[patch] FastISel omitting unused parameters hurts debug info

[JF Bastien]

Would it be possible to configure this behavior? i.e. to have "fast but not
silly isel", and "fast and debuggable isel"? The later should probably be
the default, but some people (however silly they may be) want fast code
from fast isel. ;-)


On Fri, Jun 7, 2013 at 10:27 AM, David Blaikie <dblaikie at gmail.com> wrote:

> [+Eric, since I forgot him on the first mail]
>
> One other avenue that I haven't explored yet is that, perhaps,
> something like r182062 might be applicable, which simply retained more
> argument frame index information without generating more machine
> registers/instructions. I don't know at all if that applies to this
> issue, but it did address a similar issue of debug info missing on
> unused parameters (PR14575).
>
> On Thu, Jun 6, 2013 at 7:52 PM, David Blaikie <dblaikie at gmail.com> wrote:
> > Hi Dan,
> >
> > Here's the patch we talked about this afternoon - with the equivalent
> > change made to the ARM side of fast ISel to match the X86 change I'd
> > made. Turns out with the ARM change something does fail -
> > CodeGen/ARM/fast-isel-call.ll. So it seems this change isn't neutral
> > to codegen. (though I'm not sure how much we need to care deeply about
> > the performance of unused parameters?) I assume the source of the
> > change is the FIXME immediately proceeding the change to *FastISel.cpp
> > - so perhaps this is an acceptable "we'll fix the underlying issue &
> > this and some other things will get better".
> >
> > I was wondering if you had any ideas on how this should be addressed
> > (is this acceptable? in which case I should just update the test case
> > (though I don't fully understand it yet). I've attached the two .s
> > files, with my change (a.s) and without my change (b.s) (around
> > r183465) along with the patch itself.
> >
> > - David
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130608/76178017/attachment.html>