Subversion Repositories QNX 8.QNX8 LLVM/Clang compiler suite

Rev

Details | Last modification | View Log | RSS feed

Rev Author Line No. Line
14 pmbaty 1
//===-- llvm/MC/MCSchedule.h - Scheduling -----------------------*- C++ -*-===//
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
//===----------------------------------------------------------------------===//
8
//
9
// This file defines the classes used to describe a subtarget's machine model
10
// for scheduling and other instruction cost heuristics.
11
//
12
//===----------------------------------------------------------------------===//
13
 
14
#ifndef LLVM_MC_MCSCHEDULE_H
15
#define LLVM_MC_MCSCHEDULE_H
16
 
17
#include "llvm/Config/llvm-config.h"
18
#include "llvm/Support/DataTypes.h"
19
#include <cassert>
20
 
21
namespace llvm {
22
 
23
template <typename T> class ArrayRef;
24
struct InstrItinerary;
25
class MCSubtargetInfo;
26
class MCInstrInfo;
27
class MCInst;
28
class InstrItineraryData;
29
 
30
/// Define a kind of processor resource that will be modeled by the scheduler.
31
struct MCProcResourceDesc {
32
  const char *Name;
33
  unsigned NumUnits; // Number of resource of this kind
34
  unsigned SuperIdx; // Index of the resources kind that contains this kind.
35
 
36
  // Number of resources that may be buffered.
37
  //
38
  // Buffered resources (BufferSize != 0) may be consumed at some indeterminate
39
  // cycle after dispatch. This should be used for out-of-order cpus when
40
  // instructions that use this resource can be buffered in a reservaton
41
  // station.
42
  //
43
  // Unbuffered resources (BufferSize == 0) always consume their resource some
44
  // fixed number of cycles after dispatch. If a resource is unbuffered, then
45
  // the scheduler will avoid scheduling instructions with conflicting resources
46
  // in the same cycle. This is for in-order cpus, or the in-order portion of
47
  // an out-of-order cpus.
48
  int BufferSize;
49
 
50
  // If the resource has sub-units, a pointer to the first element of an array
51
  // of `NumUnits` elements containing the ProcResourceIdx of the sub units.
52
  // nullptr if the resource does not have sub-units.
53
  const unsigned *SubUnitsIdxBegin;
54
 
55
  bool operator==(const MCProcResourceDesc &Other) const {
56
    return NumUnits == Other.NumUnits && SuperIdx == Other.SuperIdx
57
      && BufferSize == Other.BufferSize;
58
  }
59
};
60
 
61
/// Identify one of the processor resource kinds consumed by a particular
62
/// scheduling class for the specified number of cycles.
63
struct MCWriteProcResEntry {
64
  uint16_t ProcResourceIdx;
65
  uint16_t Cycles;
66
 
67
  bool operator==(const MCWriteProcResEntry &Other) const {
68
    return ProcResourceIdx == Other.ProcResourceIdx && Cycles == Other.Cycles;
69
  }
70
};
71
 
72
/// Specify the latency in cpu cycles for a particular scheduling class and def
73
/// index. -1 indicates an invalid latency. Heuristics would typically consider
74
/// an instruction with invalid latency to have infinite latency.  Also identify
75
/// the WriteResources of this def. When the operand expands to a sequence of
76
/// writes, this ID is the last write in the sequence.
77
struct MCWriteLatencyEntry {
78
  int16_t Cycles;
79
  uint16_t WriteResourceID;
80
 
81
  bool operator==(const MCWriteLatencyEntry &Other) const {
82
    return Cycles == Other.Cycles && WriteResourceID == Other.WriteResourceID;
83
  }
84
};
85
 
86
/// Specify the number of cycles allowed after instruction issue before a
87
/// particular use operand reads its registers. This effectively reduces the
88
/// write's latency. Here we allow negative cycles for corner cases where
89
/// latency increases. This rule only applies when the entry's WriteResource
90
/// matches the write's WriteResource.
91
///
92
/// MCReadAdvanceEntries are sorted first by operand index (UseIdx), then by
93
/// WriteResourceIdx.
94
struct MCReadAdvanceEntry {
95
  unsigned UseIdx;
96
  unsigned WriteResourceID;
97
  int Cycles;
98
 
99
  bool operator==(const MCReadAdvanceEntry &Other) const {
100
    return UseIdx == Other.UseIdx && WriteResourceID == Other.WriteResourceID
101
      && Cycles == Other.Cycles;
102
  }
103
};
104
 
