Skip to content

[FEA]: Identify gaps in tensor algorithms coverage #9030

@gevtushenko

Description

@gevtushenko

Is this a duplicate?

Area

CUB

Is your feature request related to a problem? Please describe.

As of #7277 @fbusato added tensor copy algorithm. We occasionally get requests for other classes of tensor algorithms, such as reductions and cumulative sums along a dimension. Before opening implementation issues, we should identify which algorithms need mdspan overloads or tensor-aware APIs.

Describe the solution you'd like

As a first step, we should review CuPy (cc @leofang) and similar array libraries to see which generic kernels they had to reimplement to support multidimensional, strided, broadcasted, or axis-based input. This should help us distinguish cases covered by existing primitives from cases where users still need custom tensor kernels.

Describe alternatives you've considered

No response

Additional context

No response

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

Status

Todo

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions