Valid Size Arguments to clEnqueueSVMMemFill and clEnqueueSVMMemcpy

[bashbaug]

Currently, the spec is unclear whether zero should be allowed for the size argument to clEnqueueSVMMemFill and clEnqueueSVMMemcpy.

For the cl_khr_unified_svm extension, we are leaning towards allowing the size to be zero for clEnqueueSVMMemFill and clEnqueueSVMMemcpy. Much like a zero-sized enqueue, when the size is zero the command will behave similarly to an enqueued marker.

Should we clarify that a size of zero is valid in general, independent of the cl_khr_unified_svm extension?

[bashbaug]

Here are some draft tests to check how an implementation behaves when it is passed a size of zero for these SVM APIs and a few others:

• https://github.com/KhronosGroup/OpenCL-CTS/pull/2431

Most of the implementations I've tested return an error in at least a few cases.

[bashbaug]

Discussed in the November 4th memory subgroup:

• We aren't going to require any new behavior for implementations that do not support cl_khr_unified_svm.
• We are going to test this behavior for implementations that do support cl_khr_unified_svm, even if the allocations are made via the older SVM APIs and not the newer unified SVM APIs.

For more detail, please see the linked CTS PR.