[SYCL] Enable checking the result of max_num_work_groups query with exceeded launch limits on more backends (hip and opencl) (#15369)

The HIP and OpenCL backend implementations of the query had a default
return `1` group implementation, which is an incorrect assumptions. They
will now be marked as _Unsupported_ with the accompanying UR chagnes
(see oneapi-src/unified-runtime#2038), so for
these cases the `kernel_queue_specific::max_num_work_groups` launch
query will rely on the fallback that returns either `1` or `0` groups
based on hardware resource limitation checks for the kernel.

---------

Co-authored-by: Aaron Greig <[email protected]>

Author: GeorgeWeb

sycl/cmake/modules/FetchUnifiedRuntime.cmake

@@ -117,13 +117,14 @@ if(SYCL_UR_USE_FETCH_CONTENT)
   endfunction()
 
   set(UNIFIED_RUNTIME_REPO "https://github.com/oneapi-src/unified-runtime.git")
-  # commit b79ebe4e98789144bcdf3832088eb6e6b5ae6967
-  # Merge: 7b4bc761 fbb6e862
-  # Author: Kenneth Benzie (Benie) <[email protected]>
-  # Date:   Fri Oct 4 16:39:59 2024 +0100
-  #     Merge pull request #2018 from wenju-he/L0-bindless-image-device-query
-  #     [L0] Fix device query of bindless image support
-  set(UNIFIED_RUNTIME_TAG b79ebe4e98789144bcdf3832088eb6e6b5ae6967)
+  # commit df6da35d6e67f2383db28dd49ab08c5c0ef541d2
+  # Merge: 67590533 55bd5636
+  # Author: aarongreig <[email protected]>
+  # Date:   Mon Oct 7 12:28:07 2024 +0100
+  #     Merge pull request #2038 from GeorgeWeb/georgi/unsupported-max-coop-wgsize
+  #     [UR][hip][opencl] Mark urKernelSuggestMaxCooperativeGroupCountExp as unsupported
+  #     instead of returning misleading default value
+  set(UNIFIED_RUNTIME_TAG df6da35d6e67f2383db28dd49ab08c5c0ef541d2)
 
   set(UMF_BUILD_EXAMPLES OFF CACHE INTERNAL "EXAMPLES")
   # Due to the use of dependentloadflag and no installer for UMF and hwloc we need

sycl/test-e2e/Basic/launch_queries/max_num_work_groups.cpp

@@ -161,10 +161,13 @@ int test_max_num_work_groups(sycl::queue &q, const sycl::device &dev) {
   // It cannot be possible to launch a kernel successfully with a configuration
   // that exceeds the available resources as in the above defined workGroupSize.
   // workGroupSize is larger than maxWorkGroupSize, hence maxWGs must equal 0.
-  // Note: Level-Zero currently always returns a non-zero value. While other
-  // backends (i.e., OpenCL, HIP) always return 1 in their implementations.
-  if (dev.get_backend() == sycl::backend::ext_oneapi_cuda) {
-    assert(maxWGs == 0 && "max_num_work_groups query failed");
+  // Note: Level-Zero currently always returns a non-zero value.
+  // TODO: Remove the backend condition once the Level-Zero API issue is fixed.
+  if (dev.get_backend() != sycl::backend::ext_oneapi_level_zero) {
+    assert(maxWGs == 0 &&
+           "max_num_work_groups query failed.\n"
+           "It should return 0 possible groups when the requested resources "
+           "by the lanuch config exceed those available in the hardware.");
   }
 
   // As we ensured that the 'max_num_work_groups' query correctly