From cbd66afc98d0231f2e64c39a9b4a5b9386d886dd Mon Sep 17 00:00:00 2001 From: Michael Zolotukhin Date: Fri, 13 May 2016 01:42:34 +0000 Subject: [PATCH] [LoopUnrollAnalyzer] Don't treat gep-instructions with simplified offset as simplified. Summary: Currently we consider such instructions as simplified, which is incorrect, because if their user isn't simplified, we can't actually simplify them too. This biases our estimates of profitability: for instance the analyzer expects much more gains from unrolling memcpy loops than there actually are. Reviewers: hfinkel, chandlerc Subscribers: mzolotukhin, llvm-commits Differential Revision: http://reviews.llvm.org/D17365 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@269387 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Analysis/LoopUnrollAnalyzer.cpp | 2 +- .../LoopUnroll/full-unroll-heuristics-cmp.ll | 2 +- .../LoopUnroll/full-unroll-heuristics-geps.ll | 28 ++++++++++++++++++++++ 3 files changed, 30 insertions(+), 2 deletions(-) create mode 100644 test/Transforms/LoopUnroll/full-unroll-heuristics-geps.ll diff --git a/lib/Analysis/LoopUnrollAnalyzer.cpp b/lib/Analysis/LoopUnrollAnalyzer.cpp index 20e79a82eb9..fe73fe05724 100644 --- a/lib/Analysis/LoopUnrollAnalyzer.cpp +++ b/lib/Analysis/LoopUnrollAnalyzer.cpp @@ -59,7 +59,7 @@ bool UnrolledInstAnalyzer::simplifyInstWithSCEV(Instruction *I) { Address.Base = Base->getValue(); Address.Offset = Offset->getValue(); SimplifiedAddresses[I] = Address; - return true; + return false; } /// Try to simplify binary operator I. diff --git a/test/Transforms/LoopUnroll/full-unroll-heuristics-cmp.ll b/test/Transforms/LoopUnroll/full-unroll-heuristics-cmp.ll index 2804848d8ef..12ad4f26bf8 100644 --- a/test/Transforms/LoopUnroll/full-unroll-heuristics-cmp.ll +++ b/test/Transforms/LoopUnroll/full-unroll-heuristics-cmp.ll @@ -1,4 +1,4 @@ -; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-dynamic-cost-savings-discount=1000 -unroll-threshold=10 -unroll-percent-dynamic-cost-saved-threshold=50 | FileCheck %s +; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-dynamic-cost-savings-discount=1000 -unroll-threshold=10 -unroll-percent-dynamic-cost-saved-threshold=40 | FileCheck %s target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" @known_constant = internal unnamed_addr constant [10 x i32] [i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1], align 16 diff --git a/test/Transforms/LoopUnroll/full-unroll-heuristics-geps.ll b/test/Transforms/LoopUnroll/full-unroll-heuristics-geps.ll new file mode 100644 index 00000000000..bc08029d4f8 --- /dev/null +++ b/test/Transforms/LoopUnroll/full-unroll-heuristics-geps.ll @@ -0,0 +1,28 @@ +; RUN: opt < %s -S -loop-unroll -unroll-max-iteration-count-to-analyze=100 -unroll-dynamic-cost-savings-discount=1000 -unroll-threshold=10 -unroll-percent-dynamic-cost-saved-threshold=40 | FileCheck %s +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" + +; When examining gep-instructions we shouldn't consider them simplified if the +; corresponding memory access isn't simplified. Doing the opposite might bias +; our estimate, so that we might decide to unroll even a simple memcpy loop. +; +; Thus, the following loop shouldn't be unrolled: +; CHECK-LABEL: @not_simplified_geps +; CHECK: br i1 % +; CHECK: ret void +define void @not_simplified_geps(i32* noalias %b, i32* noalias %c) { +entry: + br label %for.body + +for.body: + %iv.0 = phi i64 [ 0, %entry ], [ %iv.1, %for.body ] + %arrayidx1 = getelementptr inbounds i32, i32* %b, i64 %iv.0 + %x1 = load i32, i32* %arrayidx1, align 4 + %arrayidx2 = getelementptr inbounds i32, i32* %c, i64 %iv.0 + store i32 %x1, i32* %arrayidx2, align 4 + %iv.1 = add nuw nsw i64 %iv.0, 1 + %exitcond = icmp eq i64 %iv.1, 10 + br i1 %exitcond, label %for.end, label %for.body + +for.end: + ret void +} -- 2.11.0