LLVM 20.0.0git
MicroOpQueueStage.h
Go to the documentation of this file.
1//===---------------------- MicroOpQueueStage.h -----------------*- 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/// \file
9///
10/// This file defines a stage that implements a queue of micro opcodes.
11/// It can be used to simulate a hardware micro-op queue that serves opcodes to
12/// the out of order backend.
13///
14//===----------------------------------------------------------------------===//
15
16#ifndef LLVM_MCA_STAGES_MICROOPQUEUESTAGE_H
17#define LLVM_MCA_STAGES_MICROOPQUEUESTAGE_H
18
21
22namespace llvm {
23namespace mca {
24
25/// A stage that simulates a queue of instruction opcodes.
26class MicroOpQueueStage : public Stage {
28 unsigned NextAvailableSlotIdx;
29 unsigned CurrentInstructionSlotIdx;
30
31 // Limits the number of instructions that can be written to this buffer every
32 // cycle. A value of zero means that there is no limit to the instruction
33 // throughput in input.
34 const unsigned MaxIPC;
35 unsigned CurrentIPC;
36
37 // Number of entries that are available during this cycle.
38 unsigned AvailableEntries;
39
40 // True if instructions dispatched to this stage don't need to wait for the
41 // next cycle before moving to the next stage.
42 // False if this buffer acts as a one cycle delay in the execution pipeline.
43 bool IsZeroLatencyStage;
44
46 MicroOpQueueStage &operator=(const MicroOpQueueStage &Other) = delete;
47
48 // By default, an instruction consumes a number of buffer entries equal to its
49 // number of micro opcodes (see field `InstrDesc::NumMicroOpcodes`). The
50 // number of entries consumed by an instruction is normalized to the
51 // minimum value between NumMicroOpcodes and the buffer size. This is to avoid
52 // problems with (microcoded) instructions that generate a number of micro
53 // opcodes than doesn't fit in the buffer.
54 unsigned getNormalizedOpcodes(const InstRef &IR) const {
55 unsigned NormalizedOpcodes =
56 std::min(static_cast<unsigned>(Buffer.size()),
57 IR.getInstruction()->getDesc().NumMicroOps);
58 return NormalizedOpcodes ? NormalizedOpcodes : 1U;
59 }
60
61 Error moveInstructions();
62
63public:
64 MicroOpQueueStage(unsigned Size, unsigned IPC = 0,
65 bool ZeroLatencyStage = true);
66
67 bool isAvailable(const InstRef &IR) const override {
68 if (MaxIPC && CurrentIPC == MaxIPC)
69 return false;
70 unsigned NormalizedOpcodes = getNormalizedOpcodes(IR);
71 if (NormalizedOpcodes > AvailableEntries)
72 return false;
73 return true;
74 }
75
76 bool hasWorkToComplete() const override {
77 return AvailableEntries != Buffer.size();
78 }
79
80 Error execute(InstRef &IR) override;
81 Error cycleStart() override;
82 Error cycleEnd() override;
83};
84
85} // namespace mca
86} // namespace llvm
87
88#endif // LLVM_MCA_STAGES_MICROOPQUEUESTAGE_H
uint64_t Size
Legalize the Machine IR a function s Machine IR
Definition: Legalizer.cpp:80
This file defines the SmallVector class.
This file defines a stage.
Lightweight error class with error context and mandatory checking.
Definition: Error.h:160
size_t size() const
Definition: SmallVector.h:78
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
Definition: SmallVector.h:1196
An InstRef contains both a SourceMgr index and Instruction pair.
Definition: Instruction.h:720
A stage that simulates a queue of instruction opcodes.
bool isAvailable(const InstRef &IR) const override
Returns true if it can execute IR during this cycle.
Error cycleEnd() override
Called once at the end of each cycle.
bool hasWorkToComplete() const override
Returns true if some instructions are still executing this stage.
Error execute(InstRef &IR) override
The primary action that this stage performs on instruction IR.
Error cycleStart() override
Called once at the start of each cycle.
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
@ Other
Any other memory.