If you use clEnqueueNDRangeKernel with the global_work_offset set to NULL or all zeros, you can use this attribute to tell the compiler that the global_work_offset is always zero.

This attribute can improve memory performance when you have memory accesses like:
A[get_global_id(x)] = ...;
Note: You can specify reqd_work_group_size, vec_type_hint, and xcl_zero_global_work_offset together to maximize performance.


Place this attribute before the kernel definition, or before the primary function specified for the kernel:
__kernel __attribute__((xcl_zero_global_work_offset))
void test (__global short *input, __global short *output, __constant short *constants) { }

See Also