[llvm] r265337 - Enable unroll for constant bound loops when TripCount is not modulo of unroll factor, reducing it to maximum power-of-2 that satisfies threshold limit.

[Evgeny Stupachenko via llvm-commits]

Before the patch the loop
for (i = 0; i < 15; i++)
  loop_body;
was not unrolled,

the loop
for (i = 0; i < 16; i++)
  loop_body;
was unrolled

the loop
for (i = 0; i < n; i++)
  loop_body;
was unrolled

Why we should avoid unrolling if threshold let us unroll a loop?
The sense of unrolling (right now) is to reduce induction variable and
compare/branch costs.

One of possible solutions is to add " && Unrolling == Runtime":
>       if (Count <= 1 && Unrolling == Runtime) {

However I still do not understand why we should avoid unrolling if
threshold let us unroll a loop?
For the cases where unroll is unprofitable there should be
corresponding heuristics.
What is your case?

Thanks,
Evgeny

On Mon, Apr 4, 2016 at 1:10 PM,  <escha at apple.com> wrote:
> What if we don’t want this behavior?
>
>       if (Count <= 1) {
>         // If there is no Count that is modulo of TripCount, set Count to
>         // largest power-of-two factor that satisfies the threshold limit.
>         Count = (std::max(UP.PartialThreshold, 3u) - 2) / (LoopSize - 2);
>         UnrolledSize = (LoopSize - 2) * Count + 2;
>         while (Count != 0 && UnrolledSize > UP.PartialThreshold) {
>           Count >>= 1;
>           UnrolledSize = (LoopSize - 2) * Count + 2;
>         }
>       }
>
> Previously, partial unrolling was always modulo the trip count; this allows
> partial unrolling to duplicate the loop, which we don’t want. What should we
> do if we don’t want this behavior?
>
> —escha
>
> On Apr 4, 2016, at 1:02 PM, Evgeny Stupachenko via llvm-commits
> <llvm-commits at lists.llvm.org> wrote:
>
> When the a TripCount is not modulo of Count (unroll factor) we jump
> into the middle of a loop:
>
> i = 0;
> while(1) {
>  a[i] = i;
>  i++;
>  if (i >= 15) break;
> }
>
> Unrolled by 2 into:
>
> i = 0;
> jump L1;
> while(1) {
>  a[i] = i;
>  i++;
> L1:
>  a[i] = i;
>  i++;
>  if (i >= 15) break;
> }
>
> Here we check this type of unroll. It could be optimized by further
> passes, however unroll should do exactly this.
>
> Thanks,
> Evgeny
>
>
>
> On Mon, Apr 4, 2016 at 12:37 PM, Sean Silva via llvm-commits
> <llvm-commits at lists.llvm.org> wrote:
>
>
>
> On Mon, Apr 4, 2016 at 12:24 PM, Zia Ansari via llvm-commits
> <llvm-commits at lists.llvm.org> wrote:
>
>
> Author: zansari
> Date: Mon Apr  4 14:24:46 2016
> New Revision: 265337
>
> URL: http://llvm.org/viewvc/llvm-project?rev=265337&view=rev
> Log:
> Enable unroll for constant bound loops when TripCount is not modulo of
> unroll factor, reducing it to maximum power-of-2 that satisfies threshold
> limit.
>
> Commit for Evgeny Stupachenko (evstupac at gmail.com)
>
> Differential Revision: http://reviews.llvm.org/D18290
>
>
> Added:
>    llvm/trunk/test/Transforms/LoopUnroll/partial-unroll-const-bounds.ll
> Modified:
>    llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp
>
> Modified: llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp?rev=265337&r1=265336&r2=265337&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp (original)
> +++ llvm/trunk/lib/Transforms/Scalar/LoopUnrollPass.cpp Mon Apr  4
> 14:24:46 2016
> @@ -639,6 +639,16 @@ static bool tryToUnrollLoop(Loop *L, Dom
>       Count = (std::max(UP.PartialThreshold, 3u) - 2) / (LoopSize - 2);
>       while (Count != 0 && TripCount % Count != 0)
>         Count--;
> +      if (Count <= 1) {
> +        // If there is no Count that is modulo of TripCount, set Count to
> +        // largest power-of-two factor that satisfies the threshold
> limit.
> +        Count = (std::max(UP.PartialThreshold, 3u) - 2) / (LoopSize - 2);
> +        UnrolledSize = (LoopSize - 2) * Count + 2;
> +        while (Count != 0 && UnrolledSize > UP.PartialThreshold) {
> +          Count >>= 1;
> +          UnrolledSize = (LoopSize - 2) * Count + 2;
> +        }
> +      }
>     }
>   } else if (Unrolling == Runtime) {
>     if (!AllowRuntime && !CountSetExplicitly) {
>
> Added:
> llvm/trunk/test/Transforms/LoopUnroll/partial-unroll-const-bounds.ll
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopUnroll/partial-unroll-const-bounds.ll?rev=265337&view=auto
>
> ==============================================================================
> --- llvm/trunk/test/Transforms/LoopUnroll/partial-unroll-const-bounds.ll
> (added)
> +++ llvm/trunk/test/Transforms/LoopUnroll/partial-unroll-const-bounds.ll
> Mon Apr  4 14:24:46 2016
> @@ -0,0 +1,29 @@
> +; RUN: opt < %s -S -unroll-threshold=20 -loop-unroll
> -unroll-allow-partial | FileCheck %s
> +
> +; The Loop TripCount is 9. However unroll factors 3 or 9 exceed given
> threshold.
> +; The test checks that we choose a smaller, power-of-two, unroll count
> and do not give up on unrolling.
> +
> +; CHECK: for.body:
> +; CHECK: store
> +; CHECK: for.body.1:
> +; CHECK: store
>
>
>
> Could these checks be tightened up? E.g. to verify that the bodies are
> actually unrolled?
>
> -- Sean Silva
>
>
> +
> +define void @foo(i32* nocapture %a, i32* nocapture readonly %b) nounwind
> uwtable {
> +entry:
> +  br label %for.body
> +
> +for.body:                                         ; preds = %for.body,
> %entry
> +  %indvars.iv = phi i64 [ 1, %entry ], [ %indvars.iv.next, %for.body ]
> +  %arrayidx = getelementptr inbounds i32, i32* %b, i64 %indvars.iv
> +  %ld = load i32, i32* %arrayidx, align 4
> +  %idxprom1 = sext i32 %ld to i64
> +  %arrayidx2 = getelementptr inbounds i32, i32* %a, i64 %idxprom1
> +  %st = trunc i64 %indvars.iv to i32
> +  store i32 %st, i32* %arrayidx2, align 4
> +  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
> +  %exitcond = icmp eq i64 %indvars.iv.next, 10
> +  br i1 %exitcond, label %for.end, label %for.body
> +
> +for.end:                                          ; preds = %for.body
> +  ret void
> +}
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
>