https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121700
--- Comment #1 from Avinash Jayakar <avinashd at gcc dot gnu.org> ---
After some investigation it seems like in powerpc, the min_profitable_estimate
value is 6 for this loop. Meaning vectorization is beneficial if the number of
iterations is greater than or equal to 6.
In the O2 optimization flag, the value of -fvect-cost-model=very-cheap, thus
the conservative analysis at tree-vect-loop.cc
1971 if (loop_cost_model (loop) == VECT_COST_MODEL_VERY_CHEAP
1972 && min_profitable_estimate > (int) vect_vf_for_cost (loop_vinfo))
1973 {
does not cause the loop to be vectorized.
However, using the dynamic model as done in O3 flags, gets the loop to be
vectorized.
But the issue persists for FLOOR_MOD_EXPR, which reports the value of
min_profitable_iters to be -1, and no other cost model will cause the loop to
be vectorized. I will look into this further.
