1 //===-- XCoreLowerThreadLocal - Lower thread local variables --------------===//
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 /// \file
10 /// This file contains a pass that lowers thread local variables on the
11 ///        XCore.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "XCore.h"
16 #include "llvm/IR/Constants.h"
17 #include "llvm/IR/DerivedTypes.h"
18 #include "llvm/IR/GlobalVariable.h"
19 #include "llvm/IR/IRBuilder.h"
20 #include "llvm/IR/Intrinsics.h"
21 #include "llvm/IR/IntrinsicsXCore.h"
22 #include "llvm/IR/Module.h"
23 #include "llvm/IR/NoFolder.h"
24 #include "llvm/IR/ReplaceConstant.h"
25 #include "llvm/IR/ValueHandle.h"
26 #include "llvm/Pass.h"
27 #include "llvm/Support/CommandLine.h"
28 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
29 
30 #define DEBUG_TYPE "xcore-lower-thread-local"
31 
32 using namespace llvm;
33 
34 static cl::opt<unsigned> MaxThreads(
35   "xcore-max-threads", cl::Optional,
36   cl::desc("Maximum number of threads (for emulation thread-local storage)"),
37   cl::Hidden, cl::value_desc("number"), cl::init(8));
38 
39 namespace {
40   /// Lowers thread local variables on the XCore. Each thread local variable is
41   /// expanded to an array of n elements indexed by the thread ID where n is the
42   /// fixed number hardware threads supported by the device.
43   struct XCoreLowerThreadLocal : public ModulePass {
44     static char ID;
45 
46     XCoreLowerThreadLocal() : ModulePass(ID) {
47       initializeXCoreLowerThreadLocalPass(*PassRegistry::getPassRegistry());
48     }
49 
50     bool lowerGlobal(GlobalVariable *GV);
51 
52     bool runOnModule(Module &M) override;
53   };
54 }
55 
56 char XCoreLowerThreadLocal::ID = 0;
57 
58 INITIALIZE_PASS(XCoreLowerThreadLocal, "xcore-lower-thread-local",
59                 "Lower thread local variables", false, false)
60 
61 ModulePass *llvm::createXCoreLowerThreadLocalPass() {
62   return new XCoreLowerThreadLocal();
63 }
64 
65 static ArrayType *createLoweredType(Type *OriginalType) {
66   return ArrayType::get(OriginalType, MaxThreads);
67 }
68 
69 static Constant *
70 createLoweredInitializer(ArrayType *NewType, Constant *OriginalInitializer) {
71   SmallVector<Constant *, 8> Elements(MaxThreads);
72   for (unsigned i = 0; i != MaxThreads; ++i) {
73     Elements[i] = OriginalInitializer;
74   }
75   return ConstantArray::get(NewType, Elements);
76 }
77 
78 
79 static bool replaceConstantExprOp(ConstantExpr *CE, Pass *P) {
80   do {
81     SmallVector<WeakTrackingVH, 8> WUsers(CE->users());
82     llvm::sort(WUsers);
83     WUsers.erase(std::unique(WUsers.begin(), WUsers.end()), WUsers.end());
84     while (!WUsers.empty())
85       if (WeakTrackingVH WU = WUsers.pop_back_val()) {
86         if (PHINode *PN = dyn_cast<PHINode>(WU)) {
87           for (int I = 0, E = PN->getNumIncomingValues(); I < E; ++I)
88             if (PN->getIncomingValue(I) == CE) {
89               BasicBlock *PredBB = PN->getIncomingBlock(I);
90               if (PredBB->getTerminator()->getNumSuccessors() > 1)
91                 PredBB = SplitEdge(PredBB, PN->getParent());
92               Instruction *InsertPos = PredBB->getTerminator();
93               Instruction *NewInst = createReplacementInstr(CE, InsertPos);
94               PN->setOperand(I, NewInst);
95             }
96         } else if (Instruction *Instr = dyn_cast<Instruction>(WU)) {
97           Instruction *NewInst = createReplacementInstr(CE, Instr);
98           Instr->replaceUsesOfWith(CE, NewInst);
99         } else {
100           ConstantExpr *CExpr = dyn_cast<ConstantExpr>(WU);
101           if (!CExpr || !replaceConstantExprOp(CExpr, P))
102             return false;
103         }
104       }
105   } while (CE->hasNUsesOrMore(1)); // We need to check because a recursive
106   // sibling may have used 'CE' when createReplacementInstr was called.
