[PATCH][X86] Add target specific combine rules to fold SSE/AVX/AVX2 blend intrinsics.

Eric Christopher echristo at gmail.com
Thu May 15 15:48:00 PDT 2014


On Thu, May 15, 2014 at 3:42 PM, Philip Reames
<listmail at philipreames.com> wrote:
>
> On 05/15/2014 01:45 PM, Alex Rosenberg wrote:
>>
>> After r208664, most of those patch is dead code since Clang no longer
>> generates these intrinsics except the blendv variants. We have to keep the
>> intrinsics as is until the next release because they may exist in IR, but we
>> should remove unused ones when possible.
>
> I want to express disagreement with the assumption here.  Using Clang as our
> *only* benchmark for when an intrinsic is worth retaining is wrong.  There
> are a number of other frontends which use LLVM and we should not remove
> something just because *one* no longer uses something.  Also, aren't most of
> these usable in code compiled via Clang?
>
> Just to be clear, this comment is getting at general philosophy, not this
> specific case.  I have no actual opinion w.r.t. the case at hand.
>

I think there's general agreement here and Craig's comment about using
AutoUpgrade is the right way to go.

-eric



More information about the llvm-commits mailing list