[Bug target/87561] [9 Regression] 416.gamess is slower by ~10% starting from r264866 with -Ofast

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87561

--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> ---
433.milc         9180        336       27.4 *    9180        349       26.3 S
433.milc         9180        335       27.4 S    9180        340       27.0 *
433.milc         9180        344       26.7 S    9180        334       27.5 S
450.soplex       8340        225       37.1 *    8340        223       37.5 S
450.soplex       8340        226       36.9 S    8340        228       36.5 S
450.soplex       8340        223       37.4 S    8340        223       37.3 *
482.sphinx3     19490        386       50.5 *   19490        392       49.8 S
482.sphinx3     19490        384       50.7 S   19490        374       52.1 *
482.sphinx3     19490        394       49.5 S   19490        368       53.0 S

comparing the fastest runtimes makes this a progression for both 433.milc
and 482.sphinx3 and no difference for 450.soplex.

I'll post the patch.

For GCC 10 we'd want to play with applying the cost model to the whole
loop nest instead.