1 //===-- AArch64CleanupLocalDynamicTLSPass.cpp ---------------------*- C++ -*-=//
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 // Local-dynamic access to thread-local variables proceeds in three stages.
11 //
12 // 1. The offset of this Module's thread-local area from TPIDR_EL0 is calculated
13 //    in much the same way as a general-dynamic TLS-descriptor access against
14 //    the special symbol _TLS_MODULE_BASE.
15 // 2. The variable's offset from _TLS_MODULE_BASE_ is calculated using
16 //    instructions with "dtprel" modifiers.
17 // 3. These two are added, together with TPIDR_EL0, to obtain the variable's
18 //    true address.
19 //
20 // This is only better than general-dynamic access to the variable if two or
21 // more of the first stage TLS-descriptor calculations can be combined. This
22 // pass looks through a function and performs such combinations.
23 //
24 //===----------------------------------------------------------------------===//
25 #include "AArch64.h"
26 #include "AArch64InstrInfo.h"
27 #include "AArch64MachineFunctionInfo.h"
28 #include "AArch64TargetMachine.h"
29 #include "llvm/CodeGen/MachineDominators.h"
30 #include "llvm/CodeGen/MachineFunction.h"
31 #include "llvm/CodeGen/MachineFunctionPass.h"
32 #include "llvm/CodeGen/MachineInstrBuilder.h"
33 #include "llvm/CodeGen/MachineRegisterInfo.h"
34 using namespace llvm;
35 
36 #define TLSCLEANUP_PASS_NAME "AArch64 Local Dynamic TLS Access Clean-up"
37 
38 namespace {
39 struct LDTLSCleanup : public MachineFunctionPass {
40   static char ID;
41   LDTLSCleanup() : MachineFunctionPass(ID) {
42     initializeLDTLSCleanupPass(*PassRegistry::getPassRegistry());
43   }
44 
45   bool runOnMachineFunction(MachineFunction &MF) override {
46     if (skipFunction(*MF.getFunction()))
47       return false;
48 
49     AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
50     if (AFI->getNumLocalDynamicTLSAccesses() < 2) {
51       // No point folding accesses if there isn't at least two.
52       return false;
53     }
54 
55     MachineDominatorTree *DT = &getAnalysis<MachineDominatorTree>();
56     return VisitNode(DT->getRootNode(), 0);
57   }
58 
59   // Visit the dominator subtree rooted at Node in pre-order.
60   // If TLSBaseAddrReg is non-null, then use that to replace any
61   // TLS_base_addr instructions. Otherwise, create the register
62   // when the first such instruction is seen, and then use it
63   // as we encounter more instructions.
64   bool VisitNode(MachineDomTreeNode *Node, unsigned TLSBaseAddrReg) {
65     MachineBasicBlock *BB = Node->getBlock();
66     bool Changed = false;
67 
68     // Traverse the current block.
69     for (MachineBasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;
70          ++I) {
71       switch (I->getOpcode()) {
72       case AArch64::TLSDESC_CALLSEQ:
73         // Make sure it's a local dynamic access.
74         if (!I->getOperand(0).isSymbol() ||
75             strcmp(I->getOperand(0).getSymbolName(), "_TLS_MODULE_BASE_"))
76           break;
77 
78         if (TLSBaseAddrReg)
79           I = replaceTLSBaseAddrCall(*I, TLSBaseAddrReg);
80         else
81           I = setRegister(*I, &TLSBaseAddrReg);
82         Changed = true;
83         break;
84       default:
85         break;
86       }
87     }
88 
89     // Visit the children of this block in the dominator tree.
90     for (MachineDomTreeNode *N : *Node) {
91       Changed |= VisitNode(N, TLSBaseAddrReg);
92     }
93 
94     return Changed;
95   }
96 
97   // Replace the TLS_base_addr instruction I with a copy from
98   // TLSBaseAddrReg, returning the new instruction.
99   MachineInstr *replaceTLSBaseAddrCall(MachineInstr &I,
100                                        unsigned TLSBaseAddrReg) {
101     MachineFunction *MF = I.getParent()->getParent();
102     const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo();
103 
104     // Insert a Copy from TLSBaseAddrReg to x0, which is where the rest of the
105     // code sequence assumes the address will be.
106     MachineInstr *Copy = BuildMI(*I.getParent(), I, I.getDebugLoc(),
107                                  TII->get(TargetOpcode::COPY), AArch64::X0)
108                              .addReg(TLSBaseAddrReg);
109 
110     // Erase the TLS_base_addr instruction.
111     I.eraseFromParent();
112 
113     return Copy;
114   }
115 
116   // Create a virtal register in *TLSBaseAddrReg, and populate it by
117   // inserting a copy instruction after I. Returns the new instruction.
118   MachineInstr *setRegister(MachineInstr &I, unsigned *TLSBaseAddrReg) {
119     MachineFunction *MF = I.getParent()->getParent();
120     const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo();
121 
122     // Create a virtual register for the TLS base address.
123     MachineRegisterInfo &RegInfo = MF->getRegInfo();
124     *TLSBaseAddrReg = RegInfo.createVirtualRegister(&AArch64::GPR64RegClass);
125 
126     // Insert a copy from X0 to TLSBaseAddrReg for later.
127     MachineInstr *Copy =
128         BuildMI(*I.getParent(), ++I.getIterator(), I.getDebugLoc(),
129                 TII->get(TargetOpcode::COPY), *TLSBaseAddrReg)
130             .addReg(AArch64::X0);
131 
132     return Copy;
133   }
134 
135   StringRef getPassName() const override { return TLSCLEANUP_PASS_NAME; }
136 
137   void getAnalysisUsage(AnalysisUsage &AU) const override {
138     AU.setPreservesCFG();
139     AU.addRequired<MachineDominatorTree>();
140     MachineFunctionPass::getAnalysisUsage(AU);
141   }
142 };
143 }
144 
145 INITIALIZE_PASS(LDTLSCleanup, "aarch64-local-dynamic-tls-cleanup",
146                 TLSCLEANUP_PASS_NAME, false, false)
147 
148 char LDTLSCleanup::ID = 0;
149 FunctionPass *llvm::createAArch64CleanupLocalDynamicTLSPass() {
150   return new LDTLSCleanup();
151 }
152