Add missing barriers in gemm scheduler

a few places in the gemm scheduler code were missing barriers;
the code likely worked OK due to heavy use of volatile / _Atomic
but there's no reason to get this incorrect

driver/level3/level3_thread.c

@@ -347,7 +347,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
       /* Make sure if no one is using workspace */
       START_RPCC();
       for (i = 0; i < args -> nthreads; i++)
-	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;};
+	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;MB;};
       STOP_RPCC(waiting1);
 
 #if defined(FUSED_GEMM) && !defined(TIMING)
@@ -409,7 +409,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
 
 	  /* Wait until other region of B is initialized */
 	  START_RPCC();
-	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;};
+	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;MB;};
 	  STOP_RPCC(waiting2);
 
           /* Apply kernel with local region of A and part of other region of B */
@@ -427,6 +427,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
         /* Clear synchronization flag if this thread is done with other region of B */
 	if (m_to - m_from == min_i) {
 	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
+	  WMB;
 	}
       }
     } while (current != mypos);
@@ -488,7 +489,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
   START_RPCC();
   for (i = 0; i < args -> nthreads; i++) {
     for (js = 0; js < DIVIDE_RATE; js++) {
-      while (job[mypos].working[i][CACHE_LINE_SIZE * js] ) {YIELDING;};
+      while (job[mypos].working[i][CACHE_LINE_SIZE * js] ) {YIELDING;MB;};
     }
   }
   STOP_RPCC(waiting3);