Ben Ashbaugh

Discussed in the May 6th memory subgroup. In our mental model, we are creating a memory object "view" of the SVM allocation, and we are not be making a copy...

Discussed in the June 24th memory subgroup. Can we just ban creating a memory object from just part of an SVM allocation and call it an error unless a) the...

Discussed in the July 15th memory subgroup: * Banning creation of memory objects from a subset of an SVM allocation simplifies spec design work, potential implementation work, and testing scope,...

Interesting, here's a shorter version of this issue: https://godbolt.org/z/67arzdq5a We removed the alignment information from the scalar OpenCL types a very long time ago (see https://github.com/KhronosGroup/OpenCL-Headers/pull/40), but we didn't remove...

Hi, it's not clear to me what you are running to produce this output. Can you please provide more detail? If possible, can you also please provide information about the...

> Calling any other API on a multi-device context that includes a custom device will result in undefined/implementation defined behavior. I think this is probably a bit too strong. As...

Discussed in the June 24th teleconference: * Even if `CL_DEVICE_TYPE_ALL` is not inclusive of custom devices (see #300), if a platform exposes both a custom device and a non-custom device,...

Can you please include the device you're running on? Are you sure your device supports `cl_khr_mipmap_image`? FWIW I tried your example on an Intel GPU that supports `cl_khr_mipmap_image` and it...

Neat idea, I can see how this would be useful. > Maybe it's better to create a new extension for this: `cl_ext_compiler_il_only` which would define `CL_DEVICE_CLC_COMPILER_AVAILABLE_EXT`? Yes, I think a...

Not that I know of, but tagging @aharon-abramson who may have more insight.