[PATCH] Don't unroll loops in loop vectorization pass when VF is one.

[James Molloy]

Hi Wei,

The important difference between loopunroll and the loop vectoriser
unrolling is alias checks and interleaving.

The normal unroller will concatenate iterations, which without good alias
analysis results in a schedule that cannot be reordered. The loop
vectoriser will use loopaccessanalysis to plant runtime pointer checks,
allowing a much better schedule.

So I don't think your patch as-is makes sense, unless loopunrollruntime has
grown runtime ptr check support recently.

Cheers,

James
On Mon, 13 Apr 2015 at 22:43, Wei Mi <wmi at google.com> wrote:

> Hi hfinkel,
>
> The patch is to fix the problem described in
> https://llvm.org/bugs/show_bug.cgi?id=23217
>
> The problem is: When loop vectorization decide not to vectorize a loop
> (VF==1), it may still unroll the loop. However, the unroll factor may be
> smaller than the factor computed in loop unroll pass, so loop unroll pass
> may unroll the already unrolled loop once more and unroll the remainder
> loop at the same time.
>
> I don't see the benefit of unrolling when VF==1. The patch is to disable
> the unrolling when VF==1 in loop vectorization pass, and let loop unroll
> pass to do the unrolling for such loop.
>
> Performance neutral for spec2000. Google internal benchmarks: detection
> improved by 5% on sandybridge and 9% on westmere, saw improved by 1.5% on
> both platforms.
>
> REPOSITORY
>   rL LLVM
>
> http://reviews.llvm.org/D9007
>
> Files:
>   lib/Transforms/Vectorize/LoopVectorize.cpp
>   test/Transforms/LoopVectorize/unroll.ll
>
> Index: lib/Transforms/Vectorize/LoopVectorize.cpp
> ===================================================================
> --- lib/Transforms/Vectorize/LoopVectorize.cpp
> +++ lib/Transforms/Vectorize/LoopVectorize.cpp
> @@ -1468,34 +1468,24 @@
>          CM.selectVectorizationFactor(OptForSize);
>
>      // Select the unroll factor.
> -    const unsigned UF =
> -        CM.selectUnrollFactor(OptForSize, VF.Width, VF.Cost);
> +    unsigned UF = CM.selectUnrollFactor(OptForSize, VF.Width, VF.Cost);
> +
> +    // If a loop is not beneficial to be vectorized, don't unroll it here.
> +    // Let the LoopUnroll pass to unroll it.
> +    if (VF.Width == 1)
> +      UF = 1;
>
>      DEBUG(dbgs() << "LV: Found a vectorizable loop (" << VF.Width << ")
> in "
>                   << DebugLocStr << '\n');
>      DEBUG(dbgs() << "LV: Unroll Factor is " << UF << '\n');
>
>      if (VF.Width == 1) {
>        DEBUG(dbgs() << "LV: Vectorization is possible but not
> beneficial\n");
>
> -      if (UF == 1) {
> -        emitOptimizationRemarkAnalysis(
> -            F->getContext(), DEBUG_TYPE, *F, L->getStartLoc(),
> -            "not beneficial to vectorize and user disabled interleaving");
> -        return false;
> -      }
> -      DEBUG(dbgs() << "LV: Trying to at least unroll the loops.\n");
> -
> -      // Report the unrolling decision.
> -      emitOptimizationRemark(F->getContext(), DEBUG_TYPE, *F,
> L->getStartLoc(),
> -                             Twine("unrolled with interleaving factor " +
> -                                   Twine(UF) +
> -                                   " (vectorization not beneficial)"));
> -
> -      // We decided not to vectorize, but we may want to unroll.
> -
> -      InnerLoopUnroller Unroller(L, SE, LI, DT, TLI, TTI, UF);
> -      Unroller.vectorize(&LVL);
> +      emitOptimizationRemarkAnalysis(
> +          F->getContext(), DEBUG_TYPE, *F, L->getStartLoc(),
> +          "not beneficial to vectorize and user disabled interleaving");
> +      return false;
>      } else {
>        // If we decided that it is *legal* to vectorize the loop then do
> it.
>        InnerLoopVectorizer LB(L, SE, LI, DT, TLI, TTI, VF.Width, UF);
> Index: test/Transforms/LoopVectorize/unroll.ll
> ===================================================================
> --- test/Transforms/LoopVectorize/unroll.ll
> +++ test/Transforms/LoopVectorize/unroll.ll
> @@ -0,0 +1,37 @@
> +; This test makes sure that loop will not be unrolled in vectorization if
> VF computed
> +; equals to 1.
> +; RUN: opt < %s -loop-vectorize -S | FileCheck %s
> +
> +; Make sure there are no geps being merged.
> +; CHECK-LABEL: @foo(
> +; CHECK: getelementptr
> +; CHECK-NOT: getelementptr
> +
> + at N = common global i32 0, align 4
> + at a = common global [1000 x i32] zeroinitializer, align 16
> +
> +define void @foo() #0 {
> +entry:
> +  %0 = load i32, i32* @N, align 4
> +  %cmp5 = icmp sgt i32 %0, 0
> +  br i1 %cmp5, label %for.body.lr.ph, label %for.end
> +
> +for.body.lr.ph:                                   ; preds = %entry
> +  %conv = sext i32 %0 to i64
> +  br label %for.body
> +
> +for.body:                                         ; preds = %
> for.body.lr.ph, %for.body
> +  %i.06 = phi i64 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
> +  %mul = mul nuw nsw i64 %i.06, 7
> +  %arrayidx = getelementptr inbounds [1000 x i32], [1000 x i32]* @a, i64
> 0, i64 %mul
> +  store i32 3, i32* %arrayidx, align 4
> +  %inc = add nuw nsw i64 %i.06, 1
> +  %cmp = icmp slt i64 %inc, %conv
> +  br i1 %cmp, label %for.body, label %for.end.loopexit
> +
> +for.end.loopexit:                                 ; preds = %for.body
> +  br label %for.end
> +
> +for.end:                                          ; preds =
> %for.end.loopexit, %entry
> +  ret void
> +}
>
> EMAIL PREFERENCES
>   http://reviews.llvm.org/settings/panel/emailpreferences/
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150414/c0f06afa/attachment.html>