[llvm] r319894 - [InstSimplify] Fold insertelement into undef if index is out of bounds

[Igor Laevsky via llvm-commits]

I reintroduced the changes in the r320568.

> On 13 Dec 2017, at 00:20, Marek Olšák <maraeo at gmail.com> wrote:
> 
> I have a patch that fixes this for Mesa.
> 
> Igor, you can commit your change again if it's been reverted.
> 
> Marek
> 
> On Tue, Dec 12, 2017 at 5:31 PM, Michel Dänzer <michel at daenzer.net> wrote:
>> 
>> Looks like this has become a case for Nicolai or Marek.
>> 
>> To give some context to Marek, the issue below can be reproduced with
>> the piglit test fcc-read-to-pbo-after-clear.
>> 
>> 
>> On 2017-12-12 04:25 PM, Sanjay Patel wrote:
>>> For reference, the extractelement with bad index should already have
>>> been simplified.
>>> https://reviews.llvm.org/rL319910 recently made that stronger, but
>>> raised some concern for AMDGPU backend tests.
>>> 
>>> On Tue, Dec 12, 2017 at 8:01 AM, Igor Laevsky via llvm-commits
>>> <llvm-commits at lists.llvm.org <mailto:llvm-commits at lists.llvm.org>> wrote:
>>> 
>>>    Hi Michael,
>>> 
>>>    In the source code I see the following operations:
>>>      %122 = extractelement <4 x i32> %121, i32 6
>>>    and:
>>>      %124 = insertelement <4 x i32> %121, i32 %123, i32 6
>>> 
>>>    This operations has undefined result
>>>    (see https://llvm.org/docs/LangRef.html#extractelement-instruction
>>>    <https://llvm.org/docs/LangRef.html#extractelement-instruction>).
>>>    Looks like a bug in the frontend. Do you know what’s the best
>>>    approach here?
>>> 
>>> 
>>>>    On 12 Dec 2017, at 13:45, Michel Dänzer <michel at daenzer.net
>>>>    <mailto:michel at daenzer.net>> wrote:
>>>> 
>>>>    On 2017-12-12 11:08 AM, Igor Laevsky wrote:
>>>>>    I reverted problematic changes in the
>>>>>    https://reviews.llvm.org/rL320466 <https://reviews.llvm.org/rL320466>
>>>> 
>>>>    Thanks.
>>>> 
>>>> 
>>>>>    Is it possible to obtain an llvm-ir of the failing test case
>>>>>    before the
>>>>>    optimizations? I can see the miscompile, but I can’t reproduce it
>>>>>    using
>>>>>    optimised good case.
>>>> 
>>>>    The attached should be the LLVM IR before optimizations.
>>>> 
>>>> 
>>>>    --
>>>>    Earthling Michel Dänzer               |
>>>>                  http://www.amd.com <http://www.amd.com>
>>>>    Libre software enthusiast             |             Mesa and X
>>>>    developer
>>>>    <fcc-read-to-pbo-after-clear.ll>
>>> 
>>> 
>>>    _______________________________________________
>>>    llvm-commits mailing list
>>>    llvm-commits at lists.llvm.org <mailto:llvm-commits at lists.llvm.org>
>>>    http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>>    <http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits>
>>> 
>>> 
>> 
>> 
>> --
>> Earthling Michel Dänzer               |               http://www.amd.com
>> Libre software enthusiast             |             Mesa and X developer