diff -ruN tesseract-4.0.0-beta.4/src/lstm/lstm.cpp tesseract-4.0.0-beta.4-openmp/src/lstm/lstm.cpp
--- tesseract-4.0.0-beta.4/src/lstm/lstm.cpp	2018-07-27 15:10:37.000000000 +0900
+++ tesseract-4.0.0-beta.4-openmp/src/lstm/lstm.cpp	2018-10-07 08:15:23.139366142 +0900
@@ -36,9 +36,7 @@
 // Macros for openmp code if it is available, otherwise empty macros.
 #ifdef _OPENMP
 #define PARALLEL_IF_OPENMP(__num_threads)                                  \
-  PRAGMA(omp parallel if (__num_threads > 1) num_threads(__num_threads)) { \
-    PRAGMA(omp sections nowait) {                                          \
-      PRAGMA(omp section) {
+  PRAGMA(omp parallel if (__num_threads > 1) num_threads(__num_threads)) {
 #define SECTION_IF_OPENMP \
         } \
     PRAGMA(omp section) \
@@ -320,6 +318,10 @@
     if (!source_.int_mode()) source_.ReadTimeStep(t, curr_input);
     // Matrix multiply the inputs with the source.
     PARALLEL_IF_OPENMP(GFS)
+#ifdef _OPENMP
+    PRAGMA(omp sections nowait) {
+    PRAGMA(omp section) {
+#endif
     // It looks inefficient to create the threads on each t iteration, but the
     // alternative of putting the parallel outside the t loop, a single around
     // the t-loop and then tasks in place of the sections is a *lot* slower.
@@ -575,7 +577,10 @@
 #endif
     // Matrix multiply to get the source errors.
     PARALLEL_IF_OPENMP(GFS)
-
+#ifdef _OPENMP
+    PRAGMA(omp sections nowait) {
+    PRAGMA(omp section) {
+#endif
     // Cell inputs.
     node_values_[CI].FuncMultiply3<GPrime>(t, node_values_[GI], t,
                                            curr_stateerr, gate_errors[CI]);