[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAOUYtQDwJ8Jc2Cu+_1N78k5QFMtYzB_vrWiLJrHatNhVb=pn7A@mail.gmail.com>
Date: Fri, 7 May 2021 16:59:52 +0100
From: Jon Chesterfield <jonathanchesterfield@...il.com>
To: musl@...ts.openwall.com
Subject: Re: Compiling to single llvm bc, working example?
Hi Rich,
It looks like no-builtin is now an attribute https://godbolt.org/z/GsfffavKM
that gets tagged on ~ everything when compiling with ffreestanding.
Introduced in https://reviews.llvm.org/D68028 with a comment saying 'not
yet fully implemented...no effect on generated code' which is still there
at head.
Thanks for the example from picolibc! Eerily similar to the fmod loop I ran
into before, except there was no -fnobuiltin-fmod at that point.
Putting D68028 and the picolibc example together, that part may now be
resolved (despite the documentation).
Cheers!
Jon
On Fri, May 7, 2021 at 4:54 PM Érico Nogueira <ericonr@...root.org> wrote:
> Em 07/05/2021 11:41, Jon Chesterfield escreveu:
> > Hello musl,
> >
> > Clang has mixed support for compiling libc from C. A couple of years ago
> > I ran into pattern matching turning fmod into a call to itself and
> > similar. I can see malloc getting into fights with alias analysis as
> > well. I'm not confident ffreestanding is propagated in the IR.
>
> For a tangible example:
>
>
> https://github.com/picolibc/picolibc/commit/4261cfecabee211af8d0dee1764bf86ed9c8e258
>
> >
> > Can anyone report success from compiling (a subset) of musl to llvm IR,
> > perhaps with the hacks necessary at the time to make the end application
> > work?
> >
> > Motivation is an architecture (amdgcn) that can't yet link machine code,
> > but someone may have done this for performance.
> >
> > Thanks,
> >
> > Jon
> >
> >
>
>
Content of type "text/html" skipped
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.
