shape_quantization.h#

Shape-inference functions for ONNX operators in the quantization 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 quantization#

Functions

void ComputeShapeQuantizeLinear(ShapesContext &ctx, const NodeProto &node, const char *x, const char *y_zero_point)#

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

QuantizeLinear produces an output y that always has the same shape as the input x. The output element type is resolved as follows:

  • when y_zero_point is supplied (third input, non-empty name) the output dtype is the dtype of y_zero_point;

  • otherwise, when the output_dtype integer attribute is set (opset 23+), it is interpreted as a TensorProto::DataType and mapped to the matching :cpp:enum:TensorType;

  • otherwise the output dtype defaults to uint8.

The axis, saturate, block_size, and precision attributes do not affect the output shape and are therefore not inspected by this function.

Parameters:

  • ctx – In/out context. Must already contain entries for x, y_scale, and y_zero_point when the latter is provided; on return it also contains an entry for node.output(0).

  • node – The QuantizeLinear NodeProto whose output should be described. node.op_type() must be "QuantizeLinear" and node must declare at least one output.

  • x – Name of the input value to read from ctx. Must be present in ctx.

  • y_zero_point – Name of the y_zero_point input value, or nullptr / empty string when the input is omitted. When non-empty it must be present in ctx.

Throws:

  • std::invalid_argument – if node.op_type() is not "QuantizeLinear", if node has no output, or if output_dtype is set to a value that does not map to a supported :cpp:enum:TensorType.

  • std::out_of_range – if x (or y_zero_point when non-empty) is not present in ctx.

void ComputeShapeDequantizeLinear(ShapesContext &ctx, const NodeProto &node, const char *x, const char *x_scale)#

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

DequantizeLinear produces an output y that always has the same shape as the input x. The output element type is resolved as follows:

  • when the output_dtype integer attribute is set (opset 23+) it is interpreted as a TensorProto::DataType and mapped to the matching :cpp:enum:TensorType;

  • otherwise, the output dtype is the dtype of x_scale.

The axis and block_size attributes do not affect the output shape and are therefore not inspected by this function.

Parameters:

  • ctx – In/out context. Must already contain entries for x and x_scale; on return it also contains an entry for node.output(0).

  • node – The DequantizeLinear NodeProto whose output should be described. node.op_type() must be "DequantizeLinear" and node must declare at least one output.

  • x – Name of the input value to read from ctx. Must be present in ctx.

  • x_scale – Name of the x_scale input value, used to derive the default output element type. Must be present in ctx.

Throws:

  • std::invalid_argument – if node.op_type() is not "DequantizeLinear", if node has no output, or if output_dtype is set to a value that does not map to a supported :cpp:enum:TensorType.

  • std::out_of_range – if x or x_scale is not present in ctx.

void ComputeShapeDynamicQuantizeLinear(ShapesContext &ctx, const NodeProto &node, const char *x)#

Computes the output :cpp:class:OptimTensor entries of a DynamicQuantizeLinear node and stores them in ctx.

DynamicQuantizeLinear produces three outputs (since opset 11 in the ai.onnx domain):

  • y — same shape as the input x, with dtype uint8;

  • y_scale — scalar (rank 0) float;

  • y_zero_point — scalar (rank 0) uint8.

The operator takes no attributes that affect output shapes.

Parameters:

  • ctx – In/out context. Must already contain an entry for x; on return it also contains entries for the (up to three) non-empty outputs of node.

  • node – The DynamicQuantizeLinear NodeProto whose outputs should be described. node.op_type() must be "DynamicQuantizeLinear" and node must declare at least one output.

  • x – Name of the input value to read from ctx. Must be present in ctx.

Throws:

  • std::invalid_argument – if node.op_type() is not "DynamicQuantizeLinear" or if node has no output.

  • std::out_of_range – if x is not present in ctx.

void ComputeShapeQLinearConv(ShapesContext &ctx, const NodeProto &node, const char *x, const char *w, const char *y_zero_point)#

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

The output shape rule matches :cpp:func:ComputeShapeConv applied to the quantized inputs x (input 0) and w (input 3). The output dtype is the dtype of y_zero_point (input 7).

Parameters:

  • ctx – In/out context. Must already contain entries for x, w, and y_zero_point; on return it also contains an entry for node.output(0).

  • node – The QLinearConv NodeProto whose output should be described. node.op_type() must be "QLinearConv" and node must declare at least one output.

  • x – Name of the input data value (rank >= 3) in ctx.

  • w – Name of the weight value (rank >= 3) in ctx.

  • y_zero_point – Name of the y_zero_point input value, used to derive the output element type. Must be present in ctx.

Throws:

  • std::invalid_argument – if node.op_type() is not "QLinearConv" or if node has no output.

  • std::out_of_range – if x, w, or y_zero_point is not present in ctx.

void ComputeShapeQLinearMatMul(ShapesContext &ctx, const NodeProto &node, const char *a, const char *b, const char *y_zero_point)#

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

The output shape rule matches :cpp:func:ComputeShapeMatMul applied to the quantized inputs a (input 0) and b (input 3). The output dtype is the dtype of y_zero_point (input 7).

Parameters:

  • ctx – In/out context. Must already contain entries for a, b, and y_zero_point; on return it also contains an entry for node.output(0).

  • node – The QLinearMatMul NodeProto whose output should be described. node.op_type() must be "QLinearMatMul" and node must declare at least one output.

  • a – Name of the input a value (rank >= 1) in ctx.

  • b – Name of the input b value (rank >= 1) in ctx.

  • y_zero_point – Name of the y_zero_point input value, used to derive the output element type. Must be present in ctx.

Throws:

  • std::invalid_argument – if node.op_type() is not "QLinearMatMul" or if node has no output.

  • std::out_of_range – if a, b, or y_zero_point is not present in ctx.