Georgi Mirazchiyski

> I think we can implement this on any backend because we can always fall back and return the same value as `max_num_work_groups`. @gmlueck I see your point. Thanks for...

> If the only way to interpret `info::device::max_compute_units` meaningfully is to know what (device, backend) combination you're running on, what do we gain by exposing it as a general query?...

> On Intel GPUs a compute unit is an Execution Unit (EU) or XE Vector Engine (XVE), which has its own register file. The maximum number of sub-groups you can...

@Pennycook and @gmlueck Circling back to this. We are agreement with your argument above and decided to end up just implementing `max_num_work_groups` with device semantics and let SYCLcompat handle the...

Unrelated passing XFAIL: https://github.com/intel/llvm/actions/runs/10809701571/job/29985720662?pr=14333#step:22:2286 ``` ******************** Unexpectedly Passed Tests (1): SYCL :: OneapiDeviceSelector/no_duplicate_devices.cpp ``` see [#15341](https://github.com/intel/llvm/issues/15341) and [#15288](https://github.com/intel/llvm/issues/15288)

Wrt the [Doxygen docs build failure](https://github.com/intel/llvm/actions/runs/10809701579/job/29989266991?pr=14333): see issue https://github.com/intel/llvm/issues/15355 for the failing Generate Doxygen documentation build action. The link pointing to the now removed kernel-fusion extension is wrong.

As @AlexeySachkov already noted (thanks). The follow-up PR that should fix the failure is this one https://github.com/intel/llvm/pull/15359.