[llvm-dev] [SCEV][ScalarEvolution] SE limitation impacting LV

[Caballero, Diego via llvm-dev]

Hi!

I would appreciate some feedback from someone with experience in SCEV/SE. D39346 tries to fix an issue in LV (PR34965) that exposes a limitation in SCEV/SE. The best solution to the LV issue might not be a fix at SCEV/SE level but we may want to report/address SCEV/SE limitation as well.

For the snippet below, LV expects SE to return a SCEVAddRecExpr for %21. However, SE returns ((4 * (zext i32 (2 + %18) to i64))<nuw><nsw> + undef)<nsw>, probably because it's not able to deduce that  %bc.resume.val1 = %bc.resume.val + 1. You can find further information in D39346 and the full test in PR34965 ("IR_after_createVectorizedLoopSkeleton" attachment).

I could open a new bug against SCEV/SE if a fix or extension is feasible at that level.

Thanks,
Diego

----

.outer:                                           ; preds = %2, %0
  ...
  %.ph2 = phi i32 [ 62, %2 ], [ 110, %0 ]
  %5 = add i32 %.ph2, 1
  ...

vector.ph:                                        ; preds = %.outer
  ...
  %ind.end = add i32 %5, %n.vec
  %ind.end2 = add i32 %.ph2, %n.vec
  ...

middle.block:                                     ; preds = %vector.body.latch
  ...

scalar.ph:                                        ; preds = %middle.block, %.outer
  %bc.resume.val = phi i32 [ %ind.end, %middle.block ], [ %5, %.outer ]
  %bc.resume.val1 = phi i32 [ %ind.end2, %middle.block ], [ %.ph2, %.outer ]
  br label %16

; <label>:16:                                     ; preds = %16, %scalar.ph
  %17 = phi i32 [ %bc.resume.val, %scalar.ph ], [ %23, %16 ]
  %18 = phi i32 [ %bc.resume.val1, %scalar.ph ], [ %17, %16 ]
  %19 = add i32 %18, 2
  %20 = zext i32 %19 to i64
  %21 = getelementptr inbounds i32, i32 addrspace(1)* undef, i64 %20
  %22 = ashr i32 undef, %4
  store i32 %22, i32 addrspace(1)* %21, align 4
 %23 = add i32 %17, 1
  %24 = icmp sgt i32 %23, 61
  br i1 %24, label %._crit_edge.loopexit.preheader, label %16

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20171114/e6657172/attachment.html>