unified memory in cuda 6 allocate data in unified memory and there is no more data migrate between cpu and gpus. they give an example of c/c++ version of how to allocate data in unified memory, and use cudamallocmanaged().
is this cudamallocmanaged can be used in pgfortran , and is pgfortran can use unified memory? if yes, can you give an example for it ?
thank you!
Hi zsh,
In the 14.7 release, we’ll add the “managed” attribute to CUDA Fortran data types so you wont need to call cudamallocmanaged. You’ll just use allocate/deallocate as normal.
Hope this helps,
Mat
is the lower version can use cudaallocmanaged? i used 11.* version.
is the lower version can use cudaallocmanaged? i used 11.* version.
No since you also need to use CUDA 6.0. PGI 11.10 only supports CUDA 4.0.
Given that technology changes, we highly recommend users keep their products under subscription so they have access to these new features as they become available. Please contact PGI Sales (sales@pgroup.com) or your local PGI reseller (How to Buy | Purchase | PGI) for information about updating your product.
- Mat