reduce_node.h File Reference

Tensor reduction operation nodes. More...

#include <optional>
#include "sdfg/data_flow/library_nodes/math/tensor/tensor_node.h"
#include "sdfg/codegen/dispatchers/block_dispatcher.h"
#include "sdfg/serializer/json_serializer.h"

Go to the source code of this file.

Classes
class	sdfg::math::tensor::ReduceNode
	Base class for tensor reduction operations. More...
class	sdfg::math::tensor::ReduceNodeSerializer< T >

Namespaces
namespace	sdfg
namespace	sdfg::math
namespace	sdfg::math::tensor

Detailed Description

Tensor reduction operation nodes.

This file defines the base class for tensor reduction operations. Reductions are operations that aggregate values along one or more tensor dimensions.

Tensor Library Nodes

ReduceNode expects scalars or flat pointers of scalars as inputs. The tensor operation is performed with linearized indices.

Reduction Operations

Reduction operations include:

• Sum, mean, std: Statistical reductions
• Min, max: Value reductions
• Softmax: Exponential normalization

Reductions can operate on:

• Specific axes (e.g., reduce along dimension 1)
• Multiple axes (e.g., reduce along dimensions 0 and 2)
• All axes (full reduction to scalar)

The keepdims parameter controls whether reduced dimensions are kept with size 1 or removed from the output shape.

Example

Creating a sum reduction:

// Reduce along last axis of shape [32, 64, 128]
std::vector<symbolic::Expression> shape = {
    symbolic::integer(32), symbolic::integer(64), symbolic::integer(128)
};
std::vector<int64_t> axes = {-1};  // Last axis
bool keepdims = false;
 
auto& sum_node = builder.add_library_node<math::tensor::SumNode>(
    block, debug_info, shape, axes, keepdims
);
 
// Output shape will be [32, 64] since last axis is reduced

See also

math::tensor::ElementWiseUnaryNode for elementwise operations

math::MathNode for expansion interface