intel-oneapi-mkl: add missing compiler libraries for thread=openmp (#42087)

var/spack/repos/builtin/packages/intel-oneapi-mkl/package.py

@@ -165,6 +165,7 @@ def setup_dependent_build_environment(self, env, dependent_spec):
 
     def _find_mkl_libs(self, shared):
         libs = []
+        threading_libs = []
 
         if self.spec.satisfies("+cluster"):
             libs.extend([self._xlp64_lib("libmkl_scalapack"), "libmkl_cdft_core"])
@@ -177,6 +178,12 @@ def _find_mkl_libs(self, shared):
                 libs.append("libmkl_intel_thread")
             else:
                 libs.append("libmkl_gnu_thread")
+
+            # this is slightly different than what link-line advisor suggests.
+            # here it uses what the compiler suggests to use to enable openmp,
+            # instead of being explicit about in which path openmp libraries
+            # are located (e.g. intel libiomp5, gcc libgomp, clang libomp).
+            threading_libs += [self.compiler.openmp_flag]
         else:
             libs.append("libmkl_sequential")
 
@@ -212,7 +219,10 @@ def _find_mkl_libs(self, shared):
         )
         lib_path = lib_path if isdir(lib_path) else dirname(lib_path)
 
+        # resolved_libs is populated as follows
+        # MKL-related + MPI-related + threading-related
         resolved_libs = find_libraries(libs, lib_path, shared=shared)
+
         # Add MPI libraries for cluster support. If MPI is not in the
         # spec, then MKL is externally installed and application must
         # link with MPI libaries. If MPI is in spec, but there are no
@@ -223,6 +233,9 @@ def _find_mkl_libs(self, shared):
                 resolved_libs = resolved_libs + self.spec["mpi"].libs
         except spack.error.NoLibrariesError:
             pass
+
+        resolved_libs += threading_libs
+
         return resolved_libs
 
     def _xlp64_lib(self, lib):