https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62283
--- Comment #1 from Henrik Holst <holst at matmech dot com> ---
(forgot to indent the end statement above.)
The expected assembler code should be something like:
movaps %xmm0, %xmm1
movups (%rdi), %xmm0
shufps $0, %xmm1, %xmm1
movups (%rsi), %xmm2
mulps %xmm1, %xmm0
addps %xmm2, %xmm0
movups %xmm0, (%rsi)
ret
