diff --git a/VAX/vax_sysdev.c b/VAX/vax_sysdev.c
index 885e64b8..00fe816b 100644
--- a/VAX/vax_sysdev.c
+++ b/VAX/vax_sysdev.c
@@ -481,16 +481,17 @@ return ((val << 24) & 0xff000000) | (( val << 8) & 0xff0000) |
     ((val >> 8) & 0xff00) | ((val >> 24) & 0xff);
 }
 
+volatile int32 rom_loopval = 0;
+
 int32 rom_read_delay (int32 val)
 {
 uint32 i, l = rom_delay;
-int32 loopval = 0;
 
 if (rom_unit.flags & UNIT_NODELAY)
     return val;
 
 /* Calibrate the loop delay factor when first used.
-   Do this 4 times to and use the largest value computed. */
+   Do this 4 times and use the largest value computed. */
 
 if (rom_delay == 0) {
     uint32 ts, te, c = 10000, samples = 0;
@@ -503,15 +504,15 @@ if (rom_delay == 0) {
    away by a good compiler. loopval always is zero.  To avoid smart compilers,
    the loopval variable is referenced in the function arguments so that the
    function expression is not loop invariant.  It also must be referenced
-   by subsequent code or to avoid the whole computation being eliminated. */
+   by subsequent code to avoid the whole computation being eliminated. */
 
         for (i = 0; i < c; i++)
-            loopval |= (loopval + ts) ^ rom_swapb (rom_swapb (loopval + ts));
+            rom_loopval |= (rom_loopval + ts) ^ rom_swapb (rom_swapb (rom_loopval + ts));
         te = sim_os_msec (); 
         if ((te - ts) < 50)                         /* sample big enough? */
             continue;
-        if (rom_delay < (loopval + (c / (te - ts) / 1000) + 1))
-            rom_delay = loopval + (c / (te - ts) / 1000) + 1;
+        if (rom_delay < (rom_loopval + (c / (te - ts) / 1000) + 1))
+            rom_delay = rom_loopval + (c / (te - ts) / 1000) + 1;
         if (++samples >= 4)
             break;
         c = c / 2;
@@ -521,8 +522,8 @@ if (rom_delay == 0) {
     }
 
 for (i = 0; i < l; i++)
-    loopval |= (loopval + val) ^ rom_swapb (rom_swapb (loopval + val));
-return val + loopval;
+    rom_loopval |= (rom_loopval + val) ^ rom_swapb (rom_swapb (rom_loopval + val));
+return val + rom_loopval;
 }
 
 int32 rom_rd (int32 pa)