[AMDGPU][Scheduler] Avoid initializing Register pressure tracker when tracking is disabled

[gandhi21299]

Summary

When register pressure tracking is disabled, the scheduler attempts to load
pressures at SReg_32 and VGPR_32. This causes an index out of bounds error.
This patch fixes this issue by inserting checks for RPTracking when pressure values
are loaded and candidates are initialized. NFC

[rampitec]

Needs test.

[kerbowa]

Can we not rename initCandidate since it is mimicking GenericScheduler::initCandidate?

[gandhi21299]

• reverted the function name back to initCandidate
• added testing for the option -misched-regpressure=false

[rampitec]

Why did extractelement.ll change? It wasn't suppose to happen.

[gandhi21299]

@rampitec Agreed, I have been examining the code around the changes I have made. I am not sure why that test was affected.

[gandhi21299]

• Use DAG to check if RPTracking is enabled.
• Eliminate diffs in extract-element.ll as this patch should not have altered that test.

[rampitec]

In D129322#3685811, @gandhi21299 wrote:

• Use DAG to check if RPTracking is enabled.
• Eliminate diffs in extract-element.ll as this patch should not have altered that test.

What exactly was the problem with extract-element.ll?

[gandhi21299]

@rampitec I was loading the pressure values at every candidate. initCandidate clears the values in Pressure so the array becomes empty for the next iteration.

llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp

void GCNMaxOccupancySchedStrategy::initialize(ScheduleDAGMI *DAG) {
   VGPRExcessLimit = std::min(VGPRExcessLimit - ErrorMargin, VGPRExcessLimit);
 }
 
 void GCNMaxOccupancySchedStrategy::initCandidate(SchedCandidate &Cand, SUnit *SU,
                                      bool AtTop, const RegPressureTracker &RPTracker,
                                      const SIRegisterInfo *SRI,
                                      unsigned SGPRPressure,
                                      unsigned VGPRPressure) {
-
   Cand.SU = SU;
   Cand.AtTop = AtTop;
 
+  if (!DAG->isTrackingPressure())
+    return;
+
   // getDownwardPressure() and getUpwardPressure() make temporary changes to
   // the tracker, so we need to pass those function a non-const copy.
   RegPressureTracker &TempTracker = const_cast<RegPressureTracker&>(RPTracker);
 
   Pressure.clear();
   MaxPressure.clear();
 
   if (AtTop)

 // This function is mostly cut and pasted from
 // GenericScheduler::pickNodeFromQueue()
 void GCNMaxOccupancySchedStrategy::pickNodeFromQueue(SchedBoundary &Zone,
                                          const CandPolicy &ZonePolicy,
                                          const RegPressureTracker &RPTracker,
                                          SchedCandidate &Cand) {
   const SIRegisterInfo *SRI = static_cast<const SIRegisterInfo*>(TRI);
   ArrayRef<unsigned> Pressure = RPTracker.getRegSetPressureAtPos();
-  unsigned SGPRPressure = Pressure[AMDGPU::RegisterPressureSets::SReg_32];
-  unsigned VGPRPressure = Pressure[AMDGPU::RegisterPressureSets::VGPR_32];
+  unsigned SGPRPressure = 0;
+  unsigned VGPRPressure = 0;
+  if (DAG->isTrackingPressure()) {
+    SGPRPressure = Pressure[AMDGPU::RegisterPressureSets::SReg_32];
+    VGPRPressure = Pressure[AMDGPU::RegisterPressureSets::VGPR_32];
+  }
   ReadyQueue &Q = Zone.Available;
   for (SUnit *SU : Q) {
 
     SchedCandidate TryCand(ZonePolicy);
     initCandidate(TryCand, SU, Zone.isTop(), RPTracker, SRI,
                   SGPRPressure, VGPRPressure);
     // Pass SchedBoundary only when comparing nodes from the same boundary.
     SchedBoundary *ZoneArg = Cand.AtTop == TryCand.AtTop ? &Zone : nullptr;

llvm/test/CodeGen/AMDGPU/schedule-regpressure.mir

 # RUN: llc -march=amdgcn -misched=converge -run-pass machine-scheduler %s -o - -debug-only=machine-scheduler 2>&1 | FileCheck %s
+# RUN: llc -march=amdgcn -misched-regpressure=false -run-pass machine-scheduler %s -o - -debug-only=machine-scheduler 2>&1 | FileCheck %s --check-prefix=NORP
+
 # REQUIRES: asserts
 
 # Check there is no SReg_32 pressure created by DS_* instructions because of M0 use
 
 # CHECK: ScheduleDAGMILive::schedule starting
 # CHECK: SU({{.*}} = DS_READ_B32 {{.*}} implicit $m0, implicit $exec
 # CHECK: Pressure Diff : {{$}}
 # CHECK: SU({{.*}} DS_WRITE_B32
 
+# NORP: ScheduleDAGMILive::schedule starting
+# NORP: GenericScheduler RegionPolicy:  ShouldTrackPressure=0
+# NORP-NOT: Pressure Diff : {{$}}
+# NORP-NOT: Pressure Diff : {{$}}
+# NORP-NOT: Pressure Diff : {{$}}
+# NORP-NOT: Pressure Diff : {{$}}
+# NORP-NOT: Pressure Diff : {{$}}
+# NORP-NOT: Pressure Diff : {{$}}
+# NORP-NOT: Bottom Pressure:
+# NORP-NOT: UpdateRegP
+# NORP-NOT: UpdateRegP
+# NORP-NOT: UpdateRegP
+
 ---
 name:            mo_pset
 alignment:       1
 exposesReturnsTwice: false
 legalized:       false
 regBankSelected: false
 selected:        false
 tracksRegLiveness: true