternOptimization, node: NodeProto, matched: List[MatchResult]) → MatchResult | None[source]#

Determines nodes around node which can be rewritten.

Parameters:

g – is a GraphBuilderPatternOptimization, it holds all the existing nodes, is able to return any information about type, shape, the node before, the node after another one.
node – the matching must determine if some nodes around this one are part of set of nodes this pattern optimizer can rewrite. From there, the function explores wherever it needs, checking any condition it needs.
matched – usually unused, it returns of nodes already matching a pattern

The method must not modify the graph. The method returns None if no match is found or an instance of class MatchResult. It must contain:

a list of nodes involved in the rewriting. It does not mean all of them will be removed but all of them are needed to do the rewriting and must not be impacted by other pattern optimizer.
A function doing the rewriting (usually method apply of the pattern class).
An existing node where the rewritten nodes can be inserted. Knowing it makes it faster to rewriter. If not specified, the optimizer will automatically determine the position of the new nodes.

class yobx.xoptim.patterns.onnx_transpose.TransposeGatherPattern(verbose: int = 0, priority: int = 0)[source]#

Removes one unnecessary transpose followed by Gather with only one index.

Model with nodes to be fused:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_X(["X FLOAT(a, b, 16, 80)"])
    I_ind(["ind INT64()"])

    Constant_0[["Constant() -#gt; ind"]]
    Transpose_1[["Transpose(., perm=[1, 0, 2, 3])"]]
    Gather_2[["Gather(., ., axis=0)"]]

    I_X -->|"FLOAT(a, b, 16, 80)"| Transpose_1
    Transpose_1 -->|"FLOAT(b, a, 16, 80)"| Gather_2
    Constant_0 -->|"INT64()"| Gather_2

    O_Y(["Y FLOAT(a, 16, 80)"])
    Gather_2 --> O_Y

    class I_X,I_ind,O_Y ioNode
    class Constant_0 constNode
    class Transpose_1,Gather_2 opNode

Outcome of the fusion:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_X(["X FLOAT(a, b, 16, 80)"])
    I_ind(["ind INT64()"])

    Gather_0[["Gather(., ., axis=1)"]]

    I_X -->|"FLOAT(a, b, 16, 80)"| Gather_0
    I_ind -->|"INT64()"| Gather_0

    O_Y(["Y FLOAT(a, 16, 80)"])
    Gather_0 --> O_Y

    class I_X,I_ind,O_Y ioNode
    class Gather_0 opNode

apply(g: GraphBuilder, transpose_node: NodeProto, gather_node: NodeProto) → List[NodeProto][source]#

The method does the rewriting. It assumes it can happen. It takes a list of nodes impacted by the rewriting assumes no other pattern optimizer will be modify them. It receives the list of nodes returned by method apply. Since it is a list of argument, method match can include None values. The method returns the new nodes. The optimizer considers that any node given to this function is removed from the graph, and any node returned by it are added. If a received node must be kept, it must be added to the list of returned node.

Parameters:: nodes – nodes returned by method match, there are then removed
Returns:: nodes to add to graph.

match(g: GraphBuilderPatternOptimization, node: NodeProto, matched: List[MatchResult]) → MatchResult | None[source]#

Determines nodes around node which can be rewritten.

Parameters:

g – is a GraphBuilderPatternOptimization, it holds all the existing nodes, is able to return any information about type, shape, the node before, the node after another one.
node – the matching must determine if some nodes around this one are part of set of nodes this pattern optimizer can rewrite. From there, the function explores wherever it needs, checking any condition it needs.
matched – usually unused, it returns of nodes already matching a pattern

The method must not modify the graph. The method returns None if no match is found or an instance of class MatchResult. It must contain:

a list of nodes involved in the rewriting. It does not mean all of them will be removed but all of them are needed to do the rewriting and must not be impacted by other pattern optimizer.
A function doing the rewriting (usually method apply of the pattern class).
An existing node where the rewritten nodes can be inserted. Knowing it makes it faster to rewriter. If not specified, the optimizer will automatically determine the position of the new nodes.

