1 //===-- NVPTXLowerAlloca.cpp - Make alloca to use local memory =====--===//
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 // For all alloca instructions, and add a pair of cast to local address for
11 // each of them. For example,
12 //
13 // %A = alloca i32
14 // store i32 0, i32* %A ; emits st.u32
15 //
16 // will be transformed to
17 //
18 // %A = alloca i32
19 // %Local = addrspacecast i32* %A to i32 addrspace(5)*
20 // %Generic = addrspacecast i32 addrspace(5)* %A to i32*
21 // store i32 0, i32 addrspace(5)* %Generic ; emits st.local.u32
22 //
23 // And we will rely on NVPTXFavorNonGenericAddrSpace to combine the last
24 // two instructions.
25 //
26 //===----------------------------------------------------------------------===//
27
28 #include "NVPTX.h"
29 #include "NVPTXUtilities.h"
30 #include "llvm/IR/Function.h"
31 #include "llvm/IR/Instructions.h"
32 #include "llvm/IR/IntrinsicInst.h"
33 #include "llvm/IR/Module.h"
34 #include "llvm/IR/Type.h"
35 #include "llvm/Pass.h"
36
37 using namespace llvm;
38
39 namespace llvm {
40 void initializeNVPTXLowerAllocaPass(PassRegistry &);
41 }
42
43 namespace {
44 class NVPTXLowerAlloca : public BasicBlockPass {
45 bool runOnBasicBlock(BasicBlock &BB) override;
46
47 public:
48 static char ID; // Pass identification, replacement for typeid
NVPTXLowerAlloca()49 NVPTXLowerAlloca() : BasicBlockPass(ID) {}
getPassName() const50 const char *getPassName() const override {
51 return "convert address space of alloca'ed memory to local";
52 }
53 };
54 } // namespace
55
56 char NVPTXLowerAlloca::ID = 1;
57
58 INITIALIZE_PASS(NVPTXLowerAlloca, "nvptx-lower-alloca",
59 "Lower Alloca", false, false)
60
61 // =============================================================================
62 // Main function for this pass.
63 // =============================================================================
runOnBasicBlock(BasicBlock & BB)64 bool NVPTXLowerAlloca::runOnBasicBlock(BasicBlock &BB) {
65 bool Changed = false;
66 for (auto &I : BB) {
67 if (auto allocaInst = dyn_cast<AllocaInst>(&I)) {
68 Changed = true;
69 auto PTy = dyn_cast<PointerType>(allocaInst->getType());
70 auto ETy = PTy->getElementType();
71 auto LocalAddrTy = PointerType::get(ETy, ADDRESS_SPACE_LOCAL);
72 auto NewASCToLocal = new AddrSpaceCastInst(allocaInst, LocalAddrTy, "");
73 auto GenericAddrTy = PointerType::get(ETy, ADDRESS_SPACE_GENERIC);
74 auto NewASCToGeneric = new AddrSpaceCastInst(NewASCToLocal,
75 GenericAddrTy, "");
76 NewASCToLocal->insertAfter(allocaInst);
77 NewASCToGeneric->insertAfter(NewASCToLocal);
78 for (Value::use_iterator UI = allocaInst->use_begin(),
79 UE = allocaInst->use_end();
80 UI != UE; ) {
81 // Check Load, Store, GEP, and BitCast Uses on alloca and make them
82 // use the converted generic address, in order to expose non-generic
83 // addrspacecast to NVPTXFavorNonGenericAddrSpace. For other types
84 // of instructions this is unnecessary and may introduce redundant
85 // address cast.
86 const auto &AllocaUse = *UI++;
87 auto LI = dyn_cast<LoadInst>(AllocaUse.getUser());
88 if (LI && LI->getPointerOperand() == allocaInst && !LI->isVolatile()) {
89 LI->setOperand(LI->getPointerOperandIndex(), NewASCToGeneric);
90 continue;
91 }
92 auto SI = dyn_cast<StoreInst>(AllocaUse.getUser());
93 if (SI && SI->getPointerOperand() == allocaInst && !SI->isVolatile()) {
94 SI->setOperand(SI->getPointerOperandIndex(), NewASCToGeneric);
95 continue;
96 }
97 auto GI = dyn_cast<GetElementPtrInst>(AllocaUse.getUser());
98 if (GI && GI->getPointerOperand() == allocaInst) {
99 GI->setOperand(GI->getPointerOperandIndex(), NewASCToGeneric);
100 continue;
101 }
102 auto BI = dyn_cast<BitCastInst>(AllocaUse.getUser());
103 if (BI && BI->getOperand(0) == allocaInst) {
104 BI->setOperand(0, NewASCToGeneric);
105 continue;
106 }
107 }
108 }
109 }
110 return Changed;
111 }
112
createNVPTXLowerAllocaPass()113 BasicBlockPass *llvm::createNVPTXLowerAllocaPass() {
114 return new NVPTXLowerAlloca();
115 }
116