Arrayfun/gpuArray CUDA kernel need to be able to remember previous steps

Question

Anders 2024-3-25

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2098621-arrayfun-gpuarray-cuda-kernel-need-to-be-able-to-remember-previous-steps

回答： Joss Knight 2024-3-29

Background

The problem can be separated into a large number of independent sub-problems.
All sub-problems share the same matrix parameters.
Each sub-problem needs to remember the indices itself has visited up to this point.
The goal is to process the sub-problems in paralell on the GPU.

Array indexing and memory allocation is not supported in this context. Is this function possible to achieve?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Joss Knight 2024-3-29

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2098621-arrayfun-gpuarray-cuda-kernel-need-to-be-able-to-remember-previous-steps#answer_1433481

This is a bit too vague to answer. Without indexing, how can each subproblem retrieve its subset of the data? If you just mean indexed assignment is not allowed then sure, you could write an arrayfun perhaps that solves some independent problem for a subset of an array, as long as all the operations are scalar and the output is scalar. Not if the subproblems are completely different algorithms though.

Anyway, sorry, but not enough information to help.