$ diff -u mhz.c ../mhz.new/mhz.c
--- mhz.c   2018-02-04 12:30:45.000000000 +0200
+++ ../mhz.new/mhz.c    2018-02-04 21:56:41.147617493 +0200
@@ -29,14 +29,14 @@
  * to parallelize. The "asm" statements are here to prevent the compiler from
  * reordering this code.
  */
-#define dont_move(var) do { asm volatile("" : "=r"(var) : "0" (var)); } while (0)
+#define produce(var) do { asm volatile("" : "=r"(var) : ); } while (0)
+#define consume(var) do { asm volatile("" : : "r"(var)); } while (0)

-#define run1cycle_ae()   do { a ^= e; dont_move(a); } while (0)
-#define run1cycle_ba()   do { b ^= a; dont_move(b); } while (0)
-#define run1cycle_cb()   do { c ^= b; dont_move(c); } while (0)
-#define run1cycle_dc()   do { d ^= c; dont_move(d); } while (0)
-#define run1cycle_ed()   do { e ^= d; dont_move(e); } while (0)
-#define run1cycle_eb()   do { e ^= b; dont_move(e); } while (0)
+#define run1cycle_ae()   do { a ^= e; } while (0)
+#define run1cycle_ba()   do { b ^= a; } while (0)
+#define run1cycle_cb()   do { c ^= b; } while (0)
+#define run1cycle_dc()   do { d ^= c; } while (0)
+#define run1cycle_ed()   do { e ^= d; } while (0)

 #define run5cycles()                                    \
    do {                                            \
@@ -74,7 +74,13 @@
  */
 static __attribute__((noinline)) void loop50(unsigned int n)
 {
-   unsigned int a = 0, b = 0, c = 0, d = 0, e = 0;
+   unsigned int a, b, c, d, e;
+
+   produce(a);
+   produce(b);
+   produce(c);
+   produce(d);
+   produce(e);

    do {
        run10cycles();
@@ -83,6 +89,12 @@
        run10cycles();
        run10cycles();
    } while (__builtin_expect(--n, 1));
+
+   consume(a);
+   consume(b);
+   consume(c);
+   consume(d);
+   consume(e);
 }

 /* performs 250 operations in a loop, all dependant on each other, so that the
@@ -92,7 +104,13 @@
  */
 static __attribute__((noinline)) void loop250(unsigned int n)
 {
-   unsigned int a = 0, b = 0, c = 0, d = 0, e = 0;
+   unsigned int a, b, c, d, e;
+
+   produce(a);
+   produce(b);
+   produce(c);
+   produce(d);
+   produce(e);

    do {
        run10cycles();
@@ -103,6 +121,12 @@
        run100cycles();
        run100cycles();
    } while (__builtin_expect(--n, 1));
+
+   consume(a);
+   consume(b);
+   consume(c);
+   consume(d);
+   consume(e);
 }

 void run_once(long count)