[DAG] Prevent NodeId pruning of TokenFactors in Instruction Selection.

Summary:
Instruction Selection preserves relative orders of all nodes save
TokenFactors which we treat specially. As a result Node Ids for
TokenFactors may violate the topological ordering and should not be
considered as valid pruning candidates in predecessor search.

Fixes PR35316.

Reviewers: RKSimon, hfinkel

Subscribers: hiraditya, llvm-commits

Differential Revision: https://reviews.llvm.org/D42701

llvm-svn: 323880

2 files changed, 79 insertions, 1 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
index e90ec20bcdb..b3a316b93b4 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
@@ -2157,7 +2157,9 @@ static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse,
   while (!WorkList.empty()) {
     Use = WorkList.back();
     WorkList.pop_back();
-    if (Use->getNodeId() < Def->getNodeId() && Use->getNodeId() != -1)
+    // NodeId topological order of TokenFactors is not guaranteed. Do not skip.
+    if (Use->getOpcode() != ISD::TokenFactor &&
+        Use->getNodeId() < Def->getNodeId() && Use->getNodeId() != -1)
       continue;
 
     // Don't revisit nodes if we already scanned it and didn't fail, we know we
diff --git a/llvm/test/CodeGen/X86/pr35316.ll b/llvm/test/CodeGen/X86/pr35316.ll
new file mode 100644
index 00000000000..1c1523b3547
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pr35316.ll
@@ -0,0 +1,76 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s --check-prefix=CHECK
+
+; This tests for a cyclic dependencies in the generated DAG. 
+
+@c = external local_unnamed_addr global i32, align 4
+@a = external local_unnamed_addr global i32, align 4
+@b = external local_unnamed_addr global i32, align 4
+
+define void @main() {
+; CHECK-LABEL: main:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl $0, {{.*}}(%rip)
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %r8d
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %edi
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %esi
+; CHECK-NEXT:    movl -{{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    cltd
+; CHECK-NEXT:    idivl {{.*}}(%rip)
+; CHECK-NEXT:    movl %eax, %ecx
+; CHECK-NEXT:    movl {{.*}}(%rip), %eax
+; CHECK-NEXT:    cltd
+; CHECK-NEXT:    idivl %esi
+; CHECK-NEXT:    andl %edi, %eax
+; CHECK-NEXT:    addl %ecx, %eax
+; CHECK-NEXT:    andl %r8d, %eax
+; CHECK-NEXT:    movl %eax, (%rax)
+; CHECK-NEXT:    retq
+entry:
+  %e = alloca i32, align 4
+  %e.0.e.0.24 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.25 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.26 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.27 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.28 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.29 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.30 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.31 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.32 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.33 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.34 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.35 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.36 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.37 = load volatile i32, i32* %e, align 4
+  %e.0.e.0.39 = load volatile i32, i32* %e, align 4
+  %tmp = load i32, i32* @a, align 4
+  store i32 0, i32* @b, align 4
+  %e.0.e.0.41 = load volatile i32, i32* %e, align 4
+  %add17 = add nsw i32 %e.0.e.0.41, 0
+  %e.0.e.0.42 = load volatile i32, i32* %e, align 4
+  %tmp1 = load i32, i32* @c, align 4
+  %e.0.e.0.43 = load volatile i32, i32* %e, align 4
+  %div = sdiv i32 %tmp1, %e.0.e.0.43
+  %and18 = and i32 %div, %e.0.e.0.42
+  %e.0.e.0.44 = load volatile i32, i32* %e, align 4
+  %div19 = sdiv i32 %e.0.e.0.44, %tmp
+  %add20 = add nsw i32 %div19, %and18
+  %and21 = and i32 %add20, %add17
+  store volatile i32 %and21, i32* undef, align 4
+  ret void
+}