class yobx.xoptim.patterns.onnx_transpose.TransposeReshapeTransposePattern(verbose: int = 0, priority: int = 0)[source]#

Swaps Reshape and Transpose in a sequence such as this one:

input is 32x4x14x4x14x128

Transpose(., perm=[0, 1, 3, 2, 4, 5])
Reshape(., 32x56x56x128)
Transpose(., perm=[0, 3, 1, 2])

By:

Transpose(., perm=[0, 1, 3, 2, 4, 5])
Transpose(., perm=[0, 5, 1, 2, 3, 4])
Reshape(., 32x128x56x56)

Model with nodes to be fused:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_xts(["xts FLOAT(32, 2, 14, 2, 13, 256)"])
    I_X(["X FLOAT(32, 256, 28, 26)"])

    Transpose_0[["Transpose(., perm=[0, 2, 3, 1])"]]
    Reshape_1[["Reshape(., [32, 2, 14, 2, 13, 256])"]]
    Transpose_2[["Transpose(., perm=[0, 1, 3, 2, 4, 5])"]]

    I_X -->|"FLOAT(32, 256, 28, 26)"| Transpose_0
    Transpose_0 -->|"FLOAT(32, 28, 26, 256)"| Reshape_1
    Reshape_1 -->|"FLOAT(32, 2, 14, 2, 13, 256)"| Transpose_2

    O_xts(["xts FLOAT(32, 2, 14, 2, 13, 256)"])
    Reshape_1 --> O_xts
    O_Y(["Y FLOAT(32, 2, 2, 14, 13, 256)"])
    Transpose_2 --> O_Y

    class I_xts,I_X,O_xts,O_Y ioNode
    class Transpose_0,Reshape_1,Transpose_2 opNode

Outcome of the fusion:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_xts(["xts FLOAT(32, 2, 14, 2, 13, 256)"])
    I_X(["X FLOAT(32, 256, 28, 26)"])

    Reshape_0[["Reshape(., [32, 256, 2, 14, 2, 13])"]]
    Transpose_1[["Transpose(., perm=[0, 2, 3, 4, 5, 1])"]]
    Transpose_2[["Transpose(., perm=[0, 1, 3, 2, 4, 5])"]]

    I_X -->|"FLOAT(32, 256, 28, 26)"| Reshape_0
    Reshape_0 -->|"FLOAT(32, 256, 2, 14, 2, 13)"| Transpose_1
    Transpose_1 -->|"FLOAT(32, 2, 14, 2, 13, 256)"| Transpose_2

    O_xts(["xts FLOAT(32, 2, 14, 2, 13, 256)"])
    Transpose_1 --> O_xts
    O_Y(["Y FLOAT(32, 2, 2, 14, 13, 256)"])
    Transpose_2 --> O_Y

    class I_xts,I_X,O_xts,O_Y ioNode
    class Reshape_0,Transpose_1,Transpose_2 opNode

apply(g: GraphBuilder, t1_node: NodeProto, reshape_node: NodeProto, t2_node: NodeProto) → List[NodeProto][source]#

The method does the rewriting. It assumes it can happen. It takes a list of nodes impacted by the rewriting assumes no other pattern optimizer will be modify them. It receives the list of nodes returned by method apply. Since it is a list of argument, method match can include None values. The method returns the new nodes. The optimizer considers that any node given to this function is removed from the graph, and any node returned by it are added. If a received node must be kept, it must be added to the list of returned node.

Parameters:: nodes – nodes returned by method match, there are then removed
Returns:: nodes to add to graph.

match(g: GraphBuilderPatternOptimization, node: NodeProto, matched: List[MatchResult]) → MatchResult | None[source]#

Determines nodes around node which can be rewritten.

