[llvm] r237544 - RegisterPressureTracker: reword stale comments.

Andrew Trick atrick at apple.com
Sun May 17 16:40:28 PDT 2015


Author: atrick
Date: Sun May 17 18:40:27 2015
New Revision: 237544

URL: http://llvm.org/viewvc/llvm-project?rev=237544&view=rev
Log:
RegisterPressureTracker: reword stale comments.

Modified:
    llvm/trunk/lib/CodeGen/MachineScheduler.cpp
    llvm/trunk/lib/CodeGen/RegisterPressure.cpp

Modified: llvm/trunk/lib/CodeGen/MachineScheduler.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/MachineScheduler.cpp?rev=237544&r1=237543&r2=237544&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/MachineScheduler.cpp (original)
+++ llvm/trunk/lib/CodeGen/MachineScheduler.cpp Sun May 17 18:40:27 2015
@@ -2611,8 +2611,7 @@ void GenericScheduler::tryCandidate(Sche
                  TryCand, Cand, PhysRegCopy))
     return;
 
-  // Avoid exceeding the target's limit. If signed PSetID is negative, it is
-  // invalid; convert it to INT_MAX to give it lowest priority.
+  // Avoid exceeding the target's limit.
   if (DAG->isTrackingPressure() && tryPressure(TryCand.RPDelta.Excess,
                                                Cand.RPDelta.Excess,
                                                TryCand, Cand, RegExcess))

Modified: llvm/trunk/lib/CodeGen/RegisterPressure.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/RegisterPressure.cpp?rev=237544&r1=237543&r2=237544&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/RegisterPressure.cpp (original)
+++ llvm/trunk/lib/CodeGen/RegisterPressure.cpp Sun May 17 18:40:27 2015
@@ -750,9 +750,11 @@ void RegPressureTracker::bumpUpwardPress
 ///
 /// This assumes that the current LiveOut set is sufficient.
 ///
-/// FIXME: This is expensive for an on-the-fly query. We need to cache the
-/// result per-SUnit with enough information to adjust for the current
-/// scheduling position. But this works as a proof of concept.
+/// This is expensive for an on-the-fly query because it calls
+/// bumpUpwardPressure to recompute the pressure sets based on current
+/// liveness. This mainly exists to verify correctness, e.g. with
+/// -verify-misched. getUpwardPressureDelta is the fast version of this query
+/// that uses the per-SUnit cache of the PressureDiff.
 void RegPressureTracker::
 getMaxUpwardPressureDelta(const MachineInstr *MI, PressureDiff *PDiff,
                           RegPressureDelta &Delta,
@@ -809,10 +811,8 @@ getMaxUpwardPressureDelta(const MachineI
 #endif
 }
 
-/// This is a prototype of the fast version of querying register pressure that
-/// does not directly depend on current liveness. It's still slow because we
-/// recompute pressure change on-the-fly. This implementation only exists to
-/// prove correctness.
+/// This is the fast version of querying register pressure that does not
+/// directly depend on current liveness.
 ///
 /// @param Delta captures information needed for heuristics.
 ///
@@ -950,6 +950,11 @@ void RegPressureTracker::bumpDownwardPre
 /// register units of that pressure set introduced by this instruction.
 ///
 /// This assumes that the current LiveIn set is sufficient.
+///
+/// This is expensive for an on-the-fly query because it calls
+/// bumpDownwardPressure to recompute the pressure sets based on current
+/// liveness. We don't yet have a fast version of downward pressure tracking
+/// analagous to getUpwardPressureDelta.
 void RegPressureTracker::
 getMaxDownwardPressureDelta(const MachineInstr *MI, RegPressureDelta &Delta,
                             ArrayRef<PressureChange> CriticalPSets,





More information about the llvm-commits mailing list