negative_set_default_device_command_queue

[rjodinchr]

At the end of negative_set_default_device_command_queue the test is trying to create a queue that should not be on device to check that clSetDefaultDeviceCommandQueue returns the proper error code.

But to do that it tries to create a queue with only CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE.

I see 2 potential issues:

• Device not supporting CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE will fail on the creation of the queue which would be a false negative.
• Are we sure that that queue won't be on-device? It's not clear to me reading the specification that CL_QUEUE_DEVICE is required for the driver to provide an on-device queue. Thus not providing it won't guarantee that the created queue won't be on-device.

[bashbaug]

Discussed in the November 4th teleconference and added to the Mobica backlog.

The first issue mentioned above looks like a simple bugfix. We should not be creating an out-of-order queue when the device does not support it.

For the second issue, the test may be correct, but we're not sure. Suggested action is to check the description in the spec and raise an issue to the group of the spec text is unclear.

[shajder]

I just created a OpenCL-Docs issue to clarify requirements on clSetDefaultDeviceCommandQueue