Currently, we apply the aligned attribute to all integers, regardless of their actual aligement.

This leads to such gems as this:

__int8_t  __attribute__((aligned(1)))

Since a byte-sized type has a size of 1(by definition), it's alignment can be 1 and only 1.

This attribute is thus unneeded, and only leads to increased memory consumption, and more confusing debug strings. It could also inhibit some optimizations(no examples of that yet, but it is not hard to imagine some of them can't deal with under / over aligned types, and ignore types with explicit alignment altogether).

While removing this attribute for byte-sized types is an easy fix, other types pose a bigger challenge.

Ideally, we'd need an API to query GCC about the aligement of intigers / floats / vectors, and then use that API to not emit this attribute when not necessary(eg. a structure composed of floats with aligement 4 is alteady correctly aligned by the virtue of containing floats).

For example, Builder::load could check if the requested align: Align differs from the one already guaranteed for a type(by checking the aligement on the rustc side of things). We could then call get_aligned if and only if this aligement differs, reducing the usage of this attribute(thus reducing memory usage).