[llvm-commits] [PATCH] fix PR12359

[Bill Wendling]

Thanks for the changes. The patch LGTM.

-bw

On Aug 31, 2012, at 11:40 AM, Michael Liao <michael.liao at intel.com> wrote:

> On Fri, 2012-08-31 at 10:51 -0700, Bill Wendling wrote:
>> On Aug 31, 2012, at 10:48 AM, Michael Liao <michael.liao at intel.com> wrote:
>> 
>>> On Fri, 2012-08-31 at 10:33 -0700, Bill Wendling wrote:
>>>> Oh! And please flesh out the test more so that it checks for exactly what you expect with no intermediate instructions.
>>> 
>>> Sorry, may I misunderstand your question? PSHUB has no immediate
>>> variant.
>>> 
>> I meant that the list of instructions you're expecting are the ones you are expecting with no other ones in there. So you want to check that the only shuffle instruction is:
>> 
> 
> I see your point. Please let the revised patch again.
> 
> - michael
> 
>>  pshufd ...
>> 
>> right? :)
>> 
>> -bw
>> 
>>> BTW, here is the revised patch following your suggestions. Thanks!
>>> 
>>> Yours
>>> - Michael
>>> 
>>>> 
>>>> -bw
>>>> 
>>>> On Aug 31, 2012, at 10:31 AM, Bill Wendling <wendling at apple.com> wrote:
>>>> 
>>>>> You will probably need to specify '-march=x86' in the testcase. Also, since the bools are used in only one place, please inline them:
>>>>> 
>>>>>  if (V2.getOpcode() == ISD::UNDEF ||
>>>>>      ISD::isBuildVectorAllZeros(V2.getNode()))
>>>>>    return V1;
>>>>> 
>>>>> And write a comment on why the "all zeros" is valid.
>>>>> 
>>>>> Thanks!
>>>>> -bw
>>>>> 
>>>>> On Aug 31, 2012, at 10:01 AM, Michael Liao <michael.liao at intel.com> wrote:
>>>>> 
>>>>>> ping again. Thanks for your review.
>>>>>> 
>>>>>> Yours
>>>>>> - Michael
>>>>>> 
>>>>>> On Tue, 2012-08-28 at 11:26 -0700, Michael Liao wrote:
>>>>>>> ping again for this simple patch. - Michael
>>>>>>> 
>>>>>>> On Sat, 2012-08-25 at 14:05 -0700, Michael Liao wrote:
>>>>>>>> Hi
>>>>>>>> 
>>>>>>>> Please review the attached from Sriram fixing PR12359. As PSHUFB will zero elements of negative indices, v16i8 shuffle vector is optimized with extra checking whether V2 is zero and treat that similar to undefined V2 to save one PSHUFB and POR.
>>>>>>>> 
>>>>>>>> Let me whether it's OK to commit.
>>>>>>>> 
>>>>>>>> Yours
>>>>>>>> - Michael
>>>>>>>> 
>>>>>>>> 
>>>>>>>> _______________________________________________
>>>>>>>> llvm-commits mailing list
>>>>>>>> llvm-commits at cs.uiuc.edu
>>>>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>>>>> 
>>>>>>> 
>>>>>>> _______________________________________________
>>>>>>> llvm-commits mailing list
>>>>>>> llvm-commits at cs.uiuc.edu
>>>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>>>> 
>>>>>> 
>>>>>> _______________________________________________
>>>>>> llvm-commits mailing list
>>>>>> llvm-commits at cs.uiuc.edu
>>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>>> 
>>>>> _______________________________________________
>>>>> llvm-commits mailing list
>>>>> llvm-commits at cs.uiuc.edu
>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>> 
>>> 
>>> <0001-fix-PR12359.patch>
>> 
> 
> <0001-fix-PR12359.patch>