[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20260311162408.GF1827@brightrain.aerifal.cx>
Date: Wed, 11 Mar 2026 12:24:08 -0400
From: Rich Felker <dalias@...c.org>
To: James Y Knight <jyknight@...gle.com>
Cc: musl@...ts.openwall.com, Xing Li <lixing@...ngson.cn>,
wanghongliang@...ngson.cn
Subject: Re: [PATCH] loongarch64: Add lsx and lasx regset definition
On Wed, Mar 11, 2026 at 10:49:15AM -0400, James Y Knight wrote:
> On Wed, Mar 11, 2026 at 12:07 AM Rich Felker <dalias@...c.org> wrote:
> >
> > On Tue, Sep 10, 2024 at 09:17:57AM +0800, Xing Li wrote:
> > > Fix the upstream binutils-gdb building error caused by
> > > the lack of lsx and lasx regset definition.
> > > ---
> > > arch/loongarch64/bits/user.h | 12 ++++++++++++
> > > 1 file changed, 12 insertions(+)
> > >
> > > diff --git a/arch/loongarch64/bits/user.h b/arch/loongarch64/bits/user.h
> > > index fd9b7b22..10e1be45 100644
> > > --- a/arch/loongarch64/bits/user.h
> > > +++ b/arch/loongarch64/bits/user.h
> > > @@ -22,3 +22,15 @@ typedef union {
> > > float f;
> > > } elf_fpreg_t;
> > > typedef elf_fpreg_t elf_fpregset_t[ELF_NFPREG];
> > > +
> > > +typedef union
> > > +{
> > > + double d[2];
> > > + float f[4];
> > > +} elf_lsxregset_t[32] __attribute__((__aligned__(16)));
> > > +
> > > +typedef union
> > > +{
> > > + double d[4];
> > > + float f[8];
> > > +} elf_lasxregset_t[32] __attribute__((__aligned__(32)));
> > > --
> > > 2.27.0
> >
> > I don't like having the alignment directive bound to the whole type
> > like this where it's depending on weird GNUC-isms that can't be
> > represented in standard C alignas, but I think it's ok for now and we
> > can find a better way to express it later if needed. I know this is
> > another patch that distros have been carrying for a long time so I'd
> > like to merge it for release.
>
>
> That code looks weird. Why does it apply the alignment attribute on
> the _array_ type, rather than the union type? E.g. why is this not
> just:
>
> typedef union
> {
> double d[2];
> float f[4];
> } __attribute__((__aligned__(16))) elf_lsxregset_t[32];
I'm not sure how GNU C __attribute__((__aligned__(...))) even binds,
which is part of why I don't like it and prefer the clearer ISO C
form. But..
> And once you have that, it's easy to take it one step further and move
> the alignment onto the union field, at which point you can have
> standard C code:
>
> typedef union
> {
> _Alignas(16) double d[2];
> _Alignas(16) float f[4];
> } elf_lsxregset_t[32];
Does that work? Or would that be trying to apply the alignment to
double where the alignment is larger than the type and thereby
invalid? This is totally on me, but I don't understand well how the
binding works.
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
