[CI] Build LLD when building AMDGPU (#18185)

This fixes the nightly oneAPI test job where we don't build HIP but
build AMDGPU and then try to compile an AMDGPU test.


[Bad](https://github.com/intel/llvm/actions/runs/14632659184/job/41058003341)


[Good](https://github.com/intel/llvm/actions/runs/14647335163/job/41105420068)

---------

Signed-off-by: Sarnie, Nick <nick.sarnie@intel.com>

Author: sarnex

buildbot/configure.py

@@ -78,10 +78,6 @@ def do_configure(args, passthrough_args):
             sycl_enabled_backends.append("level_zero")
             sycl_enabled_backends.append("level_zero_v2")
 
-    # lld is needed on Windows or for the HIP adapter on AMD
-    if platform.system() == "Windows" or (args.hip and args.hip_platform == "AMD"):
-        llvm_enable_projects += ";lld"
-
     libclc_enabled = args.cuda or args.hip or args.native_cpu
     if libclc_enabled:
         llvm_enable_projects += ";libclc"
@@ -171,6 +167,10 @@ def do_configure(args, passthrough_args):
     if args.disable_preview_lib:
         sycl_preview_lib = "OFF"
 
+    # lld is needed on Windows or when building AMDGPU
+    if platform.system() == "Windows" or "AMDGPU" in llvm_targets_to_build:
+        llvm_enable_projects += ";lld"
+
     install_dir = os.path.join(abs_obj_dir, "install")
 
     cmake_cmd = [