KhronosGroup
Added comparability verification for GL associated devices query
I've tested this PR with linux and windows, currently I am not able to launch the test on MacOS - I will appreciate confirmation from this platform.
GLES version does not work for me, neither on ubuntu nor on windows. Is this sample operational ?
@bashbaug thanks for your feedback. The confusion around cl_context_properties comes from the differencies in creating context between platforms. Please note that for OSX and X11 existing setup does not use CL_CONTEXT_PLATFORM property to create context. Then, if I will use clGetContextInfo(CL_CONTEXT_PROPERTIES) to collect properties to be used later with clGetGLContextInfoKHR the test fails at my machine. I am aware this is inconsistency either in spec or driver implementation (most probably clGetGLContextInfoKHR implementation in my nvidia driver). Spec for clGetGLContextInfoKHR says:
properties points to an property list whose format and valid contents are identical to the properties argument of clCreateContext
which means your suggestion should work. Now if I extend list of properties for creation of a context with CL_CONTEXT_PLATFORM my X11 setup passes the test so we could try unifying properties list among platforms but I have no OSX device to verify that (thus I separated clCreateContext and clGetGLContextInfoKHR props). Can you do OSX test so I could extend props list with CL_CONTEXT_PLATFORM ?
I think this question is in reference to https://github.com/KhronosGroup/OpenCL-CTS/pull/2231#discussion_r1960998417 - is that correct? If so, note that replying to review comments tends to thread better, and then we can resolve the discussion once it's closed.
This seems like a problem:
Please note that for OSX and X11 existing setup does not use CL_CONTEXT_PLATFORM property to create context.
When CL_CONTEXT_PLATFORM is not specified in the properties, the default platform is "implementation-defined", so if this is working currently then we are getting lucky. It sounds like if we fix this and pass a CL_CONTEXT_PLATFORM then we can use the CL_CONTEXT_PROPERTIES query to make the new test more self-contained, so this definitely gets my vote.
I don't have an OSX system setup to test with. If no other working group members do either, I think we should just do our best, which probably means including the CL_CONTEXT_PLATFORM for OSX as well.
