awkehwe82397rfaowUI

Plans for double-precision support?

Discussion created by awkehwe82397rfaowUI on Jan 7, 2010

Are there any plans to turn the experimental double precision support of the SDK 2.0 into full support? Any idea of how long it will take if there are plans?

Just to let others know:

I've read the double precision info page at http://developer.amd.com/support/KnowledgeBase/Lists/KnowledgeBase/DispForm.aspx?ID=88. It says "Use of double-precision floating point arithmetic on GPUs other than ATI Radeon™ HD 5800 series GPUs and ATI Radeon™ HD 5900 series GPUs may result in unexpected behavior."

I can confirm that on my machine, if you try enabling double precision support with #pragma OPENCL EXTENSION cl_khr_fp64 : enable and use the cpu to do some simple double precision arithmetic, it will calculate correctly but the built in math functions are not implemented.

Outcomes