aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Transforms/IPO/InlineSimple.cpp
blob: e62a8b13445f31620fe5afb5784782d9fefaf754 (plain)
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
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
//===- InlineSimple.cpp - Code to perform simple function inlining --------===//
//
// This file implements bottom-up inlining of functions into callees.
//
//===----------------------------------------------------------------------===//

#include "Inliner.h"
#include "llvm/Function.h"
#include "llvm/iMemory.h"
#include "llvm/Support/CallSite.h"
#include "llvm/Transforms/IPO.h"

namespace {
  struct SimpleInliner : public Inliner {
    int getInlineCost(CallSite CS);
  };
  RegisterOpt<SimpleInliner> X("inline", "Function Integration/Inlining");
}

Pass *createFunctionInliningPass() { return new SimpleInliner(); }


// getInlineCost - The heuristic used to determine if we should inline the
// function call or not.
//
int SimpleInliner::getInlineCost(CallSite CS) {
  Instruction *TheCall = CS.getInstruction();
  const Function *Callee = CS.getCalledFunction();
  const Function *Caller = TheCall->getParent()->getParent();

  // Don't inline a directly recursive call.
  if (Caller == Callee) return 2000000000;

  // InlineCost - This value measures how good of an inline candidate this call
  // site is to inline.  A lower inline cost make is more likely for the call to
  // be inlined.  This value may go negative.
  //
  int InlineCost = 0;

  // If there is only one call of the function, and it has internal linkage,
  // make it almost guaranteed to be inlined.
  //
  if (Callee->use_size() == 1 && Callee->hasInternalLinkage())
    InlineCost -= 30000;

  // Add to the inline quality for properties that make the call valueable to
  // inline.  This includes factors that indicate that the result of inlining
  // the function will be optimizable.  Currently this just looks at arguments
  // passed into the function.
  //
  for (CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end();
       I != E; ++I) {
    // Each argument passed in has a cost at both the caller and the callee
    // sides.  This favors functions that take many arguments over functions
    // that take few arguments.
    InlineCost -= 20;

    // If this is a function being passed in, it is very likely that we will be
    // able to turn an indirect function call into a direct function call.
    if (isa<Function>(I))
      InlineCost -= 100;

    // If a constant, global variable or alloca is passed in, inlining this
    // function is likely to allow significant future optimization possibilities
    // (constant propagation, scalar promotion, and scalarization), so encourage
    // the inlining of the function.
    //
    else if (isa<Constant>(I) || isa<GlobalVariable>(I) || isa<AllocaInst>(I))
      InlineCost -= 60;
  }

  // Now that we have considered all of the factors that make the call site more
  // likely to be inlined, look at factors that make us not want to inline it.
  // As soon as the inline quality gets negative, bail out.

  // Look at the size of the callee.  Each basic block counts as 20 units, and
  // each instruction counts as 10.
  for (Function::const_iterator BB = Callee->begin(), E = Callee->end();
       BB != E; ++BB)
    InlineCost += BB->size()*10 + 20;

  // Don't inline into something too big, which would make it bigger.  Here, we
  // count each basic block as a single unit.
  for (Function::const_iterator BB = Caller->begin(), E = Caller->end();
       BB != E; ++BB)
    InlineCost++;

  return InlineCost;
}