This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: TLS support for ARM
On Sat, Oct 08, 2005 at 07:48:45PM +0100, Richard Earnshaw wrote:
> I really think making the compiler do the right thing as automatically
> as possible is keeping things simple (for the user at any rate). One of
I spent a lot of time working with the MIPS options. They did a lot of
this kind of guessing, and it was an absolute nightmare, because it's so
hard to predict what a combination will do.
> the nasty things about using the compiler is that so many of the options
> cannot be used together without generating an error. This can make
> Makefile maintenance a nightmare. There's no real reason to error out
> here because we know what to do when we can't honour the users request
> and at worst that just produces slightly slower code. So I think the
> logic should be:
Errors don't bother me: they're nice and clear. You asked for
something, we couldn't do it. Maybe the user is working in a custom
system where the use of __aeabi_read_tp is inappropriate for some
reason. They asked the compiler not to do it, but we did anyway.
>
> tls_model = soft;
> if (hard_tls_requested)
> if (CPU and instruction set support hard tls)
> tls_model = hard;
>
> This is easy to understand, easy to document and not much harder to
> implement.
I think it's a bad idea. I'll implement it if you want me to, of
course, but I'd at least like a warning in the
hard_tls_requested-but-not-using-it-anyway case.
--
Daniel Jacobowitz
CodeSourcery, LLC