[WIP][SYCL] Fix kernel name mangling for CUDA/HIP with Microsoft ABI host

sycl/test-e2e/IntermediateLib/dynamic_app_linux.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: level_zero && linux
+// REQUIRES: linux
 
 // build shared library
 // RUN: rm -rf %t.dir; mkdir -p %t.dir

sycl/test-e2e/IntermediateLib/multi_lib_app.cpp

@@ -1,6 +1,3 @@
-// UNSUPPORTED: cuda || hip
-// UNSUPPORTED-TRACKER: CMPLRLLVM-69415
-
 // DEFINE: %{fPIC_flag} =  %if windows %{%} %else %{-fPIC%}
 // DEFINE: %{shared_lib_ext} = %if windows %{dll%} %else %{so%}
 
@@ -29,6 +26,26 @@
 
 // RUN:  env UR_L0_LEAKS_DEBUG=1 %{run} %t.out
 
+// UNSUPPORTED: cuda, hip
+// UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/CMPLRLLVM-69415
+//
+// This test intentionally uses the SAME kernel name ("class same") in three 
+// different shared libraries to verify that the runtime can properly isolate
+// and distinguish kernels from different libraries.
+//
+// CUDA and HIP runtimes identify kernels solely by their mangled name without
+// library-level tracking. When multiple shared libraries contain kernels with
+// identical names, the runtime encounters name collisions and cannot determine
+// which kernel to invoke. This is an architectural limitation of CUDA/HIP.
+//
+// Level Zero runtime maintains library handles and can distinguish between
+// identically-named kernels from different libraries, hence this test passes
+// on Level Zero.
+//
+// For typical usage with distinct kernel names across libraries, see
+// dynamic_app_linux.cpp which tests basic dynamic loading/unloading and
+// works on all backends.
+
 // XFAIL: target-native_cpu
 // XFAIL-TRACKER: https://github.com/intel/llvm/issues/20142
 

sycl/test-e2e/Printf/int.cpp

@@ -6,8 +6,8 @@
 //
 // UNSUPPORTED: target-amd
 // FIXME: The 'short' type gets overflown with sporadic values on CUDA.
-// XFAIL: cuda
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/14734
+// TODO: Re-enable XFAIL if cross-ABI mangling fix doesn't resolve the issue.
+// Previously: XFAIL: cuda (https://github.com/intel/llvm/issues/14734)
 
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out | FileCheck %s

sycl/test-e2e/Printf/mixed-address-space.cpp

@@ -2,8 +2,8 @@
 // for constant and generic address space can be used in the same module.
 //
 // UNSUPPORTED: target-amd
-// XFAIL: cuda && windows
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/14733
+// TODO: Re-enable XFAIL if cross-ABI mangling fix doesn't resolve the issue.
+// Previously: XFAIL: cuda && windows (https://github.com/intel/llvm/issues/14733)
 // FIXME: Drop the test once generic AS support is considered stable and the
 //        dedicated constant AS overload of printf is removed from the library.
 //

sycl/test-e2e/Printf/percent-symbol.cpp

@@ -5,8 +5,8 @@
 // [1]: https://en.cppreference.com/w/cpp/io/c/fprintf
 //
 // UNSUPPORTED: target-amd
-// XFAIL: cuda && windows
-// XFAIL-TRACKER: https://github.com/intel/llvm/issues/14733
+// TODO: Re-enable XFAIL if cross-ABI mangling fix doesn't resolve the issue.
+// Previously: XFAIL: cuda && windows (https://github.com/intel/llvm/issues/14733)
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out | FileCheck %s
 // FIXME: Remove dedicated constant address space testing once generic AS