> "Robin Dapp" <rdapp....@gmail.com> writes:
>> But here I intended to just change the encoded value in memory and to
>> prevent it from aliasing with other encoded values.
>> In an actual instruction the values we "padded" (or that are beyond the
>> vector length) are ignored. For the purpose of hashing a mask of
>> "1010" and "10101010" should be distinct which they are not right now
>> because we don't stop at the length so to say.
>
> But what modes do "1010" and "10101010" represent here? And which code
> treats them as equal? (It should be ok if the hashes are equal, as long
> as the equality function distinguishes them.)
I realize I didn't fully detail that in the original post, sorry :/
In my test case we have two functions one with the first mask ("1010" for
V4BI) and another one performing a similar operation just for
V8BI ("10101010").
In optimize_constant_pool we use native_encode_rtx to encode the value
into a buffer (of size GET_MODE_SIZE which is one byte here).
IIRC the whole function works at byte granularity.
htab->find_slot_with_hash
then finds the same hash for the V4BI and V8BI masks, and we wrongly unify it.
--
Regards
Robin
