The original pattern was not exercised by any pre-existing test. I tried but
failed to come up with a testcase that would expand to
float_extend ∘ vec_duplicate
rather than
vec_duplicate ∘ float_extend.
Ok, so we indeed don't have a test and the intrinsics tests unfortunately are
no help either. It's high time we move these propagations to gimple...
What's not tested doesn't exist so I won't insist on keeping the previous
order. Checking other patterns, we have instances of both variants.
At least for our port we should decide for a canonical order and I'd agree that
the longer we stay in the scalar domain the better. This means that in time
the remaining 8(?) instances should be migrated as well.
--
Regards
Robin
