[llvm-commits] [llvm] r120230 - in /llvm/trunk/tools/llvmc/src: Base.td.in Hooks.cpp

[Mikhail Glushenkov]

Hi Eric,

On Sun, Nov 28, 2010 at 09:41:47PM -0800, Eric Christopher wrote:
>
> [...]
>
> I'm not particularly happy with this direction at the moment. Ideally I
> think that if we want to support the full set of gcc command line options
> (in particular the abomination that is the arm version) we'll need to
> explicitly factor the backends to support them instead of trying to translate
> each march option (which is, tbh, what we really support at the moment)
> into a set of mattr flags. We would then need, at least a:
>
> a) target processor compiled for
> b) target cpu family compiled for (arm only afaik)
> c) target processor tuned for
>
> in order to support -march, -mcpu, and -mtune (in that order).

How difficult of an undertaking would that be for someone who never worked on a
LLVM backend? I'm not very fond of the current solution, but backend hacking is
an uncharted territory for me.

> In particular the implementation that you have here won't work for any
> other processor that I know of at the moment.

Yes, it is incomplete.

--
()  ascii ribbon campaign - against html e-mail
/\  www.asciiribbon.org   - against proprietary attachments