107   CE->destroyConstant();
108   return true;
109 }
110 
111 static bool rewriteNonInstructionUses(GlobalVariable *GV, Pass *P) {
112   SmallVector<WeakTrackingVH, 8> WUsers;
113   for (User *U : GV->users())
114     if (!isa<Instruction>(U))
115       WUsers.push_back(WeakTrackingVH(U));
116   while (!WUsers.empty())
117     if (WeakTrackingVH WU = WUsers.pop_back_val()) {
118       ConstantExpr *CE = dyn_cast<ConstantExpr>(WU);
119       if (!CE || !replaceConstantExprOp(CE, P))
120         return false;
121     }
122   return true;
123 }
124 
125 static bool isZeroLengthArray(Type *Ty) {
126   ArrayType *AT = dyn_cast<ArrayType>(Ty);
127   return AT && (AT->getNumElements() == 0);
128 }
129 
130 bool XCoreLowerThreadLocal::lowerGlobal(GlobalVariable *GV) {
131   Module *M = GV->getParent();
132   if (!GV->isThreadLocal())
133     return false;
134 
135   // Skip globals that we can't lower and leave it for the backend to error.
136   if (!rewriteNonInstructionUses(GV, this) ||
137       !GV->getType()->isSized() || isZeroLengthArray(GV->getType()))
138     return false;
139 
140   // Create replacement global.
141   ArrayType *NewType = createLoweredType(GV->getValueType());
142   Constant *NewInitializer = nullptr;
143   if (GV->hasInitializer())
144     NewInitializer = createLoweredInitializer(NewType,
145                                               GV->getInitializer());
146   GlobalVariable *NewGV =
147     new GlobalVariable(*M, NewType, GV->isConstant(), GV->getLinkage(),
148                        NewInitializer, "", nullptr,
149                        GlobalVariable::NotThreadLocal,
150                        GV->getType()->getAddressSpace(),
151                        GV->isExternallyInitialized());
152 
153   // Update uses.
154   SmallVector<User *, 16> Users(GV->users());
155   for (unsigned I = 0, E = Users.size(); I != E; ++I) {
156     User *U = Users[I];
157     Instruction *Inst = cast<Instruction>(U);
158     IRBuilder<> Builder(Inst);
159     Function *GetID = Intrinsic::getDeclaration(GV->getParent(),
160                                                 Intrinsic::xcore_getid);
161     Value *ThreadID = Builder.CreateCall(GetID, {});
162     Value *Addr = Builder.CreateInBoundsGEP(NewGV->getValueType(), NewGV,
163                                             {Builder.getInt64(0), ThreadID});
164     U->replaceUsesOfWith(GV, Addr);
165   }
166 
167   // Remove old global.
168   NewGV->takeName(GV);
169   GV->eraseFromParent();
170   return true;
171 }
172 
173 bool XCoreLowerThreadLocal::runOnModule(Module &M) {
174   // Find thread local globals.
175   bool MadeChange = false;
176   SmallVector<GlobalVariable *, 16> ThreadLocalGlobals;
177   for (GlobalVariable &GV : M.globals())
178     if (GV.isThreadLocal())
179       ThreadLocalGlobals.push_back(&GV);
180   for (unsigned I = 0, E = ThreadLocalGlobals.size(); I != E; ++I) {
181     MadeChange |= lowerGlobal(ThreadLocalGlobals[I]);
182   }
183   return MadeChange;
184 }
185