[PATCH] [PowerPC, Darwin] fix for PR15821

[Bill Schmidt]

On Thu, 2013-05-02 at 19:36 -0400, David Fang wrote:
> Hi Bill,
> 
> > On Wed, 2013-05-01 at 16:25 -0500, Bill Schmidt wrote:
> >> David, the patch looks generally good.  I'll remove the FIXME since it's
> >> now fixed.  Rather than cloning anon-aggr.ll, I'll just add the new
> >> variants to that test case as well.  I probably won't get around to
> >> committing it until tomorrow or the next day.
> >
> > Hi David,
> >
> > With the test case as written, I don't see the same code generation that
> > you do.  I think we need to have a -mcpu=<machine> parameter in the RUN
> > command to make sure we get universal code gen no matter which host
> > machine is building the test.  What cpu are you using to get the code
> > gen being tested for?
> 
> Just checked my .s files.
> 
> For powerpc-darwin8, I have:
> .machine ppc7400
> 
> For powerpc64-darwin8, I have:
> .machine ppc64
> 

Hm.  Using -mcpu=g4 (PowerPC 7400) gives me the code generation expected
by your DARWIN32 checks, so that's progress.  However, I can't get the
proper 64-bit code gen yet.  Using either -mcpu=ppc64 or -mcpu=g5 gives
me the attached assembly results, using current trunk to build llc.  Any
ideas why I see the differences?

Thanks,
Bill

> Fang
> 
> > (If you're unsure, run llc from the command line and look for
> > the .machine directive in the generated assembly file.)
> >
> > Thanks,
> > Bill
> >
> >>
> >> Thanks,
> >> Bill
> >>
> >> On Wed, 2013-05-01 at 17:08 -0400, David Fang wrote:
> >>> Hi,
> >>>
> >>> Attached is a patch for PR 15821, which fixes PowerPC CodeGen on Darwin.
> >>> It re-applies the same patch from PR 14779, and adds a test for darwin8,
> >>> targeting ppc32 and ppc64.
> >>>
> >>> http://llvm.org/bugs/show_bug.cgi?id=15821
> >>> https://github.com/fangism/llvm/commit/d27de53554abf4443ad386ea451991e4ef32f9ae
> >>>
> >>> Some PPC maintainers have already commented in the thread.
> >>> Please review and commit on my behalf.
> >>> Thanks.
> >>>
> >>> Fang
> >>>
> >>> _______________________________________________
> >>> llvm-commits mailing list
> >>> llvm-commits at cs.uiuc.edu
> >>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> >
> 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: anon_aggr.s
Type: text/x-tex
Size: 2895 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130502/7404e2ec/attachment.tex>