On 2/18/26 13:07, Théo Papadopoulo wrote:
On 2/18/26 11:33 AM, Victor Do Nascimento wrote:
- Finally, a simplification is made. Before, we had the check:
if ((vf.is_constant () && pow2p_hwi (new_alignment.to_constant ()))
|| (!vf.is_constant () && pow2p_hwi (align_factor_c)))
Given that we have `new_alignment = vf * align_factor_c', I do not
believe anything can be derived from `new_alignment' that is now known
already in `align_factor_c '. That is, a power of 2 has only one
prime factor: 2. If `align_factor_c' is not itself a power of 2,
the multiplication by `vf' cannot correct that, such that the check
for `new_alignment' seems redundant. The predicate can thus
simplified to `if (pow2p_hwi (align_factor_c))'. The seems sound to
me, should my logic be correct.
This simplification is certainly wrong if vf is constant and not a power
of 2 and if align_factor_c is a power of 2.... if that can happen.
Theo.
Thanks Theo. Yes, you're right.
Given the patch is to fix non power-of-2 `align_factor_c' cases, I
reasoned only about cases with that as a starting point, neglecting to
think about other hypothetical cases where the `align_factor_c' value
is fine to begin with and what "messes up" the power-of-2 criterion is
the `vf' value.
Thanks for bringing that to my attention...
Onto v3 we go.
Many thanks,
Victor
