[llvm] r187267 - SLP Vectorier: Don't vectorize really short chains because they are already handled by the SelectionDAG store-vectorizer, which does a better job in deciding when to vectorize.

[Nick Lewycky]

Nadav Rotem wrote:
> Author: nadav
> Date: Fri Jul 26 18:07:55 2013
> New Revision: 187267
>
> URL: http://llvm.org/viewvc/llvm-project?rev=187267&view=rev
> Log:
> SLP Vectorier:  Don't vectorize really short chains because they are already handled by the SelectionDAG store-vectorizer, which does a better job in deciding when to vectorize.
>
>
> Modified:
>      llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp
>      llvm/trunk/test/Transforms/SLPVectorizer/X86/reduction2.ll
>
> Modified: llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp?rev=187267&r1=187266&r2=187267&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp (original)
> +++ llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp Fri Jul 26 18:07:55 2013
> @@ -898,8 +898,12 @@ int BoUpSLP::getTreeCost() {
>     DEBUG(dbgs()<<  "SLP: Calculating cost for tree of size "<<
>           VectorizableTree.size()<<  ".\n");
>
> -  if (!VectorizableTree.size()) {
> -    assert(!ExternalUses.size()&&  "We should not have any external users");
> +  // Don't vectorize tiny trees. Small load/store chains or consecutive stores
> +  // of constants will be vectoried in SelectionDAG in MergeConsecutiveStores.

Typo, vectoried should be vectorized.

Nick

> +  if (VectorizableTree.size()<  3) {
> +    if (!VectorizableTree.size()) {
> +      assert(!ExternalUses.size()&&  "We should not have any external users");
> +    }
>       return 0;
>     }
>
>
> Modified: llvm/trunk/test/Transforms/SLPVectorizer/X86/reduction2.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SLPVectorizer/X86/reduction2.ll?rev=187267&r1=187266&r2=187267&view=diff
> ==============================================================================
> --- llvm/trunk/test/Transforms/SLPVectorizer/X86/reduction2.ll (original)
> +++ llvm/trunk/test/Transforms/SLPVectorizer/X86/reduction2.ll Fri Jul 26 18:07:55 2013
> @@ -16,11 +16,13 @@ define double @foo(double* nocapture %D)
>     %3 = getelementptr inbounds double* %D, i32 %2
>     %4 = load double* %3, align 4
>     %A4 = fmul double %4, %4
> +  %A42 = fmul double %A4, %A4
>     %5 = or i32 %2, 1
>     %6 = getelementptr inbounds double* %D, i32 %5
>     %7 = load double* %6, align 4
>     %A7 = fmul double %7, %7
> -  %8 = fadd double %A4, %A7
> +  %A72 = fmul double %A7, %A7
> +  %8 = fadd double %A42, %A72
>     %9 = fadd double %sum.01, %8
>     %10 = add nsw i32 %i.02, 1
>     %exitcond = icmp eq i32 %10, 100
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>