From 7330e358f6b5c696fe58a63671e64dc81d92dee8 Mon Sep 17 00:00:00 2001 From: Erik Eckstein Date: Mon, 2 Feb 2015 12:45:34 +0000 Subject: [PATCH] Fix: SLPVectorizer crashes with assertion when vectorizing a cmp instruction. The commit r225977 uncovered this bug. The problem was that the vectorizer tried to read the second operand of an already deleted instruction. The bug didn't show up before r225977 because the freed memory still contained a non-null pointer. With r225977 deletion of instructions is delayed and the read operand pointer is always null. llvm-svn: 227800 --- llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp | 1 + .../Transforms/SLPVectorizer/X86/crash_cmpop.ll | 56 ++++++++++++++++++++++ 2 files changed, 57 insertions(+) create mode 100644 llvm/test/Transforms/SLPVectorizer/X86/crash_cmpop.ll diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp index 08af1da..d1a8438 100644 --- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp +++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp @@ -3943,6 +3943,7 @@ bool SLPVectorizer::vectorizeChainsInBlock(BasicBlock *BB, BoUpSLP &R) { // and the iterator may become invalid value. it = BB->begin(); e = BB->end(); + break; } } } diff --git a/llvm/test/Transforms/SLPVectorizer/X86/crash_cmpop.ll b/llvm/test/Transforms/SLPVectorizer/X86/crash_cmpop.ll new file mode 100644 index 0000000..18a96e5 --- /dev/null +++ b/llvm/test/Transforms/SLPVectorizer/X86/crash_cmpop.ll @@ -0,0 +1,56 @@ +; RUN: opt < %s -basicaa -slp-vectorizer -S + +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" +target triple = "x86_64-apple-macosx10.10.0" + +define void @testfunc(float* nocapture %dest, float* nocapture readonly %src) { +entry: + br label %for.body + +for.body: + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] + %acc1.056 = phi float [ 0.000000e+00, %entry ], [ %add13, %for.body ] + %s1.055 = phi float [ 0.000000e+00, %entry ], [ %cond.i40, %for.body ] + %s0.054 = phi float [ 0.000000e+00, %entry ], [ %cond.i44, %for.body ] + %arrayidx = getelementptr inbounds float* %src, i64 %indvars.iv + %0 = load float* %arrayidx, align 4 + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 + %arrayidx2 = getelementptr inbounds float* %dest, i64 %indvars.iv + store float %acc1.056, float* %arrayidx2, align 4 + %add = fadd float %s0.054, %0 + %add3 = fadd float %s1.055, %0 + %mul = fmul float %s0.054, 0.000000e+00 + %add4 = fadd float %mul, %add3 + %mul5 = fmul float %s1.055, 0.000000e+00 + %add6 = fadd float %mul5, %add + %cmp.i = fcmp olt float %add6, 1.000000e+00 + %cond.i = select i1 %cmp.i, float %add6, float 1.000000e+00 + %cmp.i51 = fcmp olt float %cond.i, -1.000000e+00 + %cmp.i49 = fcmp olt float %add4, 1.000000e+00 + %cond.i50 = select i1 %cmp.i49, float %add4, float 1.000000e+00 + %cmp.i47 = fcmp olt float %cond.i50, -1.000000e+00 + %cond.i.op = fmul float %cond.i, 0.000000e+00 + %mul10 = select i1 %cmp.i51, float -0.000000e+00, float %cond.i.op + %cond.i50.op = fmul float %cond.i50, 0.000000e+00 + %mul11 = select i1 %cmp.i47, float -0.000000e+00, float %cond.i50.op + %add13 = fadd float %mul10, %mul11 + + ; The SLPVectorizer crashed in vectorizeChainsInBlock() because it tried + ; to access the second operand of the following cmp after the cmp itself + ; was already vectorized and deleted. + %cmp.i45 = fcmp olt float %add13, 1.000000e+00 + + %cond.i46 = select i1 %cmp.i45, float %add13, float 1.000000e+00 + %cmp.i43 = fcmp olt float %cond.i46, -1.000000e+00 + %cond.i44 = select i1 %cmp.i43, float -1.000000e+00, float %cond.i46 + %cmp.i41 = fcmp olt float %mul11, 1.000000e+00 + %cond.i42 = select i1 %cmp.i41, float %mul11, float 1.000000e+00 + %cmp.i39 = fcmp olt float %cond.i42, -1.000000e+00 + %cond.i40 = select i1 %cmp.i39, float -1.000000e+00, float %cond.i42 + %exitcond = icmp eq i64 %indvars.iv.next, 32 + br i1 %exitcond, label %for.end, label %for.body + +for.end: + ret void +} + -- 2.7.4