[SYCL][E2E] limit the xfail to the failing cases in a SG32/joint_matrix_bfloat16.cpp

[dkhaldi]

No description provided.

[YuriPlyakhin]

this patch LGTM.
Just a side question: do we really need to keep

if (combinations[i].nsize == 8) { // architecture::intel_gpu_dg2*

in this test? Any plan to support SG=32 on DG2?

[dkhaldi]

if (combinations[i].nsize == 8) { // architecture::intel_gpu_dg2*

in this test? Any plan to support SG=32 on DG2?

Yes I also saw it was there at the same time as unsupported for DG2.
I think it is fine to keep it there to make the query comprehensive as this is really about the N value (in case we do support N=8 (not DG2 SG32)).

[dkhaldi]

@intel/llvm-gatekeepers, This should be ready to merge.
The failure (SYCL :: KernelCompiler/sycl_device_globals.cpp) in CI is already reported in #19965