AMDGPU: Mark test as XFAIL in expensive_checks builds
[llvm-project.git] / llvm / lib / Target / PowerPC / PPCMachineScheduler.cpp
blob03712879f7c4985729b66bff07d9ef2f2d95b28f
1 //===- PPCMachineScheduler.cpp - MI Scheduler for PowerPC -------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
9 #include "PPCMachineScheduler.h"
10 #include "MCTargetDesc/PPCMCTargetDesc.h"
12 using namespace llvm;
14 static cl::opt<bool>
15 DisableAddiLoadHeuristic("disable-ppc-sched-addi-load",
16 cl::desc("Disable scheduling addi instruction before"
17 "load for ppc"), cl::Hidden);
18 static cl::opt<bool>
19 EnableAddiHeuristic("ppc-postra-bias-addi",
20 cl::desc("Enable scheduling addi instruction as early"
21 "as possible post ra"),
22 cl::Hidden, cl::init(true));
24 static bool isADDIInstr(const GenericScheduler::SchedCandidate &Cand) {
25 return Cand.SU->getInstr()->getOpcode() == PPC::ADDI ||
26 Cand.SU->getInstr()->getOpcode() == PPC::ADDI8;
29 bool PPCPreRASchedStrategy::biasAddiLoadCandidate(SchedCandidate &Cand,
30 SchedCandidate &TryCand,
31 SchedBoundary &Zone) const {
32 if (DisableAddiLoadHeuristic)
33 return false;
35 SchedCandidate &FirstCand = Zone.isTop() ? TryCand : Cand;
36 SchedCandidate &SecondCand = Zone.isTop() ? Cand : TryCand;
37 if (isADDIInstr(FirstCand) && SecondCand.SU->getInstr()->mayLoad()) {
38 TryCand.Reason = Stall;
39 return true;
41 if (FirstCand.SU->getInstr()->mayLoad() && isADDIInstr(SecondCand)) {
42 TryCand.Reason = NoCand;
43 return true;
46 return false;
49 bool PPCPreRASchedStrategy::tryCandidate(SchedCandidate &Cand,
50 SchedCandidate &TryCand,
51 SchedBoundary *Zone) const {
52 // From GenericScheduler::tryCandidate
54 // Initialize the candidate if needed.
55 if (!Cand.isValid()) {
56 TryCand.Reason = NodeOrder;
57 return true;
60 // Bias PhysReg Defs and copies to their uses and defined respectively.
61 if (tryGreater(biasPhysReg(TryCand.SU, TryCand.AtTop),
62 biasPhysReg(Cand.SU, Cand.AtTop), TryCand, Cand, PhysReg))
63 return TryCand.Reason != NoCand;
65 // Avoid exceeding the target's limit.
66 if (DAG->isTrackingPressure() &&
67 tryPressure(TryCand.RPDelta.Excess, Cand.RPDelta.Excess, TryCand, Cand,
68 RegExcess, TRI, DAG->MF))
69 return TryCand.Reason != NoCand;
71 // Avoid increasing the max critical pressure in the scheduled region.
72 if (DAG->isTrackingPressure() &&
73 tryPressure(TryCand.RPDelta.CriticalMax, Cand.RPDelta.CriticalMax,
74 TryCand, Cand, RegCritical, TRI, DAG->MF))
75 return TryCand.Reason != NoCand;
77 // We only compare a subset of features when comparing nodes between
78 // Top and Bottom boundary. Some properties are simply incomparable, in many
79 // other instances we should only override the other boundary if something
80 // is a clear good pick on one boundary. Skip heuristics that are more
81 // "tie-breaking" in nature.
82 bool SameBoundary = Zone != nullptr;
83 if (SameBoundary) {
84 // For loops that are acyclic path limited, aggressively schedule for
85 // latency. Within an single cycle, whenever CurrMOps > 0, allow normal
86 // heuristics to take precedence.
87 if (Rem.IsAcyclicLatencyLimited && !Zone->getCurrMOps() &&
88 tryLatency(TryCand, Cand, *Zone))
89 return TryCand.Reason != NoCand;
91 // Prioritize instructions that read unbuffered resources by stall cycles.
92 if (tryLess(Zone->getLatencyStallCycles(TryCand.SU),
93 Zone->getLatencyStallCycles(Cand.SU), TryCand, Cand, Stall))
94 return TryCand.Reason != NoCand;
97 // Keep clustered nodes together to encourage downstream peephole
98 // optimizations which may reduce resource requirements.
100 // This is a best effort to set things up for a post-RA pass. Optimizations
101 // like generating loads of multiple registers should ideally be done within
102 // the scheduler pass by combining the loads during DAG postprocessing.
103 const SUnit *CandNextClusterSU =
104 Cand.AtTop ? DAG->getNextClusterSucc() : DAG->getNextClusterPred();
105 const SUnit *TryCandNextClusterSU =
106 TryCand.AtTop ? DAG->getNextClusterSucc() : DAG->getNextClusterPred();
107 if (tryGreater(TryCand.SU == TryCandNextClusterSU,
108 Cand.SU == CandNextClusterSU, TryCand, Cand, Cluster))
109 return TryCand.Reason != NoCand;
111 if (SameBoundary) {
112 // Weak edges are for clustering and other constraints.
113 if (tryLess(getWeakLeft(TryCand.SU, TryCand.AtTop),
114 getWeakLeft(Cand.SU, Cand.AtTop), TryCand, Cand, Weak))
115 return TryCand.Reason != NoCand;
118 // Avoid increasing the max pressure of the entire region.
