[CodeGen] Fix hashing for MO_ExternalSymbol MachineOperands.

We were hashing the string pointer, not the string, so two instructions
could be identical (isIdenticalTo), but have different hash codes.

This showed up as a very rare, non-deterministic assertion failure
rehashing a DenseMap constructed by MachineOutliner.  So there's no
"real" testcase, just a unittest which checks that the hash function
behaves correctly.

I'm a little scared fixing this is going to cause a regression in
outlining or MachineCSE, but hopefully we won't run into any issues.

Differential Revision: https://reviews.llvm.org/D61975

llvm-svn: 362281

1 files changed, 10 insertions, 0 deletions

diff --git a/llvm/unittests/CodeGen/MachineOperandTest.cpp b/llvm/unittests/CodeGen/MachineOperandTest.cpp
index cfedfa9c883..faa471f2260 100644
--- a/llvm/unittests/CodeGen/MachineOperandTest.cpp
+++ b/llvm/unittests/CodeGen/MachineOperandTest.cpp
@@ -398,4 +398,14 @@ TEST(MachineOperandTest, PrintPredicate) {
   ASSERT_TRUE(OS.str() == "intpred(eq)");
 }
 
+TEST(MachineOperandTest, HashValue) {
+  char SymName1[] = "test";
+  char SymName2[] = "test";
+  MachineOperand MO1 = MachineOperand::CreateES(SymName1);
+  MachineOperand MO2 = MachineOperand::CreateES(SymName2);
+  ASSERT_NE(SymName1, SymName2);
+  ASSERT_EQ(hash_value(MO1), hash_value(MO2));
+  ASSERT_TRUE(MO1.isIdenticalTo(MO2));
+}
+
 } // end namespace