Intel GPU BARRIER(CLK_GLOBAL_MEM_FENCE) behaviour

[andradx]

Can anyone please direct me to a documentation source on the intel Ivy Bridge GPU? Namely, I'm trying to figure out if BARRIER(CLK_GLOBAL_MEM_FENCE) will behave similarly to ADM or Nvidia devices. If there is an over-subscription of hardware threads will it draw a device deadlock with threads spinning on threads that cannot spawn reaching the barrier. TY for your help.

[Maxim_S_Intel]

If workitems within a workgroup follow the same path ( e.g. when your barrier is under if- clause), then no deadlock should occur by the best of my knowledge. All vendors follow the same OpenCL* pec so I suggest you rely on the specific description and recomendations from the OpenCL* standard as well.