119 if (DAG->isTrackingPressure() &&
120 tryPressure(TryCand.RPDelta.CurrentMax, Cand.RPDelta.CurrentMax, TryCand,
121 Cand, RegMax, TRI, DAG->MF))
122 return TryCand.Reason != NoCand;
124 if (SameBoundary) {
125 // Avoid critical resource consumption and balance the schedule.
126 TryCand.initResourceDelta(DAG, SchedModel);
127 if (tryLess(TryCand.ResDelta.CritResources, Cand.ResDelta.CritResources,
128 TryCand, Cand, ResourceReduce))
129 return TryCand.Reason != NoCand;
130 if (tryGreater(TryCand.ResDelta.DemandedResources,
131 Cand.ResDelta.DemandedResources, TryCand, Cand,
132 ResourceDemand))
133 return TryCand.Reason != NoCand;
135 // Avoid serializing long latency dependence chains.
136 // For acyclic path limited loops, latency was already checked above.
137 if (!RegionPolicy.DisableLatencyHeuristic && TryCand.Policy.ReduceLatency &&
138 !Rem.IsAcyclicLatencyLimited && tryLatency(TryCand, Cand, *Zone))
139 return TryCand.Reason != NoCand;
141 // Fall through to original instruction order.
142 if ((Zone->isTop() && TryCand.SU->NodeNum < Cand.SU->NodeNum) ||
143 (!Zone->isTop() && TryCand.SU->NodeNum > Cand.SU->NodeNum)) {
144 TryCand.Reason = NodeOrder;
148 // GenericScheduler::tryCandidate end
150 // Add powerpc specific heuristic only when TryCand isn't selected or
151 // selected as node order.
152 if (TryCand.Reason != NodeOrder && TryCand.Reason != NoCand)
153 return true;
155 // There are some benefits to schedule the ADDI before the load to hide the
156 // latency, as RA may create a true dependency between the load and addi.
157 if (SameBoundary) {
158 if (biasAddiLoadCandidate(Cand, TryCand, *Zone))
159 return TryCand.Reason != NoCand;
162 return TryCand.Reason != NoCand;
165 bool PPCPostRASchedStrategy::biasAddiCandidate(SchedCandidate &Cand,
166 SchedCandidate &TryCand) const {
167 if (!EnableAddiHeuristic)
168 return false;
170 if (isADDIInstr(TryCand) && !isADDIInstr(Cand)) {
171 TryCand.Reason = Stall;
172 return true;
174 return false;
177 bool PPCPostRASchedStrategy::tryCandidate(SchedCandidate &Cand,
178 SchedCandidate &TryCand) {
179 // From PostGenericScheduler::tryCandidate
181 // Initialize the candidate if needed.
182 if (!Cand.isValid()) {
183 TryCand.Reason = NodeOrder;
184 return true;
187 // Prioritize instructions that read unbuffered resources by stall cycles.
188 if (tryLess(Top.getLatencyStallCycles(TryCand.SU),
189 Top.getLatencyStallCycles(Cand.SU), TryCand, Cand, Stall))
190 return TryCand.Reason != NoCand;
192 // Keep clustered nodes together.
193 if (tryGreater(TryCand.SU == DAG->getNextClusterSucc(),
194 Cand.SU == DAG->getNextClusterSucc(), TryCand, Cand, Cluster))
195 return TryCand.Reason != NoCand;
197 // Avoid critical resource consumption and balance the schedule.
198 if (tryLess(TryCand.ResDelta.CritResources, Cand.ResDelta.CritResources,
199 TryCand, Cand, ResourceReduce))
200 return TryCand.Reason != NoCand;
201 if (tryGreater(TryCand.ResDelta.DemandedResources,
202 Cand.ResDelta.DemandedResources, TryCand, Cand,
203 ResourceDemand))
204 return TryCand.Reason != NoCand;
206 // Avoid serializing long latency dependence chains.
207 if (Cand.Policy.ReduceLatency && tryLatency(TryCand, Cand, Top)) {
208 return TryCand.Reason != NoCand;
211 // Fall through to original instruction order.
212 if (TryCand.SU->NodeNum < Cand.SU->NodeNum)
213 TryCand.Reason = NodeOrder;
215 // PostGenericScheduler::tryCandidate end
217 // Add powerpc post ra specific heuristic only when TryCand isn't selected or
218 // selected as node order.
219 if (TryCand.Reason != NodeOrder && TryCand.Reason != NoCand)
220 return true;
222 // There are some benefits to schedule the ADDI as early as possible post ra
223 // to avoid stalled by vector instructions which take up all the hw units.
224 // And ADDI is usually used to post inc the loop indvar, which matters the
225 // performance.
226 if (biasAddiCandidate(Cand, TryCand))
227 return TryCand.Reason != NoCand;
229 return TryCand.Reason != NoCand;
232 void PPCPostRASchedStrategy::enterMBB(MachineBasicBlock *MBB) {
233 // Custom PPC PostRA specific behavior here.
234 PostGenericScheduler::enterMBB(MBB);
237 void PPCPostRASchedStrategy::leaveMBB() {
238 // Custom PPC PostRA specific behavior here.
239 PostGenericScheduler::leaveMBB();
242 void PPCPostRASchedStrategy::initialize(ScheduleDAGMI *Dag) {
243 // Custom PPC PostRA specific initialization here.
244 PostGenericScheduler::initialize(Dag);
247 SUnit *PPCPostRASchedStrategy::pickNode(bool &IsTopNode) {
248 // Custom PPC PostRA specific scheduling here.
249 return PostGenericScheduler::pickNode(IsTopNode);