105
/// Summarize the scheduling resources required for an instruction of a
106
/// particular scheduling class.
107
///
108
/// Defined as an aggregate struct for creating tables with initializer lists.
109
struct MCSchedClassDesc {
110
  static const unsigned short InvalidNumMicroOps = (1U << 13) - 1;
111
  static const unsigned short VariantNumMicroOps = InvalidNumMicroOps - 1;
112
 
113
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
114
  const char* Name;
115
#endif
116
  uint16_t NumMicroOps : 13;
117
  uint16_t BeginGroup : 1;
118
  uint16_t EndGroup : 1;
119
  uint16_t RetireOOO : 1;
120
  uint16_t WriteProcResIdx; // First index into WriteProcResTable.
121
  uint16_t NumWriteProcResEntries;
122
  uint16_t WriteLatencyIdx; // First index into WriteLatencyTable.
123
  uint16_t NumWriteLatencyEntries;
124
  uint16_t ReadAdvanceIdx; // First index into ReadAdvanceTable.
125
  uint16_t NumReadAdvanceEntries;
126
 
127
  bool isValid() const {
128
    return NumMicroOps != InvalidNumMicroOps;
129
  }
130
  bool isVariant() const {
131
    return NumMicroOps == VariantNumMicroOps;
132
  }
133
};
134
 
135
/// Specify the cost of a register definition in terms of number of physical
136
/// register allocated at register renaming stage. For example, AMD Jaguar.
137
/// natively supports 128-bit data types, and operations on 256-bit registers
138
/// (i.e. YMM registers) are internally split into two COPs (complex operations)
139
/// and each COP updates a physical register. Basically, on Jaguar, a YMM
140
/// register write effectively consumes two physical registers. That means,
141
/// the cost of a YMM write in the BtVer2 model is 2.
142
struct MCRegisterCostEntry {
143
  unsigned RegisterClassID;
144
  unsigned Cost;
145
  bool AllowMoveElimination;
146
};
147
 
148
/// A register file descriptor.
149
///
150
/// This struct allows to describe processor register files. In particular, it
151
/// helps describing the size of the register file, as well as the cost of
152
/// allocating a register file at register renaming stage.
153
/// FIXME: this struct can be extended to provide information about the number
154
/// of read/write ports to the register file.  A value of zero for field
155
/// 'NumPhysRegs' means: this register file has an unbounded number of physical
156
/// registers.
157
struct MCRegisterFileDesc {
158
  const char *Name;
159
  uint16_t NumPhysRegs;
160
  uint16_t NumRegisterCostEntries;
161
  // Index of the first cost entry in MCExtraProcessorInfo::RegisterCostTable.
162
  uint16_t RegisterCostEntryIdx;
163
  // A value of zero means: there is no limit in the number of moves that can be
164
  // eliminated every cycle.
165
  uint16_t MaxMovesEliminatedPerCycle;
166
  // Ture if this register file only knows how to optimize register moves from
167
  // known zero registers.
168
  bool AllowZeroMoveEliminationOnly;
169
};
170
 
171
/// Provide extra details about the machine processor.
172
///
173
/// This is a collection of "optional" processor information that is not
174
/// normally used by the LLVM machine schedulers, but that can be consumed by
175
/// external tools like llvm-mca to improve the quality of the peformance
176
/// analysis.
177
struct MCExtraProcessorInfo {
178
  // Actual size of the reorder buffer in hardware.
179
  unsigned ReorderBufferSize;
180
  // Number of instructions retired per cycle.
181
  unsigned MaxRetirePerCycle;
182
  const MCRegisterFileDesc *RegisterFiles;
183
  unsigned NumRegisterFiles;
184
  const MCRegisterCostEntry *RegisterCostTable;
185
  unsigned NumRegisterCostEntries;
186
  unsigned LoadQueueID;
187
  unsigned StoreQueueID;
188
};
189
 
