Parallella Community

Are there any special functions I can use to read/write other core's local memory from OpenCL kernel code, or do I need to build the other core addresses manually? How can I determine the row/col id from the id returned by get_global_id()?

There are, or rather were, but I need to re-validate them. I am working on a project where I happen to need to
do this so I can get back to you with a more clear answer soon. FYI here is the API that once worked on an earlier prototype platform - I will just list the calls and maybe you can read between the lines since they are designed to support exactly what you are trying to do. I do not want to post the full API spec until I can re-validate since I do not want to post something that people cannot actually use.

typedef threadspec_t;

// get thread handle based on OpenCL local ID
threadspec_t get_thread( int dim, unsigned int* ltid );

// copy contents of array into same array of another thread
void* memsync( void* ptr, size_t n, threadspec_t thrs, int flags );

// copy contents of array to a different array of another thread
void* memsend( void* src, size_t n, threadspec_t thrs, int flags );

// receive contents of an array from another thread
void* memrecv( void* dst, size_t n, threadspec_t thrs, int flags );

memsend/memrecv must be paired.

Awesome! Thanks again for your quick, detailed response. Looking forward to this capability!

djm

Try adding a barrier after the loop that initializes the local memory but before you attempt to write into other core's memory.

See if you get same results.

Yes, that works perfectly. I should have seen that. Thanks very much!

Hi djm,

Could you post the working version of the code, please? I'm yet to get my head around this inter-core data passing.

nick

Thanks.