Does OpenCL1.1 on Mali-T760 MP4 support cl_ext_device_fission?

Hi, everyone

Background:

I am working on a project utilizing MACE(https://github.com/XiaoMi/mace a CNN inference framework designed for mobile devices like Mali GPU). This framework uses OpenCL to perform computation on Mali GPU. And I have an Android device containing a Mali-T760 MP4. It allows me to try MACE as a jni library on my android project.

When I call MACE's API to start some time-consuming statements, which does computations on Mali GPU, I observe severe FPS drop of Android GUI. Here are screen shots from Mali Graphics Debugger and Streamline.

(Note that at the center of the screen shot, regular frame behavior is stalled )
(Here, we can see the FPS drop whenever Mali Clock reaches a peak value)

After googling around for a while and some dig inside OpenCL 1.1 Specification Docs, I think OpenCL is taking as much resources as possible and preventing others make use of GPU during OpenCL's computation. And there is an extension of OpenCL called "cl_ext_device_fission"(www.khronos.org/.../cl_ext_device_fission.txt) which sub-divides an OpenCL device into multiple sub-devices and execute any work enqueued on the sub-device.

Question:

1.  Does OpenCL1.1 on Mali-T760 MP4 support cl_ext_device_fission?

2. Are there any other ways to reserve part of the GPU for android GUI rendering while OpenCL is doing computation?