1 files changed, 17 insertions, 9 deletions
diff --git a/target/arm/translate-a64.h b/target/arm/translate-a64.h
index dd9c09f89b..63d958cf50 100644
--- a/target/arm/translate-a64.h
+++ b/target/arm/translate-a64.h
@@ -67,18 +67,26 @@ static inline void assert_fp_access_checked(DisasContext *s)
 static inline int vec_reg_offset(DisasContext *s, int regno,
                                  int element, TCGMemOp size)
 {
-    int offs = 0;
+    int element_size = 1 << size;
+    int offs = element * element_size;
 #ifdef HOST_WORDS_BIGENDIAN
     /* This is complicated slightly because vfp.zregs[n].d[0] is
-     * still the low half and vfp.zregs[n].d[1] the high half
-     * of the 128 bit vector, even on big endian systems.
-     * Calculate the offset assuming a fully bigendian 128 bits,
-     * then XOR to account for the order of the two 64 bit halves.
+     * still the lowest and vfp.zregs[n].d[15] the highest of the
+     * 256 byte vector, even on big endian systems.
+     *
+     * Calculate the offset assuming fully little-endian,
+     * then XOR to account for the order of the 8-byte units.
+     *
+     * For 16 byte elements, the two 8 byte halves will not form a
+     * host int128 if the host is bigendian, since they're in the
+     * wrong order.  However the only 16 byte operation we have is
+     * a move, so we can ignore this for the moment.  More complicated
+     * operations will have to special case loading and storing from
+     * the zregs array.
      */
-    offs += (16 - ((element + 1) * (1 << size)));
-    offs ^= 8;
-#else
-    offs += element * (1 << size);
+    if (element_size < 8) {
+        offs ^= 8 - element_size;
+    }
 #endif
     offs += offsetof(CPUARMState, vfp.zregs[regno]);
     assert_fp_access_checked(s);