TensorTypes.h File
Tensor dtype, axis semantics, layout token, and minimal DLPack-like structs. More...
Included Headers
Namespaces Index
| namespace | simaai |
| namespace | neat |
| namespace | dlpack |
|
Minimal DLPack-shaped structs for zero-copy interop with NumPy/PyTorch. More... | |
Classes Index
| struct | DLDevice |
|
DLPack device descriptor: type + numeric ID. More... | |
| struct | DLDataType |
|
DLPack scalar dtype: code + bit width + SIMD lane count. More... | |
| struct | DLTensor |
|
DLPack non-owning tensor view: pointer + shape + strides + dtype + device. More... | |
| struct | DLManagedTensor |
|
DLPack managed tensor: owning wrapper that calls deleter on destruction. More... | |
Description
Tensor dtype, axis semantics, layout token, and minimal DLPack-like structs.
This header defines the small enum vocabulary every tensor consumer needs: data type (TensorDType), per-axis semantic tag (TensorAxisSemantic, the long-term direction), legacy layout token (TensorLayout, transitional), plus minimal DLPack-shaped structs for zero-copy bridges to NumPy/PyTorch via the standard DLPack interchange format.
- See Also
TensorCore.h for the Tensor struct that uses these
- See Also
"TensorAxisSemantic" rationale (§0.10 of the design deep dive)
Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.8.