[Bug target/87317] Missed optimisation: merging VMOVQ with operations that only use the low 8 bytes

[glisse at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87317

Marc Glisse <glisse at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Target|                            |x86_64-*-*

--- Comment #1 from Marc Glisse <glisse at gcc dot gnu.org> ---
What I see:

- we could implement _mm_cvtsi128_si32 using gcc extensions instead of a
builtin

- I think we have code to simplify (vec_select (vec_concat ...) ...) when we
select everything from the same half, but we do not handle an intermediate
subreg: (vec_select:V8QI (subreg:V16QI (vec_concat:V2DI ...

- sse4_1_zero_extendv8qiv8hi2 is described as taking a v16qi as input, not a
v8qi