shape_reduction.h#

Shape-inference functions for ONNX operators in the reduction family.

namespace onnx_light

Alias that makes onnx-light headers compatible with code that references ONNX_LIGHT_NAMESPACE (the macro used in the standard onnx package).

Set to ONNX_LIGHT_NAMESPACE so both names resolve to the same namespace.

Symbol-visibility attribute for the public onnx-light C++ API.

Defined as empty because onnx-light does not require explicit __declspec(dllexport) or __attribute__((visibility("default"))) annotations — visibility is controlled at the shared-library level. The macro is provided so that vendored ONNX headers that decorate their declarations with ONNX_API compile without modification.

Namespace alias so that ONNX C++ code (and consumers such as onnxruntime) that refers to the literal onnx namespace — rather than the ONNX_NAMESPACE macro — resolves to the onnx-light namespace. The standard onnx package lives in namespace onnx; onnx-light uses onnx_light (via ONNX_LIGHT_NAMESPACE), so this alias keeps onnx-light a true drop-in. It is only introduced when the onnx-light namespace differs from onnx.

namespace onnx_optim

namespace shapes

namespace reduction#

Functions

void ComputeShapeReduceSum(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#

Computes the output :cpp:class:OptimTensor of a ReduceSum node and stores it in ctx.

ReduceSum reduces the input tensor along a set of axes. The output dtype always matches the input dtype (type constraint T); the output shape is the input shape with the reduced axes either dropped (keepdims=0) or replaced by 1 (keepdims=1, the default).

The way axes are specified depends on the opset:

opset < 13: axes is a repeated INTS attribute. If absent every dimension is reduced.
opset >= 13: axes is an optional second input. When the input is missing or empty the behaviour is controlled by the noop_with_empty_axes attribute (default 0): 0 means reduce all axes, 1 means identity (no axes reduced).

When the axes input is provided this function looks at the :cpp:func:OptimTensor::ValueAsShape annotation of axes to identify the reduced axes. If the annotation is missing the output rank can still be inferred when keepdims=1 (same rank as the input, with each previously-known dimension kept as is and reduced positions left symbolic) or when the number of axes is known via the shape of axes (keepdims=0: output rank = input rank − number of axes, with every dimension marked symbolic).

Parameters:

ctx – In/out context. Must already contain an entry for data and, when provided and non-empty, axes. On return it also contains an entry for node.output(0).
node – The ReduceSum NodeProto whose output should be described. node.op_type() must be "ReduceSum" and node must declare at least one output.
data – Name of the data input value to read from ctx. Must be present in ctx.
axes – Name of the axes input value to read from ctx for opset >= 13, or nullptr / empty string when the axes input is omitted (use the attribute / default “reduce all” behaviour). For opset < 13 the value is ignored and the axes attribute is consulted.

Throws:

std::invalid_argument – if node.op_type() is not "ReduceSum", if node has no output, or if an axis value is out of range.
std::out_of_range – if data (or axes when given) is not present in ctx.

void ComputeShapeReduceMax(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceMax node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeReduceMin(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceMin node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeReduceL1(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceL1 node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeReduceL2(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceL2 node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeReduceSumSquare(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceSumSquare node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeReduceLogSum(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceLogSum node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeReduceLogSumExp(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceLogSumExp node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeReduceProd(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceProd node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeReduceMean(ShapesContext &ctx, const NodeProto &node, const char *data, const char *axes)#: Computes the output :cpp:class:OptimTensor of a ReduceMean node. Shape/attribute semantics are the same as :cpp:func:ComputeShapeReduceSum and the output dtype matches the input dtype.

void ComputeShapeArgReduce(ShapesContext &ctx, const NodeProto &node, const char *data)#

Computes the output :cpp:class:OptimTensor of an ArgMax or ArgMin node and stores it in ctx.

ArgMax/ArgMin reduce the input tensor along a single axis (attribute, default 0; accepts negative values from opset 11). The output dtype is always tensor(int64) (independent of the input dtype); the output shape is the input shape with the reduced axis either dropped (keepdims=0) or replaced by 1 (keepdims=1, the default).

The select_last_index attribute introduced in opset 12 does not affect the output shape and is therefore ignored here.

Parameters:

ctx – In/out context. Must already contain an entry for data. On return it also contains an entry for node.output(0).
node – The ArgMax or ArgMin NodeProto whose output should be described. node.op_type() must be either "ArgMax" or "ArgMin" and node must declare at least one output.
data – Name of the data input value to read from ctx. Must be present in ctx.

Throws:

std::invalid_argument – if node.op_type() is not "ArgMax" or "ArgMin", if node has no output, if the input has rank 0, or if the axis attribute is out of range.
std::out_of_range – if data is not present in ctx.