compositor layer performance diffs with S8 exynos and S8 snapdragon 835

hello everyone, i use compositor layer with sdk 1.18 for a single ui panel in unity 2017.1. With some test, i found overlay ui demo in Oculus sample framework( also 1.18 ) performance diff in S8 with different cpu (exynos and 835), which exynos get fps 60 and 835 only get 50fps. And in my project exynos get 60 while 835 onlyget 35fps, while scene only have a sinlge ui panel rendered with overlay.

 

So, i upgrade oculus sdk into 1.22 using unity 2017.3.0f3, the demo in framework get fps all 60fps nomatter  exynos or 835. But when i upgrade my project which has more complex model and only one ui panel with overlay tech, the performance with 835 get fps 45 fps while exynos get 60. there has optimize but not enough, when i turn off overlay, all cpu get fps 60 fps.

More infomation, 835 run apps whatever demo or my project, there has obvious dither when i turn my head, looks like time wrap not work in s8 835. is some advise to using overlay tech in s8 835? maybe i am not get the right way to using it. but s7 and s8 with exynos all get 60fps with same code and works well........

edit-2018-1-29:
i test build-in overlay demo(sdk 1.22 &2017.3) with unity profiler, the result is below:


the left is 835 and right is exynos, the most time cost different is PLAYER_SEND_FRAME_COMPLETE. after a few of frames, PLAYER_SEND_FRAME_COMPLETE
cost 9-10ms while exynos cpu just cost 1-2m. i have no idea about how to reduce the cost time of FRAMEEVENTS.XREndFrame which's father is  PLAYER_SEND_FRAME_COMPLETE in profiler tree.

the profiler compare below is my project(sdk 1.18 2017.1)

left is exynos & right is 835

is someone can help me figure it out? thanks for your attention