Suggestion Description
cudaMemcpy2D accepts fortran pointers to arrays with ranks greater than 2., while hip equivalent does not.
Is this a conscious choice? I have code which uses memcpy2d on 3- and 4-d arrays. I can obviously rewrite, but was hoping to keep changes with respect to cuda version minimal.
Operating System
No response
GPU
No response
ROCm Component
No response