Mercurial > hg > Members > tobaru > cbc > CbC_llvm
comparison lib/Target/PowerPC/PPCTOCRegDeps.cpp @ 95:afa8332a0e37
LLVM 3.8
author | Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp> |
---|---|
date | Tue, 13 Oct 2015 17:48:58 +0900 |
parents | |
children | 1172e4bd9c6f |
comparison
equal
deleted
inserted
replaced
84:f3e34b893a5f | 95:afa8332a0e37 |
---|---|
1 //===-- PPCTOCRegDeps.cpp - Add Extra TOC Register Dependencies -----------===// | |
2 // | |
3 // The LLVM Compiler Infrastructure | |
4 // | |
5 // This file is distributed under the University of Illinois Open Source | |
6 // License. See LICENSE.TXT for details. | |
7 // | |
8 //===----------------------------------------------------------------------===// | |
9 // | |
10 // When resolving an address using the ELF ABI TOC pointer, two relocations are | |
11 // generally required: one for the high part and one for the low part. Only | |
12 // the high part generally explicitly depends on r2 (the TOC pointer). And, so, | |
13 // we might produce code like this: | |
14 // | |
15 // .Ltmp526: | |
16 // addis 3, 2, .LC12@toc@ha | |
17 // .Ltmp1628: | |
18 // std 2, 40(1) | |
19 // ld 5, 0(27) | |
20 // ld 2, 8(27) | |
21 // ld 11, 16(27) | |
22 // ld 3, .LC12@toc@l(3) | |
23 // rldicl 4, 4, 0, 32 | |
24 // mtctr 5 | |
25 // bctrl | |
26 // ld 2, 40(1) | |
27 // | |
28 // And there is nothing wrong with this code, as such, but there is a linker bug | |
29 // in binutils (https://sourceware.org/bugzilla/show_bug.cgi?id=18414) that will | |
30 // misoptimize this code sequence to this: | |
31 // nop | |
32 // std r2,40(r1) | |
33 // ld r5,0(r27) | |
34 // ld r2,8(r27) | |
35 // ld r11,16(r27) | |
36 // ld r3,-32472(r2) | |
37 // clrldi r4,r4,32 | |
38 // mtctr r5 | |
39 // bctrl | |
40 // ld r2,40(r1) | |
41 // because the linker does not know (and does not check) that the value in r2 | |
42 // changed in between the instruction using the .LC12@toc@ha (TOC-relative) | |
43 // relocation and the instruction using the .LC12@toc@l(3) relocation. | |
44 // Because it finds these instructions using the relocations (and not by | |
45 // scanning the instructions), it has been asserted that there is no good way | |
46 // to detect the change of r2 in between. As a result, this bug may never be | |
47 // fixed (i.e. it may become part of the definition of the ABI). GCC was | |
48 // updated to add extra dependencies on r2 to instructions using the @toc@l | |
49 // relocations to avoid this problem, and we'll do the same here. | |
50 // | |
51 // This is done as a separate pass because: | |
52 // 1. These extra r2 dependencies are not really properties of the | |
53 // instructions, but rather due to a linker bug, and maybe one day we'll be | |
54 // able to get rid of them when targeting linkers without this bug (and, | |
55 // thus, keeping the logic centralized here will make that | |
56 // straightforward). | |
57 // 2. There are ISel-level peephole optimizations that propagate the @toc@l | |
58 // relocations to some user instructions, and so the exta dependencies do | |
59 // not apply only to a fixed set of instructions (without undesirable | |
60 // definition replication). | |
61 // | |
62 //===----------------------------------------------------------------------===// | |
63 | |
64 #include "PPCInstrInfo.h" | |
65 #include "MCTargetDesc/PPCPredicates.h" | |
66 #include "PPC.h" | |
67 #include "PPCInstrBuilder.h" | |
68 #include "PPCMachineFunctionInfo.h" | |
69 #include "PPCTargetMachine.h" | |
70 #include "llvm/ADT/STLExtras.h" | |
71 #include "llvm/ADT/Statistic.h" | |
72 #include "llvm/CodeGen/MachineFrameInfo.h" | |
73 #include "llvm/CodeGen/MachineFunctionPass.h" | |
74 #include "llvm/CodeGen/MachineInstr.h" | |
75 #include "llvm/CodeGen/MachineRegisterInfo.h" | |
76 #include "llvm/MC/MCAsmInfo.h" | |
77 #include "llvm/Support/CommandLine.h" | |
78 #include "llvm/Support/Debug.h" | |
79 #include "llvm/Support/ErrorHandling.h" | |
80 #include "llvm/Support/TargetRegistry.h" | |
81 #include "llvm/Support/raw_ostream.h" | |
82 | |
83 using namespace llvm; | |
84 | |
85 #define DEBUG_TYPE "ppc-toc-reg-deps" | |
86 | |
87 namespace llvm { | |
88 void initializePPCTOCRegDepsPass(PassRegistry&); | |
89 } | |
90 | |
91 namespace { | |
92 // PPCTOCRegDeps pass - For simple functions without epilogue code, move | |
93 // returns up, and create conditional returns, to avoid unnecessary | |
94 // branch-to-blr sequences. | |
95 struct PPCTOCRegDeps : public MachineFunctionPass { | |
96 static char ID; | |
97 PPCTOCRegDeps() : MachineFunctionPass(ID) { | |
98 initializePPCTOCRegDepsPass(*PassRegistry::getPassRegistry()); | |
99 } | |
100 | |
101 protected: | |
102 bool hasTOCLoReloc(const MachineInstr &MI) { | |
103 if (MI.getOpcode() == PPC::LDtocL || | |
104 MI.getOpcode() == PPC::ADDItocL) | |
105 return true; | |
106 | |
107 for (const MachineOperand &MO : MI.operands()) { | |
108 if ((MO.getTargetFlags() & PPCII::MO_ACCESS_MASK) == PPCII::MO_TOC_LO) | |
109 return true; | |
110 } | |
111 | |
112 return false; | |
113 } | |
114 | |
115 bool processBlock(MachineBasicBlock &MBB) { | |
116 bool Changed = false; | |
117 | |
118 for (auto &MI : MBB) { | |
119 if (!hasTOCLoReloc(MI)) | |
120 continue; | |
121 | |
122 MI.addOperand(MachineOperand::CreateReg(PPC::X2, | |
123 false /*IsDef*/, | |
124 true /*IsImp*/)); | |
125 Changed = true; | |
126 } | |
127 | |
128 return Changed; | |
129 } | |
130 | |
131 public: | |
132 bool runOnMachineFunction(MachineFunction &MF) override { | |
133 bool Changed = false; | |
134 | |
135 for (MachineFunction::iterator I = MF.begin(); I != MF.end();) { | |
136 MachineBasicBlock &B = *I++; | |
137 if (processBlock(B)) | |
138 Changed = true; | |
139 } | |
140 | |
141 return Changed; | |
142 } | |
143 | |
144 void getAnalysisUsage(AnalysisUsage &AU) const override { | |
145 MachineFunctionPass::getAnalysisUsage(AU); | |
146 } | |
147 }; | |
148 } | |
149 | |
150 INITIALIZE_PASS(PPCTOCRegDeps, DEBUG_TYPE, | |
151 "PowerPC TOC Register Dependencies", false, false) | |
152 | |
153 char PPCTOCRegDeps::ID = 0; | |
154 FunctionPass* | |
155 llvm::createPPCTOCRegDepsPass() { return new PPCTOCRegDeps(); } | |
156 |