[llvm-bugs] [Bug 6246] SLP vectorization when performing scalar operations on vector elements

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=6246

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
                 CC|                            |mkuper at google.com,
                   |                            |spatel+llvm at rotateright.com
         Resolution|FIXED                       |---

--- Comment #6 from Simon Pilgrim <llvm-dev at redking.me.uk> ---
Reopening this. If we look at the equivalent c source then it appears that the
vectorizer isn't working for non 128-bit vectors:

#include <x86intrin.h>

__m128 _vadd128(__m128 a, __m128 b) {
 return _mm_setr_ps(
   a[0] + b[0],
   a[1] + b[1],
   a[2] + b[2],
   a[3] + b[3]);
}

__m256 _vadd256(__m256 a, __m256 b) {
 return _mm256_setr_ps(
   a[0] + b[0],
   a[1] + b[1],
   a[2] + b[2],
   a[3] + b[3],
   a[4] + b[4],
   a[5] + b[5],
   a[6] + b[6],
   a[7] + b[7]);
}

clang -S -O3 pr6246.c -march=btver2 -o - -emit-llvm

; Function Attrs: norecurse nounwind readnone ssp uwtable
define <4 x float> @_vadd128(<4 x float> %a, <4 x float> %b) local_unnamed_addr
#0 {
entry:
  %0 = fadd <4 x float> %a, %b
  ret <4 x float> %0
}

; Function Attrs: norecurse nounwind readnone ssp uwtable
define <8 x float> @_vadd256(<8 x float> %a, <8 x float> %b) local_unnamed_addr
#0 {
entry:
  %vecext = extractelement <8 x float> %a, i32 0
  %vecext1 = extractelement <8 x float> %b, i32 0
  %add = fadd float %vecext, %vecext1
  %vecext2 = extractelement <8 x float> %a, i32 1
  %vecext3 = extractelement <8 x float> %b, i32 1
  %add4 = fadd float %vecext2, %vecext3
  %vecext5 = extractelement <8 x float> %a, i32 2
  %vecext6 = extractelement <8 x float> %b, i32 2
  %add7 = fadd float %vecext5, %vecext6
  %vecext8 = extractelement <8 x float> %a, i32 3
  %vecext9 = extractelement <8 x float> %b, i32 3
  %add10 = fadd float %vecext8, %vecext9
  %vecext11 = extractelement <8 x float> %a, i32 4
  %vecext12 = extractelement <8 x float> %b, i32 4
  %add13 = fadd float %vecext11, %vecext12
  %vecext14 = extractelement <8 x float> %a, i32 5
  %vecext15 = extractelement <8 x float> %b, i32 5
  %add16 = fadd float %vecext14, %vecext15
  %vecext17 = extractelement <8 x float> %a, i32 6
  %vecext18 = extractelement <8 x float> %b, i32 6
  %add19 = fadd float %vecext17, %vecext18
  %vecext20 = extractelement <8 x float> %a, i32 7
  %vecext21 = extractelement <8 x float> %b, i32 7
  %add22 = fadd float %vecext20, %vecext21
  %vecinit.i = insertelement <8 x float> undef, float %add, i32 0
  %vecinit1.i = insertelement <8 x float> %vecinit.i, float %add4, i32 1
  %vecinit2.i = insertelement <8 x float> %vecinit1.i, float %add7, i32 2
  %vecinit3.i = insertelement <8 x float> %vecinit2.i, float %add10, i32 3
  %vecinit4.i = insertelement <8 x float> %vecinit3.i, float %add13, i32 4
  %vecinit5.i = insertelement <8 x float> %vecinit4.i, float %add16, i32 5
  %vecinit6.i = insertelement <8 x float> %vecinit5.i, float %add19, i32 6
  %vecinit7.i = insertelement <8 x float> %vecinit6.i, float %add22, i32 7
  ret <8 x float> %vecinit7.i
}

-- 
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/20161112/da353f99/attachment.html>