[FEA]: Add out of bound assertion to cuda::std::mdspan on operator()
Closed this issue · 2 comments
Is this a duplicate?
- I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct
Area
Thrust
Is your feature request related to a problem? Please describe.
cuda::std::mdspan doesn't trigger, in debug, an assertion when an out of bound access is performed via operator(), contrary to cuda::std::span (with its operator[]).
Describe the solution you'd like
cuda::std::mdspan should trigger an assertion, in debug, in both host and device just like cuda::std::span.
We will also have to update the following: https://github.com/NVIDIA/cccl/blob/main/docs/libcudacxx/standard_api/container_library/mdspan.rst
Describe alternatives you've considered
Using cuda::std::span but we want to promote cuda::std::mdspan as a safe multi-dimensional accessor
Additional context
Came in a discussion with @bernhardmgruber on slack. The goal is for it to be used in the "The Modern CUDA Developer’s Toolbox" future blogpost.
@fbusato since you are working on mdspan at the moment, maybe you have time to add an assertion to operator() and operator[].