support double precision for Geforce GPU

OpenCL
1

Hi,
I have a question about using double in Geforce GPU.
The extension of my GPU includes cl_khr_fp64 and the OpenCl version is 1.1.
So, I think I can use double type in my kernel.
However, if I add #pragma OPENCL EXTENSION cl_khr_fp64: enable, I got error while building.
If I remove it, the kernel can be built and run.

Any suggestions are appreciated.

Thanks a lot!

EJ

2

That should enable double precision and I’ve used it many times myself with GeForce GPUs. What error do you get? Please post a minimum kernel that demonstrates this error.

3

Thanks. It does support. I was mislead by the output error information.