190
/// Machine model for scheduling, bundling, and heuristics.
191
///
192
/// The machine model directly provides basic information about the
193
/// microarchitecture to the scheduler in the form of properties. It also
194
/// optionally refers to scheduler resource tables and itinerary
195
/// tables. Scheduler resource tables model the latency and cost for each
196
/// instruction type. Itinerary tables are an independent mechanism that
197
/// provides a detailed reservation table describing each cycle of instruction
198
/// execution. Subtargets may define any or all of the above categories of data
199
/// depending on the type of CPU and selected scheduler.
200
///
201
/// The machine independent properties defined here are used by the scheduler as
202
/// an abstract machine model. A real micro-architecture has a number of
203
/// buffers, queues, and stages. Declaring that a given machine-independent
204
/// abstract property corresponds to a specific physical property across all
205
/// subtargets can't be done. Nonetheless, the abstract model is
206
/// useful. Futhermore, subtargets typically extend this model with processor
207
/// specific resources to model any hardware features that can be exploited by
208
/// scheduling heuristics and aren't sufficiently represented in the abstract.
209
///
210
/// The abstract pipeline is built around the notion of an "issue point". This
211
/// is merely a reference point for counting machine cycles. The physical
212
/// machine will have pipeline stages that delay execution. The scheduler does
213
/// not model those delays because they are irrelevant as long as they are
214
/// consistent. Inaccuracies arise when instructions have different execution
215
/// delays relative to each other, in addition to their intrinsic latency. Those
216
/// special cases can be handled by TableGen constructs such as, ReadAdvance,
217
/// which reduces latency when reading data, and ResourceCycles, which consumes
218
/// a processor resource when writing data for a number of abstract
219
/// cycles.
220
///
221
/// TODO: One tool currently missing is the ability to add a delay to
222
/// ResourceCycles. That would be easy to add and would likely cover all cases
223
/// currently handled by the legacy itinerary tables.
224
///
225
/// A note on out-of-order execution and, more generally, instruction
226
/// buffers. Part of the CPU pipeline is always in-order. The issue point, which
227
/// is the point of reference for counting cycles, only makes sense as an
228
/// in-order part of the pipeline. Other parts of the pipeline are sometimes
229
/// falling behind and sometimes catching up. It's only interesting to model
230
/// those other, decoupled parts of the pipeline if they may be predictably
231
/// resource constrained in a way that the scheduler can exploit.
232
///
233
/// The LLVM machine model distinguishes between in-order constraints and
234
/// out-of-order constraints so that the target's scheduling strategy can apply
235
/// appropriate heuristics. For a well-balanced CPU pipeline, out-of-order
236
/// resources would not typically be treated as a hard scheduling
237
/// constraint. For example, in the GenericScheduler, a delay caused by limited
238
/// out-of-order resources is not directly reflected in the number of cycles
239
/// that the scheduler sees between issuing an instruction and its dependent
240
/// instructions. In other words, out-of-order resources don't directly increase
241
/// the latency between pairs of instructions. However, they can still be used
242
/// to detect potential bottlenecks across a sequence of instructions and bias
243
/// the scheduling heuristics appropriately.
244
struct MCSchedModel {
245
  // IssueWidth is the maximum number of instructions that may be scheduled in
246
  // the same per-cycle group. This is meant to be a hard in-order constraint
247
  // (a.k.a. "hazard"). In the GenericScheduler strategy, no more than
248
  // IssueWidth micro-ops can ever be scheduled in a particular cycle.
249
  //
250
  // In practice, IssueWidth is useful to model any bottleneck between the
251
  // decoder (after micro-op expansion) and the out-of-order reservation
252
  // stations or the decoder bandwidth itself. If the total number of
253
  // reservation stations is also a bottleneck, or if any other pipeline stage
254
  // has a bandwidth limitation, then that can be naturally modeled by adding an
255
  // out-of-order processor resource.
256
  unsigned IssueWidth;
257
  static const unsigned DefaultIssueWidth = 1;
258
 
259
  // MicroOpBufferSize is the number of micro-ops that the processor may buffer
260
  // for out-of-order execution.
261
  //
262
  // "0" means operations that are not ready in this cycle are not considered
263
  // for scheduling (they go in the pending queue). Latency is paramount. This
264
  // may be more efficient if many instructions are pending in a schedule.
265
  //
266
  // "1" means all instructions are considered for scheduling regardless of
267
  // whether they are ready in this cycle. Latency still causes issue stalls,
268
  // but we balance those stalls against other heuristics.
269
  //
270
  // "> 1" means the processor is out-of-order. This is a machine independent
271
  // estimate of highly machine specific characteristics such as the register
272
  // renaming pool and reorder buffer.
273
  unsigned MicroOpBufferSize;
274
  static const unsigned DefaultMicroOpBufferSize = 0;
275
 
276
  // LoopMicroOpBufferSize is the number of micro-ops that the processor may
277
  // buffer for optimized loop execution. More generally, this represents the
278
  // optimal number of micro-ops in a loop body. A loop may be partially
279
  // unrolled to bring the count of micro-ops in the loop body closer to this
280
  // number.
281
  unsigned LoopMicroOpBufferSize;
282
  static const unsigned DefaultLoopMicroOpBufferSize = 0;
283
 
