Is persistent L2 equal to cuda::annotated_ptr<int const, cuda::access_property::persisting?

cuda::annotated_ptr<int const, cuda::access_property::persisting> a_p {a}, b_p{b};

Well… I do not see any size limitation in the page: cuda::annotated_ptr | libcu++

Is it the same to persistant L2?

It is not the same. Persitance API reserve a portion of the cache. access_propertiy::persisting does not. It is only a hint

1 Like

Thanks!! So now I am using several streams, but the persistent L2 API has to be bound to one stream… How can I solve it? I… have to use only one stream?

I have not used the L2 API before. What happens if you set 50% for stream 1 and 50% for stream 2? Does it reset stream 1?

Oh…Actually my algorithm is, stream 1 produce data, and it will be used for stream 2… So…Maybe I persist it in stream 1. But not sure whether stream 2 can read it…Well, let me design an experiment to test it.