[Bug target/96918] Failure to optimize vector shift left+shift right+or to pshuf

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96918

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The __builtin_ia32_* builtins are implementation detail, we don't guarantee
them between releases, they can be added or removed, the only supported thing
are the _mm* intrinsics in the headers.  E.g. sometimes we replace a builtin
with just generic C code if it can do e.g. with generic vectors what the
builtin was doing before.
So, for testcases, unless we test for something that can't be reproduced with
the intrinsics and can be only reproduced with the builtins, we strongly prefer
using the supported intrinsics.