[Bug tree-optimization/32663] [4.3 regression]: revision 126369 went into an infinite loop

[hjl at lucon dot org]

------- Comment #5 from hjl at lucon dot org  2007-07-07 16:54 -------
Before the change, PRE generates:

  # prephitmp.165_1394 = PHI <storetmp.156_472(127), storetmp.156_452(130)>
  # prephitmp.165_1395 = PHI <storetmp.156_1216(127), storetmp.156_466(130)>
  # nlast_26 = PHI <0(127), nlast_442(130)>
  # lasp_21 = PHI <1(127), lasp_454(130)>
  D.1457_436 = prephitmp.165_1395;
  ii.62_440 = ii;
  D.1361_441 = ii.62_440 + -1;
  nlast_442 = (*iwhi_303(D))[D.1361_441];
  D.1459_443 = nlast_442 + -1;
  D.1460_444 = D.1459_443 * nlast_442;
  D.1461_445 = D.1460_444 / 2;
  D.1462_446 = D.1457_436 + D.1461_445;
  storetmp.156_466 = D.1462_446;
  *nswe_430(D) = D.1462_446;
  D.1463_447 = prephitmp.165_1394;
  D.1464_448 = D.1463_447 + 1;
  storetmp.156_452 = D.1464_448;
  *ncat_431(D) = D.1464_448;
  D.1466_451 = nlast_26 + lasp_21;
  (*inat_192(D))[D.1463_447] = D.1466_451;
  D.1465_453 = prephitmp.165_1394;

After change, PRE generates:

  # prephitmp.165_1395 = PHI <storetmp.156_1216(127), storetmp.156_466(130)>
  # nlast_26 = PHI <0(127), nlast_442(130)>
  # lasp_21 = PHI <1(127), lasp_454(130)>
  D.1457_436 = prephitmp.165_1395;
  ii.62_440 = ii;
  D.1361_441 = ii.62_440 + -1;
  nlast_442 = (*iwhi_303(D))[D.1361_441];
  D.1459_443 = nlast_442 + -1;
  D.1460_444 = D.1459_443 * nlast_442;
  D.1461_445 = D.1460_444 / 2;
  D.1462_446 = D.1457_436 + D.1461_445;
  storetmp.156_466 = D.1462_446;
  *nswe_430(D) = D.1462_446;
  D.1463_447 = D.1465_453;
  D.1464_448 = D.1463_447 + 1;
  *ncat_431(D) = D.1464_448;
  D.1466_451 = nlast_26 + lasp_21;
  (*inat_192(D))[D.1463_447] = D.1466_451;

D.1465_453 is never initialized.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32663