A pile of long over-due refactorings here. There are some very, *very*

minor behavior changes with this, but nothing I have seen evidence of in
the wild or expect to be meaningful. The real goal is unifying our logic
and simplifying the interfaces. A summary of the changes follows:

- Make 'callIsSmall' actually accept a callsite so it can handle
  intrinsics, and simplify callers appropriately.
- Nuke a completely bogus declaration of 'callIsSmall' that was still
  lurking in InlineCost.h... No idea how this got missed.
- Teach the 'isInstructionFree' about the various more intelligent
  'free' heuristics that got added to the inline cost analysis during
  review and testing. This mostly surrounds int->ptr and ptr->int casts.
- Switch most of the interesting parts of the inline cost analysis that
  were essentially computing 'is this instruction free?' to use the code
  metrics routine instead. This way we won't keep duplicating logic.

All of this is motivated by the desire to allow other passes to compute
a roughly equivalent 'cost' metric for a particular basic block as the
inline cost analysis. Sadly, re-using the same analysis for both is
really messy because only the actual inline cost analysis is ever going
to go to the contortions required for simplification, SROA analysis,
etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@156140 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 35 insertions, 43 deletions

diff --git a/lib/Analysis/CodeMetrics.cpp b/lib/Analysis/CodeMetrics.cpp
index 316e7bc934..acda34ba14 100644
--- a/lib/Analysis/CodeMetrics.cpp
+++ b/lib/Analysis/CodeMetrics.cpp
@@ -22,7 +22,11 @@ using namespace llvm;
 /// callIsSmall - If a call is likely to lower to a single target instruction,
 /// or is otherwise deemed small return true.
 /// TODO: Perhaps calls like memcpy, strcpy, etc?
-bool llvm::callIsSmall(const Function *F) {
+bool llvm::callIsSmall(ImmutableCallSite CS) {
+  if (isa<IntrinsicInst>(CS.getInstruction()))
+    return true;
+
+  const Function *F = CS.getCalledFunction();
   if (!F) return false;
 
   if (F->hasLocalLinkage()) return false;
@@ -79,8 +83,24 @@ bool llvm::isInstructionFree(const Instruction *I, const TargetData *TD) {
 
   if (const CastInst *CI = dyn_cast<CastInst>(I)) {
     // Noop casts, including ptr <-> int,  don't count.
-    if (CI->isLosslessCast() || isa<IntToPtrInst>(CI) || isa<PtrToIntInst>(CI))
+    if (CI->isLosslessCast())
+      return true;
+
+    Value *Op = CI->getOperand(0);
+    // An inttoptr cast is free so long as the input is a legal integer type
+    // which doesn't contain values outside the range of a pointer.
+    if (isa<IntToPtrInst>(CI) && TD &&
+        TD->isLegalInteger(Op->getType()->getScalarSizeInBits()) &&
+        Op->getType()->getScalarSizeInBits() <= TD->getPointerSizeInBits())
       return true;
+
+    // A ptrtoint cast is free so long as the result is large enough to store
+    // the pointer, and a legal integer type.
+    if (isa<PtrToIntInst>(CI) && TD &&
+        TD->isLegalInteger(Op->getType()->getScalarSizeInBits()) &&
+        Op->getType()->getScalarSizeInBits() >= TD->getPointerSizeInBits())
+      return true;
+
     // trunc to a native type is free (assuming the target has compare and
     // shift-right of the same width).
     if (TD && isa<TruncInst>(CI) &&
@@ -126,7 +146,7 @@ void CodeMetrics::analyzeBasicBlock(const BasicBlock *BB,
           isRecursive = true;
       }
 
-      if (!isa<IntrinsicInst>(II) && !callIsSmall(CS.getCalledFunction())) {
+      if (!callIsSmall(CS)) {
         // Each argument to a call takes on average one instruction to set up.
         NumInsts += CS.arg_size();
 
diff --git a/lib/Analysis/InlineCost.cpp b/lib/Analysis/InlineCost.cpp
index 3e3d2ab753..615a7e6b7e 100644
--- a/lib/Analysis/InlineCost.cpp
+++ b/lib/Analysis/InlineCost.cpp
@@ -398,10 +398,7 @@ bool CallAnalyzer::visitPtrToInt(PtrToIntInst &I) {
   if (lookupSROAArgAndCost(I.getOperand(0), SROAArg, CostIt))
     SROAArgValues[&I] = SROAArg;
 
