[llvm-commits] [PATCH, PowerPC] Fix MC instruction encodings

[Hal Finkel]

----- Original Message -----
> From: "Ulrich Weigand" <Ulrich.Weigand at de.ibm.com>
> To: "Chandler Carruth" <chandlerc at google.com>
> Cc: "Hal Finkel" <hfinkel at anl.gov>, "Commit Messages and Patches for LLVM" <llvm-commits at cs.uiuc.edu>
> Sent: Tuesday, November 13, 2012 2:25:14 PM
> Subject: Re: [llvm-commits] [PATCH, PowerPC] Fix MC instruction encodings
> 
> Chandler Carruth <chandlerc at google.com> wrote on 13.11.2012 21:08:11:
> > On Tue, Nov 13, 2012 at 11:23 AM, Ulrich Weigand
> > <Ulrich.Weigand at de.ibm.com> wrote:
> > > Hal Finkel <hfinkel at anl.gov> wrote on 13.11.2012 20:01:08:
> > >
> > >> LGTM. We should add test cases (but I think that we can't do
> > >> that
> > >> until we have the asm parser working, and that's blocked by
> > >> TableGen
> > >> problems), so those can wait.
> > >
> > > Checked in as revisions 167860..167863.
> > >
> > > Agree on the test cases ...
> >
> > FWIW, I don't think this is a good strategy. It is *essential* that
> > we
> > prioritize all of the efforts necessary to fix the underlying
> > infrastructure first, so that we can build up a proper regression
> > test
> > suite each time these types of encoding fixes go into the tree.
> >
> > I would ask that folks who care about PPC focus their efforts on
> > resolving these underlying issues first, and then come back to
> > encoding bugs.
> 
> I agree that we urgently need to get the asm parser working, and
> then create a full set of test cases for instruction encodings.
> We're working on it ...
> 
> However, right now, there is *no* test whatsoever for any encoding,
> so IMHO it doesn't really make sense to block patches that fix
> obviously broken encodings until the infrastructure is in place.
> 
> As to the testing I did:  I built all of test-suite twice, once
> using clang set to use the integrated assembler, and once using
> clang set to use the system assembler, and then compared all the
> resulting object files.
> 
> With the set of patches just applied, I'm getting 100% identical
> .text sections between integrated and system assembler, for all
> those object files.
> 
> While this is obviously not automated, I'm planning to regularly
> re-run such tests until we have the asm parser and and encoding
> test suite in place, to ensure we don't regress.
> 

I'd like to add that once we have the asm parser in place, I plan to automatically generate encoding tests from the test-suite binaries. That process will generate tests for these encodings.

 -Hal

> Bye,
> Ulrich
> 
> 

-- 
Hal Finkel
Postdoctoral Appointee
Leadership Computing Facility
Argonne National Laboratory