1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
| //===- llvm/CodeGen/TargetSchedule.h - Sched Machine Model ------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file defines a wrapper around MCSchedModel that allows the interface to
// benefit from information currently only available in TargetInstrInfo.
// Ideally, the scheduling interface would be fully defined in the MC layer.
//
//===----------------------------------------------------------------------===//
#ifndef LLVM_CODEGEN_TARGETSCHEDULE_H
#define LLVM_CODEGEN_TARGETSCHEDULE_H
#include "llvm/ADT/Optional.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/Config/llvm-config.h"
#include "llvm/MC/MCInstrItineraries.h"
#include "llvm/MC/MCSchedule.h"
namespace llvm {
class MachineInstr;
class TargetInstrInfo;
/// Provide an instruction scheduling machine model to CodeGen passes.
class TargetSchedModel {
// For efficiency, hold a copy of the statically defined MCSchedModel for this
// processor.
MCSchedModel SchedModel;
InstrItineraryData InstrItins;
const TargetSubtargetInfo *STI = nullptr;
const TargetInstrInfo *TII = nullptr;
SmallVector<unsigned, 16> ResourceFactors;
unsigned MicroOpFactor; // Multiply to normalize microops to resource units.
unsigned ResourceLCM; // Resource units per cycle. Latency normalization factor.
unsigned computeInstrLatency(const MCSchedClassDesc &SCDesc) const;
public:
TargetSchedModel() : SchedModel(MCSchedModel::GetDefaultSchedModel()) {}
/// Initialize the machine model for instruction scheduling.
///
/// The machine model API keeps a copy of the top-level MCSchedModel table
/// indices and may query TargetSubtargetInfo and TargetInstrInfo to resolve
/// dynamic properties.
void init(const TargetSubtargetInfo *TSInfo);
/// Return the MCSchedClassDesc for this instruction.
const MCSchedClassDesc *resolveSchedClass(const MachineInstr *MI) const;
/// TargetSubtargetInfo getter.
const TargetSubtargetInfo *getSubtargetInfo() const { return STI; }
/// TargetInstrInfo getter.
const TargetInstrInfo *getInstrInfo() const { return TII; }
/// Return true if this machine model includes an instruction-level
/// scheduling model.
///
/// This is more detailed than the course grain IssueWidth and default
/// latency properties, but separate from the per-cycle itinerary data.
bool hasInstrSchedModel() const;
const MCSchedModel *getMCSchedModel() const { return &SchedModel; }
/// Return true if this machine model includes cycle-to-cycle itinerary
/// data.
///
/// This models scheduling at each stage in the processor pipeline.
bool hasInstrItineraries() const;
const InstrItineraryData *getInstrItineraries() const {
if (hasInstrItineraries())
return &InstrItins;
return nullptr;
}
/// Return true if this machine model includes an instruction-level
/// scheduling model or cycle-to-cycle itinerary data.
bool hasInstrSchedModelOrItineraries() const {
return hasInstrSchedModel() || hasInstrItineraries();
}
/// Identify the processor corresponding to the current subtarget.
unsigned getProcessorID() const { return SchedModel.getProcessorID(); }
/// Maximum number of micro-ops that may be scheduled per cycle.
unsigned getIssueWidth() const { return SchedModel.IssueWidth; }
/// Return true if new group must begin.
bool mustBeginGroup(const MachineInstr *MI,
const MCSchedClassDesc *SC = nullptr) const;
/// Return true if current group must end.
bool mustEndGroup(const MachineInstr *MI,
const MCSchedClassDesc *SC = nullptr) const;
/// Return the number of issue slots required for this MI.
unsigned getNumMicroOps(const MachineInstr *MI,
const MCSchedClassDesc *SC = nullptr) const;
/// Get the number of kinds of resources for this target.
unsigned getNumProcResourceKinds() const {
return SchedModel.getNumProcResourceKinds();
}
/// Get a processor resource by ID for convenience.
const MCProcResourceDesc *getProcResource(unsigned PIdx) const {
return SchedModel.getProcResource(PIdx);
}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
const char *getResourceName(unsigned PIdx) const {
if (!PIdx)
return "MOps";
return SchedModel.getProcResource(PIdx)->Name;
}
#endif
using ProcResIter = const MCWriteProcResEntry *;
// Get an iterator into the processor resources consumed by this
// scheduling class.
ProcResIter getWriteProcResBegin(const MCSchedClassDesc *SC) const {
// The subtarget holds a single resource table for all processors.
return STI->getWriteProcResBegin(SC);
}
ProcResIter getWriteProcResEnd(const MCSchedClassDesc *SC) const {
return STI->getWriteProcResEnd(SC);
}
/// Multiply the number of units consumed for a resource by this factor
/// to normalize it relative to other resources.
unsigned getResourceFactor(unsigned ResIdx) const {
return ResourceFactors[ResIdx];
}
/// Multiply number of micro-ops by this factor to normalize it
/// relative to other resources.
unsigned getMicroOpFactor() const {
return MicroOpFactor;
}
/// Multiply cycle count by this factor to normalize it relative to
/// other resources. This is the number of resource units per cycle.
unsigned getLatencyFactor() const {
return ResourceLCM;
}
/// Number of micro-ops that may be buffered for OOO execution.
unsigned getMicroOpBufferSize() const { return SchedModel.MicroOpBufferSize; }
/// Number of resource units that may be buffered for OOO execution.
/// \return The buffer size in resource units or -1 for unlimited.
int getResourceBufferSize(unsigned PIdx) const {
return SchedModel.getProcResource(PIdx)->BufferSize;
}
/// Compute operand latency based on the available machine model.
///
/// Compute and return the latency of the given data dependent def and use
/// when the operand indices are already known. UseMI may be NULL for an
/// unknown user.
unsigned computeOperandLatency(const MachineInstr *DefMI, unsigned DefOperIdx,
const MachineInstr *UseMI, unsigned UseOperIdx)
const;
/// Compute the instruction latency based on the available machine
/// model.
///
/// Compute and return the expected latency of this instruction independent of
/// a particular use. computeOperandLatency is the preferred API, but this is
/// occasionally useful to help estimate instruction cost.
///
/// If UseDefaultDefLatency is false and no new machine sched model is
/// present this method falls back to TII->getInstrLatency with an empty
/// instruction itinerary (this is so we preserve the previous behavior of the
/// if converter after moving it to TargetSchedModel).
unsigned computeInstrLatency(const MachineInstr *MI,
bool UseDefaultDefLatency = true) const;
unsigned computeInstrLatency(const MCInst &Inst) const;
unsigned computeInstrLatency(unsigned Opcode) const;
/// Output dependency latency of a pair of defs of the same register.
///
/// This is typically one cycle.
unsigned computeOutputLatency(const MachineInstr *DefMI, unsigned DefOperIdx,
const MachineInstr *DepMI) const;
/// Compute the reciprocal throughput of the given instruction.
double computeReciprocalThroughput(const MachineInstr *MI) const;
double computeReciprocalThroughput(const MCInst &MI) const;
double computeReciprocalThroughput(unsigned Opcode) const;
};
} // end namespace llvm
#endif // LLVM_CODEGEN_TARGETSCHEDULE_H
|