Gimple lowering question

[Andrew MacLeod]

I had a question asked of me, and now I'm passing the buck.
   extern void *memcpy(void *, const void *, unsigned int);
   extern int memcmp(const void *, const void *, unsigned int);
   typedef unsigned long bits32;
   typedef unsigned char byte;

   static const byte orig[10] = {
       'J', '2', 'O', 'Z', 'F', '5', '0', 'F', 'Y', 'L' };

   static byte test[10];

   int
   verify (void)
   {
      return 0 == memcmp (test, orig, 10 * sizeof (orig[0]));
   }

   int
   benchmark (void)
   {
      memcpy (test, orig, 10 * sizeof (orig[0]));
      return 0;
   }


Target is arm-none-eabi, and when compiled with -Os

After the gimple lowering, the verify routine remains the same, but the 
benchmark () routine  is transformed from a memcpy and becomes:


   ;; Function benchmark (benchmark, funcdef_no=1, decl_uid=4718,
   cgraph_uid=4, symbol_order=3)

   int benchmark ()
   {
      int D.4726;

      MEM <unsigned char[10]> [(char * {ref-all})&test] = MEM <unsigned
   char[10]> [(char * {ref-all})&orig];
      D.4726 = 0;
      goto <D.4727>;
      <D.4727>:
      return D.4726;
   }


It appears that forwprop is then transforming the statement to
  <bb 2> :
  MEM <unsigned char[10]> [(char * {ref-all})&test] = "J2OZF50FYL";
  return 0;

And in the final output, there are now 2 copies of the original 
character data:

orig:
        .ascii  "J2OZF50FYL"
        .space  2
.LC0:
        .ascii  "J2OZF50FYL"
        .bss


and I presume that new string is a copy of the orig text that forwprop 
has created for some reason.

Whats going on, and is there a way to disable this?  Either at the 
lowering stage or in forwprop?   At -Os, they are not thrilled that a 
bunch more redundant text is being generated in the object file.   This 
is a reduced testcase to demonstrate a much larger problem.

I don't see this happening on my x86 box.  the memcpy's are not lowered 
to MEMs there under any circumstances I can find.

This is true for at least gcc13 through trunk.  It was not true back in 
the heyday of gcc8.

 Andrew