9039b0cd25
This allows us to use it in QTypedArrayData<T>::AlignmentDummy, which is
used in __builtin_assume_aligned() as a hint to the compiler. This
increases the value we were passing as the compiler hint from 4 to 8 on
32-bit platforms and from 8 to 16 on 64-bit platforms. We actually do
align to a bit higher than even that, but that's not an ABI guarantee
we're making.
However, it looks like GCC on 32-bit platforms is buggy, so we work
around it. Commit r240248 ("Make max_align_t respect _Float128.",
63012d9a57edc950c5f30242d1e19318b5708060 in Git) increased
std::max_align_t to 16 bytes on all platforms, saying "Such an increase
is of course an ABI change" and "I think glibc malloc alignment should
also increase to 16-byte 32-bit x86", but such a change to glibc was
never implemented. Moreover, there are systems that don't use glibc,
such as when using GCC on other OSes.
This is not a change in Qt behavior anywhere. i386 does not (usually) do
alignment checks, so most code wouldn't notice this at all.
Change-Id: I79e700614d034281bf55fffd178fa0aee2344307
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
(cherry picked from commit 4d57b30194f31865132312c67abf796cb58a0f8f)
Reviewed-by: Qt Cherry-pick Bot <cherrypick_bot@qt-project.org>