Operations

cpptensor ships a broad operation set. Most ops register kernels in the runtime registry, which initializes lazily on first use.

Elementwise & unary#

• Arithmetic: +, -, *, /
• Unary math: elementwise functions over a tensor
• Comparisons: produce bool tensors

Tensor c = a + b;
Tensor mask = a > b;   // bool tensor

Reductions#

sum, mean, and other reductions collapse along dimensions. These are numerically careful — see Numerics & dtypes.

Tensor s = a.sum();
Tensor m = a.mean();

Shape operations#

reshape, transpose, concat, and stack rearrange data. Many return views where possible; materialize a contiguous copy when needed (see Tensors & views).

Linear algebra#

Adding new ops#

The repo documents how to add new ops and kernels (docs/HowToAddNewOpsWithKernel.md) and tracks operator status and semantics (docs/Ops_Status.md) in the source tree.