[llvm] r296898 - [LoopUnrolling] Peel loops with invariant backedge Phi input

[Maxim Kazantsev via llvm-commits]

Hello Douglas,

I've taken a look and it seemed that my patch has revealed this bug, yet it reason lies in other place and is related to peeling irreducible loops. It seems that Michael has already proposed a solution patch, I hope it helps! Please let me know if you need more assessment from my side.

Best regards,
Max


-----Original Message-----
From: Yung, Douglas [mailto:douglas.yung at sony.com] 
Sent: Wednesday, March 8, 2017 8:20 AM
To: Maxim Kazantsev <max.kazantsev at azul.com>
Cc: Sanjoy Das <sanjoy at playingwithpointers.com>; llvm-commits at lists.llvm.org
Subject: RE: [llvm] r296898 - [LoopUnrolling] Peel loops with invariant backedge Phi input

Hi Max,

This change is causing an assertion failure in one of our tests. I have put the details in PR32178, can you take a look at it?

Douglas Yung

> -----Original Message-----
> From: llvm-commits [mailto:llvm-commits-bounces at lists.llvm.org] On 
> Behalf Of Sanjoy Das via llvm-commits
> Sent: Friday, March 03, 2017 10:19
> To: llvm-commits at lists.llvm.org
> Subject: [llvm] r296898 - [LoopUnrolling] Peel loops with invariant 
> backedge Phi input
> 
> Author: sanjoy
> Date: Fri Mar  3 12:19:15 2017
> New Revision: 296898
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=296898&view=rev
> Log:
> [LoopUnrolling] Peel loops with invariant backedge Phi input
> 
> Summary:
> If a loop contains a Phi node which has an invariant input from back 
> edge, it is profitable to peel such loops (rather than unroll them) to 
> use the advantage that this Phi is always invariant starting from 2nd 
> iteration. After the 1st iteration is peeled, other optimizations can 
> potentially simplify calculations with this invariant.
> 
> Patch by Max Kazantsev!
> 
> Reviewers: sanjoy, apilipenko, igor-laevsky, anna, mkuper, reames
> 
> Reviewed By: mkuper
> 
> Subscribers: mkuper, mzolotukhin, llvm-commits
> 
> Differential Revision: https://reviews.llvm.org/D30161
> 
> Added:
>     llvm/trunk/test/Transforms/LoopUnroll/peel-loop-not-forced.ll
> Modified:
>     llvm/trunk/lib/Transforms/Utils/LoopUnrollPeel.cpp
> 
> Modified: llvm/trunk/lib/Transforms/Utils/LoopUnrollPeel.cpp
> URL: http://llvm.org/viewvc/llvm-
> project/llvm/trunk/lib/Transforms/Utils/LoopUnrollPeel.cpp?rev=296898&
> r1=29689
> 7&r2=296898&view=diff
> ======================================================================
> ========
> --- llvm/trunk/lib/Transforms/Utils/LoopUnrollPeel.cpp (original)
> +++ llvm/trunk/lib/Transforms/Utils/LoopUnrollPeel.cpp Fri Mar  3
> +++ 12:19:15 2017
> @@ -71,6 +71,31 @@ void llvm::computePeelCount(Loop *L, uns
>    if (!L->empty())
>      return;
> 
> +  // Try to find a Phi node that has the same loop invariant as an 
> + input from  // its only back edge. If there is such Phi, peeling 1 
> + iteration from the  // loop is profitable, because starting from 2nd 
> + iteration we will have an  // invariant instead of this Phi.
> +  if (auto *BackEdge = L->getLoopLatch()) {
> +    BasicBlock *Header = L->getHeader();
> +    // Iterate over Phis to find one with invariant input on back edge.
> +    bool FoundCandidate = false;
> +    PHINode *Phi;
> +    for (auto BI = Header->begin(); Phi = dyn_cast<PHINode>(&*BI); ++BI) {
> +      Value *Input = Phi->getIncomingValueForBlock(BackEdge);
> +      if (L->isLoopInvariant(Input)) {
> +        FoundCandidate = true;
> +        break;
> +      }
> +    }
> +    if (FoundCandidate) {
> +      DEBUG(dbgs() << "Peel one iteration to get rid of " << *Phi
> +                   << " because starting from 2nd iteration it is always"
> +                   << " an invariant\n");
> +      UP.PeelCount = 1;
> +      return;
> +    }
> +  }
> +
>    // Bail if we know the statically calculated trip count.
>    // In this case we rather prefer partial unrolling.
>    if (TripCount)
> 
> Added: llvm/trunk/test/Transforms/LoopUnroll/peel-loop-not-forced.ll
> URL: http://llvm.org/viewvc/llvm-
> project/llvm/trunk/test/Transforms/LoopUnroll/peel-loop-not-
> forced.ll?rev=296898&view=auto
> ======================================================================
> ========
> --- llvm/trunk/test/Transforms/LoopUnroll/peel-loop-not-forced.ll 
> (added)
> +++ llvm/trunk/test/Transforms/LoopUnroll/peel-loop-not-forced.ll Fri 
> +++ Mar  3 12:19:15 2017
> @@ -0,0 +1,25 @@
> +; RUN: opt < %s -S -loop-unroll | FileCheck %s
> +
> +define i32 @invariant_backedge_1(i32 %a, i32 %b) { ; CHECK-LABEL:
> + at invariant_backedge_1
> +; CHECK-NOT:     %plus = phi
> +; CHECK:       loop.peel:
> +; CHECK:       loop:
> +; CHECK:         %i = phi
> +; CHECK:         %sum = phi
> +entry:
> +  br label %loop
> +
> +loop:
> +  %i = phi i32 [ 0, %entry ], [ %inc, %loop ]
> +  %sum = phi i32 [ 0, %entry ], [ %incsum, %loop ]
> +  %plus = phi i32 [ %a, %entry ], [ %b, %loop ]
> +
> +  %incsum = add i32 %sum, %plus
> +  %inc = add i32 %i, 1
> +  %cmp = icmp slt i32 %i, 1000
> +  br i1 %cmp, label %loop, label %exit
> +
> +exit:
> +  ret i32 %sum
> +}
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits