Check that `clSetKernelArg` called with `arg_size` of 0 for an argument in the `local` address space returns `CL_INVALID_ARG_SIZE`

[kpet]

We could potentially widen this issue to cover all negative testing for clSetKernelArg

Work plan:

test_api

• CL_INVALID_KERNEL - kernel is not a valid kernel object
• CL_INVALID_ARG_INDEX - arg_index is not a valid argument index
• CL_INVALID_ARG_VALUE - image declared with the read_only qualifier and arg_value refers to an image object created with cl_mem_flags of CL_MEM_WRITE_ONLY, PR Added test to verify negative result of clSetKernelArg with CL_INVALID_ARG_VALUE #2445
• CL_INVALID_ARG_VALUE - image argument is declared with the write_only qualifier and arg_value refers to an image object created with cl_mem_flags of CL_MEM_READ_ONLY, PR Added test to verify negative result of clSetKernelArg with CL_INVALID_ARG_VALUE #2445
• CL_INVALID_MEM_OBJECT - argument declared to be a memory object when the specified arg_value is not a valid memory object
• CL_INVALID_SAMPLER - type sampler_t when the specified arg_value is not a valid sampler object
• CL_INVALID_DEVICE_QUEUE - type queue_t when the specified arg_value is not a valid device queue object
• CL_INVALID_ARG_SIZE - arg_size does not match the size of the data type for an argument that is not a memory object
• CL_INVALID_ARG_SIZE - argument is a memory object and arg_size != sizeof(cl_mem)
• CL_INVALID_ARG_SIZE - arg_size is zero and the argument is declared with the local qualifier, PR Added test to verify negative result of clSetKernelArg with CL_INVALID_ARG_SIZE and local qalified argument #2449
• CL_INVALID_ARG_SIZE - argument is a sampler and arg_size != sizeof(cl_sampler)
• CL_INVALID_MEM_OBJECT - argument declared to be a depth image, depth image array, multi-sample image, multi-sample image array, multi-sample depth image, or a multi-sample depth image array when the specified arg_value does not follow the rules described above for a depth memory object or memory array object argument

[shajder]

After reviewing CTS source code I identified below missing nevative coverage for clSetKernelArg:

edit: work plan moved to issue description to align with github sub-tasks scheme

[bashbaug]

CL_INVALID_ARG_VALUE - arg_value specified is not a valid value

Is this case sufficiently well-specified? What would be considered "not a valid value" that is not covered by another error condition?

Or to ask the question another way - what exactly would be tested for this error condition?

Note that we've been going through a similar exercise recently for SVM kernel arguments.

[shajder]

Is this case sufficiently well-specified?

I had nothing specific on my mind - just followed spec for clSetKernelArg. I conclude from your question this is not a real-life case, just a general phrase, is that right ?

[bashbaug]

I conclude from your question this is not a real-life case, just a general phrase, is that right ?

Yes, I think we need to ask ourselves if there is any condition that would be considered "not a valid value" that is not covered by other error conditions, and if not, should we remove this error condition from the spec?

Noting also: clSetKernelArg tends to be the most frequently called OpenCL API for most applications so it is very important for it to be fast. I haven't checked whether our implementation properly handles all of these error conditions, but if it doesn't and it'd be expensive to handle them, we should be sure adding the extra overhead is the right tradeoff.