Archives Discussions

[Fuxianjun]

Assume there are n work-groups in a kernel, and each work-group works out an array on local memory with the same length(i.e.array0,array1,array2,……,array(n-1), all arrays with the same length and on different local memory), how to add them up to an  array in the same kernel ?

I'm afraid that because there is no synchronization with different work-group, it can not insure every array is added up.

Dose async_work_group_copy() work on this problem? If yes ,how to use it ?

Thanks !

[himanshu_gautam]

Fuxianjun,

You can use Events to solve the sync issue.

define an event say e1 for the async_work_group_copy() command.

then specify this event in the wait_group_evets for the commands using the data.

Refer to OpenCL spec 1.1 pg 227 for detailed explaination