Expand description
Tensor linear algebra on Ruda device tensors.
Use einsum for index expressions or construct an OperationDescriptor
and reuse a Plan with different buffers and scalar coefficients.
Structs§
- Einsum
Plan - A reusable einsum expression with fixed input shapes, strides and storage types.
- Operand
Descriptor - Associates index names and a unary transform with a tensor descriptor.
- Operation
Descriptor - A buffer-independent operation. Input order is the order passed to its constructor.
- Plan
- A reusable index map and execution description, independent of buffer addresses.
- Ruda
Tensor - The basic tensor primitive struct.
- Tensor
Descriptor - Extents and element strides, relative to the start of a tensor’s buffer view.
Enums§
- Binary
Op - Compute
Type - Arithmetic precision, including operand transforms and scalar coefficients.
- DType
- Error
- Errors detected before a device operation is submitted.
- Reduction
Op - UnaryOp
Functions§
- contract
- Sum the product of A and B over modes absent from the output.
- einsum
- Evaluate an einsum expression on device tensors without reading values back to the host.
- einsum_
with_ options - elementwise_
binary - Named-axis broadcasting followed by a binary operation.
- elementwise_
trinary - Combine three broadcast operands as
(A op_ab B) op_abc C. - permute
- Materialize a permutation in a new packed row-major tensor; do not return an aliased view.
- reduce
- Reduce modes absent from
output_modes, in the specified output order.