get_global_id(0) cause much latency ?

Thanks for the replies

I did more testing and find that if I didn't use the "reqd_work_group_size(2, 1, 1)", including "begin=get_global_id(0)"(still not using it) it would run slower then single pipeline;

but when having the reqd_work_group_size attribute, including or comment out "begin=get_global_id(0)" made no difference, both slower then single pipeline.

the difference are like 13ms and 1.5ms each frame.

I guess it's like you said because of the pipeline length. Could you tell me where did you see the 32 and 192 cycle number? I didn't see them in the report.html

Thanks