Implemented a work-around for issue with forthcoming compiler

Author: oleksandr-pavlyk

dpctl/tensor/libtensor/include/kernels/elementwise_functions/proj.hpp

@@ -125,6 +125,30 @@ sycl::event proj_contig_impl(sycl::queue exec_q,
         exec_q, nelems, arg_p, res_p, depends);
 }
 
+template <typename argTy>
+sycl::event
+proj_workaround_contig_impl(sycl::queue exec_q,
+                            size_t nelems,
+                            const char *arg_p,
+                            char *res_p,
+                            const std::vector<sycl::event> &depends = {})
+{
+    using resTy = typename ProjOutputType<argTy>::value_type;
+
+    const argTy *arg_tp = reinterpret_cast<const argTy *>(arg_p);
+    resTy *res_tp = reinterpret_cast<resTy *>(res_p);
+
+    sycl::event e = exec_q.submit([&](sycl::handler &cgh) {
+        cgh.depends_on(depends);
+        cgh.parallel_for({nelems}, [=](sycl::id<1> id) {
+            size_t i = id[0];
+            res_tp[i] = ProjFunctor<argTy, resTy>{}(arg_tp[i]);
+        });
+    });
+
+    return e;
+}
+
 template <typename fnT, typename T> struct ProjContigFactory
 {
     fnT get()
@@ -135,8 +159,14 @@ template <typename fnT, typename T> struct ProjContigFactory
             return fn;
         }
         else {
-            fnT fn = proj_contig_impl<T>;
-            return fn;
+            if constexpr (std::is_same_v<T, std::complex<double>>) {
+                fnT fn = proj_workaround_contig_impl<T>;
+                return fn;
+            }
+            else {
+                fnT fn = proj_contig_impl<T>;
+                return fn;
+            }
         }
     }
 };