Ignore IMPLICIT_DEF instructions when computing physreg liveness.

While they appear to provide a normal clobbering def, they don't
in the case of the awkward IMPLICIT_DEF+INSERT_SUBREG idiom. It
would be good to change INSERT_SUBREG; until then, this change
allows post-regalloc scheduling to cope in a mildly conservative
way.

llvm-svn: 60583

1 files changed, 7 insertions, 0 deletions

diff --git a/llvm/lib/CodeGen/PostRASchedulerList.cpp b/llvm/lib/CodeGen/PostRASchedulerList.cpp
index ec56feaff96..b19525cbaa8 100644
--- a/llvm/lib/CodeGen/PostRASchedulerList.cpp
+++ b/llvm/lib/CodeGen/PostRASchedulerList.cpp
@@ -353,6 +353,13 @@ bool SchedulePostRATDList::BreakAntiDependencies() {
        I != E; ++I, --Count) {
     MachineInstr *MI = &*I;
 
+    // After regalloc, IMPLICIT_DEF instructions aren't safe to treat as
+    // dependence-breaking. In the case of an INSERT_SUBREG, the IMPLICIT_DEF
+    // is left behind appearing to clobber the super-register, while the
+    // subregister needs to remain live. So we just ignore them.
+    if (MI->getOpcode() == TargetInstrInfo::IMPLICIT_DEF)
+      continue;
+
     // Check if this instruction has an anti-dependence that we're
     // interested in.
     DenseMap<MachineInstr *, unsigned>::iterator C = CriticalAntiDeps.find(MI);