Truncation of Float to Half in Intel Run Time Driver using OpenCL 2.0

[Raghu_K_]

Hello,

    The latest OpenCL Intel driver download uses OpenCL 2.0 and seem to truncate Floating point values to half values.  Does anyone know a workaround or a way to get a version of Intel openCL drivers that do not do this truncation?

Thanks,

[Raghu_K_]

BTW, this problem is seen with Intel drivers found at https://software.intel.com/en-us/articles/opencl-drivers but not seen when I use drivers from http://developer.amd.com/tools-and-sdks/opencl-zone/amd-accelerated-parallel-processing-app-sdk/

[Robert_I_Intel]

Hi Raghu,

Which Intel processor are you running on? Which OS? Which version of the driver are you using? Do you have a reproducer that clearly shows the issue?

[Raghu_K_]

Hi Robert, we use i5 or i7 destops running win 8.1.  We downloaded the latest openCL driver from the link I sent that uses openCL 2.0.  You can use the pair programming box to reproduce the problem.  Thanks, Raghu

[Robert_I_Intel]

Hi Raghu,

I am not familiar with the pair programming box - what is it? Is it a reproducer? Could you provide a link?

Thanks!