284
  // LoadLatency is the expected latency of load instructions.
285
  unsigned LoadLatency;
286
  static const unsigned DefaultLoadLatency = 4;
287
 
288
  // HighLatency is the expected latency of "very high latency" operations.
289
  // See TargetInstrInfo::isHighLatencyDef().
290
  // By default, this is set to an arbitrarily high number of cycles
291
  // likely to have some impact on scheduling heuristics.
292
  unsigned HighLatency;
293
  static const unsigned DefaultHighLatency = 10;
294
 
295
  // MispredictPenalty is the typical number of extra cycles the processor
296
  // takes to recover from a branch misprediction.
297
  unsigned MispredictPenalty;
298
  static const unsigned DefaultMispredictPenalty = 10;
299
 
300
  bool PostRAScheduler; // default value is false
301
 
302
  bool CompleteModel;
303
 
304
  unsigned ProcID;
305
  const MCProcResourceDesc *ProcResourceTable;
306
  const MCSchedClassDesc *SchedClassTable;
307
  unsigned NumProcResourceKinds;
308
  unsigned NumSchedClasses;
309
  // Instruction itinerary tables used by InstrItineraryData.
310
  friend class InstrItineraryData;
311
  const InstrItinerary *InstrItineraries;
312
 
313
  const MCExtraProcessorInfo *ExtraProcessorInfo;
314
 
315
  bool hasExtraProcessorInfo() const { return ExtraProcessorInfo; }
316
 
317
  unsigned getProcessorID() const { return ProcID; }
318
 
319
  /// Does this machine model include instruction-level scheduling.
320
  bool hasInstrSchedModel() const { return SchedClassTable; }
321
 
322
  const MCExtraProcessorInfo &getExtraProcessorInfo() const {
323
    assert(hasExtraProcessorInfo() &&
324
           "No extra information available for this model");
325
    return *ExtraProcessorInfo;
326
  }
327
 
328
  /// Return true if this machine model data for all instructions with a
329
  /// scheduling class (itinerary class or SchedRW list).
330
  bool isComplete() const { return CompleteModel; }
331
 
332
  /// Return true if machine supports out of order execution.
333
  bool isOutOfOrder() const { return MicroOpBufferSize > 1; }
334
 
335
  unsigned getNumProcResourceKinds() const {
336
    return NumProcResourceKinds;
337
  }
338
 
339
  const MCProcResourceDesc *getProcResource(unsigned ProcResourceIdx) const {
340
    assert(hasInstrSchedModel() && "No scheduling machine model");
341
 
342
    assert(ProcResourceIdx < NumProcResourceKinds && "bad proc resource idx");
343
    return &ProcResourceTable[ProcResourceIdx];
344
  }
345
 
346
  const MCSchedClassDesc *getSchedClassDesc(unsigned SchedClassIdx) const {
347
    assert(hasInstrSchedModel() && "No scheduling machine model");
348
 
349
    assert(SchedClassIdx < NumSchedClasses && "bad scheduling class idx");
350
    return &SchedClassTable[SchedClassIdx];
351
  }
352
 
353
  /// Returns the latency value for the scheduling class.
354
  static int computeInstrLatency(const MCSubtargetInfo &STI,
355
                                 const MCSchedClassDesc &SCDesc);
356
 
357
  int computeInstrLatency(const MCSubtargetInfo &STI, unsigned SClass) const;
358
  int computeInstrLatency(const MCSubtargetInfo &STI, const MCInstrInfo &MCII,
359
                          const MCInst &Inst) const;
360
 
361
  // Returns the reciprocal throughput information from a MCSchedClassDesc.
362
  static double
363
  getReciprocalThroughput(const MCSubtargetInfo &STI,
364
                          const MCSchedClassDesc &SCDesc);
365
 
366
  static double
367
  getReciprocalThroughput(unsigned SchedClass, const InstrItineraryData &IID);
368
 
369
  double
370
  getReciprocalThroughput(const MCSubtargetInfo &STI, const MCInstrInfo &MCII,
371
                          const MCInst &Inst) const;
372
 
373
  /// Returns the maximum forwarding delay for register reads dependent on
374
  /// writes of scheduling class WriteResourceIdx.
375
  static unsigned getForwardingDelayCycles(ArrayRef<MCReadAdvanceEntry> Entries,
376
                                           unsigned WriteResourceIdx = 0);
377
 
378
  /// Returns the default initialized model.
379
  static const MCSchedModel &GetDefaultSchedModel() { return Default; }
380
  static const MCSchedModel Default;
381
};
382
 
383
} // namespace llvm
384
 
385
#endif