arm/msac: Enable NEON optimizations on more platforms

changed milestone to %1.4.1

added ARM performance labels

requested review from @mstorsjo

mentioned in merge request !1613 (merged)

I do not know if this makes a difference. This combination of conditions was added in 2019 with fc3777b4. It could be that things have changed since then and __ARM_NEON is sufficient.

Things haven't changed as such, but on one hand I think the conditions added in fc3777b4 can have been a bit excessively verbose, just to make things clear.

But on second thought; with MSVC targeting ARM, it won't define __ARM_NEON (because it uses its own defines and I'm not sure if it has any define specific for "NEON is allowed unconditionally everywhere"), so this does indeed opt into using this assembly in that combination. The __APPLE__ case is probably redundant, but it doesn't hurt either.

Note that with !1609 (merged) (which looks roughly quite good to me, it won't probably see big changes), the conditions in the runtime detection file probably will look a bit different.

Thus, I think this change probably is fine. (In one sense, it might be nice to have all the _WIN32 and __APPLE__ cases more logically grouped with ARCH_ARM, but that makes a more complicated condition, so this is probably fine.)

Can you amend the commit message, saying that this enables the assembly for MSVC targeting ARM, while the change for __APPLE__ should be a no-op and just is added for consistency?

approved this merge request

mentioned in merge request !1611 (merged)

added 1 commit

• 8c209190 - arm/msac: Enable NEON optimizations on more platforms

Compare with previous version

merged