Enable OpenMP task parallelism for C kernels

AMSS_NCKU_source/lopsided_kodis_c.C

@@ -43,7 +43,13 @@ void lopsided_kodis(const int ex[3],
     const size_t nz = (size_t)ex3 + 3;
     const size_t fh_size = nx * ny * nz;
 
-    double *fh = (double*)malloc(fh_size * sizeof(double));
+    static thread_local double *fh = NULL;
+    static thread_local size_t cap = 0;
+    if (fh_size > cap) {
+        free(fh);
+        fh = (double*)aligned_alloc(64, fh_size * sizeof(double));
+        cap = fh_size;
+    }
     if (!fh) return;
 
     symmetry_bd(3, ex, f, fh, SoA);
@@ -243,6 +249,4 @@ void lopsided_kodis(const int ex[3],
             }
         }
     }
-
-    free(fh);
 }