topic Sorry for the delayed reply. in OpenCL* for CPU

Reimplementing work_group_scan_exclusive_add

Mathias_M_1 — Thu, 18 Aug 2016 21:59:56 GMT

Hi there,

I have reimplemented the function work_group_scan_exclusive_add, which is supported by OpenCL 2.0. My OpenCL source code looks like this:

#if __OPENCL_VERSION__ < 200
inline uint work_group_scan_exclusive_add(const uint value)
{
...
}
#endif

When I try to compile this with the Intel OpenCL driver 6.4.0.25 (Windows 8.1 x64) for the OpenCL 1.2 CPU device, the following error message is generated by the intel OpenCL compiler:

error: redeclaration of 'work_group_scan_exclusive_add' must have the 'overloadable' attribute

Why is the function work_group_scan_exclusive_add already defined for the OpenCL 1.2 CPU device?
How it is possible to circumevent this compile error ?

Best regards,
Mathias

Sorry for the delayed reply.

Jeffrey_M_Intel1 — Sat, 10 Sep 2016 01:53:10 GMT

Sorry for the delayed reply. You may need to use an older runtime only supporting CPU OpenCL 1.2 to avoid this problem.

https://software.intel.com/en-us/articles/opencl-drivers

(See the CPU runtimes section. OpenCL 2.0 was introduced for the 16.1 runtimes.)

I'm checking for other alternatives and will get back to you if I find any.

After further internal

Jeffrey_M_Intel1 — Wed, 14 Sep 2016 05:57:58 GMT

After further internal discussion we're going to investigate this as a bug. As a workaround the overloadable attribute might help:

__kernel __attribute__((overloadable)) uint work_group_scan_exclusive_add(const uint value)