summaryrefslogtreecommitdiff
path: root/lib/Transforms/IPO/PartialInlining.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Transforms/IPO/PartialInlining.cpp')
-rw-r--r--lib/Transforms/IPO/PartialInlining.cpp82
1 files changed, 31 insertions, 51 deletions
diff --git a/lib/Transforms/IPO/PartialInlining.cpp b/lib/Transforms/IPO/PartialInlining.cpp
index a9cfd8ded6fb..4907e4b30519 100644
--- a/lib/Transforms/IPO/PartialInlining.cpp
+++ b/lib/Transforms/IPO/PartialInlining.cpp
@@ -202,10 +202,8 @@ struct PartialInlinerImpl {
std::function<AssumptionCache &(Function &)> *GetAC,
std::function<TargetTransformInfo &(Function &)> *GTTI,
Optional<function_ref<BlockFrequencyInfo &(Function &)>> GBFI,
- ProfileSummaryInfo *ProfSI,
- std::function<OptimizationRemarkEmitter &(Function &)> *GORE)
- : GetAssumptionCache(GetAC), GetTTI(GTTI), GetBFI(GBFI), PSI(ProfSI),
- GetORE(GORE) {}
+ ProfileSummaryInfo *ProfSI)
+ : GetAssumptionCache(GetAC), GetTTI(GTTI), GetBFI(GBFI), PSI(ProfSI) {}
bool run(Module &M);
// Main part of the transformation that calls helper functions to find
@@ -217,7 +215,7 @@ struct PartialInlinerImpl {
// outline function due to code size.
std::pair<bool, Function *> unswitchFunction(Function *F);
- // This class speculatively clones the the function to be partial inlined.
+ // This class speculatively clones the function to be partial inlined.
// At the end of partial inlining, the remaining callsites to the cloned
// function that are not partially inlined will be fixed up to reference
// the original function, and the cloned function will be erased.
@@ -271,7 +269,6 @@ private:
std::function<TargetTransformInfo &(Function &)> *GetTTI;
Optional<function_ref<BlockFrequencyInfo &(Function &)>> GetBFI;
ProfileSummaryInfo *PSI;
- std::function<OptimizationRemarkEmitter &(Function &)> *GetORE;
// Return the frequency of the OutlininingBB relative to F's entry point.
// The result is no larger than 1 and is represented using BP.
@@ -282,7 +279,8 @@ private:
// Return true if the callee of CS should be partially inlined with
// profit.
bool shouldPartialInline(CallSite CS, FunctionCloner &Cloner,
- BlockFrequency WeightedOutliningRcost);
+ BlockFrequency WeightedOutliningRcost,
+ OptimizationRemarkEmitter &ORE);
// Try to inline DuplicateFunction (cloned from F with call to
// the OutlinedFunction into its callers. Return true
@@ -337,7 +335,7 @@ private:
std::unique_ptr<FunctionOutliningInfo> computeOutliningInfo(Function *F);
std::unique_ptr<FunctionOutliningMultiRegionInfo>
- computeOutliningColdRegionsInfo(Function *F);
+ computeOutliningColdRegionsInfo(Function *F, OptimizationRemarkEmitter &ORE);
};
struct PartialInlinerLegacyPass : public ModulePass {
@@ -362,7 +360,6 @@ struct PartialInlinerLegacyPass : public ModulePass {
&getAnalysis<TargetTransformInfoWrapperPass>();
ProfileSummaryInfo *PSI =
getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI();
- std::unique_ptr<OptimizationRemarkEmitter> UPORE;
std::function<AssumptionCache &(Function &)> GetAssumptionCache =
[&ACT](Function &F) -> AssumptionCache & {
@@ -374,14 +371,7 @@ struct PartialInlinerLegacyPass : public ModulePass {
return TTIWP->getTTI(F);
};
- std::function<OptimizationRemarkEmitter &(Function &)> GetORE =
- [&UPORE](Function &F) -> OptimizationRemarkEmitter & {
- UPORE.reset(new OptimizationRemarkEmitter(&F));
- return *UPORE.get();
- };
-
- return PartialInlinerImpl(&GetAssumptionCache, &GetTTI, NoneType::None, PSI,
- &GetORE)
+ return PartialInlinerImpl(&GetAssumptionCache, &GetTTI, NoneType::None, PSI)
.run(M);
}
};
@@ -389,7 +379,8 @@ struct PartialInlinerLegacyPass : public ModulePass {
} // end anonymous namespace
std::unique_ptr<FunctionOutliningMultiRegionInfo>
-PartialInlinerImpl::computeOutliningColdRegionsInfo(Function *F) {
+PartialInlinerImpl::computeOutliningColdRegionsInfo(Function *F,
+ OptimizationRemarkEmitter &ORE) {
BasicBlock *EntryBlock = &F->front();
DominatorTree DT(*F);
@@ -403,8 +394,6 @@ PartialInlinerImpl::computeOutliningColdRegionsInfo(Function *F) {
} else
BFI = &(*GetBFI)(*F);
- auto &ORE = (*GetORE)(*F);
-
// Return if we don't have profiling information.