-  // A ptrtoint cast is free so long as the result is large enough to store the
-  // pointer, and a legal integer type.
-  return TD && TD->isLegalInteger(IntegerSize) &&
-         IntegerSize >= TD->getPointerSizeInBits();
+  return isInstructionFree(&I, TD);
 }
 
 bool CallAnalyzer::visitIntToPtr(IntToPtrInst &I) {
@@ -428,10 +425,7 @@ bool CallAnalyzer::visitIntToPtr(IntToPtrInst &I) {
   if (lookupSROAArgAndCost(Op, SROAArg, CostIt))
     SROAArgValues[&I] = SROAArg;
 
-  // An inttoptr cast is free so long as the input is a legal integer type
-  // which doesn't contain values outside the range of a pointer.
-  return TD && TD->isLegalInteger(IntegerSize) &&
-         IntegerSize <= TD->getPointerSizeInBits();
+  return isInstructionFree(&I, TD);
 }
 
 bool CallAnalyzer::visitCastInst(CastInst &I) {
@@ -445,24 +439,7 @@ bool CallAnalyzer::visitCastInst(CastInst &I) {
   // Disable SROA in the face of arbitrary casts we don't whitelist elsewhere.
   disableSROA(I.getOperand(0));
 
-  // No-op casts don't have any cost.
-  if (I.isLosslessCast())
-    return true;
-
-  // trunc to a native type is free (assuming the target has compare and
-  // shift-right of the same width).
-  if (TD && isa<TruncInst>(I) &&
-      TD->isLegalInteger(TD->getTypeSizeInBits(I.getType())))
-    return true;
-
-  // Result of a cmp instruction is often extended (to be used by other
-  // cmp instructions, logical or return instructions). These are usually
-  // no-ops on most sane targets.
-  if (isa<CmpInst>(I.getOperand(0)))
-    return true;
-
-  // Assume the rest of the casts require work.
-  return false;
+  return isInstructionFree(&I, TD);
 }
 
 bool CallAnalyzer::visitUnaryInstruction(UnaryInstruction &I) {
@@ -636,21 +613,11 @@ bool CallAnalyzer::visitCallSite(CallSite CS) {
     default:
       return Base::visitCallSite(CS);
 
-    case Intrinsic::dbg_declare:
-    case Intrinsic::dbg_value:
-    case Intrinsic::invariant_start:
-    case Intrinsic::invariant_end:
-    case Intrinsic::lifetime_start:
-    case Intrinsic::lifetime_end:
     case Intrinsic::memset:
     case Intrinsic::memcpy:
     case Intrinsic::memmove:
-    case Intrinsic::objectsize:
-    case Intrinsic::ptr_annotation:
-    case Intrinsic::var_annotation:
-      // SROA can usually chew through these intrinsics and they have no cost
-      // so don't pay the price of analyzing them in detail.
-      return true;
+      // SROA can usually chew through these intrinsics, but they aren't free.
+      return false;
     }
   }
 
@@ -662,7 +629,7 @@ bool CallAnalyzer::visitCallSite(CallSite CS) {
       return false;
     }
 
-    if (!callIsSmall(F)) {
+    if (!callIsSmall(CS)) {
       // We account for the average 1 instruction per call argument setup
       // here.
       Cost += CS.arg_size() * InlineConstants::InstrCost;
@@ -706,6 +673,11 @@ bool CallAnalyzer::visitCallSite(CallSite CS) {
 }
 
 bool CallAnalyzer::visitInstruction(Instruction &I) {
+  // Some instructions are free. All of the free intrinsics can also be
+  // handled by SROA, etc.
+  if (isInstructionFree(&I, TD))
+    return true;
+
   // We found something we don't understand or can't handle. Mark any SROA-able
   // values in the operand list as no longer viable.
   for (User::op_iterator OI = I.op_begin(), OE = I.op_end(); OI != OE; ++OI)
diff --git a/lib/Transforms/Scalar/TailRecursionElimination.cpp b/lib/Transforms/Scalar/TailRecursionElimination.cpp
index e21eb9dca2..074d0325d9 100644
--- a/lib/Transforms/Scalar/TailRecursionElimination.cpp
+++ b/lib/Transforms/Scalar/TailRecursionElimination.cpp
@@ -391,7 +391,7 @@ TailCallElim::FindTRECandidate(Instruction *TI,
   if (BB == &F->getEntryBlock() && 
       FirstNonDbg(BB->front()) == CI &&
       FirstNonDbg(llvm::next(BB->begin())) == TI &&
-      callIsSmall(F)) {
+      callIsSmall(CI)) {
     // A single-block function with just a call and a return. Check that
     // the arguments match.
     CallSite::arg_iterator I = CallSite(CI).arg_begin(),