https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118681
--- Comment #4 from Diego Ramirez <dieram3 at gmail dot com> --- I think the way to fix this is finding the the first pool block size `B` such as `B >= size` and `B & (alignment - 1) == 0` (or in other words `gcd(B, alignment) == alignment)`). The starting point in `pool_sizes` would be found with `lower_bound(pool_sizes, max(size, alignment))`, and then we'd do a linear search. I see that in `pool_sizes`, powers of 2 are at most 4 elements apart (256 to 512), so the linear search would be 4 iterations in the worse case. Another option would be to change the pool block sizes, but that seems like a bigger endeavor.
