[Bug target/91796] Sub-optimal YMM register allocation.

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91796

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hjl.tools at gmail dot com,
                   |                            |jakub at gcc dot gnu.org,
                   |                            |uros at gcc dot gnu.org,
                   |                            |vmakarov at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The #c0 issue is a RA decision, what registers it decides to use for each
pseudo register.
As for #c1, the optimizers see that what you've typed there is a constant and
optimizes it into a constant (which is desirable).  Whether that particular
constant is better loaded into a register using a memory load or two
instructions depends on various factors, we obviously have special cases for
constants that can be built using a single non-memory instruction (e.g. zero
vector or all ones vector), whether to do that for other constant is something
that needs to be thought out, benchmarked etc.