https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54904
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|pinskia at gcc dot gnu.org |unassigned at gcc dot
gnu.org
Target| |x86_64-linux-gnu
Status|ASSIGNED |NEW
Keywords| |missed-optimization
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I think I accidently set this to assigned rather than just new.
x86_64 produces now:
movdqa .LC0(%rip), %xmm0
movabsq $1229782938247303441, %rax
movups %xmm0, (%rdi)
movq %rax, 15(%rdi)
Which is slightly better.
x86_64 with -mavx2 produces:
movl $17, %eax
vmovd %eax, %xmm0
movabsq $1229782938247303441, %rax
vpbroadcastb %xmm0, %xmm0
vmovdqu %xmm0, (%rdi)
movq %rax, 15(%rdi)
which is only slightly better but we need to produce 0x1111111111111111 in rax,
then why not use that for the broadcast ?
aarch64 is reasonible though:
movi v31.16b, 0x11
str q31, [x0]
str d31, [x0, 15]
