[LLVMdev] Please benchmark new x86 vector shuffle lowering, planning to make it the default very soon!
Andrea Di Biagio
andrea.dibiagio at gmail.com
Fri Sep 19 13:44:34 PDT 2014
On Fri, Sep 19, 2014 at 9:39 PM, Chandler Carruth <chandlerc at google.com> wrote:
> FWIW, I agree with Quentin. I'm actually confident I have test cases
> covering all of these.
Ok then. In that case, I will keep investigating to see if I can find
something more.
Cheers,
Andrea
>
> On Fri, Sep 19, 2014 at 1:22 PM, Andrea Di Biagio
> <andrea.dibiagio at gmail.com> wrote:
>>
>> 3. When zero extending 2 packed 32-bit integers, we should try to
>> emit a vpmovzxdq
>> Example:
>> vmovq 20(%rbx), %xmm0
>> vpshufd $80, %xmm0, %xmm0 # %xmm0 = %xmm0[0,0,1,1]
>>
>> Before:
>> vpmovzxdq 20(%rbx), %xmm0
>
>
> This one is already fixed.
More information about the llvm-dev
mailing list