[Bug target/78762] Regression: Splitting unaligned AVX loads also when AVX2 is enabled

[venkataramanan.kumar at amd dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78762

Venkataramanan <venkataramanan.kumar at amd dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |venkataramanan.kumar at amd dot co
                   |                            |m

--- Comment #8 from Venkataramanan <venkataramanan.kumar at amd dot com> ---
(In reply to Jakub Jelinek from comment #7)
> Actually, it seems the unaligned load splitting is there because of
> sandybridge?
> DEF_TUNE (X86_TUNE_AVX256_UNALIGNED_LOAD_OPTIMAL,
> "256_unaligned_load_optimal",
>           ~(m_NEHALEM | m_SANDYBRIDGE | m_GENERIC))
> Though the Nehalem in there is weird, there weren't any Nehalem CPUs with
> AVX support, right?  And if it is just something for Sandybridge which
> doesn't support AVX2, then either for AVX2 we should use a different tuning
> parameter (which might be set for all CPUs?), or do the ISA specific generic
> tuning.

My understanding is that avx256 unaligned loads are good for all the processors
other than NEHALEM, SANDYBRIDGE and GENERIC. In other words splitting is set
for only these targets. 

Before I can comment on whether AMD targets needs this tuning (or) what is the
behavior we want to see for generic, allow me to benchmark and get back.