1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
|
//===-- AMDGPUAlwaysInlinePass.cpp - Promote Allocas ----------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
/// \file
/// This pass marks all internal functions as always_inline and creates
/// duplicates of all other functions a marks the duplicates as always_inline.
//
//===----------------------------------------------------------------------===//
#include "AMDGPU.h"
#include "llvm/IR/Module.h"
#include "llvm/Transforms/Utils/Cloning.h"
using namespace llvm;
namespace {
class AMDGPUAlwaysInline : public ModulePass {
static char ID;
public:
AMDGPUAlwaysInline() : ModulePass(ID) { }
bool runOnModule(Module &M) override;
const char *getPassName() const override { return "AMDGPU Always Inline Pass"; }
};
} // End anonymous namespace
char AMDGPUAlwaysInline::ID = 0;
bool AMDGPUAlwaysInline::runOnModule(Module &M) {
std::vector<Function *> FuncsToClone;
for (Function &F : M) {
if (!F.hasLocalLinkage() && !F.isDeclaration() && !F.use_empty() &&
!F.hasFnAttribute(Attribute::NoInline))
FuncsToClone.push_back(&F);
}
for (Function *F : FuncsToClone) {
ValueToValueMapTy VMap;
Function *NewFunc = CloneFunction(F, VMap);
NewFunc->setLinkage(GlobalValue::InternalLinkage);
F->replaceAllUsesWith(NewFunc);
}
for (Function &F : M) {
if (F.hasLocalLinkage() && !F.hasFnAttribute(Attribute::NoInline)) {
F.addFnAttr(Attribute::AlwaysInline);
}
}
return false;
}
ModulePass *llvm::createAMDGPUAlwaysInlinePass() {
return new AMDGPUAlwaysInline();
}
|
