[Bug tree-optimization/33404] Predictive commoning + ivopts possibly introducing extra sign extensions.

["pinskia at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #1 from pinskia at gcc dot gnu dot org  2008-07-23 01:55 -------
Hmm, I think the sign/zero extension is needed.
Even doing:
void fred(unsigned short in, unsigned short *out1)
{
    int i;
    for (i=0;i<100;i++)
        out1[i+1] = out1[i]*in;
}
--- CUT ---
We will get a zero extension in the assembly (on Powerpc-linux-gnu).
.L2:
        add 9,4,11
        addi 11,11,2
        mullw 0,3,0
        sth 0,2(9)
        rldicl 0,0,0,48
        bdnz .L2
--- CUT ---
But it is needed.  If we look at -O2 case:
.L2:
        lhz 0,0(4)
        mullw 0,3,0
        sthu 0,2(4)
        bdnz .L2
You will see that we load the code as zero extended.

Also I get:
<bb 3>:
  D.1263 = (short unsigned int) D__lsm0.41 * pretmp.27;
  D__lsm0.41 = (short int) D.1263;
  MEM[base: out1 + ivtmp.49, offset: 2] = D__lsm0.41;
  ivtmp.49 = ivtmp.49 + 2;
  if (ivtmp.49 != 200)

for the loop on the trunk.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33404