Parameters:

g – is a GraphBuilderPatternOptimization, it holds all the existing nodes, is able to return any information about type, shape, the node before, the node after another one.
node – the matching must determine if some nodes around this one are part of set of nodes this pattern optimizer can rewrite. From there, the function explores wherever it needs, checking any condition it needs.
matched – usually unused, it returns of nodes already matching a pattern

The method must not modify the graph. The method returns None if no match is found or an instance of class MatchResult. It must contain:

a list of nodes involved in the rewriting. It does not mean all of them will be removed but all of them are needed to do the rewriting and must not be impacted by other pattern optimizer.
A function doing the rewriting (usually method apply of the pattern class).
An existing node where the rewritten nodes can be inserted. Knowing it makes it faster to rewriter. If not specified, the optimizer will automatically determine the position of the new nodes.

class yobx.xoptim.patterns.onnx_transpose.TransposeTransposePattern(verbose: int = 0, priority: int = 0)[source]#

Removes two consecutive transpose if the second one put the tensor in origin shape.

Model with nodes to be fused:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_xs(["xs FLOAT(1, 1, 32, 128)"])

    Transpose_0[["Transpose(., perm=[1, 0, 3, 2])"]]
    Transpose_1[["Transpose(., perm=[0, 1, 3, 2])"]]

    I_xs -->|"FLOAT(1, 1, 32, 128)"| Transpose_0
    Transpose_0 -->|"FLOAT(1, 1, 128, 32)"| Transpose_1

    O_xm1(["xm1 FLOAT(1, 1, 32, 128)"])
    Transpose_1 --> O_xm1

    class I_xs,O_xm1 ioNode
    class Transpose_0,Transpose_1 opNode

Outcome of the fusion:

        graph TD

    classDef ioNode fill:#dfd,stroke:#333,color:#333
    classDef initNode fill:#cccc00,stroke:#333,color:#333
    classDef constNode fill:#f9f,stroke:#333,stroke-width:2px,color:#333
    classDef opNode fill:#bbf,stroke:#333,stroke-width:2px,color:#333

    I_xs(["xs FLOAT(1, 1, 32, 128)"])

    Transpose_0[["Transpose(., perm=[1, 0, 2, 3])"]]

    I_xs -->|"FLOAT(1, 1, 32, 128)"| Transpose_0

    O_xm1(["xm1 FLOAT(1, 1, 32, 128)"])
    Transpose_0 --> O_xm1

    class I_xs,O_xm1 ioNode
    class Transpose_0 opNode

apply(g: GraphBuilder, node: NodeProto, next_node: NodeProto) → List[NodeProto][source]#

The method does the rewriting. It assumes it can happen. It takes a list of nodes impacted by the rewriting assumes no other pattern optimizer will be modify them. It receives the list of nodes returned by method apply. Since it is a list of argument, method match can include None values. The method returns the new nodes. The optimizer considers that any node given to this function is removed from the graph, and any node returned by it are added. If a received node must be kept, it must be added to the list of returned node.

Parameters:: nodes – nodes returned by method match, there are then removed
Returns:: nodes to add to graph.

match(g: GraphBuilderPatternOptimization, node: NodeProto, matched: List[MatchResult]) → MatchResult | None[source]#

Determines nodes around node which can be rewritten.

Parameters:

g – is a GraphBuilderPatternOptimization, it holds all the existing nodes, is able to return any information about type, shape, the node before, the node after another one.
node – the matching must determine if some nodes around this one are part of set of nodes this pattern optimizer can rewrite. From there, the function explores wherever it needs, checking any condition it needs.
matched – usually unused, it returns of nodes already matching a pattern

The method must not modify the graph. The method returns None if no match is found or an instance of class MatchResult. It must contain:

a list of nodes involved in the rewriting. It does not mean all of them will be removed but all of them are needed to do the rewriting and must not be impacted by other pattern optimizer.
A function doing the rewriting (usually method apply of the pattern class).
An existing node where the rewritten nodes can be inserted. Knowing it makes it faster to rewriter. If not specified, the optimizer will automatically determine the position of the new nodes.

yobx.xoptim.patterns.onnx_transpose#

This Page