if (!PSI->hasInstrumentationProfile())
return std::unique_ptr<FunctionOutliningMultiRegionInfo>();
@@ -414,8 +403,7 @@ PartialInlinerImpl::computeOutliningColdRegionsInfo(Function *F) {
auto IsSingleEntry = [](SmallVectorImpl<BasicBlock *> &BlockList) {
BasicBlock *Dom = BlockList.front();
- return BlockList.size() > 1 &&
- std::distance(pred_begin(Dom), pred_end(Dom)) == 1;
+ return BlockList.size() > 1 && pred_size(Dom) == 1;
};
auto IsSingleExit =
@@ -567,10 +555,6 @@ PartialInlinerImpl::computeOutliningInfo(Function *F) {
return is_contained(successors(BB), Succ);
};
- auto SuccSize = [](BasicBlock *BB) {
- return std::distance(succ_begin(BB), succ_end(BB));
- };
-
auto IsReturnBlock = [](BasicBlock *BB) {
TerminatorInst *TI = BB->getTerminator();
return isa<ReturnInst>(TI);
@@ -607,7 +591,7 @@ PartialInlinerImpl::computeOutliningInfo(Function *F) {
if (OutliningInfo->GetNumInlinedBlocks() >= MaxNumInlineBlocks)
break;
- if (SuccSize(CurrEntry) != 2)
+ if (succ_size(CurrEntry) != 2)
break;
BasicBlock *Succ1 = *succ_begin(CurrEntry);
@@ -681,7 +665,7 @@ PartialInlinerImpl::computeOutliningInfo(Function *F) {
// peeling off dominating blocks from the outlining region:
while (OutliningInfo->GetNumInlinedBlocks() < MaxNumInlineBlocks) {
BasicBlock *Cand = OutliningInfo->NonReturnBlock;
- if (SuccSize(Cand) != 2)
+ if (succ_size(Cand) != 2)
break;
if (HasNonEntryPred(Cand))
@@ -766,19 +750,19 @@ PartialInlinerImpl::getOutliningCallBBRelativeFreq(FunctionCloner &Cloner) {
bool PartialInlinerImpl::shouldPartialInline(
CallSite CS, FunctionCloner &Cloner,
- BlockFrequency WeightedOutliningRcost) {
+ BlockFrequency WeightedOutliningRcost,
+ OptimizationRemarkEmitter &ORE) {
using namespace ore;
- if (SkipCostAnalysis)
- return true;
-
Instruction *Call = CS.getInstruction();
Function *Callee = CS.getCalledFunction();
assert(Callee == Cloner.ClonedFunc);
+ if (SkipCostAnalysis)
+ return isInlineViable(*Callee);
+
Function *Caller = CS.getCaller();
auto &CalleeTTI = (*GetTTI)(*Callee);
- auto &ORE = (*GetORE)(*Caller);
InlineCost IC = getInlineCost(CS, getInlineParams(), CalleeTTI,
*GetAssumptionCache, GetBFI, PSI, &ORE);
@@ -1270,14 +1254,14 @@ std::pair<bool, Function *> PartialInlinerImpl::unswitchFunction(Function *F) {
if (F->user_begin() == F->user_end())
return {false, nullptr};
- auto &ORE = (*GetORE)(*F);
+ OptimizationRemarkEmitter ORE(F);
// Only try to outline cold regions if we have a profile summary, which
// implies we have profiling information.
if (PSI->hasProfileSummary() && F->hasProfileData() &&
!DisableMultiRegionPartialInline) {
std::unique_ptr<FunctionOutliningMultiRegionInfo> OMRI =
- computeOutliningColdRegionsInfo(F);
+ computeOutliningColdRegionsInfo(F, ORE);
if (OMRI) {
FunctionCloner Cloner(F, OMRI.get(), ORE);
@@ -1357,11 +1341,11 @@ bool PartialInlinerImpl::tryPartialInline(FunctionCloner &Cloner) {
// inlining the function with outlining (The inliner uses the size increase to
// model the cost of inlining a callee).
