[llvm-bugs] [Bug 39921] [X86] AVX2 should use an extract_subvector and phadd for the first step of a pairwise v8i32 addition reduction

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=39921

Simon Pilgrim <llvm-dev at redking.me.uk> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Fixed By Commit(s)|r359491                     |r359491,r362327
         Resolution|---                         |FIXED
             Status|NEW                         |RESOLVED

--- Comment #2 from Simon Pilgrim <llvm-dev at redking.me.uk> ---
(In reply to Simon Pilgrim from comment #1)
> But Intel targets can get stuck as the 'fast shuffle' attribute gets in the
> way:
> 
> pairwise_reduction8i32: # @pairwise_reduction8i32
>   vmovdqa .LCPI0_0(%rip), %ymm1 # ymm1 = [0,2,4,6,4,6,6,7]
>   vpermd %ymm0, %ymm1, %ymm1
>   vmovdqa .LCPI0_1(%rip), %ymm2 # ymm2 = [1,3,5,7,5,7,6,7]
>   vpermd %ymm0, %ymm2, %ymm0
>   vpaddd %xmm0, %xmm1, %xmm0
>   vpshufd $232, %xmm0, %xmm1 # xmm1 = xmm0[0,2,2,3]
>   vpshufd $237, %xmm0, %xmm0 # xmm0 = xmm0[1,3,2,3]
>   vpaddd %xmm0, %xmm1, %xmm0
>   vpshufd $229, %xmm0, %xmm1 # xmm1 = xmm0[1,1,2,3]
>   vpaddd %xmm1, %xmm0, %xmm0
>   vmovd %xmm0, %eax
>   vzeroupper
>   retq

Resolving - the fast-variable-shuffle issue was fixed at rL362327:

pairwise_reduction8i32: # @pairwise_reduction8i32
  vextracti128 $1, %ymm0, %xmm1
  vphaddd %xmm1, %xmm0, %xmm0
  vphaddd %xmm0, %xmm0, %xmm0
  vpshufd $229, %xmm0, %xmm1 # xmm1 = xmm0[1,1,2,3]
  vpaddd %xmm1, %xmm0, %xmm0
  vmovd %xmm0, %eax
  vzeroupper
  retq

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20190602/97fd50f1/attachment-0001.html>