fix: SLPVectorizer crashes for unreachable blocks containing not schedulable instructions.

In unreachable blocks it's legal to have instructions like "%x = op %x".
Such instuctions are not schedulable. Therefore the SLPVectorizer has to check for
unreachable blocks and ignore them.

Fixes bug 20646.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@216256 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 48 insertions, 0 deletions

diff --git a/lib/Transforms/Vectorize/SLPVectorizer.cpp b/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 9fea4450a10..3aced0b4c40 100644
--- a/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -976,6 +976,14 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth) {
   Instruction *VL0 = cast<Instruction>(VL[0]);
   BasicBlock *BB = cast<Instruction>(VL0)->getParent();
 
+  if (!DT->isReachableFromEntry(BB)) {
+    // Don't go into unreachable blocks. They may contain instructions with
+    // dependency cycles which confuse the final scheduling.
+    DEBUG(dbgs() << "SLP: bundle in unreachable block.\n");
+    newTreeEntry(VL, false);
+    return;
+  }
+  
   // Check that every instructions appears once in this bundle.
   for (unsigned i = 0, e = VL.size(); i < e; ++i)
     for (unsigned j = i+1; j < e; ++j)
diff --git a/test/Transforms/SLPVectorizer/X86/unreachable.ll b/test/Transforms/SLPVectorizer/X86/unreachable.ll
new file mode 100644
index 00000000000..8d609571be1
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/unreachable.ll
@@ -0,0 +1,40 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7
+
+; Check if the SLPVectorizer does not crash when handling
+; unreachable blocks with unscheduleable instructions.
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.9.0"
+
+define void @foo(i32* nocapture %x) #0 {
+entry:
+  br label %bb2
+
+bb1:                                    ; an unreachable block
+  %t3 = getelementptr inbounds i32* %x, i64 4
+  %t4 = load i32* %t3, align 4
+  %t5 = getelementptr inbounds i32* %x, i64 5
+  %t6 = load i32* %t5, align 4
+  %bad = fadd float %bad, 0.000000e+00  ; <- an instruction with self dependency,
+                                        ;    but legal in unreachable code
+  %t7 = getelementptr inbounds i32* %x, i64 6
+  %t8 = load i32* %t7, align 4
+  %t9 = getelementptr inbounds i32* %x, i64 7
+  %t10 = load i32* %t9, align 4
+  br label %bb2
+
+bb2:
+  %t1.0 = phi i32 [ %t4, %bb1 ], [ 2, %entry ]
+  %t2.0 = phi i32 [ %t6, %bb1 ], [ 2, %entry ]
+  %t3.0 = phi i32 [ %t8, %bb1 ], [ 2, %entry ]
+  %t4.0 = phi i32 [ %t10, %bb1 ], [ 2, %entry ]
+  store i32 %t1.0, i32* %x, align 4
+  %t12 = getelementptr inbounds i32* %x, i64 1
+  store i32 %t2.0, i32* %t12, align 4
+  %t13 = getelementptr inbounds i32* %x, i64 2
+  store i32 %t3.0, i32* %t13, align 4
+  %t14 = getelementptr inbounds i32* %x, i64 3
+  store i32 %t4.0, i32* %t14, align 4
+  ret void
+}
+