RFA: Fix rtl-optimization/57439

[Joern Rennecke]

bootstrapped / regtested on i686-pc-linux-gnu
build / regtested for i686-pc-linux-gnu X sh-elf
Also, Andreas Schwab reported in the PR that he has successfully  
tested this patch on m68k.
2013-05-29  Joern Rennecke <joern.renne...@embecosm.com>

        PR rtl-optimization/57439
        * postreload.c (move2add_valid_value_p): Check that we have
        a zero subreg_regno_offset when accessing the register in
        the requested mode.

Index: postreload.c
===================================================================
--- postreload.c        (revision 199387)
+++ postreload.c        (working copy)
@@ -1726,10 +1726,25 @@ move2add_record_sym_value (rtx reg, rtx
 static bool
 move2add_valid_value_p (int regno, enum machine_mode mode)
 {
-  if (reg_set_luid[regno] <= move2add_last_label_luid
-      || !MODES_OK_FOR_MOVE2ADD (mode, reg_mode[regno]))
+  if (reg_set_luid[regno] <= move2add_last_label_luid)
     return false;
 
+  if (mode != reg_mode[regno])
+    {
+      if (!MODES_OK_FOR_MOVE2ADD (mode, reg_mode[regno]))
+       return false;
+      /* For big endian, the starting regno for the subreg might be
+        different.  */
+      int s_off = subreg_lowpart_offset (mode, reg_mode[regno]);
+      s_off = subreg_regno_offset (regno, reg_mode[regno], s_off, mode);
+      if (s_off != 0)
+       /* We could in principle adjust regno, check reg_mode[regno] to be
+          BLKmode, and return s_off to the caller (vs. -1 for failure),
+          but we currently have no callers that could make use of this
+          information.  */
+       return false;
+    }
+
   for (int i = hard_regno_nregs[regno][mode] - 1; i > 0; i--)
     if (reg_mode[regno + i] != BLKmode)
       return false;