1 //===-- AMDGPUAlwaysInlinePass.cpp - Promote Allocas ----------------------===//
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 /// \file
11 /// This pass marks all internal functions as always_inline and creates
12 /// duplicates of all other functions and marks the duplicates as always_inline.
13 //
14 //===----------------------------------------------------------------------===//
15 
16 #include "AMDGPU.h"
17 #include "llvm/IR/Module.h"
18 #include "llvm/Transforms/Utils/Cloning.h"
19 
20 using namespace llvm;
21 
22 namespace {
23 
24 static cl::opt<bool> StressCalls(
25   "amdgpu-stress-function-calls",
26   cl::Hidden,
27   cl::desc("Force all functions to be noinline"),
28   cl::init(false));
29 
30 class AMDGPUAlwaysInline : public ModulePass {
31   bool GlobalOpt;
32 
33 public:
34   static char ID;
35 
36   AMDGPUAlwaysInline(bool GlobalOpt = false) :
37     ModulePass(ID), GlobalOpt(GlobalOpt) { }
38   bool runOnModule(Module &M) override;
39   StringRef getPassName() const override { return "AMDGPU Always Inline Pass"; }
40 };
41 
42 } // End anonymous namespace
43 
44 INITIALIZE_PASS(AMDGPUAlwaysInline, "amdgpu-always-inline",
45                 "AMDGPU Inline All Functions", false, false)
46 
47 char AMDGPUAlwaysInline::ID = 0;
48 
49 bool AMDGPUAlwaysInline::runOnModule(Module &M) {
50   std::vector<GlobalAlias*> AliasesToRemove;
51   std::vector<Function *> FuncsToClone;
52 
53   for (GlobalAlias &A : M.aliases()) {
54     if (Function* F = dyn_cast<Function>(A.getAliasee())) {
55       A.replaceAllUsesWith(F);
56       AliasesToRemove.push_back(&A);
57     }
58   }
59 
60   if (GlobalOpt) {
61     for (GlobalAlias* A : AliasesToRemove) {
62       A->eraseFromParent();
63     }
64   }
65 
66   auto NewAttr = StressCalls ? Attribute::NoInline : Attribute::AlwaysInline;
67   auto IncompatAttr
68     = StressCalls ? Attribute::AlwaysInline : Attribute::NoInline;
69 
70   for (Function &F : M) {
71     if (!F.hasLocalLinkage() && !F.isDeclaration() && !F.use_empty() &&
72         !F.hasFnAttribute(IncompatAttr))
73       FuncsToClone.push_back(&F);
74   }
75 
76   for (Function *F : FuncsToClone) {
77     ValueToValueMapTy VMap;
78     Function *NewFunc = CloneFunction(F, VMap);
79     NewFunc->setLinkage(GlobalValue::InternalLinkage);
80     F->replaceAllUsesWith(NewFunc);
81   }
82 
83   for (Function &F : M) {
84     if (F.hasLocalLinkage() && !F.hasFnAttribute(IncompatAttr)) {
85       F.addFnAttr(NewAttr);
86     }
87   }
88   return false;
89 }
90 
91 ModulePass *llvm::createAMDGPUAlwaysInlinePass(bool GlobalOpt) {
92   return new AMDGPUAlwaysInline(GlobalOpt);
93 }
94