if (!SkipCostAnalysis && Cloner.OutlinedRegionCost < SizeCost) {
- auto &ORE = (*GetORE)(*Cloner.OrigFunc);
+ OptimizationRemarkEmitter OrigFuncORE(Cloner.OrigFunc);
DebugLoc DLoc;
BasicBlock *Block;
std::tie(DLoc, Block) = getOneDebugLoc(Cloner.ClonedFunc);
- ORE.emit([&]() {
+ OrigFuncORE.emit([&]() {
return OptimizationRemarkAnalysis(DEBUG_TYPE, "OutlineRegionTooSmall",
DLoc, Block)
<< ore::NV("Function", Cloner.OrigFunc)
@@ -1384,7 +1368,8 @@ bool PartialInlinerImpl::tryPartialInline(FunctionCloner &Cloner) {
if (CalleeEntryCount)
computeCallsiteToProfCountMap(Cloner.ClonedFunc, CallSiteToProfCountMap);
- uint64_t CalleeEntryCountV = (CalleeEntryCount ? *CalleeEntryCount : 0);
+ uint64_t CalleeEntryCountV =
+ (CalleeEntryCount ? CalleeEntryCount.getCount() : 0);
bool AnyInline = false;
for (User *User : Users) {
@@ -1393,11 +1378,10 @@ bool PartialInlinerImpl::tryPartialInline(FunctionCloner &Cloner) {
if (IsLimitReached())
continue;
-
- if (!shouldPartialInline(CS, Cloner, WeightedRcost))
+ OptimizationRemarkEmitter CallerORE(CS.getCaller());
+ if (!shouldPartialInline(CS, Cloner, WeightedRcost, CallerORE))
continue;
- auto &ORE = (*GetORE)(*CS.getCaller());
// Construct remark before doing the inlining, as after successful inlining
// the callsite is removed.
OptimizationRemark OR(DEBUG_TYPE, "PartiallyInlined", CS.getInstruction());
@@ -1412,7 +1396,7 @@ bool PartialInlinerImpl::tryPartialInline(FunctionCloner &Cloner) {
: nullptr)))
continue;
- ORE.emit(OR);
+ CallerORE.emit(OR);
// Now update the entry count:
if (CalleeEntryCountV && CallSiteToProfCountMap.count(User)) {
@@ -1433,9 +1417,10 @@ bool PartialInlinerImpl::tryPartialInline(FunctionCloner &Cloner) {
if (AnyInline) {
Cloner.IsFunctionInlined = true;
if (CalleeEntryCount)
- Cloner.OrigFunc->setEntryCount(CalleeEntryCountV);
- auto &ORE = (*GetORE)(*Cloner.OrigFunc);
- ORE.emit([&]() {
+ Cloner.OrigFunc->setEntryCount(
+ CalleeEntryCount.setCount(CalleeEntryCountV));
+ OptimizationRemarkEmitter OrigFuncORE(Cloner.OrigFunc);
+ OrigFuncORE.emit([&]() {
return OptimizationRemark(DEBUG_TYPE, "PartiallyInlined", Cloner.OrigFunc)
<< "Partially inlined into at least one caller";
});
@@ -1517,14 +1502,9 @@ PreservedAnalyses PartialInlinerPass::run(Module &M,
return FAM.getResult<TargetIRAnalysis>(F);
};
- std::function<OptimizationRemarkEmitter &(Function &)> GetORE =
- [&FAM](Function &F) -> OptimizationRemarkEmitter & {
- return FAM.getResult<OptimizationRemarkEmitterAnalysis>(F);
- };
-
ProfileSummaryInfo *PSI = &AM.getResult<ProfileSummaryAnalysis>(M);
- if (PartialInlinerImpl(&GetAssumptionCache, &GetTTI, {GetBFI}, PSI, &GetORE)
+ if (PartialInlinerImpl(&GetAssumptionCache, &GetTTI, {GetBFI}, PSI)
.run(M))
return PreservedAnalyses::none();
return PreservedAnalyses::all();
generated by cgit v1.2.3 (git 2.25.1) at 2025-10-07 04:48:11 +0000