for loop pipelined with NDRange

thank you for reply again.

I want to know how can two thread in a work group issued at the same time?

if I have my kernel code absence of SIMD, the execution will be like:

read(); --->WI 2

cal(); --->WI 1

write(); --->WI 0

or

--->WI 1...waiting for WI 0 finish

read();

cal();

write(); --->WI 0

and If I set SIMD to 2, Is the execution like:

read(); --->WI 3, WI 4

cal(); --->WI 2, WI 3

write(); --->WI 0, WI 1

or

--->WI 2, WI 3...waiting

read();

cal();

write(); --->WI 0, WI 1

or SIMD will only increase the performance of reading global memory like:

read(); --->WI 2, WI 3...waiting

cal(); --->WI 1

write(); --->WI 0