Track register pressure a bit more carefully (weird corner case).

This solves a problem where a def machine operand has no uses but has
not been marked dead. In this case, the initial RP analysis was being
extra precise and determining from LiveIntervals the the register was
actually dead. This caused us to omit the register from the RP
tracker's block live out. That's all good, but the per-instruction
summary still accounted for it as a valid def. This could cause an
assertion in the tracker later when we underflow pressure.

This is from a bug report on an out-of-tree target. It is not
reproducible on well-behaved targets. I'm just making an obvious fix
without unit test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@200941 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/RegisterPressure.cpp b/lib/CodeGen/RegisterPressure.cpp
index 092ecdd9bb256e185ee6671412dd01df9aa7c3e8..249789a4b16bb58122386aec689d4d0954d152cc 100644 (file)
--- a/lib/CodeGen/RegisterPressure.cpp
+++ b/lib/CodeGen/RegisterPressure.cpp
@@ -506,7 +506,14 @@ bool RegPressureTracker::recede(SmallVectorImpl<unsigned> *LiveUses,
         DeadDef = LRQ.isDeadDef();
       }
     }
-    if (!DeadDef) {
+    if (DeadDef) {
+      // LiveIntervals knows this is a dead even though it's MachineOperand is
+      // not flagged as such. Since this register will not be recorded as
+      // live-out, increase its PDiff value to avoid underflowing pressure.
+      if (PDiff)
+        PDiff->addPressureChange(Reg, false, MRI);
+    }
+    else {
       if (LiveRegs.erase(Reg))
         decreaseRegPressure(Reg);
       else