How to get CL_KERNEL_GLOBAL_WORK_SIZE values? about opencl-clhpp

[noma]

For some entries here there are missing entries here.

Please try workaround I wrote you on stackoverflow.

    size_t result[3];
    kernel.getWorkGroupInfo<decltype(result)>(device, CL_KERNEL_GLOBAL_WORK_SIZE, result);

This seems to be a bug in cl.hpp and cl2.hpp.

[LeMoussel]

With your code, I got

error: no matching function for call to 'cl::Kernel::getWorkGroupInfo<size_t [3]>(cl::Device&, int, size_t [3])'
     miner_kernel_v3.getWorkGroupInfo<decltype(result)>(device, CL_KERNEL_GLOBAL_WORK_SIZE, result);
                                                                                                  ^
In file included from opencl2_error.hpp:7:0,
                 from CLclient.cpp:25:
cl2.hpp:5771:12: note: candidate: template<class T> cl_int cl::Kernel::getWorkGroupInfo(const cl::Device&, cl_kernel_work_group_info, T*) const
     cl_int getWorkGroupInfo(
            ^~~~~~~~~~~~~~~~
cl2.hpp:5771:12: note:   template argument deduction/substitution failed:
CLclient.cpp:412:98: note:   cannot convert 'result' (type 'size_t [3] {aka long long unsigned int [3]}') to type 'long long unsigned int (*)[3]'
     my_kernel.getWorkGroupInfo<decltype(result)>(my_device, CL_KERNEL_GLOBAL_WORK_SIZE, result);

I do this

cl::array<size_t, 3> kernel_global_work_size;
my_kernel.getWorkGroupInfo<cl::array<size_t, 3>>(my_device, CL_KERNEL_GLOBAL_WORK_SIZE, &kernel_global_work_size);

But I got OpenCL error -30 (Invalid Value)

[LeMoussel]

In cl2.hpp, I add this :
F(cl_kernel_work_group_info, CL_KERNEL_GLOBAL_WORK_SIZE, cl::detail::size_t_array) \
an do this
cl::array<size_t, 3> kernel_global_work_size = my_kernel.getWorkGroupInfo<CL_KERNEL_GLOBAL_WORK_SIZE>(my_device);

No compilation error but OpenCL error -30 (Invalid Value)

[noma]

Ok, so the workaround, using the other variant call (with cl::array of course...) basically works, and the header can be easily fixed.

Now with the Invalid value we are back at the specification, which is not a bug, but requires carefully reading the specification. I though this was the original problem, as you initially didn't post any compilation errors.

Please carefully read the second part of my stack overflow answer, I guess you are using a normal kernel on normal device, in which case the specification basically tells you, that CL_INVALID_VALUE is all you will get from that call.

[LeMoussel]

Yep I use a normal kernel on normal device and as indicated then it is normal to have CL_INVALID_VALUE

If device is not a custom device or kernel is not a built-in kernel, clGetKernelArgInfo returns the error CL_INVALID_VALUE.

[noma]

So problem solved?