[Bug tree-optimization/25621] Missed optimization when unrolling the loop (splitting up the sum) (only with -ffast-math)

[jv244 at cam dot ac dot uk]

------- Comment #4 from jv244 at cam dot ac dot uk  2007-07-03 19:30 -------
Now, I get the same timings for the hand-optimised loop and compiled loop if I
use the option:

gfortran -O3 -ffast-math -ftree-vectorize -march=native  -funroll-loops
-fvariable-expansion-in-unroller test.f90

whereas -funroll-loops is quite common to add, -fvariable-expansion-in-unroller
is not. Could one have a heuristic that switches that on by default if
-funroll-loops (and -ffast-math) ? For S31 the timings are:

> gfortran -O3 -ffast-math -ftree-vectorize -march=native  -funroll-loops test.f90
> time ./a.out
real    0m6.618s

> gfortran -O3 -ffast-math -ftree-vectorize -march=native  -funroll-loops -fvariable-expansion-in-unroller test.f90
> time ./a.out
real    0m4.457s

so a 50% improvement. 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25621