[Bug target/29900] [4.3 regression] stackreg to go wrongs after fwprop

[hjl at lucon dot org]

------- Comment #7 from hjl at lucon dot org  2006-11-20 14:26 -------
A smaller testcase:

bash-3.1$ cat foo.f90
   SUBROUTINE foo(IU, JU, IIL, JJL, XI, YJ, ALBTAB, XALB)
   INTEGER, INTENT(IN) :: IU, JU, IIL, JJL
   REAL, INTENT(IN) ::  XI, YJ
   REAL, DIMENSION(4, 5), INTENT(IN) :: ALBTAB
   REAL, INTENT(OUT) :: XALB

   XALB=ALBTAB(IU,JU)*(XI-IIL)*(YJ-JJL)+ALBTAB(IIL,JJL)*(IU-XI)*(JU-YJ)
   END SUBROUTINE foo

With -O2, I got

(insn:TI 35 67 88 2 (set (reg:SF 11 st(3))
        (minus:SF (reg:SF 8 st)
            (reg:SF 11 st(3)))) 606 {*fop_sf_1_i387} (nil)
    (expr_list:REG_DEAD (reg:SF 8 st)
        (nil)))

SF 11 is never set before.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29900