[Bug middle-end/104151] [9/10/11/12 Regression] x86: excessive code generated for 128-bit byteswap

[ubizjak at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104151

--- Comment #12 from Uroš Bizjak <ubizjak at gmail dot com> ---
(In reply to Uroš Bizjak from comment #10)
> (In reply to Hongtao.liu from comment #4)
> > Also there's separate issue, codegen for below is not optimal
> > gimple:
> > _11 = VIEW_CONVERT_EXPR<vector(2) long unsigned int>(a_3(D))
> > asm:
> >         mov     QWORD PTR [rsp-24], rdi
> >         mov     QWORD PTR [rsp-16], rsi
> >         movdqa  xmm0, XMMWORD PTR [rsp-24]
> > 
> > 
> > I think this issue has been recorded in several existed PRs.
> 
> Maybe this can be solved with secondary_reload when GPR and XMM regs are
> involved.

PR104306 is an experiment.